Nucl - Phys.B v.579

Nuclear Physics B 579 (2000) 316
www.elsevier.nl/locate/npe
Looking for TeV-scale strings

and extra-dimensions
E. Accomando a , I. Antoniadis b , K. Benakli c
a Center for Theoretical Physics, Texas A&M University, College Station, TX 77843-4242, USA
b Centre de Physique Thorique, Ecole Polytechnique, 91128 Palaiseau, France
c Unit mixte du CNRS et de lEP, UMR 7644, CERN Theory Division CH-1211, Genve 23, Switzerland
Received 21 January 2000; accepted 22 February 2000
Abstract
In contrast to the old heterotic string case, the (weakly coupled) type I brane framework allows
to have all, part or none of the standard model gauge group factors propagating in large extradimensions of TeV1 size. We investigate the main experimental signatures of these possibilities,
related to the production of KaluzaKlein excitations of gluons and electroweak gauge bosons.
A discovery through direct observation of resonances is possible only for compactification scales
below 6 TeV. However effects due to exchange of virtual KaluzaKlein excitations could be observed
for higher scales. We find that LHC can probe compactification scales as high as 20 TeV for
excitations of gluons and 815 TeV for excitations of electroweak gauge bosons. Finally, in the
case where no gauge boson feels the extra-dimension, we find that effective contact interactions due
to massive string mode oscillations dominate those due to the exchange of KaluzaKlein excitations
of gravitons and could be used to obtain bounds on the string scale. 2000 Elsevier Science B.V.
All rights reserved.
1. Introduction
A lot of efforts have been devoted to understand possible patterns of new physics beyond
the standard model. One of the most spectacular possibilities is the proposal of existence of
large extra-dimensions compactified in the TeV range [1]. Such a scenario is easily realized
if the fundamental string or quantum gravity scale is low [213].
Previous studies of the phenomenological implications of such extra-dimensions were
focused on the simplest scenario where all standard model gauge bosons propagate in the
same compact space [1433].
Although this is the simplest scenario favored by gauge coupling unification [3448], it
does not constitute the most general case. The different factors of the standard model gauge
group may arise from branes of different kinds, extended in different compact directions.
0550-3213/00/$ see front matter 2000 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 1 2 3 - 1
E. Accomando et al. / Nuclear Physics B 579 (2000) 316
As a result, TeV-dimensions can be longitudinal along the world-volume of some branes

and transverse to others.
A number of important questions arise in each of these cases: what are the experimental
bounds on the size of such extra-dimensions? Do these bounds allow the probe of extradimensions at future collider experiments? How can one distinguish the corresponding
signals from other possible origin of new physics, such as models with new gauge
bosons?
These questions can be addressed only in a well defined theoretical framework, where
scenaria with large extra-dimensions can be realized consistently, such as perturbative
type I string theory with fundamental scale Ms of the order of a few TeV. 1 Our
analysis concerns then longitudinal dimensions felt by gauge interactions, associated to
compactification scales R 1 somewhat smaller than the string scale, R 1 < Ms . The
reason is that for string size dimensions, KK (KaluzaKlein) excitations have masses
comparable to string massive modes and their effects become indistinguishable. On the
other hand, transverse dimensions larger than the string length are associated to superheavy
open string winding modes which decouple at low energies; 2 they can be probed only
through gravitational interactions associated to closed string KK modes.
Existing lower bounds on the compactification scale in the various cases of large extradimensions, come from indirect effects of KK modes in low or high energy measurements
and are of the order of 2 to 4 TeV. These leave very little hope for production of many KK
excitations on-shell at LHC and none for other ongoing machines. In the most optimistic
case, one will be able to observe just the first excitation, and thus, additional information
or high precision will be necessary to bring evidence for its higher dimensional origin.
In Section 2 we discuss the brane picture in models of type I strings on orbifolds. This
will allow us to illustrate different possibilities for localizing different parts of the observed
particles on different branes. We obtain five classes of models that we study in the following
sections.
In Section 3, we discuss the standard scenario where all the three gauge factors of
the standard model feel the extra-dimension. New bounds on the upper value of the
compactification scale accessible at LHC are obtained. These arise from the increase of
luminosity compared to [30] and the careful study of gluon and W channels that have not
received much of attention in the past (see [3133]).
In Section 4, we discuss the collider signatures of the three other cases where only part
of the standard model gauge group feels the large extra-dimension.
Section 5 deals with the configuration where there are no longitudinal dimensions. We
point out that in this case the effective contact interactions are dominated by effects due
to tree-level exchange of massive string oscillation modes and the usually considered
processes of virtual KK gravitons are sub-dominant. Finally Section 6 summarizes our
quantitative results.
1 For notational simplicity, in this work we refer to type I any type I0 vacuum obtained by T-dualities from the
original type I theory.
2 This is true if bulk fields propagate always in more than two transverse dimensions, or if tadpoles cancel
locally [4].
2. Bulk and boundary states within the brane picture

In this work we are interested in collider experiments taking place at energies above the
electroweak scale. Effects of the electroweak breaking can therefore be neglected and one
can consider brane configurations that lead to unbroken SU (3)c SU (2)w U (1)Y gauge
symmetry. Although no real realistic model has emerged yet from type I vacua [1013,
4958] we can discuss the main properties associated to different possible classes of such
models.
In the generic case, the 3 group factors of the standard model may arise from different
collections of coincident branes. If the space dimensionality p of a set of branes (p-branes)
is bigger than 3, there are p 3 longitudinal compact dimensions felt by the brane gauge
interactions associated to the KK excitations with masses
Mn2 =
n2
Rk2
(1)
for integer n. Here, Rk denotes generically the radii of longitudinal dimensions. The
remaining 9 p transverse dimensions are felt only by gravity. The brane world-volume
fields have no KK excitations along these directions but (superheavy) winding modes with
masses |n|R Ms2 , where R denotes generically the radii of transverse dimensions. The
size of compact dimensions is constrained by present observations to be less than TeV1
for the longitudinal ones and less than millimeter for transverse. Below, we consider a
particular dimension with a compactification scale in the TeV region, roughly an order of
magnitude smaller than the string scale R 1 O(101 )Ms . The 3 sets of branes associated
with the three standard model gauge group factors can then be either longitudinal or
transverse with respect to this particular direction.
Since the coupling constant of the gauge group living on the longitudinal branes is
reduced by the size of the large dimension RMs compared to those of the transverse
branes, if SU (3) has KK modes all three group factors must have. Otherwise it is difficult
to reconcile the suppression of the strong coupling at the string scale with the observed
reverse situation. As a result, there are 5 distinct cases that we denote (l, l, l), (t, l, l),
(t, l, t), (t, t, l) and (t, t, t), where l (t) refers to branes longitudinal (transverse) to the
large dimension and the three positions in the brackets correspond to the 3 gauge group
factors of the standard model SU (3)c SU (2)w U (1)Y . In the first two cases, we will
also comment on the anisotropic possibility of having more than one dimensions larger
than the string length. We will ignore the possibility that a gauge factor is associated to a
linear combination of l and t type of branes. Since the longitudinal component gives rise
to interactions suppressed by RMs , this can be of importance only if the t component is
suppressed by a small mixing angle.
Similarly, there are three kinds of matter fields. Those which live on longitudinal or
on transverse branes and those which arise as massless modes of open strings stretched
between two different types of branes and live on the corresponding intersections. The first
two types have KK and winding modes, respectively, along the large dimension, while
the latter have no excitations and behave as the Z2 twisted (boundary) states of heterotic
strings on orbifolds. 3 The boundary states couple to all KK modes of gauge fields in the
same way. 4 These couplings violate obviously momentum conservation in the compact
direction and make all massive KK excitations unstable. If all quarks and leptons are of
the first two kinds of matter fields living on branes, their interactions preserve the internal
momentum and KK excitations (or windings) can be produced only in pairs and are stable.
Naively, such a possibility is excluded, for instance from the non-observation of stable
excitations bounded to atoms. However, this assumes their production during the early
history of the universe which might be suppressed if the maximal temperature is very
low as suggested in [5961]. In the following we ignore this possibility and assume that
all matter fields are localized in the large dimension, with the exception of the (t, t, t) case
which we treat in Section 5.
In addition to the KK excitations of massless modes there might be massive modes with
different quantum numbers. Their corresponding massless modes are projected out since
they transform non-trivially under the orientifold projection that leads to chiral spectrum.
These odd states can be new gauge bosons that enlarge the gauge symmetry or extra
fermions and scalars. They can decay only to localized matter fields, but their coupling
vanishes at the lowest order. Non-vanishing couplings can be obtained by taking derivatives
along the extra-dimension and are therefore suppressed by powers of RMs . For instance,
the lowest order of such operators (in non-supersymmetric models) involves three scalars
one of which is odd and has a derivative acting on it. These states are in general model
dependent and will not be discussed here.
3. The standard (l, l, l) scenario

In the (l, l, l) scenario all gauge bosons of the standard model SU (3)c SU (2)w
U (1)Y have KK excitations. Because matter fields are localized, their interactions do not
preserve the momenta in the extra-dimension and single KK excitations can be produced.
This means for example that QCD processes q q G(n) with q representing quarks
and G(n) massive KK excitations of gluons are allowed. In contrast processes such as
GG G(n) are forbidden as gauge boson interactions conserve the internal momenta. The
exchange of virtual KK modes leads to effects in low energy processes that are constrained
by fits to high precision data [2329]. For instance, fit of measured values of MW , ll and
had lead to R 1 & 3.5 TeV. Inclusion of QW measurement, which does not give a good
agreement with the standard model itself, raises the bound to R 1 & 3.9 TeV [29].
For the LHC experiment, this scenario predicts that new resonances in three different
channels l + l , l and dijets could be observed at the same mass if the compactification
scale is low enough. The resonances due to the SU (3) gluon excitations have quite large
widths due to the strong coupling value. They are thus spread and difficult to detect
already for compactification scales of the order of 5 TeV. In Fig. 1 we show the shape of
3 In contrast to the heterotic case open strings do not lead to Z twisted matter with N > 2.
N
4 Actually, when one is restricted to physical states with positive KK momentum n (see Eq. (1)), the coupling
of massive excitations is enhanced by a factor
2 relative to the lowest (massless) one.
Fig. 1. First resonances in the LHC experiment due to a KK excitation of gluon for one
extra-dimension at 4 and 5 TeV.
a typical expected signal for the production of a gluon excitation for R 1 = 4 and 5 TeV
at the LHC (assuming s = 14 TeV and L = 100 fb1 ). Our results have been obtained
using the CTEQ4L structure functions [62]. We have also included the multiplicative K
factor [63], taken to be K = 1.1, and summed over all jets, top excluded. Moreover, we
have implemented a rapidity cut, || 6 0.5, on both jets and required the invariant mass
to be Mjj 0 > 2 TeV, which reduces the SM background and gives the optimal ratio S/ B
especially for large masses.
The strength of the coupling constant of gluon excitations implies on the other hand a
large excess in the total number of events from the QCD expected ones. We found that
looking for an excess of dijet events could be the most efficient channel to constrain the
size of extra-dimensions. In Fig. 2, we present the effect of these gluonic KK states on
the total number of dijet events. The significance, given by the ratio |NT NSM |/ NSM
where NT is the total number of events and NSM is the SM background, indicates that
the LHC could probe in this hadronic channel values of the compactification scale up to
about 20 TeV. This becomes of the order of 15 TeV for a lower luminosity, L = 10 fb1 ,
as used in [30]. The Tevatron run II, with luminosity of L = 20 fb1 , could instead probe
compactification scales up to about 4 TeV.
(n)
The KK excitations W of the W bosons could show up in clean final states consisting
of one charged lepton and missing energy (l l with l = e, ). The differential crosssection in the transverse mass has the shape illustrated for R 1 = 4, 5 and 6 TeV in Fig. 3.
We have required the charged lepton to be in the central region, || 6 1. This cut reduces
the SM background by a factor of about 50%, while decreasing the pure signal only by
Fig. 2. Number of standard deviation in number of observed dijets from the expected standard model
value, due to the presence of a TeV-scale extra-dimension of compactification radius R.
Fig. 3. First resonances in the LHC experiment due to a KK excitation of W for one extra-dimension
at 4, 5 and 6 TeV. We plot the differential cross section as function of the transverse mass for the W s.
Fig. 4. First resonances in the LHC experiment due to a KK excitation of photon and Z for one
extra-dimension at 4 TeV. From highest to lowest: excitation of photon + Z, photon and Z boson.
35% due to the large masses considered. The discovery limit of such a resonance is around
6 TeV.
Approximately the same values of compactification scales, R 1 . 6 TeV, could be
discovered through the observation of a resonance in the l + l final state (with l = e, ).
The interference in the overlap of the resonances (n) of the photon and Z (n) leads to an
a priori distinctive deep as noticed in [3033]. The shape of the resonance is then quite
different from the one of a BrightWigner due to the excitation of only one gauge boson,
as can be seen in Fig. 4.
In a way similar to the case of excitations of gluons, the exchange of virtual KK
(n)
excitations W , (n) and Z (n) lead to deviations from the standard model expectation
in the number of total events. These allow to probe higher compactification scales. Fig. 5
shows that, at 95% confidence level, the LHC could exclude values of compactification
(n)
scales up to 12, 14 and 15 TeV from the (n) + Z (n) , W and combined channels,
respectively. Here, for the dilepton final state, we have required one lepton to be in the
central region, |l | 6 1, the other one having a looser cut |l 0 | 6 2.4. This gives about
the same acceptance as above. Moreover, we have chosen a 400 GeV lower bound on the
transverse and invariant mass, in order to optimize the significance.
4. The cases (t, l, l), (t, l, t) and (t, t, l)

The difference between the (t, l, l) and the (l, l, l) scenario is that in the former the
gluons do not feel the extra-dimensions. Only the KK excitations W(n) , (n) and Z (n)
10
Fig. 5. Number of standard deviation in the number of l + l pairs and l l pairs produced from the
expected standard model value due to the presence of one extra-dimension of radius R.
are present and lead to the same effects as for the (l, l, l) case. The limits on the extradimensions follow from the previous section: 6 TeV for discovery and 15 TeV for the
exclusion bounds.
In the (t, l, t) case, only the SU (2) factor arises from a set of branes longitudinal to
the TeV-scale extra-dimension. As all the states charged under U (1)Y are localized in this
case, one has no more the freedom to choose the Higgs to be localized or not as it was
for the cases (l, l, l) and (t, l, l) but only the first possibility exists. The massive modes
are KK excitations of W and W3 . The latter will lead to a deficit in l + l half of the
corresponding one in the l channel (as two gauge bosons contribute here). Moreover,
the forward-backward asymmetry is maximal. The limits on this scenario arise from the
(n)
exchange of W and are again 6 TeV for discovery and 14 TeV for the exclusion bounds.
The third case is the one of (t, t, l) channel where only U (1)Y feels the extra-dimension.
Similarly to the previous case, the Higgs doublets have to be identified with localized
states. In this case the limits are weaker, the exclusion bound is in fact around 8 TeV, as
can be seen in Fig. 6.
Finally, let us comment about some of the possibilities that we neglected. In the scenaria
discussed above, we have assumed that only one extra-dimension is large and longitudinal
to some sets of branes. In the (t, l, l) and the (l, l, l) cases one could imagine that the
different gauge factors arise from different sets of branes feeling different dimensions. In
this case, the (n) + Z (n) resonances split into resonances of W3 and U (1)Y located at
different masses. Another possibility to notice is the case where part of U (1)Y is t and part
11
Fig. 6. Number of standard deviation in the number of l + l pairs produced from the expected
standard model value due to the presence of one extra-dimension of radius R in the case of (t, l, t),
i.e., W3 KK excitations and the case of (t, t, l), i.e., U (1)Y boson KK excitations.
is l; in this scenario however one would observe a signal which is hard to distinguish from
a generic extra U (1)0 . A good statistics would be needed to see the deviation in the tail of
the resonance as being due to effects additional to those of the U (1)0 itself.
5. Limit on the string scale

Finally we discuss the case (t, t, t) where all the extra-dimensions are transverse to the
3-brane where the standard model lives. In this case standard model particles have no
KK excitations and the main experimental signals are due either to gravitational effects
associated to closed strings propagating in the ten-dimensional bulk or to the exchange of
massive string oscillation modes. The gravitational effects were extensively studied in the
literature based on graviton exchange in the effective field theory [6488]. In the context
of type I string models, these effects are part of the one-loop diagram of open strings
(annulus) that can be seen as closed strings exchanged in the transverse channel (cylinder).
However these are subdominant, O(gs2 ) with gs the string coupling, compared to the tree
level exchange of open string oscillation modes, which are of order O(gs ). 5
Indeed, the exchange of virtual gravitons is described in the effective field theory by an
amplitude of the form S(s)T with [64,65]:
5 This observation is also made recently by M. Peskin in [89].
12
T = T T
1
T T
1 + d
(2)
and
S=
1 X 1
,
Mp2 n s nE22
R
(3)
is the energy momentum tensor and s the center of mass energy. The sum
where
in S is divergent for a number d > 1 of transverse dimensions. From the string theory
point of view this corresponds to an ultraviolet one-loop divergence appropriately cutoff by the string scale. The final result is finite and model (compactification) dependent.
A phenomenological approach followed in [14,6688] is to estimate S as
S = gs2
A
,
Ms4
(4)
M2
where A = log ss for d = 2 and A = d22 for d > 2. The coupling constant factor gs2
arises from the type I relation Mp2 = Ms2 (Ms R )d /gs2 and the fact that the divergent sum
is cut-off at the string scale Ms .
On the other hand, the tree-level open string four-point amplitude is:
A(1, 2, 3, 4) =
(1 s/Ms2 ) (1 t/Ms2 )
1
K(1, 2, 3, 4)
gs
2
Ms
(1 s/Ms2 t/Ms2 )
+ [s t] + [s u],
(5)
where K contains kinematic factors as well as gauge indices (ChanPaton factors) [90].
We have introduced the Mandelstam kinematic invariants:
s = (p1 + p2 )2 ,
t = (p1 + p3 )2 ,
u = (p1 + p4 )2
(6)
with pi the momentum of state i. Expanding the Gamma-functions at low energies in

powers of s/Ms2 , and summing over the three permutations, one finds for the 4-fermion
amplitude:

2 s2
2 t2
As
At
1+
1
+
+
g
A(1, 2, 3, 4) ' gs
s
s
12 Ms4
t
12 Ms4

2 u2
Au
1+
.
(7)
+ gs
u
12 Ms4
Note that the leading contribution gs ( As s + At t + Auu ) reproduces the point-like particle
result. In As we have absorbed the kinematical as well as the trace on ChanPaton trace
factors:
As = T r(1 2 3 4 )u 1 u2 u 3 u4 ,
(8)
where i are the ChanPaton factors, ui are the associated spinors and we dropped
numerical factors. The At and Au are obtained by replacing the cyclic order 1234 by 1342
and 1423, respectively. For example, in the usual Bhabha scattering, Au vanishes due to
13
the traces on ChanPaton factors and the two terms with poles in s and t channels remain.
The next terms describe effective contact interactions. A comparison with Eq. (4) shows
that they are enhanced by a string-loop factor gs1 with respect to the field theory estimate
for KK graviton exchanges. Although the precise value of gs requires a detailed analysis
of threshold corrections, for a rough estimate one could take gs ' 1/25, that implies
an enhancement by an order of magnitude.
As a result, one can not get reliable experimental bounds on the string scale just based
on fits of differential cross-sections with field theoretical estimates for the contribution of
graviton KK exchange, but should use processes with missing energy due to emission of
on-shell gravitons in the bulk [64,65]. Alternatively, the four-point contact interactions
analyzed above may also be used to provide direct bounds on the string scale. Note however
that these interactions are in general model dependent.
Above we have considered the case of 4-fermion interactions which receive lowest order
contributions at tree-level from massless fields. However, one could also consider processes
for which the dominant lowest order contributions arise from exchange of massive open
string oscillator modes. Such an example is the four-point amplitude of Abelian gauge
fields, which gives rise to leading contact interactions encoded in the BornInfeld action
[90].
6. Conclusions
We have described various possible models allowed by the type I picture of the braneworld scenario. Assuming that matter fermions are localized in some large TeV extradimension and that there is no inverse hierarchy of the tree-level gauge couplings compared
to the observed (loop-corrected) ones, we found 5 classes of models, denoted as (l, l, l),
(t, l, l), (t, l, t), (t, t, l) and (t, t, t). For these models we have attempted to answer the
questions raised in the introduction. What are the experimental bounds on the size of
such extra-dimensions? Lower bounds on the compactification scale from low energy
measurements are of the order of 3 to 4 TeV.
Do these bounds allow the probe of extra-dimensions at future collider experiments? We
have found that new resonances can be discovered at LHC for scales as high as around
6 TeV for all the four cases (l, l, l), (t, l, l), (t, l, t) and (t, t, l). This implies that only
a small window of 34 to 6 TeV for the compactification scale is left. Improvement of
low energy precision bounds from existing experiments (LEP, Tevatron, etc.) are thus of
great importance as they may allow to narrow further this window. Higher scales can be
probed through indirect effects of exchanges of virtual KK excitations. We found that the
most efficient channel is through KK excitations of gluons which give sensitivity up to
R 1 . 20 TeV. The W , photon + Z and U (1)Y boson excitations allow to probe values
R 1 . 14, 12 and 8 TeV, respectively.
How can one distinguish the corresponding signals from other possible origin of new
physics, such as models with new gauge bosons? The discussion on this issue follows the
one for the (t, l, l) case in [3033]. We have pointed out that scenaria (l, l, l), (t, l, l) and
14
(t, l, t) predict the existence of resonances in three or two different final states. As for the
(t, l, l) case in [3033], we have pointed out that models (l, l, l) and (t, l, t) predict the
existence of resonances in three and two final channels, respectively. These resonances are
located at the same mass value. This property is not shared by most of other new gauge
boson models. Moreover, the heights and widths of the resonances are directly related to
those of standard model gauge bosons in the corresponding channels. Also, in the case of
excitations of the photon + Z a deep due to the interference between the two bosons should
be observed just before the resonance.
For the last case (t, t, t), we pointed out that there are open-string tree-level contact
terms that are dominant with respect of those induced by the exchange of KK excitations
of gravitons. Analysis of these terms should allow in principle to get stronger bounds
on the string scale. The precise coefficients, however, of the higher-dimensional effective
operators are in general model (compactification) dependent and further investigation is
needed.
Acknowledgement
We wish to thank T. Kamon and Y. Oz for discussions at various stages of the work.
KB wishes also to thank A. Delgado, A. Pomarol, M. Quiros and J. Lykken for discussion.
This work is supported in part by the EEC under TMR contract ERBFMRX-CT96-0090
and in part by NSF grant No. PHY-9722090.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
I. Antoniadis, Phys. Lett. B 246 (1990) 377.

N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 429 (1998) 263.
I. Antoniadis, N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 436 (1998) 257.
I. Antoniadis, C. Bachas, Phys. Lett. B 450 (1999) 83.
J.D. Lykken, Phys. Rev. D 54 (1996) 3693.
I. Antoniadis, B. Pioline, Phys. Rev. B 550 (1999) 41.
K. Benakli, Y. Oz, hep-th/9910090.
K. Benakli, Phys. Rev. B 60 (1999) 104002.
C.P. Burgess, L.E. Ibez, F. Quevedo, Phys. Lett. D 447 (1999) 257.
G. Shiu, S.-H.H. Tye, Phys. Rev. B 58 (1998) 106007.
Z. Kakushadze, S.-H.H. Tye, Phys. Rev. B 548 (1999) 180.
L.E. Ibez, C. Muoz, S. Rigolin, hep-ph/9812397.
G. Aldazabal, L.E. Ibez, F. Quevedo, hep-th/9909172.
I. Antoniadis, K. Benakli, Phys. Lett. B 326 (1994) 69.
I. Antoniadis, C. Muoz, M. Quirs, Phys. Rev. B 397 (1993) 515.
I. Antoniadis, K. Benakli, M. Quirs, Phys. Lett. B 331 (1994) 313.
K. Benakli, Phys. Lett. B 386 (1996) 106.
I. Antoniadis, M. Quirs, Phys. Lett. B 392 (1997) 61.
I. Antoniadis, S. Dimopoulos, G. Dvali, Phys. Rev. B 516 (1998) 70.
A. Pomarol, M. Quirs, Phys. Lett. B 438 (1998) 225.
I. Antoniadis, S. Dimopoulos, A. Pomarol, M. Quirs, Phys. Rev. B 544 (1999) 503.
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
[50]
[51]
[52]
[53]
[54]
[55]
[56]
[57]
[58]
[59]
[60]
[61]
[62]
[63]
[64]
[65]
[66]
[67]
[68]
[69]
[70]
15
A. Delgado, A. Pomarol, M. Quirs, hep-ph/9812489.

P. Nath, M. Yamaguchi, hep-ph/9902323; hep-ph/9903298.
M. Masip, A. Pomarol, hep-ph/9902467.
W.J. Marciano, Phys. Rev. D 60 (1999) 093006.
A. Strumia, hep-ph/9906266.
R. Casalbuoni, S. De Curtis, D. Dominici, R. Gatto, hep-ph/9907355.
C.D. Carone, hep-ph/9907362.
A. Delgado, A. Pomarol, M. Quirs, hep-ph/9911252.
I. Antoniadis, K. Benakli, M. Quiros, Phys. Lett. B 460 (1999) 176.
P. Nath, Y. Yamada, M. Yamaguchi, hep-ph/9905415.
T.G. Rizzo, J.D. Wells, hep-ph/9906234.
T.G. Rizzo, hep-ph/9909232.
K.R. Dienes, E. Dudas, T. Gherghetta, Phys. Lett. B 436 (1998) 55; Phys. Rev. B 537 (1999)
47.
D. Ghilencea, G.G. Ross, Phys. Lett. B 442 (1998) 165.
Z. Kakushadze, hep-th/9811193.
C.D. Carone, hep-ph/9902407.
A. Delgado, M. Quirs, hep-ph/9903400.
P. Frampton, A. Rasin, hep-ph/9903479.
A. Prez-Lorenzana, R.N. Mohapatra, hep-ph/9904504.
Z. Kakushadze, T.R. Taylor, hep-th/9905137.
K. Huitu, T. Kobayashi, hep-ph/9906431.
Y. Abe, C. Hattori, M. Ito, M. Matsunaga, T. Matsuoka, hep-ph/9902472.
T. Kobayashi, J. Kubo, G. Zoupanos, hep-ph/9812221.
H.-C. Cheng, B.A. Dobrescu, C.T. Hill, hep-ph/9906327.
C. Bachas, JHEP 9811 (1998) 23.
I. Antoniadis, C. Bachas, E. Dudas, hep-th/9906039.
N. Arkani-Hamed, S. Dimopoulos, J. March-Russell, hep-th/9908146.
A. Sagnotti, hep-th/9302099.
C. Angelantonj, M. Bianchi, G. Pradisi, A. Sagnotti, Ya.S. Stanev, Phys. Lett.B 385 (1996) 96.
Z. Kakushadze, G. Shiu, Phys. Rev. D 56 (1997) 3686; Phys. Rev. B 520 (1998) 75.
M. Berkooz, R. Leigh, Phys. Rev. B 483 (1997) 187.
G. Zwart, hep-th/9708040.
Z. Kakushadze, Phys. Rev. B 512 (1998) 221.
G. Aldazabal, A. Font, L.E. Ibez, G. Violero, hep-th/9804026.
J. Lykken, E. Poppitz, S.P. Trivedi, Phys. Rev. B 543 (1998) 105.
G. Aldazabal, D. Badagnani, L.E. Ibez, A. Uranga, hep-th/9904071.
C. Angelantonj, I. Antoniadis, G. DAppollonio, E. Dudas, A. Sagnotti, hep-th/9911081.
N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Rev. D 59 (1999) 086004.
K. Benakli, S. Davidson, Phys. Rev. D 60 (1999) 025004.
L.J. Hall, D. Smith, Phys. Rev. D 60 (1999) 085008.
H.J. Lai, J. Huston, S. Kuhlmann, F. Olness, J. Owens, D. Soper, W.K. Tung, H. Weerts, Phys.
Rev. D 55 (1997) 1280.
R. Hamberg, W.L. van Neerven, T. Matsuura, Phys. Rev. B 359 (1991) 343.
G.F. Giudice, R. Rattazzi, J.D. Wells, hep-th/9811292.
E.A. Mirabelli, M. Perelstein, M.E. Peskin, hep-ph/9811337.
T. Han, J. Lykken, R.-J. Zhang, hep-ph/9811350.
J.L. Hewett, hep-ph/9811356.
E. Dudas, J. Mourad, hep-th/9911019.
T.G. Rizzo, Phys. Rev. D 60 (1999) 115010.
P. Mathews, S. Raychaudhuri, K. Sridhar, hep-ph/9811501.
16
[71]
[72]
[73]
[74]
[75]
[76]
[77]
[78]
[79]
[80]
[81]
[82]
[83]
[84]
[85]
[86]
[87]
[88]
[89]
[90]
K. Cheung, W.-Y. Keung, hep-ph/9903294.

K. Cheung, G. Landsberg, hep-ph/9909218.
H. Davoudiasl, hep-ph/9904425.
P. Mathews, P. Poulose, K. Sridhar, hep-ph/9905395.
H.-W. Yu, L.H. Ford, gr-qc/9907037.
T. Han, D. Rainwater, D. Zeppenfeld, hep-ph/9905423.
K. Agashe, N.G. Deshpande, hep/9902263.
M.L. Graesser, hep-ph/9902310.
D. Atwood, S. Bar-Shalom, A. Soni, hep-ph/9903538; hep-ph/9906400.
D. Bourilkov, hep-ph/9907380.
M. Besancon, hep-ph/9909364.
S. Nussinov, R. Shrock, Phys. Rev. D 59 (1999) 105002.
G. Shiu, R. Shrock, S.H. H. Tye, hep-ph/9904262.
T. Banks, M. Dine, A. Nelson, hep-th/9903019.
E. Halyo, hep-ph/9904432.
A. Donini, S. Rigolin, hep-ph/9901443.
J. Lykken, S. Nandi, hep-ph/9908505.
B. Grzadkowski, J.F. Gunion, hep-ph/9910456.
M. Peskin, talk at ITP Santa Barbara, http://www.itp.ucsb.edu/online/susy_c99/.
J. Polchinski, String Theory, Cambridge Press, 1998.

Golden measurements at a neutrino factory

A. Cervera a,1 , A. Donini b,2 , M.B. Gavela b,3 , J.J. Gomez Cdenas a,4 ,
P. Hernndez c,5 , O. Mena b,6 , S. Rigolin d,7
a Departamento de Fsica Atmica y Nuclear and IFIC, Universidad de Valencia, Valencia, Spain
b Departamento de Fsica Terica, Universidad Autnoma de Madrid, 28049 Madrid, Spain
c Theory Division, CERN, 1211 Geneva 23, Switzerland
d Department of Physics, University of Michigan, Ann Arbor, MI 48105 USA
Received 29 February 2000; accepted 5 April 2000
Abstract
The precision and discovery potential of a neutrino factory based on muon storage rings is studied.
For three-family neutrino oscillations, we analyse how to measure or severely constraint the angle
13 , CP-violation, MSW effects and the sign of the atmospheric mass difference 1m223 . We present a
simple analytical formula for the oscillation probabilities in matter, with all neutrino mass differences
non-vanishing, which clarifies the subtleties involved in disentangling the unknown parameters. The
appearance of wrong-sign muons at three reference baselines is considered: 732 km, 3500 km, and
7332 km. We exploit the dependence of the signal on the neutrino energy, and include as well realistic
background estimations and detection efficiencies. The optimal baseline turns out to be O(3000 km).
Analyses combining the information from different baselines are also presented. 2000 Elsevier
Science B.V. All rights reserved.
PACS: 14.60Pq
Keywords: Neutrino oscillations; CP-violation; Neutrino factory
1. Introduction
The atmospheric [16] plus solar [714] neutrino data point to neutrino oscillations
[1517] and can be easily accommodated in a three-family mixing scenario.
1 anselmo.cervera@cern.ch
2 donini@daniel.ft.uam.es
3 gavela@garuda.ft.uam.es
4 gomez@hal.ific.uv.es
5 pilar.hernandez@cern.ch; On leave from Dept. de Fsica Terica, Universidad de Valencia.
6 mena@delta.ft.uam.es
7 srigolin@umich.edu
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 2 2 1 - 2
18
A. Cervera et al. / Nuclear Physics B 579 (2000) 1755
Let U , with (e , , )T = U (1 , 2 , 3 )T , be the leptonic CabibboKobayashi

Maskawa (CKM) matrix in its most conventional parametrization [18,19]: 8
U U23 U13 U12
1
0
0
c13
0 c23 s23
0
0 s23 c23
s13 ei
0
1
0
s13 ei
c12
0 s12
c13
0
s12
c12
0
0
0
1
(1)
with s12 sin 12 , and similarly for the other sines and cosines. Oscillation experiments
are sensitive to the neutrino mass differences and the four parameters in the mixing matrix
of Eq. (1): three angles and the Dirac CP-odd phase.
The Super-Kamiokande [1,2] data on atmospheric neutrinos are interpreted as oscillations of muon neutrinos into neutrinos that are not e s, with a mass gap that we denote 9 by
1m223 . Roughly speaking, the measured mixing angle 23 is close to maximal and |1m223|
is in the range 103 102 eV2 . The solar neutrino deficit is interpreted either as MSW
(matter enhanced) oscillations [16,17] or as vacuum oscillations (VO) [15] that deplete
the original e s, presumably in favour of s or alternatively into sterile neutrinos. The
corresponding squared mass differences O(105 104 ) eV2 for the large mixing angle MSW solution (LMAMSW), O(106 ) eV2 for the small mixing angle MSW solution
(SMAMSW), or O(1010 ) eV2 for VO are significantly below the range deduced from
atmospheric observations. We identify this mass difference with 1m212 in this parametrization. Its sign is constrained by solar data: while the SMAMSW solution exists only for
positive 1m212 , in the LMAMSW range there is also a small window at negative values
[20].
These oscillation signals will be confirmed and further constrained in ongoing and
planned atmospheric, solar and long baseline reactor experiments [2125], as well as in
future long baseline accelerator neutrino experiments [26,27]. In a few years they will
answer the question of sterile neutrinos contributing or not to present data. The MSW
effect is expected to play a major role in explaining the solar deficit and both solar and
reactor experiments will also clarify whether Nature has chosen the LMAMSW rather
than SMAMSW or VO solutions.
The atmospheric neutrino parameters will be known with better precision as well.
Experimental information relevant for a more precise knowledge of the atmospheric
neutrino fluxes will be available [2830]. Also, projected long baseline accelerator
experiments will improve the precision of |1m223| and 23 . For instance, |1m223| is
expected to be measured at MINOS with an accuracy below 10% if |1m223 | > 3 103 eV2
[31].
Nevertheless, there is a strong case for going further in the fundamental quest of the
neutrino masses and mixing angles, as a necessary step to unravel the fundamental new
scale(s) behind neutrino oscillations. In ten years from now no significant improvement is
expected in the knowledge of:
8 Notice, though, that our convention for the sign of is opposite to the one used in Ref. [19].
9 1m2 m2 m2 throughout the paper.
ij
j
i
19
The angle 13 , which is the key between the atmospheric and solar neutrino realms,
for which the present CHOOZ bound is sin2 13 6 5 102 [32,33].
The sign of 1m223 , which determines whether the three-family neutrino spectrum is
of the hierarchical or degenerate type (i.e., only one heavy state and two almost
degenerate light ones, or the reverse).
Leptonic CP-violation.
The precise study of matter effects in the propagation through the Earth: a modelindependent experimental confirmation of the MSW effect will not be available.
The most sensitive method to study these topics is to measure the transition probabilities
involving e and e , in particular e (e ) ( ). This is precisely the golden
measurement at the neutrino factory [34]. Such a facility is unique in providing high
energy and intense e (e ) beams coming from positive (negative) muons which decay in
the straight sections of a muon storage ring [3537]. Since these beams contain also ( )
(but no ( )!), the transitions of interest can be measured by searching for wrong-sign
muons: negative (positive) muons appearing in a massive detector with good muon charge
identification capabilities.
The first exploratory studies of the use of a neutrino beam with these characteristics [38]
were done in the context of two-family mixing. In this approximation, the wrong-sign
muon signal in the atmospheric range is absent, since the atmospheric oscillation is
. The enormous physics reach of such signals in the context of three-family neutrino
mixing was first realized in [39], where the authors put the emphasis on the measurement
of the angle 13 and CP-violation (see also [40]). The latter may be at reach if the
solar deficit corresponds to the LMAMSW solution [39,41,42]. Recently, it has also
been shown [44] that the precision in the knowledge of the atmospheric parameters 23
and |1m223| can reach the percent level at a neutrino factory, using muon disappearance
measurements. Furthermore, it was pointed out [43,44] that the sign of 1m223 can
also be determined at long baselines, through sizeable matter effects. The importance of
measuring precisely Earth matter effects in a clean and model-independent way, both for
the understanding of the fundamental parameters and for their intrinsic interest has been
stressed in Refs. [4548].
The aim of this paper is to identify the optimal baselines for studying the above itemized
topics. This requires to include in the analysis the maximum information that can be
attained at any fixed baseline. While all previous analyses have been based in energyintegrated quantities, we will take into account the neutrino energy dependence of the
wrong-sign muon signals, together with the information obtained from running in the two
different beam polarities.
We will consider in turn scenarios in which the solar oscillation lies in the SMAMSW
or VO range and in the LMAMSW range. In the latter, the dependence of the oscillation
probabilities on the solar parameters 12 , 1m212 , and on the CP-odd phase, , is sizeable at
terrestrial distances and complicates the measurement of 13 due to the presence of other
unknowns (mainly ). This potential difficulty was first pointed out in [41]. The previous
analysis of the sensitivity to 13 [39] neglected solar parameters and is thus only valid
for the SMAMSW and VO solutions, or the LMAMSW if 13 is large enough. In the
20
present paper a higher statistics is considered, allowing to explore smaller values of 13 ,

and the remark is very pertinent. We will discuss in detail the issue of how to disentangle
13 and , guided by an approximate analytical formula for the oscillation probabilities in
matter including two distinct mass differences. As we will see, the choice of the correct
baseline is essential to solve this problem.
We shall consider the following reference set-up: neutrino beams resulting from the
decay of 2 1020+ s and/or s per year in a straight section of an E = 50 GeV muon
accumulator. A long baseline (LBL) experiment with a 40 kT detector and five years of data
taking for each polarity is considered. Alternatively, the same results could be obtained in
one year of running for the higher intensity option of the machine, providing 1021 useful
+ s and s per year. A realistic detector of magnetized iron [49] will be considered and
detailed estimates of the corresponding expected backgrounds and efficiencies included
in the analysis. Three reference detector distances are discussed: 732 km, 3500 km and
7332 km.
A preliminary version of this work was presented in [5052].
2. Fluxes and charged currents

2.1. The neutrino factory
One of the most encouraging outcomes of the recent neutrino factory workshop at
Lyon [53] was the convergence of the various machine designs existing previously, to
an essentially unified design [54], based on a muon accumulator with either a triangular
or a bow-tie shape. Both geometries permit two straight sections pointing in different
directions, allowing two different baselines.
It was also agreed in Lyon that the beam power on target should not exceed 4 MW. This,
in turn, limits the production of muons to 1021 per year, out of which only 2025% are
useful (that is, decay in the straight sections pointing towards the detectors). Agreement
was also found upon other important parameters, namely the machine dual polarity (i.e.,
the ability to store + and , although not simultaneously) and the maximum realistic
energy to accelerate the stored muons, which was fixed at 50 GeV.
Ultimate sensitivity to the neutrino mixing matrix parameters, in particular to 13 and ,
require a data set as large as possible. The analysis presented in this paper assumes a total
data set of 1021 useful + decays and 1021 useful decays.
2.2. Number of events
In the muon rest-frame, the distribution of muon antineutrinos (neutrinos) and electron
neutrinos (antineutrinos) in the decay e + e (e ) + ( ) is given by:
1
d2 N
=
[f0 (x) P f1 (x) cos ],
(2)
dx d
4
where E denotes the neutrino energy, x = 2E /m and P is the average muon
polarization along the beam directions. is the angle between the neutrino momentum
21
Table 1
Flux functions in the muon rest-frame as in Ref. [55]
, e
e
f0 (x)
f1 (x)
2x 2 (3 2x)
12x 2 (1 x)
2x 2 (1 2x)
12x 2 (1 x)
vector and the muon spin direction and m is the muon mass. The positron (electron) flux
is identical to that for muon neutrinos (antineutrinos), when the electron mass is neglected.
The functions f0 and f1 are given in Table 1 [55].
In the laboratory frame, the neutrino fluxes, boosted along the muon momentum vector,
are given by:
d2 N ,
dy d

4n
E4 y 2 (1 cos ) 3m2 4E2 y(1 cos )
2
6
L m

P m2 4E2 y(1 cos ) ,
2

24n 4 2
d2 Ne ,e
2
=
E
y
(1
cos
)
m
2E
y(1
cos
)
dy d
L2 m6

P m2 2E2 y(1 cos ) .
(3)
q
Here, = 1 m2 /E2 , E is the parent muon energy, y = E /E , n is the number of
useful muons per year obtained from the storage ring and L is the distance to the detector.
is the angle between the beam axis and the direction pointing towards the detector,
assumed to be located in the forward direction of the muon beam. The angular divergence
will be taken as constant, 0.1 mr.
Unlike traditional neutrino beams obtained from and K decays, the fluxes in Eq. (3)
present a leading quadratic dependence on E and E . This is an important feature:
the number of neutrinos produced with a given E < E is independent of E . As a
consequence, the oscillation signal does not decrease with increasing E . In Appendix A
we present our numerical results for the e (e ) and ( ) fluxes.
The charged current neutrino and antineutrino interaction rates can be computed using
the approximate expressions for the neutrinonucleon cross sections with an isoscalar
target,
N 0.67 1042
E
m2 ,
GeV
N 0.34 1042
E
m2 .
GeV
(4)
It follows that the number of charged current (CC) events at a neutrino factory scales
cubically with energy. In Appendix A we also include our numerical results for the rates
of e and production, from a beam.
22
3. Oscillation probabilities
3.1. In vacuum
Atmospheric or terrestrial experiments have an energy range such that 1m2 L/E
1 for the smaller (1m212 ) but not necessarily for the larger (1m223 ) of these mass
gaps. Even then, solar and atmospheric transitions are not (provided 13 6= 0) two separate
two-generation oscillations. For |1m212 | |1m223|, neutrino oscillation probabilities at
terrestrial distances are accurately described by only three parameters, 23 , 1m223 = 1m213
and 13 :

113 L
2
sin2 213 sin2
,
Pe ( e ) = s23
2

113 L
2
,
sin2 213 sin2
Pe ( e ) = c23
2

113 L
4
sin2 223 sin2
,
(5)
P ( ) = c13
2
where
1ij
1m2ij
2E
(6)
Eqs. (5) are a very good approximation when the solar parameters lie in the SMAMSW
or VO range. The present best fit value for 13 is in the range 6 8 [56,57], although it
is compatible with zero within errors. Among the transitions in Eq. (5) the channels e
, have clearly the best sensitivity to a small 13 . Experimentally, the measurement of
e oscillations through the appearance of wrong-sign muons is far superior to that of
e oscillations through detection. In [39], it was shown that the sensitivity to 13
of the former can improve the present limits, which are mainly set by Chooz [32], by at
least two orders of magnitude.
At the neutrino factory, precision measurements for |1m223| and 23 can also be
performed. Measurements of appearance or disappearance may be competitive with
wrong-sign signals for small values of 13 , because of the cosine dependence of the
corresponding probabilities in Eqs. (5). We do not develop this topic further in the present
work, see [44].
In the LMAMSW scenario, for which the effects of 1m212 may be relevant at the
neutrino factory, a good and simple approximation for the e transition probability is
obtained by expanding to second order in the small parameters, 13 , 112 /113 and 112 L:

2
2
2 113 L
2
2
2 112 L
+ c23 sin 212 sin
Pe ( e ) = s23 sin 213 sin
2
2

1
1
L
L
L
1
13
12
13
sin
,
(7)
+ J cos
2
2
2
where here and throughout the paper the upper/lower sign in the formulae refers to
neutrinos/antineutrinos, and
J c13 sin 212 sin 223 sin 213
23
(8)
is the usual combination of mixing angles appearing in the Jarlskog determinant. The first
term in Eq. (7) is quadratic in sin 13 , whereas the leading term in 112 is linear. The latter
may then be significant or even dominant for very small values of 13 [41], when 1m212
and 12 are in the range allowed by the LMAMSW solution.
At short distances, such as 732 km, Eq. (7) can be further approximated by:

112 L 113 L
113 L 2
+ J cos
2
2
2
2

112 L
2
sin2 212
.
+ c23
2

2
sin2 213
Pe ( e ) = s23
(9)
The CP-odd term in the probability (i.e., the one proportional to sin ) has dropped because
it is of higher order in 1ij L. The comparison of the two polarities is then not useful
(except for doubling the statistics ), since the CP-conjugated channels measure the same
probability. Furthermore, all terms in Eq. (9) have the same dependence on the neutrino
energy and the baseline, and consequently is very hard to disentangle them. As a result we
expect a large correlation between the parameters 13 , , 112 . Even though long baseline
(LBL) reactor experiments [58] will provide a measurement of |1m212 | if the LMAMSW
solution is at work, the parameters 13 and will have to be determined simultaneously
at the neutrino factory. It will then be necessary to go to longer baselines where the
energy dependence of the different terms in Eq. (7) differs, and where the comparison
of the neutrino and antineutrino probabilities provides non-trivial information to separate
from 13 .
It is uncertain [20] whether solar experiments will determine the sign of 1m212 if it lies in
the LMAMSW range. If it remains unknown, it implies an ambiguity in the determination
of : to reverse the sign of 1m212 in Eq. (7) is equivalent to replacing by + . Notice,
though, that whether there is CP-violation or not is independent of whether the phase in
any parametrization is or + .
It is possible to construct a measurable CP-odd asymmetry, which in vacuum is
proportional to sin . In Refs. [39] and [42], the authors considered the following
integrated asymmetry (see also [5968]):
{N[ ]/No [e ]}+ {N[+ ]/No [e+ ]}
.
A CP
e =
{N[ ]/No [e ]}+ + {N[+ ]/No [e+ ]}
(10)
The sign of the decaying muons is indicated by a subindex, N[+ ](N[ ]) are the
measured number of wrong-sign muons, and No [e+ ](No [e ]) are the expected number
of e (e ) charged current interactions in the absence of oscillations. The significance of
this asymmetry (i.e., the asymmetry divided by its statistical error) scales with the baseline
and neutrino energy in the following way:

p 123 L
A CP
.
E sin
(11)

2
A CP
24
The best sensitivity to a non-zero CP-odd asymmetry is found at the maximum of the
atmospheric oscillation. At the corresponding distance, however, matter effects are already
important and should be taken into account, as we proceed to discuss in the next subsection.
3.2. In matter
Of all neutrino species, only e and e have charged-current elastic scattering amplitudes
on electrons. This, as is well known, induces effective masses = 2E A, where the
signs refer to e and e and A is the matter parameter,
(12)
A = 2 GF ne ,
with ne the ambient electron number density [16,17].
Matter effects may be important if A is comparable to, or bigger than, the quantity 1ij
for some mass difference and neutrino energy, and if distances are large enough for the
probabilities to be in the non-linear region of the oscillation.
For the Earths crust, with density 2.8 g/cm3 and roughly equal numbers of protons,
neutrons and electrons, A 1013 eV. The typical neutrino energies we are considering
are tens of GeVs. For instance, for E = 30 GeV (the average e energy in the decay of
E = 50 GeV muons) A = 1.1 104 eV2 /GeV 123 . This means that matter effects
will be important at long distances. Notice that at L = 732 km and 3500 km the neutrino
path remains in the Earth crust, whereas for 7332 km the deeper flight path meets a denser
medium and A = 1.5 104 eV2 /GeV.
3.2.1. Neglecting solar parameters: VO or SMAMSW solutions
Consider the case when 112 is negligible compared to 123 , L1 and A. In the
approximation of constant ne , the transition probability in matter governing the appearance
of wrong-sign muons can then be read from [69]:

113 2 2 B L
2
,
(13)
sin2 213
sin
Pe ( e ) ' s23
B
2
where
B
[113 cos 213 A]2 + [113 sin 213 ]2 .
(14)
For A = 0, Eq. (13) reduces to the corresponding vacuum result: the first line in Eq. (5).
At short distances, that is for B L/2 sufficiently small, the sinus in Eq. (13) can be
expanded and

113 L 2
2
sin2 213
,
(15)
Pe ( e ) s23
2
which also coincides with the vacuum behaviour for small 113 L/2, even when A
113 cos 213 . Eq. (15) is a good approximation up to 3000 km [39]. Note that the
leading dependence on 13 is quadratic as in vacuum.
In contrast with the vacuum result, the probability in matter depends on the sign of
1m213 . It follows from Eqs. (13) and (14) that a change on this sign is equivalent to a
25
Fig. 1. The signal-over-noise ratio of the CP-odd asymmetry of Eq. (10) as a function of the distance.
The thick line corresponds to the integrated asymmetry, while the dashed lines correspond to the
asymmetry computed in five energy bins of equal width 1E = 10 GeV. The neutrino mixing
parameters correspond to the SMAMSW solution to the solar anomaly: 1m223 = 2.8 103 eV2 ,
1m212 = 6 106 eV2 , sin2 212 = 6 103 , 23 = 45 , 13 = 13 and = 90 . The muon energy
is E = 50 GeV and the matter parameter A is varied with the distance as in [70].
CP transformation 10 , that is, interchanging the probability of neutrinos and antineutrinos.

Thus matter effects induce by themselves a non-vanishing CP-odd asymmetry, and the
best sensitivity to the sign is achieved when the sensititivity to the asymmetry of Eq. (10)
is maximal. Fig. 1 shows the significance of this asymmetry as a function of the baseline.
The maximum sensitivity to the sign is thus expected at O(7000) km, although it is already
very good at much shorter distances: notice the large number of standard deviations in the y
axis.
It is important to stress that having sizeable matter effects does not necessarily imply
having sensitivity to the sign. For example if A 113 cos 213 , the sensitivity is lost,
even though matter effects are important at large distances. The optimal sensitivity occurs
for A 113 cos 213 , which is an energy dependent condition. In Fig. 1 the asymmetry
resulting from each of five energy bins of width 1E = 10 GeV is also shown: the
asymmetries in different bins peak at different distances. This dependence suggests that
using the information in energy bins can further improve the measurement of the sign
of 1m223 .
3.2.2. With solar parameters: LMAMSW solution
In the LMAMSW solar scenario, the effects of 1m212 are not negligible over
terrestrial distances, given the high intensity of the neutrino factory. The exact oscillation
probabilities in matter when no mass difference is neglected have been derived analytically
10 Recall that in this approximation 1m2 = 1m2 .
23
13
26
in [71]. However, the physical implications of the formulae in [71] are not easily derived.
A convenient and precise approximation is obtained by expanding to second order in the
following small parameters: 13 , 112 /123 , 112 /A and 112 L. The result is (details of the
calculation can be found in Appendix C):

e

113 2 2 B
112 2 2 AL
L
2
2
2
2
sin
sin
+ c23 sin 212
Pe ( e ) = s23 sin 213
e
2
A
2
B

e

AL
B L
113 L
112 113
sin
sin
cos
,
(16)
+ J
e
A B
2
2
2
e |A 113 |. Once again, this expression reduces to the vacuum result, Eq. (7),
where B
in the limit A 0. As already remarked in the previous subsection, a reversal of the sign
of 112 can be simply traded by an shift of in , and we will stick to positive 112 in the
numerical exercises below.
We have numerically compared Eq. (16) with the exact formulae of [71], in the range
1 < 13 < 10 . Consider for instance the average energy E = 30 GeV and the following
set of values: 1m223 = 2.8 103 eV2 , 12 = 22.5 , 23 = 45 and = 90 . For 1m212 =
1 104 eV2 , the difference is < 10% (< 20%) at L 3000 (7000) km. For 1m212 =
1 105 eV2 , the error diminishes to < 2.5% (< 10%) at L 3000 (7000) km. Slightly
better accuracy is obtained for = 0 .
As before, matter effects in Eq. (16) induce an asymmetry between neutrinos and
antineutrinos oscillation probabilities even for vanishing . For this reason the CP-odd
asymmetry, Eq. (10), is not the most transparent observable to determine the optimal
distance for measuring . A better way of addressing the issue is to ask at what distance
the significance of the terms which depend on is maximal.
For 13 > 1 , the dominant contributions in Eq. (16) are:

e
113 2 2 B
L
2
2
sin
,
P1 s23 sin 213
e
2
B

e

AL
B L
112 113
113 L
sin
sin
,
cos cos
P2 J
e
A B
2
2
2

e

AL
B L
112 113
113 L
sin
sin
.
(17)
sin sin
P3 J
e
A B
2
2
2
In Fig. 2 we show the significance of the -dependent terms, P2 and P3 , as function of L.
The significance is defined as the fraction of wrong-sign muons for positive muons in
the beam resulting from a given term, over the statistical error in the measurement of
the total number of wrong-sign muons. Results for several values of the neutrino energy
are depicted.
The term in cos , P2 , is more significant than P3 at short distances. Unfortunately, this
sensitivity to through P2 is fake, because at short distances there is no way to separate P2
from the leading term, P1 : they have similar energy dependence and do not differ in the
two polarities, as illustrated in Fig. 3. In order words, a change in can be compensated by
a change in 13 to keep the total probability unchanged.
27
Fig. 2. (a) Scaling with L of the ratio of the number of wrong-sign muons induced solely by P2
in Eq. (17) to the statistical error in the total number of wrong-sign muons, for = 0 ; (b) the same
for wrong-sign muons induced by P3 , for = 90 . The normalization of the y axis is arbitrary.
The oscillation parameters are: 1m223 = 2.8 103 eV2 , 1m212 = 1 104 eV2 , 12 = 22.5 ,
23 = 45 and 13 = 8 .
Fig. 3. Neutrino energy dependence of the different terms in Eqs. (17): E2 P1 (solid line),
E2 P2 = 0 (dashed line) for and E2 P3 for = 90 (dotted line), at L = 732 km, for
neutrinos (a) and antineutrinos (b). The parameters correspond to the LMAMSW solution:
1m223 = 2.8 103 eV2 , 1m212 = 1 104 eV2 , 12 = 22.5 , 23 = 45 and 13 = 8 .
At 3500 km the situation is very different, though: as Fig. 4 illustrates, the energy
dependence of the three terms is quite different and furthermore P3 which changes
sign with the beam polarity is considerably larger. The comparison of the number of
wrong-sign muons detected running in the two polarities and the binning in energy of the
signal are thus strong analysis tools to disentangle 13 and at baselines around 3500 km.
Notice that this optimal distance is in nice agreement with the previous studies based on
the significance of integrated asymmetries [39,42,72], updated in Fig. 5 for the present
set-up.
The summary of this long discussion is that baselines much larger than 732 km are
needed, for the following reasons:
28
Fig. 4. The same as in Fig. 3 but for L = 3500 km.
Fig. 5. The signal-over-noise ratio of the CP-odd asymmetry of Eq. (10) as a function of the distance,
after subtracting the fake matter induced asymmetry. The thick line corresponds to the integrated
asymmetry, while the dashed lines correspond to the asymmetry computed in five energy bins of
equal width 1E = 10 GeV. The neutrino mixing parameters correspond to the LMAMSW solution
to the solar anomaly: 1m223 = 2.8 103 eV2 , 1m212 = 1 104 eV2 , 12 = 22.5 , 23 = 45 ,
13 = 13 and = 90 . The muon energy is E = 50 GeV and the matter parameter A is varied with
the distance as in [70].
In the SMAMSW or VO scenarios, the sign of 1m223 can only be determined for
distances such that matter effects are sizeable and the CP asymmetries they induce
measurable. This happens at L = O (3000 km) or larger.
In the LMAMSW scenario, there is a strong correlation between 13 and at short
distances. It is necessary to go far so that the terms most sensitive to them show
a different energy dependence, and the signals in the CP-conjugate channels differ
sizeably, allowing the simultaneous measurement of both parameters.
These expectations will be sustained by a detailed energy analysis in the following
sections.
29
4. Detection of wrong sign muons

4.1. A large magnetized calorimeter
For the present study we consider a Large Magnetized Iron Calorimeter such as the one
proposed in [49]. The apparatus, shown in Fig. 6 is a huge cylinder, of 10 m radius and
20 m length. It is made of 6 cm thick iron rods intersped with 2 cm thick scintillators
segmented along their length. Its fiducial mass is 40 kT. A superconducting coil generates
a solenoidal magnetic field of 1 T.
The detector axis is oriented to form a few degrees with the direction of the neutrino
beam. Thus, a neutrino crossing the detector sees a sandwich of iron and scintillator. The
coordinates of the track transverse to the cylinder axis are measured from the location of
the scintillator rods, while the longitudinal coordinate is measured from their segmentation.
As discussed in [49], the performance of the device would be similar to the one expected
for the MINOS detector [73].
Neutrino interactions in such a detector have a clear signature. For example, a
charged current (CC) event is characterized by a muon, which is seen as a penetrating
long track, plus a shower resulting from the interactions of the hadrons in the event. Fitting
the muon track determines its charge and momentum, and a calorimetric measurement
allows the determination of the hadronic momentum vector. In contrast, for a e CC event
the electromagnetic shower due to the prompt electron cannot be disentangled from the
hadronic shower in an event-by-event basis, and therefore, the event looks similar to a
neutral current (NC), which is characterized by having no penetrating track.
A realistic simulation of the response of an iron calorimeter must be able to compute:
(1) whether the primary muon characterizing a CC event was identified or not, (2) whether
non prompt muons arising from the decays of hadrons were missidentified as primary
muons and (3) the muon and hadronic momentum vectors.
Fig. 6. Sketch of the large calorimeter for the neutrino factory.
30
Fig. 7. Relative error in the measurement of the muon momentum as a function of the muon
momentum.
To address the above points we have written a Monte Carlo simulation based in the
GEANT 3 package [74]. The apparatus is simulated with the correct geometry, and
neutrino interactions are generated at random points in the fiducial volume. Then, every
particle produced in the interaction is followed until it decays, exits the detector or
undergoes a nuclear interaction.
When the muon track is very short, it cannot be disentangled from the other tracks in the
hadronic jets. The peak of the hadron shower occurs at about 10 cm from the interaction
vertex and essentially no hadronic activity remains at 100 cm from it. We thus impose the
conservative criteria that, in order to be reconstructed, a muon track must be longer than
100 cm. About 99.2% of all CC events at 50 GeV produce primary muons that satisfy
this condition.
All muons, either primary or arising from the decay of hadrons, are tracked through the
entire volume, and a hit is recorded each time a scintillator rod is crossed. The tracks can
then be fitted to obtain the muon momentum. However, once the distance travelled by the
muon is known, it is more efficient to use a simple smearing, which takes into account
correctly both the effect of detector resolution and the multiple scattering.
As an illustration of the smearing procedure, the relative error in the measurement of
the muon momentum as a function of the muon momentum is shown in Fig. 7. The
resolution decreases rapidly for low momentum muons (for P < 5 GeV, a much better
resolution would be obtained from the measurement of the muon range), and improves
rather smoothly for large momentum (P /P 4% for P 7 GeV, while P /P
3% for P 50 GeV). The fact that P /P is almost constant for large momentum is
due to the dominance of the multiple scattering term in the resolution.
Hadrons, in contrast, are followed until they decay or undergo a nuclear interaction. In
the latter case their momenta are added to the hadronic energy vector. Finally, both the
magnitude and direction of the hadronic vector was smeared to account for the resolution
of the detector. For details we refer to [49].
A large sample of neutrino interactions, corresponding to 107 charged and neutral
currents, 5 106 e charged and neutral currents and the same data samples for the opposite
polarity were analyzed for this study.
31
4.2. Search for wrong-sign muon events

We consider first the ( , e ) neutrino beams originating from a + beam of E =
50 GeV (the dependence of the backgrounds with the energy of the muon beam was
discussed in [49], where it was shown that optimal performance is obtained at the highest
possible energy). The bulk of the events in the detector are charged currents, signaled
by the presence of a positive primary muon in the event, and e neutral currents, which
are events with no primary leptons, and e charged currents, for which we assume that
the primary electron is not identified. On top of those events, one searches for wrong sign
arising from the produced via the oscillation e . Table 2 shows the number
of interactions corresponding to a total of 1021 useful + decays and a 40 kT detector at
our reference baselines. For illustration, we will consider the oscillation parameters for the
signal in the LMAMSW scenario: 1m223 = 4 103 eV2 , 1m212 = 104 eV2 , 13 = 13 ,
12 = 22.5 and 23 = 45 , as in the tables of Appendix B. Notice however that our results
for the fractional background and efficiencies should be rather insensitive to the particular
choice of parameters.
The potential backgrounds to the wrong sign events signaling the presence of
oscillations are:
1. CC events in which the positive muon is not detected, and a secondary negative
muon arising from the decay of , K and D hadrons fakes the signal. The most
dangerous events are those with D , which yield an energetic muon with a
spectrum similar to the signal.
2. e CC events, for which it is assumed that the primary electron is never detected.
Charm production is not relevant for this type of events since the charmed hadrons in
the hadronic jet are predominantly positive. Instead, fake arise from the decay of
negative pions and kaons in the hadronic jet.
3. and e NC events. Fake arise in this case also predominantly from the decay
of negative pions and kaons, since charm production is suppressed with respect to the
case of CC.
At first sight these backgrounds seem impressive. Fortunately, simple kinematical cuts
can suppress them very efficiently. One exploits the fact that for signal events the
candidate is harder and more isolated from the hadronic jet than for background events.
We thus perform a simple analysis based in two variables, namely: the momentum of the
Table 2
Data samples expected in a 40 kT detector for 1021 useful + decays.
e oscillations with parameters as in Appendix B
Baseline (km)
CC
e CC
+ e NC
(signal)
732
3500
7332
3.5 107
1.2 106
1.2 105
5.9 107
2.4 106
5.1 105
3.1 107
1.2 106
2.1 105
1.1 105
1.0 105
3.8 104
32
Fig. 8. Definition of the kinematical variables used in this study.

Table 3
Fractional backgrounds and signal selection efficiency for the
wrong sign muon search with + decays
CC
e CC
+ e NC
(signal)
1.0 107
5.0 107
1.0 106
3 101
Table 4
Events surviving the cuts in a 40 kT detector for 1021 useful +
decays. Oscillation parameters as in Appendix B
Baseline (km)
CC
e CC
+ e NC
(signal)
732
3500
7332
3.5
0.1
< 0.1
30
1.2
< 0.2
31
1.2
< 0.2
3.3 104
3 104
1.3 104
muon, P , and a variable measuring the isolation of the muon, Qt = P sin (see Fig. 8).
To illustrate the rejection power of this analysis, Fig. 9 shows the efficiency for signal
detection and the fractional backgrounds as a function of P and Qt for charged and
neutral currents. Also shown is the signal-to-noise ratio, S/N , defined as the ratio of the
signal selection efficiency and the error in the subtraction of the number of background
events that pass the cuts, Nb . The error is taken to be Gaussian for large Nb ( Nb ) and
Poisson otherwise. Notice that charm production is the dominant background for CC,
while decay dominates the NC backgrounds.
Inspection of Fig. 9 shows that the S/N is rather flat for P > 5 GeV and Qt > 0.5 GeV.
Cutting at P > 7.5 GeV, Qt > 1.0 GeV, maximizes S/N . However, given the flatness of
the S/N ratio, one can vary these values generously with very little difference in the final
results. Table 3 shows the fractional background contamination and the signal efficiency,
while Table 4 shows the number of background and signal events that pass the cuts for
the data sample discussed above. Notice that the residual backgrounds are quite sizeable
at L = 732 km, small at L = 3500 km and negligible at L = 7332 km. It is possible to
33
Fig. 9. Signal and backgrounds for CC and NC events.
optimize the analysis for very long baseline in order to achieve higher efficiency. However,
we have chosen to use the same set of cuts for the three baselines.
A potential source of fake wrong sign muons not discussed here is that due to a
wrong measurement of the charge. In [49] it was estimated that, for E = 50 GeV, this
background could be reduced to a very small level, of the order of 106 or less. We have
not included this source of background in the analysis.
The same exercise has to be repeated when a beam is considered. The resulting
neutrino beams are now and e and the signal events are . Fig. 10 shows the efficiency
for signal detection and the fractional backgrounds, as a function of P and Qt , for
34
Fig. 10. Signal and backgrounds for CC and NC events.
charged and neutral currents. Tables 5, 6 and 7 summarize the results obtained (the cuts
are the same than for the + analysis). Finally, Fig. 11 shows the signal efficiency and the
fractional backgrounds for + s and s, as a function of the neutrino energy.
In summary, our study shows that a large magnetized iron calorimeter allows the
detection, with high efficiency ( 3040%) of the golden-plated wrong-sign muon signal.
The different backgrounds to this signal can be efficiently controlled using simple cuts,
which exploit the different kinematics between signal and background events. The charm
background can be suppressed to circa 107 , taking advantage of the high degree of
35
Table 5
Data samples expected in a 40 kT detector for 1021 useful decays
Baseline (km)
CC
e CC
+ e NC
(signal)
732
3500
7332
6.8 107
2.3 106
2.6 105
3.0 107
1.3 106
3.0 105
3.1 107
1.2 106
2.0 105
5.0 104
1.6 104
0.2 104
Table 6
Fractional backgrounds and signal selection efficiency for the wrong-sign
muon search with decays
CC
e CC
+ e NC
(signal)
1.0 107
5.0 107
2.0 106
4 101
Table 7
Events surviving the cuts in a 40 kT detector for 1021 useful decays
Baseline (km)
CC
e CC
+ e NC
(signal)
732
3500
7332
7
0.2
< 0.1
15
0.7
< 0.2
62
2.4
< 0.4
2 104
5.4 103
8 102
collimation with the hadronic jet of charmed hadrons produced by 50 GeV muons. Instead,
decays of energetic pions or kaons in NC events contaminate the signal at the 106 level.
The optimization of the S/N ratio points to the intermediate distances of O(3500) km as
the optimal baseline.
5. Analysis in energy bins

A conservative estimate for the neutrino energy resolution in a detector of the type
described in the previous section is 1E /E 20%. For a beam of 50 GeV and
the statistics of oscillated neutrinos expected in the range of parameters considered, it is
reasonable to bin the data in five bins of equal width 1E = 10 GeV.
be the total number of wrong-sign muons detected when the factory is run in
Let Ni,p
polarity p = + , , grouped in 5 energy bins specified by i = 1 to 5, and three possible
distances, = 1 (732 km), 2 (3500 km), 3 (7332 km).
In order to simulate a typical experimental situation we generate a set of data ni,p as
follows: for a given value of the oscillation parameters, the expected number of events,
, is computed; taking into account backgrounds and detection efficiencies per bin, b
Ni,p
i,p
36
Fig. 11. Signal efficiency and total fractional backgrounds for the polarities + (up) and (down)
as a function of the neutrino energy.
, as given in Fig. 11, we then perform a Gaussian (or Poisson, depending on the
and i,p
number of events) smearing to mimic the statistical uncertainty. All in all,
ni,p
+ b ) b
smear(Ni,p
i,p
i,p
i,p
i,p
(18)
The data are then fitted to the theoretical expectation as a function of the neutrino
parameters under study, using a 2 minimization,
2
X X ni,p Ni,p
,
(19)
2 =
ni,p
p
i
where ni,p is the error of ni,p (we include no error in the efficiencies). We perform and
compare six different fits using: 12 , 22 , 32 for the three distances, and the combinations
12 + 22 , 22 + 32 , 12 + 22 + 32 to illustrate the gain in case the neutrino factory shoots
37
to more than one location, a natural scenario given the ring configurations under study. For
simplicity, we will consider a fit in at most two parameters at a time. All numerical results
below will be obtained with the exact formulae for the oscillation probabilities.
6. SMAMSW or vacuum solar deficit

For the SMAMSW or VO scenarios, the influence of solar parameters on the neutrino
factory signals will be negligible 11 , and CP-violation out of reach. Besides its capability
to reduce the errors on 23 and |1m223| to 1% [44], the factory would still be a unique
machine to constrain/measure 13 [39] and the sign of 1m223 [43,44].
Consider first 13 . In Fig. 12, we show the exclusion plot at 90% CL, on the 1m223
(in the range allowed by Super-Kamiokande) versus sin2 13 plane, obtained with the
full unbinned statistics and the two polarities. The same results, but including as well
background errors and detection efficiencies are shown in Fig. 13. The statistical treatment
in the presence of backgrounds is done as in [75]. Notice that the sensitivity is better at
L = 3500 km than at 732 km when efficiencies and backgrounds are included. The latter
are responsible for it. The sensitivity at 7332 km is also worse than at 3500 km, due to the
loss in statistics. In conclusion, the sensitivity to 13 can be improved by three-four orders
of magnitude with respect to the present limits. This is consistent with the results of [39]
given the different statistics used.
The second major topic would be to perform the first precise measurements related to
matter effects, in order to determine the sign of 1m223 [44] and the size of the matter
parameter, A ne .
We have studied the determination of the sign of 1m223 , assuming that the absolute value
has by then been measured with a precision of 10%. We have explored the region around
the best fit values of Super-Kamiokande: |1m223| = 2.8 103 eV2 and 23 = 45 . We
perform a 2 analysis on the 1m223, 13 plane, as described in last section. The conclusion
is that, for data generated within the range 13 = 110 and |1m223 | in the range allowed
by Super-Kamiokande, a misidentification of the sign of 1m223 can be excluded at 99% CL
at 3500 km and 7332 km, but not at the shortest distance, 732 km. This conclusion agrees
with the analysis of Ref. [44], which did not include the energy dependence information.
We have further studied how the matter parameter of Eq. (12) and the angle 13 can be
measured simultaneously. Fig. 14 shows the result of a 2 fit as described in Section 5.
Only statistical errors have been included in this figure. The corresponding results
including as well background errors and detection efficiencies are shown in Fig. 15. At 732
km there is no sensitivity to the matter term, as expected. However, already at 3500 km,
A can be measured with a 10% precision. At the largest baseline, the precision in A
improves although at the expense of loosing precision in 13 due to the loss in statistics.
The level of precision discussed here might even be interesting for geophysicists [76,77].
The above conclusions also hold for the vacuum solution to the solar deficit.
11 In practice, for the numerical results of this section, the central values in the SMAMSW range are taken:
1m212 = 6 106 eV2 and sin2 212 = 0.006.
38
Fig. 12. Asymptotic sensitivity to sin2 13 as a function of 1m223 at 90% CL for L = 732 km (dashed
lines), 3500 km (solid lines) and 7332 km (dotted lines), in the SMAMSW solution. Only statistical
errors are included.
Fig. 13. As in Fig. 12, including as well background errors and detection efficiencies.
7. LMAMSW
Assume now the LMAMSW scenario. Fixed values of the atmospheric parameters are
used in this section, 1m223 = 2.8 103 eV2 and maximal mixing, 23 = 45 . A precision
of 1% in these parameters is achievable through muon disappearance measurements at the
neutrino factory [44]. This level of uncertainty is not expected to affect the results of this
section.
Let us start discussing the measurement of the CP phase versus 13 . We have studied
numerically how to disentangle them in the range 110 and 0 6 6 180 .
Consider first the upper solar mass range allowed by the LMAMSW solution: 1m212 =
104 eV2 . Fig. 16 shows the confidence level contours for a simultaneous fit of 13 and ,
for data corresponding to 13 = 8 , = 54 , including only statistical errors in the
39
Fig. 14. 68.5, 90, 99% CL resulting from a simultaneous fit of 13 and A. The parameters used to
generate the data are denoted by a star, while the baseline(s) used in the fit is indicated in each plot.
Only statistical errors included.
Fig. 15. The same as in Fig. 14 but including backgrounds and efficiencies.
40
analysis. Fig. 17 shows the same analysis taking into account the background errors and
detection efficiencies of Fig. 11. The correlation between and 13 is very large at the
shortest baseline 732 km, as argued in Section 3. The phase is then not measurable
and this indetermination induces a rather large error on the angle 13 . However, at the
intermediate baseline of 3500 km the two parameters can be disentangled and measured.
At the largest baseline, the sensitivity to is lost and the precision in 13 becomes worse
due to the smaller statistics. The combination of the results for 3500 km with that for any
one of the other distances improves the fit, but not in a dramatic way. Just one detector
placed at O (3000 km) may be sufficient: a precision of few tenths of degree is attained
for 13 and of a few tens of degrees for .
Similar figures are obtained for data corresponding to smaller values of 13 , as shown
in Fig. 18 for 13 = 2 . The pattern is maintained as well for different values of : see
Fig. 19 for data corresponding to = 0 and 13 = 8 . This last figure also proves that,
if the sign of 1m212 is known by the time the neutrino factory will be operative, = 0
is distinguishable from = 180 with just one baseline. This exemplifies the power of
the analysis of the energy dependence. Recall in any case that a -ambiguity on has no
bearing on the existence of CP-violation, and we will go on considering positive values
of 1m212 .
The sensitivity to CP-violation decreases linearly with 1m212 . At the central value
allowed by the LMAMSW solution, 1m212 = 5 105 eV2 , CP-violation can still be
discovered, as shown in Fig. 20. At the lower value allowed, 1m212 = 1 105 eV2 , the
sensitivity to CP-violation is lost with the experimental set-up considered, as shown in
Fig. 21.
We have quantified what is the minimum value of 1m212 for which a maximal CP-odd
phase, = 90 , can be distinguished at 99% CL from = 0 . The result is shown in Fig. 22:
1m212 > 2 105 eV2 , with very small dependence on 13 , in the range considered.
One word of caution is pertinent: up to now we assumed |1m212| and sin 212 known
by the time the neutrino factory will be operational. Otherwise, the correlation of these
parameters with 13 would be even more problematic than that between and 13 , as
illustrated in Fig. 23 for |1m212|. The error induced on the measurement of 13 by the
present uncertainty in |1m212| is much larger than that stemming from the uncertainty
on . Fortunately, LBL reactor experiments will measure |1m212| and sin 212 if it lies in
the LMAMSW range. Even if the error in these measurements is as large as 50%, the
problem would be much less serious. We have checked that such uncertainty does not
affect our results concerning the sensitivity to , and only induces an error in 13 that can
be read from Fig. 23.
Concerning the measurement of the matter parameter, we have considered simultaneous
fits of 13 and A, for two values of the CP-phase: = 0, /2. The confidence level contours
obtained are very similar to those in Fig. 14 for the SMAMSW solution. This indicates
that there is no dangerous correlation between A and 13 in the presence of sizeable dependent terms, and both parameters can be safely measured at 3500 km. However, it
is important to stress that the simultaneous measurement of the three parameters 13 ,
41
Fig. 16. 68.5, 90, 99% CL contours resulting from a 2 fit of 13 and . The parameters used to
generate the data are depicted by a star and the baseline(s) which is used for the fit indicated in
each plot. Only statistical errors are included.
Fig. 17. The same as Fig. 16 with backgrounds and efficiencies included.
42
Fig. 18. As Fig. 17 for different values of the parameters denoted by the star.
Fig. 19. As Fig. 17 for different values of the parameters denoted by the star.
43
Fig. 20. Fit for 1m212 = 5 105 eV2 including backgrounds errors and detection efficiencies. The
star indicates the parameters used to generate the data.
Fig. 21. As Fig. 20, for 1m212 = 1 105 eV2 .
44
Fig. 22. Lower limit in 1m212 at which a maximal CP phase (90 ) can be distinguished from a
vanishing phase at 99% CL, as a function of 13 at L = 3500 km. Background errors and efficiencies
are included.
Fig. 23. Simultaneous fit to 13 and 1m212 . The range shown in the vertical axis is the presently
allowed LMAMSW range. The star indicates the parameters used to generate the data and the
CP-odd phase is set to zero. Backgrounds and detection efficiencies are included.
45
and A will increase the errors with respect to the two-parameter fits performed here.
In this respect the combination of two baselines: O(3500 km) and O(7332 km) may be
helpful.
8. Summary
The neutrino beams obtained from muon storage rings will be excellent for precision
neutrino physics. The appearance of wrong-sign muons is a powerful neutrino oscillation
signal, which allows to improve considerably our knowledge of the leptonic flavour sector.
Two very important questions are the optimal energy for the decaying muons and the
optimal detection distance(s), in view of the physics goals. The higher the parent muon
energy, the larger the oscillation signals. This fact, together with the requirement of low
backgrounds and good detection efficiencies, lead to consider muon energies as high as
possible within realistic machine designs. Energies of a several tens of GeV are currently
under discussion, assumed here to be E = 50 GeV, for definiteness.
Energy and detection distance are intertwined in the oscillation pattern of neutrinos
propagating in matter. We have derived an analytical approximate formula for the
oscillation probabilities in matter, which helps to understand how the sensitivity to the
most interesting quantities scales with the neutrino energy and distance.
We have shown that an analysis in neutrino energy bins, combined with a comparison of
the signals obtained with the two polarities, allows to disentangle the unknown parameters
at long enough baselines. In particular, for the LMAMSW solution, 13 and can
be simultaneously measured. We have also studied realistic backgrounds and detection
efficiencies. The overall conclusion is that the intermediate baseline of O(3000 km) is
optimal for the physics goals considered in this paper (see Fig. 24 for an artistic view of
possible locations).
Quantitatively, our two parameter fits at 3500 km indicate:
The angle 13 can be measured with a precision of tenths of degrees, down to values
of 13 = 1 . The asymptotic sensitivity to sin2 13 can be improved by three orders of
magnitude or more.
If the solar deficit corresponds to solar parameters in the LMAMSW range, CPviolation may be tackled. The phase can be determined with a precision of tens of
degrees, for the central values allowed for |1m212|, and maximal CP-violation can be
disentangled from no CP-violation at 99% CL for values of |1m212| > 2 105 eV2 .
The sign of the atmospheric mass difference, 1m223 , can be determined at 99% CL,
for 13 within the range 13 = 110 and |1m223 | in the range allowed by SuperKamiokande data.
A model independent confirmation of the MSW effect will be feasible, and the matter
parameter A measured within a 10% precision, or better if combined with the longest
baseline: 7332 km.
In the case of the LMAMSW solution, the combination of the two longest baselines
may be useful if a multiparameter fit becomes necessary.
Fig. 24. Artistic view of possible baselines of O(3000 km) at which a neutrino factory, located at Geneva (Switzerland) or Geneva (Illinois), could shoot. The inner and outer
circles correspond to 2000 and 4000 km, respectively. Note that the circles have not been projected onto the Earth surface.
46
47
Even though non-zero neutrino masses are barely established, the neutrino sector of the
theory can be convincingly argued to herald physics well beyond the standard model. It is
in this perspective that a neutrino factory should be built.
Acknowledgements
We acknowledge useful conversations with: J. Bernabeu, A. de Gouvea, A. De Rjula,
F. Dydak, J. Ellis and C. Quigg. We further thank A. De Rjula for suggestions
and a critical reading of the manuscript. A.D. acknowledges the INFN for financial
support. S.R. acknowledges the European Union for financial support through contract
ERBFMBICT972474. The work of A.D., M.B.G. and S.R. was also partially supported by
CICYT project AEN/97/1678.
Appendix A. Non-oscillated statistics

Table 8 contains the results for the e and fluxes for the parent muon energy E =
50 GeV and for n = 2 1020 useful muons per year, per 5 operational years. The result
for three muon polarizations are shown: P = 0, 0.3 (the natural polarization) and
1. For P = 0 our results agree with [46] when the same number of useful muons is
considered.
For a + ( ) beam of E = 50 GeV, the average neutrino and antineutrino energies
are, for P+ ( ) = 0, hE ( ) i = 35 GeV, and hEe ( e ) i = 30 GeV. For P+ ( ) =
1(+1), we get hE ( ) i = 30 GeV, hEe ( e ) i = 30 GeV.
Table 9 contains the numerical results for the number of CC interaction rates for e and
fluxes in a beam in a 40 kT detector with n = 2 1020 useful muons per year per 5
operational years. These results represent the number of leptons of a given flavour observed
at the detector in case no neutrino oscillation occurs, neglecting detection efficiencies.
Appendix B. Oscillated statistics

As an illustration, we give the oscillated fluxes for three values of the atmospheric mass
difference, 1m223 = 2, 4, 6 103 eV2 . The rest of the leptonic parameters are 1m212 =
104 eV2 , 23 = 45 , 13 = 13 and 12 = 22.5. The matter parameter is taken to be
A = 1.1 104 eV2 /GeV for the baselines of L = 732 km and L = 3500 km, while A =
1.5 104 eV2 /GeV for L = 7332 km. The in principle measurable quantities are the
number of leptons of a given flavour and charge reaching the detector at a given baseline.
Table 10 shows the total number of leptons of different flavours ( e + ,
e e , e + and for a beam) at 732 km. Table 11 shows
the analogous results from + decays. Detection efficiencies are not included. The whole
exercise is repeated for the baselines L = 3500 km and L = 7332 km in Tables 12, 13 and
14, 15, respectively.
48
Table 8
Neutrino and antineutrino fluxes for L = 732, 3500 and 7332 km per m2 per 5 operational years when
2 1020 muons decay in the straight section of the storage ring. This fluxes have been averaged over
an angular divergence of 0.1 mr
E = 50 GeV
L (km)
/1012
e /1012
/1012
e /1012
732
0
0.3
1
123
110
81.5
122
159
244
123
110
81.5
122
159
244
3500
0
0.3
1
5.35
4.83
3.56
5.35
6.95
10.7
5.35
4.83
3.56
5.35
6.95
10.7
7332
0
0.3
1
1.22
1.10
0.81
1.22
1.58
2.43
1.22
1.10
0.81
1.22
1.58
2.43
Table 9
Neutrino and antineutrino charged currents interaction rates for L = 732, 3500 and 7332 km per
40 kT and per 5 operational years when 2 1020 muons decay in the straight section of the storage
ring. These fluxes have been averaged over an angular divergence of 0.1 mr
E = 50 GeV
L (km)
N /105
N e /105
N /105
Ne /105
732
0
0.3
1
692
603
394
300
390
600
352
306
200
590
768
1180
3500
0
0.3
1
30.4
26.4
17.2
13.1
17.1
26.2
15.4
13.4
8.75
25.8
33.6
51.6
7332
0
0.3
1
6.90
6.01
3.92
3.00
3.88
5.98
3.50
3.05
1.99
5.88
7.66
11.8
Appendix C. Perturbative expansion of oscillation probabilities

In this appendix, we describe the perturbative expansion we have performed to obtain
Eq. (16). The problem is to diagonalize the neutrino mass matrix,
A 0 0
0
0
0
(20)
M U 0 112
0 U + 0 0 0,
0 0 0
0
0
113
49
Table 10
Calculated charged currents event rates for beam assuming neutrino oscillations in a 40 kT
detector, for a L = 732 km baseline as a function of E , 1m223 , for different polarizations of the
parent muon. We have considered 1 1021 negative muons decays (2 1020 useful muons/year 5
operational years)
P
1m223
N /105
Ne+ /105
N+ /103
Ne /103
N + /103
N /104
0.002
0.004
0.006
691
684
673
300
299
298
13.8
51.3
110
24.1
91.7
201
14.3
52.9
113
20.6
80.9
178
0.3
0.002
0.004
0.006
601
595
584
390
389
387
18.0
66.7
143
22.0
83.8
184
18.6
68.8
147
19.4
76.1
167
0.002
0.004
0.006
391
387
378
600
598
596
27.7
103
220
18.6
70.8
155
28.6
106
226
16.5
64.2
140
Table 11
Calculated charged currents event rates for + beam assuming neutrino oscillations in a 40 kT
parent muon. We have considered 1 1021 positive muons decays (2 1020 useful muons/year 5
operational years)
P+
1m223
N+ /105
Ne /105
N /103
Ne+ /103
N /103
N + /104
0.002
0.004
0.006
351
348
342
590
588
586
28.9
110
239
11.7
43.6
94.1
28.0
106
232
10.5
41.4
91.5
0.3
0.002
0.004
0.006
305
302
297
767
765
762
37.5
142
311
11.2
41.9
90.1
36.4
138
302
9.86
38.7
85.4
0.002
0.004
0.006
200
196
192
1180
1170
1172
57.8
219
478
9.52
35.2
75.3
56.0
212
464
8.36
32.7
71.6
with U as given in Eq. (1). The exact diagonalization of this matrix has been done in [71].
Since we are interested only in the case in which 1m212 is small, it is adequate to use
perturbation theory to compute corrections to first order in this quantity. This leads to
much simpler analytical formulae.
In the limit 1m212 = 0, the diagonalization of this matrix is very simple [69],
1
0
2 (113 A B ) 0
(0)
U .
(21)
0
0
0
M U
1
0
0 2 (113 A + B )
The matrix of eigenvectors is
50
Table 12
operational years)
P
1m223
N /104
Ne+ /104
N+ /102
Ne /103
N + /102
N /103
0.002
0.004
0.006
282
232
170
130
128
126
65.8
163
227
21.1
81.9
174
75.3
187
259
187
631
1150
0.3
0.002
0.004
0.006
244
198
143
169
166
164
85.5
212
295
17.9
70.5
150
97.9
243
337
176
585
1054
0.002
0.004
0.006
156
120
81.0
260
256
253
132
327
454
15.3
58.7
121
151
373
518
145
463
792
Table 13
operational years)
P+
1m223
N+ /104
Ne /104
N /103
Ne+ /102
N /103
0.002
0.004
0.006
143
118
86.0
254
240
221
25.7
96.4
195
60.4
164
249
22.9
88.1
181
100
348
655
0.3
0.002
0.004
0.006
124
100
72.4
330
312
287
33.4
125
254
62.4
165
246
29.8
115
235
92.7
319
592
0.002
0.004
0.006
79.3
60.8
40.8
507
480
441
51.4
193
390
50.2
124
173
45.8
176
362
77.0
254
450
U U23 (23 )U13 (M ),
N + /103
(22)
where M is defined by:

tan 2M
113 sin 213

.
113 cos 213 A
(23)
M is to be taken in the first (second) quadrant if 113 cos 213 A is positive (negative).
At first order in 112 , the perturbation to Eq. (21) (in the basis of non-perturbated
eigenvectors) is:
51
Table 14
operational years)
P
1m223
N /103
Ne+ /103
N+ /102
Ne /103
N + /102
N /103
0.002
0.004
0.006
527
256
89.1
299
296
292
2.38
16.7
35.8
3.97
28.2
78.8
1.94
15.1
33.5
159
406
523
0.3
0.002
0.004
0.006
451
212
75.1
388
385
380
3.10
21.6
46.6
3.95
27.3
74.1
2.52
19.6
43.6
146
362
452
0.002
0.004
0.006
273
110
42.5
597
592
584
4.77
33.3
71.6
3.89
25.2
63.3
3.88
30.1
67.0
115
257
287
Table 15
operational years)
P+
1m223
N+ /103
Ne /103
N /103
Ne+ /102
N /103
N + /103
0.002
0.004
0.006
265
125
52.8
577
513
399
6.06
38.1
94.8
1.16
11.5
31.2
5.84
37.8
95.0
85.6
225
295
0.3
0.002
0.004
0.006
227
104
47.6
750
667
519
7.88
49.5
123
1.20
11.0
28.5
7.59
49.1
123
78.6
201
255
0.002
0.004
0.006
137
54.8
35.7
1150
1030
798
12.1
76.2
190
11.7
75.6
190
62.4
144
161
0
0
(1)
M
U U 0 112
0
0
0
0 U U .
0
129
10.0
22.3
(24)
The eigenvalues at first order in 112 are:

(1)
(0)
2
112 cos2 M ,
1 = 1 + s12
(0)
2
(1)
2 = 2 + c12 112 ,
2
112 sin2 M .
3 = 3 + s12
(1)
(0)
The corresponding (not normalized) eigenvectors are,
(25)
52

2 1 sin
i
sin 212 cos M 112 s12
12
M e
v1(1) = 1,
,
,
113 + A B
2B

sin 212 cos M 112
sin 212 sin M ei 112
(1)
, 1,
,
v2 =
113 A + B
113 + A + B
2

s12 sin 2M 112 ei sin 212 sin M 112 ei
(1)
,
,1 ,
v3 =
2B
131 + A + B
(26)
where M 13 M .
With these results, it is easy to compute the oscillation probabilities, keeping consistently
terms up to first order in 112 . We obtain:

2
2
2 B L
Pe ( e ) = s23 sin (2M ) sin
2

2 2
2 B L 112
s23 s12 sin(4M ) sin(2M ) sin

2
B

112 L
2
+ sin (2M ) cos(2M ) sin(B L)

2

B L
112
+ sin(212) sin(223) sin(2M ) sin
2
(0)

(0) L
L
sin 1
cos 3
2
2

cos M cos M sin M sin M
(0)
(0)
1
3

B L 1
,
(27)
sin M sin M cos sin
(0)
2

B L
2
sin2 (2M ) sin2
Pe ( e ) = c23
2

B L 112
2 2
s12 sin(4M ) sin(2M ) sin2
c23
2
B

112 L
+ sin2 (2M ) cos(2M ) sin(B L)
2

B L
112
sin(212) sin(223 ) sin(2M ) sin
2
(0)
(0)
3 L
1 L
cos
sin
2
2

sin M sin M
cos M cos M
(0)
(0)
1
3

B L 1
,
sin M sin M cos sin
2
(0)
3
(28)
53

B L
Pe e ( e e ) = 1 sin2 (2M ) sin2
2

B L 112
2
sin(4M ) sin(2M ) sin2
+ s12
2
B

112 L
.
+ sin2 (2M ) cos(2M ) sin(B L)
2
(29)
These formulae are valid for all values of 13 and to first order in 112 . It is rather
straightforward to check that they reduce to the vacuum result for A 0.
In Section 3, we have further considered an expansion in which not only 112 but also
13 are small. We have kept terms up to second order: i.e., O(1212 13 0 ), O(112 13 ) and
O(1012 13 2 ). The latter two can be obtained from Eqs. (27,28) and (29), by performing
an expansion in 13 . On the other hand, the terms of O(1212 13 0 ) are absent in that
approximation, they appear at next order in the expansion. They can be easily obtained,
though, starting directly from the diagonalization of the mass matrix with 13 = 0:
1
0
0
2 (A + 112 C )

0
1
M U23 (23 )U12 M

(A + 112 + C )
0
0
113
0
) U23 (23 ) ,
U12 (M
q
with C 1212 + A2 2A112 cos 212 and
0
sin 2M
112 sin 212

.
C
The expansion of the corresponding probabilities to second order in 112 gives,

112 2 2 AL
2
2
sin
,
Pe ( e ) = c23 sin 212
A
2

112 2 2 AL
2
sin2 212
sin
,
Pe ( e ) = s23
A
2

112 2 2 AL
2
.
sin
Pe e ( e e ) = 1 sin 212
A
2
(30)
(31)
(32)
From the first of these equations we obtain the term in 1212 of Eq. (16).
References
[1] Super-Kamiokande Collaboration, Y. Fukuda et al., Phys. Lett. B 433 (1998) 9; Phys. Lett.
B 436 (1998) 33; Phys. Rev. Lett. 81 (1998) 1562; Phys. Rev. Lett. 82 (1999) 2644; Phys. Lett.
B 467 (1999) 185.
[2] T. Kajita, Nucl. Phys. B (Proc. Suppl.) 77 (1999) 123.
[3] Kamiokande Collaboration, S. Hatakeyama et al., Phys. Rev. Lett. 81 (1998) 2016.
[4] Soudan-2 Collaboration, E. Peterson, Nucl. Phys. B (Proc. Suppl.) 77 (1999) 111.
[5] W.W. Allison, Phys. Lett. B 449 (1999) 137.
54
[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]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
[50]
MACRO Collaboration, F. Ronga et al., Nucl. Phys. B (Proc. Suppl.) 77 (1999) 117.
R. Davis et al., Phys. Rev. Lett. 20 (1968) 1205.
B.T. Cleveland et al., Astrophys. J. 496 (1998) 505.
K. Lande et al., Nucl. Phys. (Proc. Suppl.) 77 (1999) 13.
T. Kirsten, Nucl. Phys. (Proc. Suppl.) 77 (1999) 26.
Kamiokande Collaboration, Y. Fukuda et al., Phys. Rev. Lett. 77 (1996) 1683.
SAGE Collaboration, D.N. Abdurashistov et al., Nucl. Phys. (Proc. Suppl.) 77 (1999) 20.
Super-Kamiokande Collaboration, Y. Fukuda et al., Phys. Rev. Lett. 82 (1999) 1810.
M. Smy et al., hep-ex/9903034.
B. Pontecorvo, J. Expt. Theor. Phys. 33 (1957) 549; J. Expt. Theor. Phys. 34 (1958) 247; Sov.
Phys. JETP 26 (1968) 984.
L. Wolfenstein, Phys. Rev. D 17 (1978) 2369; Phys. Rev. D 20 (1979) 2634.
S.P. Mikheyev, A.Yu. Smirnov, Sov. J. Nucl. Phys. 42 (1986) 913.
Z. Maki, M. Nakagawa, S. Sakata, Prog. Theor. Phys. 28 (1962) 970.
Particle Data Book, Eur. Phys. J. C 3 (1998) 1.
A. de Gouvea, A. Friedland, H. Murayama, hep-ph/0002064.
Super-Kamiokande home-page, http://www-sk.icrr.u-tokyo.ac.jp/doc/sk.
SNO home-page, http://snodaq.phy.queensu.ca/SNO/sno.html.
BOREXINO home-page, http://almime.mi.infn.it.
HERON home-page, http://www.physics.brown.edu/research/heron.
HELLAZ home-page, http://sg1.hep.fsu.edu/hellaz.
K. Nishikawa, Nucl. Phys. (Proc. Suppl.) 77 (1999) 198.
B.C. Barish, Nucl. Phys. (Proc. Suppl.) 70 (1999) 227.
M.G. Catanesi et al., Proposal to study hadron production for the neutrino factory and for the
atmospheric neutrino flux, CERN-SPC-99-35, 1999.
AMS Collaboration, J. Alcaraz et al., Phys. Lett. B 461 (1999) 387.
S. Ahlen et al., Nucl. Instrum. Meth. A 350 (1994) 351.
D.A. Petyt, A study of parameter measurement in a long-baseline neutrino oscillation
experiment, Thesis submitted to Univ. of Oxford, England, 1998.
M. Apollonio et al., hep-ex/9907037.
M.C. Gonzalez-Garca, C. Pea-Garay, hep-ph/0001129.
S. Geer, Phys. Rev. D 57 (1998) 6989.
C.M. Ankenbrandt et al., Muon Collider Collaboration, Phys. Rev. ST Accel. Beams 2 (1999)
081001.
B. Autin et al., CERN-SPSC/98-30, SPSC/M 617 (October 1998).
S. Geer, C. Johnstone, D. Neuffer, FERMILAB-PUB-99-121.
S. Geer, R. Raja, Workshop on Physics at the first muon collider and at the front end of a muon
collider, AIP Conf. Proc. 435 (1998).
A. de Rjula, M.B. Gavela, P. Hernndez, Nucl. Phys. B 547 (1999) 21.
B. Autin, A. Blondel, J. Ellis, Prospective study of muon storage rings, CERN 99-02, ECFA
99-197.
K. Dick, M. Freund, M. Lindner, A. Romanino, Nucl. Phys. B 562 (1999) 29.
A. Donini, M.B. Gavela, P. Hernndez, S. Rigolin, hep-ph/9909254.
J. Ellis, hep-ph/9911440.
V. Barger, S. Geer, R. Raja, K. Whisnant, hep-ph/9911524.
M. Campanelli, A. Bueno, A. Rubbia, hep-ph/9905240.
V. Barger, S. Geer, K. Whisnant, Phys. Rev. D 61 (2000) 053004.
I. Mocioiu, R. Shrock, hep-ph/9910554.
M. Freund, M. Lindner, S.T. Petcov, A. Romanino, hep-ph/9912457.
A. Cervera, F. Dydak, J. Gmez-Cadenas, Nufact 99 Workshop, July 5-9th, Lyon.
ECFA plenary meeting, December 3 (1999).
[51]
[52]
[53]
[54]
[55]
[56]
[57]
[58]
[59]
[60]
[61]
[62]
[63]
[64]
[65]
[66]
[67]
[68]
[69]
[70]
[71]
[72]
[73]
[74]
[75]
[76]
[77]
55
CERN working group on the neutrino factory, December 6 (1999).

CERN plenary meeting on the muon collider, January 19 (2000).
Nufact 99 Workshop, July 5-9th, Lyon; http://lyoninfo.in2p3.fr/nufact99/.
The Neutrino Factory, Muon Collider Collaboration, K.T. McDonald et al., physics/9911009.
T.K. Gaisser, Cosmic Rays and Particle Physics, Cambridge University Press, 1990.
G.L. Fogli, E. Lisi, A. Marrone, G. Scioscia, Phys. Rev D 59 (1999) 033001, hep-ph/9904465.
A. de Rjula, M.B. Gavela, P. Hernndez, hep-ph/0001124.
S. Schonert, Nucl. Phys. (Proc. Suppl.) 70 (1999) 195.
J. Arafune, M. Koike, J. Sato, Phys. Rev. D 56 (1997) 3093, Phys. Rev. (Erratum) D 60 (1999)
119905.
M. Tanimoto, Prog. Theor. Phys. 97 (1997) 9091; Phys. Lett B 435 (1998) 373; Phys. Lett.
B 462 (1999) 115.
H. Minakata, H. Nunokawa, Phys. Lett. B 413 (1997) 369; Phys. Rev. D 57 (1998) 4403.
S.M. Bilenky, C. Giunti, W. Grimus, hep-ph/9705300; Phys. Rev. D 58 (1998) 033001.
K.R. Schubert, hep-ph/9902215.
J. Bernabu, hep-ph/9904474.
H. Fritzsch, Z.-Z. Xiang, hep-ph/9909304.
A. Romanino, hep-ph/9909425.
M. Koike, J. Sato, hep-ph/9909469.
J. Sato, 9910442.
O. Yasuda, hep-ph/9809205, TMUP-HEL-9810.
R. Gandhi, C. Quigg, M. Hall Reno, I. Sarcevic, Astropart. Phys. 5 (1996) 81.
H.W. Zaglauer, K.H. Schwarzer, Z. Phys. C 40 (1988) 273.
A. Donini, M.B. Gavela, P. Hernndez, S. Rigolin, hep-ph/9910516.
P-875, A long-baseline neutrino oscillation experiment at Fermilab, NuMI-L-63 (1995);
http://www.hep.anl.gov/NDK/Hypertext/numi.html.
Geant 3.21 CERN Program Library, Long write up W5013.
G.J. Feldman, R.D. Cousin, Phys. Rev. D 57 (1998) 3873.
A.M. Dziewonski, D.L. Anderson, Phys. Earth Planet. Int. 25 (1981) 297.
A.M. Dziewonski, in: D.E. James (Ed.), The Encyclopedia of Solid Earth Geophysics, Van
NostrandReinhold, New York, 1989, p. 331.

Computation of quark mass anomalous dimension

at O(1/Nf2) in quantum chromodynamics
M. Ciuchini a , S.. Derkachov b,c,1 , J.A. Gracey a,2 , A.N. Manashov d,e,3
a Theoretical Physics Division, Department of Mathematical Sciences, University of Liverpool,
Liverpool, L69 7ZF, United Kingdom

b Institut fr Theoretische Physik, Universitt Leipzig, Augustusplatz 10, D-04109 Leipzig, Germany
c Department of Mathematics, St. Petersburg Technology Institute, Sankt Petersburg, Russia
d Institut fr Theoretische Physik, Universitt Regensburg, D-93040 Regensburg, Germany
e Department of Theoretical Physics, State University of St. Petersburg, Sankt Petersburg, 198904 Russia
Received 10 December 1999; accepted 29 March 2000
Abstract
We present the formalism to calculate d-dimensional critical exponents in QCD in the large Nf
expansion where Nf is the number of quark flavours. It relies in part on demonstrating that at the
d-dimensional fixed point of QCD the critical theory is equivalent to a non-abelian version of the
Thirring model. We describe the techniques used to compute critical two- and three-loop Feynman
diagrams and as an application determine the quark wave function, , and mass renormalization
critical exponents at O(1/Nf2 ) in d dimensions. Their values when expressed in relation to fourdimensional perturbation theory are in exact agreement with the known four-loop MS results.
Moreover, new coefficients in these renormalization group functions are determined to six-loops
and O(1/Nf2 ). The computation of the exponents in the Schwinger Dyson approach is also provided
and an expression for in arbitrary covariant gauge is given. 2000 Elsevier Science B.V. All rights
reserved.
PACS: 11.10.Gh; 11.15.Pg; 12.38.-t; 11.10.Kk
Keywords: Large Nf method; Renormalization; Quark mass anomalous dimension; Perturbation theory
1. Introduction
Recent developments in the area of multiloop calculations in quantum chromodynamics
(QCD) have included the determination of the MS four-loop -function [1], and the
four-loop quark mass anomalous dimension [2,3]. These calculations involved current
1 sergey.derkachov@itp.uni-leipzig.de
2 jag@amtp.liv.ac.uk
3 manashov@heps.phys.spbu.ru
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 2 0 9 - 1
M. Ciuchini et al. / Nuclear Physics B 579 (2000) 56100
57
state of the art analytic and symbolic manipulation techniques in the evaluation of the
order of 10, 000 Feynman diagrams. Such high-order calculation are necessary to refine
our understanding, for example, of the running of the quark masses [2,3]. Indeed such
four-loop calculations have built on the early lower-order work of several decades of
[410] for the QCD -function and [1114] for the quark mass anomalous dimension.
Whilst we believe the results of [1] and [2,3] will not be superseded by a five-loop
calculation for quite some time, it is possible to probe the higher-order structure of the
QCD renormalization group (RG) functions using techniques alternative to conventional
perturbation theory. One such approach is the large Nf expansion where Nf is the number
of quark flavours. In this method the Feynman diagrams contributing to the determination
of an RG function are reordered according to the number of quark loops and evaluated by
considering at leading order those graphs which are simply a chain of bubbles. The next
order is represented by those bubble chains which have either one vertex or self energy
correction and so on. Clearly this large Nf reordering will cover the information already
contained in explicit perturbative loop calculations but will reveal new information beyond
the current orders. For simple scalar field theories such as the O(N) 4 theory or the O(N)
nonlinear model, the large N technique has been developed to determine information
on the RG functions at O(1/N 3 ) [15,16]. That impressive programme of Vasilev et al.
exploited ideas from critical renormalization group theory and determined rather than
the RG functions themselves, the related critical exponents. These correspond to the RG
functions evaluated at the d-dimensional fixed point of the -function which in O(N) 4
theory is the WilsonFisher fixed point. The techniques developed in the early work on
scalar theories [1517] have been applied to QCD at leading order in 1/Nf in [18,19].
Further, information on the anomalous dimension of the twist-2 operators fundamental
to the operator product expansion used in deep inelastic scattering has been produced as
a function of the operator moment [20,21]. Although it is important that these calculations
are extended to next order in 1/Nf , it turns out that there are several issues which need to
be addressed. First, no rigorous proof exists of the critical equivalence of QCD and the so
called non-abelian Thirring model (NATM) which underlies all the large Nf computations.
The original observation of the connection between the models in [22] was effectively at
O(1/Nf ). Second, the analytic regularization commonly used in 1/N calculations breaks
the gauge invariance of the theory. As far as we are aware all other regularizations spoil
the masslessness of the propagators and this therefore makes higher-order calculations
virtually impossible. We note that dimensional regularization, which is ordinarily used
in conventional perturbation theory, is not applicable in large Nf work since the theory
remains divergent in arbitrary dimensions. This second obstacle is much more serious and
might seem to destroy the possibility of developing a sensible 1/Nf scheme. The resolution
of the issues of the proof of the critical equivalence of QCD and the NATM as well as
demonstrating that the critical exponents evaluated in the latter model using non-gauge
invariant regularizations do in fact match those of QCD are of extreme importance and
represent the central results of this paper. By way of application and in order to support the
scheme which will be developed in practical calculations we examine the quark and mass
anomalous dimensions and determine them at new order in the large Nf expansion which
58
is O(1/Nf2 ). A preliminary version of our results was given in [23].

There are various motivations for examining these two RG functions. First, as in conventional perturbation theory one always needs to deduce the wave-function renormalization
of the fundamental fields of a theory before considering the renormalization of the other
parameters and operators in a Lagrangian. Likewise in the large Nf approach the evaluation of the wave function exponent, , needs to be determined first. Indeed in the context
of the phenomenological application of the large Nf technique the programme of evaluating anomalous dimensions of the twist-2 operators of deep inelastic scattering at O(1/Nf2 )
cannot proceed prior to the determination of at the same order. Second, although is
fundamental it depends on the covariant gauge parameter introduced in the gauge fixing
of the NATM Lagrangian. Consequently, it is not as fundamental a quantity as a gauge
independent exponent. Therefore, to further understand the large Nf method in QCD at
O(1/Nf2 ) we have undertaken to consider the gauge independent quark mass anomalous
dimension. Indeed as part of our calculations we address the issue of the choice of gauge in
the large Nf method. Whilst it may seem that the evaluation of two exponents at O(1/Nf2 )
is a formidable task, it turns out that the computation of the quark mass dimension does not
represent a significant amount of additional calculation as there is, in fact, a close relation
between the fermionic graphs for the wave-function and mass dimension which we will
establish.
It is worth noting that our detailed calculation relies on the novel method of [24,25] for
the computation of critical exponents in 1/N expansion. In comparison with the standard
technique [15] based on the direct solution of the SchwingerDyson (SD) equations the
latter approach allows us to express the critical exponents through the corresponding
renormalization constants whose explicit form is very similar to those computed using
dimensional regularization. The immediate advantage is that it allows one to use techniques
similar to the infrared rearrangement of conventional high order perturbative calculations to
evaluate the three-loop critical-point Feynman diagrams needed for the present calculation.
This approach, which was originally developed to analyse the O(N) nonlinear model
appears to be universal, and can be easily extended to the case we will consider. Together
with some calculational shortcuts it makes the evaluation of almost all the relevant graphs
quite straightforward and simplifies computations significantly. In addition we will also
present the SchwingerDyson formalism for the determination of at O(1/Nf2 ) in QCD.
This builds naturally on the earlier large Nf calculations in quantum electrodynamics
(QED) [2628]. In those papers the original technique of [15] in solving the SD equations
was followed and in extending the formalism to QCD here will provide an interested reader
with a unique opportunity to compare both approaches. Finally, it is worth stressing that the
consistency of our final results for the quark and mass anomalous dimensions at O(1/Nf2 )
with the corresponding four-loop perturbative results of [2,3,1114] provides a nontrivial
check on the validity of either approach.
The paper is organised as follows. Section 2 is devoted to introducing background
formalism and providing the proof of the critical equivalence of QCD and NATM. In
Section 3 we develop the large Nf method explicitly in the context of QCD and the NATM.
The details of the evaluation of the three-loop diagrams to determine the quark anomalous
59
dimension are described at length in Section 4. Section 5 is devoted to the development

of the SchwingerDyson formalism to determine the quark anomalous dimension. The
computation of the extra three-loop graphs needed to determine the mass anomalous
dimension is given in Section 6. The final results for the quark anomalous dimension and
mass anomalous dimension are collected together in Section 7 where we also derive new
information on the coefficients of the respective RG functions. Several appendices contain
results which were fundamental to our calculations.
2. Background
The QCD Lagrangian in (d = 4 2)-dimensional Euclidean space reads:
a
1 a a
1
/ iI + 2 F
F
+
( A)2 + ca D c ,
L = iID
2
4g
2g
(2.1)
where iI is the quark field belonging to the fundamental representation of the colour
group, 1 6 I 6 Nf , Aa is the gluon field, ca and ca are the ghost fields in the adjoint
representation of the colour group, is the covariant gauge parameter and g is the coupling
a and the covariant derivative are defined by
constant. The field strength tensor F

a
= Aa Aa + f abc Ab Ac and D = iAa T a ,
F
where T a are the colour group generators in the corresponding representation and f abc
are the structure constants with [T a , T b ] = if abc T c . To ensure the coupling constant
g is dimensionless below four dimensions we rescale it in the standard way by setting
g M g, where the parameter M has dimensions of mass. The partition functions of
QCD are defined as
Z

(2.2)
O1 (x1 ) . . . On (xn ) = Z 1 D8 O1 (x1 ) . . . On (xn ) exp{S},
, c,
where 8 {A, ,
c} is the set of fundamental fields, Oi (xi ) represent a basic field
or a composite operator and Z is a normalizing factor. As usual, the divergences arising
in the calculation of (2.2) are removed by the renormalization procedure at each order of
perturbation theory. Namely, provided that the averaging in (2.2) is carried out with the
renormalized action all correlators of the fields taken at different spacetime points will be
finite. The renormalized action SR (8, e) (e {g, }) has the form of the action (2.1) with
the fields and parameters being replaced by the bare ones:
SR (8, e) = S(80 , e0 ),
80 = Z8 8,
e0 = Ze e.
Here Z8 = {ZA , Z , Zc } and Ze = {Zg , Z } are the renormalization constants. However

this procedure does not guarantee the finiteness of the Green functions with operator
insertions, which contain additional divergences. To remove these extra divergences one
should renormalize the composite operators as well. In general, the renormalized operator
reads
X
Zik Ok ,
(2.3)
[Oi ]R =
k
60
where the operators Ok have canonical dimensions equal to or less than those of the original
operator Oi , and Zik is the mixing matrix of renormalization constants. An operator is
called multiplicatively renormalized if the matrix Zik is diagonal, giving [Oi ]R = Zi Oi
where there is no summation over i.
The renormalization group equation (RGE) for the one-particle irreducible n-point
Green function with the insertion of k multiplicatively renormalizable composite operators
reads
!
k
X
Oi 0(x1 , . . . , xn+k , M, g, ) = 0,
(2.4)
MM + g g + n8 8 +
i=1
where we use the shorthand notation n8 8 = nA A + n + nc c . The RGE functions

g , , 8 and Oi which are the respective beta functions and the anomalous dimensions
of the fields and composite operators, are defined as follows:
d
d
g(M),
= M
(M),
dM
dM
d
d
ln Z8 ,
ln ZOi .
Oi = M
(2.5)
8 = M
dM
dM
Formally, the parameters g and enter (2.4) on the same footing and should be considered
as independent charges. However, the gauge fixing parameter is introduced into the
theory by hand and cannot enter an expression for a physical quantity. Thus for proper
(gauge invariant) objects the term drops out of (2.4), which then takes the form of an
RGE of a single charge theory.
Our subsequent analysis relies heavily on the existence of a nontrivial infrared (IR),
stable fixed point g of the d-dimensional -function, g (g ) = 0, for large values of Nf .
The -function has been calculated in MS using dimensional regularization up to O(a 6 )
terms, where a = (g/2)2 = s / , in [410]. We only record the first few terms here,
while the full four-loop result can be found in [1]:

2
11
a (a) = (d 4)a + TF Nf CA a 2
3
6

1
5
17
+ CF TF Nf + CA TF Nf CA2 a 3
2
6
12

11
79
1
205
CF TF2 Nf2 +
CA TF2 Nf2 + CF2 TF Nf
CF CA TF Nf
72
432
16
288

1415 2
2857 3 4
CA TF Nf +
CA a + O a 5 ,
(2.6)
864
1728
g = M
from which it follows that

3
1
1
2
3
+

[27C
+
45C
]
+
O

33C
+
O
.
a =
A
F
A
TF Nf
4TF2 Nf2
Nf3
The Casimirs for a general classical Lie group are defined by

T a T a = CF I,
f acd f bcd = CA ab .
Tr T a T b = TF ab ,
(2.7)
(2.8)
61
It immediately follows from (2.4) that the Green functions of gauge invariant operators
are scale invariant at the critical-point g . In other words,
G(xi ) = Di G(xi ),
where Di is the scaling dimension of the corresponding Green function. Moreover, due
to the IR nature of the fixed point, this index determines the power of the leading term
of the IR asymptotic behaviour of the Green functions (pi 0) near the critical-points
and this is stable against the perturbation of the action by IR irrelevant operators [29]. On
the other hand, Green functions of gauge dependent objects such as the propagators of the
basic fields which will in general depend on are not invariant under scale transformations.
Although one may restrict attention from the outset to gauge-independent quantities, since
they have physical meaning, it is also possible and convenient to choose so that all Green
functions are scale invariant. Evidently, this is equivalent to the condition (g , ) = 0.
Since
(g, ) = 2( + A + g /g),
(2.9)
one concludes that the equation (g , ) = 0 has two solutions. One is = 0 whilst the
other is A (g , ) = . Bearing in mind that our main aim is the development of the
1/Nf expansion we choose the first solution, = 0, since the latter gives Nf , which
leads to problems in the construction of the 1/Nf scheme. The origin of the above two
solutions for becomes more transparent if one tries to write down the most general form
of the gluon propagator satisfying the requirements of both scale and gauge invariance.
Indeed, scale invariance yields:
G (p) =

M 2
k
+ BP
AP
,
2
(p )
k
(2.10)
and P
where P
are the transverse and longitudinal projectors, respectively, and both A
and B are constants. As is well known [29], radiative corrections do not contribute to the
k
k
longitudinal part of gluon propagator. Hence, G = g 2 M 2 P p2 . This implies that
if 6= 1 then must vanish, = 0. On the other hand, for 6= 0 one must have = 1
which is easy to check is equivalent to A = corresponding to the canonical dimension
of the field. Earlier work concerning the relation of scaling and conformal symmetry in the
context of gauge theories has been given in [30,31]. We also note that in an abelian gauge
theory the condition A = follows directly from the Ward identities that implies the
scale invariance of all Green functions at the critical-point g = g in any gauge [28].
As is well known from the theory of the critical phenomena [29] physical systems which
look quite different at the microscopic level may exhibit the same behaviour at the phase
transition point. An example of this universality is the fixed point relation between the
Heisenberg ferromagnet and d-dimensional 4 field theory. In what follows we construct
the theory belonging to the same universality class as large Nf QCD but which has
a simpler structure. To this end we develop the 1/Nf expansion correlators of the type
given in (2.2) and analyse their behaviour in the IR region. As usual, the first step is to
integrate over the fermion fields in the functional integral to obtain the effective action for
the gluon field:
62

M 2 a 2 M 2
2
SAeff Nf tr ln / iA
/aT a +
+
(A)
F
4g 2
2 g 2

a
+ ca D c ,
(2.11)
where bearing in mind that g2 1/Nf we have set g 2 = g 2 /Nf . Assuming that Nf is
a large parameter, one can evaluate the integral with action (2.11) by the saddle point
method that generates the systematic expansion for the correlators.
We can now examine which terms in (2.11) contribute to the leading IR asymptotics of
the correlators. We start our analysis with the gluon propagator in the Landau gauge, = 0,
which in leading order in 1/Nf is

d/21 M 2 2 1
+
p
P
G (p) = Nf1 a p2
2g 2

2 1

1
p
1
2 1d/2
= Nf a p
P
.
(2.12)
1+
2
2a g M 2
Since 2 = 4 d > 0 the contribution to the propagator from the second term in the
a )2 , is suppressed by (p/M)2 in the IR region as compared with
brackets of (2.11), (F
the contribution form the fermion loop which is the first term in the brackets. Thus the
asymptotic form of the gluon propagator in the IR limit is fully determined by the fermion
loop contribution:
1d/2
IR a
P .
(2.13)
p2
G (p) '
Nf
Careful analysis shows that diagrams with triple and quartic gluon vertices do not
contribute to the leading IR asymptotics of the correlators either. To see this we note
a )2 in the action (2.11) may be considered as an ultraviolet
at first that the term (F
regulator with M playing the role of a cutoff. Indeed, due to this term the gluon propagator
decreases as 1/p2 as p , which makes all diagrams convergent. We note that the
appearance of divergences of the type A2 is prohibited by gauge invariance. Next, if we
let G denote an arbitrary diagram, then for rescaling of the variables in the momentum
integrals, li Mli0 , we find that up to some prefactor M k , the parameter M enters
the integrand together with the external momenta only. In other words, G(qext , M) =
M k G(qext/M). Therefore the limit qext 0 corresponds to the limit M , which
can be regarded as the removal of the regularization. Further, since the most IR singular
contributions we are interested in come from the integration over the region where li0 0,
one can replace the gluon propagator by its asymptotic form (2.13). Then it can be easily
checked by simple dimensional analysis that the leading IR singularities come from the
diagrams without gluon self-interaction vertices. On the formal level this follows from the
a )2 is accompanied by the factor M 2 and vanishes in the M limit.
fact that term (F
So one concludes that this term does not influence the critical properties of the theory and
according to the general scheme [29] should be excluded from the action. Therefore we
obtain the theory defined by the action

N
/ a T a + f /2A 2 + ca ca + f abc ca Ab cc , (2.14)
SNATM = / iA
2
63
which in the Landau gauge has the same critical behaviour as QCD to all orders in the
1/Nf expansion. Of course, for gauge-independent quantities it is true in any gauge. We
have adapted the usual form of the gauge fixing condition to ensure that the transverse
and longitudinal parts of the gluon propagator have the same momentum dependence. It
is interesting to note that in order to arrive at (2.14) one can start from the theory with
iA
/ a T a ) with
manifest gauge invariance which is determined by the action S = (/
ghost and gauge fixing terms in turn arising from the application of the FaddeevPopov
procedure to the functional integral.
Power counting shows that the theory (2.14) is renormalizable within the 1/Nf
expansion and the renormalized action has the form:
/aT a +
SR = Z1 / iZ2 A
+ Z4f abc c Ab cc ,
Nf /2 2

A + Z3 ca c
2
(2.15)
where we assume, of course, that a gauge-invariant regularization is used. Due to the

SlavnovTaylor identities the renormalization constants Zi are related by
Z2 Z11 = Z4 Z31 .
(2.16)
This equation was used in the exponent formulation to determine the ghost anomalous
dimension at O(1/Nf ) in the Landau gauge in [18]. As was proved above in the Landau
gauge, the critical properties of QCD and this new theory which we shall refer to as the
non-abelian Thirring model (NATM) are identical. Therefore one can use the NATM to
deduce the QCD RG functions. This equivalence at leading order in 1/Nf was noted
in [22] and used to deduce various exponents at O(1/Nf ) in [18,20,21,2628,32]. The
extension of these calculations to O(1/Nf2 ) requires special care. The main one is the
necessity of using a gauge-invariant regularization which was not crucial at O(1/Nf ). The
conventional dimensional regularization is not applicable here, since the gluon propagator
behaves as (p2 )1d/2 and the theory remains logarithmically divergent in any dimension d.
To our knowledge most other invariant regularizations such as higher derivatives spoil
the masslessness of the propagators, which makes higher-order calculations virtually
impossible. Usually in 1/N calculations the analytical regularization of [15] is used.
However, this breaks gauge invariance.
We now consider how to reconcile gauge invariance with the calculational advantage
of using massless propagators. First, we break the gauge invariance of (2.14) from the
beginning by introducing a new coupling, , for the ghost-gluon vertex of (2.14). For = 1
we recover the original model but the theory remains renormalizable for arbitrary as well.
The only effect will be that the identity (2.16) will no longer hold. The bare coupling 0 is
related to the renormalized one, , by
0 = Z = Z4 Z1 Z21 Z31 ,
(2.17)
where the Zi will now also depend on . Let us suppose that we used an invariant method
such as regularization by higher derivatives [33] to regularize this extended theory. Then
it immediately follows from (2.16) and (2.17) that the equality = 1 for the renormalized
64
coupling implies that 0 = 1 as well. Therefore we conclude that = 1 is a fixed point,

(1) = 0. The existence of this fixed point is the key point and it does not depend on
the regularization used. So using any other regularization can only change the position
of the fixed point with in general = 1 + O(1/Nf ). What is important, though, is that
the anomalous dimensions calculated at the critical-point ( ) are scheme independent
and, hence, coincide with the anomalous dimensions deduced in the original model (2.14).
Therefore one can use the regularization which is most convenient from the computational
point of view. Moreover, since we do not need to consider diagrams with external ghost
legs, then the only diagrams depending on are those with a ghost loop. As is evident from
counting powers of 1/Nf these are themselves O(1/Nf2 ). So at this order it is sufficient to
set = 1.
3. Methods
As was shown above the RG functions of QCD at the critical-points can be deduced
from the more simple NATM. The use of the -regularization [15,17] allows us to
retain massless propagators which means that the calculation of the higher-order Feynman
integrals can be achieved. However, the price to be paid for this is the loss of the property
of multiplicative renormalizability. Indeed, formally, the regularized action has the form:
i A
/aT a +
S = /
2
Nf
()/2 Aa + ca ca
2
2 M
+ f abc ca Ab cc 12 Aa (x)M 2 K (x, y)Aa (y)
+ 12 Aa (x)K (x, y)Aa (y).
(3.1)
We use K (x, y) = K (x y) for the inverse fermion loop in coordinate space with

1
n0 2 ()
2x x
= 2 2 21 g
K (x)
,
(3.2)
4 |x |
x2
where n = Nf TF Trspinor I . The regularized kernel K (x, y) is defined by
K (x) = C()K (x)|x|2
(3.3)
and the constant C() = 1 + O() will be specified below. The last term in (3.1) forms
part of the interaction, while the penultimate one is assigned to the free part of the action.
Since the last two terms in the action (3.1) are nonlocal, they are not renormalized and
therefore the full renormalized action takes form
2
Nf
R
= Z1 / iZ2 A
/aT a +
()/2 Aa + Z3 ca ca
S
2
2 M
+ Z4f abc ca Ab cc 12 Aa (x)M 2 K (x, y)Aa (y)
+ 12 Aa (x)K (x, y)Aa (y).
(3.4)
R cannot be brought into the form (3.1) by the redefinition of the fields and
Obviously, S
constants. Consequently, the theory with action (3.1) is not multiplicatively renormalizable.
65
A more detailed discussion of this topic can be found in the [17,34]. The absence of
multiplicative renormalizability means the following. We recall that in multiplicatively
renormalizable theories the Green functions calculated within two different subtraction
schemes are related to each other as follows:
GIn (x1 , . . . , xn , g) = An GII
n (x1 , . . . , xn , Zg g),
n
and Z8 is the field renormalization constant. In the case
where the coefficient An = Z8
n
any longer with An
under consideration this equality holds as well. However, An 6= Z8
depending on n in a nontrivial way. This of course does not contradict the statement about
the IR equivalence with QCD, since the IR asymptotics remains unchanged. Nevertheless,
the absence of multiplicative renormalizability prevents us from using the standard method
of RG analysis for the determination of critical exponents. The more widely used method
for this is the method of self-consistency equations, which is based on the direct solution of
the SchwingerDyson (SD) equations [15,17,26,32]. Another approach has been developed
recently in [24,25] and will be briefly discussed below in the context of QCD.
3.1. Extended model

The convenience of the RG method from the computational point of view consists of
the possibility of calculating critical exponents via the renormalization constants given
in (2.5). To retain this calculational advantage, following the lines of [17,24,25], we restore
the multiplicative renormalizability of the model in question by attaching two additional
charges, u and v, to the last two terms of the action (3.1). Namely, we consider the model
with the action:
u
v
Suv = SNATM Aa (x)M 2 K (x, y)Aa (y) + Aa (x)K (x, y)Aa (y). (3.5)
2
2
Obviously, the initial model (3.1) is recovered by the special choice of the parameters
u = v = 1. As distinct from the initial model the propagator of the gluon field in the
model (3.5) with u and v couplings has a more complicated form. Indeed, at v 6= 1 the
last term in the action (3.5) does not ensure the exact cancellation of the simple fermion
loop insertion in the gluon line and one should sum up all such insertions (see Fig. 1), that
yield for the transverse part of the gluon propagator:

1
(v 1) (v 1)2 2
t +
t
+
,
(3.6)
G (p; u, v) G (p) 1 +
u
u
u2
with t C()(M 2 /p2 ). This theory is obviously renormalized. In addition to the standard
renormalization constants Z1 , . . . , Z4 , which depend now on the couplings u and v, one
1
u
(1 v)
u2
(1 v)2
u3
Fig. 1. The effective gluon propagator for u, v 6= 1.
66
should add two new ones Zu and Zv to take account of the renormalization of the couplings
u and v:
u0 = uM 2 Zu ,
2
Zu = Zv = ZA
.
v0 = vZv ,
Being multiplicatively renormalizable this model can be analysed with standard RG

methods. In particular the basic RG functions are:
8 = M M ln Z8 ,
u = M M u = 2u( A ),
= M M = 2( A ),
v = M M v = 2vA ,
= MM ,
(3.7)
and
1/2
Z = Z1 ,
1/2
Zc = Z3 ,
Z = Z4 Z31 Z1 Z21 ,
ZA = Z2 Z11 ,
2
Z = ZA
.
(3.8)
Before discussing the question of the relation of the RG functions of the extended model to
the critical exponents of QCD, we derive a compact expression for the RG functions at the
point u = v = 1. Henceforth we will adopt the minimal subtraction (MS) scheme in which
the general renormalization constants have the form
Z=1+
X
1 (n)
Z (u, v, , ).
n
n=0
Then taking into account the finiteness of the RG functions and using (3.7) one obtains
= M M ln Z|u=v=1 = 2(uu )Z (1) (u, v, , )|u=v=1 .
(3.9)
Since there are no derivatives with respect to and v, we can set = = 1 + O(1/Nf )
and v = 1 from the very beginning. In this case only the first term in (3.6) survives and,
therefore, the operation (uu ) applied to a diagram gives simply the number of the
gluon lines, nA , in the latter:
(uu )0(u, v, , )|u=v=1 = nA 0(1, 1, , ).
The formula (3.9) can be rewritten as follows:
X (0) (1)
nA Z0 (1, 1, , ),
= 2
(3.10)
(3.11)
where the sum runs over the set of all diagrams. Thus to calculate the RG functions defined
by (3.9) one can put u = v = 1 from the very beginning but take into account the number
of gluon lines in each individual graph. To represent (3.11) in a compact form we attach
the factor g to the gluon propagator:
G gG ,
(3.12)
so that each diagram with k internal gluon lines acquires a factor g k . Then for the
anomalous dimensions of the basic fields, (3.9) takes the form:
(1)
(3.13)
8 = 2gg Z8 g=1 ,
where the renormalization constants Z8 are calculated within the model with u = v = 1
and the modified gluon propagator (3.12).
67
Similarly, the matrix of anomalous dimensions of a system of composite operators {Oi }

which mix under renormalization is given by
(1)
(3.14)
ik = 2gg Zik g=1 + ik nk,8 8 ,
where the mixing matrix Zik is defined in the standard way from the condition for the
Green functions of the renormalized operators OiR ,
OiR = Zik Ok ,
(3.15)
(1)
to be finite. Again, Zik is the coefficient of the simple pole in . One can easily note the
obvious resemblance of the formul (3.13) and (3.14) with those used in the MS scheme in
dimensional regularization. Therefore using this approach gives a simple and transparent
way to compute the critical exponents of operators and, importantly, all the machinery of
expansions can be adapted for the case in question as well.
3.2. Critical exponents and RG functions
In the previous subsection we have developed the effective tool for the computation of
the RG functions RG = RG (u, v)|u=v=1 in the NATM. The nontrivial exercise is to show
that these RG functions coincide with the corresponding critical exponents crit , which
determine the scaling properties of the correlators. Generally speaking, this is not so and
RG 6= crit . Indeed, the RG equation for 1-irreducible n-point Green functions with an
operator insertion in the extended uv model reads:

ik
[MM + U U + + n8 8 ] ik + RG
0k (p1 , . . . , pn+1 ; u, v, , ) = 0.
(3.16)
Bearing in mind the critical equivalence between QCD and the NATM in the Landau
gauge, we set = 0 and = and do not display their explicit dependence in the Green
functions. Since (0) = ( ) = 0 from (3.7), the above equation takes the form:

ik
(3.17)
0k (p1 , . . . , pn+1 ; u, v, , ) = 0.
[MM + U U n8 8 ] ik + RG
Evidently, due to the presence of the term U U in (3.17) the latter does not describe the
scaling properties of the Green function 0k (p1 , . . . , pn+1 ; u, v) under scale transformation
even at the point u = v = 1. To make this more transparent, taking into account (3.7), we
rewrite (3.17) as follows:

ik
0k {pl } = 2 ik A (u + v )0k {pl }; u, v u=v=1 . (3.18)
[MM n8 8 ] ik + RG
ik can be considered as the matrix of the anomalous
It is evident that the matrix RG
dimensions only if the right side of (3.18) is equal to zero. However, since in the Landau
gauge A 6= 0 the right side does not vanish identically. This means that knowledge of the
ik alone is not sufficient for the calculation of the critical exponents.
matrix RG
The problem of the relation of the RG and crit has been analysed in [24,25] where it
had been shown for the example of the O(N) nonlinear sigma model that the difference
RG crit is O(1/N 3 ). Thus, up to O(1/N 2 ) the formul (3.13) and (3.14) give the true
68
answer for the critical exponents. To prove this statement one has to show that the right
side of (3.18) is O(1/N 3 ). Since the anomalous dimension of the gluon field is O(1/Nf ),
it is sufficient to prove that the quantity (u + v )0k ({pl }; u, v)|u=v=1 vanishes at O(1/Nf )
for any Green function. The proof given in [24] is reduced to the problem of checking this
property and without any change can be simply adapted to the present model. More details
can be found in [24,25]. Thus we have justified the use of formul (3.13) and (3.14) for
the computation of critical exponents in the NATM or QCD in the Landau gauge up to
O(1/Nf2 ).
It is worth considering what happens in other gauges. Of course, as has been stressed
earlier, in this case it is only sensible to consider the anomalous dimensions of gaugeinvariant operators. In general, the anomalous dimensions of gauge-invariant operators
a )2 calculated with (3.13) and (3.14) in the extended model will
or (F
such as
be gauge dependent. Indeed, we broke the gauge invariance of the model explicitly by
introducing the coupling for the ghost-gluon vertex and implicitly by the regularization
we have introduced. So it is not surprising that the correlators of gauge-invariant objects
in this model appear to be gauge dependent. However, it follows immediately from the
critical equivalence of QCD and the NATM that at the critical-point = all dependence
on in the correlators of gauge-invariant operators factorizes
0 (xi ),
0(xi , ) = A0 ( )e
(3.19)
where e
0(xi ) is already independent of . Then the RG equation for the correlator of two
gauge-invariant operators O1 (x) and O2 (y) can be cast in the form
0 (x, y) = 0,
(MM + O1 + O2 + )e
(3.20)
where = ln A( ). Bearing in mind that the explicit breaking of gauge invariance

occurs at O(1/Nf ) only, ( = 1 + O(1/Nf )), one can deduce that is O(1/Nf3 ). This
immediately leads us to the conclusion that up to O(1/Nf2 ) the RG functions of the gaugeinvariant operator given by the formul (3.13) and (3.14) do not depend on the gauge fixing
parameter.
Finally, since u and v as well as are proportional to A from (3.7), one can nullify
all these beta functions simultaneously choosing the value of the gauge fixing parameter
from the condition

1
(2 1)
+O
A =
.
(3.21)
A (A ) = 0,
( 1)
Nf
In this gauge the RG equation in the extended uv model takes the desired form

ik
0k {pl } = 0,
[MM n8 8 ]ik + RG
(3.22)
from which it follows that the RG functions RG are the genuine anomalous dimensions
crit .
We conclude this section by summarizing the main results:
In the Landau gauge the critical exponents of QCD at the critical-point can be
calculated with standard RG methods via the renormalization constants in the
extended NATM with the help of the formul (3.13) and (3.14) up to O(1/Nf2 ).
69
Up to O(1/Nf2 ) the correspondence between the critical exponents of gauge-invariant

operators in QCD and NATM holds in any gauge.
For the special value of the gauge fixing parameter = A the formul (3.13) and
(3.14) yield the anomalous dimensions of the gauge invariant operators in all orders
of 1/Nf expansion.
In the next sections we will demonstrate the effectiveness of the above approach by
at
calculating the anomalous dimensions of the quark field and the mass operator
second order in the 1/Nf expansion. The explicit values of the renormalization constants
Zi and the basic RG functions at O(1/Nf ) are given in Appendix B.
4. Calculation of graphs
In this section we discuss the technical details of the calculation of the diagrams
relevant for the determination of the anomalous quark dimension. The technique for the
evaluation of massless Feynman diagrams is widely documented. (See, for example, [16,
35].) Nevertheless, for completeness, we provide a list of basic formul, such as the rules
for the integration of chains of propagators for scalar, spinor and vector fields and the
uniqueness relation for the different type of vertices in Appendix A. In what follows
we shall concentrate mainly on the calculational problems specific to the case under
consideration.
First, we consider the diagrams with external quark legs which are illustrated in Fig. 2. It
transpires that these are not as complicated to evaluate as may appear at first sight. Indeed
diagrams (a) and (b) are trivial in that they are equivalent to simple chain graphs in the
language of [15]. Although (c) has been considered in the case of QED we reevaluate it
here in the context of the method used to determine diagrams (d) and (e) where the latter
differ from each other by a colour factor.
First, for (c) and (d) it is convenient to choose the flow of the external momentum p in
both diagrams to be along the fermion line connecting the external vertices. The diagrams
(a)
(b)
(d)
(c)
(e)
Fig. 2. Diagrams contributing to the computation of 2 . The first graph represents the gluon self
energy diagrams of Fig. 4.
70
are linearly divergent. Differentiating them with respect to the external momentum p one
obtains a set of logarithmically divergent diagrams, which differ from the initial ones by
the insertion of a new vertex ( ) in each fermion line. This is due to the result:
p
/ + q/
p
/ + q/
p
/ + q/
.
p (p + q)2
(p + q)2
(p + q)2
(4.1)
We recall that after subtraction of the divergent subgraphs, in other words after the
application of the R0 operation, the residues of the poles in in the diagrams do not depend
on the external momenta. Therefore, we are free to choose an arbitrary route for the flow
of the external momenta in the each of the resulting diagrams to simplify the subsequent
calculations. Our choice is the following. For diagram (c) with the insertion of the vertex
( ) in the vertical fermion line we direct the external momenta flow from the upper
to the bottom vertices. When the insertion is in the right or left fermion line we choose
the route of the external momenta flow from this new vertex to be along the fermion line
nearest to the incoming (outgoing) vertex. After this rearrangement all diagrams are easily
integrated since they are reduced to simple chains. For diagram (d) we choose the inserted
vertex adjacent to the external one as the new external vertex, and direct the momentum
along the fermion line connecting them. Then, the resulting diagram is also reduced to
chains but with the insertion of the two loop master diagram in one of the lines, see Fig. 3.
The evaluation of this two loop diagram is straightforward.
We now turn to the discussion on the calculation of the diagrams contributing to the
gluon self-energy. These are shown in Fig. 4 where (b), (c) and the graph involving ghosts,
(d), are again trivial to determine and do not deserve further comment. On the contrary,
Fig. 3. New external momenta routing for diagram (d) in Fig. 4.
(a)
(b)
(e)
(c)
(d)
(f )
Fig. 4. The diagrams contributing to the gluon self-energy at O(1/Nf2 ).
71
graphs (a), (e) and (f) are tougher to evaluate and we give details of their determination.
As (e) and (f) are related by the same up to the colour factor, we will focus on the former.
Again (a) was evaluated in the QED calculation but we reconsider it here due to the fact
that new techniques were required to evaluate (e) which can more easily be appreciated in
a two loop topology. We recall first the standard method for calculating the diagrams with
these topologies in the example of the nonlinear model [15]. It will allow us to avoid the
unnecessary complications related to the nontrivial -matrix structure of these diagrams in
the case of QCD. So, we first consider diagram (a) of Fig. 4, where now all the lines are
assumed to be scalar ones, with respective indices (2 ) and ( 1) for the wavy and
arrowed lines. The diagram is superficially convergent but has two divergent subgraphs. If
one could set = 0 then the diagram could be easily integrated due to the uniqueness of
the 3-point vertex. To determine this diagram it was suggested in [15] to subtract and add
diagrams which have the same singularities as the initial one, but which can be explicitly
computed.
Since the singularities cancel in the difference one can take the limit 0. Then all
diagrams can be calculated due to uniqueness of the 3-point vertex. The extension of this
approach to the case of QCD is possible but leads to the certain problems. The first one
is the increase in the number of diagrams to be evaluated. In other words five instead of
the one for the nonlinear -model. The second and more difficult one is that for = 0
the quarkgluon vertex is not unique. Using the terminology of [35] it is one step from
uniqueness, which suggests the application of the integration by parts method. This also
increases the number of integrals to be considered. The calculation of diagrams with
propagators having nontrivial tensor structure is more involved compared to scalar ones.
Therefore it is desirable to try and keep the number of diagrams during a calculation to
a minimum. The approach which is the most economical in this sense is likely to be the
one advocated in [35]. Again, we explain the idea in the simple example of the model.
We shift the indices of the upper lines in diagram (a) by an amount and the lower ones
by . In other words we set 1 1 + in the top lines. It is important to note
that here is a temporary regularization and ought not to be confused with the parameter
which is conventionally used as the dimensional regularization in standard perturbative
calculations. The analytic expression for such a diagram can now be written as
1
F (, ),
(4.2)
where F (, ) is a regular function in the vicinity of the point = = 0. What is

important is that we have shifted the indices of the lines in such a way that the pole structure
of the diagram has remained unchanged. Next, we need to know this diagram up to constant
term in at = 0. Due to the obvious symmetry the function F only depends on
2 , and its expansion near = = 0 is
72

F (, ) = F0 + F1 + 2 F3 + 2 F4 + O 3 , 2 , 4 .
(4.3)
Since the first two terms of the expansion we are interested in are independent of , one is
allowed to evaluate the diagram at any value of . It is convenient to set = /2 which
results in the uniqueness of the upper vertex and the diagram is immediately integrable. It
is worth noting that one has only to calculate a single diagram instead of five. Applying this
procedure twice to the diagram with topology (e) one reproduces the known answer [15]
with a minimum amount of calculation. For further discussion it will be important that one
is able to determine the subsequent, 2 term in the expansion (4.3) as well. Indeed, bearing
in mind that the pole term F (0, )/ is fully determined by the counterterm diagrams
or, equivalently, is given by the application of the R0 operation to the graph in question,
one can easily deduce the coefficient F4 . This in its turn allows us to fix the value of the
coefficient F3 .
It now seems reasonable to apply this idea to the QCD diagrams. However, one obstacle
remains in that the quarkgluon vertex is not unique for the choice of exponents we are
restricted to. Though when the propagator of the vector fields has the conformal form

1
2x x
conf
,
(4.4)
G (x) = 2
(x )
x2
then the 3-point fermion vector vertex is unique if the sum of the vertex indices, 2 + ,
is equal to d + 1, where is the index of the fermion line. For this unique vertex the
relation given in Appendix A holds. However the propagator of the gluon field does not
have a conformal form. Indeed, at first order in the 1/Nf expansion in the Landau gauge,
it reads:

e
G
2(1 ) x x
,
(4.5)
G (x) = 2 1 +
(2 3 + 2) x 2
(x )
e is some constant. Of course, bearing in mind that the calculation of the diagrams
where G
with a longitudinal gluon propagator is rather trivial one may represent (4.5) in the
following form:
k
G (x) = A(, )Gconf
(x) + B(, )G (x),
k
(4.6)
where A and B are some constants and G (x) is purely longitudinal in momentum
space. In such a decomposition it is easy to check that the constants A and B are
singular as 0, A B 1/. The reason for this is that when = 0 the conformal
propagator (4.4) coincides with the longitudinal one. If these constants were finite then
one could calculate the diagram for the conformal and longitudinal part of the propagator
separately. The diagrams with longitudinal propagator would be trivial to integrate, whilst
for those with a conformal propagator could be evaluated with the methods we discussed
earlier. Now, the singularity in the coefficients of A and B lead to additional difficulties.
Namely, due to the additional factor 1/ in the coefficients A and B one should evaluate
this diagram with higher accuracy in . However it is a reasonable price to pay for the
considerable simplification which arises from the uniqueness of the triple vertex.
In the following we shall mainly discuss diagram (e) of Fig. 4 since its evaluation
was the most difficult. We will focus our discussion on the more important points as the
73
intermediate steps are not difficult to reproduce. However, to give an impression about
the effectiveness of the suggested approach it is instructive to discuss the calculation of
the QED type, Fig. 4(a), first. Using the decomposition (4.6) one reduces the problem
to the calculation of the diagram with the conformal propagator (4.4) with index =
1 , where the calculation of the diagram with longitudinal propagator is trivial. Due
to the singularity of the coefficient A() this diagram should be evaluated up to terms
linear in . Following the scheme discussed above in the case of the scalar diagram we
introduce the additional regularizing parameter in the upper and lower lines. Then
the analytic expression for this diagram is given by the same formula (4.3), where all the
necessary Lorentz indices are implied. For = /2 the diagram can be integrated due
to the uniqueness of the upper vertex, while the coefficient F4 , as was explained earlier,
can be extracted from the one-loop counterterm diagrams. Thus, we have reduced the
determination of the photon self-energy diagram to the calculation of four chains.
We now consider the calculation of the three-loop gluon self-energy diagram.
First, since the longitudinal part of the gluon polarization tensor 5 (p) can be easily
calculated, it is sufficient to calculate the above diagram with contracted indices, 5 ,
which simplifies the algebra significantly. Moreover, the contributions of all the diagrams
in Fig. 4 to the longitudinal part of the gluon polarization cancel identically. Further, since
the calculation of the diagram with the longitudinal propagator is trivial, we shall discuss
only those with the conformal ones. For this diagram we again introduce the additional
regularization in the fermion lines as shown in Fig. 5. The diagram, being superficially
convergent with two divergent two-loop subgraphs, the analytic expression for it is given
as follows:
1
0 + 1 + 2 2 + 3 3 + 2 4 + 2 5
5(, ) =

+ O 4 , 2 2 , 4 .
(4.7)
We recall that our purpose is to determine the function 5(, 0) up to the 3 terms. In
other words we are interested in the coefficients 0 , 1 , 2 and 3 .
The strategy of the calculation is the following. For = one can exploit the uniqueness
of the lower right and upper left vertices to evaluate the diagram, which allows us to find
the coefficients 0 and 1 and the combination of coefficients 2 + 4 and 3 + 5 .
To extract 2 and 3 we evaluate the coefficients 4 and 5 . The former is determined
by the counterterm diagrams and its calculation is straightforward, while the evaluation of
Fig. 5. Three-loop gluon self-energy diagram with and regularizations.
74
(a)
(b)
(c)
Fig. 6. The three resulting diagrams for Gconf Gconf case.
the latter is more nontrivial and will be discussed below. At first, however, we consider in
more detail the computation of the diagram for a special value of the parameter . We take
= , which restores the uniqueness of the two vertices.
Using the uniqueness relation one can express the initial diagram as the sum of twoloop diagrams, which after some algebra can be reduced to the ones shown in Fig. 6 and
a chain integral. Here the wavy line with labels and denotes the conformal propagator
2
Gconf
. The dashed line with label and index is used for the propagator x /(x ) . In
diagram (c) the vertical double line is used to represent x/ x/ where each x/ enters from a
different fermion cycle. We discuss briefly the calculation of each of the diagrams shown
in Fig. 6.
Diagram (a). This diagram enters the expansion with a coefficient proportional to
1 = so we need to calculate it up to O(). To this end we shift the indices
of upper fermion line by , and the lower ones by . The analytic structure of this
modified diagram is the following:
R()
+ R 0 (, )

1
R0 + 2 R1 + O 4 + R00 + R10 + O 2 , 2 ,
(4.8)
=
where we took into account the fact that due to the symmetry the diagram
depends on 2 only. Further, the diagram can be integrated at = due to uniqueness
of the GrossNeveu type at the upper vertex. Again, the coefficient R1 is determined
by the counterterm diagrams and can be easily evaluated. This is sufficient to
determine 5a (, 0) with the required accuracy.
Diagram (b). At first we note that this diagram is finite as 0 and we need it
up to O(2 ). Again, we shift the indices of the wavy lines by + and . Then the
expansion of this new diagram in and reads:
5a (, ) =
5b (, ) = R0 + 2 R1 + 2 R2 + .
(4.9)
When = 0 the diagram can be calculated exactly which allows us to find the
coefficient R2 . Next, it can be checked that the diagram can be integrated for = 2
as well. Indeed, representing ( 2x x /x 2 )(z x) as
z x
z2
(z x)2
+ x
,
2
x
x2
75
where we choose the coordinates of the left and right vertices to be 0 and z,
respectively, and the upper vertex to be x, one obtains three diagrams. The first two
of them can be calculated due to the uniqueness of the upper vertex, whilst the last is
a simple chain. We note that one needs to introduce an additional regularization , by
for example shifting the index of the lower dashed line of the original diagram by ,
to make each of the resulting diagrams finite. Of course, the sum of the diagrams is
finite in the 0 limit. Again it is sufficient to determine 5b (, 0) with the required
accuracy: 5b (, 0) = 5b (, 2) 5b (0, 2).
Diagram (c). The last diagram enters the expansion with the coefficient ( 1)2 =
2 . So one needs to calculate it with O(1) accuracy. The calculation runs along
the same lines as above. We shift the indices of the upper lines by +, and the
lower ones by , which evidently does not influence the 0 terms. To simplify the
-matrix structure of the diagrams it is convenient to represent the tensor product of
b
the -matrix, where each -matrix enters from different traces, as G
b is the numerator of the
b 0 0 , where x is the coordinate of the upper vertex and G
G
conformal propagator (4.4). We repeat this exercise for and . Then,
using the following identities:
b (x) = x/ ,
x/ G
b (x) = /
G
x x/ /x 2 ,
both traces can be cast into the form suitable for the inversion transformation. After
the inversion the upper vertex decouples from the base (left vertex) and the diagram
can be immediately integrated.
The evaluation of the coefficient 5 is grounded on the following observation. Since in
the momentum space representation the conformal propagator takes the form

e
A
conf
k
P (p)
(4.10)
G (p) = 2 1+ P (p)
2( 1 + )
(p )
the contributions to the 2 term come only from the momentum integral involving
two longitudinal propagators P k P k , or longitudinal and transverse ones, P k P .
Moreover, because the transverse propagator enters the expansion (4.10) with an additional
factor , the contribution from the latter can be extracted again from the consideration of
the counterterm diagrams alone, and does not require much work. Thus the nontrivial part
of the calculation is the determination of the diagram in question with the longitudinal
propagators. We shall use the momentum representation and at the first step contract one
momentum from the numerator of each longitudinal gluon propagators into the fermion
traces. After this the initial diagram reduces to four integrals. Two of these are identical
whilst another is a simple chain integral. Therefore, we are left with two nontrivial
diagrams which are shown in Fig. 7. For later convenience we replace the regulator
in the right fermionic triangles by . We consider the left-hand diagram first and denote
it by G(, , ). We look for the 2 term in the expansion of function G(, , ). It is
easy to see that for 0 with and fixed that G 0 as well. Indeed in the 0
limit the left-hand triangle turns into a fermion loop, which is transverse, so that when it is
contracted with the incoming momenta it vanishes. Thus one concludes that G(, , ) =
76
Fig. 7. Two diagrams contributing to the evaluation of 5 . The indices of lines are given in the
momentum space version.
e
e 0, 0), where
G(,
, ) and for our purpose it is therefore sufficient to calculate G(0,
we take into account the finiteness of the diagram. This implies the following method for
its evaluation. We ascribe the label 1 to the regulators and on the upper fermion lines
and replace 1 and 1 . Similarly for the lower lines we replace by 2 and
by 2 there. Hence, the diagram now depends on four variables G(1 , 2 , 1 , 2 ). Then
e 0, 0) = G(, 0, , 0) + G(, 0, 0, ) + G(0, , 0, , 0) + G(0, , 0, ) and each of
2 G(0,
these four diagrams can be reduced to chain integrals which completes the evaluation of
this graph.
The calculation of the right-hand diagram of Fig. 7 is more involved. In momentum
space it corresponds to
5(, ) =
(p)
5 (, )P
(p2 )1+2
(4.11)
where the transverse projector arises from the fermion loop and the nontrivial piece, 5 ,
comes from the two-loop master diagram. We begin with the analysis of the analytic
structure of the two-loop integral. First, it is superficially divergent, so one has:
R()
+ regular terms.
Second, for the same reason as we discussed for the previous graph, 5 (, ) 0 as
0. Third, the residue at the pole is independent of . Indeed, since the divergence is
logarithmic, the residue does not depend on the external momenta. Then, choosing a new
route for the external momenta flow, for example, along the vertical line, one can see that
all the dependence on disappears. Since 5 (, 0) = 0, we conclude that R = 0 and
therefore the diagram is finite. Moreover, it is not hard to check that the terms proportional
to are even in , and therefore we have:
h
i
p p
2
B
+
C
.
(4.12)
D
+

5 (, ) = 2 A +
p2
5 (, ) =
For = the diagram with contracted indices can be reduced, in coordinate space, to the
sum of chain integrals and one with a unique vertex of the GrossNeveu type. Next, it
can be checked that g 5 (, ) = 0 which results in the following constraints on the
coefficients:
D = 0,
2A + B + C = 0.
77
We recall that due to the presence of the transverse projector in (4.11) we are only interested
in the coefficient A. However, it is easier to calculate coefficients B and C and find A from
the above constraint. To determine C we repeat the technique used for the evaluation of
the previous diagram. In other words, we set 1(2) for the upper and lower lines,
respectively. Then, C is equal to the coefficient of the term in the sum of the two
diagrams 5 (, 1 = 0, 2 = ) + 5 (, 1 = , 2 = 0), which are easy to integrate.
To deduce B we first transform the master diagram into coordinate space, giving
1
e (x, , ).
5
(x 2 )2
e with
It is easy to check that for = 0, only the term proportional to B survives in 5
2
e (x, , ) B( 2 x x /x )+O(). Next, we use tetrahedral symmetry whereby
5
we add the new line, x x /(x 2 )1+ , to the diagram to obtain a vacuum graph. The
integration over x yields the pole in which is independent of which line we place
the regularizing parameter [37]. Thus we can place the regulator on the vertical line
and change the order of integration where we integrate over the upper vertex last. After
some algebra the new diagram can be reduced to chains which allows us to determine the
coefficient B and, hence, the coefficient A which we are interested in.
Clearly, the evaluation of the three-loop contribution to the gluon propagator is a tedious
exercise. To ensure that we have determined it correctly, aside from the checks we will
discuss later, we have also undertaken to calculate it by another method. As this is equally
as long an exercise we will briefly summarize the main steps. It is based on the original
method of subtractions of [15] but differs from the one outlined above in that the original
Feynman diagram is broken up into a sum of scalar integrals by taking spinor traces and the
relevant Lorentz projections. Although this results in a large set of integrals the majority of
them are in fact reducible to simple chain integrals or graphs involving the two-loop selfenergy master diagram. The latter can readily be evaluated by uniqueness methods. The
remaining three-loop graphs fall into two classes. Either they are divergent, and therefore
their simple pole and finite part with respect to need to be determined, or they are fully
-finite. In the former case we were able to apply integration by parts and related methods
to again reduce them to integrals which were chains, master diagrams or additionally
three-loop integrals which were evaluated by subtraction methods to the finite part. The
purely -finite integrals, in fact, represented the most difficult to determine and were of
the form given in Fig. 8 where n = 0, 1, 2 or 3, and k, l and q are the loop momenta with
p the external momentum. In this notation the negative integer exponent corresponds to a
numerator factor in the integrand, ((k l)2 )n . The n = 0 case had been treated previously
in [15]. To determine the other graphs we considered a related diagram which involved
fermion lines constructed in such a way that when the n traces were computed the original
scalar diagram of Fig. 8 emerged. The additional integrals which accompanied this were
again either chains or two-loop self-energy graphs, aside of course from the new fermionic
graph. This was evaluated by applying a Fourier transform to map it to coordinate space
before taking the spinor traces. However, to avoid intermediate singularities in this step
which arise from anti-uniqueness we introduced a temporary analytic regularization, , in
78
Fig. 8. Finite three-loop diagrams contributing to gluon self-energy.
the exponents of the propagators. The advantage of transforming to coordinate space is that
the resulting scalar integrals all have a vertex where the scalar uniqueness relation can be
applied to leave chain integrals. For completeness, we note that the values of the integral
of Fig. 8 are

a 3(1)a(2 3)
( 1)a 4 (1)a 2(2 2)
8() + 9 2 () +
20()
(2 3)( 2)

2
(13 35 + 23)
(4 7)(3 4)9()
+
(4.13)
+
(2 3)( 2)( 1) (2 3)( 1)2 ( 2)
for n = 2 and

( 1)a 3 (1)a(2 3)
a 4 (1)a 2 (2 2)
( 1) 8() + 9 2 ()
2(2 1)0()
(2 1)(2 3)( 2)
+
3(124 483 + 632 30 + 4)9()

(2 1)(2 3)( 2)
(585 2324 + 3393 2302 + 78 12)

(2 1)(2 3)( 1)( 2)

(4.14)
for n = 3. Including these with the other values we have verified that the same value for
the three-loop gluon self-energy is obtained in the Landau gauge.
This completes our detailed discussion of the evaluation of the three-loop self-energy
diagrams and we now collect all the contributions and record the final values for all
the relevant Feynman diagrams. First, we consider the gluon polarization tensor. To fix
normalization we write down the expressions for the propagators in momentum space:
G =
i/
p
,
p2
Ggh =
1
,
p2
G =
p p i
G M 2 h
(1
n(p2 )1+
p2
where the amplitude G in the gluon propagator reads:

G=
(4) 0(2)
.
20 2 ()0(2 )
(4.15)
Further, the contribution from the ith graph of Fig. 4 to the polarization tensor 5 can be
written in momentum space as

ab,i
(p) =
ab
(p2 )1+ni
k
+ B i P
Ai P
,
79
(4.16)
Here P and P k are the transverse and longitudinal projectors and ni is the number of the
gluon lines in the diagram. The calculation yields for the coefficients B i :
B a = (CF CA /2) R1 ,
B d = CA R2 ,
B b+c = CF R1 ,
B e+f = CA (R1 + R2 ),
(4.17)
where

(2 1)( 2)
1+
,
R1 =
(4) ( 1)0()
(2 1)( 2)
R2 =
0 2 ()a(2 1)
.
2(4)
(4.18)
P
As can be seen they are all finite and their sum vanishes: i B i = 0.
Instead of recording the values for Ai we list the values for the coefficients i , which
are related to the former as
i Bi
,
Ai =
(2 1)
and, as is easy to see, they are given by the trace of the corresponding contributions to the
polarization tensor, i = i . We find:

CA
a
R()
= 2 CF
2

3( 1)
1
1
1+
2()
,
( 2)(2 1)
( 2)(2 1)
( 1)2
(4.19)

b = c = CF R()
d =
ef

2( 1)
1
1+
+
,
( 2)(2 1)
( 2)
CA
0 2 () a(2 1),
(4)

1
2 0 3 () a(2 1)
R(2) 1
1+
= CA
2
2
( 1)
( 1) (2 1)2

2
9 2 () + 8() 62()
+
(2 1)
(84 763 + 1982 193 + 62)
9()
(2 1)2 (2 3)( 1)( 2)

2(87 606 + 2065 3744 + 3463 1492 + 34 6)
+
(2 1)2 (2 3)( 1)2 ( 2)
(104 343 + 272 + 10 12)
2 9()
(2 1)
( 1)2 ( 2)(2 1)(2 3)
(4.20)
(4.21)
80

2
.
2( 1)2 (2 1)
(4.22)
Here
R() =
( 2)(2 1)2 a( 1 + )a(1 )

,
(4) ( 1 + )a(1 + )
and the functions 9(), 8() and 2() are defined as follows:
9() = (2 3) + (3 ) (1) ( 1),
8() = 0 (2 3) 0 (3 ) 0 ( 1) + 0 (1),
(4.23)
2() = 0 ( 1) 0 (1),
where (x) = (ln 0(x))0 . Taking into account (4.16) the renormalized gluon propagator is

G
1
(1)
2
2
k
r
+
ln
p
/M
+
P
P
1
+
(4.24)
G (p) =
,
A
n(p2 )1
n
where
(1)
A =
(i)
ni 51 =
r=

CA 0
( 1)
1+
,
2( 2)
(2 1)
(4.25)
(i)
50

1
0
6 ( 1)
4( 1)
CF
2()
+
2( 2)
(2 1)
( 1)2

CA
( 1) 9 2 () + 8() 32()
+
(2 1)
+
(85 924 + 2703 3012 + 124 12)9()

2( 2)(2 1)(2 3)
(167 1206 + 4205 7764 + 7423 3492 + 84 12)

2( 1)( 2)(2 1)(2 3)
2

( + 2 2)
+
,
(4.26)
+
2
( 1)
2( 1)
(i)
(i)
and the functions 50 and 51 are defined as

(i)
= 0 (2 1) G
5(i)
1

+ 5(i)
0
(4.27)
Finally, 0 is twice the anomalous quark dimension in Landau gauge which is recorded in
(B.3):
0 =
( 2)(2 1)0(2)
.
0 2 ()0( + 1)0(2 )
(4.28)
81
We note that the part of the polarization tensor corresponding to the QED contribution is
independent of the gauge fixing parameter which is consistent with a general analysis.
For example, see [29].
Next we turn to the quark propagator and record the values for the graphs of Fig. 2. First,
for completeness we write down the value for the graph which contributes to the fermion
propagator at O(1/Nf ). Straightforward calculations give:

2( 1)
4M 2 CF 0
2
1
+
+
+
O
(4.29)
i/
p
2
(2 1)( 2)
( 1)
p2
for the above graph which results in the following anomalous quark dimension at O(1/Nf ):

CF 0
(1)
1+
(4.30)
=
2
(2 1)( 2)
and the renormalized propagator

( 1)0
i/
p
CF
(1)
2
2
+ q ln p /M
G (p) = 2 1 +
.
(4.31)
n
( 2)
p
P
Here and elsewhere we use the notation A = k A(k) /nk .
Below we list the KR0 values of the graphs of Fig. 2, which are necessary to determine
the quark anomalous dimension at O(1/Nf2 ). The value of each graph after application
of the KR0 operation will be denoted by the respective capital letter. So, for example,
p B. In addition A(i) corresponds to the value of the
applying KR0 to graph (b) of 2 gives i/
ith subgraph of Fig. 4. We find:

1
B (i)
(ni 2) (i)
(i)
(i)
5 +
50
1
, (4.32)
A = CF 0
2ni 2 1 2ni
(2 1)
( 2)
B=

2

CF2 02
1
1+
,
8
( 2)(2 1)
2

2

1
CA 02
1
1
+
C = CF CF
2
4
( 2)(2 1)
2 2

2
2( 1)( 1)
,
+
( 2)2 (2 1)
D+E=
(4.33)
(4.34)

CA 02 ( 1 + )[( 2)(2 1) + ]
12
( 2)2 (2 1)2
44 103 + 92 + 4 2
3( 1)
2()
( 2)(2 1)
2( 1)( 2)(2 1)

2
( 4 + 2)
1+
.
(4.35)
+
( 2)2 ( 1)(2 1)
2
+
82
According to (3.13) the anomalous quark dimension at O(1/Nf2 ) is given by the sum of the
residues at simple poles in multiplied by the number of gluon lines in the diagrams. The
final result for 2 will be given in Section 7.
5. SchwingerDyson approach
In this section we present the SchwingerDyson (SD) formalism to determine the
exponent 2 . There are various motivations for this. One is that in the context of the
newer approach of the previous sections it will allow an interested reader to compare and
contrast the technology involved. Though in fact it will not represent a significant amount
of extra work given that the difficult Feynman integrals have now been evaluated. Second,
it provides another check on our calculations as it is clear these are quite technical. Third,
we would like to discuss some distinctions in the treatment of the theories with abelian and
non-abelian gauge symmetries in the SD approach. Our first comment concerns the issue
of choosing the gauge fixing parameter, , entering the gluon propagator. To appreciate the
subtlety of this choice we recall that in the original construction of Vasilev et al. for scale
invariant theories such as 4 theory [15], one omits at the outset the contribution of the bare
propagator in the SD equation, which is less singular in the infrared region compared to
the loop corrections. This converts the SD equations into self-consistency equations which
determine the exponents of the propagators. In respect of the discussion of the previous
sections concerning QCD and the NATM, scale invariance is only present in two gauges.
These are the Landau gauge, = 0, and that given by = A in (3.21). Therefore, strictly
speaking, our subsequent SD analysis will only be valid for these particular gauges. By
contrast in an abelian gauge theory such as QED the gauge propagator is scale invariant in
any gauge in the 1/Nf expansion. To understand how the choice of these two gauges arises
from demanding scale invariance one can repeat the arguments given in Section 2. Indeed,
on general grounds the form of the full gluon propagator in the massless limit will take the
form

e
k k
k k
B
ek .
e
(5.1)
A (k) = 2 2 + B
(k )
k
(k 2 )
Clearly to have a scale invariant gluon propagator the exponents of the momentum
prefactor in the transverse and longitudinal pieces both ought to be the same. This can
come about in two ways subject to the physical restriction of retaining the transverse part.
One case is to have no longitudinal part which gives rise to the Landau gauge = 0. The
other is to match the powers of the exponents in each term. This requires the gluon to have
zero anomalous dimension: + = 0. Since this exponent is gauge dependent then one can
in principle solve this equation order by order in large Nf perturbation theory to determine
the explicit value of . It is this solution which we denote by A . It is worth noting that in
QED [26,27], the photon anomalous dimension vanishes in all gauges by virtue of the QED
Ward identity and therefore the ansatz for the photon propagator was automatically scale
invariant. Alternatively another point of view can be taken on this problem. One can solve
the SD equations in a non-abelian gauge theory taking a scale invariant ansatz for the gluon
83
propagator with an arbitrary gauge fixing parameter. This parameter is then considered as
an input parameter which is to be determined. Therefore, it is not hard to understand that
solving the SD equations one will find that the latter are only consistent for two particular
values of . These will be = 0 and = A .
In light of these observations we now proceed with the SD calculation but restrict
ourselves to the Landau gauge. Therefore, the ansatz for the gluon propagator in this
calculation is,

e
k k
B
e
(5.2)
A (k) = 2 2
(k )
k
in momentum space. The gluon propagator exponent is given by
= 1 ,
(5.3)
where is the quarkgluon vertex anomalous dimension. For the quark and ghost fields
we define their momentum space propagators as
e
e/
k
C
A
e =
,
c(k)
= 2
(5.4)
(k)
2
(k )
(k )
e B
e and C
e are the momentum independent amplitudes. The analogous coordinate
where A,
space amplitudes will be A, B and C, respectively. In the SD critical-point method [15],
one determines the unknown exponent by representing the two point SD equations at
the appropriate order which is O(1/Nf2 ). The representation of the equations form a set
of algebraic equations with various unknowns one of which is whilst the others are
e B
e and C.
e Eliminating the latter variables allows one
combinations of the amplitudes A,
to deduce .
To illustrate these points we focus initially on the quark equation as there are additional
features which need to be considered in the treatment of the gluon equation which will be
discussed later. If we denote by 1 and 2 the values of the respective two- and three-loop
graphs of Fig. 9 in coordinate space then the SD equation at O(1/Nf2 ) can be represented
algebraically by
+
2+2 2
+ CF (CF CA /2) x 2
z 1
0 = r( 1) + 0 zm2 x 2

2 3+3
3
+ CF CA x
TF Nf z 2 ,
(5.5)
where

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

0 = 1 +
(5.6)
Fig. 9. Quark SchwingerDyson equation to O(1/Nf2 ).
84
corresponds to the first graph of Fig. 9 and z = A2 B. The analytic regularization has been
included by shifting the vertex anomalous dimension from to + . As in the original
method of [15] we have chosen the coordinate space representation of the Schwinger
Dyson equation. To relate to the momentum space version one applies the usual Fourier
transform
Z
a()
eikx
1
=
(5.7)
(x 2 )
22
(k 2 )
k
to the version of (5.5) prior to the powers of x 2 being cancelled. The coordinate space
propagators can be determined from (5.2) and (5.4) and are taken to be

x x
B
2
A/
x
A (x) = 2 +
,
(x) = 2 ,
(2 2 1) x 2
(x )
(x )
C
(5.8)
c(x) = 2 .
(x )
The inverse quark propagator which is the origin of the first term of (5.5) is given by
1 (x) =
r() =
r( 1)/
x
,
A(x 2)22+1
a( )
2 ( )a()
where
(5.9)
(5.10)
In (5.5) the quantity m corresponds to the vertex renormalization constant which will
absorb the divergences arising in the two- and three-loop corrections. To make this more
explicit we define their -dependence as
i =
Ki
+ i0 ,
(5.11)
where the O() terms are not important at this order in 1/Nf . Therefore the quark Dyson
equation is rendered finite by defining the counterterm formally as
m1 =

(2 3)z1
(CF CA /2)K1 + CA z1 K2 ,
4(2 1)( 2)TF
where we have set:

1
m1
+O
,
m=1+
Nf
Nf2

z2
1
z1
+ 2 +O
z=
.
Nf
Nf
Nf3
(5.12)
(5.13)
However, this leaves terms in the SD equation involving ln x 2 which would otherwise
spoil the simple scaling behaviour at the fixed point. To avoid this one defines the vertex
anomalous dimension to be

(2 3)z1
(CF CA /2)K1 + 2CA z1 K2 .
(5.14)
1 =
2(2 1)( 2)TF
With these definitions the finite quark Dyson equation to O(1/Nf2 ) is

0 = r( 1) + CF
00 z1 (00 z2 + 02 z1 + 2m1 01 z1 )
+
Nf
Nf2
+ CF (CF CA /2)
z12
z13
Nf
Nf2
0 + CF CA
2 1
85
20 ,
where the first term contains 1 and 2 and we have expanded 0 as

(02 + 03 )
1
2
+O
; .
0 = 00 + 01 +
Nf
Nf2
(5.15)
(5.16)
Therefore, from (5.15) we can determine a relation for 2 in terms of z2 and the finite
correction graphs as
1
2
z2 1
+ 1 + 02 + 2m1 01 + (CF CA /2)z1 10 + CA z12 20
, (5.17)
z1
2
00
P
i
(1) /(2T ) in earlier notation.
where =
F
i=1 i /Nf and 1 =
For the gluon SD equation there is the added complication of dealing with the transverse
and longitudinal components of the equation. Whilst this has been dealt with for QED in
[27] we will recall the important features. In coordinate space the inverse gluon propagator
is [27,36]:

m()
2(2 ) x x
1
+
, where
(5.18)
A (x) =
B(x 2 )2
(2 2 1) x 2
2 =
m() =
(2 2 + 1)(2 2 1)a( )
.
4 2 ( )2 a()
(5.19)
We recall [27,36] that this is deduced by inverting the gauge-dependent momentum space
gluon propagator on the physical transverse subspace before Fourier transforming back to
coordinate space. Therefore, we can formally represent the coordinate gluon SD equation
as

2(2 + ) x x
0 = m( ) +
(2 2 1 2) x 2

2x x
2 2 +

4zTF Nf m x
x2
h

x x i
2+2
1 + 2 1
(CF CA /2)TF Nf z2 x 2
x
h

x x i
2 2 3 2 3+3
2 + 2 2
+ CA TF Nf z x
x
2
2
2
CA (2 2 1)a ( )(x )c + y x x
,
(5.20)
2( )a()
x2
where 1 and 1 relate to the coordinate space value of the two-loop diagram of Fig. 10
and 2 and 2 correspond to that of the three-loop one. The final term of (5.20) is the
ghost contribution from the ghost graph of Fig. 10. However, its treatment requires special
care. Although we have constructed the gluon equation in coordinate space, the ghost
contribution is determined from its evaluation in momentum space since the ghost couples
86
0 = A1
+
Fig. 10. Gluon SchwingerDyson equation to O(1/Nf2 ).
via a derivative to the gluon. This explains the appearance of the amplitude combination
e in (5.20). In particular the ghost graph of Fig. 10 is
e2 B
y = C

e2
p p
CA ( , , 2 + 1) 2 C
+
2(2
+
1)
.
(5.21)
2(2 + 1)(p2 )2 1
p2
Although the ghost was treated in large Nf in the Landau gauge in [18] where its
anomalous dimension and vertex anomalous dimension were computed at O(1/Nf ), the
value of y1 was not recorded. To determine it we consider the ghost SD equation of Fig. 11.
In momentum space it becomes
0 = 1 + (2 1)yC
A
( 1, , + )
,
2( 1)( + )( )
whence
(5.22)
0()c,1
,
(5.23)
(2 1)CA
P
i
where c =
i=1 c,i /Nf . Alternatively, using the SlavnovTaylor identity in the Landau
gauge we have:
y1 =
y1 =
0( + 1)0
.
2( 2)(2 1)TF
(5.24)
We note that in (5.20) the x-dependence of the final term involves c . This emerges after
using the SlavnovTaylor identity in rationalising powers of x 2 in the initial representation
of the SD equation.
Whilst there appears to be two components of (5.20) to analyse, it turns out that only one
is important. This is the transverse component when expressed in momentum space using
the Fourier transform. Although this may appear to neglect information contained in the
longitudinal part of the equation with respect to momentum space, it turns out in fact that
the sum of contributions to the longitudinal projection of (5.20) in momentum space from
the various graphs in arbitrary gauge is actually zero. The cancellation in the CF sector
was discussed in [27]. For the CA sector the contributions from the longitudinal sectors of
the last three graphs of Fig. 10 sum to zero which can be verified from the explicit values
of the graphs. This is also true for nonzero .
0 = c1 +
Fig. 11. Ghost SchwingerDyson equation at O(1/Nf ).
87
Therefore, returning to (5.20) and restoring the common factor of 1/(x 2 )2+ in each
term, we transform the equation to momentum and retain only the transverse part. Inverting
this component to coordinate space yields the relevant part of the gluon SD equation for
determining the critical exponents:
0=
2( + )m( ) 8( 1)zTF Nf m2 (x 2 )+
(2 2 + 1 + 2)
(2 1)

1
2+2
2 2
(CF CA /2)TF Nf z x
1 +
2(2 1 )

2
2 2 3 2 3+3
+ CA TF Nf z x
2 +
2(2 1 2 2)
CA 2 (2 2 1)a 2( )(x 2 )c + y
,
(5.25)
4(2 + 1)( )a()
which now requires renormalization. This is performed in the same way that (5.5) was
rendered finite. Repeating the procedure here and ensuring that there is no anomalous
scaling behaviour, we find:
0=
2( )m() 8( 1)zTF Nf
(2 2 + 1)
(2 1)

10
X1
+
(CF CA /2)TF Nf z2 10 +
2(2 1) 2(2 1)2

0
2
X2
2 2 3
0
+
+ CA TF Nf z 2 +
2(2 1) (2 1)2
CA 2 (2 2 1)a 2( )y
,
(5.26)
4(2 + 1)( )a()
where we have defined
Pi
Xi
+ i0 ,
+ i0
i =
(5.27)
i =
for i = 1 and 2, and

(2 1)z1
CA
X1
CF
P1 +
m1 =
16( 1)
2
2(2 1)

X2
,
CA z1 P2 +
2(2 1)

(2 1)z1
CA
X1
CF
P1 +
1 =
8( 1)
2
2(2 1)

X2
.
(5.28)
2CA z1 P2 +
2(2 1)
Although it may seem that (5.12) and (5.14) will give different values from (5.28) when the
explicit values for the residues of the simple poles in are substituted, they are equivalent.
Moreover, the value of 1 will agree with (5.14). Having made the gluon SchwingerDyson
equation finite, it is elementary to expand (5.26) to O(1/Nf2 ) to obtain an expression for z2 .
We find:
88

1
2
z2
1
+
+
= (1 + 1 ) 9() +
z1
2( 1) ( 2) (2 3)

41
(2 3)0()y1
(2 1)
+ CA
2
8( 1) (2 1)
4(2 1)z1

10
CA
X1
z1 10 +
+ CF
+
2
2(2 1) 2(2 1)2

20
X2
+
.
CA z12 20 +
2(2 1) (2 1)2
(5.29)
Therefore with this value for z2 we can establish a formal expression for 2 in arbitrary
gauge which depends only on the values of the various graphs of Figs. 911 by eliminating
z2 in (5.17). For completeness, we quote the coordinate space values of the integrals which
will determine 2 from these equations. First, we recall that the graphs of the QED sector
are
4
(2 3)2 0 2 ()

2(2 1)2 ( 2)2
2( 1)2 (2 3)
+ 4( 1)2
+
2

2
8(2 1) ( 2)2
,
(5.30)
4(2 1) (2 1)( 2)
(2 3)

16
3
1
(2 1)( 2)
=
+
3( 1)2()
1 =
2
( 1)
(2 3)0 2 ()

2(2 1)( 2)
.
(5.31)
(2 3)
1 =
For the three-loop graphs we have:

8(2 1)2 ( 2)
(2 3)3 2 0 4 ()0

(2 1)( 2)( 1)
6( 1)( 2)2()
2 =
86 485 + 1184 1303 + 342 + 35 12

( 1)(2 3)
and
2 =
8
(2 3)2 2 0 4 ()0

2(2 1)2 ( 1)( 2)

+ 4(2 1)( 1)( 2) 8() + 9 2 () 62()

29()
84 763 + 1982 193 + 62
(2 3)
(5.32)
89

8 47 346 + 1235 2224 + 1973 782 + 16 3
.(5.33)
+
( 1)(2 3)
The longitudinal component is related to 2 by

64
( 1)2 a 3 (1)
.
(2
1)(
2)
+
2 = 22 +
(2 3)2 2 0 4 ()0
2(2 3)a(3 )
(5.34)
Therefore, substituting the values of the integrals in the formal expressions for 1 and m1 ,
we have explicitly:
CF 0 (2 1)( 3)CA 0
,
(5.35)
2TF
8(2 1)( 2)TF
CF 0
CA 0
.
(5.36)
1 =
TF
2( 2)TF
The relation between the coordinate space variables and the momentum space ones
of earlier sections is determined from the Fourier transform of the asymptotic scaling
e
forms. For instance, it is easy to deduce that the respective quark amplitudes A and A
are related by
m1 =
e = ia( 1)A ,
(5.37)
A
( 1)
whilst the respective gluon amplitudes are determined from comparing the coefficients of
the component. We find
e = 2( )a()B .
B
(2 2 1)
(5.38)
Thus
( )a 2 ( 1)a()z
,
( 1)2 (2 2 1)
which implies
z =
z 1 =
z 2 =
2z1
(2 3)0()
and

2
2z1 1
z2
(2 3)0()
(2 3)
(5.39)
(5.40)
(5.41)
upon expanding in powers of 1/Nf . Moreover, variables involving the ghost amplitudes are
related by y = a 2 ( )a()y. Similarly, we can relate the values of the three-loop integrals
expressed in either coordinate or momentum space. For instance, if the value of the threeloop gluonic graph in coordinate space is
h
x x i
1
(5.42)
2
2
(x 2 )212
x2
and in momentum space
h
p p i
1
e2 ;
e2 +
(5.43)
(p2 )1+2
p2
90
then we have:

e2
4( 1 + )2 a 6 ( 1)a 2(1 )
e2
,
2( 1 2)
( 1)6 (2 3 + 2)2 a(2 1 2)
e2
4( 1 + )2 (2 1 2)a 6( 1)a 2(1 )
.
2 =
6
2
( 1 2)( 1) (2 3 + 2) a(2 1 2)
2 =
(5.44)
Likewise,
2 =
e2
( 3)( 1 + )3 a 5 ( 1)
.
5
3
3
16( 1) (2 3 + 2) a ( 1 + )a( 3)
(5.45)
So, for example, the finite parts of the momentum space integrals are given by
3 4
2
e20 = 8( 1) a (1)a (2 2)
(2 1)2 0()

4 9 2 () + 8() 62()
a 3 (1)
(2 1)(2 3)( 2)a(3 )
(327 1766 + 5245 9524 + 9273 4682 + 180 48)

( 1)2 (2 1)(2 3)( 2)2

4
(45 164 + 323 402 + 25 6)
9() +
( 1)
(2 1)(2 3)( 1)( 2)2

1
,
(5.46)
( 1)2
+
8a 4 (1)a 2(2 2)
e20
e20 =
(2 1)2 0()

a 3(1)
( 1)3
4(2 3)( 2)
(5.47)
2( 1)
(2 3)( 2)
a(3 )
and
e20 =
2( 1)(2 1)2 ( 2)2

2 0 3 ()0

(45 64 133 + 402 40 + 10)
.
2()
62 ( 1)2 ( 2)
(5.48)
6. Mass anomalous dimension

Having presented the formalism to calculate the quark anomalous dimension at O(1/Nf2 )
we are now turn to the quark mass anomalous dimension at the same order. Perturbatively
the mass anomalous dimension is known to four-loop accuracy. Thus comparison of results
obtained in both perturbation theory and the 1/Nf expansion will provide a nontrivial test
on the validity of our results since, for instance, only three-loop diagrams are considered
at O(1/Nf2 ) in the large Nf formalism. To complete the calculation we discuss some
observations here which allow us to relate the values of most of the graphs contributing
91
to the mass dimensions at O(1/Nf2 ) to those used for the quark anomalous dimensions. It
transpires that this greatly reduces the amount of calculation and will provide a method for
tackling the evaluation of the anomalous dimensions of other operators.
As is well known the mass anomalous dimension coincides with the anomalous
dimension of the mass operator :

m =
+ 2 .
(6.1)
Here is the quark anomalous dimension and the RG function

is expressed via the
renormalization constant of the mass operator by
= MM ln Z
,
2
R = Z
[]
Z [
0 0 ]0 .
(6.2)
To determine the renormalization constant Z

at second order in the 1/Nf expansion one
has to compute the divergences of the one-particle irreducible 2-point Green function with
the insertion of the mass operator. The corresponding diagrams can be obtained from the
vertex in
diagrams contributing to the quark self-energy of Fig. 2 by the insertion of a
the quark lines connecting the external vertices. At the level of diagrams this equivalent to
the insertion of the unit operator in a fermion line
p
/ p
/
p
/
2 2.
p2
p p
As was discussed in Section 4, the differentiation of the propagator diagrams with respect
to p , where p is the external momentum, results in the same set of the diagrams as for
the mass operator. The only difference is that the inserted vertex in this case is ( ).
We shall exploit this fact and show below that provided the Landau gauge is used then
the corresponding contributions to Z1 = Z2 and Z
from all diagrams from both sets,
except for two which arise from graphs (b) and (c) of Fig. 2 when the insertion of the new
vertex is in the middle quark line, are related to one other by
(i)
(i)
Z
=
Z1
.
( 2)
(6.3)
So given that is known then to determine the mass dimension m at O(1/Nf2 ) is

from the two extra diagrams.
sufficient to compute the contributions to Z1 and Z
However, their calculation does not lead to any difficulties using the methods discussed
in Section 4.
We now proceed to the proof of the relation (6.3). First, we use the freedom we have
to choose the external momentum flow in a diagram arbitrarily when calculating the
contribution to the renormalization constants by directing the external momentum flow
through the inserted vertex and out through the nearest external vertex. After this all the
diagrams in question take the form shown in Fig. 12. All that one needs to know about the
coloured block in Fig. 12 is its Lorentz structure

1
q q/
(6.4)
0 (q, ) = 2 n A() + 2 B() ,
(q ) i
q
where ni is the number of the gluon lines in the block and the exact form of the functions
A() and B() are irrelevant for our present discussion. Since the gluon propagator which
92
Fig. 12. External momenta routing in the quark 2-point function with an operator insertion. The large
black dot stands for the Vi = {I, ( )}.
joins this generalized vertex is purely transverse in the Landau gauge the second term in
(6.4) will not contribute to the final answer. Thus the momentum integral for the diagram
in Fig. 12 can be cast in the form
Z
(q)
p q/ ) Vi q/ P
d2 q (/
.
(6.5)
A()
(2)2 (p q)2 (q 2 )+(ni +1)
We are interested in the pole terms of this integral only, and bearing in mind that after
the subtraction of the divergent subgraphs which also have the same form, the latter is
independent of p. Therefore one can rewrite (6.5) as
Z

1
1
d2 q
Vi
,
(6.6)
A() Vi
2
(2)2 (p q)2 (q 2 )1(ni +1)
where we have used the fact that up to p-dependent terms one can replace q q by q 2 /(2)
in the integrand. It is easy to deduce that the combination in the prefactor in (6.6) is equal to
(2 1) for V = I , and to (2 1)( 2)/ for V = . Then one can immediately
conclude that irrespective of the exact form of the constant A() which corresponds to the
structure of the coloured block in Fig. 12, the contributions of the corresponding diagram
to the renormalization constants Z1 and Z
are indeed related by Eq. (6.3). To complete
the calculation of m one needs to compute the contributions from the graphs originating
from those (b) and (c) in Fig. 2. These, as well as the determination of the graphs when the
gluon propagator is taken in an arbitrary gauge, can readily be performed.
7. Discussion
We are now in a position to assemble our results. First, using the formalism of Section 3
we record that the d-dimensional expression for the quark anomalous dimension is
(2) =
CF 02
2( 2)(2 1)

1
2( 1)2 ( 3)
+ 3( 1) 2()
CF
( 2)
( 1)2

CA (124 723 + 1262 75 + 11)
+
2
(2 1)(2 3)( 2)
( 1) 9 2 () + 8()
93
(85 924 + 2703 3012 + 124 12)9()

2( 2)(2 1)(2 3)

2 (23 112 + 12 4)
(2 4 + 2)
2( 1)( 2)
4( 1)( 2)(2 1)
+
(7.1)
We note that as an initial check on the result setting CF = 1 and CA = 0 one obtains the
expression for the quark anomalous dimension of the abelian Thirring model. Again, in
agreement with general results [29], we find that in this case (2) is independent of the
gauge fixing parameter. Further, collecting all the results for the quark mass anomalous
dimension, we find:

1
2( 1)2 ( 3)
+ 3( 1) 2()
m,2 = 2
( 2)
( 1)2

CF2 02
(22 4 + 1)
+
( 2)
( 2)2 (2 1)

(124 723 + 1262 75 + 11)
( 1) 9 2 () + 8()
(2 1)(2 3)( 2)
(85 924 + 2703 3012 + 124 12)9()
2(2 1)(2 3)( 2)

CF CA 02
2 (2 3)2
,
(7.2)
4( 2)( 1) ( 2)2 (2 1)
P
r
where m (g ) =
r=1 m,r /n . We have performed the calculations with a nonzero gauge
parameter to check that it cancels exactly since the mass dimension is a gauge-independent
quantity. Aside from the internal checks on the integration which we discussed earlier,
there are other checks on the correctness of (7.1) and (7.2). First, the terms of (7.1)
and (7.2) involving CF2 have both been derived previously in QED in [27] and [28],
respectively. However, a more stringent check rests in the relation of the results with the
critical renormalization group equation since the RG functions evaluated at g ought to
agree with (7.1) and (7.2) at the same order in 1/Nf and O( 4 ). It is elementary to verify
that our results are in agreement with the Landau gauge anomalous dimension of [10,39].
Also the result for m (g ) is consistent with [2,3,11,14].
Having verified this agreement we can now determine new information on the structure
of the renormalization group functions in principle to all orders in the strong coupling
constant. However, we will record values of the coefficients to six-loops by writing:

3
25
1
(a) = TF Nf + CF CA CF a 2
4
16
32
+
+ 320TF2 Nf2 + 432CF TF Nf 4592CA TF Nf + 216CF2
94

CF a 3
+ 1728 (3) 5148 CF CA + 9155 1242 (3) CA2
4608

X

1
cr0 TFr1 Nfr1 + cr1 TFr2 Nfr2 CF a r + O
+
3
N
f
r=4
(7.3)
for the Landau gauge and

3
10
3
97
TF Nf CF CA CF a 2
m (a) = CF a +
2
24
16
48

X

1
+
mr0 TFr1 Nfr1 + mr1 TFr2 Nfr2 CF a r + O
Nf3
r=3
(7.4)
for the mass dimension where the order symbol means that contributions which are thirdorder in 1/Nf and O( 7 ) at criticality are ignored. Although the full four-loop MS function is only available it may seem that it is not possible to decode the Landau gauge
information in (7.1) and (7.2) beyond four-loops since this requires a to O( 6 ). However,
recalling that we are computing within the 1/Nf expansion we can use the information
contained in the QCD -function exponent = 0 (a )/2 which has been determined at
O(1/Nf ) in [19]. From [19], we deduce that the relevant corrections are

3
33
27
45
CA
CF + CA 2
+
a =
TF Nf
4
4
4

99
237
77
53
3
CF +
CA +
CF + CA 4
+
16
32
16
32

3
288 (3) + 214)CF + (480 (3) 229 CA 5
256

1
3168 (3) 2592 (4) 1506 CF
256

1
+ 3792 (3) 4320 (4) + 1359 CA 6 + O 7
2
TF Nf2

1
+O
.
(7.5)
Nf3
Therefore, with this value we deduce the new exact contributions to (7.3) and (7.4):
35
,
1296

CA
CF
19 18 (3) +
293 + 288 (3) ,
c41 =
72
1152

1
83 144 (3) ,
c50 =
7776

CF
5616 (3) 3888 (4) 659
c51 =
31104

CA
7776 (4) 8928 (3) 1783 ,
+
62208
c40 =

1
65 144 (4) + 80 (3) ,
15552

CF
4212 (4) 2592 (5) 432 (3) 2219
c61 =
46656

CA
4608 (5) 1440 (4) 4744 (3) 925
+
31104
95
c60 =
(7.6)
and
5 (3) (4)
65
+
,
162
18
2592

4483
5 (4) (5) 11 (3)
CF
m51 =
8
3
48
20736

18667
8 (5) 17 (4) 671 (3)
CA ,
+
9
36
1296
124416
m50 =
(7.7)

1
560 (3) + 720 (4) 1728 (5) + 451 ,
46656
CF
46704 (3) 23328 (4) + 51840 (5) 41472 2(3)
m61 =
186624

25920 (6) 12283
CA
56728 (3) 59472 (4) 112896 (5) + 55296 2(3)
+
186624

+ 112320 (6) 22709 .
(7.8)
m60 =
We have given the explicit expression for c41 in order to compare with the full MS gauge
calculation of [39] which was carried out for the particular colour group SU(Nc ) and note
exact agreement. Further, we note that in [23] it was pointed out how the asymptotic Pad
approximant predictions of [40] for m50 and m51 compared with the exact values we have
determined.
Although we have concentrated on the relation of our results to four-dimensional
perturbation theory, we can also quote values of the exponents in three dimensions. These
will be important for alternative critical-point investigations of the non-abelian Thirring
model and QCD equivalence such as the lattice. For the wave-function, we find from (7.1):

8CF
4CF (3 2)
+
64 6 2 CF + 8 2 + 14 18 2 CA
=
2
2
2
3 TF Nf
9 4 TF Nf

1
.
(7.9)
+O
Nf3
However, for the physically more interesting mass exponent we deduce:

32CF
32CF [(56 6 2 )CF (18 + 2 )CA ]
1
+
+
O
.
m = 2
3 TF Nf
9 4 TF2 Nf2
Nf3
(7.10)
It would be interesting to see how the numerical values of these exponents compare with
the lattice. Indeed the dominant contribution from the O(1/Nf2 ) correction in the latter
exponent arises from the CA term.
96
In conclusion we have provided an elegant alternative to computing perturbative

information on QCD at a new order in the large Nf expansion. Although the evaluation of
the Feynman integrals has formed a significant part of the discussion, is technically quite
involved and provides useful techniques for future massless integral evaluation, we believe
we have demonstrated that the computation of other renormalization group functions in
QCD is viable in principle. An example of this is the twist-2 operators which arise in
deep inelastic scattering. Furthermore, the relation of the NATM to QCD has been put on
a firmer ground. Indeed we believe that this critical relation between both models deserves
further investigation.
Acknowledgements
This work was carried out with the support of the Russian Foundation of Basic Research,
Grant 97-01-01152 (SED and ANM), by DFG Project N KI-623/1-2 (SED), a University
of Pisa Exchange Research Fellowship (MC), PPARC through an Advanced Fellowship
(JAG) and GSI (ANM). Also SED and ANM would like to thank Prof. A.N. Vasilev
for useful discussions. Invaluable to the calculations were the symbolic manipulation
programme F ORM [41] and computer algebra package R EDUCE [42].
Appendix A. Basic integration rules

In this appendix we give formul for the Fourier transformation, integration of chains
and uniqueness relations for the vertices of different type. The following notation is used.
We define
a(x1 , . . . , xn ) =
n
Y
a(xi ),
i=1
where a() = 0( 0 )/ 0() with 0 = ( ). The Fourier transforms of various functions

are given by:
Z
0
eipx
22
d2 x 2 = a() 2 0 ,
(x )
(p )
Z
x
a() p
0
,
d2 x eipx 2 +1 = i 22 1
(p2 ) 0
(x )
Z
0
i
x x
a() 22 1 h
0 p p
,
(A.1)
d2 x eipx 2 +1 =
(x )
(p2 ) 0
p2
Z
d2 x eipx
x x x
(x 2 )+1
0 22 h
p p p i
0
= i a()
+
p
+
p
2(
+
1)
p
,
0
(p2 ) +1
p2
0
97

i
2 0 h
eipx
2x x
a() 2
0 p p
+
2
(
1)
,
=
d x 2
0
(x )
x2
(p2 )
p2

Z
0
(2 1) a() 22 1
2 0 x x
d2 k ipx P
e
=
+
.
(A.2)
(2)2
(k 2 )
(4) (x 2 ) 0
(2 1) x 2
Z
The integration of the basic chain with various tensor generalizations are given by:
Z 2
a(, , 0 + 0 )
1
d x
=
,
((z x)2 ) (x 2 )
z2(+)
Z 2
a(, , 0 + 0 ) 0
x
z
d x
=
,
((z x)2 ) (x 2 )
0 + 0
(z2 )+
Z
x (z x)
d2 x
((z x)2 ) (x 2 )
0 0 a(, , 0 + 0 + 1) h
z z i
0
=
+
2(
+
1
2( 0 + 0 + 1) (z2 )+1
z2
1
x/ (/z x/ )
d2 x
= 0 0 a(, , 0 + 0 + 1) 2 +1 ,
((z x)2 ) (x 2 )
(z )
h
x x i
1
d2 x
+
B
A
((z x)2 ) (x 2 )
x2

0
0
0
B 0 ( 0 ) z z
B
a(, , + )
+
A+
.
=
(z2 )+
2( 0 + 0 )
( 0 + 0 ) z2
(A.3)
(A.4)
Next we list the uniqueness relations for the three type of vertices which arose in
our calculations where the appropriate uniqueness value for each vertex is recorded in
parentheses beside each rule.
Boson vertex ( + + = 2).
= a(, , )
GrossNeveu vertex ( + + = 2 1).
a(, , )
98
Vector-fermion vertex ( + + = 2 1).
a(,
,
( 1) + 2 0 3x (z, y)
Here
3x (z, y) =
(z x) (y x)
.
(z x)2 (y x)2
As usual the line with an arrow with index denotes the fermion propagator, x/ /(x 2 ) ,
a simple line corresponds to the boson propagator 1/(x 2) and a wavy one is used to
denote the conformal propagator (4.4).
Appendix B. Renormalization constants

In this appendix we collect the values of the renormalization constants Zi and the basic
RG functions at O(1/Nf ). The calculations are rather straightforward and so we only list
the results. We found:

gCF 0
1+
,
Z1 = 1
2n
(2 1)( 2)

g0
g 2 CA 0 1 +
(CF CA /2) 1 +
,
Z2 = 1
2n
(2 1)( 2)
8n
2

g2 CA 0
1
,
Z3 = 1 +
4n( 2)
(2 1)

g
g CA 0
+
,
(B.1)
Z4 = 1
8n( 2) (2 1) 2
where
( 2)(2 1)0(2)
0 2 ()0( + 1)0(2 )
and g is defined in (3.12). Using (3.13), (3.14) and (3.7) one finds:

(1)
(1)
,
= 2 = CF 0 1 +
(2 1)( 2)

CA 0
1
,
c(1) =
4( 2)
(2 1)
0 =
(B.2)
(B.3)
(B.4)

(1)
A =

CA 0
( 1)
1+
.
2( 2)
(2 1)
99
(B.5)
For we obtain
= (A + 2c 4 )

(2 1)
( 1)CA 0
( 2(2 1)) 2( 1) +
.
=
4( 2)(2 1)
( 1)
(B.6)
One can see that = 1 is indeed a zero of the beta function. Also, for A we find
(1)
A =
(2 1)
,
( 1)
(B.7)
and is simplified in this gauge to

=
(2 1)CA 0 2
( 1).
4( 2)( 1)
(B.8)
Finally, for completeness we record the value of the amplitude combination z required
in the SD formalism at O(1/Nf2 ). It is
z 1 =
0( + 1)0
2(2 1)( 2)
z 2 =

3( 1)0( + 1)CF 02
1
2()
2(2 1)2 ( 2)2

( 1)2
(B.9)
and
( 1)0( + 1)CA 02
4(2 1)2 ( 2)2

(85 924 + 2703 3012 + 124 12)9()
32() +
2( 1)(2 1)(2 3)( 2)
2
(2 + 2 2)
+
9 2 () 8() +
2
4( 1)
2( 1)2 2

(167 1206 + 4205 7764 + 7423 3492 + 84 12)
,
2(2 1)(2 3)( 1)2 ( 2)2
(B.10)
where the gluon propagator in the SD formalism is taken as

e
B
k k
e
A (k) = 2 (1 ) 2
(k )
k
(B.11)
in momentum space. These will be useful in the calculation of gauge-independent critical

exponents at O(1/Nf2 ). For example, if one chooses a particular gauge to simplify those
calculations, such as the Feynman gauge used in standard perturbative calculations, then
one will require the value of the relevant variables for a general gauge parameter.
100
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
T. van Ritbergen, J.A.M. Vermaseren, S.A. Larin, Phys. Lett. B 400 (1997) 379.
J.A.M. Vermaseren, S.A. Larin, T. van Ritbergen, Phys. Lett. B 405 (1997) 327.
K.G. Chetyrkin, Phys. Lett. B 404 (1997) 161.
D.J. Gross, F.J. Wilczek, Phys. Rev. Lett. 30 (1973) 1343.
H.D. Politzer, Phys. Rev. Lett. 30 (1973) 1346.
W.E. Caswell, Phys. Rev. Lett. 33 (1974) 244.
D.R.T. Jones, Nucl. Phys. B 75 (1974) 531.
E.S. Egorian, O.V. Tarasov, Theor. Math. Phys. 41 (1979) 26.
O.V. Tarasov, A.A. Vladimirov, A.Yu. Zharkov, Phys. Lett. B 93 (1980) 429.
S.A. Larin, J.A.M. Vermaseren, Phys. Lett. B 303 (1993) 334.
D.V. Nanopoulos, D.A. Ross, Nucl. Phys. B 157 (1979) 273.
R. Tarrach, Nucl. Phys. B 183 (1981) 384.
O. Nachtmann, W. Wetzel, Nucl. Phys. B 187 (1981) 333.
O. Tarasov, JINR preprint P2-82-900.
A.N. Vasilev, Yu.M. Pismak, J.R. Honkonen, Theor. Math. Phys. 46 (1981) 157; Theor. Math.
Phys. 47 (1981) 291.
A.N. Vasilev, Yu.M. Pismak, J.R. Honkonen, Theor. Math. Phys. 50 (1982) 127.
A.N. Vasilev, M.Yu. Nalimov, Theor. Math. Phys. 56 (1982) 643.
J.A. Gracey, Phys. Lett. B 318 (1993) 177.
J.A. Gracey, Phys. Lett. B 322 (1994) 141; Nucl. Phys. B 480 (1996) 73.
J.F. Bennett, J.A. Gracey, Nucl. Phys. B 517 (1998) 241.
A. Hasenfratz, P. Hasenfratz, Phys. Lett. B 297 (1992) 166.
M. Ciuchini, S.. Derkachov, J.A. Gracey, A.N. Manashov, Phys. Lett. B 458 (1999) 117.
S.. Derkachov, A.N. Manashov, Nucl. Phys. B 522 (1998) 301.
S.. Derkachov, A.N. Manashov, Theor. Math. Phys. 116 (1998) 379.
J.A. Gracey, J. Phys. A 24 (1991) L431.
J.A. Gracey, Nucl. Phys. B 414 (1994) 614.
J. Zinn-Justin, Quantum Field Theory and Critical Phenomena, Clarendon, Oxford, 1990.
A.N. Vasilev, M.M. Perekalin, Yu.M. Pismak, Theor. Math. Phys. 55 (1982) 323.
A.N. Vasilev, M.M. Perekalin, Yu.M. Pismak, Theor. Math. Phys. 60 (1984) 317.
J.A. Gracey, Int. J. Mod. Phys. A 8 (1993) 2465.
A.A. Slavnov, Nucl. Phys. B 31 (1971) 301.
A.N. Vasilev, A.S. Stepanenko, Theor. Math. Phys. 94 (1982) 127.
D.I. Kazakov, Phys. Lett. B 133 (1983) 406.
A.N. Vasilev, Yu.M. Pismak, J.R. Honkonen, Theor. Math. Phys. 48 (1981) 750.
S.G. Gorishnii, A.P. Isaev, Theor. Math. Phys. 62 (1985) 232.
P. Ramond, Field Theory: A Modern Primer, Frontiers in Physics Lecture Note Series, Vol. 51,
Benjamin Cummings, Reading, MA, 1981.
K.G. Chetyrkin, A. Rtey, hep-ph/9910332.
J. Ellis, I. Jack, D.R.T. Jones, M. Karliner, M.A. Samuel, Phys. Rev. D 57 (1998) 2665.
J.A.M. Vermaseren, FORM, Version 2.3, CAN publication, 1991.
A.C. Hearn, REDUCE Users Manual, Version 3.4, Rand publication CP78, 1991.

One-loop Khler potential in non-renormalizable

theories
Andrea Brignole 1
Istituto Nazionale di Fisica Nucleare, Sezione di Padova, Dipartimento di Fisica G. Galilei, Universit di
Padova, Via Marzolo n.8, I-35131 Padua, Italy
Received 26 January 2000; accepted 30 March 2000
Abstract
We consider a general d = 4, N = 1 globally supersymmetric Lagrangian involving chiral
and vector superfields, with arbitrary superpotential, Khler potential and gauge kinetic function.
We compute perturbative quantum corrections by employing a component field approach that
respects supersymmetry and background gauge invariance. In particular, we obtain the full one-loop
correction to the Khler potential in supersymmetric Landau gauge. Two derivations of this result are
described. The non-renormalization of the superpotential and the quadratic correction to the Fayet
Iliopoulos terms are further checks of our computations. 2000 Elsevier Science B.V. All rights
reserved.
PACS: 11.10.Gh; 11.30.Pb; 12.60.Jv
Keywords: Supersymmetry; Khler potential; Quantum corrections
1. Introduction
Perturbative quantum corrections have a peculiar form in supersymmetric theories. In
the case of d = 4, N = 1 theories, in particular, the non-renormalization theorem [1]
establishes that the superpotential remains uncorrected, whereas the Khler potential
generically receives quantum corrections. The anomalous dimensions of chiral superfields
are the simplest example of the latter effects, but a richer structure emerges if the full
field dependence of such corrections is taken into account. Full one-loop corrections to the
Khler potential have been recently computed, both in the WessZumino model [2] and
in more general renormalizable models [35]. For the most general renormalizable N = 1
theory, the one-loop correction to the Khler potential was found to have a very compact
form in supersymmetric Landau gauge. The result reads [5]:
1 brignole@padova.infn.it
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 2 1 1 - X
102
A. Brignole / Nuclear Physics B 579 (2000) 101116

M2
M2V
1
2
2
1K =
2 Tr MV log 2 1
,
Tr M log 2 1
32 2
(1)
where M2 and M2V are the (chiral superfield dependent) mass matrices in the chiral
and vector superfield sectors, respectively, and is an ultraviolet cutoff. One-loop
corrections to the Khler potential have also been investigated in non-renormalizable
N = 1 models, divergent contributions being the main focus. For instance, quadratically
divergent corrections to the Khler potential in general models were computed in [6].
Quadratic and logarithmic divergences were also studied in general supergravity models
with diagonal gauge kinetic function [79] or in models with chiral superfields only [10].
In [11], the Wilsonian evolution of the Khler potential was studied in non-renormalizable
models with an abelian vector superfield and/or gauge singlet chiral superfields. Divergent
and finite corrections in specific models were also evaluated in [12].
The main purpose of this paper is to generalize the result (1) to non-renormalizable
theories, i.e., to compute the full (divergent and finite) one-loop correction to the Khler
potential in a general globally supersymmetric theory. Our perturbative calculation starts
from a tree-level Lagrangian in which the superpotential, the Khler potential, the gauge
kinetic function, the gauge group and the matter representations are arbitrary. Upon
quantizing the theory, a supersymmetric gauge fixing term is added, and we choose to
preserve supersymmetric background gauge invariance. This framework is then translated
to the component field level and quantum corrections are computed in terms of component
Feynman diagrams. Notice that we do not choose the WessZumino gauge, supplemented
by a gauge fixing term for the component vector fields. Instead, we keep all components
of quantum supermultiplets and use supersymmetric Landau gauge. Thus our component
computations are equivalent to superfield computations, and a superfield language can be
used to interpret our results. In particular, we obtain the full one-loop correction to the
Khler potential, which is our main result. In spite of the fact that the interactions are
considerably more complicated in comparison to the renormalizable case, we find that the
logarithmically divergent and finite one-loop corrections to the Khler potential can be cast
in the same form as in Eq. (1), with generalized mass matrices M2 and M2V . In addition
to that, the Khler potential receives a quadratically divergent correction, in agreement
with [6]. A consistency check based on supersymmetric background gauge invariance is
also discussed. The non-renormalization of the superpotential and the quadratic correction
to the FayetIliopoulos terms are further checks of our computations.
2. Theoretical framework
We consider a general d = 4, N = 1 globally supersymmetric theory defined by a treelevel Lagrangian of the form (see, e.g., [13,14]):
Z
Z
2Vb

2
ba
e + 2a V
L=
d w() + h.c. + d 4 K ,

Z
2 1
a bb
b
(2)
+
d fab ()W W + h.c. .
4
103
The theory has a general (possibly product) gauge group G, with hermitian generators Ta
c
b=V
ba Ta
satisfying the Lie algebra [Ta , Tb ] = icab
Tc . The associated vector superfields V
b=W
b a Ta = 1 D D e2VbD e2Vb. The chiral superfields
have superfield strengths W
8
i
= { } belong to a general (reducible) representation of G. Supersymmetric gauge
b
b b
b
b
b=
ba Ta is chiral.
where
transformations read e2V e2i e2V e2i and e2i ,
The FayetIliopoulos coefficients a are real and may be nonvanishing only for the abelian
factors of the gauge group G. The superpotential w, the Khler potential K and the gauge
kinetic function fab are only constrained by gauge invariance, and are otherwise arbitrary.
)
transforms
is G-invariant, K(,
is real and G-invariant, and fab ()
In more detail: w()
as a symmetric product of adjoint representations of G. These constraints are expressed by
the identities:
i 0,
a )
wi ()(T

i
K ,
(T )

Ki ,
a T
a
,
d
d
c )
+ icbc
i icac
fdb ()
fad (),
fabi ()(T
(3)
(4)
(5)
i (Ta )i j , (T
a ) (Ta ) (T a ) , wi ()
w()/
i , and so on.
where (Ta )
j
Further identities can be obtained by differentiating the ones above. We also recall that,
in the special case of a renormalizable theory, w is at most cubic in the chiral superfields,
K is quadratic (i.e., canonical) and fab is constant (i.e., canonical). Here we aim at full
generality and do not impose renormalizability.
The supersymmetric Lagrangian above is the most general one that contains no more
than two spacetime derivatives on component fields. 1 It could arise as a low-energy limit
of a more fundamental theory. In fact, since we take w, K, fab to be generic functions,
not restricted by renormalizability, the Lagrangian L in Eq. (2) necessarily describes an
effective theory, valid below some cutoff scale . For consistency reasons, cannot be
much larger than the scale whose inverse powers control the non-renormalizable terms
in w, K, fab . Anyhow, we will not address the origin of the Lagrangian itself. We will
just take L as a general classical bare Lagrangian and study the corresponding oneloop corrections. In principle, such self-corrections could be matched to those of the
hypothetical underlying theory, if the latter theory were known.
Quantum corrections to a classical supersymmetric Lagrangian can be computed by
various techniques. We find it convenient to use the background field method [1,14]. The
V
b are split into background (still denoted by ,
V
b) and quantum (, V ) parts.
superfields ,
A supersymmetric gauge fixing Lagrangian Lgf is added in order to break the quantum
gauge invariance, and a corresponding ghost Lagrangian Lgh is introduced. If we choose
to preserve supersymmetric background gauge invariance, we can take [1,14]:
+ ,
e
b
2V
(6)
b 2V V
b
V
e e
e ,
(7)
1 This is already a nontrivial generalization of the renormalizable case. A further generalization would be the
inclusion of supersymmetric higher derivative terms, which are additional non-renormalizable terms. We leave
this to future investigation.
104
Lgf =
1
8
Z
d 4 2V
a
2V
a
(8)
where the background vector superfield has been put in a convenient form, is a gauge
b
b
b
b
parameter and eV D eV , eV D eV are background gauge covariant supersymmetric derivatives. Ghost superfields interact with vector superfields only, and we will
not need the explicit expression of Lgh . In the abelian case, the splitting of the vector
bV
b + V and the gauge fixing term (8) becomes
superfield in (7) reduces to V
Z
1
(9)
d 4 D2 V a D 2 V a .
Lgf =
8
Making these simpler choices in the non-abelian case is certainly allowed, but does not
lead to a background gauge invariant effective action. Since we find it useful to preserve
the latter property, we proceed in the way explained above.
Once the replacements (6), (7) have been made in (2) and the gauge fixing and ghost
terms have been added, the resulting Lagrangian can be expanded in powers of the
quantum superfields. The zero-th order part is just the original Lagrangian (2) for the
b). The terms linear in quantum superfields do
V
classical (background) superfields, L(,
not contribute to the (one-particle-irreducible) effective action and can be dropped. The
b), is the relevant one for the
V
part bilinear in quantum superfields, Lbil (, V , . . . ; ,
computation of one-loop quantum corrections. One has to integrate out the quantum
superfields in the theory defined by L + Lbil , either diagrammatically or by direct
functional methods. The result of this operation gives the one-loop-corrected effective
action, a functional of background superfields only. If a (super) derivative expansion of the
latter functional is performed, the lowest order terms can be interpreted as corrections to
the basic functions in (2). In other words, one-loop corrections to the effective Lagrangian
will have the form:
Z
Z

a
e2Vb + 21 V
+ h.c. + d 4 1K ,
(10)
1L =
d 2 1w()
ab + ,
bW
b terms and
where the dots stand for terms containing supercovariant derivatives (W
higher derivative terms), which we will not compute. Our purpose is to compute the
corrections 1w, 1K and 1a .
3. Component field approach

The expected form (10) of quantum corrections relies on the assumption that quantum
superfields are integrated out in a supersymmetric way. This is automatic if the perturbative
computations are performed at the superfield level. Here we choose to work with
component fields, instead. Nevertheless, we retain the whole off-shell structure of quantum
supermultiplets and literally translate the framework described above to the component
level. In particular, instead of simplifying the structure of vector supermultiplets by using
the WessZumino gauge, supplemented by a gauge fixing term for the component vector
fields only, we work with full supermultiplets and use the manifestly supersymmetric gauge
105
fixing term (8). Thus integrating out component fields is literally equivalent to integrating
out superfields, and supersymmetric background gauge invariance can also be preserved.
In order to fix the notation, we recall the component expansion of chiral and vector
superfields: 2
i
1
i , (11)
i i 2
i = i + 2 i + F i i
4
2
i
i
a
a
a
a
a
V = C + i i + G Ga + Aa
2
2

i
i
+ i a a i a a
2
2

1
a
1
a
(12)
+ D 2C ,
2
where C a , D a are real scalar fields, i , F i , Ga are complex scalar fields, Aa are real
vector fields, and i , a , a are complex Weyl fields. The expansions above apply to the
quantum superfields i and V a . Similar component expansions hold for the background
ba , and one can eventually obtain the full component expansion of Lbil ,
superfields i and V
to be used for the computation of 1L. However, since supersymmetry constrains one-loop
corrections to have the form (10), a convenient choice of the background superfields can
simplify the computation of the functions 1K and 1w. For instance, in order to compute
the function 1K, we could choose a background in which the only nonvanishing fields are
the scalars i (x) contained in i . Then Eq. (10) predicts that the one-loop computation
)
j . Hence the one-loop correction
should produce terms of the form 1Kj (,
)
to the Khler metric could be identified and the functional form of the one1Kj (,
loop-corrected Khler potential could be reconstructed by integration. Alternative choices
of the background can be even more convenient. In what follows, we will make this choice:
bi ,
i = i + F
ba ,
ba = 1 D
V
2
(13)
(14)
bi , D
ba are taken to be constant
where both the physical scalars i and the auxiliary fields F
(i.e., spacetime independent). If we specialize (10) to this background, we infer that the
one-loop computation should generate the terms:

b j

b
bi + h.c. + 1Kj ,
F
F
F
1L = 1wi ()

a
j
(Ta )
b +
+ 1Kj ,
+ 1a D
(15)
bW
b and
b, D
b and correspond to the (omitted) W
The dots stand for higher order terms in F
higher derivative terms in (10). The computational advantage of using a background with
b and D
b is obvious: the one-loop diagrams to be evaluated have vanishing
constant ,
F
external momenta. In other words, in such a background 1L is the one-loop correction to
b, D),
b considered as a function of both physical and auxiliary
F
the effective potential Veff (,
2 Our conventions are slightly different from those of Ref. [13]. For instance, we use the spacetime metric
), and the supersymmetric derivatives
g = diag(+1, 1, 1, 1), the Pauli matrices = (1, E ), = (1, E
D = / i , D = / + i .
106
scalar fields. In the case of renormalizable models, one-loop computations of this object
by either component or superfield techniques can be found, for instance, in Refs. [1519]
or Refs. [2024], respectively. Here we are interested in obtaining the first few terms in
the auxiliary field expansion of the effective potential, in the general non-renormalizable
theory defined above. Once the results of the one-loop computation have been cast in the
form (15), the functions 1w and 1K can be reconstructed from their derivatives 3 1wi ,
1Kj , 1Kj , and the coefficients 1a can be easily identified, too. Notice that the oneloop correction to the Khler potential can be reconstructed in two different ways, thanks
to background gauge invariance: this allows us to make a nontrivial consistency check. In
fact, we will proceed as follows. First, we will expand the Lagrangian in a background with
b and vanishing D
b (Section 4). Then we will compute the one-loop corrections
constant ,
F
b, from which 1w and 1K can be reconstructed
to the terms linear and quadratic in F
(Sections 5 and 6). In the final part (Section 7) we will instead consider a background
b and vanishing F
b. Then we will compute the one-loop corrections to the
with constant ,
D
b
terms linear in D. This will allow us to identify 1a and to derive 1K in a different way.
4. Quantum bilinears and propagators
b fields,
We take the background chiral superfields in the form (13), with constant ,
F
and plug the background-quantum splitting (6) in (2). At the same time, we take vanishing
background vector superfields. 4 In this case, the background-quantum splitting (7) reduces
b V in (2), Lgf in (8) reduces to (9), and Lgh is not
to the simple replacement V
relevant because ghosts do not interact with the (chiral) background. The Lagrangian is
then expanded and only the part bilinear in the quantum fields , V is retained, Lbil =
L + LV V + LV . Integration by parts is used whenever convenient, and some relations
that follow from (4) are used to rearrange the terms generated by the expansion of the
Khler potential. The terms that make up Lbil have a complicated dependence on and
b. Thus each of the three parts in Lbil can
a simple dependence (at most quadratic) on F
b)
bF
b)
b)
bF
b)
(0)
(F
(F
(0)
(F
(F
in turn be decomposed as L = L + L + L , LV V = LV V + LV V + LV V ,
(0)
b)
(F
bF
b)
(F
LV = LV + LV + LV , in a self-explanatory notation.
b, i.e., the quantum bilinears in a pure
First of all we list the terms that do not depend on F
constant background:

j
j
b
j + w
bij F i j 12 i j + h.c. , (16)
L(0)
= Kj 2 + F F + i

(0)
bab 1 ( Aa Aa ) Ab Ab + 1 D a D b + i a b
LV V = H
4
2

+b
Sab 12 Aa Ab + C a D b 12 C a 2C b + Ga Gb

a b 2i a b + h.c.
2
2
1
12 Aa + 12 D a 2C a + Ga Ga
3 Up to irrelevant constant terms in 1w or harmonic terms in 1K.

4 Background gauge invariance will not be exploited directly till Section 7.

+ i a + i a a i a ,

(0)
a K
bj j i Aa + D a 2C a
LV = T

i 2F j Ga + i 2 j a i a + h.c.
107
(17)
(18)
Most of the dependence on is left implicit. In particular, we have used the abbreviations:

,
bj Kj ,
(19)
w
bij wij (),
K

bab 1 fab ()
+ fab ,
H
2

a Kj ,
b Kj ,
(Tb )
(Ta )
b
j + T
j.
(20)
Sab T
bab and b
bj and w
bij (H
Sab ) have the meaning of dependent metric and
The matrices K
bj is hermitian and w
bij is symmetric, whereas
masses in the (V ) sector. Notice that K
b:
bab and b
Sab are real and symmetric. Next we list the terms that depend on F
H

b)
F
i k
k
bj 1 w
b
b F k 1 k + h.c.,
=F
L(
(21)
2 bikj + Kkj F + Kkj
2

b
bj 1 (H
bab )j a b + (b
Sab )j 12 a b i 2C a Gb + h.c.,
(22)
L(VFV) = F
2

b)
(F
bj K
bkj (Ta )
bk (Ta )k + K
k 2F C a + i 2 a i 2 Ga
LV = F
j

bkj k Ga + h.c.,
a K
(23)
i 2 T

bF
b)
(F
b F
bj K
b k ` ,
b k ` + 1 K
b k ` + K
=F
(24)
L
bF
b)
(F
LV V
bF
b)
F
L(V
j k`
j k`
j k `
b F
bj (b
=F
Sab )j C a C b ,

k a
b F
a ` K
bj 2 (Ta )`K
b + T
b
=F
`kj
`k j C + h.c.
(25)
(26)
Having completed the list of quantum bilinears in the chosen background, we could in
b, D
b =0), which corresponds to
F
principle compute the one-loop effective potential Veff (,
b external legs. In practice,
diagrams with an arbitrary number of (zero momentum) and F
b expansion of
as explained above, we are only interested in the first few terms in the F
the effective potential. This amounts to compute diagrams with an arbitrary number of
b external legs. In order to take into account
external legs and a small number of F
the full dependence, we will proceed as follows. In the remainder of this section, we
will consider the quantum bilinears in the pure constant background and compute the
-dressed
propagators for the quantum fields. In the next two sections, we will consider the
b and treat them as b-dependent interaction
quantum bilinears that also depend on F
and F
vertices, to be joined by the -dressed
quantum propagators. In particular, we will compute

b, which will allow us to reconstruct 1w
the -dressed
one- and two-point functions of F

and 1K, respectively.
The quantum propagators in a constant background are obtained by inverting the
quadratic forms that appear in (16)(18). We omit the details of the derivation and write
the results directly in momentum space. We use a compact matrix notation and denote
b w
b, b
b and
by K,
b, H
S the dependent matrices defined in (19) and (20). The matrices K
w
b should not be confused with the functions K and w used to define them by double
108
differentiation (the context should make this distinction clear, despite the slight abuse of
notation).
(0)
propagators for the components of quantum

From L , Eq. (16), we obtain the -dressed
chiral superfields:

1 i
i
b 2w
bK
b1T w
b
,
(27)
= i Kp

i
1
i
bK
b1T w
b 2w
b
,
(28)
F F = ip2 Kp

i j
1
ij
b 2w
bK
b1T w
b1T ,
bK
b w
(29)
F = i Kp

i
1
i
bK
b1T w
b 2w
b
,
(30)
= ip Kp

i j
1
ij
bK
b1T w
b1T .
b 2 w
bK
b w
(31)
= i Kp
propagators for the components of

From L(0)
V V , Eq. (17), we obtain the -dressed
quantum vector superfields:

1 ab
a b
p p b 2 b1 ab
p p 2
i 2
S
, (32)
p b
Hp S
A A = i g +
p2
p

1 ab
a b
S
,
(33)
G G = i p2 b

a b
1
ab
bp2 b
S
,
(34)
D D = ip2 H

a b
1
ab
bp2 b
S
,
(35)
D C = i H
1 ab
1 b 2 b1 ab
1
i 2 p2 b
S
,
Hp S
2
p
p
1 ab
a b
b 2
S
,
p b
= ip H

a b
1 ab
bp2 b
S
,
= i H
Ca Cb
=i
a b
p
= i
p2
1 ab
p
2
b
b
i
Hp S
p2
S
p2 b
(36)
(37)
(38)
1 ab
(39)
We have not yet taken into account the terms in L(0)

V , Eq. (18), which mix the
components of chiral and vector superfields. Such terms could be treated as insertions,
to be eventually resummed. This task becomes very easy if supersymmetric Landau
gauge is used, that is, if the special value = 0 of the gauge parameter is chosen. It is
well known that ordinary Landau gauge simplifies both the computation and the form
of the ordinary effective potential [25], because the vector field propagator is transverse
and annihilates mixed (scalarvector) terms. Similarly, supersymmetric Landau gauge
simplifies the computation of the effective Khler potential [3,5], because the vector
superfield propagator is supertransverse and annihilates mixed (chiralvector) terms.
So we will stick to this choice. 5 The nice properties of supersymmetric Landau gauge
are transferred to the component level, as they should. In particular, the mixed terms
5 We will not discuss the dependence of the one-loop Khler potential. For such a study in renormalizable
theories, where the dependence affects finite terms, see [5] (and also [4] for the case =1).
109
in (18) become irrelevant (in the constant background), so mixed V propagators are
not generated and the and V V propagators found above are not modified. Indeed, the
combinations of vector superfield components contained in (18) are the same that appear
in the gauge fixing Lagrangian (last line of (17)), and those components become nonpropagating for 0. As a cross-check, it is easy to verify explicitly that, for 0, the
mixed terms in (18) are annihilated by the vector supermultiplet propagators. 6
5. No corrections to the superpotential

b, at one-loop level. This
Our next computation is the -dressed
one-point function of F
bi in the effective Lagrangian (15), so we can verify whether
F
gives us the term 1wi ()
or not the superpotential receives a one-loop correction 1w. The quantum bilinears
b, Eqs. (21)(23), should be contracted with propagators, in order to
proportional to F
b tadpole. In supersymmetric Landau gauge, the required propagators are absent
close the F
b tadpole come from
for most of those terms. The only possible contributions to the F
the third and fourth terms in (21), which can be closed with propagators. However, the
corresponding contributions have equal magnitude and opposite sign, so they cancel each
other:
b
F
b
+ F
= 0.
(40)
b tadpole vanishes, which we interpret as 1w = 0. For generic , the

Thus the total F
same result is obtained as a consequence of more complicated cancellations among several
component diagrams involving both chiral and vector supermultiplets (we have checked
this on specific examples). The absence of one-loop corrections to the superpotential in the
general theory under study can be regarded as an explicit check of the well known nonrenormalization theorem [1] (see also [2628]). We recall that the literature contains some
examples which violate the theorem at the one- or two-loop level, due to infrared effects
associated to massless particles [2933,12]. We do not find such violations, because the
background field generates effective mass terms and thus acts as an effective infrared
regulator in field space [25,27].
6. Corrections to the Khler potential

We now move to the main computation, which is the -dressed
two-point function of
b, at one-loop level and vanishing external momenta, in supersymmetric Landau gauge.
F
b F
)
bj in the effective Lagrangian (15), so we
F
This will give us the term 1Kj (,
will obtain 1K. The contribution of each diagram to the effective Lagrangian is given
6 In more detail: F G terms do not contribute because now G has vanishing propagator; mixed terms of the type
A are annihilated by the A propagator as usual; similarly, the structure of propagators in the (D, C) and
(, ) sectors is such that the (D, C) and (, ) terms are annihilated, too.
110
R
explicitly, and the integration in (Minkowski) momentum space, d 4 p/(2)4 , is denoted
R
by p . Although the integrals have quadratic or logarithmic ultraviolet divergences, for
the time being we do not select a specific regularization. In this respect, notice that no
dangerous shifts in the loop momentum p are needed, since the diagrams are evaluated at
vanishing external momenta. Three classes of diagrams have to be considered: they involve
the contributions of multiplets only, V multiplets only, or both.
bF
b)
(F
(a) Pure loops. Using the first interaction term in L , Eq. (24), and the first and
b
F)
, Eq. (21), we obtain these contributions:
second ones (+ h.c.) in L(
b
F
b F
bj
=F
b
F

1
bj Kp
bK
b1T w
b 2w
i Tr K
b
,
(41)
b
F
b
F
1 b bj
F
= F
2

1

b K
b 1 w
b 2 w
bK
b1T w
b1 w
bT p2 w
i Tr Kp
b w
bK
bj ,
(42)

1
1
b 2w
bK
b1T w
bK
b1T w
b 2 w
b Kp
bj ,
ip2 Tr Kp
b K
b K
(43)
b
F
b
F
b F
bj
= F
b
F
b
F
b F
bj
= F

1
b 2w
bK
b1T w
b1T K
bT
i Tr Kp
b b
wK
bT p2 w
b1 b
K
bK
w
b
F
b F
bj
=F
1

b1 K
bj ,
w
bK
(44)
b
F

1

b 2 w
bK
b1T w
b1 w
b1 K
bT p2 w
b K
b 1 w
bj , (45)
i Tr Kp
b w
bK
bK
b
F
b F
bj
=F
b
F
Z
p

1
1

b Kp
bK
b 2 w
bK
b1T w
bK
b1T w
b1T w
b 2w
i Tr Kp
b K
b w
bj . (46)
111
b denote the matrices defined in (19), and w

bj , K
bj , . . .
We recall that here w
b and K
bj , K
denote derivatives of those matrices (i.e., third derivatives of w, third and fourth derivatives
of K). Two additional diagrams can be built, using the third and fourth terms (+ h.c.) in
b)
F
, Eq. (21). However, they cancel each other:
L(
b
F
b
F
b
+ F
b
F
= 0.
(47)
bb
(b) Pure V loops. Using the interaction terms in L(VFVF ) , Eq. (25), and L(VFV) , Eq. (22), we
obtain:
Z
C
C

1
i
b
bj
bp2 b
=F F
Tr b
Sj H
S
,
(48)
2
b
b
F
F
p
p
b
F
b
F
b
F
b
F
b
F
b F
bj
=F

1
1
b H
bj H
bp2 b
bp2 b
ip2 Tr H
S
S
H
,
(49)

1
1
i
b
bp2 b
bp2 b
Tr b
S H
S
S
Sj H
,
p2
(50)
b
F
b F
bj
=F
Z
p
b
F
b
F
b F
bj
= F
b F
bj
= F

1
1
bj H
bp2 b
bp2 b
H
i Tr b
S H
S
S
,
(51)

1
1
b
b H
bp2 b
bp2 b
Sj H
i Tr H
S
S
.
(52)
bj , H
bj , . . . denote derivatives of the matrices b
b
S and H
In the above expressions, b
Sj , H
defined in (20). Also, the trace operation does not include the trace in spinor space: the
latter has already been performed.
(c) Mixed V loops. Two nonvanishing diagrams can be built using the interaction
b)
(F
terms in LV , Eq. (23), but they cancel each other:
b
F
F
C
F
C
b
F
b
+ F
b
F
= 0.
(53)
This cancellation is a further effect of supersymmetric Landau gauge at the component

level. In superspace, the vanishing of mixed V loops should automatically follow from
mixed V vertices being annihilated by the supertransverse V propagator, as in the
renormalizable case [5].
b F
bj as a second
Now we have to sum the diagrams above and express the coefficient of F
derivative (with respect to and ). After some manipulations, the results for the and
V sectors can be cast in the required form:
112
b
F
b
F
b F
bj
=F

i 1
b + 1 Tr log Kp
b 2 w
bK
b1T w
Tr log K
b j ,
2
2
2
p
(54)

i
bp2 b
S j .
Tr log H
2
p
(55)
b
F
b
F
b F
bj
=F
Z
p
Finally, from the comparison of Eqs. (54) and (55) with Eq. (15), we can read off the
one-loop correction to the Khler potential:
Z

i 1
b + 1 Tr log Kp
b 2 w
bK
b1T w
=
Tr log K
b
1K ,
2
2
2
p
p

bp2 b
S .
Tr log H
(56)
We can go one step further and perform the momentum integration. If we do a Wick
rotation and regulate the momentum integral with a simple ultraviolet cutoff , the result
reads:
2

b log det H
b
= log det K
1K ,
2
16

M2
M2V
1
2
2
2
Tr
M
1
,
Tr
M
log
log
V
32 2
2
2
(57)
where M2 and M2V are field dependent mass matrices in the chiral and vector sectors:
b1/2 w
b1T w
b1/2 ,
bK
bK
M2 K
b1/2b
b1/2.
M2V H
SH
(58)
)
We recall that, in the above expressions, all the dependence on (,
is contained in the
b
b
b
matrices K, w
b, H , S defined in (19) and (20). The functional dependence of 1K is what
we were looking for. Indeed, if we recall that Eq. (15) was derived from Eq. (10), it is clear
)
that we can go back to superspace and replace the arguments (,
of 1K with general
b
2
V
in the case of a background gauge invariant
),
or even with (,
e )
superfields (,
quantization. So Eq. (57) is our final result: it gives the full one-loop correction to the
Khler potential in a closed form, for the general theory under study. The first line of
Eq. (57) contains quadratically divergent contributions, whereas the second line contains
logarithmically divergent and
If we evaluate the momentum integral
R
R finite contributions.
in d = 4 2 dimensions ( p = 2 d d p/(2)d ) instead of using a momentum cutoff
in d = 4, the first line of Eq. (57) should be omitted, and the replacement log 2
1/ + 1 + log(42 ) should be made in the second line. 7
In the special case of renormalizable theories, the superpotential is at most cubic and
bj = j = const.,
the metric is canonical in both the chiral and vector sectors, so K
k
2
b
b
a , Tb }.
In this limit, the first
w
bij = mij + hij k , Hab = ab /ga = const., Sab = {T
7 Notice that this regularization corresponds to dimensional reduction [34], since the spinor algebra has been
performed in d = 4.
113
line of (57) becomes irrelevant and the second line reproduces the result derived in [5],
if ga2 is identified with 2g 2 . Incidentally, we recall that the result of [5] was obtained
by computing two superdiagrams, i.e., one in each sector, after resumming insertions.
Our component field approach is not much more involved. Using -dressed
propagators
bF
b
amounts to resumming insertions, and our result for 1K originates from only three F
component diagrams in the renormalizable case, i.e. (42) in the chiral sector and (48), (50)
in the vector sector.
In the case of non-renormalizable theories, the quadratically divergent contributions in
the first line of (57) agree with the results of Ref. [6], obtained with superfield methods.
In our approach, those contributions can be traced back to the component diagrams (41),
(43) and (49). We have also tried to make a comparison with Ref. [79], in which detailed
computations of the one-loop bosonic effective action in general supergravity theories with
diagonal gauge kinetic function were presented, and the flat limit was also considered.
Only the divergent contributions were evaluated, and part of them was interpreted as
a correction to the Khler potential. A component field approach different from ours was
used. Here we have insisted on preserving supersymmetry and supersymmetric background
gauge invariance. In [79] the WessZumino gauge was used, and special emphasis
was given to ordinary background gauge invariance and scalar field reparameterization
covariance. We recall that the WessZumino gauge generally leads to a loss of manifest
supersymmetry, since vector supermultiplets are integrated out in a non-supersymmetric
way. To compensate for this, a special R -type gauge fixing for the component vector fields
was introduced in [79], with = 1. This particular prescription was argued to restore
supersymmetry, since the anomalous dimensions of component scalar fields were found
to coincide with the supersymmetric ones of the associated chiral superfields, in the flat
limit. Although this coincidence may be partly accidental, 8 the divergent part of the oneloop Khler potential reconstructed in [79] seems to agree with ours. Strictly speaking,
a slight difference can be found, for another reason. Indeed, the derivatives in w
bij are
reparameterization covariant ones in the formulae of [79] (see also [10]). However, this
apparent discrepancy is not a physical one: it depends on the way the background-quantum
splitting of chiral supermultiplets is performed. 9 If desired, our result could be made
reparameterization covariant a posteriori, e.g. by reinterpreting the derivatives in w
bij as
i
to a more general holomorphic Killing vector vai ().
covariant ones and promoting (Ta )

This could perhaps be confirmed by a supersymmetric normal coordinate expansion [6,36],
which however goes beyond the scope of the present paper.
8 For instance, extending that coincidence to the fermionic components of chiral superfields would require
some additional modification. As a further example of one-loop computations in the WessZumino gauge, we
may recall the component approach employed in [35] to study the divergences in a general renormalizable
theory. In this case ordinary Landau gauge was used for the component vector fields, and the scalar field
anomalous dimensions did not coincide with the chiral superfield ones. The latter were reconstructed from the
renormalization of superpotential parameters.
9 We recall that the perturbative computations and the resulting effective action depend on both that choice and
other ones, such as the background-quantum splitting of vector supermultiplets and the choice of the gauge fixing
function (or parameter). All such ambiguities are expected to disappear at the level of the physical S-matrix.
114
7. A consistency check
We conclude by presenting an alternative derivation of 1K. This derivation is based
on background gauge invariance, and the agreement of the final result with the one
found above provides an interesting consistency check. We recall that, although we have
presented our general framework in a background gauge invariant way, the latter property
has not been exploited in the previous section, where the functional form of 1K has been
b. Strictly speaking, what we obtained
computed by using a background with vanishing V
).
If the background-quantum splitting of vector superfields and the gauge
was 1K(,
fixing term are chosen as in (7) and (8), the result should be automatically promoted
e2Vb).
b,
We want to check this explicitly, so we switch on a nonvanishing V
to 1K(,
b field, as in Eq. (14). At the same time, we take
which we take to consist of a constant D
b. Then
the background superfield in the form (13), with constant but vanishing F
b
we compute the terms linear in D in the one-loop effective Lagrangian (or potential),
)(T
ba [1Kj (,
a )
j + 1a ]. Thus 1K can
which Eq. (15) predicts to have the form D
be reconstructed and compared to the previous result, and 1a can be identified as well.
In order to check all this, we first have to expand the Lagrangian in the new background
b
and find the terms bilinear in quantum fields (Lbil ) that have a linear dependence on D.
b
We omit the full list because only a few among such terms give a nonvanishing D tadpole,
b to a mixed V
in supersymmetric Landau gauge. For instance, terms that couple D
bilinear cannot contribute, due to the absence of mixed propagators. Also, terms that
b to ghost bilinears are independent and could at most contribute to the Fayet
couple D
Iliopoulos term, but the actual contribution is zero because ghosts belong to the adjoint
representation, which is vector-like. Special care is needed to study the effect of the gauge
ba to the components of
fixing Lagrangian (8), because it generates terms that couple D
a
b
c
quantum supermultiplets V , V with strength cbc / . Since this coefficient is divergent
in the limit 0, a small nonvanishing should be kept in intermediate steps, and the
b
c
mixed V terms in L(0)
V , Eq. (18), should be taken into account. When pure V V
b tadpole, the integrand is zero because the propagators
propagators are used to close the D
are bc-symmetric whereas the structure constants are antisymmetric. 10 If mixed V
insertions are used, at least two of them are needed to close the loop. However, since they
pick up only the dependent parts of the adjacent V propagators, the singular 1/ factor
is multiplied by a factor at least O( 2 ), so the result is again zero in the limit 0. After
completing the inspection of these and other terms, we find that the only terms that can
b tadpole are: 11
give a nonvanishing D

b
(D)
ba K
bkj (Ta )
bk (Ta )kj + K
L = D
k j ,
(59)

i
b
(D)
d
ba 1 (b
Sbc )j (Ta )
j C b C c + fbd cac
(D b C c b c ) + h.c.
(60)
LV V = D
2
2
10 A similar mechanism also kills other terms in the Lagrangian.
11 We remark that the background-quantum splitting of the vector superfield, Eq. (7), plays a crucial role here.
bV
b + V would not give the desired result, for a non-abelian gauge group.
The simple splitting V
115
b tadpole are, in matrix notation:

The contributions to the -dressed
D
b
D
ba
=D
b
D
C
C
b
D
D
C
b
D

1

b a +K
bj (Ta )
bK
b1T w
b 2 w
i Tr KT
j Kp
b
,
(61)
1 ba
= D
2

1
i
b
bp2 b
Tr H
S
j + h.c.,
Sj (Ta )
2
p
(62)
1 ba
= D
2

1
bp2 b
bj (Ta )
i Tr H
S H
j + h.c.,
(63)
b
D
ba
= D

1
bj (Ta )
bp2 b
i Tr H
S
j + h.c.
H
(64)
On writing the expressions in (63) and (64), we have used the identity (5). The expression
b = b
wj (Ta )
j
in (61) can be put in a more suitable form by using the relation w
bTa + TaT w
for the matrix w
b, which follows from the identity (3). After some manipulations, the
ba in the and V sectors can be cast in the form prescribed by Eq. (15):
coefficient of D
Z

i 1
b
D
a
j
b
b + 1 Tr log Kp
b 2 w
bK
b1T w
= D (Ta )
Tr log K
b j
2
2
2
p
p
Z
i
ba Tr Ta
+D
,
(65)
p2
p
Z

i
b
D
ba (Ta )
bp2 b
j
S j.
(66)
Tr log H
V =D
2
p
p
From the comparison of Eqs. (65) and (66) with Eq. (15), we can easily identify 1K
and 1a . The correction to the FayetIliopoulos coefficients is visible in the last term of
Eq. (65), and confirms a well-known result [37]:
Z
2
i
=
Tr Ta .
(67)
1a = Tr Ta
p2 16 2
p
The correction to the Khler potential can be read off from the first part of Eq. (65) and from
Eq. (66): the expression of 1K is identical to that obtained in Eq. (56) using a different
method. This completes our consistency check.
References
[1]
[2]
[3]
[4]
M.T. Grisaru, M. Rocek, W. Siegel, Nucl. Phys. B 159 (1979) 429.

I.L. Buchbinder, S. Kuzenko, Z. Yarevskaya, Nucl. Phys. B 411 (1994) 665.
B. de Wit, M.T. Grisaru, M. Rocek, Phys. Lett. B 374 (1996) 297.
A. Pickering, P. West, Phys. Lett. B 383 (1996) 54.
116
[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]
M.T. Grisaru, M. Rocek, R. von Unge, Phys. Lett. B 383 (1996) 415.
M.T. Grisaru, M. Rocek, A. Karlhede, Phys. Lett. B 120 (1983) 110.
M.K. Gaillard, V. Jain, Phys. Rev. D 49 (1994) 1951.
M.K. Gaillard, Phys. Lett. B 342 (1995) 125, Phys. Lett. B 347 (1995) 284.
M.K. Gaillard, V. Jain, K. Saririan, Phys. Lett. B 387 (1996) 520, Phys. Lett. D 55 (1997) 883.
J. Bagger, E. Poppitz, L. Randall, Nucl. Phys. B 455 (1995) 59.
T.E. Clark, S.T. Love, Phys. Lett. B 388 (1996) 577, Phys. Rev. D 56 (1997) 2461, Phys. Rev.
D 60 (1999) 025005.
I.L. Buchbinder, M. Cvetic, A.Y. Petrov, hep-th/9903243 and hep-th/9906141.
J. Wess, J. Bagger, Supersymmetry and Supergravity, 2nd edition, Princeton University Press,
Princeton, NJ, 1992.
S.J. Gates, M.T. Grisaru, M. Rocek, W. Siegel, Superspace, Benjamin/Cummings, Reading,
MA, 1983.
K. Fujikawa, W. Lang, Nucl. Phys. B 88 (1975) 77.
M. Huq, Phys. Rev. D 14 (1976) 3548.
L. ORaifeartaigh, G. Parravicini, Nucl. Phys. B 111 (1976) 516.
W. Lang, Nucl. Phys. B 114 (1976) 123.
R.D.C. Miller, Phys. Lett. B 124 (1983) 59, Nucl. Phys. B 229 (1983) 183.
M. Huq, Phys. Rev. D 16 (1977) 1733.
M.T. Grisaru, F. Riva, D. Zanon, Nucl. Phys. B 214 (1983) 465.
R.D.C. Miller, Nucl. Phys. B 228 (1983) 316.
P.P. Srivastava, Phys. Lett. B 132 (1983) 80;
F. Feruglio, J.A. Helayl-Neto, F. Legovini, Nucl. Phys. B 249 (1985) 533.
S. Coleman, E. Weinberg, Phys. Rev. D 7 (1973) 1888.
N. Seiberg, Phys. Lett. B 318 (1993) 469.
E. Poppitz, L. Randall, Phys. Lett. B 389 (1996) 280.
S. Weinberg, Phys. Rev. Lett. 80 (1998) 3702.
P. West, Phys. Lett. B 258 (1991) 375, Phys. Lett. B 261 (1991) 396.
I. Jack, D.R.T. Jones, P. West, Phys. Lett. B 258 (1991) 382.
D.C. Dunbar, I. Jack, D.R.T. Jones, Phys. Lett. B 261 (1991) 62.
I.L. Buchbinder, S.M. Kuzenko, A.Y. Petrov, Phys. Lett. B 321 (1994) 372.
I.L. Buchbinder, A.Y. Petrov, Phys. Lett. B 461 (1999) 209.
W. Siegel, Phys. Lett. B 84 (1979) 193.
R. Barbieri, S. Ferrara, L. Maiani, F. Palumbo, C.A. Savoy, Phys. Lett. B 115 (1982) 212.
L. Alvarez-Gaum, D.Z. Freedman, S. Mukhi, Ann. Phys. (NY) 134 (1981) 85.
E. Witten, Nucl. Phys. B 188 (1981) 513.
W. Fischler, H.P. Nilles, J. Polchinski, S. Raby, L. Susskind, Phys. Rev. Lett. 47 (1981) 757.

Scalar quartic couplings in type IIB supergravity

on AdS5 S 5
G. Arutyunov a,1,2 , S. Frolov b,1,3
a Sektion Physik, Universitt Mnchen, Theresienstr. 37, D-80333 Mnchen, Germany
b Department of Physics and Astronomy, University of Alabama, Box 870324, Tuscaloosa, AL 35487-0324, USA
Received 11 February 2000; revised 24 March 2000; accepted 29 March 2000
Abstract
All quartic couplings of scalar fields s I that are dual to extended chiral primary operators in N = 4
SYM4 are derived by using the covariant equations of motion for type IIB supergravity on AdS5 S 5 .
It is shown that despite some expectations if one keeps the structure of the cubic terms untouched, the
quartic action obtained contains terms with two and four derivatives. It is shown that the quartic action
vanishes on shell in the extremal case, e.g., k1 = k2 + k3 + k4 . Consistency of the truncation of the
quartic couplings to the massless multiplet of the N = 8, d = 5 supergravity is proven and the explicit
values of the couplings are found. It is argued that the consistency of the KK reduction implies nonrenormalization of n-point functions of n 1 operators dual to the fields from the massless multiplet
and one operator dual to a field from a massive multiplet. 2000 Elsevier Science B.V. All rights
reserved.
1. Introduction and summary

The AdS/CFT correspondence [13] provides a powerful method of studying correlation
functions in conformal field theories, in particular, in D = 4, N = 4 supersymmetric
YangMills theory (SYM4 ). According to the proposal by [2,3], the generating functional
of Green functions in SYM4 at large N and at strong t Hooft coupling coincides
with the on-shell value of the type IIB supergravity action on AdS5 S 5 . Thus, the
computation of an n-point Green function requires the knowledge of the supergravity
action up to the nth order. In particular, the quadratic [4] and cubic actions [57] for
physical fields of type IIB supergravity determine the normalization constants of twoand three-point Green functions [829]. In principle, by using the quadratic action [4]
and the covariant equations of motion for type IIB supergravity [3032] one can easily
1 On leave of absence from Steklov Mathematical Institute, Gubkin str. 8, GSP-1, 117966, Moscow, Russia.
2 arut@theorie.physik.uni-muenchen.de
3 frolov@bama.ua.edu
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 2 1 0 - 8
118
G. Arutyunov, S. Frolov / Nuclear Physics B 579 (2000) 117176
compute any three-point function of gauge invariant operators in SYM4 in the supergravity
approximation, the problem that can be hardly solved in perturbative SYM4 even at the
one-loop approximation.
The problem of computing four-point functions [3345] is obviously much more
involved, and consists in general of two independent parts one first has to derive
the relevant part of the supergravity action up to the fourth order, and then to find the
minimum of the action that amounts to computing the corresponding exchange and contact
Feynman diagrams. However, as was pointed out in [33], in the simplest cases of massless
modes of dilaton and axion fields, the relevant part of the supergravity action was known.
Computing the corresponding 4-point functions was initiated in [33], and completed in
[40]. Unfortunately, these modes correspond to rather complicated operators tr(F 2 + )
and tr(F F + ), and not much seems to be known about their four-point functions in
perturbative SYM4 . Nevertheless, the analysis of the four-point functions of these operators
performed in [45] allows one to conclude that at strong t Hooft coupling all operators with
large anomalous dimensions, which are dual to massive string states, decouple.
The important and simplest operators in SYM4 are single-trace 4 chiral primary
operators (CPOs) [5] that are of the form OkI = tr( (i1 ik ) ). It is well known that all
other operators in SYM4 corresponding to type IIB supergravity fields are descendents of
CPOs.
Type IIB supergravity on AdS5 S 5 contains in its particle spectrum [47,48] scalar
fields s I that are mixtures of the five form field strength on S 5 and the trace of the
graviton on S 5 . At the linear approximation to the supergravity, namely these scalars
correspond to CPOs, as one can see from their transformation properties with respect to
the superconformal group of SYM4 .
Although the correlation functions of CPOs are the simplest ones to compute in SYM4 ,
the corresponding calculation in the supergravity approximation is nontrivial due to the
absence of a relevant action for the scalars s I . The quadratic and cubic actions for the
scalars s I have been found and used to calculate all three-point functions of normalized
CPOs in [5]. 5
To compute four-point functions of CPOs one first has to know all cubic terms that
involve two scalar fields s I , and the s I -dependent quartic terms, and then to find the onshell value of the supergravity action.
In the present paper as the first step in this direction we determine all necessary cubic
and quartic terms by using the quadratic action [4] and the covariant equations of motion
for type IIB supergravity [3032]. It is clear that one can consider only the sector of type
IIB supergravity that depends on the graviton and the four-form potential. There are four
different types of cubic vertices describing interaction of two scalars s I with (i) symmetric
4 Throughout the paper we refer to a single-trace CPO as a CPO. Multi-trace operators may also belong to the
same short representation of the supersymmetry algebra as was shown in [46].
5 To get rid of higher-derivative terms in the equations of motion for scalars s I a derivative-dependent field
redefinition was made in [5]. By this reason the scalars s I used in [5] correspond not to CPOs but to extended
CPOs involving products of CPOs and their descendants [6]. However, three-point functions of the extended
CPOs coincide with the ones of CPOs for generic values of conformal dimensions of CPOs.
119
tensor fields coming from the AdS5 components of the graviton, (ii) with vector fields, (iii)
with scalar fields coming from the S 5 components of the graviton, and (iv) with scalar fields
t I that are mixtures of the trace of the graviton on the sphere and the five form field strength
on the sphere. Although all the cubic terms were recently found in [6,7], we will see that
the derivation done in the papers should be reconsidered. The reason is that this time, since
we are interested in cubic corrections to equations of motion for scalars s I , dealing with
quadratic terms, we have to take into account quadratic corrections to equations of motion
for the gravity fields.
Actually, it is straightforward, although cumbersome, to find cubic corrections to
equations of motion by decomposing the covariant equations of motion up to the third
order, and keeping only relevant terms. The main problem in deriving the quartic couplings
comes from the fact that the equations of motion such obtained are non-Lagrangian, and
one should perform a very complicated and fine analysis to reduce the equations of motion
to a Lagrangian form. In particular, although the original equations contain terms with six
derivatives, we will show that one can remove these terms completely by means of a chain
of field redefinitions, and by using nontrivial identities between spherical harmonics of
different types. Even after removing the terms with six derivatives, the resulting equations
containing terms with four and two derivatives (and without derivatives, of course) are
still non-Lagrangian. To make the equations Lagrangian one should again redefine the
scalar fields, and use the identities. Since any mistake in the computation would destroy
the possibility of obtaining Lagrangian equations of motion, we are pretty sure that the
quartic couplings we found are correct.
The fact that the covariant equations of motion are non-Lagrangian, and one has to
perform nontrivial field redefinitions to reduce them to a Lagrangian form, explicitly shows
that the gravity fields entering the covariant equations of motion cannot correspond to any
operator in SYM4 . The quartic action presented in the next section is in fact written for the
scalar fields s I corresponding not to CPOs, but to extended CPOs, as was discussed in [6].
In principle it seems possible to find an action for the scalars dual to CPOs by performing
the field redefinitions reversed to the ones used to reduce the equations of motion to a
Lagrangian form. 6 However, the resulting action for the new scalars will be much more
complicated and will contain higher-derivative terms with six derivatives. Its worth noting
that the equations of motion derived from the new action certainly differ from the original
ones despite the fact that one made reversed transformations.
We show that despite some expectations, if we keep the structure of the cubic terms
untouched, the action obtained contains quartic terms with two and four derivatives, and
there is no field redefinition allowing one to remove these terms. Thus, the problem of
computing the four-point functions of CPOs will require computing two new types of
Feynman diagrams: (i) exchange diagrams involving massive tensor fields of second rank,
and (ii) contact diagrams with four-derivative quartic vertices. All other necessary diagrams
were computed in [36,40,41].
6 Note that the reversed transformations should be made at the level of the quartic action, but not at the level of
equations of motion.
120
In our previous paper [6] we argued that quartic couplings of the scalars s I had to
vanish in the extremal case when, say, k1 = k2 + k3 + k4 . This conjecture was based on
the fact that all exchange Feynman diagrams vanished and contact Feynman diagrams had
singularity in the extremal case, thus nonvanishing quartic couplings would contradict to
the AdS/CFT correspondence. Although the vanishing of the quartic couplings obtained in
the present paper is not manifest, we show that this important property does take place after
an additional field redefinition. This means that 4-point extremal correlators of extended
CPOs vanish, and also implies the non-renormalization theorem [29] for the corresponding
extremal correlators of single-trace CPOs. It is clear that since the quartic couplings vanish
then there should exist such a representation of the quartic couplings, that makes the
vanishing explicit. We, however, have not looked for such a representation yet.
The quartic couplings we found allow us to study the problem of the consistency of
the KaluzaKlein (KK) reduction down to five dimensions. 7 It is customarily believed
that the S 5 compactification of type IIB supergravity admits a consistent truncation to the
massless multiplet, which can be identified with the field content of the gauged N = 8,
d = 5 supergravity [51,52]. Consistency means that there is no term linear in massive
KK modes in the untruncated supergravity action, so that all massive KK fields can
be put to zero without any contradiction with equations of motion. From the AdS/CFT
correspondence point of view the consistent truncation implies that any n-point correlation
function of n 1 operators dual to the fields from the massless multiplet and one operator
dual to a massive KK field vanishes because, as one can easily see there is no exchange
Feynman diagram in this case.
It is obvious that the cubic couplings found in [6,7] obey the consistency condition
allowing therefore truncation to the fields from the massless multiplet at the level of the
cubic action. In this paper we show that after an additional simple field redefinition the
quartic vertices we found indeed vanish when one of the four fields is not from the massless
multiplet, proving thereby the consistency of the reduction at the level of the quartic scalar
couplings. This in particular provides an additional argument that the scalars s I (and, in
general, any supergravity field) correspond not to CPOs but rather to extended CPOs.
Indeed, if we assume that the consistent truncation takes place at all orders in gravity
I
fields, we get that correlators of the form hO2I1 O2I2 O2n1 OkIn i vanish for k > 3. This
is certainly not the case for single-trace CPOs, and we are forced to conclude once more
that supergravity fields are in general dual to extended operators which are admixtures
of single-trace operators and multi-trace ones. 8 Since an extended operator is uniquely
determined by a single-trace one, it is natural to assume that if a correlation function of
extended operators vanishes then there exists a kind of a non-renormalization theorem for
an analogous correlation function of single-trace operators. If we further assume that type
IIB string theory on AdS5 S 5 respects the consistent truncation, then the vanishing of
n-point correlation functions of n 1 extended operators dual to the supergravity modes
7 For a recent discussion of the consistency problem see [49,50], and references therein.
8 Note that the lowest modes s may be dual only to single-trace CPOs. It is possible that any field from the
2
massless supergravity multiplet is dual to a single-trace operator.
121
from the massless multiplet, and one extended operator dual to a massive KK mode seems
to imply that:
At large N the n-point functions of the corresponding single-trace operators are
2 N.
independent of t Hooft coupling = gYM
If the consistent truncation is valid at quantum level, that seems to be plausible because
of a large amount of supersymmetry, then these n-point functions are independent of gYM
for any N .
In particular this conjecture is applied to n-point functions of n 1 CPOs O2 and a CPO
O4 . It would be interesting to check this in perturbation theory.
We also use the quartic couplings to find quartic action for the scalars s2I from the
massless multiplet. The 4-derivative terms vanish in this case, and we arrive at an action
with 2-derivative and non-derivative quartic couplings. We do not compare the action
obtained with the one of the gauged N = 8, d = 5 supergravity on the AdS5 background.
This problem will be considered together with the problem of computing 4-point functions
of CPOs O2 dual to the scalars from the massless multiplet in a latter paper.
The plan of the paper is as follows. In Section 2 we recall equations of motion for the
graviton and the four-form potential, introduce notations, and represent the action obtained.
In Section 3 we prove that there is the consistent reduction of type IIB supergravity on
AdS5 S 5 down to five dimensions at the level of the quartic action. In Section 4 we
show that the quartic couplings vanish in the extremal case. In Section 5 we discuss the
structure of the cubic corrections to the equations of motion due to the contributions of
the gravity fields. In Section 6 we explain what steps one should undertake to reduce the
equations of motion to a Lagrangian form. In conclusion we discuss unsolved problems. In
Appendix A we present the values of the quartic couplings obtained, and in the Appendix
B we summarize several important identities involving spherical harmonics of different
kinds.
2. Quartic Couplings
Quartic couplings of scalars s I may be derived from cubic corrections to the covariant
equations of motion [3032] for type IIB supergravity. Only the graviton and the four-form
potential give relevant contributions to cubic terms. The equations of motion of the metric
and the 4-form potential are
1
M ...M F M6 ...M10 ,
5! 1 10
1
= FMM1 ...M4 FNM1 ...M4 .
3!
FM1 ...M5 =
(2.1)
RMN
(2.2)
Here M, N, . . . , = 0, 1, . . . 9, and we use the following notation

FM1 ...M5 = 5[M1 AM2 ...M5 ] = M1 AM2 ...M5 + 4
terms,
i.e., all antisymmetrizations are with weight 1. The dual forms are defined as
122
G,
1
,
e01...9 =
G
M1 ...M10 = GM1 N1 GM10 N10 N1 ...N10 ,
1
1
(F )M1 ...Mk = M1 ...M10 F Mk+1 ...M10 = N1 ...N10 GM1 N1 GMk Nk FNk+1 ...N10 .
k!
k!
01...9 =
In the units in which the radius of S 5 is set to be unity, the AdS5 S 5 background solution
looks as

1
ds 2 = 2 dx02 + ij dx i dx j + d52 = gMN dx M dx N ,
x0
Rab = 4gab ,
Rabcd = gac gbd + gad gbc ,
R = 4g ,
R = g g g g ,
F = ,
Fabcde = abcde ,
(2.3)
where a, b, c, . . . , and , , , . . . , are the AdS and the sphere indices respectively and ij
is the 4-dimensional Minkowski metric. We represent the gravitational field and the 4-form
potential as
GMN = gMN + hMN ,
AMNP Q = A MNP Q + aMNP Q ,
F = F + f.
The gauge symmetry of the equations of motion allows one to impose the de Donder gauge:
1
h() h g h .
(2.4)
5
This gauge choice does not remove all the gauge symmetry of the theory, for a
detailed discussion of the residual symmetry see [47]. As was shown in [47], the gauge
condition (2.4) implies that the components of the 4-form potential of the form a and
aa can be represented as follows:
ha = h() = aM1 M2 M3 = 0,
a = b,
aa = a .
(2.5)
It is also convenient to introduce the dual 1- and 2-forms for aabcd and aabc :
aabcd = abcde Qe ,
aabc = abcde de .
(2.6)
Then the solution of the first-order self-duality equation can be written as

Qa = a b,
ab = [a b] .
(2.7)
To write down the action for scalars s I that can be used to compute 4-point correlation
functions of CPOs in SYM4 we need to expand fields in spherical harmonics 9
X
X
h (x, y) =
I1 (x)Y I1 (y),
b(x, y) =
bI1 (x)Y I1 (y),
X I
hab1 (x)Y I1 (y),
2 Y k = k(k + 4)Y k = f (k)Y k ,
hab (x, y) =
9 Here and in what follows we suppose that the spherical harmonics of all types are orthogonal with weight 1,
R
R I
R
J
Y()
= I J , and summation over , is assumed. Namely this
i.e., Y I Y J = I J , YI YJ = I J , Y()
normalization was used in [4].
123
X
X
ha (x, y) =
hIa5 (x)YI5 (y),
a (x, y) =
aI5 (x)YI5 (y),

2 4 Yk = (k + 1)(k + 3)Yk ,
X
k
I14
k
I14 (x)Y()
(y),
2 10 Y()
= (k 2 + 4k + 8)Y()
.
h() (x, y) =
We also need to make a number of fields redefinitions, the simplest ones required to
diagonalize the linear equations of motion are: 10
k = 10ksk + 10(k + 4)tk ,
bk = sk + tk ,
+ gab k + a b k ,
4
4
sk +
tk ,
k =
k+1
k+3
2k(k 1)
2(k + 4)(k + 5)
sk
tk ,
k =
k+1
k+3
Cak = hka + 4(k + 1)ak .
Aka = hka 4(k + 3)ak ,
hkab
k
= ab
(2.8)
(2.9)
(2.10)
(2.11)
(2.12)
Note that we use the off-shell shift of hab (2.9) that was used to find the quadratic action
for type IIB supergravity on AdS5 S 5 in [4]. It differs from the on-shell shift [47] by
higher-order terms.
The field redefinitions that are needed to make the equations of motion Lagrangian and
to remove higher-derivative terms from quadratic terms in the equations of motion will be
discussed in the next sections.
Then the action for the scalars s I may be written in the form
Z
4N 2
d 5 x ga
S(s) =
(2)5
L2 (s) + L2 (t) + L2 () + L2 (ab ) + L2 (Aa ) + L2 (Ca )
+ L3 (s) + L3 (t) + L3 () + L3 (ab ) + L3 (Aa ) + L3 (Ca )
(0)
(2)
(4)
+ L4 + L4 + L4 .
(2.13)
Here the quadratic terms are given by [4]

X 32k(k 1)(k + 2) 1
1 2 2
a
a sk sk m sk ,
(2.14)
L2 (s) =
k+1
2
2

X 32(k + 2)(k + 4)(k + 5)
1
1
a tk a tk m2t tk2 ,
(2.15)
L2 (t) =
k+3
2
2

X
1
1
a k a k f (k)k2 ,
(2.16)
L2 () =
4
4

X
1
1
1
1
k
k
c ab
c kab + a kab c cb
a cck b kba + c aak c bbk
L2 (ab ) =
4
2
2
4

k ab 1

2
1
+ 2 f (k) ab
k + 2 + f (k) aak
,
(2.17)
4
4
10 We often denote I1 as or as , and a similar notation for other fields.
k
1
124

X k+1 1
2 1
2
Fab (Ak ) m2A Aka
,
2(k + 2)
4
2

X k+3 1
2 1
2
Fab (C k ) m2C Cak
,
L2 (Ca ) =
2(k + 2)
4
2
L2 (Aa ) =
(2.18)
(2.19)
where the masses of the particles are

m2 = k(k 4),
m2A = k 2 1,
m2t = (k + 4)(k + 8),
m2 = m2 = f (k) = k(k + 4),
m2C = (k + 3)(k + 5)
and Fab (A) = a Ab b Aa .

The cubic terms were found in [57], and may be written as follows
L3 (s) = SI1 I2 I3 s I1 s I2 s I3 ,
2

2

27 12 1 12 4 1 2 3
,
SI1 I2 I3 = a123
3(k1 + 1)(k2 + 1)(k3 + 1)
L3 (t) = TI1 I2 I3 s I1 s I2 t I3 ,
a12327 ( + 4)(1 + 2)(2 + 2)3 (3 1)(3 2)(3 3)(3 4)
(k1 + 1)(k2 + 1)(k3 + 3)
I1 I2 I3
L3 () = I1 I2 I3 s s ,
4p123( + 2)
(3 1)(3 2),
I1 I2 I3 =
(k1 + 1)(k2 + 1)

I3
L3 (ab ) = GI1 I2 I3 a s I1 b s I2 ab

I I
1
1 2
a I1
I2
2
cI3
1
2
s a s + m1 + m2 f3 s s c ,
2
2
4( + 2)( + 4)3 (3 1)
a123,
GI1 I2 I3 =
(k1 + 1)(k2 + 1)
L3 (Aa ) = AI1 I2 I3 s I1 a s I2 AIa3 ,
2(k3 + 1)(3 1/2)( 1)( + 1)( + 3)
t123 ,
AI1 I2 I3 =
(k1 + 1)(k2 + 1)(k3 + 2)
TI1 I2 I3 =
(2.20)
(2.21)
(2.22)
(2.23)
(2.24)
L3 (Ca ) = CI1 I2 I3 s I1 a s I2 CaI3 ,

8(k3 + 3)(3 1/2)(3 3/2)(3 5/2)( + 3)
t123 .
(2.25)
CI1 I2 I3 =
(k1 + 1)(k2 + 1)(k3 + 2)
Here the summation over I1 , I2 , I3 is assumed, and we use the following notations:
1
1 = (k2 + k3 k1 ),
2
1
3 = (k1 + k2 k3 ),
2Z
a123 = Y I1 Y I2 Y I3 ,
Z
t123 = Y I1 Y I2 YI3 ,
1
2 = (k1 + k3 k2 ),
2
= k1 + k2 + k3 ,
Z
I3
p123 = Y I1 Y I2 Y()
,
125
and for any function fi f (ki ).

The quartic terms represent our main result and are given by
(0)
I1 I2 I3 I4
L(0)
4 = SI1 I2 I3 I4 s s s s ,

(2)
(2)
(2)
L4 = SI1 I2 I3 I4 + AI1 I2 I3 I4 s I1 a s I2 s I3 a s I4 ,
(2)
(2)
(2)
SI1 I2 I3 I4 = SI2 I1 I3 I4 = SI3 I4 I1 I2 ,
(2)
(2)
A(2)
I1 I2 I3 I4 = AI2 I1 I3 I4 = AI3 I4 I1 I2 ,

(4)
(4)
(4)
L4 = SI1 I2 I3 I4 + AI1 I2 I3 I4 s I1 a s I2 b2 (s I3 a s I4 ),
(4)
(4)
(4)
(2.26)
(2.27)
(2.28)
SI1 I2 I3 I4 = SI2 I1 I3 I4 = SI3 I4 I1 I2 ,
(2.29)
(4)
(4)
A(4)
I1 I2 I3 I4 = AI2 I1 I3 I4 = AI3 I4 I1 I2 .
(2.30)
The explicit values of the quartic couplings are collected in Appendix A.

3. Reduction to the gauged N = 8 five-dimensional supergravity
There is much evidence that the S 5 compactification of the IIB supergravity admits a
consistent truncation to the massless graviton multiplet, which can be identified with the
field content of the gauged N = 8, d = 5 supergravity [51,52]. Consistency means that
there is no term linear in massive KK modes in the untruncated action so that all massive
KK fields can be put to zero. As was noted in [49,50] the cubic couplings (2.20)(2.25)
obviously obey this condition allowing therefore the consistent truncation to the massless
gravity multiplet.
In this section we show that after an additional field redefinition the found quartic
vertices (2.26)(2.30) indeed vanish when one of the four fields is not from the massless
multiplet, proving thereby the consistency of the reduction at the level of the quartic scalar
couplings.
Recall that the gauged N = 8 five-dimensional supergravity has in particular 42 scalars
with 20 of them forming the singlet of the global invariance group SL(2, R). These 20
scalars comprise the 20 irrep. of SO(6) and correspond to the IIB supergravity fields 11 skI
with k = 2. The five-dimensional scalar Lagrangian consists of the kinetic energy and the
potential. The maximal number of derivatives appearing in the Lagrangian is two and that is
due to the non-linear sigma model type kinetic energy. We have however found the quartic
4-derivative vertices that cannot be shifted away by any field redefinition. Thus, a highly
nontrivial check of the relation between the compactification of the ten-dimensional theory
and the gauged supergravity in five dimensions as well of the results obtained involves
showing that the 4-derivative vertices vanish for the modes from the massless multiplet.
We start to analyse the consistency of the truncation with the quartic couplings of 4I
derivative vertices and assume the fields s2I2 , s23 , s2I4 belong to the massless multiplet.
(4)
(4)
Upon substituting k2 = k3 = k4 = 2 the couplings (A0 )I1 I2 ,I3 I4 and (A1 )I1 I2 ,I3 I4 turn to
11 In this section we use the explicit notation s I for a scalar transforming in I = (0, k, 0)-irrep.
k
126
zero. The other couplings are non-zero and, therefore, the only possibility is that their sum
should vanish. Moreover, according to the above discussion this vanishing should hold
regardless of the fact if the remaining field s1 is in a massive or in the massless graviton
multiplet.
Among the couplings we consider there is a distinguished one, namely, (At 2 )(4)
I1 I2 ,I3 I4
since it involves another type of SO(6) tensors. To deal with this coupling we note that
for k3 = k4 = 2 there exists only two values of k5 for which t345 does not vanish, namely,
k5 = 1 and k5 = 3. Then one represents

(3.1)
(f5 1)2 t125t345 = (f5 5)(f5 21) + 24(f5 1) 80 t125 t345 ,
so that (f5 5)(f5 21) vanishes for both k5 = 1 and k5 = 3. By using relations (B.8)
and (B.9) the remaining terms in the last formula may be now reduced to involve the same
type of the SO(6) tensors as the rest of the 4-derivative quartic couplings.
Finally, we sum up all couplings assuming that three fields are from massless multiplet
and the fourth one is skJ . The resulting expression L4 contains the tensor (aJ I4 I5 aI2 I3 I5
aJ I3 I5 aI2 I4 I5 ) as a multiplier 12, which for given three fields from the massless multiplet
restricts a number of possible fields skI to a finite number. Namely, k can be equal only to
2, 4, 6. The case k = 6 is the most simple one, since in this case the only value of k5 for
which the tensor does not vanish is 4. Thus, we can extend the summation index over the
whole set and use the fact that
X
X
X
aJ I2 I5 aI3 I4 I5 =
aJ I4 I5 aI2 I3 I5 =
aJ I3 I5 aI2 I4 I5 .
(3.2)
I5
I5
I5
Hence, for k = 6 the sum of the couplings vanishes.

If k = 4 then there are two possible values of k5 : k5 = 2, 4. Evaluating L4 for these
values of k5 we find
128 X
(aJ I4 I5 aI2 I3 I5 aJ I3 I5 aI2 I4 I5 ),
L4 =
3
k5
where sum is over k5 = 2 and k5 = 4. Thus, for k = 4 the sum L4 vanishes by virtue
of (3.2).
Finally we have k = 2 that allows for k5 three values: k5 = 0, 2, 4 and corresponds to the
case when all fields are from the massless multiplet. Substituting k = 2 we find
1 X
(aJ I4 I5 aI2 I3 I5 aJ I3 I5 aI2 I4 I5 )(k5 2)(k5 4)f5 (k5 + 6)(k5 + 8). (3.3)
L4 =
324
I5
For these values of k5 the r.h.s. here vanishes identically. Thus, we have shown that there
is no 4-derivative linear couplings of the massive fields with the massless ones and that the
4-derivative vertices are absent for fields from the massless multiplet.
The analysis of the couplings of the 2-derivative and non-derivative vertices proceeds in
the same manner. For the SO(6) tensors involving vector spherical harmonics one can use
the formula
12 We do not assume here a summation over the index I .
5
127

(f5 1)3 t125 t345 = (f5 + 23)(f5 5)(f5 21) + 496(f5 1) 1920 t125 t345 (3.4)
to make the nonreducible part vanishing when three of four fields are from the massless
multiplet. For the SO(6) tensors involving tensor spherical harmonics the corresponding
representations look as

(3.5)
f52 p125 p345 = (f5 12)2 + 24(f5 12) + 144 p125 p345 ,

3
2
f5 p125 p345 = f5 (f5 12) + 24f5 (f5 12) + 144(f5 12) + 1728 p125p345 (3.6)
and they are based on the fact that for k3 = k4 = 2 tensor p345 is nonzero only for k5 = 2.
The relevant part of the quartic Lagrangian involving two derivatives can be written as
follows
X

SJ I2 I3 I4 a skJ s2I2 a s2I3 s2I4
L2 =
I2 ,I3 ,I4

I
I
+ AJ I2 I3 I4 a skJ s2I2 skJ a s2I2 a s23 s2I4 s23 a s2I4 .
In this section and throughout the paper we often use the following notations
l = a125a345,
m = a145a235,
n = a135a245,
where we do not assume summation over the index 5.

Calculating the coefficients in L2 , we see that they have the structure
S = S l l + S m m + S n n,
A = An (n m).
Omitting total-derivative terms, taking into account the symmetry of the coefficients in
I3 , I4 , and using linear equations of motion, one can rewrite this Lagrangian in the form
X

aJ I2 I5 aI3 I4 I5 skJ 2SJl I2 I3 I4 + 4AnJ I2 I3 I4 s2I2 a s2I3 a s2I4
L2 =
I2 ,I3 ,I4

I
+ 2SJmI2 I3 I4 2SJn I2 I3 I4 4AnJ I2 I3 I4 a s2I2 s23 a s2I4
8SJ I2 I3 I4 skJ s2I2 s2I3 s2I4 .
I2 ,I3 ,I4
To remove the 2-derivative terms, we make the shift

skJ skJ +
1
1
I
I
JJ I2 I3 I4 s2I2 s23 s2I4 + LJ I2 I3 I4 s2I2 s23 s2I4 ,
k
k
where

1
J = S m + S n + 2An l,
2

1
n
l
m
n
L = A + 2S S S
l.
6
Computing L we see that L is completely symmetric in I2 , I3 , I4 for k = 4, 6. After the

shift all 2-derivative terms are removed and we get the following Lagrangian
X

I
8SJ I2 I3 I4 + 12 m2k (JJ I2 I3 I4 + LJ I2 I3 I4 ) skJ s2I2 s23 s2I4 .
L2 =
I2 ,I3 ,I4
128
This Lagrangian should be summed up with the non-derivative terms of the quartic
(0)
Lagrangian whose couplings will be denoted by SJ I2 I3 I4 . Thus the complete Lagrangian
for non-derivative terms is given by
X

(0)
I
4SJ I2 I3 I4 + 8SJ I2 I3 I4 + 12 m2k (JJ I2 I3 I4 + LJ I2 I3 I4 ) skJ s2I2 s23 s2I4 .
L0 =
I2 ,I3 ,I4
One can easily check that for k = 6 this Lagrangian vanishes, because in this case k5 can
be equal only to 4.
In the case of k = 4 there are two possible values of k5 : k5 = 2, 4, and we get at first
sight a nonzero result which has the form
X

X
aJ I2 I5 aI3 I4 I5 +
aJ I2 I5 aI3 I4 I5 skJ s2I2 s2I3 s2I4 .
L0 =
k(I5 )=2
k(I5 )=4
However, now we can use identity (B.5) to show that

X
8 X
aJ I2 I5 aI3 I4 I5 s4J s2I2 s2I3 s2I4 =
aJ I2 I5 aI3 I4 I5 s4J s2I2 s2I3 s2I4 .
7
k(I5 )=4
k(I5 )=2
Taking into account this relation we find that non-derivative Lagrangian L0 vanishes too.
Thus we have shown that at least at the level of the quartic action for scalars s I there is a
consistent dimensional reduction of the type IIB supergravity to the gauged supergravity on
the AdS5 background. We would like to stress again that this reduction requires non-trivial
redefinitions of the fields.
Now we can compute the quartic couplings for the case when all four fields are from the
massless multiplet. Summing up the nonvanishing quartic couplings of the two-derivative
vertices we find out that the answer contains the terms involving the SO(6) tensors of
the form aI1 I3 I5 aI2 I4 I5 and aI1 I4 I5 aI2 I3 I5 . Integrating by parts it is possible to convert the
tensor indices to the normal order, namely to aI1 I2 I5 aI3 I4 I5 This also leads to the additional
contributions to the quartic couplings of the non-derivative vertices. Recall that k(I5 ) runs
now the set 0, 2, 4.
Again it is useful to note that identity (B.5) implies a number of relations between the
Lagrangian terms involving tensors a125 a345 . In particular, when all the fields are from the
massless multiplet one finds the relation
X
aI1 I2 I5 aI3 I4 I5 s2I1 s2I2 s2I3 s2I4
k(I5 )=4
X
1 X
I
I
aI1 I2 I5 aI3 I4 I5 s2I1 s2I2 s23 s2I4 +
aI1 I2 I5 aI3 I4 I5 s2I1 s2I2 s23 s2I4 .
4
k(I5 )=2
k(I5 )=0
Actually there is no sum over k(I5 ) = 0 since in this case I5 is just the trivial representation.
I
Analogously, multiplying both sides of (B.5) by s2I1 a s2I2 s23 a s2I4 and then integrating by
parts and using the previous relation one obtains the following formula:
X
I
I I
I
aI1 I2 I5 aI3 I4 I5 s21 a s22 s23 a s24
k(I5 )=0,2,4
129
8 X
5 X
I
I
aI1 I2 I5 aI3 I4 I5 s2I1 s2I2 s23 s2I4 +
3
3
k(I5 )=0
k(I5 )=2
These relations allow one to exclude from the Lagrangian for the scalar fields from the
massless multiplet the contributions of the representations I5 with k(I5 ) = 4.
In this way we find the following values of the quartic couplings of the 2-derivative
vertex

52 29 X
aI1 I2 I5 aI3 I4 I5 a s2I1 s2I2 a s2I3 s2I4
L(2)
AdS5 = 27
k(I5 )=2
213
27

I
aI1 I2 I5 aI3 I4 I5 a s2I1 s2I2 a s23 s2I4
(3.7)
k(I5 )=0
and of the non-derivative vertex

52 211 X
(0)
I
LAdS5 =
9
(3.8)
k(I5 )=2
Note that the contribution of the trivial representation completely disappears from the nonderivative quartic coupling.
The quartic action can be further simplified by substituting the integrals of spherical
harmonics for their explicit value. By using (B.12) one gets
X
25 3 X I1 I2 I5 I3 I4 I5
aI1 I2 I5 aI3 I4 I5 = 2 3
C
C
,
5
k(I5 )=2
I5
where C I1 I2 I3 = hC I1 C I2 C I3 i. One can easily establish the following summation formula

X
1
1
1
I
CijI Ckl
= ik j l + il j k ij kl
2
2
6
I
that steams from the fact that the l.h.s. of the expression above is a fourth rank tensor
of SO(6), symmetric and traceless both in (ij ) and (kl) indices with the normalization
condition CijI CijI = 20. Applying this formula one gets

X
24 3
1 I1 I2 I3 I4
I1 I2 I3 I4
I1 I2 I4 I3
aI1 I2 I5 aI3 I4 I5 = 2 3 C
+C

,
(3.9)
5
3
k(I5 )=2
where the shorthand notation C I1 I2 I3 I4 = CiI11i2 CiI22i3 CiI33i4 CiI44i1 for the trace product of four
matrices C I was introduced. It remains to note that for k5 = 0 the normalization condition
P
for scalar spherical harmonics gives k(I5 )=0 aI1 I2 I5 aI3 I4 I5 = 13 I1 I2 I3 I4 . By exploiting
this formula together with (3.9) the two-derivative Lagrangian may be reduced to the
following form:
(2)
LAdS5 =

214
I
CI1 I2 I3 I4 a s2I1 s2I2 a s23 s2I4 .
3
9
(3.10)
One can also introduce the fields sij CijI s I that provide the natural parametrization of
the coset space SL(6, R)/SO(6). Although it is clear that namely this form of the action
130
should be compared to the one of the gauged N = 8, d = 5 supergravity, we will not do

this here. We only note that the comparison requires to perform the field redefinitions of
the type s I1 s I1 + jI1 I2 I3 I4 s I2 s I3 s I4 to convert the Lagrangian terms with two derivatives
to the form (3.10).
4. Quartic couplings in the extremal case

In our previous paper [6] we conjectured that quartic couplings of scalars s I vanish in
the extremal case when, say, k1 = k2 + k3 + k4 . In this section we show that the quartic
couplings we found do satisfy the property after an additional shift of the fields. In principle
by using the shift one can find such a representation of the quartic couplings, that makes
the vanishing explicit.
The vanishing of the quartic couplings means that correlation functions of extended
CPOs vanish in the extremal case [6] and also implies the non-renormalization theorem
[29] for the corresponding extremal correlators of single-trace CPOs.
To prove the vanishing we find convenient to use different 4-derivative vertices. Namely,
one can easily show that the following relations are valid on-shell
Z
(4)
A1234 s1 a s2 b2 (s3 a s4 )
Z
Z
(4)
(4)
a
b
= 2A1234 a s1 b s2 s3 s4 4A1234 s1 a s2 s3 a s4
Z
2

1
2
2
2
A(4)
m
m
m
s1 s2 s3 s4 .
2
3
4
4 1234 1
Z
(4)
s1 a s2 b2 (s3 a s4 )
S1234
Z

(4)
(4)
= S1234 a s1 a s2 b s3 b s4 + S1234 m21 + m22 + m23 + m24 4
Z
Z

1 (4)
(4.1)
s1 s2 s3 s4 .
s1 a s2 s3 a s4 + S1234 m21 + m22 m23 + m24
4
Thus we replace all 4-derivative vertices by the ones with only one derivative on each field.
This representation has also the advantage that in this case the Hamiltonian reformulation
of the quartic action is straightforward, and, therefore, as was shown in [16], there is no
need to add boundary terms.
We assume for definiteness that k1 = k2 + k3 + k4 . It is easy to show, by using the
description of spherical harmonics as restrictions of functions, vectors and tensors on
the R6 in which the sphere S 5 is embedded [5,6], that the tensors (we do not assume
summation over I5 here)
t125 t345
and p125 p345
vanish in the extremal case, and that the tensors a125a345 , a135a245 and a145a235 differ from
zero only if k5 = k3 + k4 , k5 = k2 + k4 , k5 = k2 + k3 respectively. Thus in all vertices we
can replace k5 by a corresponding function of k2 , k3 , k4 , and, then the only dependence
131
on k5 is in tensors a125 a345 , a135 a245 and a145a235 which are obviously symmetric in 2,
3, 4.
We single out the field s I1 and write the relevant part of the quartic 4-derivative vertices
in the form
X

(4)
(4)
(4)
(4)
SI1 I2 I3 I4 AI1 I3 I2 I4 + AI1 I4 I3 I2 a s I1 a s I2 b s I3 b s I4 ,
Lext = 4
I2 ,I3 ,I4
where we sum over the representations satisfying the extremality condition. Now, we
substitute the values of k5 discussed above, and k1 = k2 + k3 + k4 in the quartic couplings,
and obtain zero.
To analyze 2-derivative terms we represent the 2-derivative Lagrangian as follows

X 1 (2)
(2)
(2)
e
e
SI1 I3 I2 I4 + AI1 I2 I3 I4 s I1 a s I2 s I3 a s I4
Lext = 4
2
I2 ,I3 ,I4

I I I I
1 e(2)
(2)
2
2
2
2
1s 2s 3s 4 ,
e
+ A
m
+
m
m
S
m
s
3
4
3
I1 I2 I3 I4
4 I1 I2 I3 I4 4
where using (4.1) we define

(2)
(2)
(4)
e
SI1 I2 I3 I4 = SI1 I2 I3 I4 + SI1 I2 I3 I4 m21 + m22 + m23 + m24 4 ,
(2)
(4)
e(2)
A
I1 I2 I3 I4 = AI1 I2 I3 I4 4AI1 I2 I3 I4 .
This time substituting k5 and k1 and symmetrizing the expression obtained in I2 and I4 ,
we get a non-zero function which is, however, completely symmetric in I2 , I3 and I4 . Thus
we can remove the 2-derivative term by using the shift

2
1 e(2)
(2)
I1
I1
e
SI1 I3 I2 I4 + AI1 I2 I3 I4 s I2 s I3 s I4 .
s s
31
2
This shift also produces an additional contribution to the non-derivative terms which is
equal to

1 (2)
2
e(2)
SI1 I3 I2 I4 + A
m22 + m23 + m24 m21 s I1 s I2 s I3 s I4 .
e
I1 I2 I3 I4
3
2
After accounting this contribution the non-derivative terms acquire the form

X (0)
1 e(2)
1
(0)
(2)
e
SI1 I3 I2 I4 + AI1 I2 I3 I4
SI1 I2 I3 I4
Lext = 4
6
2
I2 ,I3 ,I4
m22 + m23 + m24 m21

+

1 e(2)
(2)
AI1 I2 I3 I4 m24 m23 e
SI1 I2 I3 I4 m24 + m23
4

1 (4)
+ S1234 m21 + m22 m23 + m24
4
132

1 (4)
A1234 m21 m22 m23 m24 s I1 s I2 s I3 s I4 .
4
Substituting k5 and k1 and symmetrizing the coefficient obtained in I2 , I3 and I4 we end
up with zero.
5. Equations of motion
The equations of motion that follow from the action (2.13) have the form

1 a2 m21 s1
= 3S125 s2 s5 2T125s2 t5 2125 s2 5

1 2
c
a
b
5
a
c
2
+ G125 2 s2 ab a s2 c5 + m1 + m2 f5 s2 c5
2

S4
,
A125 2a s2 Aa5 + s2 a Aa5 C125 2a s2 C5a + s2 a C5a
s1

32(k5 + 2)(k5 + 4)(k5 + 5) 2
a t5 m2t t5 = T125s1 s2 ,
k5 + 3
2
2
a 5 m 5 = 2125 s1 s2 ,
Eqab ()

1
1
1
1 2 5
c 5
c 5
5
f5 1 ab
c ab + a cb + b ca +
2
2
2
2

1
1
1
1
cd
c
2 d
c
gab c d 5 a b c5 + gab c d5 gab f5 + 1 c5
2
2
2
2

1
1
c
2
2
= G125 a s1 b s2 gab c s1 s2 gab m1 + m2 f5 s1 s2 ,
2
4

k5 + 1
2 A5 b a A5b m2A A5a = A125s1 a s2 ,
2(k5 + 2) b a

k5 + 3
b2 Ca5 b a Cb5 m2C Ca5 = C125 s1 a s2 .
2(k5 + 2)
(5.1)
(5.2)
(5.3)
(5.4)
(5.5)
(5.6)
(5.7)
Here 32k(k 1)(k + 2)/(k + 1), the summation over 2 and 5 is assumed, and the
masses of all fields except s depend on k5 .
To obtain these equations of motion from the covariant equations (2.1) and (2.2) we
first need to decompose them up to the third order, and then to perform a number
of fields redefinitions to make the equations Lagrangian. It is convenient to begin by
considering quadratic terms in the covariant equations because as we will see they also
give contributions to cubic terms. It is also useful to single out contributions coming from
different fields.
133
5.1. Contribution of scalars s

We begin by considering the contribution of the scalars s coming from quadratic terms in
the equations of motion for s that are obtained from the covariant equations (2.1) and (2.2).
Decomposing these equations up to the second order in fields, and keeping only terms
quadratic in s, one can represent the equation for s in the following form 13

s
s
s
s2 s5 + E125
a s2 a s5 + F125
a b s2 a b s5
a2 m21 s1 = D125

s
s
+ R125
s2 b2 m25 s5 + T125
a s2 a b2 m25 s5 .
(5.8)
We see that the r.h.s of (5.8) contains terms proportional to linear part of the equations of
motion: (b2 m25 )s5 . Although such terms do not give contributions to quadratic terms,
and by this reason were neglected in [5], they do contribute to cubic terms. To remove the
higher-derivative terms on the first line of (5.8) one should make the field redefinition [5]
s
s20 s50 + Ls125 a s20 a s50 ,
s1 = s 0 1 + J125
where
s
,
2Ls125 = F125
(5.9)

s
s
2J125
+ Ls125 m22 + m25 m21 8 = E125
.
Then (5.8) takes the form

s
s20 s50 + cubic terms,
a2 m21 s10 = V125
where

3
s
s
s
= D125
J125
m22 + m25 m21 = S125 .
V125
1
(5.10)
To simplify the form of the cubic terms we make an additional shift 14

s
s
s1 s1 + F125
a s2 a J534
s3 s4 + Ls534 a s3 a s4
s
s
s
J534
s2 s3 s4 + 2J125
Ls534 s2 a s3 a s4
+ 2J125
and represent the equation in the form

a2 m21 s1
s
= V123
s2 s3 + (s0)1234s2 s3 s4 + (s2a)1234 a s2 s3 a s4
+ (s2b)1234s2 a s3 a s4 + (s4a)1234 a s2 b s3 a b s4
+ (s4b)1234s2 b b s3 a b s4 + (s6a)1234 a s2 b c s3 a b c s4 ,
(5.11)
where the coefficients are given by

s
s
s
s
s
J534
+ R125
2J125
D534 ,
(s0)1234 = 2V125
s
s
s
(s2a)1234 = 2 T125 2L125 D534 ,
13 We do not present the explicit values of the coefficients here and below because they are pretty complicated
and not very instructive.
14 Here and in what follows we omit the primes on redefined fields.
134
s
s
s
s
(s2b)1234 = 2V125
Ls534 + R125
2J125
E534,
s
s
s
(s4a)1234 = 2 T125 2L125 E534 ,
s
s
s
2J125
F534 ,
(s4b)1234 = R125
s
s
s
.
(s6a)1234 = 2 T125 2L125 F534
(5.12)
The coefficients s1234 in this equation depend on SO(6) tensors of the form
k5n
a125 a345,
k5 (k5 1)(k5 + 1)(k5 + 2)
n > 0.
5.2. Contribution of scalars t

To obtain the contribution of the scalars t to cubic terms in the equations of motion for s
we need to decompose the covariant equations (2.1) and (2.2) up to the second order in
fields, and to keep the terms of the form st in the equation for s and terms quadratic in s in
the equation for t. We represent the equations for s and t in the following form

t
t
t
s2 t5 + N125
a s2 a t5 + P125
a b s2 a b t5
a2 m21 s1 = K125

t
t
+ R125
s2 b2 m2t t5 + T125
a s2 a b2 m2t t5 ,
(5.13)

t
t
t
s3 s4 + E345
a s3 a s4 + F345
a b s3 a b s4 .
(5.14)
a2 m2t t5 = D345
To get rid of the higher-derivative terms in (5.13) we perform the following redefinition of
the fields s:
s1 s1 + Jts125s2 t5 + Lst125a s2 a t5 ,
where
t
,
2Lst125 = P125

t
2Jts125 + Lst125 m22 + m2t m21 8 = N125
.
Then Eq. (5.13) takes the form

t
Jts125 s2 b2 m2t t5
a2 m21 s1 = Vts125s2 t5 + R125

t
+ T125
Lst125 a s2 a b2 m2t t5 ,
(5.15)
where

2
t
Jts125 m22 + m2t m21 = T125 .
Vts125 = K125
1
(5.16)
To take into account the terms proportional to the linear part of the equation for t, one
should use Eq. (5.14). We also need to reduce Eq. (5.14) to the canonical form (5.2). To
this end we perform the shift of the field t [6]
t
s3 s4 + Lt345 a s3 a s4 .
t5 t5 + J345
This shift removes all terms with derivatives, and we end up with Eq. (5.2). Finally we
make the same shift of t in Eq. (5.15), and represent the equation in the form
135

a2 m21 s1
= Vts125s2 t5 + (t0)1234s2 s3 s4 + (t2a)1234 a s2 s3 a s4
+ (t2b)1234s2 a s3 a s4 + (t4a)1234 a s2 b s3 a b s4
+ (t4b)1234s2 b b s3 a b s4 + (t6a)1234 a s2 b c s3 a b c s4 ,
(5.17)
where
t
t
t
+ R125
Jts125 D345
,
(t0)1234 = Vts125J345
t
t
Lst125 D345
,
(t2a)1234 = 2 T125
t
t
Jts125 E345
,
(t2b)1234 = Vts125Lt345 + R125
t
t
s
(t4a)1234 = 2 T125 Lt 125 E345 ,
t
t
Jts125 F345
,
(t4b)1234 = R125
t
t
s
(t6a)1234 = 2 T125 Lt 125 F345 .
(5.18)
The coefficients t1234 in this equation depend on SO(6) tensors of the form
k5n
a125a345 ,
(k5 + 2)(k5 + 3)(k5 + 4)(k5 + 5)
n > 0.
Summing up the contributions of scalars s I and t I we find that the resulting coefficients
only depend on the following SO(6) tensors:
f5n a125 a345,
1
a125 a345,
f5 + 3
n > 1,
1
a125a345.
f5 5
5.3. Contribution of scalars

Decomposing the covariant equations (2.1) and (2.2) up to the second order in fields,
and keeping the terms of the form s in the equation for s and terms quadratic in s in the
equation for , we represent the equations for s and as follows
a2 m21 s1 = K125 s2 5 + N125 a s2 a 5 + P125 a b s2 a b 5

+ R125 s2 a2 m2 5 ,
(5.19)
2
2
a
a b
(5.20)
a m 5 = D345 s3 s4 + E345 a s3 s4 + F345 a b s3 s4 .
Following the same steps as above, we make the following field redefinitions to remove
higher-derivative terms, and to reduce (5.20) to the canonical form (5.3)
s
s2 5 + Ls125 a s2 a 5 ,
s1 s1 + J125
5 5 + J345 s3 s4 + L345 a s3 a s4 ,
(5.21)
where
2(4 + k32 + 4k3k4 + k42 f5 )
p345 ,
(k3 + 1)(k4 + 1)
4p345
.
L345 =
(k3 + 1)(k4 + 1)
J345 =
(5.22)
136
Then Eq. (5.19) takes the form

a2 m21 s1
s
= V125
s2 3 + (0)1234s2 s3 s4 + (2a)1234 a s2 s3 a s4
+ (2b)1234s2 a s3 a s4 + (4a)1234 a s2 b s3 a b s4
+ (4b)1234s2 b b s3 a b s4 + (6a)1234 a s2 b c s3 a b c s4 ,
(5.23)
where

s
s
J345 + R125 J125
D345 ,
(0)1234 = V125
(2a)1234 = 2Ls125 D345 ,
s
s
L345 + R125 J125
E345,
(2b)1234 = V125
(4a)1234 = 2Ls125 E345 ,

s
(4b)1234 = R125 J125
F345 ,
(6a)1234 = 2Ls125 F345 .
(5.24)
The coefficients 1234 in this equation depend on SO(6) tensors of the form
f5n p125 p345 ,
n = 0, 1, 2, 3.
The surprizing fact is that this equation is Lagrangian. Namely, it can be derived from the
following Lagrangian:
L = L2 (s)

1
+ a J125 s1 s2 + L125 b s1 b s2 a J345 s3 s4 + L345 c s3 c s4

4

1
+ f5 J125 s1 s2 + L125 b s1 b s2 J345 s3 s4 + L345 c s3 c s4

4
+ 125 s1 s2 J345 s3 s4 + L345 b s3 b s4 ,
(5.25)
where J125 and L125 are given by (5.22). One can easily see that if one makes a redefinition
of inverse to (5.21): 5 5 J125 s1 s2 L125 a s1 a s2 , then the quartic terms in (5.25)

will be removed from the action, but we obtain additional cubic higher-derivative terms.
5.4. Contribution of massive gravitons
We loosely refer to symmetric tensor fields coming from the AdS5 components of the
graviton as massive gravitons. To account for the massive graviton contribution we first
need to derive equations of motion for the massive gravitons. In principle to obtain these
equations one should consider the Einstein equations (2.2) not only with the indices (a, b),
but also with the indices (a, ) and (, ). The reason is that the equations for a ab
and aa that are constraints and, therefore should be a consequence of a true equation, do
not follow from (2.2) if one restricts oneself by considering only indices (a, b). To find the
137
true equation for the massive gravitons, it is convenient to replace (2.2) by the following
equivalent equation

1
1
1
M2 ...M4
M1 ...M5
FMM2 ...M4 FN
gab FM1 ...M5 F
.
(5.26)
RMN gMN R =
2
3!
5
An important property of the equation is that after performing the off-shell shift of hab
(2.9), its linear part coincides with the linear part of (5.5)

(1)

1
1
1
= Eqab ().
(5.27)
Fa Fb gab FM1 ...M5 F M1 ...M5
Rab gab R
2
3!
5
Decomposing Eq. (5.26) up to the second order in fields, and keeping the terms quadratic
in s, we can represent the equations for ab in the following form
Eqab () + 123 a s1 b s2 + 123c a s1 c b s2

+ 123 c d a s1 c d b s2 + 123 a (b s1 s2 ) + b (a s1 s2 )

+ 123 a ( c s1 b c s2 ) + b ( c s1 a c s2 )

+ 123 a b (s1 s2 ) + 123 a b c d s1 c d s2

+ 123 a c d s1 c d b s2 + b c d s1 c d a s2 + gab C
= 0.
(5.28)
Here C denotes the following contribution:

1
2
3
s1 s2 + T123
a s1 a s2 + T123
a b s1 a b s2
C = T123
1
L123 a b c s1 a b c s2 .
2
(5.29)
To remove the higher-derivative and total-derivative terms we perform the following shift
of the massive gravitons:
3
= 0 ab + a b3 + b a3 + gab 3 + J123 a s1 b s2 + L123 c a s1 c b s2 . (5.30)
ab
3
Here J123 and L123 depend on the coefficients , , as follows:

L123 = 123,

1
J123 = 123 L123 m21 + m22 f3 18
2
and the cubic vertex G123 is expressed through them as

1
G123 = 123 + J123 m21 + m22 f3 6 L123 m21 + m22 .
2
To get rid of the total derivative terms with the coefficients , , , , one also has to
impose the following relation
3
1
f3 a3 a 3 + U123 s2 a s1 + H123 c s1 c a s2 = 0,
2
4
where
(5.31)
138

1 2
m2 J123 3m22 L123 + 2123 + 2123 + 2m22 123 ,
2

1
H123 = m21 3 L123 + 123 123 .
2
Thus only the coefficient has not been fixed yet. Actually, a change of the coefficient
(with the simultaneous change of according to (5.31)) results only in a change of the
interaction of the trace aa of the massive graviton with the scalars s. In particular, one
can choose in such a way that only traceless part of a massive graviton interacts with
the scalars s. However, this choice leads to the appearance of quartic couplings with 6
derivatives. Terms with 6 derivatives are absent only if we choose the cubic vertex as in
Eq. (2.23). This vertex is a natural generalization of the interaction vertex of a massless
graviton with scalar fields. To determine we take the trace of Eq. (5.28) and represent the
resulting equation as
U123 =

3
a b ab a cc 2(f + 3)cc + 123 a s1 a s2
2
+ 123 a b s1 a b s2 + 123 a b c s1 a b c s2 + 123 s1 s2 = 0.
(5.32)
Then, by requiring that after the shift (5.30) Eq. (5.32) coincides with the trace of (5.5) and
assuming that has the form
3 = A123 s1 s2 + B123 a s1 a s2 + C123 a b s1 a b s2 ,
we find the following relations

3
33
J123
+ 2f3 L123 + 123 10(f3 + 3)C123 = 0,
(5.33)
8H123 +
2
2

3
8H123 m22 4 + 8U123 2(f3 + 6)J123 + 6 m21 4 m22 4 L123
2
3
+ 123 10(f3 + 3)B123 = G123 ,
2
3 2 2
2
8U123m1 + m1 m2 (3L123 J123 ) + 123 10(f3 + 3)A123
2

3 2
= m1 + m22 f3 G123 .
4
In particular, one can show that C123 = 0.
To find the massive graviton contribution to the equations for s we also need to know
0 ) and a ( 0 a ). The simplest way to derive the
equations of motion for b ab ( b ab
a
a
0 satisfies Eq. (5.5), and then to use the graviton
equations is first to take into account that ab
redefinition (5.30) to find a ab and aa . Differentiating and taking the trace of (5.5) one
can easily obtain

2 1

1
3G123
5
a
m21 m22 f3 m21 + m22 + f32 s1 s2 ,
0 a3 =
2f3 (f3 + 3)
4
6
12

G
123
3
b
m22 m21 + f3 a s1 s2 + a 0 b3 .
b 0 ab =
f3
139
These equations also explain why we took the interaction vertex of massive gravitons in
the form (2.23), namely, there are no terms with derivatives in the r.h.s. of the equation
for 0 aa under this choice.
Now we can proceed with the massive graviton contribution to the equations for s.
To find the equations we should decompose the covariant equations of motion up to the
second order, make the shift (2.9), and keep only the terms of the form s. To simplify
the consideration we also find convenient not to shift the trace of the original gravitons haa
(however we do not decompose hab in the sum of a traceless tensor and a trace) but to
take into account the contribution of haa later. This can be easily done because the equation
of motion for haa follows from the Einstein equation (2.2) with indices (, ). Then the
equations for s take the form

g
g
g
3
3
+ R123 a s2 b ab
+ T123 s2 aa3 ,
(5.34)
a2 m21 s1 = V123 a b s2 ab
g
where V123 = 21 G123 .

Finally, substituting the massive graviton redefinition (5.30) in (5.34), and performing
the following redefinition of s to simplify the equation
g
s1 s1 + V123 b s2 b3 ,
we represent the contribution of the massive gravitons in the form

a2 m21 s1

1 a
1 2
0c
g
a
b
05
0c
2
= V125 s2 ab a s2 c5 + m1 + m2 f5 s2 c5
2
4
+ (g0)1234 s2 s3 s4 + (g2a)1234 a s2 s3 a s4 + (g2b)1234s2 a s3 a s4
+ (g4a)1234 a s2 b s3 a b s4 + (g4b)1234s2 a b s3 a b s4
+ (g4c)1234 a b s2 a s3 b s4 + (g6a)1234 a s2 b c s3 a b c s4
+ (g6c)1234 a b s2 c a s3 c b s4 .
(5.35)
The coefficients g1234 in this equation depend on SO(6) tensors of the form
f5n a125 a345,
n > 1,
1
a125 a345.
f5 + 3
5.5. Contribution of the trace of massive gravitons

It is known [47] that at linear order the graviton trace haa is equal to 35 . By using the
Einstein equation (2.2) with indices (, ) one can easily find that the combination
3
aa = haa +
5
is equal to
1
2
2
s1 s2 + 123
a s1 a s2 + 123
a b s1 a b s2 .
aa = 123
Taking into account that the terms of the form s aa enter the equation for s as follows

a
c
+ 123 a s2 a c3
a2 m2a s1 = 123s2 a3
140
we obtain the contribution of aa

a2 m21 s1
= (tr0)1234s2 s3 s4 + (tr2a)1234 a s2 s3 a s4 + (tr2b)1234s2 a s3 a s4
+ (tr4a)1234 a s2 b s3 a b s4 + (tr4b)1234s2 b b s3 a b s4
+ (tr6a)1234 a s2 b c s3 a b c s4 .
(5.36)
The coefficients tr1234 in this equation depend on SO(6) tensors of the form
f5n a125 a345,
n > 1,
1
a125 a345.
f5 5
5.6. Contribution of vector fields

In this subsection Va denotes either the vector field Aa or Ca . The contribution of the
vector fields to the equations of motion for the scalars s, and to the equations of motion for
the vector fields may be written in the form

V
V
V 2 a 5
a s 2 Va5 + N125
a b s 2 a Vb5 + R125
s Va
a2 m2 s 1 = K125

V
a 2
2 5
b
5
2 5
+ T125 s b Va a Vb mV Vb ,
(5.37)
V
V
s1 a s2 + E125
b s1 a b s2
a2 Vb5 a b Va5 m2V Va5 = a V + D125
V
+ F125
b c s1 a b c s2 ,
(5.38)
where the constants D, E, F are antisymmetric in 1, 2, and V has the following

dependence on s
V
a s1 a s2 .
V = QV123 s1 s2 + H123
To get rid of higher-derivative terms in Eq. (5.37) we perform the following shift
1 V a
s2 Va3 .
s1 s1 + N123
2
Then the equation acquires the form

1
a2 m2 s1 = V123 a s2 Va3 + R123 s2 a Va3 N123 b s2 b a Va3
2

1
+ T123 N123 a s2 b2 Va3 b a Vb3 m2V Vb3 .
2
To remove higher-derivative and total-derivative term from Eq. (5.38), and to reduce it to
the canonical form we make the following fields redefinition [6]
Va3 = V 0 a
3
1
V
a V + J453
s4 a s5 + LV453 b s4 a b s5 .
m2V
Finally we should substitute the redefinition in Eq. (5.39), and represent it in the form

a2 m2 s1

1
a
05
a 05
= V125 s2 Va + s2 Va + (V 0)1234s2 s3 s4
2
141
+ (V 2a)1234 a s2 s3 a s4 + (V 2b)1234s2 a s3 a s4
+ (V 4a)1234 a s2 b s3 a b s4 + (V 4b)1234s2 a b s3 a b s4
+ (V 6a)1234 a s2 b c s3 a b c s4 .
Summing up the contributions of the vectors A and C we see that the coefficients V1234
depend on the SO(6) tensors of the form
f5n t125 t345 ,
n > 0,
1
t125 t345 .
f5 5
5.7. Contribution of contact terms

Finally we have to take into account the contribution of contact terms that appear when
we decompose the covariant equations of motion (2.1) and (2.2) up to the third order in the
fields, and keep only terms cubic in the scalars s. This contribution has the form

a2 m21 s1
= (c0)1234s2 s3 s4 + (c2a)1234 a s2 s3 a s4 + (c2b)1234s2 a s3 a s4
+ (c4a)1234 a s2 b s3 a b s4 + (c4b)1234s2 a b s3 a b s4
+ (c4c)1234a b s2 a s3 b s4 + (c6a)1234 a s2 b c s3 a b c s4
+ (c6c)1234 a b s2 b c s3 a c s4 .
(5.39)
The coefficients c1234 in this equation depend on SO(6) tensors of the form
f5n a125 a345,
n > 1.
6. Analysis of the equations

In this section we explain what steps one should undertake to obtain Lagrangian
equations of motion from the original ones. Looking at the contributions derived in the
previous section we see that the cubic corrections to the equations of motion for the
scalars s I have the form

a2 m21 s1
= (w0)1234 s2 s3 s4 + (w2a)1234 a s2 s3 a s4
+ (w4a)1234 a s2 b s3 a b s4 + (w4b)1234s2 a b s3 a b s4
+ (w6a)1234 a s2 b c s3 a b c s4
+ (w6c)1234 a b s2 b c s3 a c s4 .
(6.1)
The coefficients w1234 in this equation may in general depend on SO(6) tensors of the form
f5n a125 a345,
f51 a125a345,
f5n t125 t345 ,

f5n p125 p345 , n > 0,
1
1
1
a125a345 ,
a125 a345,
t125 t345,
f5 + 3
f5 5
f5 5
and tensors obtained from them by permutation of the indices 1, 2, 3, 4.
142
However, by using the identities (B.8), (B.9), (B.10) and (B.11) from the appendix, we
may reduce the tensors t125 t345 and f5 t125 t345 to the tensors f5n a125a345 , n > 1, and
p125 p345 and f5 p125 p345 to the tensors f5n a125a345, n > 1, f52 t125 t345, f515 a125 a345 ,
and
1
f5 5 t125 t345 ,
and tensors obtained from them by permutation of the indices 1, 2, 3, 4.
1
f5 +3 a125 a345 ,
1
1
f5 5 t125 t345 completely disappear from the total contribution, and the tensor f5 5 a125 a345
Then we find that (after the additional shift) the tensors of the form
occurs only in the terms without derivatives, and with two derivatives.
6.1. 6-derivative terms
We begin our analysis of (6.1) with the six-derivative terms. We see that the equation
contains in particular the following term coming from the vectors contribution after the use
of the identities (B.8) and (B.9)

(6.2)
a2 m21 s1 = w1234f5 (a135 a245 a145a235 ) a s2 b c s3 a b c s4 .
All other terms in the coefficients w6a and w6c only depend on tensors f5n a125a345, n =
0, 1, 2. To compare (6.2) with the other contributions we perform the shift
s1 s1 + J1234 a s2 b s3 a b s4 ,
(6.3)
and choose
J1234 = j1234f5 (a135a245 a145a235),
where
j1234 =

1 1
1
2
(w1234 + w1324) + w1234 w1324 = w1234 w1324.
2 3
3
3
This results in the following change of Eq. (6.2)

a2 m21 s1
= 2(j1324 j1432)f5 a125a345 a b s2 b c s3 a c s4
+ j1324(f1 + f2 + f3 + f4 3f5 )a125a345 a s2 b c s3 a b c s4
+ (dw4a)1234 a s2 b s3 a b s4 + (dw2a)1234 a s2 s3 a s4 ,
where
(dw4a)1234 = f5 (a135 a245 a145a235 )

j1234 m22 + m23 + m24 m21 18 2j1243
2j1423f5 (a125a245 a135a245),

(dw2a)1234 = 2 (j1234 + j1243)(a135 a245 a145a235 )

+ j1324(a125 a345 a145a235 ) f5 m23
are the additional contributions to the coefficients w4a and w2a. Now we can use the
symmetry of the 4-derivative term and the 2-derivative term under the permutation of
the indices 2, 3, and 2, 4 respectively, and identity (B.5) from the appendix to express
143
f5 a145a235 and the whole right hand side of the equation only through a125a345 . Then the
coefficients dw4a and dw2a acquire the form
h

(dw4a)1234 = f5 (2j1324 + j1234) m22 + m23 + m24 m21 18

2(2j1342 j1423 + j1432 + j1243)

(f1 + f2 + f3 + f4 ) j1234 m22 + m23 + m24 m21 18
i
2j1243 a125a345 ,

(dw2a)1234 = 2 (j1234 + j1243)(f1 + f2 + f3 + f4 f5 )

+ (j1324 j1342 2j1432 2j1423)f5 m23 a125 a345.
Thus we reduced all 6-derivative terms to terms which depend only on the tensors
f5n a125a345 .
Now summing up the vector fields contribution with the contributions of all the other
fields we get that the term w6c has the following structure
(w6c)1234 = (w6c0)1234a125a345 + (w6c1)1234f5 a125a345
f52
a125a345,
4
(6.4)
where (w6c1)1234 is a function symmetric under permutation of 2, 3, and 4, and we denote

= (k1 1)k1(k1 + 2)(k2 + 1)(k3 + 1)(k4 + 1). Thus we may use the identity
1
w1234 f5 a125a345 = w1234(f1 + f2 + f3 + f4 )a125 a345,
3
valid for any function symmetric in 2, 3, 4. So, we see that w6c does not depend on the
tensor f5 a125a345 , and, moreover wc60 should be symmetrized in 2, 3, 4 because it is
multiplied by a symmetric tensor a b s2 b c s3 a c s4 .
Looking at the term w6a we see that this term has the same form (6.4), and moreover,
the coefficient w6a1 is also symmetric in 2, 3, 4, and, therefore, can be reduced to w6a0.
The term wa62 proportional to f52 a125a345 comes from massive gravitons, vectors and
scalars and is given explicitly

f2
a2 m21 s1 = 5 a145a235 a s2 b c s3 a b c s4 .
2
(6.5)
We may reduce the term to the structure a b s2 b c s3 a c s4 by performing the shift

s1 s1
f52
a145 a235 b s2 c s3 b c s4
8
that results in

f2
a2 m21 s1 = 5 a125a345a b s2 b c s3 a c s4
4
2

f5
f2
a125a345 + 5 a145a235 m22 + m23 + m24 m21 18
+
2
8
a s2 b s3 a b s4
144
m23
f52
f2
a125a345 a s2 s3 a s4 + m22 5 a125a345s2 a s3 a s4 .
2
4
(6.6)
Summing up the coefficient on the first line of (6.6) with w6c we obtain that the final
contribution does not depend on the tensor f52 a125 a345 . So the new coefficient w6c depends
only on the tensor a125a345 , and we can easily reduce it to w6a by means of the shift
1
s1 s1 + (w6c)1234 b s2 c s3 b c s4 .
2
This results in

a2 m21 s1 = 2(w6c)1234 a s2 b c s3 a b c s4

1
2
2
2
2
2(w6c)1234 + (w6c)1234 m2 + m3 + m4 m1 18
2
a s2 b s3 a b s4 + (w6c)1234m23 a s2 s3 a s4 .
Adding the coefficient 2(w6c) from the first line of the equation to w6a we obtain a new
coefficient that is symmetric in 2, 3, and, therefore, the first term on the r.h.s. of the equation
can be transformed to the structure a b s2 b c s3 a c s4 by using (6.3). Symmetrizing
the coefficient in front of the 6-derivative term we obtain zero. This shift also produces
additional contributions to the coefficients w4a and w2a.
Thus we have shown that all 6-derivative terms could be shifted away.
We proceed with 4-derivative terms for which, we take into account all the additional
contributions appeared in the previous subsection due to our way of working with
2
6-derivative terms. The coefficient w4a contains the term 16(f51) t125t345 that gives
Lagrangian contribution to the equations of motion. Other contributions are nonlagrangian,
and we analyze them by decomposing the coefficients w4a and w4b in Laurent series in f5 .
6.2.1. 4-derivative terms with 1/f5
These terms give the following contribution to the equations of motion

a2 m21 s1 = (w4ad)1234 a s2 b s3 a b s4 + (w4bd)1234s2 a b s3 a b s4 .
We decompose w4ad into parts symmetric and antisymmetric in 3, 4, and shift its
symmetric part to the 4b-structure by using the field redefinition
s1 s1 + J1234s2 a s3 a s4 .
The resulting 4-derivative vertices can be written in the form:

2(f1 f2 )(f3 f4 )
a125a345 a s2 b s3 a b s4
a2 m21 s1 =
f5
(f1 f2 )(f3 f4 )
a125a345 a b s2 s3 a b s4 .
f5
Finally performing the shift
(6.7)
s1 s1 + J1234 b s2 s3 b s4
145
(6.8)
and using the symmetry of the vertex w.r.t. 3, 4, we represent the final result for the
4-derivative vertex as follows

3(f1 f2 )(f3 f4 )
a125a345 a s2 b s3 a b s4
a2 m21 s1 =
f5

(f1 f2 )(f3 f4 ) 2
+
m2 + m23 + m24 m21 8 a s2 s3 a s4 .
2f5
The 4-derivative term represents a Lagrangian contribution to the equation of motion,
which can be derived from the Lagrangian of the form
Z
(4)
(6.9)
L = A1234 s1 a s2 b2 (s3 a s4 ),
where the quartic coupling A41234 is antisymmetric in 1, 2 and 3, 4, and symmetric under
the interchange (1, 2) and (3, 4), and is given by
3
(4)
(f1 f2 )(f3 f4 ).
A1234 =
8f5
The equations of motion that follow from the Lagrangian are

a
b
a2 m21 s1 = 8A(4)
1234 s2 s3 a b s4

a
4 m23 + m24 4 A(4)
1234 s2 s3 a s4

a
2 m24 m23 A(4)
1234 s2 s3 a s4

(4)
m23 + m24 4 m24 m23 A1234s2 s3 s4 .
It is clear that the 4-derivative term cannot be removed by any field redefinition.
6.2.2. 4-derivative terms with f53
The contribution of the terms with f53 is given by

f3
a2 m21 s1 = 5 (3l + 8m) a s2 b s3 a b s4
64
f3
5 (3l + 4m + 4n)s2 a b s3 a b s4 .
128
Performing the shift
s1 s1 + J1234s2 b s3 b s4 ,
(6.10)
f53
where 2J1234 = 128 (3l + 4m + 4n) is symmetric in 3, 4, we obtain the Lagrangian

4-derivative term

f3
a2 m21 s1 = 5 (n m) a s2 b s3 a b s4
16

f3
+ 5 (3l + 4n + 4m) m22 + m23 + m24 m21 8 s2 a s3 a s4 ,
256
that can be again derived from a vertex of the form (6.9).
146

Here we first consider the term of the 4a-type (the term of the 4b-type is also nonzero
and we consider it later):
a

b
a2 m21 s1 = hn1234 n + hm
1234 m s2 s3 a b s4 .
Here hn , hm denote the coefficients of the corresponding structures n, m. We can rewrite
this equation as follows

a2 m21 s1 = hn1234n + hn1243m a s2 b s3 a b s4
+ 1432 a125a345 a b s2 a s3 b s4 ,
where

n
1234 = hm
1234 h1243 .
To convert the equation to the one containing only the 4b-type structure s2 a b s3 a b s4
we make the field redefinition
1
s1 s1 + 1432 a125a345 b s2 s3 b s4 + j1234s2 b s3 b s4 ,
2
where

1
j1234 = hn1234 n + hn1243 m 1432 l .
4
Then the equation transforms as follows

1
a2 m21 s1 = 1423 n + 1324 m + hn1234n + hn1243 m 1432 l
2
s2 a b s3 a b s4

j1234 m22 + m23 + m24 m21 8 s2 a s3 a s4

1
1432 l m22 + m23 + m24 m21 8 a s2 s3 a s4 .
2
Now we sum the r.h.s. of the equation with the contribution of 4b-type and get a Lagrangian
4-derivative term

(2l n m)f52
(28 + 3f1 + 3f2 + 3f3 + 3f4 )
a2 m21 s1 =
32
s2 a b s3 a b s4

j1234 m22 + m23 + m24 m21 8 s2 a s3 a s4

1
1432 l m22 + m23 + m24 m21 8 a s2 s3 a s4 .
2
It is convenient, however, to reduce the 4-derivative term to the term of 4a-type by means
of a field redefinition of the form,
s1 s1 + J1234s2 a s3 a s4
and by using the symmetry of the 4a-type term under the permutation of the indices 2, 3.
The resulting equation looks as
147

f 2 (n m)
3(f1 + f2 + f3 + f4 ) 28 a s2 b s3 a b s4
a2 m21 s1 = 5
16

f 2 (2l n m)
3(f1 + f2 + f3 + f4 ) 28
5
64

m22 + m23 + m24 m21 8 s2 a s3 a s4

j1234 m22 + m23 + m24 m21 8 s2 a s3 a s4

1
1432 l m22 + m23 + m24 m21 8 a s2 s3 a s4 ,
2
and the 4-derivative term can be obtained from the vertex of the form (6.9).
We can reduce the 4b-type term to the 4a-type one by means of the shift
s1 s1 + j1234s2 b s3 b s4 ,
1
j1234 = (w4b)1234f5 .
2
This results in

a2 m21 s1 = (w4a 2 w4b)1234 a s2 b s3 a b s4

j1234 m22 + m23 + m24 m21 8 s2 a s3 a s4 .
Representing
l
n
m
l + P1234
n + P1234
m,
(w4a 2 w4b)1234 = P1234
using the identity (B.5), and changing the summation indices 2 and 3 we rewrite the
equation in the form

l
n
m
m
+ P1324
P1324
P1234
f5 a125 a345 a s2 b s3 a b s4
a2 m21 s1 = P1234
m
+ P1324
(f1 + f2 + f3 + f4 ) a s2 b s3 a b s4

j1234f5 m22 + m23 + m24 m21 8 s2 a s3 a s4 .
The 4-derivative term represents a Lagrangian contribution as can be seen by decomposing

the coefficients in front of f5 a125a345 in parts antisymmetric and symmetric in 3, 4:

a2 m21 s1 = Y a + Y s 1234f5 a125a345 a s2 b s3 a b s4
m
+ P1324
(f1 + f2 + f3 + f4 ) a s2 b s3 a b s4

j1234f5 m22 + m23 + m24 m21 8 s2 a s3 a s4 ,
where
3(f1 f2 )(f3 f4 )
,
16
3(f1 + f2 + f3 + f4 2)(f1 + f2 + f3 + f4 12)
s
.
Y1234
=
8
a
=
Y1234
It is convenient to get rid of the symmetric in 3, 4 4-derivative contribution by using the

following identity
148
1
w1234 f5 a125a345 = w1234 (f1 + f2 + f3 + f4 )a125 a345
3

+ f5 (a135a245 a145 a235) ,
valid for any function w1234 symmetric in 2, 3. The final contribution is given by

a
s
f5 l + 13 Y1234
f5 (n m) a s2 b s3 a b s4
a2 m21 s1 = Y1234

s
m
+ 13 Y1234
+ P1324
(f1 + f2 + f3 + f4 ) a s2 b s3 a b s4

j1234f5 m22 + m23 + m24 m21 8 s2 a s3 a s4 .
Thus, we are left only with the antisymmetric 4-derivative Lagrangian contribution and the
additional 4a-type terms without f5 .
6.2.5. 4-derivative terms without f5
Just as above we use the shift
1
j1234 = (w4b0)1234,
2
to get rid of the 4b structure, and take into account the additional contribution coming
from the terms with f5 . Then we symmetrize the resulting coefficient w4a0 in 2 and 3,
and decompose it into parts symmetric and antisymmetric in 3 and 4. Then we shift the
symmetric part back to the 4b structure and get the equation

a2 m21 s1 = (L4a0)1234a125a345 a s2 b s3 a b s4
s1 s1 + j1234s2 b s3 b s4 ,
+ (L4b0)1234a125a345 s2 a b s3 a b s4

+ 2(L4b0)1234 m22 + m23 + m24 m21 8 a125 a345s2 a s3 a s4

j1234 m22 + m23 + m24 m21 8 s2 a s3 a s4 ,
(6.11)
where
21(f1 f2 )(f3 f4 )
,
16
7(2f1 f2 + 2f3 f4 ) (f1 + f2 )(f3 + f4 )
.
(L4b0)1234 =
8
(L4a0)1234 =
Both the 4-derivative terms are Lagrangian. The antisymmetric term can be derived from
a Lagrangian of the form (6.9), and the symmetric term can be obtain from the following
Lagrangian
Z

(4)
(6.12)
s1 a s2 b2 s3 a s4 ,
L = S1234
4
is symmetric in 1, 2 and 3, 4, and symmetric under the
where the quartic coupling S1234
interchange (1, 2) and (3, 4), and is given by
(4)
S1234 = (L4b0)1234.
4
Equations of motion that follow from the Lagrangian are
149

(4)
a2 m21 s1 = 4S1234
s2 a b s3 a b s4
(4)
+ 4 m23 + m24 6 S1234
s2 a s3 a s4

(4)
+ m23 + m24 4 m23 + m24 S1234
s2 s3 s4 .
This completes considering 4-derivative terms.
We proceed with 2-derivative terms for which, we should take into account all the
additional contributions appeared because of the shifts used in the previous subsections,
and contributions which appear when one represents the 4-derivative terms as variations of
the vertices of the types (6.9) and (6.12).
The coefficient w2a contains four Lagrangian terms proportional to
f52 p125 p345 ,
1
a125a345
f5 5
(f5 1)3 t125 t345 ,
(f5 1)2 t125 t345
and
that can be found in Section 2.

We find convenient to represent the contribution of the other 2-derivative terms in the
form

a2 m21 s1 = (w2a)1234a125 a345 a s2 s3 a s4 + (w2b)1234a125a345s2 b s3 a s4 ,
where the coefficients w2a and w2b may depend on f5 .
This equation is non-Lagrangian, and we again analyze it by decomposing the
coefficients w2a and w2b in Laurent series in f5 .
6.3.1. 2-derivative terms with 1/f5
Taking into account all additional contributions we represent the 2-derivative contribution in the form

1
a2 m21 s1 = (A2ad)1234 a125a345 a s2 s3 a s4
f5
1
+ (S2ad)1234 a125a345 a s2 s3 a s4
f5
1
+ (L2bd)1234 a125 a345s2 a s3 a s4 .
f5
Here we decompose the coefficient of the 2a type on the antisymmetric A2ad and
symmetric S2ad parts with respect to permutation of the indices 3, 4.
The antisymmetric part is Lagrangian and the coefficient is given by
(A2ad)1234 =
1
(f1 f2 )(f3 f4 )
2
36 + f1 + f2 + f3 + f4 20(k1 + k2 + k3 + k4 )

+ 10(k1 + k2 )(k3 + k4 ) .
150
The corresponding 2-derivative term can be derived from the following Lagrangian
Z
(6.13)
s1 a s2 s3 a s4 ,
L = A(2)
1234
where the quartic coupling A21234 is antisymmetric in 1, 2 and 3, 4, and symmetric under
the interchange (1, 2) and (3, 4). The equations of motion that follow from the Lagrangian
are

(2)
a
2
2
a2 m21 s1 = 4A(2)
1234 s2 s3 a s4 m4 m3 A1234 s2 s3 s4 ,
and, therefore, we have
1
A(2)
1234 = 4 (A2ad)1234 f a125 a345 .
5
Now we shift the remaining type 2a structure to the type 2b one and get

1
a2 m21 s1 = (A2ad)1234 a125a345 a s2 s3 a s4
f5

1
1
+ L2bd S2ad
a125a345 s2 a s3 a s4
2
f
1234 5
1
1
(S2ad)1234 m22 + m23 + m24 m21
a125 a345s2 s3 s4 .
4
f5
(6.14)
Now we see that the 2b structure turns out to be Lagrangian with

(S2bd)1234 =
1
(k1 k2 )(k3 k4 )(f1 f2 )(f3 f4 ).
2
The corresponding 2-derivative term can be derived from the Lagrangian

Z
(2)
s1 a s2 s3 a s4 ,
L = S1234
(6.15)
(6.16)
2
is symmetric in 1, 2 and 3, 4, and symmetric under the
where the quartic coupling S1234
interchange (1, 2) and (3, 4). The equations of motion that follow from the Lagrangian are

(2)
(2)
s2 a s3 a s4 + m24 + m23 S1234
s2 s3 s4 ,
(6.17)
a2 m21 s1 = 2S1234
and, therefore, we have

(2)
=
S1234
1
(S2bd)1234 a125a345 .
2
f5
We omit considering terms with f54 , f53 and f52 , because their analysis goes the same
line as before. We just remark that we used the shift
s1 s1 + J1234s2 s3 s4
to remove the terms completely symmetric with respect to permutation of indices 2, 3 and
4 from the equations of motion.
151

By using a field redefinition, we shift the type 2b term to the type 2a one, and represent
the equation in the form

a
a2 m21 s1 = (w2)l1234l + (w2)n1234n + (w2)m
1432 m f5 s2 s3 a s4 .
This equation can be further rewritten as follows

a2 m21 s1 = (w2)l1234 (w2)n1234 (w2)n1432 + (w2)m
1432
f5 a125a345 a s2 s3 a s4
+ (w2)n1234 (f1 + f2 + f3 + f4 ) a s2 s3 a s4 ,
where we use identity (B.5) and the symmetry of the tensor a s2 s3 a s4 in 2 and 4.
Introducing the notation
F1234 = (w2)l1234 (w2)n1234 (w2)n1432 + (w2)m
1432 ,
and by using again (B.5) we rewrite the equation as follows

a2 m21 s1 = F a + 13 F s 1234f5 a125a345 a s2 s3 a s4
s
13 F1324
f5 a125a345s2 a s3 a s4

s
+ (w2)n1234 + 13 F1234
(f1 + f2 + f3 + f4 )a125a345 a s2 s3 a s4 ,
where F a and F s denote the parts of F antisymmetric and symmetric in 2 and 4

respectively. Finally we use a field redefinition to shift the 2b structure to the 2a one,
decompose the resulting 2a coefficient into parts symmetric and antisymmetric in 3 and 4

s
= S1234 + A1234,
F a + 13 F s 1234 + 23 F1324
and shift the symmetric part to the 2b structure. The resulting equation with Lagrangian
2-derivative terms A1234 and S1234 looks as follows

a2 m21 s1 = A1234f5 a125 a345 a s2 s3 a s4 12 S1234 f5 a125a345 s2 a s3 a s4

+ (w2)n + 13 F s 1234(f1 + f2 + f3 + f4 )a125a345 a s2 s3 a s4

14 S1234 m22 + m23 + m24 m21 f5 a125a345 s2 s3 s4

s
+ 16 F1234
(6.18)
m22 + m23 + m24 m21 f5 a125a345 s2 s3 s4 .
The consideration of the terms without f5 is simple. We sum up all the additional
contributions, shift the 2b structure to the 2a one, and finally we use a shift to remove
the part symmetric in 2, 3 and 4. After these steps we obtain a Lagrangian term. As before
we shift the symmetric part to the 2b structure to have a simple Lagrangian.
6.4. Non-derivative terms
The consideration of non-derivative terms is the simplest one. Summing up all
contributions we immediately obtain Lagrangian terms for all cases except the case with
f5 and without f5 . The equation of motion for the term with f5 has the form

a2 m21 s1 = Q1234f5 a125a345s2 s3 s4 .
152
Now we write the equation as

f5
a2 m21 s1 = (2Q1234a125a345 + Q1324a135 a245)s2 s3 s4
3
and then apply the identity (B.5). We get
(6.19)

f5
a2 m21 s1 = (2Q1234 Q1324 Q1342)a125 a345s2 s3 s4
3
1
+ Q1324 (f1 + f2 + f3 + f4 )s2 s3 s4 .
(6.20)
3
Here the term:
f5
(2Q1234 Q1324 Q1342 )a125a345
3
appears to be Lagrangian, and the additional term without f5 makes the total contribution
to the term without f5 Lagrangian as well.
Thus we showed that the equations of motion for the scalars s I can be reduced to the
Lagrangian form by means of a number of field redefinitions.
7. Conclusion
In this paper we derived all quartic couplings of the scalars dual to extended chiral
primary operators in N = 4 SYM4 by using the covariant equations of motion for type IIB
supergravity. The quartic terms appeared to contain vertices with two and four derivatives.
The appearance of 2-derivative vertices was of course expected. Some of the 4-derivative
vertices may be removed by such a field redefinition that changes the structure of cubic
terms, namely, one gets scalar cubic terms with two derivatives, and cubic terms describing
K.
non-minimal interaction of two scalars with vector fields of the form VI J K a s I b s J Fab
However, we do not know if all of the 4-derivative terms can be removed in such a way.
It would be interesting to clarify this point because the derivation of the CallanSymanzik
equations in the AdS/CFT framework performed in [53] was based on a gravity action
which does not contain terms with four or more derivatives.
Since we know the gravity action for the scalars s I up to the fourth order, we can
start computing 4-point functions of CPOs. In general this will require calculating two
new types of Feynman diagrams: (i) contact diagrams with 4-derivative vertices, and
(ii) exchange diagrams involving massive gravitons. It is not difficult to show that all
contact diagrams with 4-derivative vertices can be reduced to a sum of terms corresponding
to simple non-derivative quartic couplings, just as this was done in [40] for the case of
contact diagrams with 2-derivative vertices. Thus the only real problem is to compute the
exchange diagrams involving massive gravitons. However, the 4-point functions of CPOs
O2 can be easily found because all necessary diagrams have been already calculated. This
problem is now under consideration.
We proved that, as was conjectured in [6], the quartic couplings obtained vanish in the
extremal case, for which k1 = k2 + k3 + k4 . This also implies the non-renormalization of
153
extremal 4-point functions of single-trace CPOs. The vanishing of the quartic couplings is
not manifest, and requires an additional field redefinition. Although the quartic couplings
can be easily used for computing any 4-point function of CPOs, it would be useful to find
such a representation for the quartic couplings that makes the vanishing in the extremal
case explicit.
We showed that the quartic couplings admit the consistent KK truncation, and argued
that the consistency of the KK reduction implies a non-renormalization theorem of n-point
functions of n 1 single-trace operators dual to the fields from the massless multiplet and
one single-trace operator dual to a field from a massive multiplet. It would be interesting
to check the non-renormalization of the 5-point function of four CPOs O2 and one CPO
O4 in perturbation theory.
The simplest example of the 4-point function of three CPOs O2 and a CPO O4 belongs,
actually, to the class of so-called next-to-extremal 4-point functions, for which k1 = k2 +
k3 + k4 2. The non-renormalization of such correlation functions was proven in [54], and
very recently checked to first order in perturbation theory in [55]. The non-renormalization
theorem also implies the vanishing of the corresponding functions of extended CPOs
and, since it is not difficult to show that there is no exchange diagram in this case, the
corresponding next-to-extremal quartic couplings of scalars s I have to vanish too. It
would be interesting to check this.
Note added
We have recently shown [56] that the relevant part of the gauged N = 8 5-dimensional
supergravity action coincides with the action for the scalar s2 we found in the paper.
Note added in proof
As was recently shown [57] the non-renormalization theorem for n-point function
of n 1 single-trace operators dual to the fields from the massless multiplet and one
single-trace operator dual to the field from the massive multiplet for n > 5 holds in a
weaker form. Namely, the corresponding correlation functions are given by sums of certain
space-time structures whose factored form is preserved by radiative corrections.
Appendix A
Here we collect the quartic couplings of the scalars s I representing our main result. The
couplings are given by sums of terms depending on various independent SO(6) tensors. To
simplify the presentation we sometimes use the following notations
x k1 ,
y k2 ,
t k3 ,
w k4 ,
z k5 ,
= (x + 1)(y + 1)(t + 1)(w + 1).

All the SO(6) tensors are given by tensors of the form F (f5 )aI1 I2 I5 aI3 I4 I5 ,
(f5 1)n tI1 I2 I5 tI3 I4 I5 and f5n pI1 I2 I5 pI3 I4 I5 , where F (f5 ) is a function of f5 , and sum-
154
mation over the index I5 is assumed. To distinguish the couplings with different functions
F we use an additional subscript in notation of a coupling.
Quartic couplings of 4-derivative vertices:

(4)
1 3
f (a145 a235 a135a245 ),
4 5

1
=
3(f1 + f2 + f3 + f4 ) 28 f52 (a145a235 a135 a245),
4
3
= (f1 f2 )(f3 f4 )f5 a125a345
4
1
(f1 + f2 + f3 + f4 2)(f1 + f2 + f3 + f4 12)
f5 (a145 a235 a135a245 ),

21
= (f1 f2 )(f3 f4 )a125 a345,
4

7
2f1 f2 + 2f3 f4 (f1 + f2 )(f3 + f4 ) a125a345,
=
4
12
= (f1 f2 )(f3 f4 )f51 a125a345,
3
= (f5 1)2 t125 t345 .
(A3 )I1 I2 I3 I4 =
(A2 )(4)
I1 I2 I3 I4
(4)
(A1 )I1 I2 I3 I4
(A0 )(4)
I1 I2 I3 I4
(4)
(S0 )I1 I2 I3 I4
(A1 )(4)
I1 I2 I3 I4
(4)
(At 2 )I1 I2 I3 I4
Quartic couplings of 2-derivative vertices:

(2)
5 4
f (a145a235 a135a245),
48 5
1
= (k1 k2 )(k3 k4 )f53 a125a345 ,
2
1
=
137 80(k1 + k2 + k3 + k4 ) + 2(f1 + f2 + f3 + f4 )
16

+ 32(k1k2 + k3 k4 ) + 24(k1 + k2 )(k3 + k4 ) f53 a125a345,
(A4 )I1 I2 I3 I4 =
(2)
(A3 )I1 I2 I3 I4
(S3 )(2)
I1 I2 I3 I4
(k1 k2 )(k3 k4 )
40 12(k1 + k2 + k3 + k4 )
4
+ 2(f1 + f2 + f3 + f4 ) + 16(k1k2 + k3 k4 )

+ (k1 + k2 )(k3 + k4 ) f52 a125 a345,
1
3741 + 2984t 342t 2 56t 3 + 31t 4 + 2984w 2272tw
=
16
+ 376t 2 w + 128t 3 w 342w2 + 376tw2 + 42t 2 w2 56w3
(A2 )(2)
I1 I2 I3 I4 =
(S2 )(2)
I1 I2 I3 I4
+ 128tw3 + 31w4 + 2984x 1760tx + 144t 2 x + 88t 3 x 1760wx

+ 832twx + 88t 2 wx + 144w2 x + 88tw2 x + 88w3 x 342x 2
155
+ 144tx 2 + 40t 2 x 2 + 144wx 2 + 192twx 2 + 40w2 x 2 56x 3

+ 88tx 3 + 88wx 3 + 31x 4 + 2984y 1760ty + 144t 2 y + 88t 3 y
1760wy + 832twy + 88t 2 wy + 144w2 y + 88tw2 y + 88w3 y
2272xy + 832txy + 192t 2 xy + 832wxy 128twxy + 192w2 xy
+ 376x 2y + 88tx 2 y + 88wx 2 y + 128x 3y 342y 2 + 144ty 2
+ 40t 2 y 2 + 144wy 2 + 192twy 2 + 40w2 y 2 + 376xy 2 + 88txy 2
+ 88wxy 2 + 42x 2y 2 56y 3 + 88ty 3 + 88wy 3 + 128xy 3 + 31y 4
(2)
(A1 )I1 I2 I3 I4
f52 a125 a345,

(t w)(x y)
1840 1964t + 160t 2 + 156t 3 + 16t 4 1964w
=
48
+ 1312tw 388t 2 w 128t 3 w + 160w2 388tw2 120t 2 w2
+ 156w3 128tw3 + 16w4 1964x + 645tx 48t 2 x 25t 3 x
+ 645wx 952twx 73t 2 wx 48w2 x 73tw2 x 25w3 x
+ 160x 2 48tx 2 56t 2 x 2 48wx 2 328twx 2 56w2 x 2
+ 156x 3 25tx 3 25wx 3 + 16x 4 1964y + 645ty 48t 2 y
25t 3 y + 645wy 952twy 73t 2 wy 48w2 y 73tw2 y
25w3 y + 1312xy 952txy 328t 2 xy 952wxy 656twxy
328w2 xy 388x 2y 73tx 2y 73wx 2 y 128x 3y + 160y 2
48ty 2 56t 2 y 2 48wy 2 328twy 2 56w2 y 2 388xy 2
(S1 )(2)
I1 I2 I3 I4
73txy 2 73wxy 2 120x 2y 2 + 156y 3 25ty 3 25wy 3

128xy 3 + 16y 4 f5 a125a345 ,
1
20979 53784t + 18666t 2 + 4056t 3 1197t 4 + 192t 5
=
48
+ 72t 6 53784w + 59648tw 17792t 2w 2816t 3 w + 1896t 4w
+ 256t 5 w + 18666w2 17792tw2 + 2736t 2 w2 + 1344t 3w2
+ 256t 3 w3 1197w4 + 1896tw4 + 98t 2 w4 + 192w5 + 256tw5
+ 72w6 + 98t 4 w2 + 4056w3 2816tw3 + 1344t 2w3 53784x
+ 65168tx 11900t 2x 3296t 3x + 1428t 4x + 208t 5 x + 65168wx
53760twx + 7296t 2wx + 4000t 3 wx + 144t 4wx 11900w2x
+ 7296tw2 x + 1760t 2w2 x + 104t 3 w2 x 3296w3x + 4000tw3x
+ 104t 2 w3 x + 1428w4 x + 144tw4 x + 208w5 x + 18666x 2
11900tx 2 + 801t 2 x 2 + 1488t 3x 2 + 173t 4 x 2 11900wx 2
+ 3840twx 2 + 4472t 2wx 2 + 704t 3 wx 2 + 801w2x 2 + 4472tw2 x 2
+ 252t 2 w2 x 2 + 1488w3x 2 + 704tw3 x 2 + 173w4 x 2 + 4056x 3
3296tx 3 + 1488t 2x 3 + 424t 3 x 3 3296wx 3 + 5632twx 3
156
+ 464t 2 wx 3 + 1488w2 x 3 + 464tw2 x 3 + 424w3 x 3 1197x 4

+ 1428tx 4 + 173t 2x 4 + 1428wx 4 + 576twx 4 + 173w2 x 4 + 192x 5
+ 208tx 5 + 72x 6 53784y + 65168ty 11900t 2y 3296t 3y
+ 1428t 4y + 208t 5y + 65168wy 53760twy + 7296t 2wy
+ 4000t 3wy + 144t 4 wy 11900w2y + 7296tw2 y + 1760t 2w2 y
+ 104t 3 w2 y 3296w3 y + 4000tw3 y + 104t 2w3 y + 1428w4y
+ 144tw4 y + 208w5y + 59648xy 53760txy + 3840t 2xy
+ 5632t 3xy + 576t 4 xy 53760wxy + 23040twxy + 3264t 2 wxy
384t 3 wxy + 3840w2xy + 3264tw2 xy 128t 2 w2 xy + 5632w3xy
384tw3 xy + 576w4 xy 17792x 2y + 7296tx 2y + 4472t 2x 2 y
+ 464t 3 x 2 y + 7296wx 2y + 3264twx 2y + 40t 2 wx 2 y + 4472w2x 2 y
+ 40tw2 x 2 y + 464w3 x 2 y 2816x 3y + 4000tx 3y + 704t 2 x 3 y
+ 4000wx 3y 384twx 3 y + 704w2 x 3 y + 1896x 4y + 144tx 4y
+ 144wx 4y + 256x 5y + 18666y 2 11900ty 2 + 801t 2 y 2
+ 1488t 3y 2 + 173t 4 y 2 11900wy 2 + 3840twy 2 + 4472t 2wy 2
+ 704t 3 wy 2 + 801w2 y 2 + 4472tw2 y 2 + 252t 2 w2 y 2 + 1488w3y 2
+ 704tw3 y 2 + 173w4 y 2 17792xy 2 + 7296txy 2 + 4472t 2xy 2
+ 464t 3 xy 2 + 7296wxy 2 + 3264twxy 2 + 40t 2 wxy 2 + 4472w2xy 2
+ 40tw2 xy 2 + 464w3 xy 2 + 2736x 2y 2 + 1760wx 2y 2 128twx 2 y 2
+ 252w2 x 2 y 2 + 1344x 3y 2 + 104tx 3 y 2 + 104wx 3y 2 + 98x 4 y 2
+ 4056y 3 3296ty 3 + 1488t 2y 3 + 424t 3 y 3 3296wy 3
+ 5632twy 3 + 464t 2 wy 3 + 1488w2y 3 + 464tw2 y 3 + 424w3y 3
2816xy 3 + 4000txy 3 + 704t 2 xy 3 + 4000wxy 3 384twxy 3
+ 704w2 xy 3 + 1344x 2y 3 + 104tx 2y 3 + 104wx 2 y 3 + 256x 3y 3
1197y 4 + 1428ty 4 + 173t 2 y 4 + 1428wy 4 + 576twy 4 + 173w2y 4
(2)
(A0 )I1 I2 I3 I4
+ 1896xy 4 + 144txy 4 + 144wxy 4 + 98x 2y 4 + 192y 5 + 208ty 5

+ 208wy 5 + 256xy 5 + 72y 6 f5 a125 a345,
(x y)(t w)
144288 + 74776t + 10752t 2 5264t 3
=
192
+ 992t 4 + 440t 5 + 32t 6 + 74776w + 37504tw 11664t 2w
+ 2016t 3w + 1400t 4w + 128t 5w + 10752w2 11664tw2
+ 4512t 2w2 + 2088t 3 w2 + 176t 4 w2 5264w3 + 2016tw3
+ 2088t 2w3 + 256t 3 w3 + 992w4 + 1400tw4 + 176t 2 w4 + 440w5
+ 128tw5 + 32w6 + 74776x 26042tx 5888t 2x + 3948t 3x
+ 888t 4 x + 46t 5 x 26042wx 7648twx + 6380t 2 wx
157
+ 1784t 3wx + 142t 4 wx 5888w2x + 6380tw2 x + 1880t 2w2 x

+ 170t 3 w2 x + 3948w3 x + 1784tw3 x + 170t 2w3 x + 888w4 x
+ 142tw4 x + 46w5 x + 10752x 2 5888tx 2 + 832t 2 x 2 + 1272t 3x 2
+ 160t 4 x 2 5888wx 2 + 768twx 2 + 1784t 2 wx 2 + 144t 3wx 2
+ 832w2 x 2 + 1784tw2 x 2 + 16t 2 w2 x 2 + 1272w3x 2 + 144tw3 x 2
+ 160w4 x 2 5264x 3 + 3948tx 3 + 1272t 2x 3 + 5t 3 x 3 + 3948wx 3
+ 1480twx 3 91t 2 wx 3 + 1272w2x 3 91tw2 x 3 + 5w3 x 3 + 992x 4
+ 888tx 4 + 160t 2 x 4 + 888wx 4 + 208twx 4 + 160w2 x 4 + 440x 5
+ 46tx 5 + 46wx 5 + 32x 6 + 74776y 26042ty 5888t 2y
+ 3948t 3y + 888t 4y + 46t 5 y 26042wy 7648twy + 6380t 2wy
+ 1784t 3wy + 142t 4 wy 5888w2y + 6380tw2 y + 1880t 2w2 y
+ 170t 3 w2 y + 3948w3 y + 1784tw3 y + 170t 2w3 y + 888w4 y
+ 142tw4 y + 46w5 y + 37504xy 7648txy + 768t 2 xy + 1480t 3 xy
+ 208t 4 xy 7648wxy + 448twxy + 1608t 2wxy + 32t 3 wxy
+ 768w2 xy + 1608tw2 xy 352t 2 w2 xy + 1480w3xy + 32tw3 xy
+ 208w4 xy 11664x 2y + 6380tx 2y + 1784t 2x 2 y 91t 3 x 2 y
+ 6380wx 2y + 1608twx 2y 571t 2 wx 2 y + 1784w2x 2 y
571tw2 x 2 y 91w3 x 2 y + 2016x 3y + 1784tx 3y + 144t 2 x 3 y
+ 1784wx 3y + 32twx 3 y + 144w2 x 3 y + 1400x 4y + 142tx 4y
+ 142wx 4y + 128x 5y + 10752y 2 5888ty 2 + 832t 2 y 2 + 1272t 3y 2
+ 160t 4 y 2 5888wy 2 + 768twy 2 + 1784t 2 wy 2 + 144t 3wy 2
+ 832w2 y 2 + 1784tw2 y 2 + 16t 2 w2 y 2 + 1272w3y 2 + 144tw3 y 2
+ 160w4 y 2 11664xy 2 + 6380txy 2 + 1784t 2 xy 2 91t 3 xy 2
+ 6380wxy 2 + 1608twxy 2 571t 2 wxy 2 + 1784w2xy 2
571tw2 xy 2 91w3 xy 2 + 4512x 2y 2 + 1880tx 2y 2 + 16t 2 x 2 y 2
+ 1880wx 2y 2 352twx 2 y 2 + 16w2 x 2 y 2 + 2088x 3y 2 + 170tx 3y 2
+ 170wx 3y 2 + 176x 4y 2 5264y 3 + 3948ty 3 + 1272t 2 y 3 + 5t 3 y 3
+ 3948wy 3 + 1480twy 3 91t 2 wy 3 + 1272w2y 3 91tw2 y 3
+ 5w3 y 3 + 2016xy 3 + 1784txy 3 + 144t 2 xy 3 + 1784wxy 3
+ 32twxy 3 + 144w2xy 3 + 2088x 2y 3 + 170tx 2y 3 + 170wx 2y 3
+ 256x 3y 3 + 992y 4 + 888ty 4 + 160t 2 y 4 + 888wy 4 + 208twy 4
(2)
(S0 )I1 I2 I3 I4
+ 160w2 y 4 + 1400xy 4 + 142txy 4 + 142wxy 4 + 176x 2y 4

+ 440y 5 + 46ty 5 + 46wy 5 + 128xy 5 + 32y 6 a125a345 ,
1
288576tw + 149552t 2w + 21504t 3w 10528t 4w
=
576
158
+ 1984t 5w + 880t 6w + 64t 7 w + 149552tw2 52084t 2w2

11776t 3w2 + 7896t 4w2 + 1776t 5 w2 + 92t 6 w2 + 21504tw3
11776t 2w3 + 1664t 3w3 + 2544t 4 w3 + 320t 5 w3 10528tw4
+ 7896t 2w4 + 2544t 3 w4 + 10t 4 w4 + 1984tw5 + 1776t 2 w5
+ 320t 3 w5 + 880tw6 + 92t 2 w6 + 64tw7 + 144288tx 74776t 2x
10752t 3x + 5264t 4x 992t 5 x 440t 6 x 32t 7 x + 144288wx
+ 26752t 2wx 6400t 3wx + 1024t 4wx + 960t 5 wx + 96t 6 wx
74776w2x + 26752tw2x 3520t 2w2 x + 2368t 3w2 x
+ 1104t 4w2 x + 144t 5 w2 x 10752w3x 6400tw3 x + 2368t 2w3 x
+ 1024t 3w3 x 112t 4 w3 x + 5264w4 x + 1024tw4 x + 1104t 2w4 x
112t 3 w4 x 992w5 x + 960tw5 x + 144t 2 w5 x 440w6 x
+ 96tw6 x 32w7 x 74776tx 2 + 26042t 2x 2 + 5888t 3 x 2
3948t 4x 2 888t 5 x 2 46t 6 x 2 74776wx 2 53504twx 2
+ 1760t 2wx 2 + 2560t 3wx 2 + 480t 4 wx 2 + 26042w2x 2
+ 1760tw2 x 2 + 96t 3 w2 x 2 + 28t 4 w2 x 2 + 5888w3 x 2 + 2560tw3 x 2
+ 96t 2 w3 x 2 256t 3w3 x 2 3948w4x 2 + 480tw4 x 2 + 28t 2 w4 x 2
888w5 x 2 46w6 x 2 10752tx 3 + 5888t 2x 3 832t 3 x 3
1272t 4x 3 160t 5 x 3 10752wx 3 + 12800twx 3 4928t 2wx 3
512t 3 wx 3 + 176t 4 wx 3 + 5888w2x 3 4928tw2 x 3 192t 2 w2 x 3
+ 128t 3 w2 x 3 832w3 x 3 512tw3 x 3 + 128t 2w3 x 3 1272w4x 3
+ 176tw4 x 3 160w5 x 3 + 5264tx 4 3948t 2x 4 1272t 3x 4
5t 4 x 4 + 5264wx 4 2048twx 4 1584t 2wx 4 64t 3 wx 4
3948w2x 4 64tw3 x 4 1584tw2 x 4 56t 2 w2 x 4 1272w3x 4
5w4 x 4 992tx 5 888t 2 x 5 160t 3 x 5 992wx 5 1920twx 5
144t 2 wx 5 888w2 x 5 144tw2 x 5 160w3 x 5 440tx 6
46t 2 x 6 440wx 6 192twx 6 46w2 x 6 32tx 7 32wx 7
+ 144288ty 74776t 2y 10752t 3y + 5264t 4 y 992t 5 y 440t 6y
32t 7 y + 144288wy + 26752t 2wy 6400t 3wy + 1024t 4wy
+ 960t 5 wy + 96t 6 wy 74776w2y + 26752tw2 y 3520t 2w2 y
+ 2368t 3w2 y + 1104t 4w2 y + 144t 5 w2 y 10752w3y 6400tw3 y
+ 2368t 2w3 y + 1024t 3w3 y 112t 4 w3 y + 5264w4 y + 1024tw4 y
+ 1104t 2w4 y 112t 3 w4 y 992w5 y + 960tw5 y + 144t 2w5 y
440w6 y + 96tw6 y 32w7 y 288576xy 53504t 2xy
+ 12800t 3xy 2048t 4 xy 1920t 5 xy 192t 6 xy 53504w2xy
512t 2 w2 xy 768t 3w2 xy 320t 4 w2 xy + 12800w3xy
159
768t 2 w3 xy 768t 3w3 xy 2048w4xy 320t 2 w4 xy

1920w5xy 192w6 xy + 149552x 2y + 26752tx 2y + 1760t 2x 2 y
4928t 3x 2 y 1584t 4x 2 y 144t 5 x 2 y + 26752wx 2y
+ 256t 2 wx 2 y + 384t 3 wx 2 y + 160t 4 wx 2 y + 1760w2x 2 y
+ 256tw2 x 2 y 256t 3 w2 x 2 y 4928w3x 2 y + 384tw3 x 2 y
256t 2 w3 x 2 y 1584w4x 2 y + 160tw4 x 2 y 144w5x 2 y
+ 21504x 3y 6400tx 3y + 2560t 2 x 3 y 512t 3 x 3 y 64t 4 x 3 y
6400wx 3y + 384t 2 wx 3 y + 384t 3 wx 3 y + 2560w2x 3 y
+ 384tw2 x 3 y + 512t 2 w2 x 3 y 512w3x 3 y + 384tw3 x 3 y
64w4 x 3 y 10528x 4y + 1024tx 4y + 480t 2 x 4 y + 176t 3 x 4 y
+ 1024wx 4y + 160t 2 wx 4 y + 480w2 x 4 y + 160tw2 x 4 y+176w3 x 4 y
+ 1984x 5y + 960tx 5y + 960wx 5 y + 880x 6y + 96tx 6 y + 96wx 6y
+ 64x 7y 74776ty 2 + 26042t 2y 2 + 5888t 3y 2 3948t 4y 2
888t 5 y 2 46t 6 y 2 74776wy 2 53504twy 2 + 1760t 2wy 2
+ 2560t 3wy 2 + 480t 4 wy 2 + 26042w2y 2 + 1760tw2 y 2 + 96t 3 w2 y 2
+ 28t 4 w2 y 2 + 5888w3y 2 + 2560tw3 y 2 + 96t 2 w3 y 2 256t 3 w3 y 2
3948w4y 2 + 480tw4 y 2 + 28t 2 w4 y 2 888w5y 2 46w6 y 2
+ 149552xy 2 + 26752txy 2 + 1760t 2 xy 2 4928t 3 xy 2
1584t 4xy 2 144t 5 xy 2 + 26752wxy 2 + 256t 2 wxy 2
+ 384t 3 wxy 2 + 160t 4 wxy 2 + 1760w2xy 2 + 256tw2 xy 2
256t 3 w2 xy 2 4928w3xy 2 + 384tw3 xy 2 256t 2w3 xy 2
1584w4xy 2 + 160tw4 xy 2 144w5 xy 2 52084x 2y 2
3520tx 2y 2 192t 3 x 2 y 2 56t 4 x 2 y 2 3520wx 2y 2 512twx 2 y 2
+ 512t 3 wx 2 y 2 192w3x 2 y 2 + 512tw3 x 2 y 2 56w4 x 2 y 2
11776x 3y 2 + 2368tx 3y 2 + 96t 2 x 3 y 2 + 128t 3x 3 y 2 + 2368wx 3y 2
768twx 3 y 2 256t 2 wx 3 y 2 + 96w2 x 3 y 2 256tw2 x 3 y 2
+ 128w3 x 3 y 2 + 7896x 4y 2 + 1104tx 4y 2 + 28t 2 x 4 y 2 + 1104wx 4y 2
320twx 4 y 2 + 28w2 x 4 y 2 + 1776x 5y 2 + 144tx 5y 2 + 144wx 5y 2
+ 92x 6y 2 10752ty 3 + 5888t 2y 3 832t 3 y 3 1272t 4y 3
160t 5 y 3 10752wy 3 + 12800twy 3 4928t 2wy 3 512t 3 wy 3
+ 176t 4 wy 3 + 5888w2 y 3 4928tw2 y 3 192t 2 w2 y 3 + 128t 3 w2 y 3
832w3 y 3 512tw3 y 3 + 128t 2 w3 y 3 1272w4y 3 + 176tw4 y 3
160w5 y 3 + 21504xy 3 6400txy 3 + 2560t 2 xy 3 512t 3 xy 3
64t 4 xy 3 6400wxy 3 + 384t 2 wxy 3 + 384t 3 wxy 3 + 2560w2xy 3
+ 384tw2 xy 3 + 512t 2 w2 xy 3 512w3xy 3 + 384tw3 xy 3
160
64w4 xy 3 11776x 2y 3 + 2368tx 2y 3 + 96t 2 x 2 y 3 + 128t 3x 2 y 3

+ 2368wx 2y 3 768twx 2 y 3 256t 2 wx 2 y 3 + 96w2 x 2 y 3
256tw2 x 2 y 3 + 128w3x 2 y 3 + 1664x 3y 3 + 1024tx 3y 3
256t 2 x 3 y 3 + 1024wx 3y 3 768twx 3 y 3 256w2 x 3 y 3
+ 2544x 4y 3 112tx 4 y 3 112wx 4y 3 + 320x 5y 3 + 5264ty 4
3948t 2y 4 1272t 3 y 4 5t 4 y 4 + 5264wy 4 2048twy 4
1584t 2wy 4 64t 3 wy 4 3948w2y 4 1584tw2 y 4 56t 2 w2 y 4
1272w3y 4 64tw3 y 4 5w4 y 4 10528xy 4 + 1024txy 4
+ 480t 2 xy 4 + 176t 3 xy 4 + 1024wxy 4 + 160t 2 wxy 4 + 480w2 xy 4
+ 160tw2 xy 4 + 176w3 xy 4 + 7896x 2y 4 + 1104tx 2y 4 + 28t 2 x 2 y 4
+ 1104wx 2y 4 320twx 2 y 4 + 28w2 x 2 y 4 + 2544x 3y 4 112tx 3y 4
112wx 3y 4 + 10x 4 y 4 992ty 5 888t 2 y 5 160t 3 y 5 992wy 5
1920twy 5 144t 2 wy 5 888w2y 5 144tw2 y 5 160w3 y 5
+ 1984xy 5 + 960txy 5 + 960wxy 5 + 1776x 2y 5 + 144tx 2y 5
+ 144wx 2y 5 + 320x 3y 5 440ty 6 46t 2 y 6 440wy 6 192twy 6
(A1 )(2)
I1 I2 I3 I4
(2)
(S1 )I1 I2 I3 I4
(At 3 )(2)
I1 I2 I3 I4
(2)
(At 2 )I1 I2 I3 I4
(2)
(St 2 )I1 I2 I3 I4
(Sp2 )(2)
I1 I2 I3 I4
(Sd )(2)
I1 I2 I3 I4
46w2 y 6 + 880xy 6 + 96txy 6 + 96wxy 6 + 92x 2y 6 32ty 7

32wy 7 + 64xy 7 a125a345,
4
= (f1 f2 )(f3 f4 )
36 20(k1 + k2 + k3 + k4 ) + f1 + f2 + f3 + f4

+ 10(k1 + k2 )(k3 + k4 ) f51 a125a345,
8
= (f1 f2 )(f3 f4 )(k1 k2 )(k3 k4 )f51 a125a345 ,
2
= (f5 1)3 t125 t345 ,
1
= (f5 1)2 t125 t345
k12 + k22 + k32 + k42 16(k1 + k2 + k3 + k4 )

+ 10(k1 + k2 )(k3 + k4 ) + 44 ,
2
= (k1 k2 )(k3 k4 )(f5 1)2 t125 t345 ,
4 2
= f5 p125p345 ,
9a125a345
(1 + k1 k2 )(1 + k1 k2 )(3 + k1 + k2 )
=
16(f5 5)
(5 + k1 + k2 )(1 + k3 k4 )(1 + k3 k4 )(3 + k3 + k4 )
(5 + k3 + k4 ).
161
Quartic couplings of non-derivative vertices:
(0)
3 5
f a125a345 ,
64 5
1
f 4 a125 a345
=
192 5

747 368(k1 + k2 + k3 + k4 ) + 65 k12 + k22 + k32 + k42

+ 132(k1k2 + k3 k4 ) 96(k1 + k2 )(k3 + k4 ) ,
1 3
f a125a345
=
64 5
3293 + 4036t 1012t 2 96t 3 + 35t 4 + 4036w 2428tw
(S5 )I1 I2 I3 I4 =
(S4 )(0)
I1 I2 I3 I4
(S3 )(0)
I1 I2 I3 I4
+ 48t 2 w + 88t 3 w 1012w2 + 48tw2 + 122t 2w2 96w3 + 88tw3

+ 35w4 + 4036x 2976tx + 360t 2 x + 40t 3 x 2976wx
+ 1056twx 8t 2 wx + 360w2 x 8tw2 x + 40w3 x 1012x 2
+ 360tx 2 + 44t 2 x 2 + 360wx 2 8twx 2 + 44w2 x 2 96x 3 + 40tx 3
+ 40wx 3 + 35x 4 + 4036y 2976ty + 360t 2 y + 40t 3 y 2976wy
+ 1056twy 8t 2 wy + 360w2 y 8tw2 y + 40w3 y 2428xy
+ 1056txy 8t 2 xy + 1056wxy 8w2 xy + 48x 2 y 8tx 2 y
8wx 2 y + 88x 3y 1012y 2 + 360ty 2 + 44t 2 y 2 + 360wy 2
(S2 )(0)
I1 I2 I3 I4
8twy 2 + 44w2 y 2 + 48xy 2 8txy 2 8wxy 2 + 122x 2y 2 96y 3

+ 40ty 3 + 40wy 3 + 88xy 3 + 35y 4 ,
1 2
f a125a345
=
64 5
8273 20116t + 9396t 2 + 1008t 3
1227t 4 + 36t 5 + 26t 6 20116w + 25644tw 2688t 2w
3544t 3w + 356t 4 w + 76t 5 w + 9396w2 2688tw2 2778t 2w2
+ 664t 3 w2 + 46t 4 w2 + 1008w3 3544tw3 + 664t 2w3 + 104t 3w3
1227w4 + 356tw4 + 46t 2 w4 + 36w5 + 76tw5 + 26w6 20116x
+ 32384tx 9032t 2x 1696t 3 x + 492t 4 x + 8t 5 x + 32384wx
23776twx + 224t 2 wx + 1152t 3wx 104t 4 wx 9032w2x
+ 224tw2 x + 1096t 2w2 x 224t 3 w2 x 1696w3 x + 1152tw3 x
224t 2 w3 x + 492w4 x 104tw4 x + 8w5 x + 9396x 2 9032tx 2
+ 332t 2 x 2 + 288t 3 x 2 + 60t 4 x 2 9032wx 2 + 2152twx 2 96t 2 wx 2
+ 144t 3 wx 2 + 332w2 x 2 96tw2 x 2 + 96t 2 w2 x 2 + 288w3 x 2
+ 144tw3 x 2 + 60w4 x 2 + 1008x 3 1696tx 3 + 288t 2 x 3 + 80t 3 x 3
1696wx 3 + 608twx 3 + 32t 2 wx 3 + 288w2x 3 + 32tw2 x 3
162
+ 80w3 x 3 1227x 4 + 492tx 4 + 60t 2 x 4 + 492wx 4 20twx 4

+ 60w2 x 4 + 36x 5 + 8tx 5 + 8wx 5 + 26x 6 20116y + 32384ty
9032t 2y 1696t 3y + 492t 4y + 8t 5 y + 32384wy 23776twy
+ 224t 2 wy + 1152t 3wy 104t 4 wy 9032w2y + 224tw2 y
+ 1096t 2w2 y 224t 3 w2 y 1696w3 y + 1152tw3 y 224t 2w3 y
+ 492w4 y 104tw4 y + 8w5 y + 25644xy 23776txy + 2152t 2 xy
+ 608t 3 xy 20t 4 xy 23776wxy + 10496twxy + 32t 2 wxy
128t 3 wxy + 2152w2xy + 32tw2 xy + 168t 2 w2 xy + 608w3 xy
128tw3 xy 20w4 xy 2688x 2y + 224tx 2y 96t 2 x 2 y
+ 32t 3 x 2 y + 224wx 2y + 32twx 2 y 16t 2 wx 2 y 96w2 x 2 y
16tw2 x 2 y + 32w3 x 2 y 3544x 3y + 1152tx 3y + 144t 2 x 3 y
+ 1152wx 3y 128twx 3 y + 144w2x 3 y + 356x 4y 104tx 4 y
104wx 4y + 76x 5y + 9396y 2 9032ty 2 + 332t 2y 2 + 288t 3y 2
+ 60t 4 y 2 9032wy 2 + 2152twy 2 96t 2 wy 2 + 144t 3wy 2
+ 332w2 y 2 96tw2 y 2 + 96t 2 w2 y 2 + 288w3y 2 + 144tw3 y 2
+ 60w4 y 2 2688xy 2 + 224txy 2 96t 2 xy 2 + 32t 3 xy 2 + 224wxy 2
+ 32twxy 2 16t 2 wxy 2 96w2 xy 2 16tw2 xy 2 + 32w3 xy 2
2778x 2y 2 + 1096tx 2y 2 + 96t 2 x 2 y 2 + 1096wx 2y 2 + 168twx 2 y 2
+ 96w2 x 2 y 2 + 664x 3y 2 224tx 3y 2 224wx 3y 2 + 46x 4y 2
+ 1008y 3 1696ty 3 + 288t 2 y 3 + 80t 3 y 3 1696wy 3 + 608twy 3
+ 32t 2 wy 3 + 288w2 y 3 + 32tw2 y 3 + 80w3 y 3 3544xy 3
+ 1152txy 3 + 144t 2 xy 3 + 1152wxy 3 128twxy 3 + 144w2 xy 3
+ 664x 2y 3 224tx 2y 3 224wx 2 y 3 + 104x 3y 3 1227y 4+492ty 4
(0)
(S1 )I1 I2 I3 I4
+ 60t 2 y 4 + 492wy 4 20twy 4 + 60w2 y 4 + 356xy 4 104txy 4

104wxy 4 + 46x 2y 4 + 36y 5 + 8ty 5 + 8wy 5 + 76xy 5 + 26y 6 ,
(w x)(t y)
f5 a125a345
=
288
163692 128440t + 28616t 2 + 2052t 3 3460t 4 + 484t 5
+ 80t 6 128440w + 72314tw 3096t 2w + 393t 3 w + 468t 4w
119t 5 w + 28616w2 3096tw2 1208t 2w2 864t 3 w2
+ 184t 4 w2 + 2052w3 + 393tw3 864t 2w3 + 41t 3 w3 3460w4
+ 468tw4 + 184t 2 w4 + 484w5 119tw5 + 80w6 128440x
+ 72314tx 3096t 2x + 393t 3 x + 468t 4 x 119t 5 x + 88352wx
31064twx + 13912t 2wx 2360t 3wx 392t 4 wx 18716w2x
+ 15049tw2x 5472t 2w2 x + 641t 3 w2 x 416w3 x 1416tw3x
163
+ 312t 2 w3 x + 1380w4 x 335tw4 x 256w5 x + 28616x 2

3096tx 2 1208t 2x 2 864t 3 x 2 + 184t 4 x 2 18716wx 2
+ 15049twx 2 5472t 2wx 2 + 641t 3 wx 2 880w2 x 2 4496tw2 x 2
+ 2392t 2w2 x 2 + 992w3x 2 + 524tw3 x 2 + 104w4 x 2 + 2052x 3
+ 393tx 3 864t 2 x 3 + 41t 3 x 3 416wx 3 1416twx 3 + 312t 2 wx 3
+ 992w2 x 3 + 524tw2 x 3 368w3 x 3 3460x 4 + 468tx 4 + 184t 2 x 4
+ 1380wx 4 335twx 4 + 104w2x 4 + 484x 5 119tx 5 256wx 5
+ 80x 6 128440y + 88352ty 18716t 2y 416t 3 y + 1380t 4y
256t 5 y + 72314wy 31064twy + 15049t 2wy 1416t 3wy
335t 4 wy 3096w2y + 13912tw2y 5472t 2 w2 y + 312t 3 w2 y
+ 393w3 y 2360tw3 y + 641t 2w3 y + 468w4 y 392tw4 y
119w5 y + 72314xy 31064txy + 15049t 2xy 1416t 3xy
335t 4 xy 31064wxy + 34928twxy 15928t 2wxy
+ 1664t 3wxy + 15049w2xy 15928tw2xy + 4313t 2w2 xy
1416w3xy + 1664tw3 xy 335w4 xy 3096x 2y + 13912tx 2y
5472t 2x 2 y + 312t 3 x 2 y + 15049wx 2y 15928twx 2y
+ 4313t 2wx 2 y 4496w2 x 2 y + 3488tw2 x 2 y + 524w3 x 2 y
+ 393x 3y 2360tx 3y + 641t 2x 3 y 1416wx 3y + 1664twx 3 y
+ 524w2 x 3 y + 468x 4y 392tx 4y 335wx 4 y 119x 5y
+ 28616y 2 18716ty 2 880t 2 y 2 + 992t 3 y 2 + 104t 4 y 2
3096wy 2 + 15049twy 2 4496t 2wy 2 + 524t 3wy 2 1208w2y 2
5472tw2 y 2 + 2392t 2w2 y 2 864w3y 2 + 641tw3 y 2 + 184w4 y 2
3096xy 2 + 15049txy 2 4496t 2xy 2 + 524t 3xy 2 + 13912wxy 2
15928twxy 2 + 3488t 2 wxy 2 5472w2xy 2 + 4313tw2 xy 2
+ 312w3 xy 2 1208x 2y 2 5472tx 2y 2 + 2392t 2x 2 y 2 5472wx 2y 2
+ 4313twx 2y 2 + 2392w2 x 2 y 2 864x 3y 2 + 641tx 3y 2 + 312wx 3y 2
+ 184x 4y 2 + 2052y 3 416ty 3 + 992t 2 y 3 368t 3 y 3 + 393wy 3
1416twy 3 + 524t 2 wy 3 864w2 y 3 + 312tw2 y 3 + 41w3 y 3
+ 393xy 3 1416txy 3 + 524t 2xy 3 2360wxy 3 + 1664twxy 3
+ 641w2 xy 3 864x 2y 3 + 312tx 2 y 3 + 641wx 2y 3 + 41x 3 y 3
3460y 4 + 1380ty 4 + 104t 2 y 4 + 468wy 4 335twy 4 + 184w2 y 4
(0)
(S0 )I1 I2 I3 I4
+ 468xy 4 335txy 4 392wxy 4 + 184x 2y 4 + 484y 5 256ty 5

119wy 5 119xy 5 + 80y 6 ,
1
a125 a345
=
576
164
18225 24300t + 292830t 2 71028t 3 111795t 4 + 33444t 5

+ 8640t 6 5124t 7 618t 8 + 192t 9 + 24t 10 24300w 256068tw
66756t 2w 120628t 3w + 177012t 4w 17836t 5w 26900t 6w
+ 132t 7 w + 800t 8 w + 32t 9 w + 292830w2 66756tw2
+ 177178t 2w2 + 160520t 3w2 94964t 4w2 31788t 5w2
+ 1416t 6w2 + 1136t 7 w2 + 112t 8 w2 71028w3 120628tw3
+ 160520t 2w3 11080t 3w3 46764t 4w3 3076t 5w3 + 2640t 6w3
+ 336t 7 w3 111795w4 + 177012tw4 94964t 2w4 46764t 3w4
+ 6638t 4w4 + 2920t 5 w4 + 224t 6 w4 + 33444w5 17836tw5
31788t 2w5 3076t 3w5 + 2920t 4w5 + 416t 5 w5 + 8640w6
26900tw6 + 1416t 2 w6 + 2640t 3w6 + 224t 4 w6 5124w7
+ 132tw7 + 1136t 2 w7 + 336t 3 w7 618w8 + 800tw8 + 112t 2 w8
+ 192w9 + 32tw9 + 24w10 24300x 256068tx 66756t 2x
120628t 3x + 177012t 4x 17836t 5x 26900t 6x + 132t 7 x
+ 800t 8 x + 32t 9 x 256068wx + 12816twx + 319756t 2wx
+ 370976t 3wx 220300t 4wx 57456t 5wx + 5828t 6wx
+ 1024t 7wx 96t 8 wx 66756w2x + 319756tw2x + 184632t 2w2 x
247264t 3w2 x 68020t 4w2 x 2724t 5w2 x + 1792t 6w2 x
+ 272t 7 w2 x 120628w3x + 370976tw3x 247264t 2w3 x
124960t 3w3 x + 6940t 4w3 x + 4000t 5w3 x + 352t 6 w3 x
+ 177012w4x 220300tw4x 68020t 2w4 x + 6940t 3w4 x
+ 1672t 4w4 x + 268t 5 w4 x 17836w5x 57456tw5x 2724t 2w5 x
+ 4000t 3w5 x + 268t 4 w5 x 26900w6x + 5828tw6 x + 1792t 2w6 x
+ 352t 3 w6 x + 132w7 x + 1024tw7 x + 272t 2 w7 x + 800w8x
96tw8 x + 32w9 x + 292830x 2 66756tx 2 + 177178t 2x 2
+ 160520t 3x 2 94964t 4x 2 31788t 5x 2 + 1416t 6x 2 + 1136t 7x 2
+ 112t 8 x 2 66756wx 2 + 319756twx 2 + 184632t 2wx 2
247264t 3wx 2 68020t 4wx 2 2724t 5wx 2 + 1792t 6wx 2
+ 272t 7 wx 2 + 177178w2x 2 + 184632tw2x 2 335076t 2w2 x 2
134456t 3w2 x 2 1458t 4w2 x 2 + 4296t 5 w2 x 2 + 848t 6 w2 x 2
+ 160520w3x 2 247264tw3x 2 134456t 2w3 x 2 + 7688t 3 w3 x 2
+ 4056t 4w3 x 2 + 592t 5w3 x 2 94964w4x 2 68020tw4x 2
1458t 2w4 x 2 + 4056t 3w4 x 2 56t 4 w4 x 2 31788w5x 2
2724tw5 x 2 + 4296t 2w5 x 2 + 592t 3w5 x 2 + 1416w6x 2
+ 1792tw6 x 2 + 848t 2 w6 x 2 + 1136w7x 2 + 272tw7 x 2 + 112w8 x 2
165
71028x 3 120628tx 3 + 160520t 2x 3 11080t 3x 3 46764t 4x 3

3076t 5x 3 + 2640t 6 x 3 + 336t 7 x 3 120628wx 3 + 370976twx 3
247264t 2wx 3 124960t 3wx 3 + 6940t 4wx 3 + 4000t 5wx 3
+ 352t 6 wx 3 + 160520w2x 3 247264tw2x 3 134456t 2w2 x 3
+ 7688t 3w2 x 3 + 4056t 4w2 x 3 + 592t 5 w2 x 3 11080w3x 3
124960tw3x 3 + 7688t 2w3 x 3 + 6720t 3w3 x 3 1096t 4 w3 x 3
46764w4x 3 + 6940tw4 x 3 + 4056t 2w4 x 3 1096t 3w4 x 3
3076w5x 3 + 4000tw5 x 3 + 592t 2 w5 x 3 + 2640w6x 3 + 352tw6 x 3
+ 336w7 x 3 111795x 4 + 177012tx 4 94964t 2x 4 46764t 3x 4
+ 6638t 4x 4 + 2920t 5 x 4 + 224t 6 x 4 + 177012wx 4 220300twx 4
68020t 2wx 4 + 6940t 3 wx 4 + 1672t 4wx 4 + 268t 5wx 4
94964w2x 4 68020tw2x 4 1458t 2 w2 x 4 + 4056t 3 w2 x 4
56t 4 w2 x 4 46764w3x 4 + 6940tw3x 4 + 4056t 2w3 x 4
1096t 3w3 x 4 + 6638w4x 4 + 1672tw4 x 4 56t 2 w4 x 4 + 2920w5x 4
+ 268tw5 x 4 + 224w6 x 4 + 33444x 5 17836tx 5 31788t 2x 5
3076t 3x 5 + 2920t 4 x 5 + 416t 5 x 5 17836wx 5 57456twx 5
2724t 2wx 5 + 4000t 3wx 5 + 268t 4 wx 5 31788w2x 5
2724tw2 x 5 + 4296t 2w2 x 5 + 592t 3w2 x 5 3076w3x 5
+ 4000tw3 x 5 + 592t 2 w3 x 5 + 2920w4x 5 + 268tw4 x 5 + 416w5 x 5
+ 8640x 6 26900tx 6 + 1416t 2x 6 + 2640t 3x 6 + 224t 4x 6
26900wx 6 + 5828twx 6 + 1792t 2wx 6 + 352t 3wx 6 + 1416w2x 6
+ 1792tw2 x 6 + 848t 2 w2 x 6 + 2640w3x 6 + 352tw3 x 6 + 224w4 x 6
5124x 7 + 132tx 7 + 1136t 2x 7 + 336t 3 x 7 + 132wx 7 + 1024twx 7
+ 272t 2 wx 7 + 1136w2 x 7 + 272tw2 x 7 + 336w3 x 7 618x 8
+ 800tx 8 + 112t 2 x 8 + 800wx 8 96twx 8 + 112w2 x 8 + 192x 9
+ 32tx 9 + 32wx 9 + 24x 10 24300y 256068ty 66756t 2y
120628t 3y + 177012t 4y 17836t 5y 26900t 6y + 132t 7 y
+ 800t 8 y + 32t 9 y 256068wy + 12816twy + 319756t 2wy
+ 370976t 3wy 220300t 4wy 57456t 5wy + 5828t 6wy
+ 1024t 7wy 96t 8 wy 66756w2y + 319756tw2y + 184632t 2w2 y
247264t 3w2 y 68020t 4w2 y 2724t 5w2 y + 1792t 6w2 y
+ 272t 7 w2 y 120628w3y + 370976tw3y 247264t 2w3 y
124960t 3w3 y + 6940t 4w3 y + 4000t 5w3 y + 352t 6 w3 y
+ 177012w4y 220300tw4y 68020t 2w4 y + 6940t 3w4 y
+ 1672t 4w4 y + 268t 5 w4 y 17836w5y 57456tw5y 2724t 2w5 y
166
+ 4000t 3w5 y + 268t 4 w5 y 26900w6y + 5828tw6 y + 1792t 2w6 y

+ 352t 3 w6 y + 132w7 y + 1024tw7 y + 272t 2 w7 y + 800w8y
96tw8 y + 32w9 y 256068xy + 12816txy + 319756t 2xy
+ 370976t 3xy 220300t 4xy 57456t 5xy + 5828t 6 xy + 1024t 7 xy
96t 8 xy + 12816wxy + 2404608twxy 48096t 2wxy
749568t 3wxy 41136t 4wxy + 4992t 5 wxy 2112t 6wxy
384t 7 wxy + 319756w2xy 48096tw2xy 403336t 2w2 xy
128736t 3w2 xy 11588t 4w2 xy + 2432t 5w2 xy + 688t 6 w2 xy
+ 370976w3xy 749568tw3xy 128736t 2w3 xy + 39680t 3w3 xy
+ 896t 4 w3 xy + 384t 5w3 xy 220300w4xy 41136tw4 xy
11588t 2w4 xy + 896t 3w4 xy + 608t 4 w4 xy 57456w5xy
+ 4992tw5 xy + 2432t 2 w5 xy + 384t 3w5 xy + 5828w6 xy
2112tw6 xy + 688t 2 w6 xy + 1024w7xy 384tw7 xy 96w8 xy
66756x 2y + 319756tx 2y + 184632t 2x 2 y 247264t 3x 2 y
68020t 4x 2 y 2724t 5 x 2 y + 1792t 6 x 2 y + 272t 7x 2 y
+ 319756wx 2y 48096twx 2y 403336t 2wx 2 y 128736t 3wx 2 y
11588t 4wx 2 y + 2432t 5wx 2 y + 688t 6wx 2 y + 184632w2x 2 y
403336tw2x 2 y 169560t 2w2 x 2 y + 17288t 3w2 x 2 y
+ 4136t 4w2 x 2 y + 1112t 5w2 x 2 y 247264w3x 2 y 128736tw3x 2 y
+ 17288t 2w3 x 2 y + 2496t 3w3 x 2 y 1856t 4w3 x 2 y 68020w4x 2 y
11588tw4x 2 y + 4136t 2w4 x 2 y 1856t 3w4 x 2 y 2724w5 x 2 y
+ 2432tw5 x 2 y + 1112t 2 w5 x 2 y + 1792w6x 2 y + 688tw6 x 2 y
+ 272w7 x 2 y 120628x 3y + 370976tx 3y 247264t 2x 3 y
124960t 3x 3 y + 6940t 4x 3 y + 4000t 5x 3 y + 352t 6 x 3 y
+ 370976wx 3y 749568twx 3y 128736t 2wx 3 y + 39680t 3wx 3 y
+ 896t 4 wx 3 y + 384t 5 wx 3 y 247264w2x 3 y 128736tw2x 3 y
+ 17288t 2w2 x 3 y + 2496t 3w2 x 3 y 1856t 4w2 x 3 y 124960w3x 3 y
+ 39680tw3x 3 y + 2496t 2w3 x 3 y 6912t 3w3 x 3 y + 6940w4 x 3 y
+ 896tw4 x 3 y 1856t 2w4 x 3 y + 4000w5x 3 y + 384tw5 x 3 y
+ 352w6 x 3 y + 177012x 4y 220300tx 4y 68020t 2x 4 y
+ 6940t 3x 4 y + 1672t 4x 4 y + 268t 5 x 4 y 220300wx 4y
41136twx 4y 11588t 2wx 4 y + 896t 3 wx 4 y + 608t 4 wx 4 y
68020w2x 4 y 11588tw2 x 4 y + 4136t 2 w2 x 4 y 1856t 3w2 x 4 y
+ 6940w3x 4 y + 896tw3 x 4 y 1856t 2w3 x 4 y + 1672w4x 4 y
+ 608tw4 x 4 y + 268w5 x 4 y 17836x 5y 57456tx 5y 2724t 2x 5 y
167
+ 4000t 3x 5 y + 268t 4 x 5 y 57456wx 5y + 4992twx 5 y

+ 2432t 2wx 5 y + 384t 3 wx 5 y 2724w2 x 5 y + 2432tw2 x 5 y
+ 1112t 2w2 x 5 y + 4000w3x 5 y + 384tw3 x 5 y + 268w4 x 5 y
26900x 6y + 5828tx 6y + 1792t 2x 6 y + 352t 3x 6 y + 5828wx 6y
2112twx 6y + 688t 2 wx 6 y + 1792w2x 6 y + 688tw2 x 6 y
+ 352w3 x 6 y + 132x 7y + 1024tx 7y + 272t 2 x 7 y + 1024wx 7y
384twx 7 y + 272w2 x 7 y + 800x 8y 96tx 8 y 96wx 8 y + 32x 9y
+ 292830y 2 66756ty 2 + 177178t 2y 2 + 160520t 3y 2 94964t 4y 2
31788t 5y 2 + 1416t 6y 2 + 1136t 7 y 2 + 112t 8 y 2 66756wy 2
+ 319756twy 2 + 184632t 2wy 2 247264t 3wy 2 68020t 4wy 2
2724t 5wy 2 + 1792t 6wy 2 + 272t 7 wy 2 + 177178w2y 2
+ 184632tw2y 2 335076t 2w2 y 2 134456t 3w2 y 2 1458t 4w2 y 2
+ 4296t 5w2 y 2 + 848t 6w2 y 2 + 160520w3y 2 247264tw3y 2
134456t 2w3 y 2 + 7688t 3w3 y 2 + 4056t 4 w3 y 2 + 592t 5 w3 y 2
94964w4y 2 68020tw4y 2 1458t 2 w4 y 2 + 4056t 3 w4 y 2
56t 4 w4 y 2 31788w5y 2 2724tw5y 2 + 4296t 2w5 y 2
+ 592t 3 w5 y 2 + 1416w6y 2 + 1792tw6y 2 + 848t 2w6 y 2 + 1136w7y 2
+ 272tw7 y 2 + 112w8 y 2 66756xy 2 + 319756txy 2 + 184632t 2xy 2
247264t 3xy 2 68020t 4xy 2 2724t 5xy 2 + 1792t 6xy 2
+ 272t 7 xy 2 + 319756wxy 2 48096twxy 2 403336t 2wxy 2
128736t 3wxy 2 11588t 4wxy 2 + 2432t 5wxy 2 + 688t 6 wxy 2
+ 184632w2xy 2 403336tw2xy 2 169560t 2w2 xy 2
+ 17288t 3w2 xy 2 + 4136t 4w2 xy 2 + 1112t 5w2 xy 2 247264w3xy 2
128736tw3xy 2 + 17288t 2w3 xy 2 + 2496t 3 w3 xy 2 1856t 4 w3 xy 2
68020w4xy 2 11588tw4 xy 2 + 4136t 2 w4 xy 2 1856t 3w4 xy 2
2724w5xy 2 + 2432tw5 xy 2 + 1112t 2 w5 xy 2 + 1792w6xy 2
+ 688tw6 xy 2 + 272w7 xy 2 + 177178x 2y 2 + 184632tx 2y 2
335076t 2x 2 y 2 134456t 3x 2 y 2 1458t 4 x 2 y 2 + 4296t 5x 2 y 2
+ 848t 6 x 2 y 2 + 184632wx 2y 2 403336twx 2y 2 169560t 2wx 2 y 2
+ 17288t 3wx 2 y 2 + 4136t 4wx 2 y 2 + 1112t 5wx 2 y 2 335076w2x 2 y 2
169560tw2x 2 y 2 + 62988t 2w2 x 2 y 2 + 10224t 3w2 x 2 y 2
3408t 4w2 x 2 y 2 134456w3x 2 y 2 + 17288tw3x 2 y 2
+ 10224t 2w3 x 2 y 2 9680t 3 w3 x 2 y 2 1458w4 x 2 y 2 + 4136tw4 x 2 y 2
3408t 2w4 x 2 y 2 + 4296w5x 2 y 2 + 1112tw5x 2 y 2 + 848w6 x 2 y 2
+ 160520x 3y 2 247264tx 3y 2 134456t 2x 3 y 2 + 7688t 3x 3 y 2
168
+ 4056t 4x 3 y 2 + 592t 5x 3 y 2 247264wx 3y 2 128736twx 3y 2

+ 17288t 2wx 3 y 2 + 2496t 3wx 3 y 2 1856t 4wx 3 y 2 134456w2x 3 y 2
+ 17288tw2x 3 y 2 + 10224t 2w2 x 3 y 2 9680t 3w2 x 3 y 2
+ 7688w3x 3 y 2 + 2496tw3x 3 y 2 9680t 2w3 x 3 y 2 + 4056w4x 3 y 2
1856tw4 x 3 y 2 + 592w5 x 3 y 2 94964x 4y 2 68020tx 4y 2
1458t 2x 4 y 2 + 4056t 3x 4 y 2 56t 4 x 4 y 2 68020wx 4y 2
11588twx 4y 2 + 4136t 2wx 4 y 2 1856t 3wx 4 y 2 1458w2 x 4 y 2
+ 4136tw2 x 4 y 2 3408t 2w2 x 4 y 2 + 4056w3x 4 y 2 1856tw3 x 4 y 2
56w4 x 4 y 2 31788x 5y 2 2724tx 5y 2 + 4296t 2x 5 y 2 + 592t 3 x 5 y 2
2724wx 5y 2 + 2432twx 5y 2 + 1112t 2 wx 5 y 2 + 4296w2x 5 y 2
+ 1112tw2 x 5 y 2 + 592w3 x 5 y 2 + 1416x 6y 2 + 1792tx 6y 2
+ 848t 2 x 6 y 2 + 1792wx 6y 2 + 688twx 6 y 2 + 848w2 x 6 y 2
+ 1136x 7y 2 + 272tx 7 y 2 + 272wx 7y 2 + 112x 8y 2 71028y 3
120628ty 3 + 160520t 2y 3 11080t 3y 3 46764t 4y 3 3076t 5 y 3
+ 2640t 6y 3 + 336t 7 y 3 120628wy 3 + 370976twy 3
247264t 2wy 3 124960t 3wy 3 + 6940t 4wy 3 + 4000t 5wy 3
+ 352t 6 wy 3 + 160520w2y 3 247264tw2y 3 134456t 2w2 y 3
+ 7688t 3w2 y 3 + 4056t 4w2 y 3 + 592t 5 w2 y 3 11080w3y 3
124960tw3y 3 + 7688t 2w3 y 3 + 6720t 3w3 y 3 1096t 4 w3 y 3
46764w4y 3 + 6940tw4 y 3 + 4056t 2w4 y 3 1096t 3w4 y 3
3076w5y 3 + 4000tw5 y 3 + 592t 2 w5 y 3 + 2640w6y 3 + 352tw6 y 3
+ 336w7 y 3 120628xy 3 + 370976txy 3 247264t 2xy 3
124960t 3xy 3 + 6940t 4xy 3 + 4000t 5xy 3 + 352t 6 xy 3
+ 370976wxy 3 749568twxy 3 128736t 2wxy 3 + 39680t 3wxy 3
+ 896t 4 wxy 3 + 384t 5 wxy 3 247264w2xy 3 128736tw2xy 3
+ 17288t 2w2 xy 3 + 2496t 3w2 xy 3 1856t 4w2 xy 3 124960w3xy 3
+ 39680tw3xy 3 + 2496t 2w3 xy 3 6912t 3w3 xy 3 + 6940w4 xy 3
+ 896tw4 xy 3 1856t 2w4 xy 3 + 4000w5xy 3 + 384tw5 xy 3
+ 352w6 xy 3 + 160520x 2y 3 247264tx 2y 3 134456t 2x 2 y 3
+ 7688t 3x 2 y 3 + 4056t 4x 2 y 3 + 592t 5 x 2 y 3 247264wx 2y 3
128736twx 2y 3 + 17288t 2wx 2 y 3 + 2496t 3 wx 2 y 3 1856t 4 wx 2 y 3
134456w2x 2 y 3 + 17288tw2x 2 y 3 + 10224t 2w2 x 2 y 3
9680t 3w2 x 2 y 3 + 7688w3x 2 y 3 + 2496tw3x 2 y 3 9680t 2w3 x 2 y 3
+ 4056w4x 2 y 3 1856tw4x 2 y 3 + 592w5 x 2 y 3 11080x 3y 3
124960tx 3y 3 + 7688t 2x 3 y 3 + 6720t 3x 3 y 3 1096t 4 x 3 y 3
169
124960wx 3y 3 + 39680twx 3y 3 + 2496t 2wx 3 y 3 6912t 3 wx 3 y 3

+ 7688w2x 3 y 3 + 2496tw2x 3 y 3 9680t 2w2 x 3 y 3 + 6720w3x 3 y 3
6912tw3 x 3 y 3 1096w4x 3 y 3 46764x 4y 3 + 6940tx 4y 3
+ 4056t 2x 4 y 3 1096t 3x 4 y 3 + 6940wx 4y 3 + 896twx 4 y 3
1856t 2wx 4 y 3 + 4056w2x 4 y 3 1856tw2 x 4 y 3 1096w3x 4 y 3
3076x 5y 3 + 4000tx 5y 3 + 592t 2 x 5 y 3 + 4000wx 5y 3 + 384twx 5y 3
+ 592w2 x 5 y 3 + 2640x 6y 3 + 352tx 6y 3 + 352wx 6y 3 + 336x 7y 3
111795y 4 + 177012ty 4 94964t 2y 4 46764t 3y 4 + 6638t 4y 4
+ 2920t 5y 4 + 224t 6 y 4 + 177012wy 4 220300twy 4 68020t 2wy 4
+ 6940t 3wy 4 + 1672t 4wy 4 + 268t 5 wy 4 94964w2y 4
68020tw2y 4 1458t 2 w2 y 4 + 4056t 3 w2 y 4 56t 4 w2 y 4
46764w3y 4 + 6940tw3 y 4 + 4056t 2w3 y 4 1096t 3w3 y 4
+ 6638w4y 4 + 1672tw4 y 4 56t 2 w4 y 4 + 2920w5y 4 + 268tw5 y 4
+ 224w6 y 4 + 177012xy 4 220300txy 4 68020t 2xy 4
+ 6940t 3xy 4 + 1672t 4xy 4 + 268t 5 xy 4 220300wxy 4
41136twxy 4 11588t 2wxy 4 + 896t 3 wxy 4 + 608t 4 wxy 4
68020w2xy 4 11588tw2 xy 4 + 4136t 2 w2 xy 4 1856t 3w2 xy 4
+ 6940w3xy 4 + 896tw3 xy 4 1856t 2w3 xy 4 + 1672w4xy 4
+ 608tw4 xy 4 + 268w5 xy 4 94964x 2y 4 68020tx 2y 4
1458t 2x 2 y 4 + 4056t 3x 2 y 4 56t 4 x 2 y 4 68020wx 2y 4
11588twx 2y 4 + 4136t 2wx 2 y 4 1856t 3wx 2 y 4 1458w2 x 2 y 4
+ 4136tw2 x 2 y 4 3408t 2w2 x 2 y 4 + 4056w3x 2 y 4 1856tw3 x 2 y 4
56w4 x 2 y 4 46764x 3y 4 + 6940tx 3y 4 + 4056t 2x 3 y 4
1096t 3x 3 y 4 + 6940wx 3y 4 + 896twx 3y 4 1856t 2wx 3 y 4
+ 4056w2x 3 y 4 1856tw2x 3 y 4 1096w3x 3 y 4 + 6638x 4y 4
+ 1672tx 4y 4 56t 2 x 4 y 4 + 1672wx 4y 4 + 608twx 4y 4 56w2 x 4 y 4
+ 2920x 5y 4 + 268tx 5 y 4 + 268wx 5y 4 + 224x 6y 4 + 33444y 5
17836ty 5 31788t 2y 5 3076t 3 y 5 + 2920t 4y 5 + 416t 5 y 5
17836wy 5 57456twy 5 2724t 2wy 5 + 4000t 3wy 5 + 268t 4 wy 5
31788w2y 5 2724tw2 y 5 + 4296t 2w2 y 5 + 592t 3w2 y 5
3076w3y 5 + 4000tw3 y 5 + 592t 2 w3 y 5 + 2920w4y 5 + 268tw4 y 5
+ 416w5 y 5 17836xy 5 57456txy 5 2724t 2xy 5 + 4000t 3xy 5
+ 268t 4 xy 5 57456wxy 5 + 4992twxy 5 + 2432t 2wxy 5
+ 384t 3 wxy 5 2724w2xy 5 + 2432tw2 xy 5 + 1112t 2 w2 xy 5
+ 4000w3xy 5 + 384tw3 xy 5 + 268w4 xy 5 31788x 2y 5
170
2724tx 2y 5 + 4296t 2x 2 y 5 + 592t 3x 2 y 5 2724wx 2y 5

+ 2432twx 2y 5 + 1112t 2 wx 2 y 5 + 4296w2x 2 y 5 + 1112tw2 x 2 y 5
+ 592w3 x 2 y 5 3076x 3y 5 + 4000tx 3y 5 + 592t 2x 3 y 5 + 4000wx 3y 5
+ 384twx 3 y 5 + 592w2 x 3 y 5 + 2920x 4y 5 + 268tx 4y 5 + 268wx 4y 5
+ 416x 5y 5 + 8640y 6 26900ty 6 + 1416t 2y 6 + 2640t 3y 6
+ 224t 4 y 6 26900wy 6 + 5828twy 6 + 1792t 2wy 6 + 352t 3wy 6
+ 1416w2y 6 + 1792tw2 y 6 + 848t 2 w2 y 6 + 2640w3y 6 + 352tw3 y 6
+ 224w4 y 6 26900xy 6 + 5828txy 6 + 1792t 2xy 6 + 352t 3xy 6
+ 5828wxy 6 2112twxy 6 + 688t 2 wxy 6 + 1792w2xy 6
+ 688tw2 xy 6 + 352w3 xy 6 + 1416x 2y 6 + 1792tx 2y 6 + 848t 2 x 2 y 6
+ 1792wx 2y 6 + 688twx 2 y 6 + 848w2 x 2 y 6 + 2640x 3y 6 + 352tx 3y 6
+ 352wx 3y 6 + 224x 4y 6 5124y 7 + 132ty 7 + 1136t 2y 7 + 336t 3 y 7
+ 132wy 7 + 1024twy 7 + 272t 2wy 7 + 1136w2y 7 + 272tw2 y 7
+ 336w3 y 7 + 132xy 7 + 1024txy 7 + 272t 2 xy 7 + 1024wxy 7
384twxy 7 + 272w2 xy 7 + 1136x 2y 7 + 272tx 2y 7 + 272wx 2y 7
+ 336x 3y 7 618y 8 + 800ty 8 + 112t 2 y 8 + 800wy 8 96twy 8
(S1 )(0)
I1 I2 I3 I4
+ 112w2 y 8 + 800xy 8 96txy 8 96wxy 8 + 112x 2y 8 + 192y 9

+ 32ty 9 + 32wy 9 + 32xy 9 + 24y 10 ,
2
= f51 a125a345(k1 k2 )(k3 k4 )(f1 f2 )(f3 f4 )
36 + 2(k1 + k2 + k3 + k4 ) + f1 + f2 + f3 + f4

2k1k2 2k3k4 ,
(f5 1)2 t125 t345
(k1 k2 )(k3 k4 ) f1 + f2 + f3 + f4
2

+ 2(k1 + k2 + k3 + k4 ) 2(k1 k2 + k3 k4 ) 36 ,
1
= f53 p125 p345 ,
2 2
= f5 p125 p345 k12 + k22 + k32 + k42

2(k1 + k2 + k3 + k4 ) + 2(k1 k2 + k3 k4 ) ,
9a125a345
(1 + k1 k2 )(1 + k1 k2 )(3 + k1 + k2 )
=
64(f5 5)
(5 + k1 + k2 )(1 + k3 k4 )(1 + k3 k4 )(3 + k3 + k4 )
(St 2 )(0)
I1 I2 I3 I4 =
(0)
(Sp3 )I1 I2 I3 I4
(Sp2 )(0)
I1 I2 I3 I4
(0)
(Sd )I1 I2 I3 I4
(5 + k3 + k4 ) 2k12 + 2k22 + 2k32 + 2k42 + 4k1 k2 + 4k3 k4

4(k1 + k2 + k3 + k4 ) 5 .
171
Appendix B
We need a number of relations between the structure constants a123 , t123 and p123 ,
defined as 15
Z
a123 = Y I1 Y I2 Y I3 ,
(B.1)
Z
(B.2)
t123 = Y I1 Y I2 YI3 ,
Z
I3
.
(B.3)
p123 = Y I1 Y I2 Y()
k one comes
In deriving the equations of motions for scalar fields sk and tk and for tensor ab
across a number of integrals of scalar spherical harmonics, all of them can be reduced
to a123 . Introducing the concise notation fi = f (ki ) = ki (ki + 4) we present below the
corresponding formulae:
Z
1
b123 = Y I1 Y I2 Y I3 = (f1 + f2 f3 )a123,
2
Z
I1
I2 I3
c123 = Y Y Y

1
3
1
=
f1 f3 f2 f3 + f1 f2 + f12
2
5
2

1 2 1 2
+ f2 + f3 4(f1 + f2 f3 ) a123.
2
2
Since any scalar function on a sphere can be decomposed in scalar spherical harmonics,
we have the following relations
Y 1 Y 2 = a123Y 3 ,
Y 1 Y 2 = b123Y 3 ,
Y 1 Y 2 = c123Y 3 .
(B.4)
These relations allow one to calculate some integrals involving 4 scalar spherical
harmonics. In particular we have
Z
a1234 = Y I1 Y I2 Y I3 Y I4 = a125a345 = a135a245 = a145a235,
Z
b1234 = Y I1 Y I2 Y I3 Y I4 = a125 b345,
Z
c1234 = Y I1 Y I2 Y I3 Y I4 = a125c345 ,
where the summation over the index 5 is assumed.
There is the following important relation
f5 (a125 a345 + a135a245 + a235a145) = (f1 + f2 + f3 + f4 )a125 a345,
15 For a detailed description of spherical harmonics on S 5 see [5,6].
(B.5)
172
that shows that among the three tensors f5 a125a345 , f5 a135 a245 and f5 a235 a145 only the
following two tensors are independent
f5 a125 a345
and f5 (a135a245 a235a145).
We also encounter tensors of the form f5n t125 t345 and f5n p125 p345 . Some of them can be
reduced to sums of tensors of the form f5n a125 a345 . To see this we note that any vector and
any traceless symmetric tensor on a sphere can be decomposed as follows
Y 1 Y 2 = t125Y5 +
b152
Y 5 ,
f5
5
5
+ 125( Y)
+ 125 ( ) Y 5 ,
( Y 1 ) Y 2 = p125 Y()
(B.6)
(B.7)
where
125 =
f2 f1
t125,
f5 5
4
q5 = f5 (f5 5),
5
1
d125,
q5

1
f5 f2 + 4 b125 + f1 b251.
d125 = c125 +
5
125 =
Note that in the paper we use the following normalizations

Z
Z
()
I
YJ = JI ,
YI YJ = JI ,
Y()
where summation over , is assumed.
By using the decompositions one can find the following relations
(f1 f2 )(f3 f4 )
1
a125a345 + f5 (a145a235 a245a135),
(B.8)
4f5
4

1
(1 f5 )t125 t345 = f52 f5 (f1 + f2 + f3 + f4 4) (a145 a235 a135a245 )
4
4 f5
(f1 f2 )(f3 f4 )a125a345
(B.9)
4f5
(f1 f2 )(f3 f4 )
1
t125 t345 d125d345
p125 p345 =
2(f5 5)
q5
1
(f1 + f2 f5 )(f3 + f4 f5 )a125 a345
20
1
+ (f1 + f3 f5 )(f2 + f4 f5 )a135a245
8
1
+ (f1 + f4 f5 )(f2 + f3 f5 )a145a235,
(B.10)
8
p125 (2 f5 )p345
Z
10 f5
= 2 ( Y 1 Y 2 ) ( Y 3 ) Y 4
d125d345
q5
(f1 f2 )(f3 f4 )
1
t125 t345 + (f1 f2 )(f3 f4 )t125 t345 ,
(B.11)
(f5 5)
2
t125 t345 =
where
173

2 Y 1 Y 2 ( Y 3 ) Y 4
1
= (8 f1 f2 )(f1 + f3 f5 )(f2 + f4 f5 )a135a245
8
1
+ (8 f1 f2 )(f1 + f4 f5 )(f2 + f3 f5 )a145a235
8
1
+ (f1 + f2 f5 )(f3 + f4 f5 )f5 a125a345 + g1324 + g1423
20
and
1
(f1 + f5 f2 )(f1 + f2 f5 )(f3 + f5 f4 )(f3 + f4 f5 )a125 a345
16f5
1
1
+ (f1 + f2 4)(f3 + f4 4)t125 t345 + (f1 + f2 + f3 + f4 8)t125
4
4
1
2
(1 f5 )t345 + t125 (1 f5 ) t345 .
4
The following integrals are also used in deriving the equations of motion for scalars sk :
Z

1
I
( ) Y I1 Y I2 Y3 = f3 + f1 f2 5 t123 ,
2
Z

1
3
I
( ) Y I1 Y I2 Y3 = f2 + f1 f3 3 t123 ,
2
5
Z
1
I
3
= (3f1 + 5f2 5f3 30)p123,
( ) Y I1 Y I2 Y()
10
Z
1
I3
= (f1 f2 f3 8)p123.
( ) Y I1 Y I2 Y()
2
g1234 =
Considering the action for the fields from the massless graviton multiplet we need the
following explicit expression for the integral a123 of scalar spherical harmonics [5,6]:
a123 = (z(k1 )z(k2 )z(k3 ))1/2
3
( 12 + 2)!2
1
2 (2)
k1 !k2 !k3 !
I1 I2 I3
C C C .
1 !2 !3 !
(B.12)
Here
z(k) =
3
2k1 (k + 1)(k + 2)
and i = 12 (kj + kl ki ), j 6= l 6= i. Notation hC I1 C I2 C I3 i is used to denote the unique

SO(6) invariant obtained by contracting 1 indices between C I2 and C I3 , 2 indices
between C I3 and C I1 , and 3 indices between C I2 and C I1 .
Acknowledgement
We are grateful to Prof. S. Theisen and to S. Kuzenko for numerous valuable discussions.
We would like to thank A. Tseytlin for valuable comments. The work of G.A. was
174
supported by the Alexander von Humboldt Foundation and in part by the RFBI grant N9901-00166, and the work of S.F. was supported by the U.S. Department of Energy under
grant No. DE-FG02-96ER40967.
References
[1] J. Maldacena, The large N limit of superconformal field theories and supergravity, Adv. Theor.
Math. Phys. 2 (1998) 231252.
[2] G.G. Gubser, I.R. Klebanov, A.M. Polyakov, Gauge theory correlators from noncritical string
theory, Phys. Lett. B 428 (1998) 105114, hep-th/9802109.
[3] E. Witten, Anti-de Sitter space and holography, Adv. Theor. Math. Phys. 2 (1998) 253291,
hep-th/9802150.
[4] G. Arutyunov, S. Frolov, Quadratic action for type IIB supergravity on AdS5 S 5 , JHEP 9908
(1999) 024, hep-th/9811106.
[5] S. Lee, S. Minwalla, M. Rangamani, N. Seiberg, Adv. Theor. Math. Phys. 2 (1998) 697, hepth/9806074.
[6] G. Arutyunov, S. Frolov, Some cubic couplings in type IIB supergravity on AdS5 S 5 and
three-point functions in SYM4 at large N , Phys. Rev. D 61 (2000) 064009, hep-th/9907085.
[7] S. Lee, AdS(5)/CFT(4) four-point functions of chiral primary operators: Cubic vertices, Nucl.
Phys. B 563 (1999) 349360, hep-th/9907108.
[8] I.Ya. Arefeva, I.V. Volovich, On large N conformal field theories, field theories in anti-de Sitter
space and singletons, hep-th/9803028.
[9] M. Henningson, K. Sfetsos, Spinors and the AdS/CFT correspondence, Phys. Lett. B 431 (1998)
6368, hep-th/9803251.
[10] W. Mck, K.S. Viswanathan, Phys. Rev. D 58 (1998) 841901.
[11] D. Freedman, S. Mathur, A. Matusis, L. Rastelli, Nucl. Phys. B 546 (1999) 96118, hepth/9804058.
[12] H. Liu, A.A. Tseytlin, D = 4 super-YangMills, D = 5 gauge supergravity, and D = 4
conformal supergravity, Nucl. Phys. B 553 (1998) 88, hep-th/9804083.
[13] G. Chalmers, H. Nastase, K. Schalm, R. Siebelink, Nucl. Phys. B 540 (1999) 247, hepth/9805105.
[14] W. Mck, K.S. Viswanathan, Phys. Rev. D 58 (1998) 106006.
[15] A. Ghezelbash, K. Kaviani, S. Parvizi, A. Fatollahi, Phys. Lett. B 435 (1998) 291298.
[16] G. Arutyunov, S. Frolov, On the origin of the supergravity boundary terms in the AdS/CFT
correspondence, Nucl. Phys. B 544 (1999) 576589, hep-th/9806216.
[17] G. Arutyunov, S. Frolov, Antisymmetric tensor field on AdS5 , Phys. Lett. B 441 (1998) 173
177.
[18] E. DHoker, D. Freedman, W. Skiba, Field theory tests for correlators in the AdS/CFT
correspondence, Phys. Rev. D 59 (1999) 045008, hep-th/9807098.
[19] P.S. Howe, E. Sokatchev, P.C. West, Phys. Lett B 444 (1998) 341, hep-th/9808162.
[20] S. Corley, Phys. Rev. D 59 (1999) 086003, hep-th/9808184.
[21] A. Volovich, JHEP 9809 (1998) 022, hep-th/9809009.
[22] W.S. lYi, Generating functionals of correlation functions of p-form currents in the AdS/CFT
correspondence, hep-th/9809132.
[23] W. Mck, K.S. Viswanathan, The graviton in the AdS/CFT correspondence: Solution via the
Dirichlet boundary value problem, hep-th/9810151.
[24] L. Chekhov, AdS/CFT correspondence on torus, hep-th/9811146.
[25] A. Koshelev, O. Rytchkov, Phys. Lett B 450 (1999) 368376, hep-th/9812238.
[26] G. Arutyunov, S. Frolov, Phys. Rev. D 60 (1999) 026004, hep-th/9901121.
175
[27] A. Polishchuk, Massive symmetric tensor field on AdS, JHEP 9907 (1999) 007, hep-th/9905048.
[28] H. Liu, A.A. Tseytlin, Dilaton-fixed scalar correlators and AdS5 S 5 -SYM correspondence,
JHEP 9910 (1999) 003, hep-th/9906151.
[29] E. DHoker, D.Z. Freedman, S.D. Mathur, A. Matusis, L. Rastelli, Extremal correlators in the
AdS/CFT correspondence, hep-th/9908160.
[30] J.H. Schwarz, Nucl. Phys. B 226 (1983) 269.
[31] J.H. Schwarz, P.C. West, Phys. Lett. B 126 (1983) 301.
[32] P.S. Howe, P.C. West, Nucl. Phys. B 238 (1984) 181.
[33] H. Liu, A.A. Tseytlin, Phys. Rev. D 59 (1999) 086002, hep-th/9807097.
[34] D. Freedman, S.D. Mathur, A. Matusis, L. Rastelli, Phys. Lett. B 452 (1999) 61, hepth/9808006.
[35] G. Chalmers, K. Schalm, The large Nc limit of four-point functions in N = 4 super YangMills
theory from anti-de Sitter supergravity, Nucl. Phys. B 554 (1999) 215236, hep-th/9810051.
[36] E. DHoker, D. Freedman, Gauge boson exchange in AdSd+1 , Nucl. Phys. B 544 (1999) 612
632, hep-th/9809179.
[37] J.H. Brodie, M. Gutperle, String corrections to 4-point functions in the AdS/CFT correspondence, Phys. Lett. B 445 (1999) 296306, hep-th/9809067.
[38] H. Liu, Scattering in anti-de Sitter space and operator product expansion, Phys. Rev. D 60 (1999)
106005, hep-th/9811152.
[39] E. DHoker, D. Freedman, Nucl. Phys. B 550 (1999) 612632, hep-th/9811257.
[40] E. DHoker, D. Freedman, S. Mathur, A. Matusis, L. Rastelli, Graviton exchange and complete
4-point functions in the AdS/CFT correspondence, Nucl. Phys. B 562 (1999) 353394, hepth/9903196.
[41] E. DHoker, D. Freedman, L. Rastelli, AdS/CFT 4-point functions: How to succeed at zintegrals without really trying, Nucl. Phys. B 562 (1999) 395411, hep-th/9905049.
[42] B. Eden, P.S. Howe, C. Schubert, E. Sokatchev, P.C. West, Simplifications of four-point
functions in N = 4 supersymmetric YangMills theory at two loops, Phys. Lett. B 466 (1999)
2026, hep-th/9906051.
[43] Sanjay, On direct and crossed channel asymptotics of four-point functions in AdS/CFT
correspondence, Mod. Phys. Lett. A 14 (1999) 14131420, hep-th/9906099.
[44] M. Bianchi, S. Kovacs, G. Rossi, Y.S. Stanev, On logarithmic behaviour in N = 4 SYM theory,
JHEP 9908 (1999) 020, hep-th/9906188.
[45] E. DHoker, S.D. Mathur, A. Matusis, L. Rastelli, The operator product expansion of N = 4
SYM and the 4-point functions of supergravity, hep-th/9911222.
[46] L. Andrianopoli, S. Ferrara, On short and long SU(2, 2/4) in the AdS/CFT correspondence,
hep-th/9812067.
[47] H.J. Kim, L.J. Romans, P. van Nieuwenhuizen, Mass spectrum of chiral ten-dimensional N = 2
supergravity on S 5 , Phys. Rev. D 32 (1985) 389.
[48] M. Gnaydin, N. Marcus, The spectrum of the S 5 compactification of the chiral N = 2 D = 10
supergravity and the unitary supermultiplets of U (2, 2|4), Class. Quant. Grav. 2 (1985) L11.
[49] H. Nastase, D. Vaman, P. van Nieuwenhuizen, Consistent nonlinear KK reduction of 11d
supergravity on AdS7 S4 and self-duality in odd dimensions, hep-th/9905075.
[50] H. Nastase, D. Vaman, P. van Nieuwenhuizen, Consistency of the AdS7 S4 reduction and the
origin of self-duality in odd dimensions, hep-th/9911238.
[51] M. Pernici, K. Pilch, P. van Nieuwenhuizen, Gauge N = 8, d = 5 supergravity, Nucl. Phys.
B 259 (1985) 460472.
[52] M. Gnaydin, L.J. Romans, N. Warner, Gauged N = 8 supergravity in five dimensions, Phys.
Lett. 154B (1985) 268274.
[53] J. de Boer, E. Verlinde, H. Verlinde, On the holographic renormalization group, hep-th/9912012.
[54] B. Eden, P.S. Howe, C. Schubert, E. Sokatchev, P.C. West, Extremal correlators in fourdimensional SCFT, hep-th/9910150.
176
[55] J. Erdmenger, M. Perez-Victoria, Non-renormalization of next-to-extremal correlators in N = 4

SYM and the AdS/CFT correspondence, hep-th/9912250.
[56] G. Arutyunov, S. Frolov, Four-point functions of lowest weight CPOs in N = 4 SYM4 in
supergravity approximation, hep-th/0002170.
[57] E. DHoker, J. Erdmenger, D. Freedman, M. Prez-Victoria, Near-extremal correlators and
vanishing supergravity couplings in AdS/CFT, hep-th/0003218.

Absorption cross section of scalar field

in supergravity background
R. Manvelyan a,1 , H.J.W. Mller-Kirsten a, , J.-Q. Liang a,b,2 ,
Yunbo Zhang a,b,c,3
a Department of Physics, University of Kaiserslautern, D-67653 Kaiserslautern, Germany
b Department of Physics and Institute of Theoretical Physics, Shanxi University, Taiyuan, Shanxi 030006, China
c Laboratory of Computational Physics, Beijing Institute of Applied Mathematics and Computational
Mathematics, Beijing 100 088, China

Abstract
It has recently been shown that the equation of motion of a massless scalar field in the background
of some specific p-branes can be reduced to a modified Mathieu equation. In the following the
absorption rate of the scalar by a D3 brane in ten dimensions is calculated in terms of modified
Mathieu functions of the first kind, using standard Mathieu coefficients. The relation of the latter to
Dougall coefficients (used by others) is investigated. The S-matrix obtained in terms of modified
Mathieu functions of the first kind is easily evaluated if known rapidly convergent low energy
expansions of these in terms of products of Bessel functions are used. Leading order terms, including
the interesting logarithmic contributions, can be obtained analytically. 2000 Elsevier Science B.V.
All rights reserved.
PACS: 11.25.-w; 02.30.Gp; 04.62.+v
1. Introduction
Recently the equations of motion of several cases of massless scalar fields propagating
in a supergravity background describing p-brane solitons have been shown to be reducible
to a Schrdinger-like equation with a singular potential and hence to a modified Mathieu
equation, so that various aspects, such as absorption probabilities, become exactly
calculable [1,2], which by AdS/CFT correspondence may yield information on correlation
Corresponding author. E-mail: mueller1@physik.uni-kl.de; tel.: (0631)-205-2375; fax: (0631)-205-3907.
1 manvel@phys.uni-kl.de; On leave from Yerevan Physics Institute.
2 jqliang@physik.uni-kl.de; jqliang@mail.sxu.edu.cn
3 ybzhang@physik.uni-kl.de; ybzhang5@yahoo.com
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 2 2 2 - 4
178
R. Manvelyan et al. / Nuclear Physics B 579 (2000) 177208
functions in a related worldvolume effective field theory. The singular potential appearing
in the coefficients of the metric is in the case of the D3-brane the Coulomb potential in
6 spatial dimensions. In view of the fact, that very few such exactly solvable cases are
known and that a Mathieu-type equation arises in a number of such problems as a result
of the invariance of the wave equation under various diagonal dimensional reductions on
the worldvolume, these theories are of exceptional importance and deserve to be studied
in full detail. The two recent investigations [1,2] of the absorption of partial waves of
a massless scalar field by D3 branes in 10 dimensions [1] and by a dyonic string in
six dimensions (or a D1/D5 brane intersection in 10 dimensions or extremal 2-charge
black hole in 5 dimensions or M2/M5 brane intersection in 11 dimensions) [2] study
the resulting modified Mathieu equation in terms of expansion coefficients introduced by
Dougall [3] in 1916 and present very few details. It is not possible to follow the calculations
of these papers without extensive work of ones own, which is made even more difficult
by singularities of expansion coefficients that require additional attention. These studies
are, in fact, complicated applications of modified Mathieu functions, which, in our opinion
become even more complicated if instead of standard Mathieu coefficients, i.e., those in
modern texts, the coefficients of Dougall are used, for the calculation of which the authors
of Ref. [1] developed in addition their own algorithm. The significance of the Mathieu
equation in such contexts can also be seen from a different angle since the equation
occurs also as the appropriately transformed small fluctuation equation in the study of
BornInfeld theory in the bosonic light-brane approximation with only an electric field
E = in p = 3 dimensions and the remaining components of the vector potential as
massless scalar fields, reduced to only one field y in the simplest case. One can show that
finite energy configurations of these fields independent of one another are not stable, but
their combination with appropriate boundary condition [4,5] (equivalent to the Dirichlet
boundary condition) is (with supersymmetry) a BPS state with corresponding Bogomolnyi
equation. Investigating the stability of this configuration, i.e., the D3-brane, with respect to
transverse fluctuations of both the throat or fundamental string and the brane, one again
arrives at an equation with the singular potential 1/r 4 [4,6,7] which can be converted
into a modified Mathieu equation [8,9]. Thus in each of these cases a Schrdinger-like
equation is obtained with the singular potential 1/r 4 . Such potentials have been the subject
of investigation 30 years ago and were motivated by the lack of understanding of weak
interactions at that time. 4 Thus the potential 1/r 4 and the associated scattering problem
had also been investigated, and various forms of the S-matrix had been given [8,9,11,12]
in terms of modified Mathieu functions or related functions for which at the latest since
the publication of Refs. [13] and [14] widely used definitions and notations exist.
In view of the scarcity of fully solvable examples of theories on a supergravity
background we consider it worthwhile to reexamine the case of the propagation of a
massless scalar field in the presence of a 3-brane by using modified Mathieu functions with
standard Mathieu coefficients and the S-matrix evaluated in terms of these. In our opinion
these calculations are more transparent than those using Dougall coefficients and are easier
4 For references to earlier literature see [10].
179
to follow with reference to modern literature on the subject. In view of the complexity of the
calculations, due also to the fact that later iterations contribute to earlier lower order terms,
we present these in some detail. Our presentation below should therefore also enable others
to follow the reasoning, and this particularly since leading order terms can be understood
without resorting to numerical methods.
In the following we first formulate the semiclassical gravity problem and reduce it to the
modified Mathieu equation. We do not rederive the S-matrix, but recapitulate in Appendix
A the main steps in the derivation, and in particular some steps that have not been written
out explicitly in Ref. [9], this being the prime reference on which our considerations are
based. We then consider briefly the gauge field theory approach in a simplified Born
Infeld version to demonstrate how this also leads to the Mathieu equation. Following this
we consider the Floquet exponent associated with Mathieu functions and show how this
has to be calculated in singular cases (such as the the cases to be considered here and in the
S-wave case already in the dominant approximation). The calculation of coefficients of
series expansions of modified Mathieu functions is then considered and the Dougall
coefficients used in Refs. [1,2] are compared with ordinary, i.e., standard, Mathieu
coefficients as in Ref. [13]. We calculate examples in singular and asymptotic cases (the
latter being those that permit one to ignore the singularities of early coefficients). Higher
order contributions are obtained with Mathematica. We show that Dougall coefficients are
more difficult to obtain than ordinary coefficients an observation that may explain why
Dougall did not evaluate any of his own coefficients in his work of 1916. We then evaluate
the relevant quantities appearing in the S-matrix and hence the absorption probabilities and
cross sections. Where comparable, our results can be seen to agree with those of Ref. [1].
The treatment presented below makes full use of the well established theory of the Mathieu
equation and can therefore point the way to explore other aspects, such as application to
double-centered D3 branes and to higher energies which have been discussed recently [15].
2. The scalar field in the D3-brane metric

The supergravity background for an extremal Dp-brane in the 10-dimensional type IIB
theory is [1618]

1
2
dt 2 + dxk2 + H dx
,
(1)
ds 2 =
H
where (r being the radial coordinate in the SO(5) symmetric space orthogonal to the
branes)
dxk2 =
p
X
dxi2 ,
2
2
dx
= dr 2 + r 2 d(8p)
(2)
i=1
and the harmonic function H is given by

H =1+
R (7p)
.
r (7p)
(3)
180
For p = 3, the case of interest here, i.e., the D3 brane coupled to the 4-form RRpotential [18,19], R with R 4 = 4gs N 02 (gs the string coupling and N the number of D3
branes) is the radius of S 5 and AdS5 in the socalled decoupling limit in which one obtains
a duality between N = 4 U (N) supersymmetric YangMills theory in 4 dimensions and
string theory in the near horizon AdS5 S 5 background [20,21]. As pointed out in Ref. [1],
for a comparison of considerations in terms of supergravity and those in terms of D-branes,
one is interested in the domain of small R, where is the energy of the field incident on
the brane.
For a massless scalar fluctuation field around the dilaton field given by [16]
e = H (3p)/4(r)
(which is constant for p = 3) in the background of this metric, the equation of motion is
1
g g = 0.
g
(4)
After separation of the S 5 harmonics Y (i ), in particular the Gegenbauer polynomial

Cl (cos ), where x = r cos , and a factor eit the radial wave function l (r) = y(r)/r 5/2
of the lth partial wave of energy of the scalar field is found to satisfy

2 R 4
l(l + 4)
1
5
2
+ + 4 l (r) = 0,
r
r 5 r
r
r2
r

2
2 R4
(l
+
3/2)(l
+
5/2)
2
+
y = 0.
(5)
r 2
r4
r2
We see that for R 4 6= 0 this is the equation of an attractive singular potential with coupling
constant g02 = 2 R 4 . For 2 > 0 an incident wave allows both transmitted and reflected
waves, and from the ratio of coefficients one can determine the S-matrix. It is convenient
to make the substitutions
r = ez ,
y = r 1/2 (r),
h = g0 = R ,
2
= g0 / h,
= (l + 2)2 ,
(6)
which convert the range of r from 0 to to that of z from to +. The equation

thereby becomes the modified Mathieu equation

d 2 2
+ 2h cosh 2z = 0.
(7)
2
dz
In view of the principal interest in the relation of our semiclassical gravity consideration
with the superconformal limit of the dual theory in the near-horizon domain, we are here
interested in waves of low energy, i.e., of small , and so in solutions of the modified
Mathieu equation around h2 = 0. The modified Mathieu equation allows series expansions
of this type in terms of exponential, hyperbolic and cylindrical functions and (surprisingly)
(h2 ), where is the Floquet exponent
in each of the cases with the same coefficients c2r
and the subscript r a positive or negative integer or zero (not to be confused with the
radial coordinate). The solutions in terms of exponentials are written Me (z, h2 ), those
in terms of hyperbolic functions cosh and sinh Mc and Ms . The solutions of the ith
181
kind are those in terms of cylindrical functions and are written M(i) (z, h2 ), where i =
1, 2, 3, 4 correspond respectively to expansions in terms of Bessel, Neumann or Hankel(1,2)
(i)
functions. The series of M (z, h2 ) converge uniformly only for |cosh z| > 1, whereas the
series of Me (z, h2 ) is uniformly convergent for all finite complex values of z as shown in
Ref. [13]. Since r = 0 corresponds to z = , the S-matrix is obtained by continuing the
(3)
solution at z = to z = . This means that a solution M has to be continued, via
matching to Me (across the domain |z| < 1), to a linear combination of M(3) and M(4)
at +. A few main steps of this calculation are given in Appendix A. The expression for
the S-matrix finally obtained is
S=
R2 1
ei ,
R 2 e2i
(8)
where
R=
(1)
(0, h2 )
M
.
(9)
(1)
M (0, h2 )
This S-matrix describes the scattering of an incident wave (component of the scalar field)
of energy off the spherically symmetric potential. One could visualise this scattering as a
spacetime curvature effect or with black hole event horizon zero as that of a potential
barrier surrounding the horizon. With the horizon shrunk to zero at the origin (implying
in the field theory a relation between mass and charge reminiscent of the Bogomolnyi
equation), this extremal case corresponds to that of a BPS state.
The absorptivity is A = 1 SS ? . The absorption cross section differs from this by a
l
of the lth partial wave in n
multiplicative factor in front. The absorption cross section abs
spatial dimensions has been derived in Ref. [22] and is given by

2n2 n/21
l +n3
l
=
(n/2
2)!(l
+
n/2
1)
1 |S|2 .
(10)
abs
n1
l
For n = 6 as in our case this l-wave (here semiclassical) absorption cross section (or socalled greybody factor) is given by
l
=
abs

8 2
(l + 1)(l + 2)2 1 |S|2 .
5
3
(11)
3. The D3-brane in BornInfeld theory

To supplement the previous section, we consider briefly the simplest version of
supersymmetric BornInfeld electrodynamics for the 3-brane. Our main intention is to
recall that the equation of small fluctuations about the D3-brane is again a modified
Mathieu equation as obtained above. In the simplest such model reduced to the static case
we write the Lagrangian
Z
L = d p x L,
1/2

p e (r).
L = 1 1 (i )2 + (i y)2 + (i i y)2 (i )2 (j y)2
(12)
182
Here Ei = F0i = i , i = 1, . . . , p, and y(xi ) originates from one of the gauge field
components Aa for a = p + 1, . . . , d 1, where d is equal to the dimension, which
represents the transverse displacements of the brane (of which we consider only one, e.g.,
A9 ). The source term of the electric field with charge e and 3 = 4 hints at spherical
symmetry. Considering only this case here and hence that of S-waves, we obtain two Euler
Lagrange equations which we can write

L
L
= c,
= 0,
r p1
r r p1
(r y)
(r y)
where c is a constant of integration. Explicitly,
e
0
= p1 ,
0
2
0
2
1/2
[1 ( ) + (y ) ]
r
y 0
c
= p1 ,
0
2
0
2
1/2
[1 ( ) + (y ) ]
r
(13)
so that
0 e 1
= .
y0
c a
(14)
Then
e2
,
r 2(p1) + e2 (1 a 2 )
( 0 )2 =
(y 0 )2 =
(ea)2
.
r 2(p1) + e2 (1 a 2 )
(15)
The p-brane and anti-p-branes are now given by

+
Z
dr q
y(r) =() ae
r
2(p1)
where r0
written
L=1
1
2(p1)
(16)
r 2(p1) r0
= e2 (a 2 1) > 1. In view of the proportionality (14) the Lagrangian can be

q
1 (1 a 2 )(i )2 p e (r).
(17)
The contribution to the energy not including the source term is for p = 3

Z
1
1 .
Ens = d 3 x p
1 (1 a 2 )(i )2
Only for a charge e which is kept fixed under a scale transformation is the energy minimal
in the limit a 2 1. This is the limit of the Bogomolnyi bound and hence for this value
of a 2 the BornInfeld configuration, i.e., the Dp-brane or string is classically stable, i.e.,
a nontopological BPS state. The reason is that for a = 1 we have 0 = y 0 which in the
original context with y = A9 implies F0r = r A9 . This again has implications for the
supersymmetry variation of the gaugino , the susy partner of the gauge field, which is
= F
(, being the original 10-dimensional indices and the appropriate combination of
10-dimensional Dirac matrices). In the case of only the electric field and the one excitation
under consideration the gaugino variation is

= 0r r A0 + 9r r A9

a=1
= 0r + 9r r A0 .
183
(18)
is a projection operator it is precisely for a = 1 that the variation

Thus, since
can be zero for some nonzero , thus preserving a fraction (half) of the number of
supersymmetries.
The energy Ens is infinite, but integrating from r = to infinity so that y =
ae/, dEns /dy, the energy per unit length of the string is finite, i.e.,
1
4e
dEns
= (1 a 2 )
= const.
dy
2
a
As shown in [4], unless a = 1 supersymmetry is completely broken (i.e., the supersymmetry variation of the gaugino would not be preserved). In this limit the throat radius r0
becomes smaller and smaller, and the brane pair moves further and further apart. If one
considers small fluctuations orthogonal to the brane and the string, one obtains the small
fluctuation equation [4]

e2 (p 2)2
2
= 0,
(19)
1r + 1 +
r 2(p1)
1 + 0r 9r
where 2 > 0 for stability. The radial part of these equations is with = r (p1)/2 and
angular momentum l

1
(p 1)(p 3)
e2 (p 2)2
d 2
2
+ 2 l(l + p 2)
+ 1+
= 0.
dr 2
r
4
r 2(p1)
Thus for p = 3 and x = r, = e 2 and for S-waves since the string cannot depend on
the angular variables of the worldvolume

2
2
d
+
1
+
= 0.
(20)
dx 2
x4
This equation is an S-wave radial Schrdinger equation for an attractive singular potential
x 4 but depends only on the single coupling parameter with constant positive
Schrdinger energy.
4. Calculation of the Floquet exponent in singular and nonsingular cases

The Floquet exponent enters the discussion of the Mathieu equation in view of the
Bloch wave property of the modified Mathieu function Me (z, h)
Me (z + i, h) = ei Me (z, h).
(21)
The Floquet exponent can be introduced in several ways. In Ref. [13] (p. 107) is
introduced by the relation

(22)
cos = yI , , h2 ,
where yI (x) is a fundamental solution of the (periodic) Mathieu equation satisfying the
boundary conditions yI (0) = 1, yI0 (0) = 1 and is the eigenvalue which, of course, is not
184
necessarily an integer. With a perturbation theory ansatz for yI (, , h2 ) around h2 = 0 the

following expansion is then shown to result (cf. Ref. [13], p. 124):
4 sin
cos = cos + h
4 ( 1)

2 cos
152 35 + 8
8
sin
+h
+ O h12 . (23)
2
3
32(
1)
64( 1) ( 4)
Alternatively one can apply directly perturbation theory to a trivial periodic solution of the
limit h2 0 as shown in Ref. [9]. In this case the following expansion is obtained
(5 2 + 7)h8
h4
+
2( 2 1) 32( 2 1)3 ( 2 4)

(9 4 + 58 2 + 29)h12
+ O h16 .
+
64( 2 1)5 ( 2 4)( 2 9)
= 2 +
(24)
This series can be reversed to yield , i.e.,

(13 25)h8
h4
2( 1) 32( 1)3 ( 4)

(453 4552 + 1291 1169)h12
+ O h16
5
2
64( 1) ( 4) ( 9)
2 =
(25)
and so
=

(8 35 + 152 )h8

+ O h12 .
3
4(1 ) 64( 4)( 1)
h4
(26)
One can easily check the agreement with the above expression for cos by evaluating
cos with of this expansion.
An obvious feature of all of these expansions is that they are singular for integral values
of . This behaviour is well-known. It means that these expansions are in these cases really
asymptotic expansions for large values of or and can be used in such cases. However,
for other values the expansions can also be convergent for sufficiently
small values of h2 as
is shown in Ref. [13]. For values of close to an integer or close to an integer one has
to expand around these as is also mentioned in Ref. [13] (pp. 124, 125). We demonstrate
this in the case of almost equal to 2. Thus we set
=2+
= 2 + ,
so that
2 2
+
(27)
2
and consider the limit 0. Expanding the cosine and sine expressions appearing in
Eq. (23) about = 4 we have ( 4 4)
cos = cos (2 + ) = cos 2 cos = 1
( 4)2 2
+
cos = cos 2 + ( 4) sin =4 + = 1
8
2
185
and
( 4)
+ .
sin = sin 2 + ( 4) cos =4 + =
2
2
Substitution into Eq. (23) and considering the approach 4 gives

h4 2 ( 4)
2
cos = 1
( 4)2 + +
8
4 ( 1) 2

2
2
2 1
8 (15 35 + 8) ( 4)
+
+h

64( 1)3 ( 4) 2 32( 1)2

h4 2
( 4)2 2
+ +
( 4) +
= 1
8
8( 1)

8
2

h
2
11 31 + 8 + + O h12 .
+
1282( 1)3
(28)
Hence (observe the cancellation of factors ( 4) in the term of O(h8 )) in the limit 0
5 2 h8
h8 2 (11 16 31 4 + 8)
+ = 1+ 9 2 + .
(29)
7
2
3
2 4 3
2 3
Comparing the expansions (27) and (29) we obtain
5
(h/2)4 .
= i
3
We see that although the coefficients of higher order terms of Eq. (23) contain factors
( 4) in the denominators and so suggest divergences, the trigonometric factors sin

in the numerator always cancel these out and thus yield a regular expansion for which is
even convergent within a certain domain around h2 = 0. One thus obtains the expansion
7i
11851i
i 5
(h/2)4 +
(30)
=2
(h/2)8 +
(h/2)12 .
3
108 5
31104 5
This
expansion has also been given in Ref. [1]. Expansions around other integral values of
are obtained similarly. Of course, the higher the value of this integer, the more terms at
the beginning of the series are identical
with those obtainable from the perturbation series
(26) above. Thus in the case of = 3, 4, we obtain from the first two terms of expansion
(26)
cos = 1 +
=3+
(h/2)4
h4
+ = 3
+
4(8)3
6
and similarly
(h/2)4
+
15
in agreement also with results of Ref. [1] up to the given order, i.e.,
1
133
311
(h/2)8 +
(h/2)12 + ,
= 3: = 3 (h/2)4 +
6
4320
1555200
1
137
305843
(h/2)8 +
(h/2)12 .
= 4: = 4 (h/2)4
15
27000
680400000
=4
(31)
186
5. Relation between Dougall and standard Mathieu coefficients

One may wonder how the Mathieu function coefficients of Dougall [3] which are used
in Refs. [1,2], are related to those in modern standard literature such as Ref. [13]. We
therefore demonstrate their precise connection here. It is crucial thereby to distinguish
between nonsingular or asymptotic cases and singular cases, as we shall see. We begin
with the nonsingular case and calculate a coefficient given in Ref. [13] (up to the first
nonleading contribution) by starting from Dougalls definition of his coefficients. We shall
see that the coefficients given in Ref. [13] obtained from simple continued fraction solution
of the basic recurrence relation of the coefficients are in this case not only easier to derive
but have also a simpler form than the coefficients of Dougall.
The modified Mathieu function in terms of exponentials is defined in Ref. [13] as the
following sum
2
Me z, h
:=
c2r
h2 e(+2r)z ,
(32)
r=
where 6= 1, 2, . In Ref. [13] (p. 131) the following relation of general validity is
given and used
Me (z, h) = Me (z, h).
(33)
This relation implies that
c2r
= c2r
and
(h2 )
c2r
(h2 )
c2r
.
=
c0 (h2 )
c0 (h2 )
(34)
We shall see explicitly that this relation holds also in what we call the singular case below.
Dougall [3] defines in his work the solution corresponding to Me (z, h) as
X
(r + /2) (+2r)z
J (, z)
=
e
(1)r
.
(/2) r=
(/2)
(35)
We therefore expect the equivalences

J (, z)
= Me (z, h2 ),
(/2)
(1)n
(h2 )
(n + /2) c2n
= 2 .
(/2)
c0 (h )
(36)
We now verify the latter of these relations for the case n = 1 in the nonsingular case
(i.e., for 6= integer + O(h2 )), i.e., we show that
c (h2 )
h2
( 2 + 4 + 7)h6
(/2 + 1)
2 2 =
+
+ ,
(/2)
c0 (h ) 4( + 1) 128( + 1)3 ( + 2)( 1)
(37)
where the expression on the right is given in Ref. [13] (p. 121). We also show thereby that
in leading order for small h2

(h/2)2n !
(n + /2)
=
1 + O h4
(/2)
(n + )!n!
(38)
187
in agreement with Ref. [13] (p. 121). The demonstration of agreement requires Eqs. (24),
(25) (cf. also [13], p. 119), i.e.,
(5 2 + 7)h8
h4
+
+
2( 2 1) 32( 2 1)3 ( 2 4)
s = l + 2,
s2 = 2 +
s = +
h4
+ .
4( 2 1)
(39)
or
(40)
This general relation is an asymptotic expansion in (i.e., for large), and can be obtained
perturbatively [9]. It is crucial, of course, to deal separately with values of close to a
singular value like = 2 (see below).
Dougall defines his coefficients by an expansion, of which the leading and next-toleading contributions are [3]
(n + /2)
=
(h/2)2n+
(n + /2 + s/2)!(n + /2 s/2)!
(
X
(h/2)4
1
(n + /2 + s/2 + 1 + p1 )
p1 =0
1
(n + /2 + s/2 + 2 + p1 )(n + /2 s/2 + 1 + p1 )(n + /2 s/2 + 2 + p1 )
)
+ .
(41)
Taking into account only the leading contribution, we have

(h/2)2n (/2 + s/2)!(/2 s/2)!
(n + /2)
=
.
(/2)
(n + /2 + s/2)!(n + /2 s/2)!
(42)
Using
(z)! =
(z 1)! sin z
(43)
and the approximation s (cf. Eq. (40)) we obtain

(h/2)2n !
(n + /2)
=
1 + O h4
(/2)
(n + )!n!
(44)
(h2 )/c (h2 ) of Ref. [13]. We see therefore that the expansion (40)
in agreement with c2n
0
plays an important role in establishing the connection between the coefficients (r + /2)
in the nonsingular case.
and c2r
We now consider the next-to-leading contribution in Eq. (41). Setting
A(1)
q =
X
p1 =0
1
(q + s/2 + 1 + p1 )(q + s/2 + 2 + p1 )
188
1
(q s/2 + 1 + p1 )(q s/2 + 2 + p1 )
we can make the partial fraction separation

X
1
1
1
=
A(1)
q
s(s 1) p1 + q + 1 + s/2 p1 + q + 2 s/2
p1 =0

1
1
1
.
s(s + 1) p1 + q + 2 + s/2 p1 + q + 1 s/2
(45)
(46)
The sums over individual terms are divergent. Thus the expression depends crucially on
taking differences. In order to deal with these we use the formula [23]
n1
X
k=0
1
= (n + y) (y),
k+y
(47)
where (y) is the derivative of the log of the gamma function (y). Thus

n1
X
1
1
p1 + q + 1 + s/2 p1 + q + 2 s/2
p1 =0
n
(q + 2 s/2) (q + 1 + s/2)
and
n1
X
1
1
p1 + q + 2 + s/2 p1 + q + 1 s/2
p1 =0
n
(q + 1 s/2) (q + 2 + s/2)
so that

1
(q + 2 s/2) (q + 1 + s/2)
s(s 1)

1
(q + 1 s/2) (q + 2 + s/2) .
s(s + 1)
We now use Eq. (40) in order to reexpress s in terms of . Then

1
(2 1)h4
1
=
+ ,
1 2
s(s 1) ( 1)
4 ( 1)2 ( + 1)

1
(2 + 1)h4
1
=
1 2
+
.
s(s + 1) ( + 1)
4 ( 1)( + 1)2
A(1)
q =
(48)
(49)
Setting q = n + /2 and dealing similarly with the arguments of the functions , we obtain
in lowest order of h

1
A(1)
q=n+/2 = ( 1) (n + 2) ( + n + 1)

1
(n + 1) ( + n + 2) .
(50)
( + 1)
189
Again we consider a difference, i.e.,

(1)
(1)
n := An+/2 An1+/2 .
We now use the formula
1
1 1
+ + + ,
2 3
n
where C is the Euler constant. Then
1
1
,
(n + 1) (n) =
(n + 2) (n + 1) =
n+1
n
and n becomes

1
1
( + n + 1) + ( + n)
n =
( 1) n + 1

1
1
( + n + 2) + ( + n + 1) .
( + 1) n
(n + 1) = C + 1 +
We now require yet another formula of the function, i.e.,

X
1
1
.
(x) = C +
n+1 x +n
(51)
(52)
(53)
i=0
With this we obtain

1
,
+n
1
( + n + 1) ( + n + 2) =
+n+1
(where in each case the dummy summation index of the second sum, i.e., i, was renamed
i 1). We therefore obtain in the dominant approximation

1
1
1
1
1
1
.
(54)
n =
( 1) n + 1 + n
( + 1) n + n + 1
( + n) ( + n + 1) =
For n = 1 this implies

1
(55)
2( + 1)( + 2)
in the dominant approximation. This difference will now have to be substituted into the
Dougall coefficient

h 2
h 4 (1)
(/2 + 1)
2 (/2 + s/2)!(/2 s/2)! 1 2 A/2+1 +
=

4 (1)
(/2)
(/2 + 1 + s/2)!(/2 + 1 s/2)! 1 h2 A/2 +
4

2
(1)
1 h2 A(1)
h
/2+1 A/2 +
.
(56)
2
(/2 + 1 + s/2)(/2 + 1 s/2)
(1)
(1)
1 = Aq=1+/2 Aq=/2 =
Inserting Eq. (55) and for s the series of Eq. (40) we obtain

c2
h2
h6 ( 2 + 4 + 7)
(/2 + 1)
10
=
+
+
O
h
=
(/2)
4( + 1) 128( 1)( + 1)3 ( + 2)
c0
(57)
190
in agreement with Ref. [13] (p. 121). To obtain the Dougall coefficients in this form is thus
seen to be rather complicated. This may explain why Dougall himself does not evaluate
any of his coefficients explicitly.
In the remainder of this section we calculate with the method of Dougall the important
coefficients (/2) for the S-wave case, i.e., s = 2, and demonstrate the agreement with
results of Ref. [1]. The results will also be needed in the next section in establishing ratios
corresponding to the ratio of Eq. (57). From Eq. (48) we obtain for the leading term in the
limit h2 0 for q = /2 and s = 2 (i.e., 2, a so-called singular case)

1

1
(1)
(/2 + 1) (/2 + 2) (/2) (/2 + 3)
A/2 =
2
6
h2 0
1

1
(58)
= (2) (3) (1) (4) .
2
6
Using Eq. (51) we obtain
1
.
18
The authors of Ref. [1] developed another algorithm in which (cf. their Appendix A)
(1)
A/2 =
A(1) = A(1)
z S[1]
(59)
and for s = 2 (their r = 1)

A(1)
z =
(z) (z + 2)
3 + 2z
+
.
3z(2 + z)
3
(60)
Here
(z) (z + 2) =

2z + 1
d
ln (z) ln (z + 2) =
dz
z(z + 1)
(61)
so that
A(1)
z =
1
3z(z + 1)(z + 2)
(62)
and for 2 one obtains again

1
18
as above. It follows that for this case with < > 0 (cf. Eq. (41)),

(h/2)4
1
(h/2)
1
(h/2)2 1 + O h4 .
(/2)
(/2 + s/2)!(/2 s/2)!
18
2
A(1)
/2 =
(63)
Fortunately for this case the expansion (57) does not seem to possess terms which diverge
for close to a positive integer 6= 1, and evidently even if it did since in our cases =
an integer 6= 1 + O(h4 ) would not affect the leading term of Eq. (59). This is radically
different when is close to a negative integer such as 2 in that case. We see from (53) that
4
if x = 2 + O(h4 ), 1/ h4 , and hence A(1)
q 1/ h , and so this term will contribute to
the leading factor in the coefficient of Eq. (41).
191
We now consider the coefficient (/2), < > 0, for which
i 5
(h/2)4 +
(64)
s = 2,
=2
3
and calculate this first with the method of Dougall. The following steps given explicitly
demonstrate clearly how singularities in the limit h2 0 arise and how they have to be
handled. Thus with Eq. (41):
(/2) =
(h/2)
(/2 + s/2)!(/2 s/2)!
(
X
(h/2)4
1
(/2 + s/2 + 1 + p1 )(/2 + s/2 + 2 + p1 )
p1 =0
1
+
(/2 s/2 + 1 + p1 )(/2 s/2 + 2 + p1 )
(h/2)2
0! 2 +
(

i 5
4
6 (h/2) !
1 (h/2)
"
(1)(2) 1 +
+
(2)(3)

i 5
4 i 5 (h/2)4
6 (h/2)
6
1
i 5
4
6 (h/2)
1+

i 5
4
6 (h/2)
p1 =2
(h/2)2
2+
(

i 5
4
6 (h/2) !
"
+ i 5
+
(2) i 6 5 (h/2)4
6 6 (h/2)4

2
(h/2)2
=
1+
i 5
2 + i 6 5 (h/2)4 !

i 5
2
Eq. (43) 1
(h/2)2 1 + (1)(1)
(h/2)4
=
6
i 5

2 2+i 5
.
= (h/2)
6
1 (h/2)
(65)
The steps above clearly show how the singular terms in the next-to-leading contribution
contribute to the dominant order.
The result (65) will now be shown to agree with the calculations of the method of
Ref. [1]. For this purpose we set in Eq. (62) z = /2 and replace by the expression
in Eq. (64). Then
A(1)
/2
8
2
=
4
3 2(i 5/3) 2 (h/2)
i 5(h/2)4
192
and
v 1 (h/2)
(1)
A/2
2+i 5
=
.
i 5
From Eq. (41) we obtain therefore

(/2) =
(h/2)2
0! 2 +

i 5
4
6 (h/2) !
v.
(66)
Using for the factorial again Eq. (43), one obtains

(h/2)2 i 6 5 (h/2)4
2
1+ .
(/2) =
i 5
(67)
Thus finally
2+i 5
(h/2)2
(/2) =
6
(68)
in agreement with Eq. (65).
6. Calculation of standard Mathieu coefficients

Our next objective is to compute a Dougall coefficient (i.e., a ratio of two quantities )
and to compare it with a standard Mathieu coefficient in the nontrivial singular case. As a
suitable example we choose the coefficient (/2 + 1) which when divided by (/2)
(calculated above) ought to agree with the Mathieu coefficient c2 /c0 according to
Eq. (36), i.e., we wish to demonstrate that
c
(/2 + 1)
= 2 .
(/2)
c0
(69)
We begin with the calculation of the Dougall coefficient. Using Eq. (41), we have
(/2 + 1) =
(h/2)2
(1 /2 + s/2)!(1 /2 s/2)!
(
X
(h/2)4
1
(1 /2 + s/2 + 1 + p1 )(1 /2 + s/2 + 2 + p1 )
p1 =0
)
1
+
(1 /2 s/2 + 1 + p1 )(1 /2 s/2 + 2 + p1 )

1
Eq. (64) for
(1) 1 +
1 (h/2)

i 5
4
6 (h/2) !
"
)
#
X
1
+
+

(2)(3) i 6 5 (h/2)4 (1) (3)(4)(1)(2)
p1 =2
1
(0)! sin 1 i 6 5 (h/2)4
i 51
(h/2)4 .

193
1
1
i 5
(70)
With Eq. (68) we obtain therefore

(/2 + 1)
2 1i 5
1 + O h4 .
= (h/2)
(/2)
2+i 5
(71)
Similarly one obtains

(/2 + 1) = 16 (h/2)4 1 + O h4 ,

(/2 + 1) 1
= 3 (h/2)2 1 + O h4 .
(/2)
(72)
Our next step is to derive the corresponding expression from the continued fraction
relation of the recurrence relation of the standard Mathieu coefficients. This recurrence
relation is given by (cf. Ref. [13], p. 117)
c2r
=
c2r2
h2 [s 2 ( + 2r)2 ]
. (73)
1
1
h2 [s 2 ( + 2r + 2)2 ] 2 2
h [s ( + 2r + 4)2 ]
Here we set r = 1 and replace by . Then we again use (64) and (a) set s = 2, and
(b) replace by the expansion given in Eq. (64). One then has to go as far as the terms
explicitly written out in the following continued fraction only to obtain the dominant
contribution:
c2
c0
1
h2 [s 2 ( + 2)2 ]
. (74)
1
h2 [s 2 ( + 4)2 ]
1
2
[s ( + 6)2 ]
2
Making the substitutions we obtain

c2
c0
1
h2 [4]
This can be seen to reduce to
c2
2i 51
= (h/2)
c0
i 5+2
h2 [ 4i3 5 (h/2)4 ]
(75)
h2 [12]
(76)
which agrees with the negative of the above Dougall coefficient as expected on the basis of
Eq. (35). One can see that the calculation here is simpler than that of both (68) and (70).
194
The reciprocal of the continued fraction relation ((73)) is (cf. Ref. [13], p. 117)
c2r2
c2r
1
h2 [s 2 ( + 2r 2)2 ]
(77)
1
1
h2 [s 2 ( + 2r 4)2 ] 2 2
h [s ( + 2r 6)2 ]
Replacing here by we have
c2r2
c2r
1
h2 [s 2 ( + 2r 2)2 ]
(78)
h2 [s 2 ( + 2r 4)2 ] 2 2
h [s ( + 2r 6)2 ]
Here we put r = 0 and again make the replacements of Eq. (64). Then
c2
c0

h2
1 + O h4
12
(79)
For r = 2 in Eq. ((78)) we obtain

c4
c2
12
h2 (1 i
5)
(80)
so that with Eq. (76)

c4
c0
c4 c2
c2
c0
.
2+i 5
(81)
In a similar way we obtain
c4
c2

h2
1 + O h4 ,
5
2
c4
c0

c4
c2

c2
c0

h4
1 + O h4 .
7
3.2
(82)
Summarising we have as leading contributions of standard Mathieu coefficients in the

singular case l = 0 or s = 2:

c2
c2
2i 51
4
=
(h/2)
,
1
+
O
h
=
c0
c0
i 5+2
c2
c0
c4
c0
c4
c0
c6
c0
c6
c0
c
(h/2)2
1 + O h4 = 2 ,
3
c0

c
,
1 + O h4 = 4
c0
2+i 5
c4
(h/2)4
4
1
+
O
h
= ,
c0
23 3
= (h/2)2
=

c
,
1 + O h4 = 6
c0
2+i 5
1
c6
(h/2)6
4
1
+
O
h
= .
c0
23 32 5
(83)
In Appendix B we give several more terms of these expansions calculated with

M ATHEMATICA.
195
7. Evaluation of the quantity R in the singular case

Having determined the standard Mathieu coefficients in the singular S-wave case, we
can proceed to evaluate the quantity R entering the S-matrix. R was defined in Ref. [9]
(see also Appendix A) as
R=
(h) =
Me (0, h)
(1)
(84)
M (0, h)
(i)
The function Me (z, h) was defined previously. The functions M (z, h), for i =
1, 2, 3, 4, are corresponding expansions of the modified Mathieu function in terms of
(1)
(2)
cylindrical functions J (z), Y (z), H (z), H (z), respectively. In particular we have
the expansion (cf. Ref. [13], p. 178)
Me (0, h)M(1)(z, h) =
c2r
h2 J+2r (2h cosh z).
(85)
r=
As shown in Ref. [13] (p. 180), a much better expansion to use in practice for M(1) (z, h)
in view of its rapid convergence, is
+
X

2 (1)
2
(1)l c2l
h M (z, h) =
h Jlr hez J+l+r hez
c2r
(86)
l=
so that in particular
+
X
2 (1)
2
(1)l c2l
h M (0, h) =
h Jlr (h)J+l+r (h).
c2r
(87)
l=
(1)
This formula is amazing. It implies that one and the same quantity M (0, h) can be
obtained from many different expansions (and so different Bessel functions) by allocating
different values to r, i.e., for example, r = 0 and 2. An analogous observation has also
been made by Dougall [3].
We begin with the evaluation of Me (0, h), i.e.,
Me (0, h) = c0 h2
(h2 )
X c2r
r
c0 (h2 )
With the help of the standard Mathieu coefficients evaluated previously we obtain

3
Me (0, h) h2 0 = c0 (0) 1
2+i 5

i 51
= Me (0, h) h2 0 .
= c0 (0)
2+i 5
(88)
(89)
The last equality follows also from the general property Me (z, h) = Me (z, h) (cf.
Ref. [13], p. 131). More terms can be calculated with M ATHEMATICA. Thus
196
Me (0, h) = 1 +
+
(h2 )
c2 (h2 ) c4 (h2 ) c6 (h2 ) c2
+
+
+
c0 (h2 ) c0 (h2 ) c0 (h2 )

c0 (h2 )
(h2 )
c4
c0 (h2 )
(h2 )
(h2 )
c8
c10
+
+
.
c0 (h2 )
c0 (h2 )
c0 (h2 )
(h2 )/c0 (h2 ) and c10

(h2 )/c0 (h2 )
also c8
(h2 )
c6
We have to take into account

contribute to orders h4 and h6 , i.e.,
4
8
(h2 )
c8
h
h
(i 5 )
202i + 35 5
=
+
,
2
c0 (h )
23 3(i + 5) 2
24 33 5(i + 5 ) 2
6
(h2 )
c10
h
i + 5
=
,
2
3
2
c0 (h )
2 3 5(i + 5 ) 2
because these
respectively. We then obtain the following result correct up to order h6 :
2
4
4i
h
h
2i
65i + 11 5
+
Me (0, h) =
+
3
1
2
i + 5 3(i + 5 )
2 3 5(i + 5 ) 2
6

11(83i + 89 5 )
h
+
+ O h7 .
22 33 5(19i 5 5 ) 2
(90)
We note here that for general values of not equal to an integer one obtains (up to and
including contributions of O(h6 ))
2
4
h
h
2
2 + 2
+ 2
Me (0, h) = 1 + 2
1 2
( 1)( 2 4) 2

2( 6 + 4 4 39 2 62) h 6
+ 2
+ O h7 .
(91)
3
2
2
( 1) ( 4)( 9) 2
(1)
Next we evaluate M (0, h) with the help of Eq. (87) choosing r = 0 and then as a check
r = 2. In the first case we obtain the expansion
M(1) (0, h) = J0 (h)J (h)
+
(h2 )
c2
c2 (h2 )
J
J1 (h)J1 (h)
(h)J
(h)
1
+1
c0 (h2 )
c0 (h2 )
c4
(h2 )
c4 (h2 )
J
J2 (h)J2 (h) + .
(h)J
(h)
+
2
+2
c0 (h2 )
c0 (h2 )
(92)
In lowest orders of h2 this is

M(1) (0, h) = J0 (h)J (h) +
c4
(h2 )
c0 (h2 )
J2 (h)J2 (h)
which when evaluated in lowest orders of h2 implies

(h/2)2
3
(h/2)2

1+
1.
2
2 + i 5 2!0!
It follows that

1
(1)
2 1 + i 5
.
M (0, h) = (h/2)
2
2+i 5
(93)
197
Here, of course, has as before the S-wave value, i.e.,
i 5
(h/2)4 + .
=2
3
If we set r = 2 in Eq. (87) we obtain

c
c0 c0
c
(1)
J3 J+1
M (0, h) = J2 J+2 2 J1 J+3 2
c4 c0
c0
c0

c
c
+ 4 J0 J+4 + 4
J
J
+
.
4
c0
c0
(94)
(95)

c
c
J4 J +
M(1) (0, h) = 0 J2 J+2 + 4
c4
c4
which when evaluated in lowest orders gives

3
1
3 27 (h/2)+4
h4
!
2( + 1)( + 2) (2 + i 5 )4 3 2
which reduces to

1
2 1 + i 5
(h/2)
2
2+i 5
in agreement with the previous result, i.e., Eq. (93).
(1)
(0, h). Again we use first the method with r = 0 in Eq. (87). We
Next we come to M
have
(1)
(0, h) = J0 (h)J (h)
M
c4 (h2 )
c0 (h2 )
c2 (h2 )
c0 (h2 )
J1 (h)J+1 (h)
J2 (h)J+2 (h) +
2
c4
(h )
c0 (h2 )
2
c2
(h )
c0 (h2 )
J1 (h)J1 (h)
J2 (h)J2 (h) + .
(96)

(1)
(0, h) = J0 (h)J (h) +
M
c4 (h2 )
c0 (h2 )
J2 (h)J+2 (h)
which when evaluated in lowest orders of h2 implies (with the help of the power expansion
of the Bessel function J (2h))

(h/2)+4
1 c
(h/2)
+ O h4 +
+ 4 (h/2)4
1
()!
2!( + 2)!
2 c0
and so
(h/2)

2

1
1
1 (1 4i 5 )
= + (h/2)2 (h/2)2 2 + i 5 .
+
2 6(2 + i 5 )
2
6
It follows that

1
(1)
M (0, h) = (h/2)2 1 i 5 .
6
(97)
198
(1)
If we set r = 2 in Eq. (87) and evaluate M
(0, h), we obtain

c
c
c
(1)
(0, h) = 0 J2 J+2 2 J1 J+3 2
J3 J+1
M
c4
c0
c0

c4
c4
+ J0 J+4 + J4 J + .
c0
c0
(98)

c0
c4
J2 J+2 + J0 J+4 + .
c4
c0
Evaluating this as before we obtain in leading orders

2+i 5 1
3
(h/2)+4
(1)
2
1
(h/2) 1
.
(99)
M (0, h) =
3
2
( + 4)!
2+i 5
Hence in leading order

(1)
2 1i 5
(100)
M (0, h) = (h/2)
6
which is seen to be in agreement with Eq. (97). With M ATHEMATICA we obtain higher
order terms, i.e.,

1+i 5 h 2 1+i 5 h 4
(1)
+
M (0, h) =
6
2
9
2

6

1
h
h
+
290 + 151i 5 + 120 5 i 5 C + ln
2160
2
2

8

1
h
h
+
,
514 73i 5 + 120 5 i 5 C + ln
3240
2
2
(101)

1i 5 h 2 1i 5 h 4
(1)
+
M (0, h) =
6
2
9
2

6

h
h
1
290 151i 5 + 120 5 + i 5 C + ln
+
2160
2
2

8

h
1
h
514 + 73i 5 + 120 5 + i 5 C + ln
+
.
3240
2
2
(102)
With these results we can evaluate and . Thus again in the dominant approximation

. 1
Me (0, h)
= c0
(h/2)2
(103)
= (1)
2
M (0, h)
and
=
Me (0, h)
(1)
M
(0, h)
= c0
2+i 5

1
(h/2)2 .
2
(104)
199
It follows that
(1)
R=
M (0, h)
M(1)(0, h)
49 + 80 C + ln h2 4
i+ 5
h
=
+
i + 5
128 5(2i + 5 )

49 + 80 C + ln h2 151i 58 5 + 120(i + 5 ) C + ln h2
h8
+
737280(7i + 5 )
4
h
2 + i 5 (5 2i 5 ) 49 + 80 C + ln h2
+
=
3
360
2

1
h
+
49 449 + 85i 5 80 743 + 232i 5 C + ln
51840
2

2 8

h
h
(105)
+ 19200 2 + i 5 C + ln
2
2
with c0 = c0 = 1 (cf. MS, p. 122, Eq. (39)).

We can now identify our quantities with those of Ref. [1]. Comparison with the Dougall
coefficients evaluated previously implies
1
1
,
=
(106)
=
(/2)
(/2)
and so in the notation of Ref. [1]
(/2)
.
(107)
R=
(/2)
A remarkable feature of the expression (105) is its unit modulus, as was also observed in
Ref. [1]. It means that R is a pure phase factor
R = ei ,
RR = 1.
(108)
When and why this behaviour occurs is discussed at the end of Appendix A.
8. The quantity R in the general case

In the general case, or for s = l + 2 sufficiently large so that no problems with
(1)
singularities arise, we can evaluate R and so M (0, h) and Me (0, h) by simply using
the power series expansions of and the standard Mathieu coefficients and, of course, the
power series expansion of Bessel functions J (2h). One then obtains

h
1
(1)
M (0, h) =
()! 2

2
4
2
2( 2 3 7)
h
h
1+ 2
1 2
( 1)2 ( 1)( 2 4) 2

4( 4 11 3 2 2 59 23) h 6
+ .
(109)
( 1)2 ( 1)3 ( 2 4)( 2 9) 2
200
This implies for R

R=
=
!
()!

h 2
2
1+
1+
2
2 1
2
2 1

h 2
2

h 2
2
2( 2 +37)
(1)2 (+1)( 24)
2( 2 37)
(+1)2 (1)( 24)

h 4
2

h 4
2
+
+

!( 1)! sin
2
h2

4
4
h
1+ 2
2
( 1) 2
+

2(4 5 + 15 4 32 3 12 2 + 64 111) h 8
+
.
2
( 2 1)4 ( 2 4)2
(110)
The first few terms of the expansion of the function Me (0, h) which is needed for
comparison with the results of Ref. [1] have been obtained in the previous section.
From Eq. (110) we extract for later reference
4

2 h2
sin 2
=
(111)
2 .
2

h 4
R
+
!( 1)! 1 + ( 24
2
2
1)
This expansion will be used below in the low order approximation of the absorptivity for
higher partial waves.
9. Calculation of the absorptivity

We consider the absorptivity in a general case, and hence allow for complex Floquet
exponents , which we set
= n + i( + i) = (n ) + i,
(112)
where n = 2, 3, 4, . . . , and and are real and of O(h4 ). In evaluating the S-matrix for
small h4 one has to be careful to make the expansions in the appropriate way. Thus we
write SS ?

1 R12 1 1? 2
?
R
SS =
(113)
i
i ?
?
ei e R 2 ei e ? 2
R
which can be rewritten

SS ? =

1 cos 2
1
R2

e2 1 R12 1 1? 2
R

+ 1? 2 e2 + i sin 2 R12
R
1
R?2
e2
e4
R2 R?2
.
(114)
Here we set
e2i 1 + if,
e2 1 + g,
(115)
where f is complex and g is real (in the S-wave case g = 2

cos 2 = 1 =f,
5
4
3 (h/2)
1
1
=f (2)2 (sin 2)2
2
2
201
+ O(h8 )). Then

(116)
and
sin 2 = <f.
(117)
Then
SS ? = (1 + g)

o 1
n
=f (1 + g) 12 + 1?2 + (1 + g) <f i 12 1?2

g 12 + 1?2 2+g
2
?2
R
R
R
R
R R
R
R
.

1
1 12 1 1?2
R
(118)
We now consider two limiting cases.
(i) 0 implying g 0.
In this case R = R ? and so 1/R 2 ' O(h4 ). This is the case of real Floquet exponents and
so excludes the case of S-waves. Here
2
4 sinR
1
2=f
?
(119)
SS =
2 ' 1
2 ' 1
2
1 + =f R22 / 1 R12
R 2 / 1 R12
1 R12
since
1
=f ' (sin 2)2 ' 2 sin2 .
2
The absorptivity A is therefore given by
2

4 sinR
sin 2
?
.
A = 1 SS
2 4
R
1 12
(120)
With the help of Eq. (111) this can be written

4

4
2
4 2 h2
4
h
8
+O h
A
2 1 + 2
( 1)2 2
!( 1)!
4(l+2)

4

2
4 2 h2
4
h
8
=
+O h
2 1 + 2
( 1)2 2
(l + 1)!(l + 2)!
(121)
in agreement with Ref. [1]. This can be easily evaluated (e.g., with M APLE), e.g., already
in the case of P -waves (i.e., in spite of singularities in higher order terms here omitted)
and yields in this case

1
he
53
2 h12
log
h4 ,
(122)
A = 2 14 1 +
3 2
1152 24
2
where is the Euler constant (also written C). This result agrees with the result in Ref. [1].
We observe, in particular, that logarithmic energy contributions arise in the expansion. The
formula (121), of course, does not apply in the case of S-waves.
202
(ii) 0 implying f 0.
In this case RR ? = 1, so that |R| O(h0 ), and we cannot expand as in the previous case.
However, g O(h4 ), so that we can expand in powers of g. Thus

1+g
g2
3

=1
+
O
g
g
1+g 1+
2 R2 R? 2
2R 2 R ? 2

= 1 + O h8 .
SS ? =
(123)
This is the case of complex Floquet exponents as in the S-wave case. In this case
SS ? = 1

9g 2
+ O g3
20
(124)
and so

9g 2
+ O g3
20

2 2
h
(125)
7 12 + ln
(h/2)12 + O h16
= 2 (h/2)8 +
9
2
in agreement with Ref. [1] and the rough lowest order calculation of Ref. [17]. We can
also compute for this particularly interesting S-wave case the amplitudes of the reflected,
transmitted and incident waves Ar , At , Ai defined in Appendix A. One finds

2i 5
1
+ O h4 ,
Ar = R =
R
3

2 5
(h/2)4 + O h8 ,
At = 2i sin =
3

1
2i 5 4 5
i
(h/2)4 + O h8 .
=
(126)
Ai = R e
i
Re
3
9
We observe that in the limit h4 0 we have At = 0 and Ar = Ai , i.e., there is only
reflection of the fluctuation or disturbance around the D-brane like reflection from a wall
and no transmission, which can be interpreted as a vanishing of the disturbance on the
brane (implying a Dirichlet boundary condition). On the basis of the analogy with the case
of the open fundamental string between brane and antibrane in BornInfeld theory we can
expect that as the energy increases, transmission (i.e., absorption) sets in and becomes the
dominant effect at high energies. This is similar to what one finds in quantum mechanics of
a potential well of depth Vo . 5 There the properly normalised transmission and reflection
coefficients T (E), R(E), where E is the energy, have the behaviour T (E) 0, R(E) 1
as E 0, but T (E) 1 and R(E) 0 as E . The high energy behaviour of the
effect considered here can presumably be investigated with the help of large-h asymptotic
expansions of modified Mathieu functions which we expect to be formally (i.e., apart
from sign and complex i changes) similar to those of periodic Mathieu functions with
a parameter q defined as the solution of

(127)
(l + 2)2 = 2h2 + 2hq + O h0 .
A=
5 For comparison consult, e.g., Ref. [24].
The Floquet exponent is then given in Refs. [13] (p. 210), [26]

1
1 + 3(q 2 + 1)/64h
e4h
+
O
.
cos + 1 =
q/2
(8h)
((1 q)/4) ((3 q)/4)
h2
203
(128)
10. Conclusions
In the above we considered the impingement of a massless scalar field on a D3 brane
in 10 dimensions and calculated the S-matrix and partial wave absorption and reflection
amplitudes and rates for this process. Instead of coefficients introduced by Dougall for the
expansion of the modified Mathieu functions involved, we used (in the low energy domain)
rapidly convergent series in terms of products of Bessel functions. We demonstrated that
the Mathieu function coefficients are such that many different expansions in terms of
products of Bessel functions all yield the same low energy power series for the modified
Mathieu functions of the first kind. We think, this is the best way to evaluate the absorption
rates of the problem in the low energy domain. The leading term matching procedures
of Refs. [16,17] maybe select dominant terms of the expansions considered here. Since
the metric considered is extremal, one can visualise the absorption of the partial waves
of the scalar field as absorption into the brane or black hole with vanishing event horizon
(examples with nonvanishing horizon have for instance been treated in [25]). Since several
other string models lead also to the modified Mathieu equation in analogous contexts, the
above considerations, which have definite advantages over those involving the coefficients
of Dougall, may be of wider interest.
Acknowledgement
R.M. and Y.Z. acknowledge support as fellows of the A. von Humboldt Foundation.
J.-Q.L. acknowledges support by the Deutsche Forschungsgemeinschaft in the framework
of the agreement with the Ministry of Education of China. H.J.W.M.-K. is indebted to
A. Hashimoto for correspondence.
Appendix A
Here we recapitulate the main steps of the derivation of the S-matrix. We follow Ref. [9],
but instead of repeating the steps there, we emphasize those which have not been written
out explicitly there. For ease of comparison we consider the repulsive potential which
means simply that the considerations below employ coupling g as in Ref. [9] instead of g0
used above. The two cases are trivially related through
g0 = ig.
(A.1)
In the repulsive case we have a regular solution yreg of Eq. (5) at r = 0, i.e., one
proportional to exp(g/r). The variable of the cylindrical functions involved in M (j ) (z, h)
204
is = 2h cosh z = (ig/r + kr). Thus in leading order for small h2 and r 0 (z )

we can write

<z
yreg = r 1/2 M(3)(z, h) ' r 1/2 H(1)() + O h2
1/2

r0
2
'
eg/r ei(+1)/2 1 + O h2 .
(A.2)
g
If we let <z here and then replace z by z, the solution has the asymptotic
behaviour eikr . The series expansion defining the Bessel function J has the following
important property for integers n

(A.3)
J 2h cosh(z + in) = exp(in)J (2h cosh z)
so that
M(1) (z + in, h) = exp(in)M(1) (z, h).
(A.4)
Since for Me (z, h) also

Me (z + in, h) = exp(in)Me (z, h)
(A.5)
we have the proportionality

Me (z, h) = (h)M(1) (z, h)
(A.6)
with (e.g.)
(h) =
Me (0, h)
M(1)(0, h)
(A.7)
As mentioned earlier, expansions in terms of cylindrical functions like (85) converge

uniformly only in domains |cosh z| > 1, whereas the expansion (32) of Me (z, h)
converges for all finite complex values of z. Hence we match M(3) (z, h) in the domain
<z < 0 to a linear combination of M(3)(z, h) and M(4) (z, h) in the domain <z > 0 by
matching both to a combination of Me (z, h) and Me (z, h) in the intermediate domain.
We have
p
(A.8)
z = log k/g r i/4.
In the domain of r close to zero we set
r 1/2 M(3) = r 1/2 (Me + Me ),

d 1/2
d 1/2
d 1/2 (3)
r M =
r Me +
r Me ,
dr
dr
dr
(A.9)
where and have to be determined. In the domain of large r we set, with constants 0 ,
0 , A, B, which have to be determined

r 1/2 0 Me + 0 Me = r 1/2 AM(3) + BM(4) ,

d 1/2 (3)
d 1/2 (4)
d 1/2
d 1/2
r Me + 0
r Me = A
r M + B
r M . (A.10)
0
dr
dr
dr
dr
205
We match the Me , Me combination (variable z) on the left to that on the right (variable
z) at <z = 0 (r = g/k ), so that

r 1/2 (Me + Me )z=+i/4 = r 1/2 0 Me + 0 Me z=i/4 ,

d 1/2
d 1/2
r Me +
r Me
dr
dr
z=+i/4

d
d
= 0
.
r 1/2 Me + 0
r 1/2 Me
dr
dr
z=i/4
Since Me (z) = Me (z) and at r = g/k, z = i/4, also

1/2
k
d
d
=
,
dr
g
dz
(A.11)
the latter become

Me + Me = 0 Me + 0 Me ,
d
d
d
d
Me + Me = 0 Me + 0 Me .
dz
dz
dz
dz
From these equations we obtain immediately
0 = ,
(A.12)
0 = .
(A.13)
From Eqs. (A.9) we obtain (W meaning Wronskian)

W [M(3) , Me ]
W [M(3) , Me ]
,
=
.
W [Me , Me ]
W [Me , Me ]
From (A.10) we obtain similarly
=
(3)
A=
B=
(4)
(3)
(A.14)
(4)
W [M , Me ]W [Me , M ] + W [M , Me ]W [Me , M ]
W [M(3) , M(4) ]W [Me , Me ]
W [M(3) , Me ]W [Me , M(3) ] W [M(3) , Me ]W [Me , M(3)]
(3)
(4)
W [M , M ]W [Me , Me ]
(i)
,
(A.15)
(j )
We now use Eq. (A.6) and Wronskians W [M , M ] [i, j ] given in Ref. [13], i.e.,
2i
4i
[1, 3] = [1, 4] =
(A.16)
[3, 4] = ,
and the circuit relation ([13], p. 169)

(1)
M = ei M(1) i sin M(4) .
(A.17)
2 sin
,
2i

W Me , M(3) = ei ,

2i
(4)
W Me , M = ei .
(A.18)
Then
W [Me , Me ] =
206
With these expressions A and B are found to be

,
A=
2i sin

1
2i
e
.
B=
2i sin
The regular solution thus continued to r = is then

yreg ' r 1/2 AM (3)(z, h) + BM (4) (z, h)
1/2
ikr i(+1/2)/2

2
+ ei/2 Beikr ei(+1/2)/2 .
Ae e
'
k
In terms of the variable z and with R / this can be written
1/2

2r
ei(+1/2)/2 2i sin e2ih cosh z
2h cosh z
1/2

<z
2r
'
ei(+1/2)/2
2h cosh z

1 2ih cosh z
ei 2ih cosh z
i
R
+i Re
e
e
.
R
R
(A.19)
(A.20)
(A.21)
If we take Ai = (R ei ei /R) as the amplitude of the incident wave, the amplitudes

Ar and At of the reflected and transmitted waves are Ar = R R1 and At = 2i sin
respectively in agreement with Ref. [1]. With the definition of the S-matrix in the partial
wave expansion of the scattering amplitude f ( ), with x = cos , for n-space dimensions
(here we have n = 6) given by [22]
eikx + f ( )
eikr
r (n1)/2
X
ikr

1
el (cos ),
'
Se + (1)l i n1 eikr P
(n1)/2
2(ikr)
(A.22)
l=0
where
2 n/21
2
(n/2 1)(l + n/2 1)Cl (cos )
and Cl (cos ) is a Gegenbauer polynomial, we obtain for this

el (cos ) =
P
R R1
eil .
(A.23)
(R ei ei /R)
It is easy to verify that for real and R ey real, unitarity is preserved, i.e., unity minus
reflection probability = transmission probability, i.e.,
S=
|R R1 |2
|2 sin |2
=
.
(A.24)
i
i
2
i
|R e e
/R|
|R e ei /R|2
We observe that this relation remains valid if the real quantity R ey and the pure phase
factor ei exchange their roles, i.e., if R becomes a pure phase factor and ei a real
1
207
exponential. The latter is precisely what happens in the S-wave case of the attractive
potential discussed above.
Appendix B
Below we give the explicit form of the first three terms of the small-h2 perturbation
of expansions of modified Mathieu functions in

expansions of the leading coefficients c2r
the S-wave case (l = 0). The nonleading terms have been obtained with M ATHEMATICA:
2
6
c2
h
i+ 5
11(5 2i 5 )
h
+
=
3
2
c0
2i + 5 2
2 5(2i + 5 ) 2

(72311i 17746 5 ) h 10
+
+ O h11 ,
7
3
3
2 3 5(2i + 5 ) 2

c2
1 h 2 3 + 16i 5 h 6 5146 391i 5 h 10
=
+
+ O h11 ,
3
3
6
5
3 2
2
2
2 3
2 3 5
c0
4

c4
11(5i + 5 )
h
i 5
2966 5 28889i h 8
+
=
+
+ O h9 ,
c0
i + 5 23 31 5(i + 5 ) 2
27 35 5(2i + 5 ) 2
4

c4
1
18 + 125i 5 h 8 1303i 5 25774 h 12
h
=
+
+
+ O h13 ,
3
6
3
8
5
2
2 3 2
2 3 5
2
2 3 5
2
c0
c6
c0
c6
c0

2
6
290i + 49 5
39559i 4597 5 h 10
h
h
=
2+i 5 2
22 33 5(i + 5 ) 2
27 35 5(i + 5 ) 2

+ O h11 ,
1

6
h
1
19 + 157i 5 h 10 132035 5159i 5 h 14
= 3 2
6 4 2
+
2
2
2 3 5 2
2 3 5
27 36 53 7

+ O h15 .
References
[1] S.S. Gubser, A. Hashimoto, Exact absorption probabilities for the D3-brane, hep-th/98051402v2.
[2] M. Cvetic, H. L, C.N. Pope, T.A. Tran, Exact absorption probability in the extremal sixdimensional dyonic background, hep-th/9901002.
[3] J. Dougall, The solution of Mathieus differential equation, in: Proceedings of the Edinburgh
Mathematical Society XXXIV, 1916, pp. 176196.
[4] C.G. Callan, J.M. Maldacena, Brane dynamics from the BornInfeld action, hep-th/9708147.
[5] A. Hashimoto, The shape of branes pulled by strings, hep-th/9711097v3, Phys. Rev. D 57 (1998)
6441.
208
[6] K.G. Savvidy, G.K. Savvidy, Von Neumann boundary conditions from BornInfeld dynamics,
Nucl. Phys. B 561 (1999) 117, hep-th/9902023.
[7] G.K. Savvidy, Electromagnetic dipole radiation of oscillating D-branes, hep-th/0001098.
[8] R.M. Spector, Exact solution of the Schrdinger equation for inverse fourth power potential,
J. Math. Phys. 5 (1964) 1185.
[9] H.H. Aly, H.J.W. Mller-Kirsten, N. Vahedi-Faridi, Scattering by singular potentials with a
perturbation-theoretical introduction to Mathieu functions, J. Math. Phys. 16 (1975) 961.
[10] H.H. Aly, W. Gttinger, H.J.W. Mller, Singular interactions, in: H.H. Aly (Ed.), Lectures in
Theoretical High Energy Physics, Gordon and Breach, New York, 1970, pp. 247322.
[11] J. Challifour, R.J. Eden, Regge poles and branch cuts for potential scattering, J. Math. Phys. 4
(1963) 359.
[12] N. Dombey, R.H. Jones, Analyticity of the scattering amplitude for singular potentials, J. Math.
Phys. 9 (1968) 986.
[13] J. Meixner, F.W. Schfke, Mathieusche Funktionen und Sphroidfunktionen, Springer, Berlin,
1954.
[14] F.M. Arscott, Periodic Differential Equations, Pergamon, Oxford, 1964.
[15] M.S. Costa, Absorption by double-centered D3 branes and the Coulomb branch of N = 4 SYM
theory, hep-th/9912073.
[16] I.R. Klebanov, W. Taylor IV, M. Van Raamsdonk, Absorption of dilaton partial waves by D
branes, hep-th/9905175.
[17] I.R. Klebanov, World-volume approach to absorption by non-dilatonic branes, hepth/9702076v2.
[18] G. Horowitz, A. Strominger, Black strings and p-branes, Nucl. Phys. B 360 (1991) 197.
[19] J. Polchinski, Dirichlet branes and RamondRamond charges, Phys. Rev. Lett. 75 (1995) 4724,
hep-th/9510135.
[20] J. Maldacena, The large N limit of superconformal field theories and supergravity, hepth/9711200.
[21] O. Aharony, S.S. Gubser, J. Maldacena, H. Ooguri, Y. Oz, Large N field theories, string theory
and gravity, hep-th/9905111.
[22] S.S. Gubser, Can the effective string see higher partial waves?, PUPT-1697, hep-th/9704195.
[23] E.R. Hansen, A Table of Series and Products, Prentice-Hall, Englewood Cliffs, 1975.
[24] F. Schwabl, Quantenmechanik, 3rd edn., Springer, Berlin, 1992.
[25] H.W. Lee, Y.S. Myung, Scattering from an AdS3 bubble and an exact AdS3 , hep-th/990354v2.
[26] R.E. Langer, The solutions of the Mathieu equation with a complex variable and at least one
parameter large, Trans. Am. Math. Soc. 36 (1934) 637, see in particular pp. 650 and 690.

Non-commutative world-volume interactions

on D-brane and DiracBornInfeld action
Mohammad R. Garousi a,b
a Department of Physics, University of Birjand, Birjand, Iran
b Institute for Studies in Theoretical Physics and Mathematics IPM, P.O. Box 19395-5746, Tehran, Iran
Received 19 October 1999; revised 16 December 1999; accepted 24 December 1999
Abstract
By integrating the SeibergWitten differential equation in a special path, we write ordinary gauge
fields in terms of their non-commutative counterparts up to three non-commutative gauge fields. We
then use this change of variables to write the ordinary abelian DiracBornInfeld action in terms
of non-commutative fields. The resulting action is then compared with various low-energy contact
terms of world-sheet perturbative string scattering amplitudes from a non-commutative Dp-brane.
We find complete agreement between the field theory and string theory results. Hence, it shows that
perturbative string theory knows the solution of the SeibergWitten differential equation. 2000
Elsevier Science B.V. All rights reserved.
1. Introduction
Recent years have seen dramatic progress in the understanding of non-perturbative
aspects of string theory [15]. With these studies has come the realization that extended
objects, other than just strings, play an essential role. An important tool in these
investigations has been Dirichlet branes [68]. D-branes are non-perturbative states on
which open string can live, and to which various closed strings including RamondRamond
states can couple.
Another interesting aspect of D-branes is that in the presence of background flux the
world-volume of D-brane becomes non-commutative [911]. Hence, at low energy the
D-brane dynamics may be described by a non-commutative gauge theory. On the other
hand, it is known that the D-brane is properly described by the DiracBornInfeld action
with appropriate background flux (see, e.g., Ref. [12]). Using this idea, Seiberg and Witten
E-mail address: biruni@iran.com (M.R. Garousi).
PII: S 0 5 5 0 - 3 2 1 3 ( 9 9 ) 0 0 8 2 6 - 3
210
M.R. Garousi / Nuclear Physics B 579 (2000) 209228
were able to find, among other things, an explicit differential equation that relates noncommutative gauge fields at various non-commutative parameters [13]. This Seiberg
Witten differential equation can be integrated to find a transformation that changes ordinary
gauge fields into non-commutative fields.
The purpose of this paper is to show that the world-sheet perturbative string theory
can capture the above transformation. To this end, we integrate the differential equation
in a special path to write ordinary gauge fields in terms of their non-commutative
counterparts up to three non-commutative fields. The resulting transformation for the
abelian case contains two different multiplication rules. One is the familiar noncommutative multiplication that appears in the definition of the non-commutative gauge
field strength and the other one, which we call 0 multiplication (see (13)), operates as
commutative multiplication rule between two non-commutative field strengths. We use
this transformation to rewrite the ordinary DBI action in terms of non-commutative fields.
The resulting field theory action which contains various new interactions between ordinary
closed string and non-commutative open string fields are then compared with appropriate
low-energy contact terms of perturbative string theory. Our results on the string theory
side are fully consistent with the new interactions on the field theory side. Hence, it shows
that perturbative string theory knows about the solution of the SeibergWitten differential
equation.
One of the outcomes of our calculations is that if one only replaces ordinary fields in
the DBI action in terms of their non-commutative counterparts, the resulting action is not
completely identical to the appropriate contact terms of string theory scattering amplitudes.
To have an action that is fully compatible with string theory, one should transform the
ordinary multiplication rule between open string fields to the 0 rule as well.
The paper is organized as follows. In Section 2 we expand the DBI action to produce
various interactions involving one closed and one or two ordinary open string fields. In
Section 3 we integrate the SeibergWitten differential equation to transform the ordinary
open string fields to their non-commutative counterparts. This transformation leads us
to propose that the ordinary multiplication rule between two open string fields in the
expansion of DBI action should be replaced by the commutative 0 rule. In Section 4,
using the conformal field theory technique, we calculate various string theory amplitudes
describing scattering of closed and open string states from the non-commutative Dpbranes. Using these string theory amplitudes, we determine various low-energy amplitudes
and contact terms and compare them with the field theory results. We conclude with a brief
discussion of our results in Section 5. Appendix A contains our conventions and some
useful comments on conformal field theory propagators and vertex operators used in our
calculations.
2. DiracBornInfeld couplings
The world-volume theory of a single D-brane in type 0 theory includes a massless
U(1) vector Aa and a set of massless scalars Xi , describing the transverse oscillations
211
of the brane [15,16]. The leading order low-energy action for these fields corresponds to
a dimensional reduction of a ten-dimensional U(1) YangMills theory. As usual in string
theory, there are higher order 0 = `2s corrections, where `s is the string length scale. As
long as derivatives of the field strengths (and second derivatives of the scalars) are small
compared to `s , the action takes a DiracBornInfeld form [1719]. To take into account
the couplings of the open string states with closed strings, the DBI action may be extended
naturally to include background closed string fields, in particular, the metric, dilaton, Kalb
Ramond and tachyon field [20,21]. In this case one arrives at the following world-volume
action:
Z
q
eab + B
eab + 2`2s Fab ),
(1)
SBI = Tp d p+1 g(T ) e det(G
where the tachyon function is g(T ) = 1 +T /4 +3T 2 /32 + [20]. Here, Fab is the abelian
field strength of the world-volume ordinary gauge field, while the metric and antisymmetric
tensors are the pull-backs of the bulk tensors to the D-brane world-volume, e.g.,
eab = Gab + 2Gi(a b) Xi + Gij a Xi b Xj .
G
(2)
In general, the closed string fields are function of world-volume and transverse coordinates,
i.e., Xa and Xi respectively, however, for simplicity we assume they are just a function of
Xa .
In order to find the interactions expected from the DBI action, we expand the action
for fluctuations around G = , B = F ab a b , = 0. The fluctuations should
be normalized as the conventional field theory modes which appear in the string vertex
operators. As a first step, we recall that the graviton vertex operator corresponds to the
string frame metric. Hence, one should transform the Einstein frame metric G to the
string frame metric g via G = e/2 g . Now with the conventions of Ref. [22], the
string mode fluctuations take the form
g = + 2h ,
= 2 ,
B = F ab a b 2b ,
T = 2,
1
aa ,
Aa = p
Tp 2`2s
1
Xi = p i .
Tp
With these normalizations, the pull back of the Einstein frame metric becomes

1
e
1 + a i b i + ,
Gab = ab 1 + + 2 hab +
Tp
2
2
where the dots represents terms with two and more closed string fields.
Now it is straightforward to expand Eq. (1) using
(3)
(4)
212
det(M0 + M)
p

2
= det(M0 ) 1 + 12 Tr M01 M 14 Tr M01 MM01 M + 18 Tr M01 M

+ 16 Tr M01 MM01 MM01 M 18 Tr M01 M Tr M01 MM01 M
3

1
+
Tr M01 M
+ 48
to produce a vast array of interactions. We are mostly interested in the interactions linear
in the closed string fluctuations, and linear or quadratic in the open string fields.
We begin with the linear couplings of the closed strings to the D-brane source itself,

1
1
ab
(5)
L0,1 = Tp c + V (hba bba ) + Tr(V ) 4 ,
2
2 2
where we defined the overall square root of the metric as det(ab + Fab ) c, and the
matrix V ab as the dual of the metric, that is
ab
(6)
V ab ( + F )1 .
Next there are interactions involving one closed string mode and one open string mode,
that is

p

1 ab
1
1
ab
L1,1 = Tp c V fba + V (hba bba ) + Tr(V ) 4
2
2
2 2

1
ab
cd
ab
i
i
hbc bbc + bc V fda + 2V hi(b a) bi[b a] ,(7)
V
2 2
where fab = a ab b aa . We will also need to compare our results with the DBI terms
that have one closed and two open string states,

1 ab
1
1
+ V ab (hba bba ) + Tr(V ) 4
V a i b i
L2,1 = c
2
2
2 2

1 ab
1 ab 2
1
cd
V fbc V fda + V fba
V ab hbc bbc + bc
4
8
2 2

1 cd
cd
i
cd
ef
ef
V d a i V fde V ff a + V fda V ff e
2
1
+ V ab a i b i + V ab (hij bij )b i a j
2 2

+ V ab hi(b a) i bi[b a] i V cd fdc

2V ab hi(b c) i bi[b c] i V cd fda .
(8)
Finally, to compare the couplings of three open string states and massless poles of string
amplitudes with the corresponding terms in the DBI theory, we will also need the following
action:
2
L2,0 = c 12 V ab a i b i 14 V ab fbc V cd fda + 18 V ab fba

(9)
= c 12 (VS )ab a i b i 14 (VS )ab fbc (VS )cd fda ,
where we have dropped some total derivative terms in the second line above.
213
3. From commutative to non-commutative variables

Taking into account that the open string vertex operators correspond to non-commutative
gauge fields [13], one should write the ordinary open string fields in terms of their noncommutative counterparts. In [13] the differential equation for a non-commutative gauge
field was found to be

bbd + 2F
bbd F
bac A c D
bd F
bac F
bab + d F
bab
bab ( ) = 1 cd 2F
F
4

bab + d F
bab A c + O F
b3 ,
bd F
(10)
D
where the field strength and product were defined to be

bab = a A b b A a i A a A b + i A b A a = a A b b A a i A a , A b ,
F
M
i
00 )|x 0 =x 00 =x .
f(x) g(x)
= e 2 ab x 0 x 00 f(x 0 )g(x
(11)
Scattering amplitudes in Section 4 reproduce different couplings for finite . Therefore,

to compare the expected coupling of the DBI action and the string amplitude we should
integrate the differential equation (10) to find the relation between the ordinary field
bab ( ). We take the
bab ( = 0), and its non-commutative counterpart, i.e., F
strength, i.e., F
integral in the special path that ab is proportional to a scalar, i.e., ab = ab , and take the
integral over from = 0 to = 1. The result is

bab 1 cd F
bbd 00 F
bbd + F
bac A c 00 d F
bab d F
bab 00 A c
bac 00 F
Fab = F
2
+ O A3 ,
where now the non-commutative 00 product is defined to be
i
a b

e 2 ab x 0 x 00 1 0
00
00
= i
)
.
f (x )g(x
f (x) g(x)
a b
x 0 =x 00 =x
2 ab x 0 x 00
To check the result, one may differentiate it to get Eq. (10) up to order O(A 3 ). For the
abelian case that we are interested in, the transformation becomes

bab cd F
bac 0 F
bbd A c 0 d F
bab + O A 3 ,
(12)
Fab = F
where the commutative 0 operates as
=
f(x) 0 g(x)
sin( 12 ab xa0 xb00 )

1
a b
2 ab x 0 x 00

0
00
f (x )g(x
)
x 0 =x 00 =x
(13)
Now if one compares Eq. (12) for infinitesimal and finite , one may conclude that to
go from the ordinary product of two open string fields at = 0 to finite one should use
the following transformation as well:
fg|=0 f 0 g|6=0 ,
(14)
where f and g are any arbitrary open string fields. We will see in Section 4 that this
multiplication rule is consistent with string theory scattering amplitudes of two open and
one closed string states.
214
Now with the help of Eqs. (12) and (14), one can write the DBI coupling (7), (8) and
(9) at = 0 in terms of non-commutative fields at 6= 0 corresponding to the open string
vertex operator. In doing so, one should first, using (14), replace ordinary multiplication of
two open string fields by 0 multiplication. Then, using (12), the ordinary fields are shifted
to their non-commutative counterparts. For example, transformation of Eq. (8) up to three
open string states becomes

1 ab
1
1
ab
+ V (hba bba ) + Tr(V ) 4

V a i 0 b i
L2,1 = c
2
2
2 2

1 ab 0 cd
1 ab 0 cd
bc
ab
V fbc V fda + V fba V fdc V
hbc bbc +
4
8
2 2

1
V cd d i 0 a i V cd fde 0 V ef ff a + V cd fda 0 V ef ff e
2
1
+ V ab a i 0 b i + V ab (hij bij )b i 0 a j + V ab hi(b a) i
2 2

bi[b a] i 0 V cd fdc 2V ab hi(b c) i bi[b c] i 0 V cd fda .
(15)
We now turn to the string theory side and evaluate these couplings using the conformal
field theory technique.
4. Scattering calculations
In this section we calculate various string scattering amplitudes. The amplitude
describing scattering of two closed strings from D-brane with a background magnetic flux
was calculated in [14]. There, by analyzing the t-channel of the amplitude, we were able
to find the linear coupling of closed string fields to the D-brane and to show that they are
consistent with the coupling in (5).
4.1. Closedopen couplings
Here, we wish to compare the coupling of one massless non-commutative open string
field and one closed string field on the D-brane to the results of the appropriate string
couplings. In field theory, this coupling can be read from Eq. (7) and the transformation
(12). In string theory side, on the other hand, this coupling is given by the string scattering
amplitude of one open and one closed string state from the D-brane, that is,
Z

(16)
ANS,NSNS dx1 d 2 z2 V NS (k1 , 1 , x1 ) V NSNS (p2 , 2 , z2 , z 2 ) .
The details of the vertex operators can be found in Appendix A. We already assumed in
Section 2 that the closed string fields in the DBI action (1) are independent of transverse
coordinates. On the string theory side, it means that the momentum of closed string
vertex operators have components only in the world-volume directions, i.e., pi = 0. The
215
techniques in calculating the above string scattering amplitude may be found in Refs. [22
24]. The final result is
p
a
Tp c
NS,NSNS
2k1a G 2 D G T
=
1
A
2

a
2k1a G 2 D G T
1 p2 D G T
1 Tr(2 D) ,
p
Tp c
NS,
1 G p2 ,
=
A
2
with D and G matrices coming from closed and open string vertices, respectively
p
(see Appendix A). We have also normalized the amplitudes at this point by i Tp c/2,
p
where 1/ Tp , and Tp c are open string, closed string and D-brane coupling constants,
respectively. Substituting the appropriate polarizations for the open and closed string fields
from Appendix A, one finds
p

A(, h) = Tp c 1 N 2 V T k1 + k1 V T 2 N 1 ,
p
A(a, h) = Tp c 1 V 2 V k1 k1 V 2 V 1

+ k1 VA 1 Tr 2 V T ,
p

Tp c
1 V V k1 k1 V V 1 + k1 VA 1 Tr(V ) 4 ,
A(a, ) =
2 2
p
Tp c
k 1 VA 1 ,
A(a, ) =
2
where here and in the scattering amplitudes in subsequent sections h stands for both
graviton and KalbRamond fields. In writing the above equations, we have used the onshell condition k1 VS 1 = 0 (see Appendix A) and momentum conservation k1a + p2a = 0.
These terms are reproduced by the following action:

p

1
1
1
L 1,1 = Tp c V ab fba + V ab (hba bba ) + Tr(V ) 4
2
2
2 2

1
V ab hbc bbc + bc V cd fda + 2V ab hi(b a) i bi[b a] i .
2 2
(17)
This is consistent with the DBI interaction (7) and the transformation (12) up to two open
string fields. Note that transformation of a can be read from the dimensional reduction
of (12).
4.2. Openopenopen couplings
Next, we turn to the coupling of three open string states. On the string theory side this
coupling is given by
Z

ANS,NS,NS dx1 dx2 dx3 V NS (1 , k1 , x1 ) V NS (2 , k2 , x2 ) V NS (3 , k3 , x3 ) ,
216
where the appropriate vertex operators are given in Appendix A. Using the world-sheet
conformal field theory, it is not difficult to perform the correlators above and show that the
integrand is invariant under SL(2, R). Fixing this symmetry, one finds
c sin(l)
k 1 G G T 3 1 G G T 2 + k 3 G G T 2 1 G G T 3
A= p
Tp

k 3 G G T 1 2 G G T 3 ,
where we have defined l 2k1 V T F V k2 = 2k1 VA k2 and VA is the antisymmetric
part of the V matrix (6). We have also normalized the amplitude by the appropriate
p
coupling factor c/2 Tp . The sin(l) factor above arises basically from two different
phase factors corresponding to two distinct cyclic orderings of the vertex operators. Each
phase factor stems from the second terms of the world-sheet propagator (A.7). Using
polarization for scalar and gauge field, one finds the following non-vanishing terms:
c sin(l)
p
1 N 2 k 1 VS 3 ,
Tp
c sin(l)
(1 VS 2 3 VS k1 + 2 VS 3 1 VS k2
A(a, a, a) = p
Tp
+ 1 VS 3 2 VS k3 ),
A(, , a) =
(18)
where VS is the symmetric part of the V matrix (6). These couplings may be reproduced
by

ic
p
2VSab a i 0 a b , i M VSab fbc 0 VScd a d , a a M ,
(19)
L 3,0 =
8 Tp
where the Moyal bracket is defined in (11) and the 0 operates on the whole Moyal bracket.
This fixes the relation between the non-commutative two-form in (11) and the background
metric and flux to be
ab = 4VAab .
(20)
The DBI interaction (8) and the transformations (14) and (12) reproduce various terms
having two, three and more open string fields. Terms which have three fields contain
two different parts. One part is just the above action and the other part which has threemomentum may be verified to be zero. In the action (19), one may use, instead of 0 ,
another multiplication rule, e.g., ordinary multiplication [25] or multiplication [13]; all
produce the same momentum space couplings (18). Hence, although the above calculations
of three open string couplings can fix the non-commutative multiplication rule in the
definition of field strength, i.e. , it can not however uniquely fix the multiplication rule
between two open string field strengths.
If one multiplies the action (19) with a closed string field, say a tachyon, then
the ordinary and 0 and any other multiplication rule produce different momentum
space couplings. In this case, string theory calculations can be used to fix uniquely the
multiplication rule. To fix the multiplication rule between two open string fields, we
calculate string theory couplings of two open and one closed string states in momentum
space. These can be extracted from string scattering amplitudes of two open and one closed
string states from the non-commutative D-brane, the calculation of which we now turn to.
217
4.3. Closedopenopen amplitudes

The scattering amplitude of one closed and two open string states can be related to the
appropriate amplitude of four open string states in type I theory [22,23]. However, type I
theory does not have an open string tachyon, so the scattering amplitude describing the
decay of two massless open strings to one closed string tachyon in type 0 is not related
to the known amplitude in type I theory. So we explicitly calculate the tachyon amplitude,
while using the idea in [22,23] we find the massless closed string amplitude from the known
amplitude of type I theory.
4.3.1. Tachyon amplitudes
The amplitudes describing interaction of one closed string tachyon and two massless
open strings is given by
Z

ANS,NS, dx1 dx2 d 2 z V NS (1 , k1 , x1 )V NS (2 , k2 , x2 )V (p3 , z3 ) ,
where the closed and open string vertex operators are given in Appendix A. Here
again, using the appropriate world-sheet propagators from Ref. [22], one can evaluate
the correlations above and show that the integrand is SL(2, R) invariant. Gauging this
symmetry by fixing z3 = i and x2 = , one arrives at

Z
A 22s2 dx1 (2s + 1)1 G G T 2
2i1 G D T p3 2 G p3 2i1 G p3 2 G D T p3
+
x1 i
x1 + i
(x1 i)sl (x1 + i)s+l ,

where the integral is taken from to +, and s = (p3 V T )2 = 2k1 VS k2 . This
integral can be carried out and the result is
A=

(2s)
ic
a1 (s + l) a2 (s l)
,
2
(1 s l) (1 s + l)
(21)
where a1 and a2 are two kinematic factors depending only on the space-time momentum
and polarization vectors
a1 = 1 G D T p3 2 G p3 ,
a2 = (s + l)1 G G T 2 + 1 G p3 2 G D T p3 .
We have also normalized the amplitude (21) at this point by the coupling factor ic/2.
A check of our calculations is that the amplitude (21) satisfies the Ward identity associated
with the gauge invariance of the open string states, i.e., the amplitude vanishes upon
substituting ia kia . This amplitude has the pole structure at m2open = n/ 0 . 1
1 We explicitly restore 0 here. Otherwise our conventions set 0 = 2.
218
4.3.2. NSNS amplitudes

Next, we evaluate the amplitude describing the decay of two massless open NS strings
into one massless closed NSNS state. Using the idea in [22,23], we relate this amplitude
to the amplitude of four massless open NS strings. Hence, we begin with the closed string
amplitude, which is given by
Z

(22)
A dx1 dx2 d 2 z3 V NS (k1 , 1 , x1 ) V NS (k2 , 2 , x2 ) V NSNS (p3 , 3 , z3 , z 3 ) .
If one evaluated the above correlators, one would find that the integrand is SL(2, R)
invariant. Similar to the tachyon amplitude, the appropriate way to fix this gauge would
be to fix operators at {z3 , x1 , z3 , x2 } = {i, y, i, }.
To relate the calculation here to that of a four-point amplitude of open superstrings [26],
we write the latter in an SL(2, R) invariant form, that is
Z
4k1 k2 4k1 k3 4k1 k4 4k2 k3 4k2 k4 4k3 k4
x13 x14 x23 x24 x34
A0 dx1 dx2 dx3 dx4 x12

a10
a20
,
(23)
x12x13 x24 x34 x13x14 x23 x24
where the kinematic factors a10 and a20 are
a10 = 4{1 2 3 4 k2 k3 + 1 4 k1 2 k4 3 + 2 3 k2 1 k3 4
+ 3 4 k 3 2 k 4 1 + 1 2 k 2 3 k 1 4 1 3 k 1 2 k 3 4
2 4 k2 1 k4 3 3 4 k4 2 k3 1 1 2 k1 3 k2 4 },
a20
(24)
= 4{k2 k4 1 4 2 3 + k2 k3 1 3 2 4 1 2 3 4 k2 k3
1 4 k 1 2 k 4 3 2 3 k 2 1 k 3 4 3 4 k 3 2 k 4 1
1 2 k 2 3 k 1 4 + 2 3 k 3 1 k 2 4 + 1 4 k 4 2 k 1 3
+ 2 4 k4 1 k2 3 + 1 3 k3 2 k1 4 },
(25)
where i are the polarization of external states. Since we are interested in this amplitude
for transforming it to the scattering amplitude of one closed and two open string states, we
do not consider the phase factor associated with the second term of the propagator (A.7). If
one fixes the SL(2, R) symmetry by fixing the operators at {x1 , x2 , x3 , x4 } = {1, x, 1, },
one finds

0
Z+1
a20
a1
0
4k1 k3 1
4k1 k2
4k2 k3
(1 + x)
(1 x)
.
(26)
A (2)
1+x 1x
1
Now the scattering amplitude (22) can be read from the amplitude (26) by replacing
2k1 p3 D,
k2 k1 G,
3 1 (3 D) ,
2k3 p3 ,
2 1 G ,
k4 k2 G,
4 2 G .
Under these transformations the amplitude (26) transforms to

Z+
a2
a1
2s1
(1 + iy)sl (1 iy)s+l
A (2)
1 + iy 1 iy
x iy,
(27)
(28)
219
and the kinematic factors (24) and (25) become

a1 = (s + l)2 G 3 D G T 1 + 2k2 G 3 D G T 2 p3 D G T 1
+ 21 G 3 D G T k1 p3 G T 2 + 22 G 3 D G T k2 p3 G T 1
+ 2k1 G 3 D G T 1 p3 D G T 2
Tr(3 D) p3 D G T 1 p3 G T 2
4k2 G 3 D G T k1 2 G G T 1 22 G 3 D p3 k2 G G T 1
2p3 D 3 D G T 1 k1 G G T 2 ,
a2 = 2s 1 G 3 D G T 2 + (s + l) Tr(3 D)1 G G T 2
(s + l)2 G 3 D G T 1 2k2 G 3 D G T 2 p3 D G T 1
21 G 3 D G T k1 p3 G T 2 22 G 3 D G T k2 p3 G T 1
2k1 G 3 D G T 1 p3 D G T 2 + 21 G 3 D p3 k1 G G T 2
+ 2p3 D 3 D G T 2 k2 G G T 1
+ 4k1 G 3 D G T k2 1 G G T 2
+ Tr(3 D)p3 G T 1 p3 D G T 2 .
(29)
The integral (28) is doable and the result is the same as Eq. (21) with the above kinematic
factors. As a check of our calculations, we have inserted the dilaton polarization (A.8) into
the kinematic factor and found that it is independent of the auxiliary vector ` . Another
check is that the amplitude satisfies the Ward identity associated with the gauge invariance
of the open string states.
4.3.3. Massless poles
Given the general form of the string amplitude in Eq. (21), one can expand this amplitude
as an infinite sum of terms reflecting the infinite tower of open string states that propagate
on the world-volume of D-brane. In the low-energy domain, i.e., 0 mopen 1, the first term
representing the exchange of massless string states dominates. In this case the scattering
amplitude (21) reduces to
A=
ic sin(l)
(a1 + a2 ) + ,
4s
(30)
where dots represent contact terms and the infinite massive poles. Making the appropriate
explicit choice of polarization, we find
ic sin(l)
(l1 N 2 ) + 1 2,
8s

ic sin(l) l
1 VS 2 + 1 V T p3 2 V p3 + 1 2,
As (a, a, ) =
4s
2

ic sin(l) l
Tr(D) + 2 4k1 V V k2
As (, , ) =
2
8 2 s
As (, , ) =
1 N 2 + 1 2,
(31)
220
As (a, a, ) =

ic sin(l)
l
Tr(D) + 2 p3 V 1 p3 V T 2 + 1 VS 2
2
8 2 s
+ 4k1 VS 2 (p3 V V 1 1 V V p3 )

4k1 V V k2 1 VS 2 + 1 2,
ic sin(l)
(p3 D 3 N 1 + 1 N 3 D p3 )k1 VS 2 ,
2s

ic sin(l) l
Tr(3 D) 4k1 V 3T V k2 1 N 2 + 1 2,
As (, , h) =
4s
2

ic sin(l)
l
Tr(3 D) p3 V 1 p3 V T 2 + 1 VS 2
As (a, a, h) =
4s
2
As (, a, h) =
4k1 V 3T V k2 1 VS 2 + 2 p3 D 3 V T 2

T
2 V 3 D p3 1 VS k2 + 1 2.
(32)
In writing explicitly the above massless poles, one finds some terms proportional to s as
well. We will add these terms which have no contribution to the massless poles of the field
theory to the contact terms in (36). These amplitudes should be reproduced in the s-channel
of the field theory. We present the calculation explicitly for the decay of two gauge fields
into a tachyon. This amplitude can be evaluated in field theory as
e a
A0s (a, a, ) = V
a
ea
G

ab

eaaa b ,
V
(33)
where the propagator and the vertices can be read from (9), (17) and (19). They are
i (VS1 )ab
,
cp s

Tp c
e a a =
p3 VAa ,
V
2

eaaa a = c sin(l)
p (1 VS 2 k1 2k1 VS 2 1 ) VSa + 1 2.
V
2 Tp
ea
G
ab
(34)
In writing the above propagator from (9), we have used the covariant gauge VSab a A b = 0.
Replacing the above propagator and vertices into (33), one finds exactly the string massless
pole As (a, a, ). Similar calculations for the other open string modes reproduce exactly the
corresponding massless poles of the string amplitudes (31).
Although the whole string scattering amplitude (21) is gauge invariant, vanishing upon
substituting ia kia , its massless pole (31) is not gauge invariant, which can be checked
explicitly. So one expects the low-energy contact terms of the string amplitude not to be
gauge invariant either. However, the combination of the massless pole and contact terms
should be gauge invariant. We turn now to the evaluation of these low-energy contact terms
of the string amplitude (21).
221
4.3.4. Contact terms

Having examined in detail the massless poles of string amplitudes, we now extract the
low-energy contact terms of the string amplitude (21). Expanding the gamma function
appearing in this amplitude, one finds

a1 a2 sin(l)
ic a1 + a2 sin(l)
+
A=
2
2
s
2
l

X
sin(l)
(2n+1)
2
(2n + 1)l
+ k O(s, l) .
(35)
+ (a1 + a2 )
l
n=1
The factor sin(l)/(l) appears for all the contact terms. This indicates that the
multiplication rule between any two open strings is 0 . However, as we will see in a
moment, some of the terms in the kinematic factors a1 and a2 are proportional to l.
Hence, these terms have an overall factor of sin(l), which produces the non-commutative
multiplication rule instead of 0 in field theory. It is important to note that these higher
order derivative terms associated with or 0 are not the ones stemming from the massive
pole of the string amplitude. These factors appear for both the low-energy contact terms
and the contact terms corresponding to massive poles. The terms in the first line of (35) are
the low-energy massless pole and contact terms, whereas the terms in the second line are
the effect of the massive poles of the string amplitude (21).
As anticipated above, not all the low-energy contact terms are gauge invariant.
Therefore, we separate the contact terms into gauge invariant and gauge non-invariant
terms. Moreover, we divide the gauge non-invariant terms into two parts: terms which
have, apart from the overall factor, no momentum and two momenta. That is
ic sin(l)
ng
g
(a1 a2 ) AM
c + Ac + Ac ,
4l
ng
(36)
where AM
c , Ac and Ac are gauge non-invariant terms which have no momentum, two
momentum and gauge invariant terms, respectively. The AM
c terms are
ic sin(l)
1 VA 2 ,
4
ic sin(l)
1 V V 2 2 V V 1
AM
c (a, a, ) =
4 2

Tr(D) + 2 1 VA 2 ,
AM
c (a, a, ) =

ic sin(l)
1 N 3 V T 2 + 2 V T 3 N 1 ,
2
ic
sin(l)
2 V T 3 V T 1 1 V T 3 V T 2
AM
c (a, a, h) =
2

12 Tr(3 D)1 VA 2 .
AM
c (, a, h) =
In all the above terms, the factor sin(l)/(l) reduces to sin(l), which produces the
operator between two open string fields. In fact it is not difficult to see that the above terms
are exactly reproduced by the following action:
222

1 ab
1
ic
1
ab
Tr(V
)
Tr
V
+
V
=
,
a
(h
b
)
+
a
L M
b
a
ba
ba
2,1
M 2
4
2
2 2

1
V ab hbc bbc + bc V cd a d , a a M
2 2

+ 2V ab hi(b a a) , i M bi[b a a] , i M ,
(37)
where the non-commutative parameter of the Moyal bracket is the one appearing in (20).
The appearance of the Moyal bracket in (37) is consistent with the transformation (12).
ng
The gauge non-invariant terms Ac are
ic sin(l)
(1 VA k1 2 VA k1 ) + 1 2,
2l

ic sin(l)
ng
Tr(D) + 2 k1 VA 1 k1 VA 2
Ac (a, a, ) =
4 2l

2k1 VA 2 (k1 V V 1 1 V V k1 ) + 1 2,
(38)

ic
sin(l)
ng
21 N 3 V T k1 + 2k1 V T 3 N 1
Ac (, a, h) =
2l
2 VA k 1 ,
ic sin(l)
ng
Tr(3 D)k1 VA 1 k1 VA 2
Ac (a, a, h) =
2l

2k1 VA 2 k1 V 3T V 1 1 V 3T V k1 + 1 2.
ng
Ac (a, a, ) =
ng
It is easy to check that, as expected, by replacing ia kia the non-zero terms of AM

c +Ac
cancel exactly the non-zero terms of the massless poles in (31). The factor sin(l)/(l)
reproduces the 0 operator between two open string fields. The above momentum space
couplings are reproduced by the following action:

1
ng
ab
0 1 cd
V b fdc + V ef (hf e bf e )
L2,1 = c Tr (VA ) a a
2
2

1
1
cd
hde bde + de V ef b ff c
+ Tr(V ) 4 V
2 2
2 2

+ 2V cd hi(d c) b i bi[d c] b i
.
(39)
Here also the appearance of the non-commutative gauge field and the derivative of the field
strength is consistent with the transformation (12).
We now turn to the gauge invariant terms. The contact terms of two open string scalars
and one closed string are
ic sin(l)
(s1 N 2 ),
4l

ic sin(l)
g
Tr(D) + 2 (s1 N 2 )
Ac (, , ) =
8 2 l
g
Ac (, , ) =

4(s + k1 V V k2 + k2 V V k1 )1 N 2 ,
(40)

g
Ac (, , h) =
223
ic sin(l) 1
2 Tr(3 D)(s1 N 2 )
4l

4k1 V 3T V k2 1 N 2 2(s l)1 N 3 N 2
+ 1 2.
These terms reproduce exactly the appropriate terms in (15). This confirms the conjectured
multiplication rule (14) between two open string fields. Now the other gauge invariant
contact terms are

ic sin(l)
2k1 VS 2 1 N 3 V T k2 k2 V T 3 N 1
2l

+ s 1 N 3 V T 2 2 V T 3 N 1

2k2 VA 2 (k1 V 3 N 1 + 1 N 3 V k1 ) ,

ic sin(l)
l
g
1 VA k 1 2 VA k 2 + 1 VA 2
Ac (a, a, ) =
4l
2

s
1 VS 2 + 1 VA k 2 2 VA k 1 1 VS k 2 2 VS k 1 ,
2

ic
sin(l)
g
Tr(D) + 2 1 VA k1 2 VA k2
Ac (a, a, ) =
8 2 l
4k1 VA 1 (k2 V V 2 2 V V k2 )

l
s
1 VA 2 1 VS 2
+ Tr(D) + 2
2
2

+ 1 VA k 2 2 VA k 1 1 VS k 2 2 VS k 1
g
Ac (, a, h) =
+ 2s1 V V 2 41 VS 2 k1 V V k2

+ 4k2 VS 1 (k1 V V 2 + 2 V V k1 ) ,

ic sin(l)
g
Tr(3 D)1 VA k1 2 VA k2
Ac (a, a, h) =
4l

4k1 VA 1 k2 V 3T V 2 2 V 3T V k2

l
s
+ Tr(3 D) 1 VA 2 1 VS 2 + 1 VA k2 2 VA k1
2
2

1 VS k2 2 VS k1 + 2s1 V 3T V 2
41 VS 2 k1 V 3T V k2 + 4k2 VS 1 k1 V 3T V 2

T
+ 2 V 3 V k 1
(41)
plus (1 2) for equations that have two gauge fields. These gauge invariant terms are not
fully consistent with the DBI terms in (15). However, adding the following terms to (15),
the resulting action reproduces all the contact terms in (41),
224

1
1
g
L 2,1 = c Tr V ab fbc 0 (VA )cd fda + V ab (hba bba )
2
2

1
1
+ Tr(V ) 4 V ab hbc bbc + bc
2 2
2 2

cd
0
ef
ab
V fde (VA ) ff a + V (hib bib )c i 0 (VA )cd fda

+ (VA )ab (hic + bic )b i 0 V cd fda .
(42)
Now the string theory contact terms in Eqs. (17), (37), (39) and (42) are exactly the
DBI interactions (7) in which using the transformation (12) with the normalization (3) its
ordinary fields are written in terms of their non-commutative fields up to three open string
fields. This ends our illustration of consistency between string theory scattering amplitudes
and ordinary DBI action in which using transformations (12) the ordinary open string fields
are written in terms of non-commutative fields.
5. Discussion
Having integrated the SeibergWitten differential equation in a special path, we write
ordinary gauge fields in terms of non-commutative fields for a finite non-commutative
two-form parameter. We then use this change of variables to express the ordinary DBI
action in terms of non-commutative fields. We have also proposed a transformation for
multiplication of two arbitrary open string fields when writing the DBI action in terms
of non-commutative variables. The resulting action was then compared with various
world-sheet perturbative string theory scattering amplitudes. We find complete agreement
between the field theory and string theory results. This indicates that the perturbative string
theory knows about the SeibergWitten differential equation.
Our calculations of string scattering amplitude of two massless open and one closed
string states confirmed our proposed commutative multiplication rule (14) between two
open string fields. It would be interesting to perform the calculation of one closed and
three open string states to find the transformation of multiplication rules between three
open string states.
The scattering amplitudes considered in this paper fixed the relation between ordinary
and non-commutative fields up to three non-commutative fields. In principle, the perturbative string theory knows about all the terms in this change of variable. So it would be
interesting to extend our method to higher point functions to find other terms of the relation.
In Subsection 4.2 we reach the conclusion that the string scattering amplitude (18)
reproduces the action (19) in field theory. In Ref. [13], Seiberg and Witten conclude to
a different action, namely one similar to (19) with instead of the 0 operator. These two
actions are identical up to some total derivative terms. In fact, using the antisymmetric
property of the non-commutative parameter, one finds
bbc 0 V cd F
bda = V ab F
bbc V cd F
bda = V ab F
bbc V cd F
bda
VSab F
S
S
S
S
S
up to some total derivative terms.
225
Our calculations of the string scattering amplitude of two open and one closed
string states from a non-commutative D-brane are also original. Using two-dimensional
conformal field theory, we performed calculations for the scattering amplitude of two
massless open string states and one closed string tachyon explicitly. Whereas, using the
idea that the scattering amplitude of open and closed string states can be read from the
appropriate amplitude of only open string states [22,23], we were able to find an expression
for the scattering amplitude of two open and one massless closed string states from noncommutative D-branes.
Acknowledgements
I would like to acknowledge useful conversations with R.C. Myers. I would also like to
thank ICTP for hospitality. This work was supported by University of Birjand and IPM.
When finalizing this paper, Ref. [29] came out which has some overlap with the results
in Section 4.
Appendix A. Perturbative string theory with background field

In perturbative superstring theories, to study the scattering amplitude of some external
string states in a conformal field theory frame, one usually evaluates the correlation
function of their corresponding vertex operators making use of some standard conformal
field theory propagators [27,28]. In a trivial flat background one uses an appropriate linear
-model to derive the propagators and define the vertex operators. In a non-trivial D-brane
background the vertex operator remains unchanged while the standard propagators need
some modification. Alternatively, one may use a doubling trick to convert the propagators
to standard form and give the modification to the vertex operators [22]. In this appendix
we consider a D-brane with constant gauge field strength/or antisymmetric KalbRamond
field in all directions of the D-brane. The modifications arising from the appropriate linear
-model appear in the following boundary conditions [15]: 2
y Xa iF a b x Xb = 0 for a, b = 0, 1, . . . , p,
X = 0 for i = p + 1, . . . , 9,
i
(A.1)
(A.2)
where Fab are the constant background fields, and these equations are imposed at y = 0.
The world-volume (orthogonal subspace) indices are raised and lowered by ab (N ij ) and
ab (Nij ), respectively. Now we have to understand the modification of the conformal field
2 Our notation and conventions follow those established in Ref. [22]. So we are working on the upper-half
plane with boundary at y = 0 which means that y is the normal derivative and x is the tangential derivative.
Our index conventions are that lowercase Greek indices take values in the entire ten-dimensional space-time, e.g.,
, = 0, 1, . . . , 9; early Latin indices take values in the world-volume, e.g., a, b, c = 0, 1, . . . , p; and middle
Latin indices take values in the transverse space, e.g., i, j = p + 1, . . . , 8, 9. Finally, our conventions set `2s =
0 = 2.
226
theory propagators arising from these mixed boundary conditions. To this end consider the
following general expression for the propagator of X (z, z ) fields:

X (z, z ) X (w, w)
D log(z w)
D (z w),
= log(z w) log(z w)
(A.3)
where D is a constant matrix. To find this matrix, we impose the boundary condition
(A.2) on the propagator (A.3), which yields
ab D ba F ab F a c D bc = 0
(A.4)
for the world-volume directions, D ij = N ij for the orthogonal directions, and D ia = 0

otherwise. Now Eq. (A.4) can be solved for D ab , that is
Dab = 2( F )(1)
ab ab = 2Vba ab ,
(A.5)
where matrix V is the dual metric that appears in the expansion of DBI action (6). Note
that the D is an orthogonal matrix, i.e., D D = .
Using the two-dimensional equation of motion, one can write the world-sheet fields in
terms of right- and left-moving components. In terms of these chiral fields, closed NSNS
and open NS vertex operators are

z), (
z), (
z), p : ,
V NSNS =: Vn X(z), (z), (z), p ::Vm X(
(x) + (x),
(x) + (x),
k :,
V NS =: Vn X(x) + X(x),
where is the super partner of the world-sheet field X and is a world-sheet superghost
field. The indices n, m refer to the superghost charge of the vertex operators, and p and k
are closed and open string momenta, respectively. In order to work with only right-moving
fields, we use the following doubling trick:
X (z) D X (z),
(z) D (z),
z) (z).
(
(A.6)
These replacements in effect extend the right-moving fields to the entire complex plane
and shift the modification arising from the mixed boundary condition from propagators to
vertex operators. Under this replacement, the world-sheet propagators between all rightmoving fields take the standard form [14] except the following boundary propagator:

i
(A.7)
X (x1 ) X (x2 ) = log(x1 x2 ) + F (x1 x2 ),
2
where (x1 x2 ) = 1(1) if x1 > x2 (x1 < x2 ). Note that the orthogonality property of
the D matrix is an important ingredient for writing the propagators in the standard form.
Under the transformation (A.6) the vertex operators become

V NSNS = Vn X(z), (z), (z), p ::Vm D X(z), D (z), (z), p : ,

V NS =: Vn X(x) + D X(x), (x) + D (x), 2(x), k : .
The vertex operator for the closed string tachyon, massless NSNS and massless NS states
are
227
V =: Vn (p, z)::Vm (p D, z ): ,
V NSNS = ( D) : Vn (p, z)::Vm (p D, z ): ,
V NS = ( G) : Vn (2k V T , x): ,
where G ab = (ab + D ab )/2 = V ba for the gauge field, G ij = (ij D ij )/2 = N ij for the
scalar field and G ai = 0 otherwise. The open string vertex operators in the (0) and (1)
pictures are
V0 (k, x) = X (x) + ik (x) (x) eikX(x),

V1 (k, x) = e(x) (x)eikX(x),
V0 (k, x) = ik (x)eikX(x),
V1 (k, x) = e(x)eikX(x).
The physical conditions for the massless open string are
k VS k = 0,
k VS = 0
and for massless closed string are p2 = 0 and p = 0, where is the closed
string polarization which is traceless and symmetric (antisymmetric) for graviton (Kalb
Ramond) and

1
`p=1
(A.8)
= ` p ` p ,
8
for the dilaton. Using the fact that D is orthogonal matrix, one finds the following
identities:
ab
ij
D GT
= G ab ,
D G T = N ij ,
G GT = GS ,
where G S is the symmetric part of the G matrix.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
A. Sen, An introduction to non-perturbative string theory, hep-th/9802051.

C. Vafa, Lectures on strings and dualities, hep-th/9702201.
J. Polchinski, Rev. Mod. Phys. 68 (1996) 1245, hep-th/9607050.
M.J. Duff, Int. J. Mod. Phys. A 11 (1996) 5623, hep-th/9608117.
J.H. Schwarz, Nucl. Phys. Proc. Suppl. 55B (1997) 1, hep-th/9607201.
J. Polchinski, TASI Lectures on D-branes, hep-th/9611050.
J. Polchinski, S. Chaudhuri, C.V. Johnson, Notes on D-branes, hep-th/9602052.
W. Taylor, Lectures on D-branes, gauge theory and M (atrices), hep-th/9801182.
A. Connes, M.R. Douglas, A. Schwarz, J. High Energy Phys. 9802 (1998) 003, hep-th/9711162.
M.R. Douglas, C. Hull, J. High Energy Phys. 9802 (1998) 008, hep-th/9711165.
F. Ardalan, H. Arfaei, M.M. Sheikh-Jabbari, J. High Energy Phys. 9902 (1999) 016, hepth/9810072.
A.A. Tseytlin, BornInfeld action, supersymmetry and string theory, hep-th/9908105.
N. Seiberg, E. Witten, J. High Energy Phys. 9909 (1999) 032, hep-th/9908142.
M.R. Garousi, J. High Energy Phys. 9812 (1998) 008, hep-th/9805078.
J. Dai, R.G. Leigh, J. Polchinski, Mod. Phys. Lett. A 4 (1989) 2073.
228
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
I.R. Klebanov, A.A. Tseytlin, Nucl. Phys. B 546 (1999) 155, hep-th/9811035.
R.G. Leigh, Mod. Phys. Lett. A 4 (1989) 2767.
C.G. Callan, C. Lovelace, C.R. Nappi, S.A. Yost, Nucl. Phys. B 308 (1988) 221.
A. Abouelsaood, C.G. Callen, C.R. Nappi, S.A. Yost, Nucl. Phys. B 280 (1987) 599.
M.R. Garousi, Nucl. Phys. B 550 (1999) 225, hep-th/9901085.
I.R. Klebanov, A.A. Tseytlin, Nucl. Phys. 547 (1999) 143, hep-th/9812089.
M.R. Garousi, R.C. Myers, Nucl. Phys. B 475 (1996) 193, hep-th/9603194.
A. Hashimoto, I.R. Klebanov, Phys. Lett. B 381 (1996) 437, hep-th/9604065.
A. Hashimoto, I.R. Klebanov, Nucl. Phys. B (Proc. Suppl.) 55B (1997) 118, hep-th/9611214.
M.M. Sheikh-Jabbari, Phys. Lett. B 450 (1999) 119, hep-th/9810179.
J.H. Schwarz, Phys. Rep. 89 (1982) 223.
M.E. Peskin, Introduction to string and superstring theory II, in: H. Haber (Ed.), From the
Planck scale to the weak scale: Toward a theory of the universe, Proceedings TASIs 86, World
Scientific, Singapore, 1987.
[28] V.A. Kostelecky, O. Lechtenfeld, W. Lerche, S. Samuel, S. Watamura, Nucl. Phys. B 288 (1987)
173.
[29] S. Hyun, Y. Kiem, S. Lee, C.-Y. Lee, Closed strings interacting with non-commutative D-branes,
hep-th/9909059.

(p + 1)-Dimensional noncommutative YangMills

and D(p 2) branes
J.X. Lu a,1 , Shibaji Roy b,2
a Randall Physics Laboratory, University of Michigan, Ann Arbor, MI 48109-1120, USA
b Saha Institute of Nuclear Physics, 1/AF Bidhannagar, Calcutta 700 064, India
Received 22 February 2000; accepted 3 March 2000
Abstract
We consider systems of nonthreshold bound states (D(p 2), Dp), for 2 6 p 6 6, in type II string
theories. Each of them can be viewed as Dp branes with a nonzero (rank two) NeveuSchwarz Bfield. We study the noncommutative effects in the gravity dual descriptions of noncommutative gauge
theories for these systems in the limit where the brane worldvolume theories decouple from gravity.
We find that the noncommutative effects are actually due to the presence of infinitely many D(p 2)
branes in the (D(p 2), Dp) system which play the dominant role over the Dp branes in the large
B-field limit. Our study indicates that Dp branes with a constant B-field represents dynamically the
system of infinitely many D(p 2) branes without B-field in the decoupling limit. This implies an
equivalence between the noncommutative YangMills in (p + 1) dimensions and an ordinary Yang
Mills with gauge group U () in (p 1) dimensions. We provide a physical explanation for the new
scale which measures the noncommutativity. 2000 Elsevier Science B.V. All rights reserved.
PACS: 11.25.-w; 11.15.-q
Keywords: Dp branes in nonzero NS B-field; Non-commutative YangMills theory
1. Introduction
Gauge theories on noncommutative spaces can arise in certain limits of string theory [1
5]. For a system of Dp branes, a constant NeveuSchwarz (NS) B-field in spatial
directions on the worldvolume is the key to making the space noncommutative. Following
Maldacenas AdS/CFT correspondence for the usual YangMills theory, it is quite natural
to look for the gravity duals of the noncommutative gauge theories (NCYM). In order to do
so, one has to identify a limit in which closed strings decouple from the open strings which
end on the Dp-branes and the dynamics for the open strings is entirely described by their
1 jxlu@umich.edu
2 roy@tnp.saha.ernet.in
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 1 5 2 - 8
230
J.X. Lu, S. Roy / Nuclear Physics B 579 (2000) 229249
low energy gauge theory on a noncommutative space. 3 Recently, this has been achieved
in general by Seiberg and Witten [6], and for special systems by Hashimoto and Itzhaki
[7] and by Maldacena and Russo [8]. The correspondence between NCYM and the gravity
duals is proposed in [7,8]. This correspondence has also been studied for many other known
systems recently in [9,10]. The thermodynamics for NCYM has also been studied in [8,10
12]. Seiberg and Witten have shown that noncommutative gauge theories are equivalent to
usual gauge theories by a change of variables. 4 This can be understood physically by the
fact that the two descriptions are due to the different choices of regulator in the sigma model
expansion and physics in a well-defined theory should be independent of such choices.
Whether we have the usual YM or NCYM depends crucially on the vanishing or the
nonvanishing of the constant worldvolume 2-form field 5 F 2 0 F + B. Here B is the
NS 2-form field and F is the worldvolume 2-form gauge field strength. So whenever we
have a constant F -field, we can potentially have a NCYM description. For convenience,
we, as usual, make a gauge choice such that we have F = 0 while B takes the given
constant value of F .
It is well-known that the constant F field on the worldvolume of a Dp brane implies that
there are infinitely long fundamental strings (for short, F-strings) and/or D(p 2) branes
in the Dp-brane from the spacetime perspective [1317] (actually there are infinitely many
such strings or D(p 2) branes). The former gives the so-called (F, Dp) nonthreshold
bound states [1416] while the latter gives (D(p 2), Dp) nonthreshold bound states [13,
15,17]. 6 This relationship between the constant F -field on the worldvolume of a Dpbrane and the F-strings and/or D(p 2)-branes living on a Dp-brane must imply a close
connection between the noncommutativity and these F-strings and/or D(p 2) branes.
The aim of this paper is to reveal the connection and possible consequences of this. In
general, if there exists a nonthreshold BPS bound state 7 consisting of p0 -branes and pbranes with p0 < p, the charge quantizations imply that 2 0 F +B is also quantized. It can
actually be determined in terms of the quantized (integral) charges q and n (see [13], for
3 We will refer to this limit as the NCYM decoupling limit as opposed to the usual decoupling limit for the
AdS/CFT correspondence of Maldacena.
4 This in general does not require the two gauge groups in the respective descriptions to be the same.
Rather we only need to preserve the gauge equivalence relation. In other words, whenever two ordinary gauge
fields are gauge-equivalent by an ordinary gauge transformation, we have two corresponding gauge-equivalent
noncommutative gauge fields which are related by a noncommutative gauge transformation. However, in this
paper, when we talk about gauge group, we always mean the unitary group U (n) associated with the number
of D-branes defined in terms of n n Hermitian matrices, not the one associated with the noncommutative,
associative algebra defined in terms of product for functions, i.e., f g = f g + (1/2)i ij i f j g + O( 2 ).
5 The constant B-field can be taken as the asymptotic value of the NS B-field in a gravity configuration. Note
that our NS 2-form B differs from that of Seiberg and Witten by a factor 2 0 , i.e., B = 2 0 B SW .
6 There actually exist [17,18] more complicated configurations of Dp-branes with nonvanishing NS B-field.
For example, the so-called ((F, D1), D3) nonthreshold bound state given in [17] has been used to discuss the
gravity dual of NCYM in [8].
7 In this paper, we always consider infinitely extended rather than compactified branes. One advantage in
choosing so is to avoid considerations of possible finite size effects as discussed in [5] and non-local light winding
modes as discussed in [3,7,19].
231
example 8 ), where the integer n is the number of p-branes and the integer q is the number
0
0
of parallel p0 -branes per (2)(pp ) 0(pp )/2 area over the (p p0 )-plane perpendicular
to the p0 -branes in the worldvolume of p-branes. We therefore expect that the noncommutativity is actually related to the integral charge q. As we will see, this is indeed true.
For concreteness, we will focus, in this paper, on the systems of nonthreshold bound
states (D(p 2), Dp), for 2 6 p 6 6, in type II string theories. The gravity configurations
of these bound states have been given in [13,17,20]. The asymptotic value of B-field always
vanishes for each of these gravity solutions. 9 As mentioned earlier, we choose to work in
the gauge where the constant flux of the worldvolume gauge field strength vanishes. By doing so, we end up with a nonvanishing asymptotic value for the B-field in each of the gravity solutions. The component of this asymptotic B-field, denoted as B , is actually given
as 10 B = tan = q/n, i.e., quantized in terms of the integer q and n described above. 11
In the NCYM decoupling limit, the large B is the key for the origin of NCYM. More
0 with
precisely, as shown in [79], in order to obtain NCYM we need B = tan = b/
0
0 and b = fixed. For fixed B , we end up only with the usual YM as demonstrated
in [17] and also pointed out in [8]. However, so far only the classical B = tan has
been used in discussing NCYM [69,33]. But as we have discussed above, B = tan is
in fact quantized according to B = tan = q/n. So the above NCYM decoupling limit
8 It can also be inferred from the contribution of 2 0 F + B to the tension of the corresponding nonthreshold
bound state as given in [1317].
9 This is also true for all known gravity solutions of Dp branes with nonvanishing B-field, see for examples
[13,1618,20].
10 In general, we could have B = a(q/n) with a an arbitrary constant. For the reasons given in footnote 7,
we always keep this dimensionless parameter a fixed even in the NCYM decoupling limit. As discussed in
detail in the following section, our philosophy of NCYM decoupling limit is to treat the coordinates in the two
codimensions of the D(p 2) and Dp branes in (D(p 2), Dp) similarly as the radial coordinate for the bound
state in the decoupling limit. The former decouples any vortex of q D(p 2) branes from the rest in the two
codimensions while the latter decouples the asymptotic region from the near-horizon region. This leads to q n
in the NCYM decoupling limit. However, if the two codimensions form a torus rather than an infinite plane
(therefore, we may have possible complications as mentioned in footnote 7), the dimensionless parameter a will
e2 0 ) with V
e2 the area for the scaled torus in the decoupling limit. Therefore, q/n V
e2 /b which
be b2 /(V
appears to be finite in the decoupling limit. This latter case will be discussed in [21]. Apart from the apparent
difference for q/n in the decoupling limit, the rest conclusions drawn in the two cases are essentially the same.
We are grateful to R.-G. Cai and N. Ohto for their question which leads us to the above discussion. Actually, the
If we denote the
e2 for the torus is not quite independent of the noncommutative parameter b.
above scaled area V
two coordinates as xp1 , xp (we discuss them in the following section), we have the noncommutative relation
So we should have V
e2 > b.
Therefore, we have q > n. This implies that whenever we have
as [xp1 , xp ] = ib.
large n, we must have large q but not the other way around.
11 The integer q resembles, to some extent to the integer n. As discussed in [13,16], if one calculates the
Noether charge associated with the infinitely many D(p 2) branes in (D(p 2), Dp), the charge itself becomes
infinity. But the charge over the 2-dimensional area (2 )2 0 perpendicular to the D(p 2) branes in the Dp
p2
p2
with Q0
= (2 )(112p)/2 0(5p)/2 where
brane worldvolume is finite and is given as (q/(2 )2 0 )Q0
the conventions for the charge is given in [22]. But Q0 /((2 )2 0 ) is just the charge units for the (p + 2)form charge associated with the Dp branes. So this charge density looks like q units p
of Dp brane charge.
p
p
Further, the tension for (D(p 2), Dp) in string metric is T (D(p 2), Dp) = (T0 /gs ) q 2 + n2 with T0 =
p
0(p+1)/2
1/[(2 )
] the Dp-brane tension units and gs the asymptotic string coupling. In the above tension
formula, since the integers q and n appear symmetrically, so, in some sense, the D(p 2) branes in (D(p 2),
Dp) appears as if they are q Dp-branes.
p2
232
0 in the low energy limit 0 0. In other words, we have q n

implies q/n = b/
in the NCYM decoupling limit. 12 One of our observations in this paper is that: in the
NCYM decoupling limit, the infinitely many D(p 2) branes in the (D(p 2), Dp) system
plays dominating role over the n Dp branes in the system. The noncommutativity actually
arises due to these D(p 2) branes and the (p + 1)-dimensional NCYM is just one way of
describing these D(p 2) branes.
Some closely related work and ideas have been presented some time ago by a number of
authors. Townsend first suggested an equivalence between a D2-brane and a condensate of
D0-branes [23]. His proposal was based on the following observation: A condensate of D0branes can be described effectively by one-dimensional super YM theory with gauge group
U (), i.e., U (n) supersymmetric gauge quantum mechanics (SGQM) with n . But
this is just another description of D2-brane which can be inferred from the description of
D = 11 closed supermembrane by a U (n) SGQM for n established in [24] in the old
days of D = 11 supermembrane theory. This is also one of the motivations for the original
Matrix theory proposal of M-theory [25]. Generalizing this to other D-branes, Ishibashi
argued that a D-brane in a constant B-field background can be described equivalently as a
collection of infinitely many lower dimensional D-branes [26]. However, the validity of the
above equivalence is not discussed in [26]. It appears that this has been remedied in [28].
The authors in the latter paper argued that a D2-brane with a constant worldvolume gauge
field strength in the SenSeiberg limit [29,30] is described by the Matrix theory quantum
mechanics action of D0-branes.
Given the requirement of constant B-field background for the above equivalence, it is
natural for these authors to seek the connection of their work to the NCYM description of
D-branes with this background. They find whether one has an ordinary or noncommutative
description depends on the gauge choice of the D brane worldvolume diffeomorphism
[27,28]. If one takes the static gauge, i.e., the coordinates parallel to the brane are fixed
and the worldvolume gauge field remains dynamical, one ends up with an ordinary Yang
Mills description. On the other hand, if one chooses the so-called constant field strength
gauge, i.e., no fluctuations of the ordinary gauge field are allowed, then one ends up
with NCYM description. In this latter gauge, the dynamical degrees of freedom, which
are usually described by the ordinary worldvolume gauge field, are now carried by the
scalar fields corresponding to the parallel coordinates of the brane. The noncommutative
gauge field appears as the fluctuations of these scalar fields with respect to the static BPS
configuration [27,31]. Especially in [31], the author pointed out that the decoupling limit
for NCYM corresponds to the SenSeiberg limit. The two descriptions can be mapped to
each other through the worldvolume diffeomorphism [27,31,32].
One of the goals in the present paper is to show that in the gravity dual description,
the (p + 1)-dimensional U (n) NCYM is equivalent to an ordinary (p 1)-dimensional
0 and with b as the noncommutative
YM with gauge group U (q) with q/n = b/
parameter. In other words, n Dp branes with B = q/n is equivalent to infinitely many
12 See the explanation given in footnote 10.
233
D(p 2) branes without B-field in the NCYM decoupling limit. Therefore, this work lends
support to what is just described above.
0 more closely. We know that the fixed
Let us now examine the relation q/n = b/
parameter b is directly related to the noncommutative geometry. The larger (smaller)
the b is, the more (less) noncommutative the space geometry will be. The presence of
this dimensionful parameter implies that a new scale is introduced into the theory under
consideration. We will spell out the meaning of this new scale in the following section. To
make sense of the system (D(p 2), Dp), we need n > 1. Large b requires large q and
0 . So for NCYM we actually need large q and small
small n from the relation q/n = b/
n rather than large n. In other words, large n really goes against the NCYM effect even
though it is good for the gravity dual description in the usual AdS/CFT correspondence. It
But from
seems as if there were an exception, i.e., we could take n large with a fixed b.
0
q/n = b/ , we see that we can increase n only at the price of increasing q for a fixed b.
0
For example, for a fixed b, when n = 1, we denote q = q0 which gives q0 = b/ . Now

we want to increase n = 1 to a large n while keeping b unchanged. Then we must have
q = nq0 .
Because the new parameter q or b accounts for the noncommutative effect in NCYM
and
and if we really want to focus on this effect, one should take large q (or large b)
small n even in the gravity dual description of NCYM which is contrary to what have been
claimed in the literature. Interpreted in another way, the real meaning for large n used in
the literature is the large q.
So in order to isolate the noncommutative effect and to find the reason behind this effect,
we should limit ourselves to small n. Without the loss of generality, we can simply set n = 1
from now on. 13 As we will demonstrate in the following sections using the gravity duals
of NCYM, whenever the gravity dual description is valid, the (D(p 2), Dp) system is
reduced to a system of infinitely many D(p 2) branes with two additional isometries and
without B-field.
This paper is organized as follows. We will present the general features valid for all
(D(p 2), Dp) systems with 2 6 p 6 6 in the following section. In Section 3, we will
provide detail evidence in supporting our above claims for the well-studied system (D1,
D3). In Section 4, we will discuss similar evidence for each of the remaining systems. We
conclude this paper in Section 5.
2. General properties
To begin with let us explain the new scale b further. In terms of the (p + 1)-dimensional
NCYM description, this new scale defines the noncommutativity. In the corresponding
ordinary (p + 1)-dimensional YangMills description, this new scale appears as a
dimensionful coupling constant in addition to the usual gauge coupling, therefore giving
rise to higher dimensional operators. All these are due to the constant background B-field.
13 When we take n = 1, we have the gauge group U (1) for NCYM. For a general n, we have the gauge
group U (n).
234
We know that the relevant energy scale is u (i.e., the energy carried by an open string
stretched between a probe and the bound state which will be defined as r = 0 u with r the
radial coordinate shortly). We also know that the physical effect of noncommutativity is not
solely determined by b but by some combination of b and u, i.e., b 1/2u. For example,
if b 1/2u 1 the noncommutative effect is still negligible as pointed out
even for large b,
implies its large
in [8] (one may naively think that large noncommutativity, i.e., large b,
physical effect). It is hard to understand this if one sticks with the either noncommutative
or ordinary YangMills description of Dp branes.
As we have discussed in the previous section and will demonstrate in this paper, in the
NCYM decoupling limit, the (D(p 2), Dp) system is reduced to that of infinitely many
D(p 2) branes with two additional isometries and without B-field. It is much easier
to understand the above in the ordinary YangMills description of these infinitely many
D(p 2) branes. Unlike the Dp branes, there are two kinds of transverse spaces to the
D(p 2) branes. One is the common transverse space shared by both of the D(p 2) and
Dp branes in (D(p 2), Dp). The corresponding field theories share the common energy
scale u. Now the D(p 2) branes have another transverse space with codimension two
with respect to the Dp brane worldvolume. Since this space is treated differently from the
previous one, there must be another energy scale for the D(p 2) branes. In other words,
the field theory describing the low energy dynamics of the infinitely many D(p 2) branes
needs two energy scales one of which is u. What is the second energy scale?
We can estimate this scale in two different ways. For convenience, let, say, the D(p 2)
branes lie along x 0 , x 1 , . . . , x p2 directions and the Dp brane along x 0 , x 1 , . . . , x p2 ,
x p1 , x p directions. As discussed in the previous section, we have q D(p 2) branes
per (2)2 0 area over the infinitely extended x p1 x p -plane. We can also view each set
of q D(p 2) branes through the x p1 x p -plane as a vortex on the plane. From the
worldvolume viewpoint, these vortices are magnetic ones. In some sense, they resemble
the vortices in type II superconductivity. Each of these vortices has an area on the
order of
0 and the distance between the centers of two nearby vortices is on the order 0 . Now
we can estimate the characteristic separation d between two such D(p 2) branes in a
given vortex as

(2)2 0 1/2
.
(2.1)
d=
q
0 obtained in the previous section in the above, we have d
If we substitute q = b/
0
1/2
/b . The energy carried by an open string stretched between two such D(p 2) branes
in a given vortex is given as d/ 0 1/b 1/2 which should define the second energy scale.
As we will discuss latter in this section, the NCYM decoupling limit also consists of
x(p1),p with u and x(p1),p fixed. The limit 0
0 0, r = 0 u, x(p1),p = ( 0 /b)
0 means a low energy field theory limit. r = 0 u means that in the gravity description,
only the near-horizon region is relevant and it also defines the first energy scale r/ 0 = u
x p (the same for xp1 ) defines the
mentioned above. In analogy to r = 0 u, xp = ( 0 /b)
distance between a dynamical probe and the vortex of q D(p 2) branes located at xp =
xp1 = 0. Hence the xp is the length of an open string stretched between the probe and the
235
vortex. Therefore, the second energy scale should also be given as xp / 0 = xp /b which is
finite. Note, however, that
stretched between two nearby vortices carries an
the open string
energy on theorder of 0 / 0 = 1/ 0 as 0 0 since the open string length is
on the order 0 . In other words, the nearby vortices have no dynamical influence on the
vortex under consideration in the NCYM decoupling limit. So we need to consider only
one vortex and the rest vortices decouple from this one.
It appears in the above that we obtain two different estimations for the second energy
scale. Let us resolve this apparent difference. The second approach gives this scale as
The xp should be viewed as the quantum-mechanical average of the corresponding
xp /b.
operator. Because of the noncommutative relation [x p1 , x p ] = ib for the corresponding
operators, we should have the value for both of x p1 and x p on the order of b 1/2 . Therefore,
we have the energy scale on the order 1/b 1/2 which agrees with what we have obtained
from the first approach. So we explain why we have a second energy scale and how this
energy scale is related to the parameter of noncommutativity b.

Because of the presence of two energy scales, i.e., u and 1/b 1/2 , we have opportunities
to examine the low energy dynamics for these D(p 2) branes. As is understood, the
lower scale is relevant for the low energy dynamics. If 1/b 1/2 u, i.e., b 1/2 u 1, the
dynamical degrees of freedom associated with the scale 1/b 1/2 are dominantly important
while those associated with the scale u are not. Roughly speaking, the ordinary (p 1)dimensional YangMills with gauge group U (q) (with q ) explores only p + 1
spacetime dimensions dynamically. Since the scale 1/b 1/2 is associated with open strings
stretched between the D(p 2) branes rather than Dp branes, the natural description of
the dynamics is in terms of these D(p 2) branes even though it may not be the best.
However, if we force ourselves to describe the dynamics in terms of Dp branes instead,
we have to do something unnatural. This unnatural twist is to make the two codimensions
noncommutative and we expect that the effect of noncommutativity is important. This is
indeed true. As we will demonstrate in the following sections. The validity of the gravity
dual description of NCYM requires b 1/2u 1. Further the gravity configuration of Dp
brane with a constant B-field is reduced to that of D(p 2) branes with two isometries
and without B-field. The importance of the noncommutative effect can also be seen from
the noncommutative product for functions f g = fg + (1/2) i ij i f j g + . If we
choose to describe the underlying dynamics in terms of Dp branes, the apparent energy
scale is u. Therefore, f uf . Since ij b for the present case, we see that the leading
2 fg. In
noncommutative modification to the usual function product is of the order bu
1/2
the present case, i.e., for b u 1, the noncommutative effect is indeed important as
anticipated.
On the other hand, if 1/b 1/2 u, i.e., b 1/2u 1, then only the dynamical degrees of
freedom associated with the energy scale u are important. In this case, the appearance
of the infinite D(p 2) branes only collectively modifies the (n = 1) Dp brane without
B-field through the correction of order b 1/2 u 1. In other words, we do not expect
the corresponding noncommutative effect to be important if we choose to describe the
underlying dynamics in terms of the Dp brane instead. This can be easily seen from the
above product of functions. The gravity description of the NCYM is not valid in this
236
region. However, if we are willing to extrapolate to this region as done in [8], the resulting
gravity description is nothing but the gravity dual of the ordinary YangMills for Dp branes
without B-field. The above can also be interpreted as the usual large n limit. We first set
q = b 0 / 0 with a finite b 0 . Then q/n = (b 0 /n)/ 0 . We take b = b 0 /n. So in the large n limit,
the U (n) NCYM is essentially the same as the corresponding ordinary U (n) YM. In other
words, with fixed q behavior (i.e., fixed b 0 ), in the large n limit, the U (n) NCYM becomes
an ordinary U(n) YM.
Let us provide further evidence in supporting our above explanation. In the presence of
constant B SW -field, Seiberg and Witten have shown in [6] that the metric seen by the open
strings on the Dp-branes is the effective open string metric GSW
MN rather than the closed
SW
(M, N = 0, 1, . . . , 9). In the NCYM decoupling limit, i.e., with GSW
string one gMN
MN and
SW
SW
SW
0
0
fixed and 0, gMN should vanish according to gMN along the directions in
B
which B SW are nonzero. They also give the relation between the open string coupling Gs
and the closed string coupling gs as

det(g SW + 2 0 B SW ) 1/2
,
(2.2)
Gs = gs
det g SW
where the metric g SW and B SW take values only along the Dp-brane directions. The low
energy YangMills coupling is given in terms of the open string coupling Gs as
2
= (2)p2 0(p3)/2Gs .
gYM
(2.3)
As pointed out in [6], the Gs and gs must scale in the following ways
Gs
= fixed,
( 0 )(3p)/2
gs
= fixed,
0(3p+r)/2
(2.4)
2 can be kept finite for a quantum theory in the decoupling limit. We note that
such that gYM
the scaling of the closed string coupling depends on the rank r of the B-field.
For our (D(p 2), Dp) systems, we always have r = 2. So gs = 0(5p)/2 g =
0[3(p2)]/2g with g fixed in the decoupling limit. If we look from the gravity dual
viewpoint of the NCYM, only the closed string coupling gs is relevant. Then the above
scaling behavior of gs is actually the same as that for simple D(p 2) branes (i.e., without
B-field) rather than simple Dp branes in the decoupling limit for the usual correspondence
discussed in [34]. For example, for p = 5, i.e., for (D3, D5) system, the gs itself is fixed
in the NCYM decoupling limit which is the same as that for the simple D3 rather than the
simple D5 branes in the usual decoupling limit. This is entirely consistent with our above
explanation. Since the gravity description is classical, we do not sense noncommutativity
of the spatial directions along which constant B-field does not vanish. Then our above
discussion implies that this description must be that of infinitely many (i.e., q) D(p 2)
branes without B-field. This is indeed true as we will demonstrate. This explains the
behavior of the closed string coupling in the decoupling limit.
However, if we look from the open string viewpoint, the open string coupling Gs is
which will be derived in the following.
relevant. From Eq. (2.4), we have Gs = 0(3p)/2g b,
Comparing this scaling behavior with what has been given in [34] for simple Dp branes in
237
the decoupling limit for the usual YM, one can see that Gs scales like that for simple Dp
branes rather than for simple D(p 2) branes as anticipated.
Now let us study the gravity configurations for (D(p 2), Dp) systems and for D(p 2)
branes with two additional isometries in the NCYM decoupling limit, respectively.
The string-frame metric, the dilaton and the B-field for (D(p 2), Dp) with 2 6 p 6 6,
first given in [13], can be expressed in a unified way as

2
2
+ dxp2 + dy i dy i ,
+ H 01 dxp1
ds 2 = H 1/2 H 1 dx02 + dx12 + + dxp2
e2 = gs2
B=
H (5p)/2
,
H0
q 01 (p1)
H dx
dx (p),
n
(2.5)
where the string coupling gs = e0 with 0 the asymptotic value of the dilaton, i = p +
1, . . . , 9, q and n are two integers defined earlier and H, H 0 are two harmonic functions
H =1+
where
r=
Qp
,
r 7p
Qp
n2
,
2
2
n + q r 7p
H0 = 1 +
(2.6)
yi yi ,
q
Qp = gs cp n2 + q 2 0(7p)/2 ,

cp = 25p (5p)/2 (7 p)/2 .
In the NCYM decoupling limit (see [8], for example), we have,

0 0,
gs = 0(5p)/2 g,
b
q
= 0.
n
(2.7)
Also the following redefined 14 x s rather than the original x s are kept fixed
x0,1,...,(p2) = x 0,1,...,(p2) ,
x(p1),p =
0
x(p1),p .
b
(2.8)
One can check easily from Eq. (2.5) that the asymptotic values for the metric, the dilaton
and the B-field with respect to the fixed xs
correspond to the closed string moduli of
Seiberg and Witten [6] 15 , respectively, if the relation between our B and the Seiberg and
Wittens B SW , i.e, B = 2 0 B SW , is used. Using Eq. (2.2), we can have Gs = 0(3p)/2 g b
as given earlier.
In the NCYM decoupling limit, any fixed r 6= 0 region will decouple from the region
defined by r = 0 u with u fixed which provides the gravity dual description of the (D(p
2), Dp) system. The vortex of q D(p 2) branes located at xp1 = xp = 0 will decouple
14 It appears that we have introduced additional scales for x
p1 , x p . As discussed earlier, these two fixed
the value of xp (or
quantities are actually related to b because the noncommutative relation [xp1 , xp ] = ib.
xp1 ) should correspond to the quantum-mechanical average of the corresponding operator. It should be on the
order of b1/2 .
15 This is a natural choice. But one could do differently as in [33] where an effective string tension is introduced
and the 0 is set to 1.
238
from the rest of identical vortices. For convenience, let us write the NCYM decoupling
limit collectively here:
0 0,
b
q
= 0,
n
gs = 0(5p)/2 g,
x0,1,...,(p2) = x0,1,...,(p2) ,
x(p1),p =
0
x(p1),p ,
b
r = 0 u,
(2.9)
g,
where b,
u, x remain fixed. For the reason explained earlier, we always take n = 1 in
the following discussion.
With this limit, the NCYM gauge coupling is given as
2
= (2)p2 b g,
gYM
(2.10)
2
which is fixed as expected. The NCYM effective gauge coupling is geff
2 up3 . In this
gYM
paper, we will focus solely on the gravity dual description of NCYM, which is expected to
be in the region where the perturbative calculations in super YangMills cannot be trusted,
i.e.,

1/(3p),
u (g b)
p < 3,
2
1
(2.11)
geff
1/(p3), p > 3.
u (1/(g b)
So we will investigate the
As discussed earlier, the relevant parameter for NCYM is b.
validity of the gravity dual description with respect to this parameter while keeping g
fixed.
With the limit Eq. (2.9), the gravity description for (D(p 2), Dp) is

2
0
2
2
+ h dx(p1)
+ dxp2
ds = f 1 (u)u2 dx02 + dx12 + + dx(p2)

+ f (u)
du2
2
+ d8p
u2

,
(au)(7p)(p3)/2
,
1 + (au)7p
0 (au)7p
dx p1 dx p ,
(2.12)
B=
b 1 + (au)7p
1/(7p) with constant cp defined earlier, and
where g = g b (5p)/2, a = b 1/2/(cp g)
functions f (u) and h are

e2 = g 2
f (u) =
2/(7p)
(cp g)
,
(au)(3p)/2
h =
1
.
1 + (au)7p
(2.13)
Note that the NCYM effective string coupling, e , is finite in the decoupling limit (since
b, g and u remain fixed). The curvature in string units is
s
s
3p
1
u
(au)3p
.
(2.14)
0 R
geff
g 4/(7p)
g b
From the NC field theory perspective, u is an energy scale. The limit u means
going to UV in the field theory. In this limit, unlike the cases for simple Dp branes without
239
B-field analysed in [34], the effective string coupling, e (au)(7p)(p5)/2, vanishes

for p < 5 while the NCYM theory becomes UV free still for p < 3 (note that we have
2 6 p 6 6 in this paper). For p > 3, the field theory breaks down and we need new degrees
of freedom there. However, from the gravity side for 5 > p > 3, both the effective string
coupling and the curvature (see the second equation in (2.12) and Eq. (2.14)) vanishes and
therefore the gravity description is perfectly good. For p = 5, the curvature still vanishes
Finally for p = 6, the string
but the effective string coupling remains finite as e = g.
effective coupling blows up and the dual description is needed. As pointed out in [9], even
for this case, the worldvolume theory with B-field differs from that without B-field in
that the former decouples from gravity while the latter does not [9,35] in their respective
decoupling limits. The reason for this is now clear that the D6 branes with constant B-field
are equivalent to a system of infinitely many (i.e., q) D4 branes without B-field in the
decoupling limit. The decoupling of gravity in the latter system must imply the same in the
former one.
Let us now consider the gravity solutions of D(p 2) branes with two additional
isometries for 2 6 p 6 6. They can be obtained from (D(p 2), Dp) simply by setting
the charge associated with Dp branes to zero. In doing so, we need to use the original
B-field (rather than the one given above where a gauge choice is made for NCYM) and we
find that as n = 0, H 0 = 1 and so B = 0. If we take the same limit as given in Eq. (2.9)
0 we end up with 16
except for q/n which is now replaced by q = b/

1
2
2
2
ds 2 = 0 f 1 (u)u2 dx02 + dx12 + + dx(p2)
+
+
d
x
d
x
p
(p1)
(au)7p

2
du
2
+ d8p
,
+ f (u)
u2
e2 = g 2 (au)(7p)(p5)/2,
B = 0,
(2.15)
where a, g and f (u) are the same as given before. As discussed before, the above
configuration should be explained as the one describing q D(p 2) branes located
at xp1 = xp = 0.
Comparing Eq. (2.15) for q D(p 2) branes with Eq. (2.12) for (D(p 2), Dp), we
see that the two configurations become the same, except for the B-field, if au 1. The
e(p1)p = 0 /b.
(Note that the
e(p1)p = 0 while the latter has a constant B
former has B
e(p1)p is defined with respect to the fixed x coordinates though the corresponding 2-form
B
is inert under the rescalings of coordinates.) For the gravity configuration of q D(p 2)
branes, any nonvanishing constant B-field along directions transverse to the D(p 2)
branes can always be gauged away [6]. So we conclude that the two configurations are
identical if au 1. Further, as we will demonstrate, au 1 implies b 1/2u 1, i.e., in the
region of strong noncommutative effects. Therefore the crucial point for us to establish the
equivalence between the q D(p 2) branes and Dp branes is to show whenever the gravity
16 If we do not take n = 1 in the discussion of (D(p 2), Dp) system above, we simply choose the b here as n
times b there.
240
description is valid, we always end up with au 1. As we will show in the following

sections, this is indeed true.
Note also that the (D(p 2), Dp) system preserves the same number (sixteen) of
supersymmetries as D(p 2) branes with two additional isometries. This is necessary
for the equivalence.
3. The (D1, D3) system

The decoupling limit for this system with classical B = tan has been studied in [8].
Here we revisit this system with quantized B = tan = q/n. As explained in the previous
sections, we shall take n = 1. Taking p = 3 in the NCYM decoupling limit Eq. (2.9), we
have
0 0,
0,
q = b/
x0,1 = x0,1 ,
x2,3 =
gs = 0 g,
0
x2,3 ,
b
r = 0 u,
(3.1)
g,
where b,
u, x remain fixed. In this limit, the asymptotic region decouples from the
near-horizon region which describes the (D1, D3) system.
In particular, q n implies that D-strings in the system plays the dominating role over
the n D3 branes. Since the noncommutative geometry in x 2 x 3 directions is controlled
by the parameter b which is in turn controlled by the large integer q, therefore the
noncommutativity is due to the infinitely many D-strings in the system. Notice that the
the closed string coupling at the IR 17 while the
open string coupling Gs = g with g = g b,
0
This tells us that from the noncommutative open string
closed string coupling gs = g.
perspective (or the noncommutative field theory description) the system looks like D3
branes while from the closed string side (or the gravity description) it behaves like Dstrings if we compare the scalings of these string couplings with those for simple D-branes
(i.e., without B-field) given in [34] in their respective decoupling limits.
In this section, using the gravity dual description of NCYM, we will provide evidence
for the claim that the (D1, D3) system is reduced to a system of q D-strings without B-field.
The gravity description for (D1, D3) has the form
2

du
2
0
1 2
2
2
2
2
2
+ d5 ,
ds = f u dx0 + dx1 + h dx2 + dx3 + f
u2
e2 = g 2 h,
0
4 4
e23 = a u ,
B
b 1 + a 4 u4
h =
1
,
1 + a 4 u4
A01 = 0
a2 =
b
,
f
b u4
,
g 4 g
f 2 = 4 g.

e0123u = 02 h u u4 ,
H
2
gf
(3.2)
17 The open string coupling or the gauge coupling is also related to the closed string coupling at IR for the usual
AdS5 /CFT 4 correspondence with fixed q and n as pointed out in [17].
241
This has been obtained by setting p = 3 in Eq. (2.12) and we have also included in
the above the RR 2-form A associated with the D-strings and the self-dual RR 5-form
e0123u are now defined with respect to the
e23 , A01 and H
field strength H . Note also that B
new coordinates through the corresponding forms which are inert under the rescalings of
coordinates.
The curvature in string units is given by Eq. (2.14) and can also be read off from the
metric in (3.2) as
0 R
1
1
p ,
geff
g
(3.3)
2 = 2 g for p = 3. The gravity description is good if the effective string coupling,

where geff
e , and the above curvature in string units are small, i.e.,
g
1 + (au)4
1,
2
geff
g 1.
(3.4)
As usual, the validity of gravity description requires large effective gauge coupling geff
which implies that the field theory description breaks down. The above equation says
au g 1/2 1.
(3.5)
As explained in the previous section, we always keep g fixed. So large g = g b means large
b which is consistent with the picture that the noncommutative effect is important for large
au in the gravity description.
large au does not
Notice that since a b 1/4/g 1/4 is large with large b and fixed g,
necessarily mean large u. Therefore the condition (3.5) for u does not mean that we
approach the boundary which is at u = . Further we have b 1/2u au 1 which implies
that the dynamical degrees of freedom are due to the D-strings as explained before. Let us
examine the configuration (3.2) under Eq. (3.5) with fixed g.
In this case, h 1/(au)4
0
e23 approaches its constant but small value /b for large b even if it is measured
and B
e0123u/ 02 1/[(g 5/4 b 11/4)au] approaches zero
in terms of 0 . The 5-form field strength H
4 /g 2 1
4 /g 2 = g(au)
in the limit. However, via Eq. (3.5), the RR 2-form A01/ 0 bu

survives. So everything goes over to that of q D-strings as given in Eq. (2.15) (with p = 3)
in the same limit Eq. (3.1). In other words, when the gravity dual description of NCYM
for D3 branes with large asymptotic B-field is valid, this gravity system is reduced to q
D-strings without B-field.
In the region g 1/2 au 1, the effective string coupling is large. So, we need to go
to the S-dual gravity description. In this case, (F, D3) is reduced to q F-strings without
RR 2-form field since under the S-duality, D-strings become F-strings ((D1, D3) becomes
(F, D3)). Under S-duality the NS 2-form B becomes a RR 2-form while the RR 2-form A
becomes a NS 2-form and this is the reason for the appearance of F-strings.
Under S-duality, we have
ls2 l 0 s gs ls2 ,
2
e e e ,
gs gs0
1
,
gs
ds 2 ds 0 gs e ds 2 ,
2
(3.6)
242
where ls = 0 is the string length scale in the original variables. So, the new metric and
the dilaton are given as,
0

02
0 1/2 H
2
2
2
2
dx0 + dx1 + dx2 + dx3 + (H 0 )1/2 dy i dy i ,
ds = (H )
H
H0
0
(3.7)
e2 = (gs0 )2 ,
H
where the harmonic functions continue to be given by Eq. (2.6) but gs and 0 in the charge
Q3 are replaced by their S-dual values through the relations in Eq. (3.6). One can check
that the metric and the dilaton are indeed those describing the (F, D3) system as given in
[16].
Following [9], the decoupling limit for this system is the same as those given in Eq.
(3.1). The only difference is that we express them in terms of ls0 and gs0 through gs = 1/gs0
and 0 = gs0 (ls0 )2 0. In the decoupling limit, we have the following for (F, D3)
(
0 2

02
0 2 1/2 (g u)
q
dx02 + dx12 + h dx22 + dx32
ds = (ls ) h
4 g0
2
)
p
du
+ d52 ,
+ 4 g 0
u2

0
(3.8)
e2 = (g 0 )2 1 + (au)4 ,
where h is given in Eq. (2.6) with the same definition for the parameter a there and g0 =
The curvature measured in ls0 is given as,
1/g = 1/(g b).
(ls0 )2 R p
1
g 0 (1 + a 4 u4 )
(3.9)
0
The gravity description is good when the effective string coupling, e , and the curvature
are small, i.e.,
1
(g 0 )1/4
au
1
(g 0 )1/2
(3.10)
Eq. (3.10) says 1 g 1/4 au g 1/2 . For

which is true if g 0 = 1/g 1. In the original g,
fixed g,
g 1 implies large b and au 1 implies h 1/(au)4 . We have again b 1/2u
au 1. Then the gravity description for (F, D3) goes over to that for q F-strings in the
similar limit as expected.
4. The remaining (D(p 2), Dp) systems

In this section we discuss the equivalence for the rest of the (D(p 2), Dp) systems.
(D0, D2):
The decoupling limit for this system is
0 , gs = ( 0 )3/2g,
q = b/
0
x0 = x0 , x1,2 = x 1,2, r = 0 u,
b
243
0 0,
(4.1)
g,
where b,
u, x remain fixed. Under this limit, the gravity description for (D0, D2) is

2

du
2
+
d
,
ds 2 = 0 f 1 (u)u2 dx02 + h dx12 + dx22 + f (u)
6
u2
(au)5/2
,
1 + (au)5
0 (au)5
dx 1 dx 2 ,
B=
b 1 + (au)5
e2 = g 2
2/5 ,
f (u) = (au)1/2(c2 g)
h =
1
,
1 + (au)5
1/5 . The curvature in string units is

where g = g b 3/2 , a = b 1/2 /(6 2 g)
s
r
1
u
au
0
.
R
4/5
geff
g
g b
(4.2)
(4.3)
The gravity description is valid if the effective string coupling, e , and the above curvature
in string units are small, i.e.,
g 4/15 au g 4/5 ,
(4.4)
which can hold only if g 1. For fixed g,

this implies large b and therefore the large
noncommutativity. We also have b 1/2 u au 1 which implies that the dynamical
degrees of freedom are due to D0 branes and the noncommutative effect is important.
The au 1 implies h 1/(au)5 . Since a b 1/5/g 1/5 , this is also large. So large au does
not necessarily mean large u. With the above, we have the metric and the dilaton reduced
to those of q D0-branes in the same limit as given in Eq. (2.15) for p = 2. Note that
B12 / 0 = 1/b is now a small constant for large b.

4/15
, the effective string coupling, e , becomes large. So, we need
In the region au g
to lift this system to eleven dimensions. It can still be described by the 11D supergravity if
the curvature in 11D Plank units remains small. We have the 11D curvature in 11D Plank
units as
lp2 R e2/3
g 4/15
1
1
geff (au)1/3(1 + a 5 u5 )1/3
(4.5)
which gives au g 2/15 . So the 11D gravity description for the system is valid if g 2/15
au g 4/15 . For this to be true, we must have g 1, therefore b becomes large for fixed
g.
Precisely with these, the gravity description for the lifted (D0, D2) branes is the same as
that for the lifted D0 branes.
(D2, D4):
The decoupling limit for this system is
244
0 0,
0,
q = b/
x0,1,2 = x0,1,2,
gs = ( 0 )1/2g,
0
x3,4 = x3,4, r = 0 u,
b
(4.6)
g,
where b,
u, x remain fixed. Under this limit, the gravity description for (D2, D4) is

2
0
ds = f 1 (u)u2 dx02 + dx12 + dx22 + h dx32 + dx42

+ f (u)
du2
+ d42
u2

,
(au)3/2
,
1 + (au)3
0 (au)3
dx 3 dx 4 ,
B=
b 1 + (au)3
e2 = g 2
f (u) = ( g)
2/3 (au)1/2,
h =
1
,
1 + (au)3
(4.7)
where
1/2,
g = (g b)
a=
b 1/2
.
( g)
1/3
The curvature in string units is

0 R
1
1
1
q
.
1/2
geff
(au) g 2/3
g bu
(4.8)
Unlike the case for D4 branes without B-field, the closed string coupling gs = ( 0 )1/2 g
vanishes (rather than blows up) in the decoupling limit. It is still proper to consider the
10D gravity description for this system. In order to have a valid gravity description, the
effective string coupling, e , and the above curvature in string units need to remain small,
i.e.,
g 2
(au)3/2
1,
1 + (au)3
au 1/g 4/3 .
(4.9)
We have two cases here depending on whether g > 1 or g < 1. For g < 1, the first equation
in (4.9) can be satisfied if au 1 or au g 4/3 . However, au 1 is imcompatible with
the second equation in (4.9) if g < 1. So we have au 1/g 4/3 > 1 for which the gravity
description is valid. In this case, since g < 1, the b is not large and a is not large, either.
Therefore au 1 does require large u. So, when the gravity description is valid, we are
approaching the boundary. Similar analysis tells us that au g 4/3 > 1 is the condition for
a valid gravity description if g > 1. Again, we need to approach the boundary, i.e., large
u to validate the gravity description. The crucial thing here is that au 1 and b 1/2u 1
hold true whenever the gravity description is valid. Now we have h 1/(au)3 . The metric
and the dilaton go over to those of q D2 branes in the same limit. But there is a difference
here from the previous cases considered, i.e., the B-field no longer remains small the way
245
e34 / 0 = 1/b.
So long as the gravity
it happens before. It actually becomes a constant, i.e., B
e23 makes no difference from zero
description for the D2 branes is concerned, a constant B
e23 . Therefore, we expect that the (D2, D4) system is reduced to q D2 branes.
B
For g < 1, the effective string coupling can never be large. However, for g > 1, the
effective string coupling, e , is large in the region au g 4/3 . In that case, we need to lift
(D2, D4) to eleven dimensions. The 11D gravity description is valid if the curvature in 11D
Planck units is small, i.e.,
lp2 R e2/3
1
1
1,
geff (1 + a 3 u3 )1/3
(4.10)
which requires au 1. So the lifted (D2, D4) can be described by 11D gravity if 1
therefore large
it means large b,
au g 4/3 . This can be true only if g 1. For fixed g,
1/3
1/3
a b /g . So, we do not necessarily have large u here. For this description, the lifted
(D2, D4) goes over to the lifted D2 branes.
(D3, D5):
The corresponding decoupling limit is
0 0,
0,
q = b/
x0,1,2,3 = x0,1,2,3,
gs = g,
0
x4,5 = x4,5,
b
r = 0 u,
(4.11)
g,
where b,
u, x remain fixed. With this limit, the gravity description of (D3, D5) is

ds 2 = 0 f 1 (u)u2 dx02 + dx12 + dx22 + dx32 + h dx42 + dx52

+ f (u)
du2
+ d32
u2

,
(au)2
,
1 + (au)2
(au)2
dx 4 dx 5 ,
1 + (au)2
1
,
h =
1 + (au)2
e2 = g 2
B=
0
b
f (u) = g(au),
(4.12)
1/2 . The curvature in string units is

where g = g,
a = b 1/2/(g)
0 R
1
1
1
.
q
geff
(au)
g
2
g bu
(4.13)
Again unlike the case for D5 branes without B-field, the string coupling gs = g is fixed
(rather than blows up). It resembles that for D3 branes without B-field in the decoupling
limit. The gravity description is good if both the effective string coupling, e , and the above
curvature in string units remain small. This requires
au
1
,
g
g < 1.
(4.14)
246
From above we have au 1 and b 1/2u 1. Then both the dilaton and the B-field become
constant. We also have h 1/(au)2 now. We do not have any condition for b which implies
that u itself must be large, i.e., we need to approach the boundary. The gravity description
goes over to that for q D3 branes in the same decoupling limit.
When g 1, we need to go to its S-dual description. The B-field becomes RR 2-form
field. The D5 branes become NS 5-branes. So, (D3, D5) goes over to (D3, NS5). Using Eq.
(3.6), we have
0 2

(g u)
dx02 + dx12 + dx22 + dx32 + h dx42 + dx52
ds 02 = (ls0 )2 h 1/2
au
2

du
2
+
d
+ au
,
3
u2

0
(4.15)
e2 = (g 0 )2 1 + a 2 u2 ,
h continues to be the one given in Eq. (4.12) and the same is for the
where g 0 = 1/g,
parameter a. The curvature in dual string units is
(ls0 )2 R =
1
.
au(1 + a 2 u2 )1/2
(4.16)
The dual gravity description is valid when we have 1 au 1/g 0 = g which ensures that
both the effective string coupling and the curvature in the dual string units remain small.
With this, h 1/(au)2 . Therefore, the (D3, NS5) goes over to the S-dual D3 branes. Note
also that b does not play any role here and we expect that we need to approach the boundary.
This is indeed true from au 1 b 1/2u g 1/2 1.
(D4, D6):
This is our last system. The decoupling limit is
0 0,
0,
q = b/
x0,1,...,4 = x0,1,...,4 ,
gs = 01/2 g,
x5,6 = x5,6, r = 0 u,
b
(4.17)
g,
where b,
u, x remain fixed. Note that this system differs from the previous ones in that
the string coupling gs blows up in the decoupling limit. But it blows up the same way as
that for D4 branes without B-field rather than that for D6 branes without B-field. Because
of this blowing up, the theory is better analysed in eleven dimensions.
For the gravity description, the effective string coupling, e , is our real concern. The
type IIA gravity description of (D4, D6) in the above limit is

2
0
ds = f 1 (u)u2 dx02 + dx12 + + dx42 + h dx52 + dx62

du2
+ d22
+ f (u)
u2
e2 = g 2
(au)3/2
,
1 + au

,
B=
247
0 au
dx 5 dx 6 ,
b 1 + au
f (u) = (g/2)
2 (au)3/2,
h =
1
,
1 + au
(4.18)
The curvature in string units is

where g = g b 1/2 , a = b 1/2/(g/2).
0 R
1
1
1
q
2
.
3/2
geff
g
(au)
3
g bu
(4.19)
The gravity description is valid if both the effective string coupling and the above curvature
in string units are small. This gives 1/g 4/3 au 1/g 4 which can hold true only if
The above also implies au 1/g 4/3 1 and
g 1. For fixed g,
this implies large b.
1/2
1/3
e56 / 0 = 1/b vanishes for large b.
So the
b u 1/g 1. Then h 1/(au). Here B
(D4, D6) system goes over to q D4-branes in the same limit.
In the region au 1/g 4 1, the effective string coupling is large. We need to lift the
(D4, D6) system to eleven dimensions. The 11D gravity description is valid if the following
curvature in 11D Planck units is small, i.e.,
lp2 R e2/3
1
1
4/3
1,
geff g au(1 + au)1/3
(4.20)
which gives au 1/g.

So au 1/g 4 1 already ensures small curvature in 11D Planck
units. Since we have au 1, b 1/2 u 1/g 3 1 and g 1, the lifted (D4, D6) goes over
to the lifted D4 branes.
5. Conclusion
In this paper, we have investigated what causes the noncommutativity in terms of gravity
dual description of NCYM for systems of nonthreshold bound states (D(p 2), Dp) with
2 6 p 6 6 in type II string theories. Our study along with previous works strongly indicates
that in the NCYM decoupling limit, the Dp branes with a constant B-field represent
dynamically a system of infinitely many D(p 2) branes without B-field. In this limit,
the corresponding (p + 1)-dimensional NCYM is just a better description of (p 1)dimensional ordinary YangMills with gauge group U (q) with q in that it has a
finite gauge group U (n) and a finite gauge coupling. The price we pay is that we have
noncommutative space in the directions of nonvanishing B-field and we need to introduce
the product for functions. In other words, we find that the noncommutativity is entirely
due to our choice in employing Dp branes to describe the dynamics of infinitely many
D(p 2) branes without B-field. We give a clear explanation about the new parameter b
which is related to the intrinsic energy scale 1/b 1/2 for the D(p 2) branes.
In particular, we have shown that the gravity configuration of (D(p 2), Dp) system
or Dp branes with a constant B-field is reduced to that of infinitely many D(p 2) branes
without B-field in the NCYM limit and in the region of valid gravity description. The
present study provides also the reason that the D6 brane with B-field can decouple from
248
the gravity while the simple D6 brane without B-field cannot in their respective decoupling
limits. We expect that D7 branes with a rank 2 B-field can decouple from gravity, too, in the
NCYM decoupling limit since it describes infinitely many D5 branes with two additional
isometries and without B-field. But this will not be true for D8 or D9 branes with a rank 2
B-field. Our study indicates that in general Dp branes with a rank r constant B-field, this
system can decouple from gravity in the NCYM limit if p0 = p r < 6. In other words,
the higher the rank r is, the better chance we have for the Dp branes to decouple from
gravity. For example, D8 branes with rank 4 B-field can decouple from gravity.
Acknowledgements
We would like to thank Nobuyuki Ishibashi for an e-mail correspondence and
for bringing our attention to reference [32], and Ricardo Schiappa for an e-mail
correspondence. Especially, we would like to thank R.-G. Cai and N. Ohta for a very
fruitful e-mail correspondence which leads us to refine some points in Sections 1 and 2.
We would also like to thank Mike Duff for reading the manuscript, Jim Liu and Leopoldo
Pando-Zayas for discussions. J.X. Lu acknowledges the support of US Department of
Energy.
References
[1] P.M. Ho, Y.S. Wu, Noncommutative geometry and D-branes, Phys. Lett. B 398 (1997) 52;
hep-th/9611233.
[2] A. Connes, M. Douglas, A. Schwartz, Noncommutative geometry and matrix theory, JHEP 802
(1998) 003; hep-th/9711162.
[3] M.R. Douglas, C. Hull, D-branes and the noncommutative torus, JHEP 9802 (1998) 008; hepth/9711165.
[4] F. Ardalan, H. Arfaei, M. Sheikh-Jabbari, Noncommutative geometry from strings and branes,
JHEP 9902 (1999) 016; hep-th/9810072.
[5] D. Bigatti, L. Susskind, Magnetic fields, branes and noncommutative geometry, hepth/9908056.
[6] N. Seiberg, E. Witten, String theory and noncommutative geometry, hep-th/9908142.
[7] A. Hashimoto, N. Itzhaki, Non-commutative YangMills and the AdS/CFT correspondence,
hep-th/9907166.
[8] J. Maldacena, J. Russo, Large N limit of noncommutative gauge theories, hep-th/9908134.
[9] M. Alishahiha, Y. Oz, M. Sheikh-Jabbari, Supergravity and large N noncommutative field
theories, hep-th/9909215.
[10] T. Harmark, N. Obers, Phase structure of noncommutative field theories and spinning brane
bound states, hep-th/9911169.
[11] J. Barbon, E. Rabinovici, On 1/N corrections to the entropy of noncommutative YangMills
theories, hep-th/9910019.
[12] R.-G. Cai, N. Ohta, On the thermodynamics of large N noncommutative super YangMills
theory, hep-th/9910092.
[13] J. Breckenridge, G. Michaud, R. Myers, Phys. Rev. D 55 (1997) 6438.
[14] H. Arfaei, M.M. Sheikh Jabbari, Nucl. Phys. B 526 (1998) 278.
249
[15] J.X. Lu, S. Roy, (m, n)-string-like Dp-brane bound states, JHEP 9908 (1999) 002; hepth/9904112.
[16] J.X. Lu, S. Roy, Non-threshold (F, Dp) bound states, Nucl. Phys. B 560 (1999) 181.
[17] J.X. Lu, S. Roy, ((F, D1), D3) bound state and its T-dual daughters, hep-th/9905014.
[18] J.X. Lu, S. Roy, (F, D5) bound state, SL(2, Z) invariance and the descendant states in type IIB
and type IIA string theory, Phys. Rev. D 60 (1999) 126002.
[19] A. Hashimoto, N. Itzhaki, On the hierarchy between noncommutative and ordinary supersymmetric YangMills, hep-th/9911057.
[20] J.G. Russo, A.A. Tseytlin, Nucl. Phys. B 490 (1997) 121.
[21] R.-G. Cai, N. Ohta (to appear).
[22] M.J. Duff, R. Khuri, J.X. Lu, Phys. Rept. 259 (1995) 213.
[23] P.K. Townsend, Phys. Lett. B 373 (1996) 68; hep-th/9512062.
[24] B. de Wit, J. Hoppe, H. Nicolai, Nucl. Phys. B 305 [FS23] (1988) 545.
[25] T. Banks, W. Fischler, S. Shenker, L. Susskind, Phys. Rev. D 55 (1997) 5112.
[26] N. Ishibashi, Nucl. Phys. B 539 (1999) 107.
[27] N. Ishibashi, A relation between commutative and noncommutative description of D-branes,
hep-th/9909176.
[28] L. Cornalba, R. Schiappa, Matrix theory star products from the BornInfeld action, hepth/9907211.
[29] A. Sen, Adv. Theor. Math. Phys. 2 (1998) 51.
[30] N. Seiberg, Phys. Rev. Lett. 79 (1997) 3577.
[31] L. Cornalba, D-brane physics and noncommutative YangMills theory, hep-th/9909081.
[32] K. Okuyama, A path integral representation of the map between commutative and noncommutative gauge fields, hep-th/9910138.
[33] M. Li, Y.S. Wu, Holography and noncommutative YangMills, hep-th/9909085.
[34] N. Itzhaki, J. Maldacena, J. Sonnenschein, S. Yankielowicz, Supergravity and large N limit of
theories with sixteen supercharges, Phys. Rev. D 58 (1998) 046004.
[35] O. Aharony, S. Gubser, J. Maldacena, H. Ooguri, Y. Oz, Large N field theory, string theory and
gravity, hep-th/990511.

AdS2/CFT1 correspondence and near-extremal

black hole entropy
Jos Navarro-Salas 1 , Pedro Navarro 2
Departamento de Fsica Terica and IFIC, Centro Mixto Universidad de Valencia-CSIC, Facultad de Fsica,
Universidad de Valencia, Burjassot-46100, Valencia, Spain
Received 20 October 1999; revised 17 January 2000; accepted 8 March 2000
Abstract
We provide a realization of the AdS2 /CFT1 correspondence in terms of asymptotic symmetries
of the AdS2 S1 and AdS2 S2 geometries arising in near-extremal BTZ and ReissnerNordstrm
black holes. We evaluate the corresponding central charges and show that Cardys formula exactly
accounts for the deviation of the BekensteinHawking entropy from extremality. We also argue that
this result can be extended to more general black holes near extremality. 2000 Elsevier Science
B.V. All rights reserved.
PACS: 04.60.Kz; 04.60.Ds
Keywords: Asymptotic symmetries; AdS gravity; Conformal symmetry; Black hole entropy
1. Introduction
Since the discovery of the thermodynamical properties of black holes a crucial open
problem has been to find a microscopical structure responsible for the Bekenstein
Hawking entropy. In the last few years this question has started to receive some answers.
The discovery of D-branes led to an explicit statistical derivation of the black hole entropy
for extremal [1] and near-extremal [2] black holes (see also the reviews [3,4]). In a different
context, Strominger has proposed [5] a unified way to account for the BekensteinHawking
entropy of black holes whose near-horizon geometries are locally similar to the BTZ black
holes [6]. The idea of the approach of [5] is that a conformal symmetry of the gravity theory
can control the asymptotic density of states, irrespective of the details of the quantum
theory, thus providing a statistical explanation to the area formula for the entropy and, in
turn, a sort of universal mechanism. Stromingers argument is based on the holographic
1 jnavarro@lie.uv.es
2 pnavarro@lie.uv.es
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 1 6 5 - 6
J. Navarro-Salas, P. Navarro / Nuclear Physics B 579 (2000) 250266
251
relation, first discovered by Brown and Henneaux [7], between gravity on AdS3 and a
two-dimensional conformal field theory on the boundary. AdS3 gravity possesses a set
of asymptotic symmetries closing down two copies of the Virasoro algebra with central
charge c = 3`/(2G), where G is Newtons constant and 1/`2 is the cosmological
constant. Using Cardys formula [8] for the boundary CFT2 one reproduces the expected
entropy. However, the validity of Cardys formula requires that the lowest eigenvalues of
the Virasoro operators L0 and L 0 vanish. As it has been pointed out in [9], this is not the
case of the boundary theory of AdS3 gravity, because it is, up to global issues, Liouville
theory [10]. The asymptotic level density of states is then controlled by the effective
central charge [11] which, for Liouville theory, turns out to be equal to one and therefore
cannot properly account for the entropy. However, the fact that the entropy fits Cardys
formula with the ordinary central charge seems to indicate that gravity theory itself can
provide relevant information about the microscopic theory, but apparently not enough to
characterise it completely. Interesting attempts to avoid the restrictions of 2 + 1 dimensions
to explain the BekensteinHawking entropy by means of symmetry principles has been
given in [1214] by considering the horizon as a boundary.
Among the family of AdSD /CFTD1 dualities [15], the pure gravity case AdS3 /CFT2
is the best understood. In contrast, the AdS/CFT correspondence in two spacetime
dimensions is quite enigmatic. Some progress has been made in [1620]. One of the aims
of this paper is to further investigate the AdS2 /CFT1 correspondence in terms of asymptotic
symmetries. In Section 2 we shall analyse the relation between the first sub-leading
terms in the asymptotic expansion of the metric field, obeying suitable AdS2 boundary
conditions, and the stress tensor of the boundary theory, as happen in higher dimensional
situations [21,22]. On general grounds, two-dimensional anti-de Sitter space naturally
arises in the near-horizon limit around the degenerate radius of coincident horizons [23].
Therefore, a way to study Maldacenas duality in D = 2 and its implication for black
holes is to consider gravity theories having black hole solutions with degenerate horizons.
Following a similar line of reasoning as in [5] we shall show that the application of Cardys
formula to the unique copy of the Virasoro algebra emerging as an asymptotic symmetry,
correctly account for the deviation of the BekensteinHawking entropy from extremality
in the near-horizon approximation. In Section 2 we consider near-extremal BTZ black
holes and in Section 3 four-dimensional ReissnerNordstrm black holes near extremality.
Finally, in Section 4 we show that the above results can be extended to any black hole with
degenerate horizons that can be properly described by a two-dimensional effective theory.
2. Dimensional reductions of AdS3 gravity and the AdS2 /CFT1 correspondence

Einstein gravity on AdS3 is described by the action

Z
2
1
d3 x g R + 2
S=
16G
`
(2.1)
and we can dimensionally reduce the theory [24,25] via a decomposition of the metric of
the form
252
2
2
ds(3)
= g dx dx + `2 2 (x) d + A (x) dx ,
, = 0, 1
(2.2)
where ds 2 = g dx dx is a two-dimensional metric, a scalar (dilaton) field and A a

KaluzaKlein U(1) gauge field. The two-dimensional effective theory is governed by the
action

Z
2
`2 2
`
2
d x g R + 2 F F
.
(2.3)
8G
4
`
The equations of motion of the gauge field imply that
`3 3
F+ = constant,
2
g
(2.4)
where x = x 0 x 1 and F+ = + A A+ . Moreover, by varying the dilaton one

obtains
3
2
(2.5)
R + 2 `2 2 F 2 = 0,
`
4
and using (2.4) one gets
R=
3 J2
2
,
`2 2 `2 4
(2.6)
where J is related with the integration constant of (2.4). The action (2.3) turns out to be
then
Z
`
d2 x g R + V () ,
(2.7)
8G
where

J2
2
V () = 2 2 4 .
`
2`

(2.8)
The most general solution of (2.7) with a linear dilaton corresponds to the dimensional
reduction of the BTZ black hole which, in the Schwarzschild gauge, takes the form (with
At = 4GJ /r 2 )
2

r
16G2 J 2
dr 2
2
,
(2.9)
+
dt
ds 2 = 2 8GM +
`
r2
(r 2 /`2 8GM + 16G2 J 2 /r 2 )
r
(2.10)
= .
`
The two event horizons are located at
s

!
J 2
2
2
r = 4GM` 1 1
(2.11)
M`
and the outer horizon give rise to the entropy [25]
r
r
`(`M + J )
`(`M J )
2r+
=
+
,
S=
4G
2G
2G
(2.12)
253
which reproduce as expected the entropy of the original three-dimensional theory. To get a
two-dimensional AdS geometry from (2.6) we have two different ways. The simplest route
is to restrict the theory to the spinless sector J = 0 with
2
,
(2.13)
`2
but the most interesting way is to fix the value of the dilaton so that the equation of motion
is verified [23]
R=
= V (),
(2.14)
where V () is given by (2.8). This implies that the value of the dilaton = 0 should be a
zero of the potential

J2
2
= 0,
(2.15)
V (0 ) = 0 2
`
2`2 04
and then
8
.
(2.16)
`2
This AdS2 geometry can be obtained performing a perturbation around the degenerate
radius r0 = `0 of the extremal solutions (V (0 ) = 0). Redefining the dilaton
R = V 0 (0 ) =
0 + ,
the action (2.7) can be expanded as follows

Z
8
`
d2 x g R + 2 + O( 2 ).
8G
`
(2.17)
(2.18)
So, to leading order in , the metric has a constant curvature R = 8/`2 . We can also see
this behaviour introducing the coordinates (t, x)
defined by (0 < 1)
t = t/,
(2.19)
r = r0 + x,
(2.20)
where r0 = r = 2` GM0 . Keeping the angular momentum |J /`| = M0 fixed, the

solutions (2.9), (2.10), where M = |J /`|(1 + k 2 ) (k is an arbitrary positive constant),
have a well defined limit when 0
2

2
1
4x
4x
2
2
2
2
a + O() dt +
a + O()
dx 2 ,
(2.21)
ds =
`2
`2
r0
(2.22)
= + x,
`
`
with a 2 = 8M0 Gk. The above constant dilaton solution admits a nice interpretation in
terms of the three-dimensional geometry. Reconstructing the three-dimensional metric
from the KaluzaKlein decomposition we obtain
2
2

1
2
4x
4x
2
2
2
2
a dt +
a
dx 2 + `2 02 d + At dt ,
(2.23)
ds(3) =
2
2
`
`
254
where the gauge field is given by

At =
2x
.
` 2 0
(2.24)
The interesting point now is that if the coordinates (, t/`) play the role of the null
coordinates (x + , x ), where the range of variation is 0 6 x + , x 6 2 , and defining
r 2 = 40 `x
(2.25)
we can rewrite (2.23) as follows

2
= r 2 dx + dx +
ds(3)
r 2 `2
dr 2 + `2 02 (dx + )2 + a 2 (dx )2
r 4 4`4 02 a 2
(2.26)
and this is just a BTZ black hole. a 2 measures the deviation of the BTZ black hole from
extremality (M0 = J /`), which is recovered when a 2 = 0. With the opposite identification
of coordinates ( , t/`) (x , x + ) we obtain BTZ black holes with the parameter a 2
measuring the deviation from the extremal situation M0 = J /`. All this suggests that
the t coordinate of AdS2 plays the role of a null-type coordinate. This interpretation can be
derived in a more transparent way by deforming the spatial circle of the KaluzaKlein
reduction (2.2) to make a nearly light like circle. Introducing the discrete light-cone
quantization coordinates [26]
t e x 0+ + e x 0
=
,
`
e2 + e2
e x 0+ e x 0
,
x =
e2 + e2
(2.27)
(2.28)
which interpolate the original coordinates (t/`, ) (for = 0) and the infinitely boosted
light-cone coordinates x 0 = e x (x = t/` ), one obtains, in an appropriate limit
, the AdS2 geometry
2

2
4x
4x
2
0 2
2
a (dx ) +
a dx 2 .
(2.29)
`2
`2
So, this reinforce the idea that the t coordinate in AdS2 plays the role of a null coordinate.
We shall now analyze the leading order term in the expansion of the action (2.7), which
is equivalent to consider the JackiwTeitelboim model of two-dimensional gravity [27,28]:
Z
`
d2 x g (R R0 ),
(2.30)
S=
8G
whose solutions are of the form

1
R0 2
R0 2
x a 2 dt 2 +
x a 2
dx 2 ,
(2.31)
ds 2 =
2
2
x
(2.32)
=
`
with a 2 = 8GM0 k. The metrics (2.31) are locally AdS2 and in order to define a quantum
theory we have to specify boundary conditions for the fields at infinity. Mimicking the
255
analysis of three-dimensional gravity [21,22] we shall assume the following asymptotic

behaviour of the metric 3

R0 2
1
x + tt + O 2 ,
gtt =
(2.33)
2
x

1
(2.34)
gtx = t3x + O 5 ,
x
x

1
2 1
x x
+ 4 +O 6 ,
(2.35)
gx x =
2
R0 x
x
x
where we have now introduced the first sub-leading terms in the expansion aiming to relate
them with a conformal field on the boundary.
t ) preserving the above boundary conditions
The infinitesimal diffeomorphisms a (x,
are

2
1
00
t
(2.36)
= ( t ) 2 (t) + O 4 ,
2
x
R0 x

1
,
(2.37)
0 (t) + O
x = x
x
where here the prime stands for the derivative with respect to the argument t. Using the
gauge diffeomorphisms

t( t )
1
t
+O 5 ,
(2.38)
=
x 4
x

1
x ( t )
+O 2 ,
(2.39)
x =
x
x
where t and x are arbitrary functions, one can easily show that the only gauge invariant
quantity is

1 R0 2
x x ,
(2.40)
tt = tt
2 2
where is a constant coefficient.
The action of the infinitesimal diffeomorphism (2.36), (2.37) on the metric induces the
following transformation for the function tt:
tt = ( t )t0t + 2tt0 ( t )
2 000
( t ).
R0
(2.41)
So, the 2d quantity tt behaves as the (chiral component of the) stress tensor of a conformal
field theory. To evaluate the central charge we need to know the coefficient . To this end
we have to work out the charges associated to the above asymptotic symmetries. Using the
decomposition of the metric
2
(2.42)
ds 2 = N 2 dt 2 + 2 dx + N x dt ,
3 These boundary conditions where first introduced in [18].
256
the bulk Hamiltonian of the theory is given by

Z

`
dx NH + N x Hx ,
H0 =
4G
where the constraints are
0 0
2,
H = +
`
Hx = 0 0 ,
(2.43)
(2.44)
(2.45)
and the momenta

= N 1 + (N x )0 ,

= N 1 + N x 0 .
(2.46)
(2.47)
The full Hamiltonian is given by

H = H0 + K,
(2.48)
where K is a boundary term necessary to have well-defined variational derivatives.

Assuming the boundary condition for the dilaton

`
1
x
( t ) + O 2
(2.49)
= +
` 2x
x
and imposing that K vanishes for a 2 = 0, the boundary term K can be worked out [18]

1
x
x
`
x
lim 0
+
K() =
4G x
`
`
` x
`

3
2
`
`
8x
(2.50)
+ 4 gx x 2 + k .
`
x
x
Using the asymptotic expansion for the metric and the dilaton we obtain

1 R0 2
x x + .
K() =
4G 2 2
(2.51)
Moreover, the equation for the dilaton

= R0
allows to relate with the remaining quantities
2
R0
x x ,
= tt
2
and then K() can be written in terms of the unique gauge invariant quantity

1 R0 2
tt
x x .
K() =
4G
2 2
(2.52)
(2.53)
(2.54)
The standard identification of K() in terms of the stress tensor [29]

K() = tt
(2.55)
257
allows us to know the coefficient , which turns out to be
.
(2.56)
=
4G
We still have to compute the central charge. At this point we have to remark that we are
dealing with a rather particular system. The boundary of AdS2 consists of two points, and
the boundary theory reduces to a purely mechanical system. However, the existence of
asymptotic symmetries described by tt tell us that these t-dependent arbitrary functions
parametrize some sort of mechanical excitations of the metric (2.31). Although one would
like to have a more explicit description of this system in terms of a conformal mechanical
system with N degrees of freedom, the analysis of the asymptotic symmetries will
be enough to account for the entropy of the black holes. According with the interpretation
of t as a null-type coordinate (t/` can be identified with x or x + ) we shall assume that t
4
varies in the interval 0 6 t < 2`. Defining the Fourier components LR
n of tt as
LR
n
1
=
2`
Z2`
dt tt` eint/` ,
(2.57)
the Poisson algebra can be expressed as follows

R R
Ln , Lm = m LR
n,
where m = ` eimt/` . Using (2.41) it is easy to get the following Virasoro algebra

c 3
R
R
n n,m
i LR
n , Lm = (n m)Ln+m +
12
(2.58)
(2.59)
with central charge (R0 = 8/`2 )

3`
.
4G
For the black holes (2.31) we have a constant value of tt
2
a ,
tt = a 2 =
2
8G
c = 3` =
(2.60)
(2.61)
and, in terms of the mass a 2 = 8GM0 k, we have

LR
0 = `tt = M0 k`.
(2.62)
Observe that to have a Virasoro algebra of the NeveuSchwartz form we must shift the
NS
R
Ramond-type generator: LR
0 L0 = L0 + c/24.
At this point we have all the ingredients to compute the asymptotic density of
states described by a representation of the Virasoro algebra (2.59). According with the
interpretation of t as a null-type coordinate our states are like left- or right-moving states
in a 2d conformal field theory, and this illustrates, in terms of asymptotic symmetries,
the general fact [16] that the conformal group in the AdS2 /CFT1 correspondence can be
regarded as one chiral component of the two-dimensional group. It is interesting to observe
4 In [18] the integration is introduced as an average procedure over generators.
258
R
that, with respect to the time t = t/, the generators LR
shift into LR
m(t ) = Lm( t ) , and
m( t )
then LR
0(t ) is equal to the deviation of the mass from the extremal case:
2
LR
0(t ) = M0 `k = (M M0 )`.
(2.63)
From (2.60) and (2.62) we can evaluate, via Cardys formula, the degeneracy of states if
M M0 is large in the microscopic sense
s
r
cL0
`2 (M M0 )
=
,
(2.64)
2
6
2G
which turns out to be just the difference between the entropy of a nearly extremal black
hole and a extremal one
s
`2 (M M0 )
.
(2.65)
1S = S Se =
2G
Therefore, the statistical entropy (2.64) just account for microscopic excitations from the
extremal macroscopic state.
We must stress now the important fact that the statistical entropy is independent of the
length of the interval of the compactified parameter t. If we choose a different periodicity
R
R
for t: 0 < t < 2, the central charge shift c c ` , but LR
0 get modified L0 ` L0 in
such a way that cLR
0 , and hence the entropy, is not sensitive to the compactification scale
2, which can be taken arbitrarily large .
Finally we would like to consider the theory obtained by reducing the 3d theory to the
spinless sector J = 0. The reduced action is just

Z
2
`
2
d x g R + 2
(2.66)
S=
8G
`
and the solutions are of the form
2
2

1
x
x
2
a
dx 2 ,
ds 2 = 2 a 2 dt 2 +
`
`2
x
=
`
(2.67)
(2.68)
with a 2 = 8GM. Arguing now as in the previous analysis we can compute the central
charge of the Virasoro asymptotic symmetry, which turns out to be
3`
G
Moreover, we also have
c=
(2.69)
LR
0 = M`
(2.70)
The above results allow us to compute the asymptotic density of states given by Cardys
formula
r
c
,
(2.71)
2
6
259
where is the eigenvalue of the Virasoro generator LNS

0 . In our case c = 3`/G and =
M` + `/(8G). For large mass c we get the following statistical entropy
s
M`2
,
(2.72)
S = 2
2G
which coincides with the thermodynamical formula (2.12) with J = 0.
We should point out that, in contrast with the analysis of [18], which uses the convention
4G = `, we have found an exact agreement between the statistical entropy of the twodimensional black hole (2.67) and the corresponding BekensteinHawking formula. The
discrepancy comes from the evaluation of the central charge. Our result is c = 3`/G and
the authors of [18] claim, using also = ` that c = 24`/(4G).
To end this section we would like to remark that the results obtained here can be extended
to de Sitter space. If in the two-dimensional theory R0 is turned into a positive quantity
R0 = 2/`2 we have metrics of the form
2
2
1

t
t
2
2
2
2
dt 2 a dx 2 .
(2.73)
ds = 2 a
`
`
The analysis of asymptotic symmetries can also be performed in a similar way. Assuming
the following asymptotic behaviour for large t
`2 t t (x)
+ 4 + ,
t2
t
t x (x)
+ ,
gt x =
t3
t2
gxx = 2 + xx (x) + ,
`
one can show that

t t
xx = xx 4
2`
gt t =
(2.74)
(2.75)
(2.76)
(2.77)
transforms as a stress tensor

0
(x) = (x)xx + 2xx 0 (x) `2 000 (x),
(2.78)
where (x) parametrize the space-time diffeomorphism preserving the boundary conditions
`4 00
(x) + ,
2t 2
t
0
= t (x) + .
x = (x)
(2.79)
(2.80)
Since de Sitter space has a natural periodicity in x, we shall assume that x varies in the
interval 0 < x 6 2. Moreover, x is a space-like coordinate and we can define the Fourier
modes of xx in the standard way
LR
n
1
=
2
Z2
dx xx einx/ .
0
(2.81)
260
We get a Virasoro algebra with central charge

c=
12`2
(2.82)
and
2
a .
(2.83)
2
So the product cL0 is independent of the radius of the space-like slice, and it could be
related with a physical quantity. In fact, the two-dimensional de Sitter geometry (2.73)
appears as the near-horizon geometry of Schwarzschildde Sitter black holes near the
degenerate solution, and Cardys formula with the above values (the constant is model
dependent) also reproduce the entropy near the degenerate solution [30].
This situation is interesting since the Virasoro generators are defined in the standard
way (the integration is performed in the x coordinate) and the theory can be regarded as
an static two-dimensional field theory with a conformal algebra generated by a Virasoro
algebra. In contrast, when the curvature is turned into a negative quantity the role of the
t and x is interchanged, and the Virasoro generators are defined in the non-conventional
way. However we have argued that the t coordinate plays the role of a null-type coordinate.
This is reminiscent to the fact that in a 2D conformal field theory the chirality condition of
the components of the stress-tensor can convert a integral over a spatial coordinate x into
and integral over t or over x + (x ).
LR
0 =
3. Near-extremal ReissnerNordstrm black holes

Let us start with the EinsteinMaxwell action in 3 + 1 dimensions
Z
q
h
2 i
1
4
d x g (4) R (4) G F (4) .
16G
(3.1)
Imposing spherical symmetry on the electromagnetic field and on the metric

2
RN
= g
dx dx +
ds(4)
1 2 2
` (x) d 2,
2
(3.2)
where d 2 is the metric on the two-sphere and ` is the Planck length (`2 = G), the action
(3.1) reduces to
2

Z
1 2
1
1
`2
R + + 2 2 F F .
(3.3)
d2 x g
2 4
2
`
8
To perform a similar analysis to that of Section 2 we need to reparametrise the fields to
eliminate the kinetic term in (3.3) and bring the action to the more reduced form (2.3). To
this end we introduce the new fields
2
,
p4
RN
= 2 g
.
=
g
(3.4)
(3.5)
The two-dimensional effective action becomes

Z
1
1
2 1/2 `2 F F
d2 x g R +
2
2`2 3/2
261
(3.6)
and the equations of motion of the electromagnetic field yield to
4 2 ` 3/2
F+ = Q,
(3.7)
where Q is an integration constant. Plugging (3.7) into (3.6) we get

Z
1
d2 x g R + V () ,
2
(3.8)
where

`2 Q2
1
1
.
V () = 2
`
2 (2)3/2
(3.9)
The general solution with a non-constant dilaton is

p
p

1
`2 Q2
`2 Q2
2
2
2 +
dx 2 ,
2M` dt +
2M`
ds = 2 +
2
2
x
= .
`
Note that the rescaling (3.4), (3.5) map the above solutions into the standard form

2GM Q2 G2
2 RN
+
= 1
ds
dt 2
r
r2

2GM Q2 G2 1 2
+
dr ,
=+ 1
r
r2
r
= 2 .
`
As is well known, there are two event horizons, located at
p
p

2 = ` M M 2 Q2 .
(3.10)
(3.11)
(3.12)
(3.13)
(3.14)
(3.15)
of the extremal
Perturbing the solution (3.10) around the degenerate radius x0 =
solution M0 = |Q|

(3.16)
M = M0 1 + k 2 ,
t
x = x0 + x,
(3.17)
t= ,
we get in the near-horizon limit 0

1
1
1
dx 2 .
(3.18)
ds 2 = 5 3 x 2 2|Q|k` dt 2 + 5 3 x 2 2|Q|k`
` |Q|
` |Q|
1 3 2
2 ` M0
RN the
So, the curvature is R0 = 2/(`5 M03 ) and a 2 = 4M0 `. Note that for the metric g
4
2
curvature is 2/(` M0 ), which corresponds to that of the RobinsonBertotti geometry.
262
Proceeding in a parallel way as in the case of near-extremal BTZ black holes, we find
here that the charges are

1
x x ,

(3.19)
K() = ( t )
` t t 2`10Q6
and these yield to the central charge
c = 12|Q|3
`4
(3.20)
if t [0, 2]. The value of LR

0 near extremality is
LR
0 = |Q| k
and using Cardys formula we obtain
r
p
cL0
= 2 2Q3 `4 1M,
1S = 2
6
where
1M = |Q| k 2 = M M0 .
(3.21)
(3.22)
(3.23)
It is now easy to see that the statistical expression (3.22) exactly agrees with the deviation
of the BekensteinHawking entropy of near-extremal black holes from the extremal case
Se = Q2 `2
q

2
S = `2 |Q| + 1M + 2|Q| 1M + (1M)2

= Se + 1S + O (1M)3/2 .
(3.24)
4. AdS2 /CFT1 correspondence and near-extremal black holes
In this section we shall generalise the argument leading to the statistical explanation
of the near-extremal BekensteinHawking entropy of BTZ and ReissnerNordstrm black
holes to a wider family of black holes. We shall consider a generic black hole solution, in
an arbitrary dimension n, which can be described by the metric
1 2 2
` d n2 ,
(4.1)
2
where ` is the Planck length of the theory. By dimensional reduction and integrating
the equations of motion of any abelian gauge field we can arrive at an effective twodimensional theory. An additional conformal rescaling of the metric and a redefinition
of the dilaton field yield into an action of the form [31,32]
Z
1
(4.2)
d2 x g R + `2 V () ,
2G
where V () is a potential function parametrising the original theory and G is a
dimensionless constant playing the role of Newton constant in two-dimensions. The
solutions for the 2D effective metric are
ds 2 = g dx dx +

1 2
ds 2 = J () 2MG` dt 2 + J () 2MG`
dr ,
r
= ,
`
263
(4.3)
(4.4)
where V () = J 0 (). The horizons are the solutions of the equation J () = 2M` and we
have a degeneration at the zeros of the potential
V (0 ) = J 0 (0 ) = 0.
If we perturb around the degenerate radius of coincident horizons

M = M0 1 + k 2 ,
t
t= ,
r = r0 + x,
the two-dimensional metric transforms into

R0 2
2
x 2kM0 G` + O() dt 2
ds =
2

1
R0 2
x 2kM0 G` + O()
+
dx 2,
2
(4.5)
(4.6)
(4.7)
(4.8)
(4.9)
where
J 00 (0 )
.
`2
Imposing boundary conditions of the form
R0 =
R0 2
x + tt + ,
2
gtx = t3x + ,
x
2 1
x x
+ 4 + ,
gx x =
R0 x 2
x
gtt =
(4.10)
(4.11)
(4.12)
(4.13)
and working in the gauge tx = 0, the charges can be worked out without difficulty because
of the simple form of the two-dimensional effective action (4.2):
K() = ( t )tt,
where tt is the stress tensor

1 R0 2

x x .
tt =
`G t t 2 2
(4.14)
(4.15)
Assuming a periodicity of 2 in t, we obtain

c=
24
,
`GR0
LR
0 = M0 k.
(4.16)
(4.17)
264
Applying now Cardys formula we get

s
4M0 `k 2
1S = 2
R0 `2 G
(4.18)
and, taking into account (4.10) and that

1
J (h ) J (0 ) ,
2G
where h is the value of the dilaton at the outer horizon, we can rewrite (4.18) as
s
2 2(J (h ) J (0 ))
.
1S =
G
J 00 (0 )
M0 `k 2 = (M M0 )` =
(4.19)
(4.20)
On the other hand, the BekensteinHawking entropy for the two-dimensional effective
theory is given by the simple expression [32]
2
h ,
G
and therefore,
S=
1S =
2
(h 0 ).
G
(4.21)
(4.22)
Expanding J () around the extremal situation (J 0 (0 ) = 0)

1 00
J (0 )(h 0 )2 +
2
and, in the near-extremal approximation, we have
J (h ) = J (0 ) +
(4.23)
2(J (h ) J (0 ))
= (h 0 )2 ,
J 00 (0 )
(4.24)
implying the equality between the statistical expression (4.20) and the thermodynamical
one (4.22).
5. Conclusions and final remarks

We have shown that the asymptotic symmetries of BTZ and ReissnerNordstrm
extremal black holes, whose near-horizon geometry is AdS2 Sn (n = 1, 2, respectively)
are powerful enough to control the deviation of the BekensteinHawking entropy of nearly
extremal black holes from the extremal situation. We have also argued that the above results
can be generalised for arbitrary black holes near extremality if they can be described by an
effective two-dimensional dilaton gravity theory.
Our approach is based on a realization of the boundary conformal field theory in terms
of the sub-leading terms in the asymptotic expansion of the metric field. The evaluation of
the charges associated with the asymptotic symmetries near extremality allows to compute
the central charge and the value of LR
0 . These values depend on an arbitrary parameter
265
5
in such a way that cLR
0 , and hence the statistical entropy, has an absolute meaning.
However, in the present context the physical excitations are associated to the would-be
gauge diffeomorphisms characterised by the functions ( t ) and these degrees of freedom
have an effective central charge ceff = 1 (see [33]). Therefore, it could appear natural to
choose in such a way that c = 1 and bypass the question of the discrepancy between c
and ceff . This type of argument was put forward in 2 + 1 gravity in [34,35] and could have
some unexpected consequences.
Nevertheless we must also say that our approach, like [1214], does not offer and explicit
construction of the underlying conformal field theory, but rather it indicates the existence
of a conformal symmetry with an unambiguous value for the product cL0 which, inserted
in Cardys formula, can be related with the black hole entropy.
Acknowledgements
This research has been partially supported by the Comisin Interministerial de Ciencia y
Tecnologa and DGICYT. P. Navarro acknowledges the Ministerio de Educacin y Cultura
for a FPU fellowship. We want to thank J. Cruz and D.J. Navarro for useful conversations.
We also thank M. Cadoni and S. Carlip for comments on the manuscript.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
A. Strominger, C. Vafa, Phys. Lett. B 379 (1996) 99; hep-th/9601029.

C.G. Callan, J.M. Maldacena, Nucl. Phys. B 472 (1996); hep-th/9602043.
J.M. Maldacena, Black holes in string theory, hep-th/9607235.
A.W. Peet, Class. Quant. Grav. 15 (1998) 3291; hep-th/9712253.
A. Strominger, JHEP 02 (1998) 009; hep-th/9712251.
M. Baados, C. Teitelboim, J. Zanelli, Phys. Rev. Lett. 69 (1992) 1849.
J.D. Brown, M. Henneaux, Commun. Math. Phys. 104 (1986) 207.
J.A. Cardy, Nucl. Phys. B 270 (1986) 186.
S. Carlip, Class. Quant. Grav. 15 (1998) 3609; hep-th/9806026.
O. Coussaert, M. Henneaux, P. van Driel, Class. Quant. Grav. 12 (1995) 2961; gr-qc/9506019.
D. Kutasov, N. Seiberg, Nucl. Phys. B 358 (1991) 600.
S. Carlip, Phys. Rev. Lett. 82 (1999) 2828; hep-th/9812013.
S. Carlip, Class. Quant. Grav. 16 (1999) 3327; gr-qc/9906126.
S.N. Solodukhin, Phys. Lett. B 454 (1999) 213; hep-th/9812056.
J.M. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231; hep-th/9711200.
A. Strominger, AdS2 quantum gravity and string theory, JHEP 9901 (1999) 007; hepth/9809027.
J. Maldacena, J. Mickelson, A. Strominger, JHEP 9902 (1999) 011; hep-th/9812073.
M. Cadoni, S. Mignemi, Phys. Rev. D 59 (1999) 081501; hep-th/9810251; Nucl. Phys. B 557
(1999) 165; hep-th/9902040.
G.W. Gibbons, P.K. Townsend, Phys. Lett. B 454 (1999) 187192; hep-th/9812034.
S. Cacciatori, D. Klemm, D. Zanon, w Algebras, conformal mechanics, and black holes, hepth/9910065.
5 A similar situation appears in [12,14].
266
[21] J. Navarro-Salas, P. Navarro, Phys. Lett. B 439 (1998) 262; hep-th/9807019.

[22] V. Balasubramanian, P. Kraus, A stress tensor for anti-de Sitter gravity, hep-th/9902121.
[23] J. Cruz, A. Fabbri, D.J. Navarro, J. Navarro-Salas, Phys. Rev. D 61 (2000) 024011. hepth/9906187.
[24] A. Achcarro, M.E. Ortiz, Phys. Rev. D 48 (1993) 3600.
[25] D. Louis-Martinez, G. Kunstatter, Phys. Rev. D 52 (1995) 3494.
[26] J.-H. Cho, T. Lee, G. Semenoff, Phys. Lett. B 468 (1999) 52; hep-th/9906078.
[27] R. Jackiw, in: S.M. Christensen (Ed.), Quantum Theory of Gravity, Hilger, Bristol, 2000, p. 403.
[28] C. Teitelboim, in: S.M. Christensen (Ed.), Quantum Theory of Gravity, Hilger, Bristol, 2000,
p. 327.
[29] P. di Francesco, P. Mathieu, D. Snchal, Conformal Field Theory, Springer, New York, 1997.
[30] D.J. Navarro, J. Navarro-Salas, P. Navarro, Holography, degenerate horizons and entropy, hepth/9911091, to appear in Nucl. Phys. B.
[31] T. Banks, M. OLoughlin, Nucl. Phys. B 362 (1991) 649.
[32] J. Gegenberg, G. Kunstatter, D. Louis-Martinez, Phys. Rev. D 51 (1995) 1781.
[33] J. Navarro-Salas, P. Navarro, JHEP 9905 (1999) 009; hep-th/9903248.
[34] M. Baados, Phys. Rev. Lett. 82 (1999) 2030; hep-th/9811162.
[35] M. Baados, Phys. Rev. D 60 (1999) 104022; hep-th/9903178.

String loop threshold corrections for N = 1

generalized Coxeter orbifolds
C. Kokorelis
CITY University Business School, Department of Investment, Risk Management and Insurance, Frobischer
Crescent, Barbican Centre, London, EC2Y 8HB, UK
Received 3 February 2000; accepted 29 February 2000
Abstract
We discuss the calculation of threshold corrections to gauge coupling constants for the, only, nondecomposable class of abelian (2, 2) symmetric N = 1 four dimensional heterotic orbifold models,
where the internal twist is realized as a generalized Coxeter automorphism. The latter orbifold was
singled out by earlier work as the only N = 1 heterotic ZN orbifold that satisfy the phenomenological
criteria of correct minimal gauge coupling unification and cancellation of target space modular
anomalies. 2000 Elsevier Science B.V. All rights reserved.
PACS: 11.25.-w; 11.25.Mj; 11.25.Db
Keywords: N = 1 ZN orbifolds; One loop moduli-dependent gauge couplings; Threshold corrections; Coxeter
orbifolds
The purpose of this paper is to examine the appearance of one-loop string threshold
corrections in the gauge couplings of the four dimensional generalized non-decomposable
N = 1 orbifolds of the heterotic string. In 4D N = 1 orbifold compactifications the
process of integrating out massive string modes, causes the perturbative one-loop threshold
corrections 1 , to receive non-zero corrections in the form of automorphic functions of the
target space modular group. At special points in the moduli space previously massive
states become massless and contribute to gauge symmetry enhancement. As a result the
appearance of massless states in the running coupling constants appears in the form of a
dominant logarithmic term [1,2].
The moduli-dependent threshold corrections of the N = 1 4D orbifolds receive non-zero
one loop corrections from orbifold sectors for which there is a complex plane of the torus
T 6 left fixed by the orbifold twist . When the T 6 can be decomposed into the direct sum
T 2 T 4 , the one-loop moduli-dependent threshold corrections (MDGTC) are invariant
under the SL(2, Z) modular group and are classified as decomposable. Otherwise, when
1 Which receive non-zero moduli-dependent corrections from the N = 2 unrotated sectors.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 1 4 1 - 3
268
C. Kokorelis / Nuclear Physics B 579 (2000) 267274
the action of the lattice twist on the T6 torus does not decompose into the orthogonal sum
T6 = T2 T4 with the fixed plane lying on the T2 torus, MDGTC are invariant under
subgroups of SL(2, Z) and the associated orbifolds are called non-decomposable. The
N = 1 perturbative decomposable MDGTC have been calculated, with the use of string
amplitudes, in [3]. The one-loop MDGTC integration technique of [3] was extended to
non-decomposable orbifolds in [4]. Further calculations of non-decomposable orbifolds
involved in the classification list of N = 1 orbifolds of [5,6] have been performed in [7].
Here we will perform the calculation of one-loop threshold corrections for the class of
Z8 orbifolds, that can be found in the classification list of [5,6], defined by the Coxeter
twist,

2i
(1, 3, 2)
= exp
8
on the root lattice of A3 A3 . This orbifold was missing from the list of calculations of
MDGTG of non-decomposable orbifolds of [4,7] and consequently its one-loop modulidependent gauge coupling threshold corrections were not calculated in [4,7]. In [8,9] we
found that this orbifold is non-decomposable and it is the only one that possesses this
property from the list of generalized Coxeter orbifolds given in [5,6]. Its twist can be
equivalently realized through the generalized Coxeter automorphism S1 S2 S3 P35 P36 P45 on
the root lattice.
Moreover, in [9], where a classification list of the non-perturbative gaugino condensation generated superpotentials and -terms of all the N = 1 four dimensional nondecomposable heterotic orbifolds was calculated, its non-perturbative gaugino condensation generated superpotential was given. In this work we will calculate its MDGTC following the technique of [4]. Our calculation completes the calculation of the threshold corrections for the classification list of four dimensional Coxeter orbifold compactifications with
N = 1 supersymmetry of [5,6].
The generalized Coxeter automorphism is defined as a product of the Weyl reflections 2
Si of the simple roots and the outer 3 automorphisms, the latter represented by the
transposition of the roots. An outer automorphism represented by a transposition which
exchanges the roots i j , is denoted by Pij and is a symmetry of the Dynkin diagram.
In string theories the one-loop gauge couplings below the string scale evolves according
to the RG equation
2
Mstring
ka
ba
1
1
= 2
+
ln
+
4a ,
2
2
2
2
ga (p ) gM
16
p
16 2
string
(2)
where Mstring is the string scale and ba is the -function coefficient of all orbifold sectors.
For decomposable orbifolds [3] the MDGTC 4a associated with the gauge couplings ga2
2 The Weyl reflection S is defined as a reflection
i
Si (x) = x 2
hx, ei i
ei ,
hei , ei i
with respect to the hyperlane perpendicular to the simple root.

3 An automorphism is called outer if it cannot be generated by a Weyl reflection.
(1)
269
corresponding to the gauge group Ga , are determined in terms of the N = 2 sectors, fixed
under both (g, h) boundary conditions, of the orbifold, namely
Z 2
Z 2 X
d
d
(h,g)
ba Z(h,g) (, ) baN=2
.
(3)
4=
2
2
F
(g,h)
(h,g)
Here, baN=2 is the -function coefficient of all the N = 2 sectors of the orbifold, ba ,
Z(h,g) the -function coefficient of the N = 2 sector untwisted under (g, h) and its partition
function (PF) respectively. The integration is over the fundamental domain F of the
PSL(2, Z). For the case of non-decomposable orbifolds the situation is slightly different,
namely
Z 2
Z 2
d X
d
(h ,g )
ba 0 0 Z(h0 ,g0 ) (, ) baN=2
.
(4)
4=
2
2
e
F
(g0 ,h0 )O
The difference with the decomposable case now is that the sum, in the first integral of
(4) is over those N = 2 sectors that belong to the N = 2 fundamental orbit O and the
e. Because the PF Z(g ,h ) , for
integration is not over F but over the fundamental domain F
0 0
non-decomposable orbifolds, is invariant under subgroups of the modular group, i.e, e,
e is generated by the action of those modular transformations that generate
the domain F
e=
e from . In the example that we examine later in this work, e = 0 (2) and F
e on the
{1, S, ST }F . In turn the fundamental orbit O is generated by the action of F
fundamental element of this orbit.
For the orbifold Z8 there are four complex moduli fields. There are three (1, 1) moduli
due to the three untwisted generations 27 and one (2, 1)-modulus due to the one untwisted
The realization of the point group is generated by
generation 27.
0 0 0 0 0 1
1 0 0 0 0 0
0 1 0 0 0 1
Q=
(5)
0 0 1 0 0 0 .
0 0 0 1 0 1
0 0
0 0
If the action of the generator of the point group leaves some complex plane invariant then
the corresponding threshold corrections have to depend on the associated moduli of the
unrotated complex plane. There are three complex untwisted moduli: three (1, 1)-moduli
and no (2, 1)-modulus due to the three untwisted 27 generations and non-existent untwisted
generation. The metric g (defined by gij = hei |ej i) has three and the antisymmetric
27
tensor field B an other three real deformations. The equations gQ = Q g and 4 bQ = Q b
determine the background fields in terms of the independent deformation parameters.
Solving the background field equations one obtains for the metric
4 By definition ( ) mean (( )T )1 .
270
R2
v
G=
2v R 2
u
u
R2
u
v
u
2v R 2
u 2v R 2
v
u
u
v
u
R2
u
R2
v
u
v
u
R2
u
v
u
u
2
2v R
u
2
R
(6)
with R, u, v < and the antisymmetric tensor field:
0
x
z
B =
y
0
y
x
0
x
z
y
0
z
x
0
x
z
y
y
z
x
0
x
z
y
0
0
y
z
x
0
x
(7)
with x, y, z <.
The N = 2 orbit is given by these sectors which contain completely unrotated planes,
O = (1, 4 ), ( 4 , 1), ( 4 , 4 ).
The element ( 4 , 1) can be obtained from the fundamental element (1, 4 ) by an
S-transformation on and similarly ( 4 , 4 ) by an ST -transformation. The partition
torus of the fixed plane takes the following form [5,6]
function for the zero mode parts Z(g,h)
torus
Z(1,
4 ) (, , G, B) =
torus
Z(
4 ,1) (, , G, B) =
P (
N)
1
V
N
torus
, G, B) =
Z(
4 , 4 ) (,
q 2 PL q PR ,
X
q 2 PL q 2 PR ,
P (
N)
1
V
N
q 2 PL q 2 PR q i(PLPR ) ,
1
(8)
P (
N)
where with
N we denote the Narain lattice of A3 A3 which has momentum vectors
PL =
p
+ (G B)w,
2
PR =
p
(G + B)w
2
(9)
and N is that part of the lattice which remains fixed under Q4 and V its volume.
N
The lattice in our case is not self dual in contrast with the case of partition functions
torus (, , g, b) of [3]. Stated differently the general result is for the case of nonZ(g,h)
decomposable orbifolds that the modular symmetry group is some subgroup of and
torus
(, , g, b) is invariant under the same subas a consequence the partition function 2 Z(g,h)
group of .
The subspace corresponding to the lattice
N can be described by the following winding
and momentum vectors, respectively:
271
n1
2
n

0
w=
0 ,

n1
n2
n1 , n2 Z,
m1
m2
m1
and p =
m2 ,
m
1
m2
m1 , m2 Z.
(10)
They are determined by the equations Q4 w = w and Q p = p. The partition function

torus
2 Z(1,
4 ) (, , g, b) is invariant under the group 0 (2), congruence subgroup of .
Before we discuss the calculation of threshold corrections let us give some details about
congruence subgroups. The homogeneous modular group 0 SL(2, Z) is defined as
the group of two by two matrices whose entries are all integers and the determinant
is one. It is called the full modular group and we symbolize it by 0 . If the above
action is accompanied with the quotient PSL(2, Z) 0 /{1} then this is called
the inhomogeneous modular group and we symbolize it by . The fundamental domain
of is defined as the set of points which are related through linear transformations
(a + b)/(c + d). If we denote = 1 + 2 then the fundamental domain of is defined
through the relation F = { C|2 > 0, |1 | 6 1/2, | | > 1}. One of the congruence
subgroups of the modular group is the group 0 (n). The group 0 (2) can be represented
by the following set of matrices acting on as (a + b)/(c + d):

a b
| ad bc = 1, c = 0 mod 2}.
(11)
0 (2) = {
4
It is generated by the elements T and ST 2 S of . Its fundamental domain is different from

e = {1, S, ST }F . In addition
the group and is represented from the coset decomposition F
the group has cusps at the set of points {, 0}. Note that the subgroup 0 (2) of SL(2, Z)
is defined as with b = 0 mod 2.
The integration of the contribution of the various sectors (g, h) is over the fundamental
domain for the group 0 (2) which is a three fold covering of the upper complex plane. By
taking into account the values of the momentum and winding vectors in the fixed directions
torus
we get for Z(1,
4)
torus
Z(1,
4 ) (, , g, b)
X
1 t 1
1 t 1
q 2 PL G PL q 2 PR G PR
=
(PL ,PR )
N
e2ip w e2 ( 2 p G
t
1 p2p t G1 Bw+2w t Gw2w t BG1 Bw2p t w)
(12)
p,w
Consider now the the following parametrization of the torus T 2 , namely define the (1, 1)
T modulus and the (2, 1) U modulus as:
p

T = T1 + iT2 = 2 b + i det g ,
p

1
(13)
G12 + det G ,
U = U1 + iU2 =
G11
272
where G is uniquely determined by wt Gw = (n1 n2 )G (n1 n2 ) and b is the value of the

B12 element of the two-dimensional matrix B of the antisymmetric field. This way one
gets
T = 4(x y) + i8v,
(14)
U = i.
(15)
Even if we have said that we expect that this Z8 orbifold does not have a h(2,1) U-modulus
field, the T 2 torus has a U-modulus. However its value for the Z8 orbifold is fixed. The
torus (, , g, b) takes now the form
partition function Z(1,
4)
torus
Z(1,
4 ) (, , T , U ) =
1 +m n2 )
2
e2i (m1n
e T2 U2
|T U n2 +T n1 U m1 +m2 |2
(16)
m1 ,m2 2Z
n1 ,n2 Z
By Poisson resummation on m1 and m2 , using the identity:

X
t
t
e[(p+) C(p+)]+2ip ]
p
= V1
det C
e[(l+) C
t
1 (l+)2i t (l+)]
(17)
we conclude
torus
2 Z(1,
4 ) (, , T , U ) =
where
M=
n1
n2
l1 /2
l2 /2
T2
1 X 2iT det A
|(1,U )A( 1)|2
e
T2 e 2 U2
,
4
(18)
AM

(19)
and n1 , n2 , l1 , l2 Z.
torus (, ) by an S-transformation on . After exchanging
From (18) one can obtain 2 Z(
4 ,1)
ni and li and performing again a Poisson resummation on li one obtains
torus
Z(
4 ,1) (, , T , U )
2
1 X 2i (m1 n1 /2+m2 n2 /2)
|T U n2 /2+T n1 /2U m1 +m2 |2
e
e T2 U2
.
=
4 m ,m Z
(20)
1 2
n1 ,n2 Z
The factor 4 is identified with the volume of the invariant sublattice in (20). The expression
torus
0
2 Z(
4 ,1) (, , T , U ) is invariant under (2) acting on and is identical to that for the
( 4 , 4 ) sector.
Thus the contribution of the two sectors ( 4 , 1) and ( 4 , 4 ) to the coefficient baN=2
of the -function is one fourth of that of the sector (1, 4 ), thus
baN=2 =
3 (1, 4)
b
.
2 a
(21)
273
The coefficient baN=2 is the contribution to the functions of the N = 2 orbit. Including
the moduli dependence of the different sectors, we conclude that the final result for the
threshold correction to the inverse gauge coupling reads

8e1E
4
4
(22)
4a (T , T , U, U ) = baN=2 ln T2 (T /2) U2 (U ) .
3 3
The value of U2 is fixed and equal to one as can be easily seen from Eq. (15). In general for
ZN orbifolds with N > 2 the value of the U modulus is fixed. The final duality symmetry
of (22) is T0 (2) with the value of U replaced with the constant value i.
Let us use (4), (22) to deduce some information about the phenomenology of the Z8
orbifold considered in this work. We want to calculate the one-loop corrected string mass
unification scale MX , that is when two gauge group coupling constants become equal, i.e.,
1/(ka ga2 ) = 1/(kb gb2 ). We further assume that the gauge group of our theory at the string
unification scale if given by G = Gi , where Gi a gauge group factor. Taking into account
(22) we get
(b
k b
k )
h
4
4 i b2(ba kab bab ka ) b
,
MX = Mstring T2 (T /2) U2 (U )
N=2
N=2
Mstring 0.7 gstring 1018 Gev,
(23)
where ki the KacMoody level associated to the gauge group factor Gi .

We will give now some details about the integration in (4) of the integral that we used
so far to derive (22). The integration of Eq. (16) is over a 0 (2) subgroup of the modular
group since (16) is invariant under a 0 (2) transformation (a + b)/(c + d) (with
ad bc = 1, c = 0 mod 2). Under a 0 (2) transformation (16) remains invariant if at the
same time we redefine our integers n1 , n2 , l1 and l2 as follows:
0

n1 n02
a c/2
n1 n2
=
.
(24)
l10 l20
l1 l2
2b d
The integral can be calculated based on the method of decomposition into modular orbits.
There are three sets of inequivalent orbits under the 0 (2), namely:
e = {1, S, ST }F
(a) The degenerate orbit of zero matrices, where after integration over F
gives as a total contribution I0 = T2 /4.
(b) The I3 orbit of matrices with non-zero determinants. The following representatives
give a non-zero contribution I1 to the integral:

k j
0 p
0 p
,
,
, 0 6 j < k, p 6= 0,
(25)
0 p
k j
k j +p
where I0 + I1 = (3/2) 4 Re ln (T /2).
(c) The orbits of matrices with zero determinant,

0 0
j p
,
, j, p Z, (j, p) 6= (0, 0).
j p
0 0
(26)
The first matrix in (26) has to be integrated over the half-band { C, 2 > 0, |1 | < h}
while the second matrix has to be integrated over a half-band with the double width in 1 .
The total contribution from the modular orbit I3 gives,
274

8
I3 = 4 Re ln (U ) ln(T2 U2 ) + E 1 ln
3 3

1
1
8
1
E 1 ln .
4 Re ln (U ) ln(T2 U2 ) +
2
2
2
3 3
Putting I0 , I1 , I3 together we get (22).
All N = 1 four dimensional orbifolds have been tested in [10] as to whether they satisfy
several phenomenological criteria, involving:
(a) Correct unification of the three gauge coupling constants at a scale MX 1016
Gev, assuming the minimal supersymmetric, Standard Model gauge group G =
SU(3) SU(2) U (1), particle spectrum with a SUSY threshold close to weak
scale, in the two cases of (i) a single overall modulus in the three complex planes
T = T1 = T2 = T3 and (ii) the anisotropic squeezing case T1 T2 , T3 ;
(b) Anomaly cancellation with respect to duality transformations of the moduli in the
planes rotated by all the orbifold twists.
The only orbifold from this study that satisfy all the phenomenological criteria, set out by
Ibez and Lst, is the Z8 orbifold that we examined in this work.
References
[1] M. Cvetic, A. Font, L.E. Ibez, D. Lst, F. Quevedo, Nucl. Phys. B 361 (1991) 194.
[2] C. Kokorelis, Gauge and gravitational couplings from modular orbits in orbifold compactifications, Phys. Lett. B 477 (2000) 313, hep-th/0001062.
[3] L. Dixon, V. Kaplunovsky, J. Louis, Nucl. Phys. B 355 (1991) 649.
[4] P. Mayr, S. Stieberger, Nucl. Phys. B 407 (1993) 725.
[5] Y. Katsuki, Y. Kawamura, T. Kobayashi, N. Ohtsubo, Y. Ono, K. Tanioka, Nucl. Phys. B 341
(1990) 611.
[6] J. Erler, A. Klemm, Commun. Math. Phys. 153 (1993) 579.
[7] D. Bailin, A. Love, W. Sabra, S. Thomas, Mod. Phys. Lett. A 9 (1994) 67; Mod. Phys. Lett.
A 10 (1995) 337.
[8] C. Kokorelis, Theoretical and Phenomenological Aspects of Superstring Theories, Ph.D. Thesis,
SUSX-TH-98-007; hep-th/9812061.
[9] C. Kokorelis, Generalized -terms from orbifolds and M-theory, hep-th/9810187.
[10] L.E. Ibez, D. Lst, Nucl. Phys. B 382 (1992) 305.

Total cross section measurements

with , 6 and protons
on nuclei and nucleons around 600 GeV/c
SELEX Collaboration
U. Dersch i , N. Akchurin p , V.A. Andreev k , A.G. Atamantchouk k ,
M. Aykac p , M.Y. Balatz h , N.F. Bondar k , A. Bravar t , P.S. Cooper e ,
L.J. Dauwe q , G.V. Davidenko h , G. Dirkes i , A.G. Dolgolenko h ,
D. Dreossi t , G.B. Dzyubenko h , R. Edelstein c , L. Emediato s ,
A.M.F. Endler d , J. Engelfried e,m , I. Eschrich i,1 , C.O. Escobar s,2 ,
A.V. Evdokimov h , I.S. Filimonov j,3 , F.G. Garcia s , M. Gaspero r ,
S. Gerzon l , I. Giller l , V.L. Golovtsov k , Y.M. Goncharenko f ,
E. Gottschalk c,e , P. Gouffon s , O.A. Grachov f,4 , E. Glmez b ,
He Kangling g , M. Iori r , S.Y. Jun c , A.D. Kamenskii h , M. Kaya p ,
J. Kilmer e , V.T. Kim k , L.M. Kochenda k , K. Knigsmann i,5 ,
I. Konorov i,6 , A.P. Kozhevnikov f , A.G. Krivshich k , H. Krger i ,
M.A. Kubantsev h , V.P. Kubarovsky f , A.I. Kulyavtsev f,c ,
N.P. Kuropatkin k , V.F. Kurshetsov f , A. Kushnirenko c , S. Kwan e ,
J. Lach e , A. Lamberto t , L.G. Landsberg f , I. Larin h , E.M. Leikin j ,
Li Yunshan g , Li Zhigang g , M. Luksys n , T. Lungov s,7 , D. Magarrel p ,
V.P. Maleev k , D. Mao c,8 , Mao Chensheng g , Mao Zhenlin g ,
S. Masciocchi i,9 , P. Mathew c,10 , M. Mattson c , V. Matveev h ,
E. McCliment p , S.L. McKenna o , M.A. Moinester l , V.V. Molchanov f ,
A. Morelos m , V.A. Mukhin f , K.D. Nelson p , A.V. Nemitkin j ,
P.V. Neoustroev k , C. Newsom p , A.P. Nilov h , S.B. Nurushev f ,
A. Ocherashvili l , G. Oleynik e,8 , Y. Onel p , E. Ozel p , S. Ozkorucuklu p ,
S. Patrichev k , A. Penzo t , S.I. Petrenko f , P. Pogodin p , B. Povh i ,
M. Procario c , V.A. Prutskoi h , E. Ramberg e , G.F. Rapazzo t ,
B.V. Razmyslovich k , V.I. Rud j , J. Russ c , Y. Scheglov i , P. Schiavon t ,
V.K. Semyatchkin h , J. Simon i , A.I. Sitnikov h , D. Skow e , V.J. Smith o ,
M. Srivastava s , V. Steiner l , V. Stepanov k , L. Stutte e , M. Svoiski k ,
N.K. Terentyev k,c , G.P. Thomas a , L.N. Uvarov k , A.N. Vasiliev f ,
D.V. Vavilov f , V.S. Verebryusov h , V.A. Victorov f , V.E. Vishnyakov h ,
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 2 0 4 - 2
278
U. Dersch et al. / Nuclear Physics B 579 (2000) 277312
A.A. Vorobyov k , K. Vorwalter i,11 , J. You c , Zhao Wenheng g ,

Zheng Shuchen g , R. Zukanovich-Funchal s
a Ball State University, Muncie, IN 47306, USA
b Bogazici University, Bebek 80815 Istanbul, Turkey
c Carnegie-Mellon University, Pittsburgh, PA 15213, USA
d Centro Brasiliero de Pesquisas Fsicas, Rio de Janeiro, Brazil
e Fermilab, Batavia, IL 60510, USA
f Institute for High Energy Physics, Protvino, Russia
g Institute of High Energy Physics, Beijing, PR China
h Institute of Theoretical and Experimental Physics, Moscow, Russia
i Max-Planck-Institut fr Kernphysik, 69117 Heidelberg, Germany
j Moscow State University, Moscow, Russia
k Petersburg Nuclear Physics Institute, St. Petersburg, Russia
l Tel Aviv University, 69978 Ramat Aviv, Israel
m Universidad Autnoma de San Luis Potos, San Luis Potos, Mexico
n Universidade Federal da Paraba, Paraba, Brazil
o University of Bristol, Bristol BS8 1TL, United Kingdom
p University of Iowa, Iowa City, IA 52242, USA
q University of Michigan-Flint, Flint, MI 48502, USA
r University of Rome La Sapienza and INFN, Rome, Italy
s University of So Paulo, So Paulo, Brazil
t University of Trieste and INFN, Trieste, Italy
Received 22 October 1999; revised 29 February 2000; accepted 27 March 2000
Abstract
Total cross sections for 6 and on beryllium, carbon, polyethylene and copper as well as
total cross sections for protons on beryllium and carbon have been measured in a broad momentum
range around 600 GeV/c. These measurements were performed with a transmission technique in the
SELEX hyperon-beam experiment at Fermilab. We report on results obtained for hadronnucleus
cross sections and on results for tot (6 N) and tot ( N), which were deduced from nuclear cross
sections. 2000 Elsevier Science B.V. All rights reserved.
PACS: 13.85.Lg; 13.85.-t; 13.75.Ev; 13.75.Gx; 24.10.Ht
Keywords: Total cross sections; Glauber model; Hyperon reactions
1 Now at Imperial College, London SW7 2BZ, UK.

2 Current Address: Instituto de Fsica da Universidade Estadual de Campinas, UNICAMP, SP, Brazil.
3 Deceased.
4 Present address: Dept. of Physics, Wayne State University, Detroit, MI 48201.
5 Now at Universitt Freiburg, 79104 Freiburg, Germany.
6 Now at Physik-Department, Technische Universitt Mnchen, 85748 Garching, Germany.
7 Current Address: Instituto de Fsica Terica da Universidade Estadual Paulista, So Paulo, Brazil.
8 Present address: Lucent Technologies, Naperville, IL.
9 Now at Max-Planck-Institut fr Physik, Mnchen, Germany.
10 Present address: Motorola Inc., Schaumburg, IL.
11 Present address: Deutsche Bank AG, 65760 Eschborn, Germany.
279
1. Introduction
Hadronic total cross sections provide one measure of the strength of the hadronic
interaction. They have been measured for a variety of reactions over a broad range of
center of mass energies. These studies revealed that with increasing center of mass (CM)
energy, hadronhadron cross sections (generally) decrease to a minimum and then start
rising again. An important current physics question is whether the rise of a specific hadron
hadron cross section is described by a power law in the CM energy. Addressing this
question requires total cross-section experiments performed with a variety of hadronic
projectiles, targets and energies covering the maximum possible range. However, for
almost 20 years, there have been few new experiments in this field. Thus, important
hadronhadron cross sections such as tot (p) and tot (Kp) are measured only up
to 380 GeV/c and tot (6 p) is only measured up to 137 GeV/c. At these maximum
laboratory momenta only a first indication of the rise of these total cross sections is
observed.
SELEX (Fermilab E781) is a fixed-target experiment at the Fermi National Accelerator
Laboratory using a hyperon beam of about 600 GeV/c. The SELEX spectrometer, designed
for spectroscopy of charm baryons, is well-suited to measure total cross sections with a
transmission technique. It has excellent scattering-angle resolution, achieved by a system
of silicon microstrip detectors.
SELEX does not have a liquid hydrogen target. Therefore, we measured the total hadronnucleus cross sections tot ( Be), tot ( C), tot ( CH2 ), tot (6 Be),
tot (6 C), tot (6 CH2 ), tot (pBe) and tot (pC) with high precision. We then deduced
the total cross sections tot (6 p) and tot ( p) using both a CH2 C subtraction technique and a method based on the Glauber model to derive hadronnucleon cross sections
from hadronnucleus cross sections.
Further, as data on hadronnucleus cross sections are extremely scarce for charged
projectiles, we also measured tot ( Cu) and tot (6 Cu). All measurements were done
during dedicated run periods in July 1997. Laboratory momenta range from 455 GeV/c to
635 GeV/c, the highest energy yet used for these studies.
2. Experimental setup
2.1. The hyperon beam
The hyperon beam is generated by selecting positively or negatively charged secondaries
around 600 GeV/c that emerge from interactions of an 800 GeV/c primary proton beam
with a beryllium production target. Its composition has not been completely measured.
However, we have measured the main particle components of the event samples, which we
selected to determine total cross sections (see Section 5.2.1). This analysis shows that at
the position of the total cross-section target the negative beam samples consist in average
of (52.5 1.6)% mesons and (47.5 1.6)% baryons. Further, we measured a 4 fraction
of (1.18 0.06)% in these samples. Other baryonic fractions (p, ) were not measured,
280
but empirical formulae (see [1]) predict they are less than 0.1%. Likewise, the K fraction
of the negative beam is estimated with [1] to be (1.6 1.0)%. Thus, we expect that the
fraction of the event samples is (50.9 1.9)% and the 6 fraction is (46.3 1.6)%.
In the event samples for positive beam we measured a meson fraction of (8.1 1.4)%
and a baryon fraction of (91.9 1.4)%. The measured 6 + fraction was (2.7 0.7)%.
Using the empirical formula given in [1], we expect that the tiny meson fraction consists
of 70% + and 30% K+ . Thus, the samples for positive beam consisted to (89.2 1.6)%
of protons.
From these compositions, one sees that as long as one can distinguish mesons from
baryons (see Section 2.2), the SELEX hyperon beam offers a unique possibility to measure
total cross sections for protons, , and 6 in a low contaminant environment.
2.2. The section of the SELEX spectrometer used for total cross-section measurements
The SELEX spectrometer is a 60 m long, 3 stage spectrometer. In total cross-section
measurements, only its upstream detectors, shown in Fig. 1, are used.
The beam spectrometer placed in front of the target, is equipped with 12 silicon
microstrip detectors to track
incoming particles. The first 4 microstrip detectors (HSDs)
have a resolution (pitch/ 12) of 14.4 m and a maximum signal integration time of 100 ns.
As this is the shortest integration time, but poorest spatial resolution, of all SELEX silicon
microstrip detectors, the HSDs serve chiefly to reject out-of-time tracks. Always, two
HSDs are housed in a single station. The average efficiency of the HSDs is 92%.
The remaining 8 silicon microstrip detectors of the beam spectrometer are grouped into
3 stations (BSSDs) mounted on a granite block inside a noise shielded cage (RF-cage).
These detectors have a resolution of 5.8 m and an average efficiency of 99.6%.
Incoming particles are identified by a transition radiation detector (BTRD) with
10 separate transition radiation detector modules (TRMs). Each module is build of a
radiator followed by 3 proportional chambers (PCs). A radiator consists of a stack of
200 polypropylene foils, each 17 m thick and spaced at 500 m. The PC gas is a
70% Xe, 30% CO2 mixture to optimize signal response time and to maximize absorption
of transition-radiation photons. Each chamber has a single anode readout amplifier.
Each BTRD PC gives a digital output when it detects an energy deposition above a fixed
threshold. The sum of all PCs detecting a signal above threshold is the TRD plane count k.
A typical probability spectrum of TRD plane counts, a BTRD signal spectrum, is shown
in Fig. 2. It shows the baryon and meson responses at low and high TRD plane counts,
respectively.
The signal components are separated by fitting the function:

2
4
X
X
n k
n k
(1)
i
i
pi (1 pi )nk +
p (1 pi )nk
pfit (k) =
k
k i
i=1
i=3
|
{z
} |
{z
}
baryon signal
meson signal
to the normalized BTRD signal spectrum. We used four binomials to account for
the four main beam components as well as to obtain an excellent description of the
281
Fig. 1. Sections of the SELEX spectrometer involved in the measurement of total cross sections.
BTRD signal spectrum. In Eq. (1), pi and i are fit-parameters with the constraints
1 = 1 + 2 + 3 + 4 and p1 , p2 < p3 , p4 and n is the maximum possible TRD plane
count. The fit-parameters pi have the meaning of a PC response probability, when a meson
(light particle) or baryon (heavy particle) passes. Thus, we obtain from (1) the meson
fraction (3 + 4 ) and the baryon fraction (1 + 2 ) of the beam.
The target is followed by the vertex spectrometer, which consists of 22 silicon microstrip
detectors grouped into 6 stations (VSSD1, . . . , VSSD5 and HSD3). All VSSDs have
a resolution of 5.8 m. Except for one plane, which has a reduced efficiency of 68%,
all VSSDs have an average efficiency of 98.8%. At the end of the vertex spectrometer,
station HSD3 is mounted to the RF cage.
Although the total cross-section measurements presented in this article are based only
on detectors placed in the beam and the vertex spectrometer, we also use other parts of the
SELEX apparatus to compute corrections. Further detectors involved in the analysis are
situated in the M1 and the M2 spectrometer (see Fig. 1), which we describe briefly.
The M1 spectrometer starts at the center of the M1 magnet and ends at the center of
282
Fig. 2. A typical BTRD signal spectrum obtained for 600 GeV/c negatively charged secondaries.
the M2 magnet. For high resolution tracking of high energy particles in the central beam
region, sets of 6 silicon microstrip detectors (LASD1 and LASD2) are mounted to the faces
of the M1 and the M2 magnet. The LASD detectors have a resolution of 14.4 m, and an
average efficiency of 95.8%. For tracking outside the central beam region, 12 planes of
wire chambers (PWCs) are installed.
The M2 spectrometer starts at the center of the M2 magnet. To enhance the momentum
resolution for high energy particles, a third station of silicon microstrip detectors (LASD3)
is mounted to the end face of the M2 magnet. This station is followed by 14 PWCs that are
grouped into 7 stations (M2 PWC1, . . . , M2 PWC7).
2.3. The targets
To optimize the precision, total cross-section measurements are done with special
targets. Great care was taken in selecting and machining adequate target materials in
order to obtain best chemical and mechanical properties (see Table 1). All targets are thin;
multiple scattering, quantified by of Molires formula is significantly lower than the
25 rad angular resolution provided by the beam and vertex spectrometer.
The carbon target is a stack of three pyrocarbon plates, each about 5 mm thick.
Pyrocarbon is composed of thin carbon layers accumulated on top of each other in a hightemperature methane atmosphere. Compared to standard graphite it offers the advantages:
no open porosity, a density close to that of a graphite monocrystal and less than 1 ppm (parts
per million) non-carbon constituents. The beam faces of the carbon plates were milled with
a diamond-powder liquid and oriented such that the beam faces of the stack are parallel to
each other.
283
Table 1
Specifics of the targets used in total cross-section measurements. L: target thickness, : density, :
expected spread in scattering angle due to multiple scattering calculated with Molires formula for
plab = 600 GeV/c, Xcoll : collision length
Target
material
beryllium
carbon
polyethylene
copper
Thickness
L [mm]
z-direction
50.92
15.46
40.86
1.00
Transverse
dimensions
x [mm]
y [mm]
30.7
30.0
30.0
30.0
51.2
30.0
25.0
30.0
Density
[g/cm3 ]
[rad]
Xcoll
[%]
1.848 0.002
2.199 0.003
0.9291 0.0008
8.96 0.009
8.3
6.0
6.3
5.7
16.86
5.40
6.66
1.05
The polyethylene target is build from a high-purity polyethylene granulate with less than
1000 ppm contaminants. Molten granulate was solidified in a vessel, where great care was
taken that no air bubbles penetrated. The material was then carefully machined to a target
block, and beam faces were flattened using a diamond pin.
For the beryllium and the copper target, standard industry products of high purity are
used.
2.4. Trigger and data acquisition
The SELEX trigger is a programmable four-stage trigger, designed to select events
involving decays of charm hadrons in a high-intensity beam environment. The first 3 levels:
T0, T1 and T2 are hardware triggers, whereas level T3 is an online software filter. In this
section, we describe only the trigger as programmed for total cross-section data-taking.
At data-taking, the trigger accepted all beam events defined by the minimum-bias
condition:
T0 = S1 S2 S3 V1 V2 V3.
(2)
S1, S2 and S3 are scintillation counters, and V1, V2 and V3 are veto counters to reject
beam halo (see Fig. 1). In definition (2), a T0-pulse indicates a particle traversing the beam
spectrometer in the direction of the target beam face. The transverse trigger acceptance is
constrained to the size of the hole in V2 (12.8 mm 12.8 mm).
In order to keep the minimum bias condition provided by the definition of T0, no
information from detectors placed downstream of the experiment target influenced the
spectrometer readout. Thus, each T0-pulse passed the T1 trigger level unbiased, and
generated a T2-pulse, which started the spectrometer readout. The online software filter
(level T3) was not used for total cross-section data-taking. Pulses of all trigger levels were
counted by scalers for each spill, and saved in a trigger log file.
The SELEX trigger controlled readout and reset of the silicon-detector system, the
basic tool in our total cross-section measurements. Except for the HSDs, all other silicon
detectors use an SVX-I chip technology for data readout [2]. SVX chips are controlled by
a sequencer SRS (silicon readout sequencer) that interacts very closely with the trigger.
284
First, it keeps the silicon detectors sensitive (for about 5 s effective integration time) and
starts the chip readout when receiving a T2-pulse. Second, the SRS resets the SVX-chips,
when a silicon-clear signal arrives. The silicon-clear signal is generated in the trigger logic
through:
Silicon clear = V5mult Cpulser (T1 T2 ).
(3)
Here, Cpulser are pulses from a gate generator running at a frequency of 20 kHz and
V5mult represents pulses generated, when the V5 veto counter (see Fig. 1) detects a high
multiplicity event. The condition (T1 T2 ) was irrelevant for total cross-section data.
2.5. Experimental conditions and recorded data
During the fixed-target run 1996/97, the TEVATRON was operated in 60 s cycles with a
spill time of 20 s. Data for total cross sections were taken during dedicated periods, with
optimized experimental conditions for this measurement.
By adjusting the flux of the 800 GeV/c proton beam, the T0-rate was optimized to run
the SELEX DAQ near, but safely below its capacity limit of 5 104 particles per spill. The
low hyperon-beam flux allowed a high silicon-clear rate, which resulted in a very low-noise
condition for the silicon-detector system and a low probability for out-of-time tracks.
During data-taking, the M1 magnet was switched off to obtain a 2.5 m field- and
material-free section, serving as fiducial region for precise reconstruction of hyperon
decays. Magnet M2 was operated with a transverse momentum kick of pTM2 = 0.84 GeV/c.
At data-taking start, after mounting an experiment target in the RF-cage, an alignment
RUN was taken to account for eventual detector displacements caused during the target
installation. Then, the position of the experiment target was alternated every 30 min
between its out and in-beam position. Thus, almost equal amounts of data were taken
with full and empty target. A RUN, started after each target-position change, comprised
typically 106 events. A total of 9.8 107 minimum-bias events were recorded with negative
beam for the targets Be, C, Cu and CH2 . With positive beam, 3.0 107 minimum-bias
events were written using the targets Be and C.
3. The principle of the transmission method

In contrast to scattering experiments, where tot is deduced from a measured scattering
angle distribution, in a transmission experiment tot is deduced from the number of
unscattered projectiles. Strictly, unscattered means zero scattering angle, but experimental
resolution and Coulomb scattering limit this to a determination of the number of projectiles
scattered by an angle , which is smaller than a maximum angle parameter max (F0 (<
max )). Thus, one infers the number of unscattered particles by extrapolating F0 (< max )
to max = 0.
A standard transmission experiment consists of three elements: beam monitor, target,
and transmission counter. The number of projectiles hitting the target under full-target
(empty-target) condition F0 (E0 ) is counted by the beam monitor placed in front of the
285
target. A transmission counter, placed downstream of the target, counts the corresponding
number of projectiles Ftr (< i ) (Etr (< i )), leaving the target within the maximum solid
angles 1 , . . . , N . Recorded counts are combined to give a set of partial cross sections
part (< i ), defined as:

F0
Etr (< i )
1
NA
log
,
(4)
with =
part (< i ) =
L
Ftr (< i )
E0
A
where is the density of scattering centers in the target, A is the atomic mass and NA is
Avogadros number.
Driving our choice of a transmission method is an important technical advantage of
Eq. (4). We do not need to know absolute efficiencies of the beam and the transmission
monitor. Their absolute values will cancel in (4) as long as they remain unchanged between
and during the full- and the empty-target RUNs (stability condition).
Taking into account the event correlations between F0 (E0 ) and Ftr (< i ) (Etr (< i )),
the statistical error of a partial cross section is given by:
s
1
1
1
1
1
+
.
(5)
part (< i ) =
L Ftr (< i ) F0 Etr (< i ) E0
In a thin target approximation (Ltot 1), a partial cross section part (< i ) is related
to the total hadronic cross section tot (see, e.g., [3]) by:
Z4
tot = part (< i )
i
d
d
Z4
d
d
d

d
CN
{z
}
|
Correction for C and CN scattering

Zi
d hadr
d hadr
(6)
d +
d .
+
d el
d inel
0
0
{z
} |
{z
}
|
elastic term
inelastic term
In equation (6), tot is inferred by first correcting partial cross sections for Coulomb scattering (C) and the Coulomb hadronic interference (CN) and then extrapolating to zero solid
angle.
Zi
4. Data analysis
4.1. Data selection
In general, total cross-section data taken for a specific target were subject to varying
experimental conditions: thresholds on silicon microstrip detectors, high voltages for
trigger scintillators, and the inclination angle between primary proton beam and production
target. Therefore, data belonging to a cross-section measurement with a specific target were
divided into as many data sets as differing conditions had to be taken into account. This
286
offered the possibility to calculate corrections and errors specifically for each experimental
condition in a later stage of the analysis. To preserve the stability condition mentioned in
section 3, a spill by spill data pre-selection was performed. Data of a spill or a whole run
were rejected:
(1) When the experimental conditions concerning the functionality of the spectrometer
(detector efficiencies, trigger performance and track reconstruction efficiencies)
suddenly changed.
(2) When it was not possible to synchronize raw data with information in the trigger log
file.
(3) When the BTRD showed instabilities or when the beam phase space lay outside the
BTRD fiducial region.
4.2. Event selection for normalization
The total cross-section determination is made by counting how many good beam tracks
are removed from the beam by interactions in the target. The normalization therefore
depends only on the number of good beam tracks, which are identified by a software
decision routine. This routine reconstructs tracks in the beam spectrometer using the HSD
and BSSD hit information. It preserves the minimum-bias condition for the selected data
by strictly avoiding event-selection rules that require information from detectors placed
downstream of the target. An event is accepted when it is possible to reconstruct a norm
track that satisfies the following criteria:
(1) Not more than a total of 150 hits in all BSSDs.
(2) At least 6 hits from BSSD planes along the track.
(3) At least one hit from an HSD plane along the track (HSD-tagging).
(4) A reduced track-fit 2 below 3.
(5) An extrapolated origin of the track at the known transverse position of the primary
production target.
(6) Track intercept and slope parameters within the beam phase space accepted by
magnetic collimation.
(7) A transverse track position at the longitudinal position of the experimental target,
which is inside the trigger acceptance window and inside BTRD acceptance.
(8) A beam momentum assigned to the track, which is 100 GeV/c around the center
of gravity value of the momentum spectrum.
Condition (3) rejects out-of-time tracks. The selection rules (4)(6) remove events in which
hyperons decay before reaching the experiment target or react with detector material in the
beam spectrometer. Constraint (7) assures also that selected tracks point to the mid-part of
the experiment target face, where the best mechanical accuracy is obtained.
About 50% of the selected events had a norm track. From the resulting set of norm tracks
for full- and empty-target conditions, we establish classes of BTRD-tagged norm tracks.
This is done by introducing cuts on the BTRD information as indicated in Fig. 2 to separate
baryonic and mesonic norm tracks. We then determine the corresponding normalization
counts F0 and E0 by summing the norm tracks over the appropriate signal region.
287
4.3. Transmission counting

When a norm track is found in the event, we try to reconstruct a single track in the vertex
spectrometer at small angle to the norm track. The single-track algorithm was efficient
and fast. It used hits of HSD3 to remove out-of-time tracks. With loose cuts on the track
parameters 98% of the norm tracks got assigned a track in the vertex spectrometer. Such
vertex tracks were finally accepted as transmitted tracks, when:
(1) There are at least 15 hits from VSSDs found within a track search corridor.
(2) The reduced track-fit 2 is below 3.
For each transmitted track, the scattering angle between norm and transmitted track is
calculated. Following the idea of [4], a four-momentum transfer t is assigned to the event
2
2 , where pbeam is the momentum of the
using the small angle approximation t pbeam
incoming particle. Transmitted tracks are assigned to t bins of width 5.0 104 GeV2 /c2 .
Note that we count transmitted tracks in t-bins, rather than in bins of solid angle as
discussed in Section 3. Summing the events in the t-bins from zero up to a maximum ti
leads to sets of transmission counts Ftr (< |ti |) and Etr (< |ti |).
4.4. Spectra of uncorrected partial cross sections
Using the counts F0 , E0 , Ftr (< |ti |), Etr (< |ti |) and the mechanical properties of the
targets, partial cross sections part (< |ti |) are calculated according to Eq. (4).
Fig. 3 shows some spectra for uncorrected partial cross sections. The strong rise of
part (< |ti |) for |t| < 0.002 GeV2 /c2 is ascribed to multiple scattering in the target and the
finite angular resolution of 25 rad. Differing levels of partial cross-section spectra for
beam particles of different kind indicate nicely the dependence of the total cross section on
the projectile type.
Fig. 3. Spectra of uncorrected partial cross sections resulting from beryllium target data sets.
288
4.5. Corrections for non-hadronic effects

Partial cross sections were corrected for single Coulomb scattering (C) and for the
CoulombNuclear interference effect (CN) evaluating the expression:
corr
part
< |ti | = part

Zti
d
d
0
dt
dt 0 .
< |ti |
0
dt C
dt 0 CN
|
{z
} |
{z
}
C correction
CN correction
Zti
(7)
Applying the Coulomb correction, a change in the extrapolated cross section of not more
than 0.5% is observed for the light targets Be, C and CH2 . For the Cu target a change of up
to 11% is noticed. The CN correction is roughly one order of magnitude smaller than the
Coulomb correction and has negligible effect on the extrapolated cross section.
4.6. The extrapolation method
As |t| approaches zero, the growth behavior of partial cross sections is governed ideally
by the elastic term in Eq. (6). At small |t|, hadronic coherent elastic scattering off nuclei
dominates. Thus, we obtain for the elastic term in Eq. (6), the expression:
Zt
0
d
dt 0
hadr
el
dt 0 =
2

tot
1 + 2 1 eBnuc t ,
16Bnuc
(8)
where Bnuc is the exponential slope observed in hadronic coherent elastic scattering off
nuclei. Therefore, we choose the functional form

(9)
f (1 , 2 , t) = 1 1 e2 t
to describe the variation of partial cross sections with respect to |ti |.
The parameters 1 and 2 are determined in fitting function (9) to differences in
corrected partial cross sections of adjacent t-bins in the range of tmin = 0.007 GeV2 /c2
to tmax = 0.03 GeV2 /c2 . The limits tmax and tmin account for experimental sensitivity
to hadronic coherent elastic scattering off nuclei. Their derivation is described in
Section 4.6.1.
Starting from the partial cross section part (< |tmin |), the total cross section tot is
determined by extrapolating the t-variation of the partial cross sections from tmin to t = 0
using the expression:

(10)
tot = part < |tmin | + 1 1 e2 tmin .
4.6.1. The limits tmin and tmax and the sensitivity of the SELEX experiment to coherent
hadronic elastic scattering off nuclei
In measurements of hadronnucleus cross sections, it is essential that the experiment
is sensitive to hadronic coherent elastic scattering off nuclei. Further, one must be able to
distinguish coherent from incoherent scattering processes off nucleons. In scattering off
289
nuclei, the nucleus can break up when the energy transfer exceeds the binding energy of
its nucleons. This leads to a contribution of incoherent scattering off nucleons for |t| >
0.015 GeV2 /c2 . In that case, the hadronic differential elastic cross section, entering the
elastic term of Eq. (6), contains two parts:

d
dt
hadr
=
el
2 (hA)

2 (hN) BN t
tot
1 + 02 eBnuc t + N(A) tot
e
.
16
| 16
{z
} |
{z
}
coherent scattering
incoherent scattering
(11)
There is a term for coherent elastic scattering off the nucleus, in which tot (hA) is the total
nuclear cross section, and a term for incoherent scattering off nucleons, in which tot (hN)
is the corresponding hadronnucleon cross section. BN is the slope parameter for scattering
off nucleons, and N(A) is a factor describing the effective number of nucleons taking part
in the incoherent process for target nuclei of mass A (see [5]).
The contribution of the incoherent term decreases the growth behavior of the elastic
term in (6) because BN is typically one or more orders of magnitude smaller than
Bnuc . An extrapolation based on partial cross sections, selected in a |t|-range far above
0.015 GeV2 /c2 would lead to a systematically lowered cross-section result, because a
fraction of the elastic processes would be ignored. Consequently, we looked for a t-interval
[tmin ; tmax ] to select partial cross sections where their growth is dominated by Bnuc .
The sensitivity of the SELEX spectrometer to hadronic coherent elastic scattering off
nuclei was verified by looking at background-subtracted differential scattering spectra.
These spectra, not acceptance-corrected, are defined by:
S(t) =

F (t) E(t)
1
.
L F0
E0
(12)
Here, is the width of the t-bins. F (t) (E(t)) is the number of scattering events found in
the full-target (empty-target) data sets that fall into the interval [|t| /2; |t| + /2].
Fig. 4 shows a typical example of an S(t) spectrum obtained for 6 scattering off carbon
nuclei. The spectrum shows three regions governed by apparently different exponential
slopes, which can be explained by contributions of Coulomb scattering, coherent elastic
scattering and incoherent elastic scattering comparable to measurements described in [5].
Determinations of the slope parameters Bnuc and BN in S(t) spectra showed the expected
order of magnitude for all targets, and Bnuc agreed quite well with data presented in [6].
Furthermore, the magnitude of Bnuc is also reflected by the size of parameter 2 in Eq. (10),
when applying the extrapolation.
From such studies, we choose tmax = 0.03 GeV2 /c2 , as this value is well inside the
region dominated by coherent hadronic elastic scattering off nuclei for all targets. The
contribution of the integrated incoherent term at this tmax is much lower than the integrated
coherent term.
To avoid large multiple-scattering corrections, we chose tmin of 0.007 GeV2 /c2 , so that
the angular resolution has negligible effect on the extrapolated total cross section.
290
Fig. 4. Differential scattering spectrum obtained for 6 carbon reactions, showing the Coulomb, the
coherent and the incoherent region.
5. Corrections
5.1. Trigger-rate corrections
The trigger rate influences the reconstruction efficiency for tracks and thus alters the
transmission ratios Tfull and Tempty per spill. Fig. 5 shows an instructive example of this
effect.
Due to the rate effect, our extrapolated total cross-section experiences a shift T0 when
the average T0-counts, calculated for all empty and all full-target spills separately, differ.
To determine the shift T0 we calculate full and empty-target transmission ratios per
spill for |t| < 0.01 GeV2 /c2 and describe their rate dependency by fitting to the expression
Tfit (T0) = 1,k + 2,k T0k .
(13)
We have studied the effect of different powers (k = 2, 3, 4) to estimate systematic errors.

We choose the average T0-rate T0, comprising all full and all empty-target spills as
reference rate for the rate correction. Thus, transmission ratios per spill are corrected by
evaluating:

T0
= Tj |t| < 0.01 GeV2 /c2 + 2,k T0k T0kj ,
(14)
Tj,k
{z
} |
|
|{z}
{z
}
uncorrected
corrected
correction
T0 . Fit-function dependent offsets
which results in a set of corrected transmission ratios Tj,k
T0,k are deduced by:
291
Fig. 5. Dependency of full-target transmission ratios on the T0-count.

T0,k
T0,k = part
< 0.01 GeV2 /c2 part < 0.01 GeV2 /c2
(15)
and averaged to a mean offset T0 . Total cross sections are then corrected by:
T0
= tot + T0 .
tot
(16)
Averaged sizes of the rate correction are presented in Table 2. We want to mention that
the copper data were taken at higher rate, where the slope of function (13) is steeper. This,
together with the small thickness of the copper target, causes large corrections.
5.2. Corrections for beam contaminants
A transition radiation detector does not make an exact particle identification because
of statistical fluctuations in X-ray generation and background from various processes.
Therefore, when selecting the baryon or the meson component of the hyperon beam by
applying cuts on the BTRD plane count, we need to account for:
(1) The meson (baryon) contamination in the baryon (meson) sample and the effect on
the total cross section.
(2) The baryon (meson) contamination in a specific sample for a measurement with
protons or 6 ( ) and the effect on the total cross section.
exp
Once the contaminant fraction is determined, the experimental cross section tot can be
corrected by the term cont using:
h
i
(2)
(1)
1
exp
(1)
log 1 + (2) eL(tot tot ) 1 .
= tot +
(17)
tot
L
{z
}
|
Correction cont
This formula was derived in [7] for a two component beam having a contamination fraction
(2) .
5.2.1. Beam contaminant determination
In a first step, total cross sections resulting from data sets are corrected for the fraction
of mesons (baryons) in a baryon sample (meson sample) using (17). Therefore, we fit
292
Systematic errors
Cross
section
tot (pBe)
tot (6 Be)
tot ( Be)
tot (pC)
tot (pC)
tot (6 C)
tot ( C)
tot (6 CH2 )
tot ( CH2 )
tot (6 Cu)
tot ( Cu)
Corrections
plab
[GeV/c]
extr
[mb]
BTRD
[mb]
fluc
[mb]
rate
[mb]
cont
[mb]
tgt
[mb]
T0
[mb]
cont
[mb]
536
638
638
457
490
598
591
589
585
609
608
0.91
1.20
0.50
0.90
1.81
1.57
1.30
2.10
1.26
163
85
0.70
0.49
0.17
2.11
1.53
1.92
1.40
2.55
0.96
41
52
0.25
0.04
0.22
0.54
0.68
1.21
1.50
0.69
0.54
76
78
0.35
0.10
0.05
0.38
1.15
1.18
0.95
0.16
0.12
41
36
0.06
0.07
0.61
0.09
0.10
0.13
0.63
0.16
0.75
0.33
2.99
0.30
0.27
0.21
0.47
0.47
0.43
0.33
0.30
0.23
1.23
1.03
1.24
0.93
0.79
11.22
3.87
6.42
3.11
3.67
2.90
754
649
0.62
0.65
1.00
0.91
0.86
1.12
1.03
1.44
1.21
3.1
4.7
Table 2
Average sizes of systematic errors and corrections. For explanation of symbols see text of Sections 5.1, 5.2 and 6.1
293
function (1) to normalized BTRD signal spectra, which are recorded for norm tracks. For
negative beam these fits yield an average baryon fraction (1 + 2 ) of (47.5 1.6)% and
an average meson fraction (3 + 4 ) of (52.5 1.6)%. For positive beam, we measure
a baryon fraction of (91.9 1.4)% and a meson fraction of (8.1 1.4)%. To deduce the
meson (baryon) contaminant fraction , we sum the meson (baryon) component of (1) over
(2)
(1)
the TRD plane count region shown in Fig. 2. Further, the difference tot tot is calculated
in taking rate corrected extrapolated cross-section results obtained for the meson and the
baryon beam component.
In a second step, we account for the main contaminant disturbing a specific measurement
for protons, 6 and . According to the expected hyperon-beam composition we correct:
(1) For the effect of 4 particles in the baryon sample, when measuring 6 A cross
sections.
(2) For the effect of 6 + particles in the baryon sample, when measuring pA cross
sections.
(3) For the effect of K particles in the meson sample, when measuring A cross
sections.
For case (1), we measure the overall fraction of 4 particles in each negative-beam data
sample and for case (2) we measure the overall fraction of 6 + particles in each positivebeam data sample. Therefore, we count the decays 6 n + , 4 30 + and
6 + n + + , reconstructed for a known number of norm tracks within the field-free
region of the M1 magnet. Fig. 6 shows some hyperon-mass spectra obtained by the decay
reconstruction.
Particle decay counts are corrected for geometrical acceptance, branching ratio and
decay losses after the target, to yield the overall hyperon contaminant fractions. Here, we
find an overall 4 fraction of (41.18 0.06)%, and an overall 6 + fraction of (2.7 0.7)%.
These fractions are then divided by the baryon fraction (1 + 2 ), known from the first step
Fig. 6. Hyperon-mass spectra obtained from reconstructed 6 , 4 and 6 + decays. The spectra are
fit to a Gaussian plus a linear background function. mX is the mean mass found for hyperon X and
m is the corresponding mass resolution resulting from the fit.
294
procedure to yield the hyperon contaminant fraction of the baryon component.

Case (3) requires knowledge of the number of K particles in the meson sample.
The SELEX spectrometer cannot differentiate between 600 GeV/c and K particles.
We estimate the overall fraction of K particles in the sample using particle-flux
parameterizations of [1]. This results in an overall K fraction of (1.6 1.0)%, which
divided by the meson fraction (3 + 4 ) yields the K contaminant fraction of the meson
component.
Calculating the contaminant correction using Eq. (17) requires knowledge of the total
cross sections tot (4 A), tot (6 + A) and tot (K A). As data on these cross sections are
either scarce or do not exist, we estimate them using approximations like:
tot (4 A) tot (4 p)
tot (pA)
,
tot (pp)
(18)
and neglect weak energy dependencies. Necessary data for hadronnucleon cross sections
are taken from [4,8] and data for pA-cross sections are taken from [9].
Averaged sizes of the contaminant correction including both correction steps are shown
in Table 2.
6. Results for hadronnucleus cross sections

Total cross sections as well as their statistical and systematic errors were determined
for each dataset separately. In order to calculate average total cross sections and average
systematic errors, we use weighted means. We present the error contributions, the data
averaging method and the final results.
6.1. Measurement errors
6.1.1. The statistical error
The dominant error contribution is the statistical error, which is governed by the
statistical uncertainty of the partial cross section part (< |tmin |), used in the extrapolation.
Further statistical error contributions, originating in other terms of the error propagated
formula (10), are negligible. The statistical errors for each measurement are presented in
Table 3.
6.1.2. Systematic errors
In this section we briefly describe the systematic errors found during the data analysis.
Table 2 gives an overview of the average sizes of these errors as well as the rate correction
and the contaminant correction.
Systematic error of the extrapolation extr
A significant systematic error contribution is the choice of tmin for extrapolation
of partial cross sections. This error is based on the RMS-spread (root mean square)
of the extrapolated total cross section when tmin is varied from 0.004 GeV2 /c2 to
0.01 GeV2 /c2 .
295
Table 3
Results for nuclear total cross sections. For explanation of symbols see text of Section 6.2
Cross
section
tot (pBe)
tot (6 Be)
tot ( Be)
tot (pC)
tot (pC)
tot (6 C)
tot ( C)
tot (6 CH2 )
tot ( CH2 )
tot (6 Cu)
tot ( Cu)
plab
[GeV/c]
tot
[mb]
536
638
638
457
490
598
591
589
585
609
608
268.6
249.1
188.7
333.6
335.4
308.9
234.1
376.4
286.1
1232
1032
stat tot
[mb]
syst tot
[mb]
tot tot
[mb]
0.7
0.9
0.8
3.1
3.6
2.1
1.5
2.0
1.3
133
77
1.3
1.3
0.9
2.4
2.9
3.8
3.1
4.1
2.0
192
162
1.5
1.6
1.2
3.9
4.6
4.3
3.5
4.5
2.4
233
179
Cut on the BTRD signal spectrum BTRD

Although contaminant and rate corrections are applied for each specific cut on the BTRD
signal spectrum, we still observe a variation of the cross section when varying the cut on the
TRD plane count by 1 unit around its nominal value. Therefore, we calculate a systematic
error, which is the maximum spread in the cross sections found in the cut variation.
Spill to spill fluctuations fluc
Here, we compare the statistical error in part (< 0.01 GeV2 /c2 ), which we calculate
from (5) with the error in part (< 0.01 GeV2 /c2 ) calculated from the experimentally
observed RMS-spread of rate corrected transmission ratios per spill. The difference in these
errors accounts for remaining non statistical spill to spill fluctuations.
Systematic error of the rate correction rate
This error takes into account the error arising from different functional attempts to
describe the rate effect presented in Section 5.1. Its value is given by the maximum spread
of the T0,k with respect to their average value T0 .
Systematic error of the contaminant correction cont
This systematic error accounts for the uncertainty in the fit parameters of the four-fold
binomial distribution (1) and for the uncertainty in the contaminant fractions for 6 + , 4
and K .
Uncertainty of the target density tgt
The target densities were measured several times, using a pycnometer and a buoyancy
method. Laboratory studies showed systematic discrepancies in the density measurement,
which are included in the density errors shown in Table 1. These errors are propagated to
an error contribution to the total cross sections, which are on a 0.1% level.
296
6.2. Data-averaging and results on hadronnucleus cross sections

6.2.1. The average total cross section
Total cross-section results tot,i , obtained from i = 1, . . . , N data sets, are combined to
an average total cross section tot with a statistical error stat tot and an average systematic
error syst tot . The results are shown in Table 3.
We average the total cross-sections tot,i that correspond to a specific measurement using
the weighted mean:
PN
1
i=1 i tot,i
tot = P
,
i =
(19)
PM
syst .
N
stat
2
(i ) + j =1 (j,i )2
i=1 i
The weight i includes the statistical error (istat ) of data set i and all j = 1, . . . , M
syst
systematic errors j,i described in Sections 6.1.1 and 6.1.2.
The statistical error is supposed to decrease when adding more data to the evaluation.
We calculate the statistical error in the averaged total cross section by:
v
u N
u
X 1
stat
.
(20)
tot = t1
(istat )2
i=1
In assigning a systematic error to an average total cross section tot we assume that the
systematic errors of the single measurements can be just averaged. Therefore, we quote as
average systematic error:
v
h
i
uP
u N PM ( syst)2
u i=1 i
j
=1
j,i
,
(21)
syst tot = t
PN
i=1 i
using the weights i defined in (19).
Further, we quote a total error tot tot of the average total cross section, which is
calculated from:
q
2
2
(22)
tot tot = stat tot + syst tot .
6.3. Comparison to existing data on hadronnucleus total cross sections
A literature survey showed that experimental data on hadronnucleus total cross sections
for charged projectiles at high energies are extremely scarce. Information for proton
nucleus and nucleus total cross sections is only provided by the Refs. [5,6,9] and
displayed together with our results in the Figs. 7, 8, 9, 10 and 11. No data were found for
6 nucleus total cross sections.
6.3.1. Comparison of nucleonnucleus total cross sections
In Figs. 7 and 8 we display a compilation of protonnucleus and neutronnucleus cross
sections extracted from [7,1012] together with our results. As can be seen, the proton
nucleus cross sections of [5] at plab = 20 GeV/c and the neutronnucleus cross sections are
380
360
tot [mbarn]
tot(NBe) [mbarn]
400
nBe data
297
295
290
285
pBe Bellettini et al.

280
pBe SELEX
275
340
270
265
320
260
255
300
10
10
plab [GeV/c]
280
260
240
(pBe) for PDG data on tot(pp)
(pBe) for tot(pp) from (27)
mod
220
mod
200
1
10
10
plab
10
[GeV/c]
Fig. 7. Summary of experiment data on tot (nBe) from [7,1012] and on tot (pBe) from [5] and
SELEX (the colour version of this figure can be seen on the Nuclear Physics Electronic website:
http://www.elsevier.nl/locate/npe). Overlaid are results from the model calculation (see Section 7).
similar. For this reason we assume that differences between neutronnucleus and proton
nucleus cross sections are negligibly small above 20 GeV/c. This allows a comparison of
our protonnucleus cross sections with corresponding neutronnucleus cross-section data
available at much higher energy.
Comparing our results with neutronnucleus cross sections at 131273 GeV/c (data
of [7]) shows that our measurements follow the trend of these data points. Averaging the
neutronberyllium total cross sections in this momentum range results in 271.0 0.6 mb,
which is close to our protonberyllium cross section at 536 GeV/c of 268.6 1.5 mb.
A similar calculation for the neutroncarbon cross section gives a mean value of 331.0
0.8 mb, which is close to our measurements of the protoncarbon cross section around
457 GeV/c of 333.6 3.9 mb.
6.4. Comparison of nucleus total cross sections
High-energy data for tot ( Be), tot ( C) and tot ( Cu) that were determined
using a transmission technique are presented in the thesis of A. Schiz [9]. Unfortunately,
the statistical errors quoted for the A total cross-sections are quite large and mask other
corrections. Luckily, on the basis of [9], a publication on hadronnucleus elastic scattering
appeared [6], where fits to elastic nucleus scattering data are performed. We extracted
298
tot [mbarn]
tot(NC) [mbarn]
500
nC data
355
350
345
pC Bellettini et al.
450
340
pC SELEX
335
330
325
400
320
315
10
10
plab [GeV/c]
350
300
(pC) for PDG data on tot(pp)
(pC) for tot(pp) from (27)
mod
mod
250
1
10
10
plab
10
[GeV/c]
Fig. 8. Summary of experiment data on tot (nC) from [7,1012] and on tot (pC) from [5] and
SELEX (the colour version of this figure can be seen on the Nuclear Physics Electronic website:
http://www.elsevier.nl/locate/npe). Overlaid are results from the model calculation (see Section 7).
nucleus total cross-sections from these fits and present them in the Figs. 9, 10 and 11.
The Figs. 9, 10 and 11 show that the SELEX results for Be, C and Cu cross
sections are quite comparable to the data from [6]. However, more precise data are needed
to do a detailed comparison.
7. Model description of hadronnucleus cross sections

In this section, we introduce a model calculation for hadronnucleus cross sections and
show how well it describes the data.
7.0.1. The Glauber model and the inelastic screening correction
As shown in [7], the Glauber model [13,14] including an inelastic screening correction [15], is very precise in describing neutronnucleus cross sections at high energy. The
Glauber model accounts for the elastic screening effect in nuclei via multiple elastic scattering between the incident hadron h and the nucleons N. As mentioned in [7], nuclear
total cross sections calculated by the Glauber model exceed experimental data. This is
compensated by taking into account the inelastic screening correction described in [15].
It accounts for inelastic reactions h + N N + X, which produce an inelastic screening
mod comprises two parts:
effect. Consequently, a model cross section tot
299
200
190
tot( Be) [mbarn]
180
170
160
150
mod
mod
( Be) for PDG data on tot( p)
Be Schiz et al.
( Be) for tot( p) from (27)
Be SELEX
140
10
10
plab
10
[GeV/c]
tot( C) [mbarn]
Fig. 9. Summary of experiment data on tot ( Be) from [6] and SELEX (the colour version of
this figure can be seen on the Nuclear Physics Electronic website: http://www.elsevier.nl/locate/npe).
Overlaid are results from the model calculation (see Section 7).
260
250
240
230
220
210
200
190
mod
( C) for PDG data on tot( p)
C Schiz et al.
mod
( C) for tot( p) from (27)
C SELEX
180
10
10
plab
10
[GeV/c]
Fig. 10. Summary of experiment data on tot ( C) from [6] and SELEX (the colour version of
this figure can be seen on the Nuclear Physics Electronic website: http://www.elsevier.nl/locate/npe).
Overlaid are results from the model calculation (see Section 7).
300
Fig. 11. Results for tot ( Cu) from [6] and SELEX.

mod
Gla
tot
A, tot (hN) = tot
A, tot (hN) 1 Kar .
(23)
Gla (A, (hN)) and an inelastic screening

These are a Glauber model cross section tot
tot
Kar
correction 1 .
The Glauber model cross section

Gla (A, (hN)) can be calculated by:
According to [14], tot
tot
(Z
Gla
(hA) = 4 <e
tot
)

A
0 )
(1 ihN
tot (hN)T (b) b db ,
1 1
2
1
T (b) =
2
J0 (qb)e
BhN q2
S(q)q dq,
4
S(q) =
q
Z
r sin(qr)(r)
dr.
(24)
0
hN
is the real to the imaginary part of the elastic scattering amplitude in the forward
Here
direction observed in hadronnucleon elastic scattering and b is the impact parameter. BhN
is the hadronic slope in hadronnucleon elastic scattering and J0 is a Bessel function of
order zero. The nuclear density (r)
is normalized as:
Z
2
(r)r
dr = 1.
(25)
The inelastic screening correction

The inelastic screening correction 1 Kar , originally formulated in [15] for proton
nucleus reactions, is generalized by:
Z ( sm
Z p)
1 Kar = 4
0 (mp +m )2
d2
dt dM 2

t =0
2
1
e
e 2 tot (hN)AT(b) F (qL , bE ) dM 2 d2 b,
e
T(bE ) =
Z+
E z) dz,
(
b,

2 mp
q L = M 2 mp
301
Z+
F (qL , b) = A
(b,
z)eiqLz dz,
(26)
Here mp is the proton mass and m is the pion mass. The double differential cross section
d2 /dt dM 2 describes the inelastic reaction h + N N + X of the incident hadron h with
a nucleon N, where the resulting final state X has an invariant mass squared of M 2 .
7.0.2. Input parameters for the total cross-section model
0 , B , (r)
and (d2 /dt dM 2 )|t =0 . All of them
Model input parameters are tot (hN), hN
hN
are extracted from experimental data with N = p.
Input parameter tot (hN)
Model calculations require values of tot (hN) for a wide range of center of mass energies
s. We fit data on pp and p total cross sections from [8] to a smooth function:
tot (hp, s) =
a0
+ a2 log2 (s).
s a1
(27)
The fit-parameters ai , their errors and the validity range of each parameterization, are
shown in Table 4. The result of each parameterization is in mb, when using s in GeV2 .
Table 4
Fit-parameters and validity range of the total cross-section parameterizations
Reaction
a0
a1
a2
Momentum range
pp
49.51 0.26
0.097 0.002
0.314 0.004
10 . . . 3000 GeV/c
55.2 7.2
0.255 0.032
0.346 0.020
80 . . . 380 GeV/c
0
Input parameter hp
0 (p ) and 0
0 from [1628] and data
We parameterize pp
(p ), using data on pp
lab
p lab
0
0
on p from [16,21,29,30], assuming that reaches a constant value when plab goes to
infinity. Our fits are
0
(plab ) = +
pp
6.8
6.6
+ 0.124
0.599
plab
for 0.8 GeV/c < plab < 2100 GeV/c,
0.92
0 p (plab ) = 0.54 + 0.54
plab
for 8.0 GeV/c < plab < 345 GeV/c,
0.742
plab
(28)
(29)
where plab is in GeV/c. Fig. 12 displays these fit-functions together with all data points
included in the fit.
302
0 (p ) and 0
Fig. 12. Our parameterizations for pp
(p ) together with experimental data from
lab
p lab
[1628].
Input parameter Bhp

For the hadronic slope parameters Bpp and B p we take the parameterizations presented
in [31]:
B
= 11.13 6.21
plab + 0.30 log{plab }, q = 0.02,

pp,1
4.94
(30)
Bpp (plab ) = Bpp,2 = 9.26 plab + 0.28 log{plab }, q 2 = 0.20,
Bpp,3 = 9.67 7.51 + 0.10 log{plab }, q 2 = 0.40,

plab
B
= 9.11 +
p,1
Bp (plab ) = Bp,2 = 6.95 +
Bp,3 = 6.13 +
0.65
plab
0.65
plab
0.65
plab
+ 0.29 log{plab },
q 2 = 0.02,
+ 0.27 log{plab },
q 2 = 0.20,
+ 0.25 log{plab },
q2
(31)
= 0.40.
Here, q 2 is in units of GeV2 /c2 . These parameterizations are linearly interpolated to

account for the dependency of BhN on both plab and q 2 .
Input parameter (d2 /dt dM 2 )|t =0
To calculate the inelastic screening correction 1 Kar , we use the parameterization of
2
(d /dt dM 2 )|t =0 for the process p + p p + X, given in [7]:
26.470 (M 2 1.17) 35.969 (M 2 1.17)2
2
d
+ 18.470 (M 2 1.17)3 4.143 (M 2 1.17)4
(32)
=
2
dt dM t =0 + 0.341 (M 2 1.17)5 for 1.17 < M 2 < 5 GeV2 /c2 ,
for M 2 > 5 GeV2 /c2 .

4.4/M 2
In addition, we also use more recent parameterizations for (d2 /dt dM 2 )|t =0 to describe
the processes p + p p + X and + p p + X, which are presented in [32] and are based
on calculations of triple-Regge diagrams in [33]. For M 2 6 M02 , these parameterizations
303
Fig. 13. The parameterizations (32) and (33) evaluated for plab = 600 GeV/c.
consist of a background term and a sum of non-energy-dependent resonance terms. In case

M 2 > M02 the parameterizations consist of a sum over contributions from triple-Regge
diagrams:
P
c (M 2 M 2 )
2
i 2 ai2 2
+ f2 2 2min , M 2 6 M02 ,
d
(M Mi ) +i
(M M ) +df
(33)
= P
2 k(0) k(0) 0 min
k(0)
dt dM 2 t =0
M
1
2 > M2 .
V
,
M
k
2k(0)
k
0
s
s
Instead of displaying the large number of parameters for Eq. (33), which are taken from
calculations in [33], we display the parameterizations (32) and (33) in Fig. 13.
Compared to (33), parameterization (32) has no s-dependence. Further, parameterization (33) is not continuous and the resonance sizes are quite different for p + p p + X.
Input parameter (r)
In the calculations, we use density distributions (r)
that are based on the harmonicoscillator model:

2
r 2 ar
(34)
e rad .
(r)
= 0 1 +
arad
This offers the possibility to calculate some integrals in an analytic way and gives a
better description of the (charge-) density distribution for light nuclei than a standard twoparameter Fermi parameterization. As reported in [7], we also find that the model does not
provide a good description of neutronnucleus total cross sections if one uses both and
arad from electron-scattering data [34]. Therefore, we used values from [34] and adjusted
mod (NA, a ) gives a best
the radius parameter arad , such that the model cross section tot
rad
description of nA-cross section data in the momentum range 10273 GeV/c. Adjusting
of arad was done for each nucleus and for each of the parameterizations (32) and (33)
separately. Table 5 gives a summary of the density parameters.
304
Table 5
Parameters of the density distribution (r)
from electronnucleus elastic scattering [34] and the

mod (a , p ) to nA cross-section data in the momentum
radius parameters resulting from a fit of tot
rad lab
range 10273 GeV/c
Nucleus
beryllium
carbon
Data from [34]

eA scattering
Fit result using (32)

mod (NA, a )
in tot
rad
Fit result using (33)

mod (NA, a )
in tot
rad
arad [fm]
arad [fm]
arad [fm]
0.611
1.067
1.791
1.687
1.89981
1.79247
2.02914
1.89277
7.0.3. Results of the model calculations

Results for nucleonnucleus model cross sections
To show the quality of our model calculation after adjusting the nuclear density paramod (Be, (pp)) and mod (C, (pp))
meter arad , we evaluated the total cross sections tot
tot
tot
tot
using function (32). This was done for data on tot (pp) taken from [8] and for values on
tot (pp) resulting from our fit (27). The calculations were done at many different values
of plab to show the behavior over the entire high momentum region. Scatter in the model
calculations (observed when experimental data on tot (pp) are used) demonstrate the sensitivity of the model to small changes in tot (pp).
Summaries of calculation and data are shown in Figs. 7 and 8. They show that the
calculations reflect quite well the cross-section data for plab > 5 GeV/c. The nBe data
of [7] in the range 131273 GeV/c suggest a rise of the nBe cross section with energy
that is also indicated by the model calculation. Our data point does not show any rise for
pBe. In the case of nC cross sections our measurements join both data at lower energy and
calculation very nicely.
Results for nucleus model cross sections
mod (Be, ( p)) and mod (C, ( p)) using
We evaluated the cross sections tot
tot
tot
tot
function (33) and the corresponding nuclear density parameter arad , which was determined
by a fit of the model cross section to neutronnucleus data. All further input parameters
are specific for p-reactions. The calculations were done for data on tot ( p) taken
from [8] and for values from function (27).
Results are shown in Figs. 9 and 10 together with data for nucleus total cross
sections from [6] and the SELEX experiment. The figures show that the calculations match
our measurements quite well and agree within errors with lower-energy data from [6].
8. Results for hadronnucleon cross sections

The hadronnucleon cross sections tot (6 N) and tot ( N) were first determined by
a CH2 C method. As this method provides hadronnucleon cross sections only with a
305
precision on the order of 10%, we improved the precision using a method which takes
advantage of the more precise hadronnucleus cross-section ratios.
8.1. Hadronnucleon cross sections using a CH2 C difference method
The hadronnucleon cross sections tot (6 N) and tot ( N) can be deduced from
corresponding cross sections measured on carbon and polyethylene by:

1
(35)
tot (hN) = tot (hCH2 ) tot (hC) ,
2
where h denotes the incident hadron. Results obtained by this method are presented in
Table 9. The quoted errors are calculated from the total errors in the hadronnucleus cross
sections given in Table 3.
8.2. Hadronnucleon cross sections deduced from hadronnucleus cross sections
In a second approach, we deduce hadronnucleon cross sections from ratios of measured
hadronnucleus cross sections. To motivate the method, we first derive empirical relations
between hadronnucleon and hadronnucleus cross section ratios, which we then refine
using the model calculation described in Section 7.
To derive empirical relations between hadronnucleon and hadronnucleus cross section
ratios we use data on hadronnucleon cross sections around 137 GeV/c from [4,8], and
obtain the hadronnucleon cross-section ratios:
tot (6 p)
tot ( p)
0.635 0.006,
0.901 0.012.
(36)
tot (pp)
tot (pp)
Next, we build nuclear cross-section ratios using our measurements for the 6 A, A
and pA cross sections from Table 3.
Our pA cross sections were measured at lower laboratory momentum than the
corresponding 6 A or A cross sections. To correct for this, we scale the pA cross
sections by a factor kscale before building the cross-section ratio. The scale factor takes into
account the growth of the pA cross section from the laboratory momentum where it was
measured to the larger laboratory momentum of the corresponding 6 A or A cross
section. Scaling factors are calculated using the model described in Section 7. They are
displayed together with the nuclear cross-section ratios in Table 6.
The nuclear ratios show that the A cross sections are about 0.7 times and the 6 A
cross sections are about 0.92 times as large as the pA cross section.
To get a first relation between hadronnucleon and hadronnucleus cross sections, we
ignore the weak energy dependence of the cross-section ratios. Calculating the ratios of
hadronnucleon to hadronnucleus cross-section ratios using the above data gives the
results presented in Table 7.
The double ratios show a small but significant deviation from one especially for ratios
involving cross sections. From this empirical observation it follows that a hadron
nucleon cross section tot (hN) can be approximately derived from the pp cross section and
a hadronnucleus cross-section ratio using the relation:
306
Table 6
Nuclear cross-section ratios. The pA-cross section is scaled by kscale to account for the discrepancy
in laboratory momenta of the cross sections used in the ratio
plab [GeV/c]
kscale
640
1.0058
= 0.695 0.014
590
1.0036
tot (6 Be)/tot (pBe) = 0.922 0.008
640
1.0058
590
1.0040
Scaled cross-section ratio

tot ( Be)/tot (pBe) = 0.698 0.006
tot
tot
( C)/
tot (pC)
(6 C)/
tot (pC)
= 0.917 0.018
Table 7
Ratios of the hadronic cross-section ratios at 137 GeV/c and the nuclear cross-section ratios around
600 GeV/c
Double ratio
tot (6 p)/tot (pp)
tot (6 Be)/tot (pBe)
(6 p)/
Result
Double ratio
Result
0.977 0.016
tot ( p)/tot (pp)

tot ( Be)/tot (pBe)
0.910 0.012
( p)/
tot
tot (pp)
tot (6 C)/tot (pC)
0.983 0.023
tot
tot (pp)
tot ( C)/tot (pC)
0.915 0.020
average ()
0.980 0.014
average ()
0.913 0.012

tot (hN) tot (pp)

tot (hA)
,
tot (pA)
(37)
where is a parameter specific for the cross section ratio (compare with Table 7). If we
set = 1 for simplicity, we see that the precision of (37) is about 10%. The precision is
improved by adequate adjusting of .
Unfortunately we cannot empirically derive from experimental cross sections for
laboratory momenta around 600 GeV/c as necessary cross-section data is missing. Thus,
as we want to deduce hadronnucleon cross sections from nuclear cross-section ratios with
best precision, we improve the relation between hadronnucleon and hadronnucleus cross
sections using the total cross-section model that was introduced in Section 7.
The idea of the model-based ratio method is the following: Rewriting (37) yields the
following relation between the experimental hadronnucleus and the model based hadron
nucleus cross-section ratios.
mod (A, tot (hN))
tot (hA)
.
= tot
(pA)
mod (A, tot (pN))
| tot{z }
| tot
{z
}
experimental theory +tot-data
(38)
Taking the ratio of model based quantities reduces the effect of uncertainties in the crosssection model. Because precise data for tot (pp) is available over a large energy range,
it is convenient to use protonnucleus cross sections in the denominator. The energy
dependence of the pp cross section is known at SELEX energies. The model is adjusted
307
to describe NA cross sections for plab > 10 GeV/c. Therefore the energy dependence of
tot (hp), which we want to determine, is determined by the known energy dependence of
the pp cross section.
To deduce the cross section tot (hN) from the measured nuclear cross-section ratio, we
mod (A, (pp)) by taking
fix tot (pN)(= tot (pp)) first and calculate the denominator tot
tot
tot (pp) from parameterization (27) evaluated at the laboratory momentum of the nuclear
cross-section ratio as given in Table 6. We adjust the model input parameter tot (hN) until
the model based total cross-section ratio in (38) equals the experimental one. At SELEX
energy we interpret the value of the parameter tot (hN) to be equal to tot (hp).
8.2.1. Results for tot (6 N) and tot ( N) using the ratio method
Results of the ratio method are presented in Table 9 together with the results from
the CH2 C method. The errors of hadronnucleon cross sections resulting from the ratio
method include both the error in the measured nuclear cross-section ratio and model
uncertainties. Model uncertainties are taken into account by adding the error of a model
cross-section ratio in quadrature to the error of the corresponding experimental crosssection ratio given in Table 6. The error in the model cross-section ratio is derived from the
discrepancy between model and measured cross sections observed for pA and A total
cross sections. Typical sizes of these discrepancies are shown in Table 8.
Further, as two different parameterizations for (d2 /dt dM 2 )|t =0 are available, we
evaluate the ratio method for both, average the results and include their difference in the
error of the mean.
Finally, we want to mention that as little data exists for 6 scattering, we insert in the
mod (6 A) for B , 0
2
2
computation of tot
6 N 6 N and (d /dt dM )|t =0 , the parameterizations
from pp-reactions.
Comparing the hadronnucleon cross sections of the ratio and the difference method, we
find that the results agree well within their errors. As final result, we average the hadron
nucleon cross-section values from all methods. These total averages are presented in the
last row of Table 9 together with a corresponding averaged laboratory momentum.
Table 8
Discrepancy between model and measured total cross sections. The measured pA cross sections are
scaled by kscale
Reaction
tot ( Be)
tot
( C)
Measured cross
section kscale
[mb]
Calculated
cross section
[mb]
Cross-section
difference
[mb]
Nominal
plab
[GeV/c]
188.7
188.8
0.1
640
234.1
231.4
2.7
590
tot (pBe)
270.2
277.0
6.8
640
tot (pC)
336.8
335.9
0.9
590
308
Table 9
The total cross sections tot (6 N) and tot ( N) resulting from all methods and their average
Method
description
tot (6 N)
[mb]
tot ( N)
[mb]
plab
[GeV/c]
difference method
ratio method, Be data
ratio method, C data
33.7 3.1
37.4 1.3
37.0 0.8
26.0 2.1
27.1 1.5
26.4 1.3
585
640
595
total average
37.0 0.7
26.6 0.9
610
8.3. Comparison to models

8.3.1. Comparisons for tot ( p)
Most of the models and parameterizations for hadronnucleon cross sections exploit the
interplay of two contributions: the Pomeron contribution, which dominates asymptotics at
high energies; and the Regge contribution, which is important at low and medium energies.
Many models (e.g., [35,36]) describe the energy dependence of total cross sections quite
well. We display in Fig. 14 experimental data from [8] and SELEX along with the
parameterization for tot ( p, s):
tot ( p, s) = 35.9 s 0.45 + 13.7 s +0.079
for plab > 10 GeV/c, tot in mb, s in GeV2 ,
(39)
from the 1996 Particle Data Group formulation [37].

We point out that so far the total cross section tot ( p) has been measured only up to
plab = 370 GeV/c [38]. Thus, the SELEX total average for tot ( N) at 610 GeV/c is the
first new measurement at higher laboratory momentum.
Fig. 14. Existing data for tot ( p) in comparison with our results and parameterization (39) of the
particle data group 1996.
309
In Fig. 14 the parameterization (39) of the Particle Data Group, which uses a Pomeron
intercept of 0.079, is overlaid to the data. Qualitative inspection of (39) suggests that it is
strongly weighted by the huge number of low energy data points and does not sufficiently
well take into account the very accurate data of [38] at high energy. Our result seems
to strengthen the trend observed in data of [38], implying a faster rise of the p cross
section with increasing energy than represented by (39). We just want to point out this
observation, which may turn out to be in conflict with the belief that the energy increase
of hadronic cross sections is universal. We do not give any quantitative estimate of the
Pomeron intercept for the p cross section. Its value is correlated to the assumed Regge
contribution at low energy and its determination requires a careful analysis of all the data.
8.3.2. Comparisons for tot (6 p)
Data on the total cross section tot (6 p) are scarce. In the past, there have been only
two hyperon-beam experiments [4,39] giving information about the behavior of tot (6 p)
in the momentum range 19136.9 GeV/c. The SELEX result for tot (6 N) provides the
first high energy data. Fig. 15 shows a compilation of data from previous experiments
together with the SELEX result. Our measurement is 2.9 mb larger than the data point at
136.9 GeV/c from [4], indicating a rise of tot (6 p) with increasing beam energy.
Overlaid on the experimental data is the prediction for tot (6 p, plab ) from H. Lipkin
(see [36]):
p 0.13
p 0.2
lab
lab
+ 13.2
tot (6 p, plab ) = 19.5
20
20
for plab > 10 GeV/c, tot in mb, plab in GeV/c.
(40)
The corresponding curve in Fig. 15 shows good agreement between our measurement and
this prediction.
It would be certainly desirable to find the Pomeron intercept for the 6 p cross section.
The lack of low energy data does not allow any reasonable estimate of the intercept.
Fig. 15. Existing data for tot (6 p) in comparison with our results and prediction (40).
310
9. Conclusions
The SELEX collaboration has measured the total cross sections tot ( Be), tot ( C),
tot ( Cu), tot ( CH2 ), tot (6 Be), tot (6 C), tot (6 Cu), tot (pBe) and tot (pC) in
a broad momentum range around 600 GeV/c using a transmission method that was adapted
to the specifics of the SELEX spectrometer. The accuracy of the results is within 0.61.5%
for Be, C and CH2 and about 17.5% for Cu.
The ratios of hadronnucleus cross sections for Be and C show that nucleus cross
sections are a about factor of 0.7 lower than corresponding protonnucleus cross sections.
Furthermore, we find that the 6 nucleus cross sections are about a factor of 0.92 smaller
than corresponding protonnucleus cross sections.
We observe that the results for tot (pBe), tot (pC), tot ( Be), tot ( C) and
tot ( Cu) join smoothly onto corresponding cross-section data at lower energy. The good
agreement of the protonnucleus and the nucleus cross sections to Glauber model
calculations with an inelastic screening correction and one adjustable parameter in the
density distribution justifies the deduction of tot (6 p) and tot ( p) from the nuclear
cross sections.
We deduced the hadronnucleon cross sections tot ( N) and tot (6 N), which we
regard as tot ( p) and tot (6 p), from our nuclear data using a CH2 C difference and
a model based ratio method. Results from the difference method have an accuracy of 8.1
9.2%, while results from the ratio method have an accuracy of 2.25.5%.
The total averages of all methods represent first measurements for tot ( p) and
tot (6 p) near 600 GeV/c. Our result for tot (6 p) shows clearly a rise of this cross
section with increasing beam energy, which agrees with the prediction of [36].
Our result for tot ( p) joins nicely onto the trend of the high energy data of [38]. As
mentioned in Section 8.3.1, the data of [38] and our result may indicate a faster increase
of the p cross section than predicted by the parameterization given by the Particle Data
Group in 1996.
This indication of a faster increase of the p cross section compared to the pp (and pp)
one can be verified only by a high statistics measurement using a beam and a hydrogen
target to avoid some systematic errors inherent to the method used in this experiment. In
our opinion a measurement of the p cross section at 600 GeV/c or higher is the only
experimentally accessible opportunity to test if the energy variation of a hadronic cross
section might be different from that for pp and (and pp) cross sections.
Acknowledgements
We are indebted to the Fermilab staff for the realization of this experiment. We wish
to thank F. Pearsall and D. Northacker for technical support. Further, we are grateful
to T. Olzanovski, V. Mallinger, U. Schwan, R. Schwan, J. Zimmer and the mechanic
workshop of the Max-Planck-Institute for nuclear physics at Heidelberg for support in
organizing, machining and measuring target materials. We thank B. Kopeliovich and
311
A.V. Tarasov for theoretical support and are very grateful to G.T. Garvey, J. Pochodzalla,
H.W. Siebert and M. Zavertyaev for various discussions and contributions to the analysis.
Last but not least it was a pleasure for us to contact H. Lipkin about model predictions.
This project was supported in part by the Bundesministerium fr Bildung, Wissenschaft,
Forschung und Technologie, the Consejo Nacional de Ciencia y Tecnologa (CONACyT),
the Conselho Nacional de Desenvolvimento Cientfico e Tecnolgico, the Fondo de Apoyo
a la Investigacin (UASLP), the Fundao de Amparo Pequisa do Estado de So Paulo
(FAPESP), the Israel Science Foundation founded by the Israel Academy of Sciences
and Humanities, the Istituto Nazionale di Fisica Nucleare (INFN), the International
Science Foundation (ISF), the National Science Foundation (Phy#9602178), NATO (grant
CR6.941058-1360/94), the Russian Academy of Science, the Russian Ministry of Science
and Technology, the Turkish Scientific and Technological Research Board (TBITAK),
the U.S. Department of Energy (DOE grant DE-FG02-91ER40664), and the U.S.Israel
Binational Science Foundation (BSF).
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]
J.L. Langland, Ph.D. Thesis, University of Iowa, 1995.

S.A. Kleinfelder et al., IEEE Trans. Nucl. Sci. 35 (1) (1988).
U. Dersch, Ph.D. Thesis, Heidelberg, 1998.
S.F. Biagi et al., Nucl. Phys. B 186 (1981) 121.
G. Bellettini et al., Nucl. Phys. 79 (1966) 609624.
A.M. Schiz et al., Phys. Rev. D 21 (1980) 30103022.
P.V.R. Murthy et al., Nucl. Phys. B 92 (1975) 269308.
Particle Data Group, C. Caso et al., Eur. Phys. J. C 3 (1998), and data on total cross sections
from computer readable files: http://pdg.lbl.gov/1998/contents_plots.html.
A.M. Schiz, Ph.D. Thesis, Yale University, 1979.
Landolt Brnstein Tables, Volume 7, Springer edition 1973.
J. Engler et al., Phys. Lett. B 32 (1970) 716719.
A. Babaev et al., Phys. Lett. B 51 (1974) 501504.
R.J. Glauber, in: Boulder Lectures, 1959, pp. 315413.
V. Franco, Phys. Rev. C 6 (1972) 748757.
V.A. Karmanov, L.A. Kondratyuk, JETP Lett. 18 (1973) 266268.
J.P. Burq et al., Nucl. Phys. B 217 (1983) 285335.
D. Gross et al., Phys. Rev. Lett. 41 (1978) 217220.
G.G. Beznogikh et al., Phys. Lett. B 39 (1972) 411413.
A.A. Vorobyov et al., Phys. Lett. B 41 (1972) 639641.
K.J. Foley et al., Phys. Rev. Lett. 19 (1967) 857859.
L.A. Fajardo et al., Phys. Rev. D 24 (1981) 4665.
P. Jenni et al., Nucl. Phys. B 129 (1977) 232252.
R.E. Breedon et al., Phys. Rev. Lett. B 216 (1989) 459465.
N. Amos et al., Phys. Rev. Lett. B 128 (1983) 343348.
U. Amaldi et al., Phys. Rev. Lett. B 66 (1977) 390394.
N. Amos et al., Nucl. Phys. B 262 (1985) 689714.
V.D. Akopin et al., Sov. J. Nucl. Phys. 25 (1977) 5155.
I.V. Amirkhanov et al., Sov. J. Nucl. Phys. 17 (1973) 636637.
K.J. Foley et al., Phys. Rev. 181 (1969) 17751793.
312
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
V.D. Apokin et al., Nucl. Phys. B 106 (1976) 413429.

J.P. Burq et al., Phys. Lett. B 109 (1982) 124127.
L.G. Dakhno, Sov. J. Nucl. Phys. 37 (1983) 590598.
M. Kazarinov et al., Sov. Phys. JETP 43 (1976) 598606.
C.W. De Jager et al., At. Data Nucl. Data Tables 14 (1974) 479508.
A. Donnachie, P.V. Landshoff, Phys. Lett. B 296 (1992) 227232.
H. Lipkin, Phys. Rev. D 11 (1975) 18271831.
Particle Data Group, R.M. Barnett et al., Phys. Rev. D 54 (1996) 191192.
A.S. Carroll et al., Phys. Lett. B 80 (1979) 423427.
J. Badier et al., Phys. Lett. B 41 (1972) 387392.

Power counting and function in NRQCD

Harald W. Griehammer 1
Nuclear Theory Group, Department of Physics, University of Washington,
Box 351 560, Seattle, WA 98195-1560, USA
Received 13 October 1998; revised 23 April 1999; accepted 27 May 1999
Abstract
A computation of the NRQCD function both in the Lorentz gauge family and in the Coulomb
gauge to one loop order endorses a velocity power counting scheme for dimensionally regularised
NRQCD. In addition to the ultrasoft scale represented by bremsstrahlung gluons and the potential
scale with Coulomb gluons and on-shell quarks, a soft rgime is identified in which energies and
momenta are of order Mv, gluons are on shell and the quark propagator becomes static. The
instantaneous gluon propagator has a non-zero vacuum polarisation only because of contributions
from this rgime, irrespective of the gauge chosen. Rules are derived which allow one to read up
from a given graph whether it is zero because of the homogeneity of dimensional regularisation.
They also apply to threshold expansion and are used to prove that ultrasoft quarks with energy and
momentum of order Mv 2 decouple from the theory. 2000 Elsevier Science B.V. All rights reserved.
PACS: 12.38.Bx; 12.39.Hg; 12.39.Jh; 11.10.Gh
Keywords: Non-relativistic QCD; Heavy quark effective theory; Effective field theory; Threshold expansion;
Renormalisation; function; Dimensional regularisation
1. Introduction
Non-relativistic QCD [1,2] takes advantage of the fact that when a heavy quark is
nearly on shell and its energy is dominated by its mass M, the resulting non-relativistic
system exhibits two small expansion parameters: the coupling constant g and the particle
velocity v. The Coulomb interaction rules the level spacing in charmonium and bottomium
because s is small enough for perturbative calculations and the relative velocity of the
quarks is v s (Mv) by virtue of the virial theorem, where the scale of g is set by the
inverse Bohr radius Mv of the system. Albeit s increases with decreasing Q2 (Mv)2 , a
window between the relativistic perturbative and the confinement rgime remains in which
both s and v is small (e.g., for bottomium, s (Mb v) 0.35). Therefore, wave functions
1 hgrie@phys.washington.edu
PII: S 0 5 5 0 - 3 2 1 3 ( 9 9 ) 0 0 3 2 5 - 9
314
H.W Griehammer / Nuclear Physics B 579 (2000) 313351
and potentials obtained by resummation of ladder diagrams involving relevant couplings as

v 0 may be used to account for bound state physics, and calculations of production cross
sections, hyperfine splittings, lifetimes, threshold properties, etc., are much facilitated. The
Abelian analogue, NRQED, to which all considerations in this article apply equally well,
has also simplified precision calculations in positronium. An incomplete list of recent
references may include [312].
and
The NRQCD Lagrangean in terms of the heavy quark (anti-quark) bi-spinors Q (Q)
gluons (D = + igA ) reads
c1 2
c2
Q D Q+
Q D 4 Q +
2M
8M 3

g 2 d1
gc3
+
Q BQ +
C Q Q Q C Q
+
2
2M
4M
g 3 e2
e1 a , a
+
F a D 2 F ,a + + LGFix ,
F F
4
480 2 M 2
LNRQCD = Q (i0 gA0 )Q +
(1.1)
where the coefficients ci , di , ei , . . . encode the ultraviolet physics: All excitations with
four-momenta of the order of M and higher are integrated out and give rise, e.g., to fourpoint interactions between quarks (d1 6= 0). The perturbative part of the coefficients can be
determined by matching NRQCD matrix elements to their QCD counterparts in the rgime
where both theories are perturbative in the coupling constant. This has been performed
to O(M 3 ) [13]. At tree level, a FoldyWouthuysen transformation gives ci = 1, e1 =
1, and loop corrections are down by powers of g, the most famous example being the
coefficient for the Fermi term related to the anomalous magnetic moment of the electron,
s
+ . Further coefficients to enter at one loop level are d1 = e2 = 1. Lorentz
c3 = 1 + 2
invariance demands c1 = c2 = 1 to all orders.
The Lagrangean (1.1) consists of infinitely many terms constrained only by the
symmetries of the theory and is non-renormalisable. Predictive power is nonetheless
established when only a finite number of terms contribute to a given order in the two
expansion parameters 2 , g and v. Besides the heavy quark mass M, the typical binding
energy and momentum scales in NRQCD are the non-relativistic kinetic energy Mv2 and
the momentum Mv of the quark, which in quarkonia appear as the energy and inverse
size of the bound state [1,2]. Since for the smallest of the three scales s (Mv 2 ) 6< 1 in
bottomium and charmonium, an expansion in g is justified only for the interactions taking
place on scales Mv and higher in the real world, but one can imagine a world in which
Mv 2 3QCD . Toponium fulfills this requirement but decays mainly weakly and very fast
so that QCD effects do not dominate.
Because the effective Lagrangean does not exhibit the non-relativistic expansion
parameter v explicitly, a power counting scheme has to be established which determines
uniquely which terms in the Lagrangean must be taken into account to render consistent
calculations and predictive power to a given accuracy in v. It is at this point that NRQCD
can serve as a toy model for nuclear physics [14] (although this grossly understates its
2 For clarity, the two will be distinguished in the following, although v , as noted above.
s
315
value 3 ): It will establish what the relevant kinematic rgimes and infrared variables are
in a theory with three (or more) separate scales, and it will demonstrate how to count
powers of the non-relativistic expansion parameter, v. Recently, velocity power counting
rules were established for a toy model [15] following Beneke and Smirnovs threshold
expansion [16], which has become very useful for understanding non-relativistic effective
theories. The relevant energy and momentum rgimes and low energy degrees of freedom
were found for dimensionally regularised non-relativistic theories. This article presents the
extension to NRQCD and as an application the calculation of the NRQCD function. It
will also be argued that the diagrammatic approach advocated here sheds a new light on
threshold expansion [16] and helps simplify calculations.
Both from a conceptual and a technical point of view the calculation of the NRQCD
function proves interesting. NRQCD is a well-defined field theory of quarks and gluons.
One can therefore investigate its renormalisation group equations under the assumption
that perturbation theory is applicable at all scales. For example, the running and mixing
of the dimension six operators has been investigated to one loop order by Bauer and
Manohar [17]. On the other hand, as NRQCD is the low energy limit of QCD, it must
reproduce the QCD function with NF light quarks below the scale M, but this has not
been demonstrated so far. The problem with the power counting developed until now [14,
18,19] was that gluons mediating the Coulomb interaction (potential gluons) seemed to
have a vanishing vacuum polarisation and hence a function which was neither gauge
invariant nor agreed with the QCD result, while the coupling of gluons on a much lower
scale which mediate bremsstrahlung processes ran as expected. The apparent impasse
which results is resolved in this article.
Straightforward as the computation may thus seem with its outcome already anticipated,
the prime goal of this article is not the result for the NRQCD function; the objective
is rather to shed more light on power counting in NRQCD and to provide a comparison
between calculations in the effective and the full theory. It will show that the Lorentz
gauge family (and indeed any standard gauge) is a legitimate gauge choice in NRQCD,
not only the Coulomb gauge. It will prove that the recently discussed soft rgime [15,16,
20] with quarks and gluons of four-momenta of order Mv is indispensable for NRQCD
to describe the correct infrared limit of QCD. It will also test the validity of the power
counting proposed. In the Coulomb gauge, it will demonstrate that the renormalisation of
the quark propagator and of the leading order quark gluon vertex are trivial, so that for the
function in this gauge, only a computation of the gluon vacuum polarisation is necessary.
From a technical point of view, dimensional regularisation for non-covariant loop integrals
proves to be a simple regulator choice. It allows to develop a set of simple diagrammatic
rules to determine which graphs yield non-zero contributions in perturbation theory. This
reduces the computational effort considerably as compared to other regularisation schemes
in which power divergences are present like, e.g., when using a lattice cut-off. Because
of the association of threshold expansion and NRQCD, such rules can be applied to
3 The most striking difference to nuclear physics is that NRQCD has no exceptionally large scattering length
to accommodate.
316
the former, too. Finally, the calculation of the NRQCD function presented here is
especially in the Feynman gauge simpler than its QCD counterpart and shows some
pedagogically intriguing aspects.
The article is organised as follows: In Section 2, the velocity power counting of NRQCD
is presented. The relevant rgimes of NRQCD are identified (Section 2.1), extending the
formalism of Luke and Savage [18] by the soft rgime of Beneke and Smirnov [16].
The section then proposes the rescaling rules necessary for a Lagrangean with manifest
velocity power counting (Section 2.2) and gives the vertex (Section 2.3) and loop velocity
power counting rules (Section 2.4). Gauge invariance of the power counting is addressed in
Section 2.5. The calculation of the NRQCD function in the Lorentz gauges in Section 3
starts by highlighting the intimate relation between threshold expansion and the proposed
NRQCD power counting, which will additionally be summarised in the Conclusions.
Diagrammatic rules developed in Section 3.2 help to facilitate the calculation considerably
and are used in Section 3.3 to prove the absence of ultrasoft heavy quarks. As a tilly, the
function is computed in the Coulomb gauge in Section 3.7. Summary and outlook conclude
the article, together with an appendix containing exemplary calculations in non-covariant
dimensional regularisation.
2. Velocity power counting

2.1. Rgimes of NRQCD
The first power counting rules for NRQCD were derived by Lepage et al. [21] in the
Coulomb gauge using the consistency of the equations of motion and a momentum cut-off.
Only Coulomb gluons with typical momenta of order Mv were considered, taking into
account retardation, but neglecting bremsstrahlung effects. The power counting also held
to all orders only after all power divergences were subtracted.
Simpler counting rules were proposed by Luke and Manohar for Coulomb interactions [14], and by Grinstein and Rothstein for bremsstrahlung processes [19]. Luke and
Savage [18] united these schemes using dimensional regularisation, so that the tree level
power counting is automatically preserved to all orders in perturbation theory up to logarithmic corrections. They also extended the formalism to include the Lorentz gauges.
Labelle [22] proposed a similar scheme in time ordered perturbation theory. Beneke and
Smirnov [16] observed that the collinear divergence of the two gluon exchange contribution
to Coulomb scattering between non-relativistic particles near threshold is not reproduced
in this version of NRQCD because the soft low energy rgime is not taken into account. A
recent article [15] has shown that an extension of the work in [14,18,19] using dimensional
regularisation, as motivated from threshold expansion [16], resolves this conflict in a toy
model. The complete power counting scheme of NRQCD is presented briefly in the following (see also [20]). In Section 3.1, the intimate relation between NRQCD as developed
in recent years [14,15,18,19] and threshold expansion is addressed in more detail.
The NRQCD propagators are read up from the Lagrangean (1.1) as
Q:
i Num
T
p2
2M
+ i
A :
i Num
,
k 2 + i
317
(2.1)
p
+ is the kinetic energy of the quark. Num are numerators
where T = p0 M = 2M
containing the appropriate colour, Dirac and flavour indices, all of which are unimportant
for the considerations in this section and will be suppressed throughout this article. Gauge
dependence of the gluon propagator for the power counting is addressed in Section 2.5.
Cuts and poles in scattering amplitudes close to threshold stem from saddle points of
the loop integrand, corresponding to bound states and on-shell propagation of particles
in intermediate states. They give rise to infrared divergences and in general dominate
contributions to scattering amplitudes. With the two low energy scales at hand, and energies
and momenta being of either scale, three rgimes are identified in which either the quark
or the gluon in (2.1) is on shell, as noted by Beneke and Smirnov [16]:
soft rgime: A
s:
potential rgime: Qp :
ultrasoft rgime: A
u:
k0 |k| Mv,
T Mv 2 , |p| Mv,
k0 |k| Mv 2 .
(2.2)
Note that the scale M has been integrated out when deriving the non-relativistic
Lagrangean (1.1), so that a hard rgime with on shell gluons and quarks of fourmomentum of order M cannot be considered.
What is the particle content of each of the three rgimes? Ultrasoft gluons Au are emitted
as bremsstrahlung or from excited states in the bound system, and hence are physical.
Soft gluons As do not describe bremsstrahlung: Because in- and outgoing quarks Qp are
close to their mass shell, they have an energy of order Mv 2 . Therefore, overall energy
conservation forbids all processes with outgoing soft gluons but without ingoing ones, and
vice versa, as their energy is of order Mv. Finally, gluons which change the quark momenta
but keep them close to their mass shell relate the (instantaneous) Coulomb interaction:
A
p:
k0 Mv 2 ,
|k| Mv.
(2.3)
When a soft gluon As
couples to a potential quark Qp , the outgoing quark is far off its mass
shell and carries energy and momentum of order Mv. Therefore, consistency requires the
existence of quarks in the soft rgime as well,
Qs :
T |p| Mv.
(2.4)
With these five fields Qs , Qp , As , Ap , Au representing quarks and gluons in the three
different non-relativistic rgimes, soft, potential and ultrasoft, NRQCD becomes selfconsistent. These fields are the infrared-relevant degrees of freedom representing one and
the same non-relativistic particle in the respective kinematic rgimes and came naturally by
identifying all possible particle poles in the non-relativistic propagators. Their interactions
are fixed by the non-relativistic Lagrangean (1.1). Section 3.3 will prove that a hypothetical
ultrasoft quark, created, e.g., by the radiation of a potential gluon off a potential quark,
decouples completely from the theory. On the other hand, Section 3 will demonstrate that
in order to obtain the correct result for the NRQCD function, all fields listed above in
318
the three rgimes have to be accounted for. Therefore, the particle content presented above
is not only consistent but both minimal and complete.
In order to guarantee that there is no overlap between interactions and particles in different rgimes, the regularisation scheme must finally be chosen such that the expansion
around one saddle point in the loop integral as performed above does not obtain any contribution from other rgimes, represented by other saddle points in the loop integrals. One
might use an energy and momentum cut-off separating the soft from the potential, and
the potential from the ultrasoft rgime, but the integrals encountered can in general not
be performed analytically and in closed form. Furthermore, by introducing another, artificial scale, cut-off regularisation jeopardises power counting and symmetries (gauge,
chiral, . . . ). Power divergences occur when the (unphysical) cut-off is removed in intermediate steps but not in the final, physical result, complicating the computation. In contradistinction, using dimensional regularisation after the saddle point expansion preserves
power counting because its homogeneity guarantees that contributions from different saddle points and rgimes do not overlap. (A simple example is given in Ref. [15].) Homogeneity will also be essential when developing diagrammatic rules to classify graphs as
zero in Section 3.2 and when showing the decoupling of the ultrasoft quark in Section 3.3.
2.2. Rescaling rules and propagators
In order to establish explicit velocity power counting in the NRQCD Lagrangean,
one rescales the space-time coordinates such that typical momenta in either rgime are
dimensionless, as proposed by Luke and Manohar [14] for the potential rgime and by
Grinstein and Rothstein [19] for the ultrasoft one:
soft: t = (Mv)1 Ts ,
1
potential: t = Mv 2 Tu ,
1
ultrasoft: t = Mv 2 Tu ,
x = (Mv)1 Xs ,
x = (Mv)1 Xs ,
1
x = Mv 2 Xu .
(2.5)
For the propagator terms in the NRQCD Lagrangean to be normalised as order v 0 , one sets
for the representatives of the gluons in the three rgimes [15,20]
soft: A
s (x, t) = (Mv)As (X s , Ts ),

3/2
A (X s , Tu ),
potential: A
p (x, t) = Mv
p
2
ultrasoft: Au (x, t) = Mv Au (Xu , Tu ),
(2.6)
and for the quark representatives

soft: Qs (x, t) = (Mv)3/2 Qs (X s , Ts ),
potential: Qp (x, t) = (Mv)3/2 Qp (X s , Tu ).
The rescaled free quark Lagrangean reads then

v 2
3
soft: d Xs dTs Qs i0 + Qs ,
2

1 2
3
potential: d Xs dTu Qp i0 + Qp .
2
(2.7)
(2.8)
(2.9)
319
Here, as in the following, the positions of the fields have been left out whenever they
coincide with the rescaled variables of the volume element. Derivatives are to be taken
with respect to the rescaled variables of the volume element unless otherwise stated. In
order to maintain velocity power counting, corrections of order v or higher must be treated
as insertions, so that one reads up the (unrescaled) quark propagators and insertion as
soft:
Qs :
potential:
Qp :
i
,
T + i
i
T
p2
2M
+ i
p2
= P.C.(v 1 ),
2M
(2.10)
(2.11)
p
Mv 2 allows one to expand the
The soft quark becomes static because T Mv 2M
non-relativistic propagator (2.1) and to treat the momentum term as an insertion.
Throughout this article, the symbol P.C.(v n ) denotes at which order in the velocity
power counting a certain term or diagram contributes. In the approach presented, each
line in a loop diagram counts as P.C.(v 0 ) because the strength of the propagator has been
set to unity in the rescaled Lagrangean (2.8), (2.9). The rescaled Lagrangean measures the
strength of the insertion relative to the propagator. Therefore, the soft quark insertion (2.10)
p2
p2
which scales like v 2 , but as 2M
/T = P.C.(v). In contradistinction to
does not count as 2M
the approach in [16], one will therefore not obtain the absolute order in v at which a given
graph contributes, but the relative order in v between graphs or vertices is read off more
easily and asserted correctly.
The gauge fixing term was included in the NRQCD Lagrangean (1.1) since the
decomposition of the Lagrangean into a free and an interacting part is gauge dependent.
Because of the difference between canonical and physical momentum, it is important to
specify the gauge before identifying to which order in v a certain rgime in the Lagrangean
contributes, as will be seen shortly. In the following, the Feynman rules for the Lorentz
gauges and for the Coulomb gauge are derived explicitly. Section 2.5 will comment on the
gauge invariance of the procedure.
The rescaled free gluon Lagrangean in the Lorentz gauges reads

1
1 2
3
(2.12)
As ,
soft: d Xs dTs As g 1
2

1
2 2
2
potential: d3 Xs dTu A
p g v 0
2

1
(2.13)
(v0 0 + i i )(v0 0 + i i ) Au ,
1

1
1
2
Au ,
(2.14)
ultrasoft: d3 Xu dTu A
u g 1
2
while in the Coulomb gauge

soft:

1
d3 Xs dTs Ai,s 2 02 ij i j Aj,s ,
2
(2.15)
320

1
A0,p 2 A0,p + Ai,p 2 ij i j v 2 02 ij Aj,p ,
2

1
3
d Xs dTs Ai,u 2 02 ij i j Aj,u .
2
d3 Xs dTu
potential:
ultrasoft:
(2.16)
(2.17)
ij
The (unrescaled) Coulomb and Lorentz gauge propagators are therefore given as (tr (k) =
i j
k k
))
ij k k2 , G (k) = (g (1 ) k 2+i
k
Coulomb gauge Lorentz gauges

ij
soft:
potential:
A
s:
Ap,0 :
itr (k)
k 2 + i
i
k 2 + i
ij
Ap :
A
u:
itr (k)
,
k 2 + i

i ij + (1 )
k 2 + i
ij
ultrasoft:
iG (k)
,
k 2 + i
i
itr (k)
k 2 + i
ki kj
k 2 +i
k 2 + i
iG (k)
.
k 2 + i
,
(2.18)
The potential gluon becomes instantaneous in both gauges as expected for the particle
mediating the Coulomb interaction. Insertions are necessary only in the potential rgime:
Coulomb gauge
Lorentz gauges
k02
= P.C.(v 2 ),
none
ik02 ij = P.C.(v 2 )
none
ik02 ij = P.C.(v 2 ),

1
i 1
ki k0 = P.C.(v).
(2.19)
The Lorentz gauge propagators and insertions for the potential and ultrasoft rgimes were
first given in [18]. Especially for the Feynman gauge = 1 (G (k) = g ), Lorentz and
Coulomb gauges in the potential rgime differ only by insertions, i.e., at higher order in v.
As seen from (2.15)(2.17), the choice of the Coulomb gauge makes A0 instantaneous
to all orders in v, and hence it contributes in the potential rgime, only. Since in this gauge,
A0 solely mediates the instantaneous Coulomb potential (physical fields are transverse
by virtue of Gau law), this result was to be expected. The field Ap is associated with
retardation effects like spinorbit coupling and the Darwin term in (1.1). In the Coulomb
gauge, the advantages of A0 only contributing in the potential rgime and of its propagator
having no insertions or admixtures with the vector components of the gauge field, are
balanced by the fact that a more lengthy and cumbersome renormalisation seems necessary.
The calculation of the NRQCD function will prove that the Lorentz gauges are a
legitimate gauge choice in NRQCD, although the number of diagrams appears to be larger
because of a larger number of vertices, as will be seen now.
321
2.3. Vertex rules

By experience, particles in the various rgimes couple: On-shell (potential) quarks
radiate bremsstrahlung (ultrasoft) gluons. In general, one must allow all couplings between
the various rgimes which obey scale conservation: Both energies and momenta must be
conserved within each rgime to the order in v one works. This will exclude for example
the coupling of two potential quarks (T Mv 2 ) to one soft gluon (q0 Mv), but not
to two soft gluons via the Q A AQ term of the Lagrangean (1.1). For a more rigorous
derivation of this rule see Section 3.3.
As an example, consider a bremsstrahlung-like process: the radiation of a soft scalar
gluon As ,0 off a soft quark, resulting in a potential quark with four momentum (Tp0 , p 0 ).
The rescaled interaction Lagrangean with its Hermitean conjugate reads

(2.20)
d3 Xs dTs gv 0 Qs (Xs , Ts )As,0 (Xs , Ts )Qp (Xs , vTs ) + H.c. .
Note that the scaling rgime of the volume element is set by the particle with the highest
momentum and energy. Therefore, one obtains the power counting of a given vertex in a
specific rgime. As the external quarks are on shell, i.e., in the potential rgime, the power
counting rules of subgraphs whose scaling rules have been determined in the soft rgime
have to be transferred to the potential rgime. This step is postponed to the next section.
The strength of the example vertex above is read up as v 0 in the soft rgime. This is clearly
also the strength of the Hermitean conjugate vertex.
The interaction Lagrangean is non-local in the rescaled variables, and in order to
maintain velocity power counting, Qp (Xs , vTs ) has to be expanded about Qp (Xs , 0) in
powers of v. The Feynman rule for this vertex is after inverting the rescaling
ig(2)4 (3) (p p 0 + q)

0
(T + q0 ) = P.C.(ev ).
exp Tp
(T + q0 )
(2.21)
The Hermitean conjugate vertex is built analogously. As the potential quark has a much
smaller energy than either of the soft particles, it can by the uncertainty relation
not resolve the precise time at which the soft quark emits or absorbs the soft gluon. This
temporal multipole expansion comes technically from the different scaling of x and t in
the three rgimes. The multipole expansion symbolised in the power counting by ev corresponds term by term to an expansion in v and should be truncated at the desired order.
In general, the coupling between particles of different rgimes will not be point-like but
contain multipole expansions for the particle belonging to the weaker kinematic rgime.
That the coupling of potential quarks to ultrasoft gluons requires a momentum multipole
expansion as in atomic physics has been observed by Grinstein and Rothstein [19], and by
Labelle [22].
Amongst the fields introduced, six scale conserving interactions are allowed within and
between the various rgimes for any coupling of one gluon to two quarks. Because only in
the Coulomb gauge, A0 is a field in the potential rgime only, its propagator always being
322
Table 1
Velocity power counting and vertices for the interaction Lagrangean gQ A0 Q. In the Coulomb
gauge, only the first two diagrams exist. The last line indicates the field for which an energy or
momentum multipole expansion has to be performed
instantaneous, only the first two interactions exist for the scalar coupling gQ A0 Q of
Table 1 in this gauge choice. The v counting for the lowest order quark-gluon interactions
from this vertex is presented in Table 1.
Note that although both describing interactions with physical gluons soft and ultrasoft couplings occur at different orders in v. On the level of the vertex rules, an overlap
of different rgimes resulting in double counting is prevented by the fact that in addition to
most of the propagators, all vertices are distinct because of different multipole expansion
rules.
This approach may be compared to power counting in loop diagrams as proposed in
Beneke and Smirnovs threshold expansion [16] where the strength of all scalar gluon
interactions is v 0 because the scalar vertex Q A0 Q of Table 1 does not contain derivatives.
Here, one can easily see that the Coulomb interaction is the only relevant coupling as v 0
and that it scales like v 1/2 . This follows immediately from the rescaling rules proposed.
Also, the suppression of bremsstrahlung processes relative to the Coulomb interaction is
evident from Table 1. In threshold expansion, these features are established by considering
scattering processes with an arbitrary number of loops.
Velocity power counting for other vertices is again obtained by rescaling and multipole
expansion. For later reference, the counting rules for the coupling of one and two gluons,
either minimally or via the Fermi term, and the rules for the three gluon interaction are
displayed in Tables 2, 3 and 4. In the second, couplings between on shell gluons in the
same rgime are all of order v 0 = c = 1 as expected for relativistic particles.
ig
Q AQ, the derivative acts on both the gluon and
In the minimal coupling term M
the quark field. Because the quark is either soft or potential, its derivative from Q A Q
is rescaled as (Mv) s . The same holds when the derivative acts on a soft or potential
gluon. But both for the one gluon part of the Fermi interaction Q BQ and for the term
Q ( A)Q of the minimal coupling, the derivative acts on an ultrasoft gluon field and must
hence be rescaled as (Mv 2 ) u . That this part of the vertex coupling is one power in v
weaker than the part where the derivative acts on the quark, is recorded in parentheses in
Table 2. The v rules for the coupling of one gluon to the quark via the Fermi term are
identical to those sub-dominant contributions to the minimal coupling vertex for ultrasoft
gluons, and to the dominant contributions in the other cases. A similar provision is made
in Table 3, but is not necessary in Table 4.
323
Table 2
ig
Velocity power counting and vertices for the interaction M
Q AQ. When the spatial
derivative acts on an ultrasoft gluon field, the power counting in parentheses is the sub-dominant
contribution as explained in the text. It coincides with the power counting for the one gluon
component of the Fermi vertex
Table 3
Velocity power counting and vertices for the gluonic interaction Lagrangean g2 f abc
( Aa Aa )Ab, Ac, . Sub-dominant contributions in parentheses
Table 4
Velocity power counting and vertices for the coupling of two gluons to a quark from the minimal
coupling or the Fermi term in the NRQCD Lagrangean (1.1)
2.4. Loop rules

As hinted upon above, the velocity power counting is not yet complete. The propagators
were constructed such that they count as order v 0 in each rgime in a Feynman diagram. As
one sees from the volume element used in (2.20), the vertex rules for the soft rgime count
powers of v with respect to the soft rgime. There, one hence retrieves the velocity power
counting of Heavy Quark Effective Theory [23,24] (HQET), in which the interactions
between one heavy (and hence static) and one or several light quarks are described.
HQET counts inverse powers of mass in the Lagrangean, but because in the soft rgime
324
Fig. 1. Power counting with soft loops. The loops in the second and third diagram obtain an inverse
power of v, the last diagram of v 2 in addition to the power counting following from the vertex rules.
Mv const, the two approaches are actually equivalent. Now, HQET becomes a subset
of NRQCD, complemented by interactions between soft (HQET) and potential or ultrasoft
particles. This has already been used by Manohar [13] to facilitate and unify the matching
of the NRQCD/HQET Lagrangean to QCD.
In NRQCD with on-shell quarks as initial and final states, the soft rgime can occur only
inside loops. Let us define as a soft blob each connected graph containing (soft) loops
which is obtained when all potential and ultrasoft lines are cut. If the diagram contains
even one soft particle, scale conservation ensures that there is at least one loop which
consists of only soft particles, be they quarks or gluons, and that it is part of the soft blob.
Inside the soft blob, the power counting for the vertices is performed in the soft rgime
and has therefore to be transferred to the potential rgime. Since soft loop momenta scale
like [d4 ks ] v 4 while potential ones like [d4kp ] v 5 , each soft blob is enhanced by an
additional factor 1/v.
Consider for example the graphs of Fig. 1: Using the Lorentz gauges, vertex power
counting gives that the leading contribution is from the exchange of two potential gluons,
coupled via Q A0 Q. There are four such vertices, so the diagram is O(g 4 ) P.C.(v 2 ) from
Table 1. The next two diagrams are O(g 6 ) P.C.(v 0 ) and O(g 2 ) P.C.(v 1 ) from the vertex
power counting in the soft rgime, but another factor 1/v must be included because there
is one soft blob in each diagram. This way, the v counting of the soft rgime is moved to
the potential one. The intermediate couplings in the second diagram take place in the soft
rgime and hence are counted in that rgime. After cutting all potential and ultrasoft lines
in the last diagram, two soft blobs are separated by the propagation of two potential quarks.
The graph is O(g 14 ) P.C.(v 0 ) from the vertices, and the loop counting gives a factor 1/v2 .
Each soft blob contributes at least four powers of g, but only one inverse power of v g 2 .
Power counting is preserved. These velocity power counting rules in loops are also verified
in explicit calculations of exemplary graphs, [15] and Appendix A.2.
There is no similar rule for ultrasoft loops: In the absence of ultrasoft quarks (see
Section 3.3), the internal ultrasoft gluon couples ultimately to a particle in the potential
or soft rgime. Those vertices are automatically counted in this stronger rgime, while
couplings between ultrasoft particles are counted in the weakest rgime. No ultrasoft
blobs can therefore be isolated by cutting all potential and soft lines. It is hence ultimately
scale conservation which forbids a non-trivial loop counting rule for the ultrasoft rgime.
Irrespective of the gauge chosen, there is only one relevant quark-gluon coupling at
tree level in the renormalisation group approach (i.e., only one which dominates at zero
325
velocity): As expected, it is the Qp Ap,0 Qp coupling providing the binding (Table 1). The
potential gluon ladders must therefore be resummed to all orders to yield the 1/r Coulomb
potential. Diagrams higher order in v are corrections. In the Coulomb gauge, all other
couplings and insertions are irrelevant, while in general, there are three marginal couplings:
Qp Au,0 Qp , Qs As,0 Qs and Qs As,0Qp . Because of the additional factor v 1 per soft blob,
graphs containing the latter two couplings can indeed be relevant and contribute as v 0
(e.g., the second graph in Fig. 1). Eventually, retardation effects from Ap will become
weaker than contributions from the soft rgime.
One finally turns to the inclusion of other relativistic particles. In the same way
as NRQCD replaces the physical gluon with one representative per rgime, any light
(relativistic) particle has to be tripled. There are therefore three ghost fields s , p , u with
the same rescaling rules as the gluon fields (2.6). Fermions require more thought: In the real
world, the kinetic energy of the b quark in bottomium is compatible to the strange quark
mass, Mb v 2 ms . For the sake of simplicity, this article assumes all light particles to have
masses very much smaller than any other scale, mq Mv 2 , so that the relativistic particles
can be treated as massless to lowest order and the denominators of the light particle
propagators are identical to the ones of the gluon in the respective rgimes. The Dirac
spinors representing light quarks scale in the three rgimes as s (Mv)3/2 p , u
(Mv 2 )3/2 , i.e., similar to the heavy quark but including its ultrasoft counterpart (see also
Section 3.3). The number of vertices per term in the rescaled interaction Lagrangean
increases as shown above. Quark-ghost couplings and heavy-light quark couplings will
appear in the Lagrangean (1.1) at O(g 4 ) to restore unitarity and can be obtained by
matching NRQCD to QCD.
With rescaling, multipole expansion and loop counting, the velocity power counting
rules are established. To witness, the rescaling rules (2.5)(2.7) provide an efficient and
well-defined way to arrive at an NRQCD power counting: After rescaling, one reads up the
order in v at which any term in the Lagrangean contributes in each of the three kinematic
rgimes and performs the appropriate multipole expansions. This establishes the Feynman
rules for NRQCD and HQET simultaneously, and classifies the strength of the vertices
in the rgime of the particle with highest energy and momentum. Finally, the rescaling is
inverted, introducing one unrescaled gluon and quark field as the infrared-relevant degrees
of freedom for each kinematic rgime. To obtain the power counting for a certain graph,
loop counting is taken into account to transfer the strength of a soft blob into the potential
rgime. Computations are performed most naturally in the original, dimensionful variables.
The rescaling rules are only needed to establish the power counting, but in order to maintain
it, one is of course not allowed to resum the multipole expansions in the unrescaled
variables. It is also interesting to note that there is no choice but to assign one and the same
coupling strength g to each interaction. Different couplings for one vertex in different
rgimes are not allowed. This is to be expected, as the fields in the various rgimes are
representatives of one and the same non-relativistic particle, whose interactions are fixed
by the non-relativistic Lagrangean (1.1).
326
2.5. Gauge invariance

The Coulomb gauge A = 0 is a natural choice in NRQED because in it, static charges
do not radiate. In NRQCD, gluons carry colour and hence the Coulomb gauge does not
have this advantage. Luke and Savage [18] showed how to establish explicit velocity power
counting in the Lorentz gauges of NRQCD, too. The set of gauges applicable to NRQCD
can be extended further: The classification of the three kinematic rgimes (2.2) itself
relied only on the typical excitation energy and momentum, and hence on gauge invariant
quantities, and on the form of the denominator in the propagators (2.1) which is unchanged
in any order of perturbation theory. 4 The perturbative quark propagator is gauge
independent. Gauge fixing will introduce gauge dependent denominators multiplying the
gauge independent denominators in the perturbative gluon propagators. As an example that
this will in general not change the identification of the soft and ultrasoft rgimes from poles
in the gluon propagators, consider the generalised axial gauge (n2 = 1, arbitrary), in
which

i
k n + n k 1 + k 2
+
g
.
(2.22)
A :
kn
k 2 + i
(n k)2
The additional denominators n k introduce no new combinations of the two infrared scales
Mv and Mv 2 for which the gluon propagator has a pole. Therefore, the decomposition into
the three rgimes (2.2) remains unchanged, as do the rescaling properties of the fields and
interactions, (2.5)(2.7) and Tables 1 to 4.
The standard gauges (axial, Weyl, Lorentz, Coulomb) will therefore all show the same
power counting and vertex rules quoted above. Details of the gluon propagator and its
insertions look different in different rgimes and gauges, and some gauges will not exhibit
certain vertices, insertions and representatives, e.g., the Coulomb gauge is unique in having
A0 contribute only in the potential rgime. Only when the gauge dependent denominator
introduces a new rgime has the power counting to be modified. This is for example the
case in a gauge with denominator k04 M 2 k 2 , in which an exceptional rgime (k0 Mv,
|k| Mv 2 ) enters. Rescaling is then performed including the new rgime.
As is well known, the Weyl gauge A0 = 0 wants a constraint quantisation: Gau law
D E = gQ Q, generating local gauge transformations, is the equation of motion derived
by varying the Lagrangean with respect to A0 . When A0 = 0, Gau law will not be
recovered as an equation of motion. In order to restore local gauge invariance, a projector
onto states obeying Gau law has therefore to be inserted into the path integral. Resolving
Gau law explicitly, the Lagrangean of the Coulomb interaction has the structure
Z
(2.23)
LCoulomb = d3 y g 2 Q Q(x)G(x y)Q Q(y),
where G(x y) is an appropriate instantaneous Greens function to Gau law with
1
. Using the rescaling rules (2.5)(2.7), one derives
dimension [mass]1 , e.g., in QED |xy|
4 Non-perturbatively, the propagators are not of the form (2.1) because of confinement and the absence of
coloured states, so that the power counting presented here may break down.
327
the Coulomb interaction between two quarks to be P.C.(v1 ) as in the other gauges.
Without constraint quantisation, the longitudinal component of the vector gauge field
mediating the Coulomb interaction in Weyl gauge QED, Al , has a static propagator i/k02 .
3. The NRQCD function
This chapter presents the computation of the perturbative part of the NRQCD function
to order g 3 , v 0 in the MS scheme using the vertex and rescaling rules derived in the
previous section. Initially, the Lorentz gauges are not only chosen because they allow a
less cumbersome renormalisation than, e.g., the Coulomb gauge, but also to demonstrate
that in the final result, the gauge parameter drops out. The SlavnovTaylor identities will
be shown to be fulfilled, and the Lorentz gauges are a legitimate gauge choice in NRQCD.
As a by-product, Section 3.7 will calculate the NRQCD function to lowest order in the
Coulomb gauge.
By construction, NRQCD and QCD must agree in the infrared limit, and especially in the
structure of collinear (infrared) divergences. Matching proved that both the soft quark and
the soft gluon are indispensable to reproduce the correct structure of collinear divergences
in a toy model given by Beneke and Smirnov [16] and confirmed the proposed counting
rules [15]. The calculation of the NRQCD function will manifest the relevance of the
soft rgime beyond infrared divergences, and it will endorse the power counting further.
Deriving the lowest order function in the following, the relation between this power
counting and threshold expansion [16] will be exemplified, and rules will be developed
which allow one to determine from the structure of a diagram whether it is zero or not to
all orders in v.
NRQCD with NF light quarks must reproduce the function of QCD below the scale M,

g3
2
11
N NF
(3.1)
QCD = R 2
(4) 3
3
to lowest order for the gauge group SU(N ) and renormalised coupling gR . This means
especially that the renormalised coupling strengths of all interactions steming from
expanding the same term in the Lagrangean in the various rgimes are the same, except
that they have to be taken at different scales: For the interaction Lagrangean gQ A0 Q,
the six vertices of Table 1 all couple with the same unrenormalised strength. Although the
number of vertices is increased in the approach presented here, the number of independent
couplings is not. Indeed, one major result of this chapter will be that renormalisation will
not destroy this symmetry because the renormalisation constants agree in all rgimes.
One can obtain the function from the ghost-gluon coupling, and the ghost and gluon
self energy. Because the relativistic sector of NRQCD is identical to the one of QCD as seen
in Section 2.3, the calculation proceeds in that case like in QCD and yields the same result.
In this article, the function is calculated by studying the heavy quark sector, namely the
renormalised coupling QR A0,R QR between the heavy quark and the scalar gluon. This
term yields immediately the coupling renormalisation. All other vertices are suppressed by
at least one power of v.
328
Because the integrals to be performed are not Lorentz invariant, standard formulae for
dimensional regularisation often do not apply. Therefore, Ref. [15] used split dimensional
regularisation, which was introduced by Leibbrandt and Williams [25] to cure the problems
arising from pinch singularities in non-covariant gauges. It treats the temporal and spatial
components of the loop integrations on an equal footing by regularising the energy and
momentum integration separately,
Z
Z
d k0 dd k
dd k
=
, 1, d 4
d
(2)
(2) (2)d
[26, Chapter 4.1]. Useful formulae for the computation of the function are given in
Appendix A.1.
3.1. NRQCD and threshold expansion: potential quark self-energy
Threshold expansion [16] and NRQCD make use of very similar basic techniques but
different formulations, as this section outlines. In order to clarify the relation, consider the
quark self-energy diagram lowest order in g which couples the scalar gluon and the quark
in NRQCD. To facilitate the presentation initially, the Feynman gauge = 1 is chosen in
which scalar and vector gauge fields do not mix when propagating. Without the scaling
rules for the various rgimes, one obtains
ip (T , p) = (ig) CF
2
i
dd q
i
,
2
d
2
2
(p+q)
(2) q0 q + i T + q0
2M + i
(3.2)
with CF ij = (t a t a )ij = N2N1 ij the Casimir operator of the fundamental representation

of SU(N ). Threshold expansion identifies the loop momentum of QCD to belong to a
hard rgime q M or to either of the three rgimes (2.2) and expands the integrand
about the various saddle points, i.e., about the values of the loop-momentum q where
poles occur, like the NRQCD classification in Section 2.1. The hard rgime subsumes the
relativistic effects contained in the coefficients ci , di , ei of the NRQCD Lagrangean (1.1)
and hence provides the matching between NRQCD and QCD [16]. In deriving the NRQCD
Lagrangean (1.1), hard momenta have been integrated out, so that the ultraviolet behaviour
of the above loop integral is arbitrary and it is not necessary to take the poles at q M
in (3.2) into account. Indeed, to be consistent, they should rather be discarded.
Let the incident quark be on shell, i.e., have a four-momentum (Tp , p) in the potential
rgime. The integral decomposes then into approximations about three saddle points: When
q is soft, the gluon propagator has a pole, and one can expand the quark propagator in
2
powers of Tp /q0 v and (p+q)
2M /q0 v. The quark becomes static:

i
i
(p + q)2 i
i
+
i Tp
+
(3.3)
2
q0 q0
2M
q0
q0 + Tp (p+q)
2
2M

H ip (T , p)soft
Z
2
= (ig) CF
n
(p+q)2
X
i
i
dd q
2M Tp
.
(2)d q02 q 2 + i q0 + i
q0 + i
n=0
329
(3.4)
The NRQCD power counting proceeds on the level of the Lagrangean instead of the
Feynman diagrams with corresponding diagrams for loop momenta in the soft rgime,
Z
(ig)2 CF
+ :
(3.5)

2
i
i
i
i
(p + q)
+ iTp
+ ,
1i
(2)d q02 q 2 + i q0 + i
2M q0 + i
q0 + i
dd q
and recovers order by order in v the result of threshold expansion. The intermediate
soft quark is static, and the higher order terms in threshold expansion are interpreted
as insertions into the soft quark propagator or as resulting from the energy multipole
expansion at the Qs As Qp vertex. In fact, using the equations of motion, a temporal

multipole expansion may be rewritten such that the energy becomes conserved at the
vertex. Now, both soft and potential or ultrasoft energies are present in the propagators,
making it necessary to expand it in ultrasoft and potential energies. An example would be
to restate the vertex (2.21) as
:

ig(2)4 (T Tp0 + q0 ) (3) p p 0 + q ,
(3.6)
and the soft propagator to contain insertions P.C.(v) for potential energies Tp0
:

X
Tp0 n
i
.
q0 + i
q0
(3.7)
n=0
The same can be shown for the momentum-non-conserving vertices, too.

When the loop momentum q is potential, threshold expansion picks the pole of the quark
propagator, and the gluon propagator is expanded as if q0 |q|, corresponding to NRQCD
diagrams with insertions in the gluon propagator:
+
+ :
(3.8)
Z
2 n
d
X

q0
i
i
d q
.
ip (T , p)potential = (ig)2 CF
(2)d q 2 + i
q 2 T + q0 (p+q)2 + i
n=0
2M
The correspondence is again complete for ultrasoft q, and the higher order terms of
threshold expansion come from the spatial multipole expansion in NRQCD:
330
:

ip (T , p)ultrasoft = (ig)2 CF
(3.9)
dd q
(2)d
q02
i
i
2
q + i T + q0

2pq+q 2
X
p2
2M
2M
n=0
T + q0
p2
2M
+ i
+ i
n
.
Threshold expansion and NRQCD concur indeed in all diagrams to be calculated for the
function. Further comparison of the two approaches is postponed to the conclusions.
3.2. Some diagrammatic rules
In general, the number of possible diagrams in power counted NRQCD is considerably
larger than in the original theory because there are at least six vertices per interaction
allowed by scale conservation, cf. Tables 1 to 4 for interactions between three and
four particles. In the preceding subsection, three diagrams were drawn for the lowest
order contribution to the potential quark self energy. The analogue problem in threshold
expansion is that each loop momentum can lie in either of the three rgimes, so that the
number of expansions for an n loop diagram grows like 3n .
The important step to cut down the computational effort in either approach is to note
that as an immediate consequence of the axioms of dimensional regularisation (e.g., [26,
Chapters 4.1 and 4.2]), integrals without scales set by external momenta or energies in the
denominator vanish because of homogeneity,
Z
dd q
q = 0.
(3.10)
(2)d
For example, applying this theorem to the potential quark self energy diagrams, contributions from soft (3.4) and potential (3.8) loop momenta are zero to all orders in the velocity
expansion. For the potential gluon contribution (3.8), this is seen by shifting the regularised
2
loop integral T + q0 (p+q)
2M q0 before using (3.10) for the energy integral. The importance of (3.10) has already been alluded to in threshold expansion [16], and here a more
thorough and formal treatment is presented.
It is this theorem (3.10) which renders most of the diagrams in NRQCD and threshold
expansion calculations zero, and it is usually not even necessary to consider the whole
diagram but only a subset of it. The following set of rules is helpful for reducing the
numbers of diagrams to be dealt with in the calculation of the function.
In order to establish these rules, recall that all graphs vanish which contain a subgraph
zero by the rules developed. The routing of the loop four-momentum inside a diagram
is arbitrary as always in dimensional regularisation so that having regularised the loop
integration, all loop four-momenta can be shifted at will like in ordinary integration. Since
they have identical denominators (Section 2.3), gluon lines in any rule can be replaced
by any relativistic particle, e.g., ghosts and light quarks. Finally and most importantly,
(a)
(b)
331
(c)
(d)
Fig. 2. The first rule as proven step by step in the text. (a) Simple case; (b) a more complicated
subdiagram; (c,d) generalisations. The overlay of double and zigzag line stands for any number of
arbitrary soft particles, the triple line for any number of potential or ultrasoft particles, all entering at
the same vertex. The blob represents vertex and propagator dressings. No time ordering is implied in
the way the diagrams are drawn.
numerators are unimportant to determine whether an integral is scale-less following (3.10).

As multipole expansion and insertions do not change the denominators of the propagators,
a diagram is therefore zero to all orders when it is zero because of (3.10) at leading order
in v. This result is also insensitive to the gauge chosen and to the specific vertex involved.
For example, because the numerators play no rle in rendering (3.4)(3.8) zero using
(3.10), the same will hold for any graph with the same diagrammatic representation but
different interactions, e.g., also for graphs in which one or both vertices are replaced with
the minimal coupled vector fields Q AQ or the Fermi interaction Q BQ.
As a first rule, consider a potential gluon between two soft particles, represented by the
overlay of double and zigzag lines in Fig. 2(a). This subdiagram is zero when the loop
energy q0 is integrated over: Assign the potential loop momentum q to the instantaneous
gluon with a propagator denominator q 2 (2.18). As the coupling of soft particles to
potential ones is energy non-conserving, the denominators of the soft particles coupling to
Ap do not contain q0 Mv 2 , either. Therefore, q0 does not occur in any denominator, and
the dimensionally regularised loop integral over q0 is zero from (3.10). The rule is extended
by noting (Fig. 2(b)) that overall energy conservation allows for the potential gluon line
to be attached with an arbitrary number of vertices into which an arbitrary number of
potential or ultrasoft gluons or quarks with collective (potential) four-momentum l enters.
The denominators of the potential gluon between two soft particles still do not contain
the loop energy, and the q0 integral vanishes again, Fig. 2(c). With this four-momentum
routing, l can be an external or loop momentum.
One can finally dress the vertices and propagators without changing the argument thanks
to scale conservation. Also, when several other potential or ultrasoft particles couple to
the same soft particle in addition to the potential gluon, a temporal multipole expansion
332
(a)
(b)
(c)
Fig. 3. The second rule in its bare (a) and dressed (b) version; (c) the generalisation analogous to
Fig. 2(c) fails. Conventions as in Fig. 2.
is necessary for each of them. Therefore, a potential gluon can never connect two soft
particles, irrespective of the number of particles coupling to the same vertices as the
potential gluon, and irrespective of other particles coupling to the potential gluon in the
intermediate state, Fig. 2(d): Diagrams with unbroken potential gluon lines between soft
particles vanish.
As predicted, the only vertex property which enters is that the coupling between particles
of different rgimes necessitates a multipole expansion which can be read up from the
diagram directly. The proof therefore extends to massive, relativistic potential particles,
like a light quark. It eliminates for example the diagram corresponding to a non-Abelian
vertex correction with a denominator
h
2 i
2
Denom. = T 0 q 2 T T 0 p p0 + q .
(3.11)
A similar rule can be proven for the potential quark between soft particles, see Fig. 3(a)
for its bare and (b) for its dressed form. The vertices are again energy non-conserving, so
q2
.
that q0 enters only in the denominator of the potential quark in the combination q0 2M
2
q
q0 , the q0 integral again does not contain a scale in the denominator.
Shifting q0 2M
In contradistinction to the previous case, this rule cannot be extended to include potential
or ultrasoft momenta coupling to the blob, Fig. 3(c). For example, a bremsstrahlung gluon
with external, ultrasoft four-momentum l renders the potential quark denominators in the
q2
q2
)(q0 + l0 2M
). Now, l0 provides the necessary scale for the q0
propagators as (q0 2M
integration. As an example of the rule, the Abelian vertex contribution with denominator

2 i
h
q2
Denom. = q0
T T 0 T 02 p0 q
2M
(3.12)
is zero, but the last of the graphs in Fig. 1 is not.

As the coupling of an ultrasoft gluon to a soft particle conserves neither energy nor
momentum, one may disconnect the ultrasoft gluon from such a vertex, Fig. 4(a). If after
cutting, a line with loop momenta to be integrated over becomes completely disconnected
from the graph, the resulting tadpole makes the graph vanish, Fig. 4(b). If only one of the
Au legs becomes disattached, the diagram can be non-zero, as the following diagram for
the Abelian vertex correction to the Qs As 0 Qp coupling demonstrates:
(a)
333
(b)
Fig. 4. (a) Soft-to-ultrasoft vertices are cut off in a third rule; (b) an example. Conventions as in
Fig. 2.
(a)
(b)
Fig. 5. The fourth rule in its (a) bare and (b) extended version, where the rule in Fig. 4(a) was used
for attaching ultrasoft gluons to the soft gluon. Conventions as in Fig. 2.

p2
0
6 0 : Denom. = T q T q0
=
. (3.13)
2M
2
See also Section 3.6 and the computation of this graph in Appendix A.2.
When several ultrasoft gluons couple to a soft particle at the same point, cutting does
usually not result in a tadpole, either. For example, the gluon-gluon vertex obtained by
cutting the two-gluon two-quark vertex
6= 0 : Denom. = (k q)2 q 2 6= 0
(3.14)
might be represented by a small blob as a reminder that it is neither energy nor momentum
conserving. In the tadpole, a scale is set by the external ultrasoft four-momentum k.
Loops made out of one soft quark and one soft gluon are restricted by another rule.
In Fig. 5(a), the denominators are q02 q 2 for the soft gluon and q0 for the soft quark.
Therefore, the diagram is without scale. The generalisation, Fig. 5(b), eliminates all soft
contributions to the potential quark self energy (3.5) and to Abelian vertex corrections
between potential and ultrasoft particles. Note that the light particle exchanged is massless
or has a mass considerably smaller than Mv 2 . A relativistic particle mass of order Mv 2 or
bigger provides a scale in the relativistic propagator, invalidating this rule. Also, as soon as
a potential particle couples to the soft gluon, the diagram is non-zero. An example is the
non-Abelian vertex correction to the Qp Ap Qp coupling with denominator
334

6= 0 : Denom. = q0 q 2 q02 (p p0 + q)2 .
(3.15)
To summarise, the principle underlying the rules is simple: One identifies a potentially
vanishing sub-set of the entire diagram and assigns the external and loop momenta
to it, taking into account multipole expansions. Then, the denominators (i.e., inverse
propagators) are written out at lowest order in v. If by shifting integration variables, one
arrives at an energy or momentum integral without scale, the diagram vanishes to all orders.
Along this line, further rules are easily established.
Two more rules are more straightforwardly derived from the propagator properties of
NRQCD directly. The first one cuts down the number of diagrams including potential
gluons. For the heavy quarks, there is no distinction between Feynman and retarded
propagators in NRQCD (2.10), (2.11) because antiparticle propagation has been eliminated
by the field transformation from the relativistic to the non-relativistic Lagrangean. The
fields Qs and Qp propagate hence only time-like into the forward light cone. Still,
Feynmans perturbation theory becomes more convenient than the time-ordered formalism,
as less diagrams have to be calculated. Soft and ultrasoft gluons propagate light-like, and
potential gluons instantaneously. Therefore, any diagram which when time ordered
involves quarks propagating outside the forward light cone are zero, as are diagrams which
cannot be drawn with both ends of a potential gluon being space-like separated. This
forwardness excludes all non-zero contributions from the potential gluon contribution
to the quark self energy in any rgime, e.g., (3.8).
Another rule is that there must at least be one pole on each side of the real q0 axis
for a diagram to be non-zero. This will, e.g., not be the case when the loop momentum
q only occurs in one quark propagator besides potential gluon propagators. Likewise, the
contribution to the Abelian vertex correction
:

q2
(q + p 0 p)2
+ i q0
+ i
Denom. = q0
2M
2M

2
2
(3.16)
T (p q)
has only poles below the real axis. Closing the contour of the q0 integration above it, the
diagram is zero. This can also be proven using dimensional regularisation.
In order to demonstrate that the rules established above are useful in reducing the number
of graphs, Fig. 6 shows that only one of the four scale-conserving graphs contributing to
the one loop soft quark self-energy survives their filter. The classification of the only nonzero diagrams for Abelian and non-Abelian one-loop corrections to one gluon vertices
addressed later, Figs. 8, 9 and 10, may serve as another example.
In conclusion, the homogeneity of dimensional regularisation (3.10) can be translated
into diagrammatic rules to systematically identify graphs which vanish to all orders in
threshold expansion or velocity expansion in NRQCD. The concept is sensitive only to
(a)
(b)
(c)
335
(d)
Fig. 6. Out of the four scale-conserving one loop contributions to the soft quark self energy, only
the first one, (a), survives; (b) is zero because it contains a sub-graph zero by Fig. 3(a); (c) vanishes
because of Fig. 2(c) or because of forwardness; (d) is eliminated by cutting off the ultrasoft gluons
(Fig. 4(a)). The resulting tadpole is zero.
the multipole expansion of the vertices involved, not to their precise nature. It is also
independent of insertions and gauge chosen. Only the denominators of the graph lowest
order in v have to be looked at.
3.3. Theorems from diagrammatic rules
Diagrammatic rules are quite effective to prove decoupling theorems to all orders in
perturbation theory. Scale conservation entered in Section 2.3 as a constraint on both
the overall energy and momentum of a vertex because both have to be conserved order
by order in v. Now, one may resort to a more formal argument: If one rescales the fields
in the Lagrangean in all possible combinations, diagrams containing scale non-conserving
vertices are eliminated because they correspond to temporal or spatial tadpoles. As an
example, consider a soft particle coupling to two potential ones. The dashed-solid line
representing any potential particle, the multipole expansion is
(2)
4 (3)

0
(p p + q) exp (T T )
(q0 ) .
q0
0
(3.17)
The temporal tadpole of the q0 integration makes all diagrams vanish in which this vertex
is embedded such that q is a loop momentum. If q0 is an external energy, its being zero
contradicts the assumption that it is of order Mv. Therefore, diagrams containing scale nonconserving vertices are zero. In threshold expansion, this follows automatically from the
observation that not the individual vertices but the loop momentum as a whole is expanded,
see Section 3.1.
Another application is the decoupling of an ultrasoft heavy quark from NRQCD.
When the particle content of NRQCD in the three different rgimes was outlined in
Section 2.1, the bremsstrahlung gluon was assumed to be the only ultrasoft field, and
an ultrasoft quark Qu was not necessary to make the theory self-consistent. An ultrasoft
quark might still be produced by the radiation of a potential gluon off a potential quark,
but it is not unavoidable. In contradistinction, a soft quark is indispensable because a
soft, on shell gluon excites a potential quark to a kinetic energy of order Mv: The
potential quark coupling to a soft gluon becomes necessarily soft. Taking advantage of
the homogeneity (3.10) of dimensional regularisation, one can show that a hypothetical
ultrasoft heavy quark decouples completely from the theory: All graphs containing it are
zero. This fact was already remarked upon by Beneke and Smirnov [16], and the formal
proof using the NRQCD techniques proceeds as follows.
336
(a)
(b)
(c)
(d)
Fig. 7. Decoupling of the heavy, ultrasoft quark. (a) primitive diagram; (b) extension to the case
of ultrasoft particles coupling to Qu ; (c) general vertex involving potential and ultrasoft particles;
(d) ultrasoft quarks decouple from any graph containing potential and ultrasoft particles. Conventions
as in Fig. 2.
Of all properties of the ultrasoft quark, the only ones needed are that its propagator
(denoted by a dotted line) is static because T Mv 2 , |p| Mv 2 ,
:
i
,
T + i
(3.18)
and that it couples momentum non-conserving to all but ultrasoft particles. These features
can of course be derived from the rescaling rules of an ultrasoft quark, Qu (x, t) =
(Mv 2 )3/2 Qu (X u , Tu ), in full analogy to Section 2.2. If it exists, Qu can only enter in
internal lines. Fermion number conservation dictates that it is produced and annihilated
in a vertex into which at least one soft or potential quark enters.
The simplest sub-diagram containing an ultrasoft quark is depicted in Fig. 7(a) and
vanishes because there is no loop momentum q in any denominator. This is still true when
an arbitrary number of ultrasoft particles couple to Qu , Fig. 7(b), and when such a coupling
occurs repetitively. Now one considers couplings to potential particles like in Fig. 7(c).
The momentum multipole expansion necessary for all ultrasoft particles separately in such
a vertex disallows q to be transferred to a denominator. Therefore, any diagram containing
an ultrasoft quark coupled to potential or ultrasoft particles is zero because of the spatial
tadpole in the q integration, Fig. 7(d). Because couplings of ultrasoft particles to soft ones
involve the same momentum multipole expansion as couplings to potential ones, the proof
extends to all graphs containing ultrasoft quarks: Every diagram with ultrasoft quarks is
zero. Recursive application dresses all vertices and propagators.
3.4. The quark self-energy
The rules cover the lowest order contributions to the potential quark self energy
as discussed in Section 3.1. The graph in (3.5) is zero because of Fig. 5, the one
337
in (3.8) because of forwardness. The only remaining graph is (3.9), whose lowest
order contribution is P.C.(v 0 ) by the power counting of Table 1. Indeed, corrections to
particle propagators, quark or gluon, must count as v 0 since the power counting was
constructed such that free particle propagators in each rgime scale like v 0 (2.6), (2.7) and
renormalisation corrections to propagators must be of the same order. Using the integral
(A.4) of Appendix A.1, the pole in four dimensions is extracted as
:

igR2 2
p2
3
[]
T
ip (T , p)pole =
C
+ P.C.(v), (3.19)
F
8 2
2
2M
where = 2 d/2 and is the renormalisation parameter. The potential quark propagator
is recovered, confirming that different rgimes do not mix under renormalisation. The nonrelativistic quark propagator does not need renormalisation of the heavy quark mass. The
quark wave function renormalisation is hence in the MS scheme
gR2 2
3
[].
(3.20)
CF
2
8
2
For the soft quark self energy, the only diagram surviving the diagrammatic filter (Fig. 6)
is at lowest order in the power counting using the Feynman rules and (A.4)
Z2 = 1 +
igR2 2
3
[]T .
(3.21)
CF
8 2
2
That the soft quark renormalisation is the same as for the potential quark is not surprising
as both stem from the same unexpanded NRQCD diagram (3.2).
:
is (T , p)|pole =
3.5. The vacuum polarisation

In contradistinction to the self energy for Qp which does not receive contributions from
soft particles, the vacuum polarisation of the potential gluon is non-zero only because of
the presence of soft gluons, independent of the gauge chosen. The scale-conserving graphs
with gluon loops are
(3.22)
The integral in the second graph does not contain q0 in the denominator and hence
is zero; the third diagram vanishes because it couples a potential gluon to a light-like
particle twice. Soft gluons are therefore indispensable to provide the gluons mediating
the Coulomb interaction with a non-zero vacuum polarisation, and hence to have a running
function. As the gluons in the soft and ultrasoft rgimes are on-shell particles and must
run in NRQCD as in QCD, there is no reason to expect the potential gluons to freeze
out in perturbation theory. The only non-zero ghost and light quark contributions come
analogously from soft ghost and soft light quark propagation in the loop.
The rescaling rules of Table 3 give the three gluon vertex to count as v 1/2 , and an
additional v 1 from the loop power counting makes the graph count as v0 at leading order.
338
Again, this is expected not only because gluons are relativistic particles, but also because
one wants to renormalise a propagator which is P.C.(v0 ) in the power counting. Without
the loop rule of Section 2.4, the graph would be of order v 1 , and the power counting
would predict that its contribution would vanish as v 0 so that the potential gluon
propagator at rest would reduce to the bare one, a clearly unacceptable conclusion. The
explicit computation (3.24) will also show that this graph is P.C.(v 0 ).
The soft and ultrasoft gluon receive their vacuum polarisations from loops with on shell
gluons, P.C.(v 0 ):
(3.23)
All other diagrams vanish by the rules developed above. Both contributions are identical
to the ordinary QCD result, as no insertions or expansions enter. The gluon vacuum
polarisation from gluon, ghost and light fermion contributions is therefore in the soft and
ultrasoft rgime [27, Eq. (2.5.132)]

ab
(k) = ab k k k 2 g (k 2 )

gR2 2 2
13
1
2
NF N
[] + finite.
(3.24)
with (k ) =
(4)2 3
2
3
Because the potential gluon vacuum polarisation does not contain insertions or multipole
expansions in the internal lines, either, the QCD result can be taken and expanded in powers
of the external energy k0 |k|. As the infinite part of (k 2 ) does not contain k, the only
change will be that to lowest order in v, the part guaranteeing transversality of the gluon
becomes

(3.25)
k k k 2 g i j ki kj + k 2 g + P.C.(v).
Renormalisation therefore keeps the contributions from the three rgimes separate and the
potential gluon propagator transversal up to higher order in v. For all three rgimes, the
gluon wave function renormalisation is in the end the one of QCD [27, Eq. (2.5.135)],

gR2 2 2
13
1
QCD
NF N
[] = Z3 .
(3.26)
Z3 = 1
(4)2 3
2
3
3.6. The vertex correction and NRQCD function
Since they probe only the relativistic sector of the theory, the renormalisation constants
for gluons, ghosts and other light particles are the same in QCD and NRQCD. The quark
wave function renormalisation is computed in the non-relativistic sector and for a nonrelativistic bispinor rather than a relativistic Dirac spinor, so that it is not surprising
QCD
g 2 2
R
= 1 (4)
that the result (3.20) differs from its QCD counterpart Z2
2 CF [] [27,
Eq. (2.5.139)] even in the dependence on , although the MS scheme was used in
both cases. As both Lagrangeans are gauge invariant and agree in the light particle
sector, the SlavnovTaylor identities of QCD must also hold in NRQCD. Therefore, the
339
Fig. 8. The Abelian vertex corrections of order g 3 from the interactions Q A0 Q which survive the
diagrammatic filter. The leading order power counting is obtained when all vertices are the leading
order scalar gluon interactions. After applying the rules of Section 3.2, the seven diagrams drawn are
the only survivors of 22 scale conserving graphs.
renormalisation Z1 of the quark-gluon vertex Q A0 Q can be inferred from Z2 , Z3 and the

three-gluon renormalisation Z1, g of QCD (and NRQCD) not to be the same as in QCD
where Z1QCD = 1
Z1 =
Z2 Z1, g
Z3
gR2 2
(CF
8 2
+ N 3+
4 ) [] [27, Eq. (2.5.145)]:

gR2 2
3+
QCD
=1+
[] 6= Z1 .
CF (3 ) N
(4)2
4
(3.27)
In the following, one will identify as the cause that all non-zero contributions from both
Abelian and non-Abelian vertex corrections to the scalar gluon vertex in NRQCD are
different from their QCD values: Most notably, the non-Abelian vertex only provides
gauge parameter corrections. As a by-product, the topologies of all one loop corrections to
interactions of one gluon with a quark will be found and their leading order velocity power
counting determined.
Which vertices may be encountered in the graphs leading in v? For every vertex, the
Q A0 Q coupling is at least one order stronger in v than any other coupling (tables 1
and 2). Any contribution is therefore suppressed in which scalar gluons in a vertex are
gc3
Q BQ
replaced with vector gluons, coupled either minimally or via the Fermi term 2M
in (1.1). Moreover, the Fermi term couples one or two gluons to the quark spin. When
only one Fermi interaction is found in any vertex correction, parity conservation requires
the strongest possible correction involving A0 as outgoing particle to be of the form of the
spinorbit term, i.e., proportional to Q (D E)Q. This cannot correct the leading
scalar gluon vertex. Two Fermi interactions are even more suppressed because vector
couplings are weaker than scalar couplings.
Each Abelian correction graph (Fig. 8) starts off at the same order as the bare graph
whose vertex correction it presents (Table 1) when only scalar gluons couple to the
quark. The only exception is the last diagram in the first row of Fig. 8, which by the
power counting is P.C.(v 1 ) while the bare Qs As 0Qp vertex is P.C.(v 0 ) from Table 1, as
confirmed by an explicit computation of this graph in Appendix A.2. This diagram does
therefore not enter in the computation of the vertex correction at lowest order.
The non-Abelian corrections to the Q A0 Q vertex at order g 3 fall into two categories:
The topology of the first class of diagrams, Fig. 9, is analogous to the one of the
340
Fig. 9. The non-Abelian vertex corrections of order g 3 which survive the diagrammatic filter. The
soft blob in the second diagram in the second row provides an additional factor 1/v by loop power
counting. The power counting indicated is the leading contribution in the Feynman gauge, when
the three gluon vertex is the standard QCD one and both one vector and one scalar gluon couples
minimally to the quark. In the Lorentz gauges, two scalar gluons couple to the quark, making all
diagrams a factor 1/v stronger and hence contribute at the same order as the Abelian corrections in
Fig. 8. Out of 25 scale-conserving diagrams, only eight survive the diagrammatic filter.
non-Abelian vertex correction in QCD. In the Feynman gauge = 1, its leading order
contribution involves two scalar gluons and one vector gluon because in that gauge,
different components of the gluon field do not mix when propagating, (2.18), and the
coupling of three scalar gluons is forbidden at order g by the Lorentz structure. Again
because for every vertex, the Q A0 Q coupling is stronger in v than any other coupling,
the NRQCD analogue to the non-Abelian QCD vertex correction to the scalar gluon vertex
is sub-leading in the Feynman gauge. The resulting power counting is reported in Fig. 9.
In the generalised Lorentz gauges, the scalar and vector components of the gauge field
mix in the gluon propagation, with the amount of mixing proportional to (1 ) (2.18).
A scalar gluon emitted from a quark will therefore partially turn into a vector gluon, which
then enters a three gluon vertex. The non-Abelian vertex corrections now have a non-trivial
Lorentz structure and evaluate to be proportional to (1 ) and (1 )2 , as is confirmed
in Appendix A.2, (A.15). A factor 1/v is gained with respect to the power counting with
one vector gluon coupling as given in Fig. 9, making the diagrams of the same, leading
order as the Abelian vertex corrections. The non-Abelian graphs can consequently be seen
as merely providing gauge corrections to the Abelian vertex result. The last diagram in the
first row and the next-to-last in the second row of Fig. 9 are sub-leading and vanish using
the equations of motion at lowest order for the outgoing potential quark, see Appendix A.2.
There is no comparable graph in QCD to the second category of non-Abelian vertex
corrections in NRQCD at order g 3 . Fig. 10 lists the topologies of all non-zero corrections
to one gluon vertices involving one gluon loop. Again, one convinces oneself easily from
Tables 3 and 4 that these diagrams are at least one order in v weaker than the Abelian vertex
corrections, even before the precise vertex structure is specified. In NRQED, this type of
diagrams occurs first at order g 5 from the e2 term in the Lagrangean (1.1) because a three
photon coupling does not occur earlier. In QCD, the lowest order three gluon coupling
is totally antisymmetric in colour space, so that its contraction with the totally symmetric
g2
Q A2 Q vanishes. The Fermi interaction has
expression in colour space from the vertex 2M
341
Fig. 10. The only non-zero diagrams of the vertex corrections of order g 3 from the Fermi term
proportional to c3 in the NRQCD Lagrangean (1.1). The soft blob in the next-to-last diagram provides
an additional factor 1/v by loop power counting. The leading order power counting is obtained when
the three gluon vertex is the standard QCD one and the gluons couple to the quark via the Fermi
term. Out of 24 scale-conserving diagrams, only six survive the diagrammatic filter.
a colour antisymmetric two gluon part because Bia = 12 ij k (j Aak k Aaj + gf abc Abj Ack ),
but is not a possible source of an order g 3 vertex correction in Fig. 10 to the minimal
coupling of scalar gluons and quarks because the spin structures do not match.
Hence, the non-Abelian nature of the gauge field enters into the NRQCD function only
via the gluon vacuum polarisation and as what may be called gauge correction represented
by non-Abelian vertex corrections. Each of the six quark gluon couplings between the
various rgimes is renormalised by one and only one non-zero Abelian and non-Abelian
vertex correction.
Since the classification of diagrams following the rules of Section 3.2 is insensitive to
the precise nature of the vertex coupling, the list leads as a by-product to a classification
of all non-zero one loop graphs in which two quarks and one gluon couple. The same is
true for the quark self energy and vacuum polarisation graphs. The v power counting of
the diagrams in Figs. 8, 9 and 10 is specific to the interactions chosen, but the fact that all
other graphs with the same topology are zero is not.
All Abelian graphs lead to the same vertex vertex correction, as demonstrated in
Appendix A.2 at exemplary graphs:

igR3 3
N a 3
a Abelian
t
[].
(3.28)
=
CF
igR t
pole
2
2
8 2
Likewise for the non-Abelian vertex corrections:

igR3 3 N a 3(1 )
t
[].
igR t a non-Abelianpole =
(4)2 2
2
The vertex renormalisation in the MS scheme,

non-Abelian ,
Z = 1 Abelian
1
pole
pole
(3.29)
(3.30)
coincides therefore with the one expected from the SlavnovTaylor identities (3.27).
The NRQCD function is computed from the scale invariance of the bare coupling
Z1
(3.31)
g = gR
Z2 Z3
342
and agrees with the anticipated result (3.1),

gR3
2
11
N NF = QCD .
NRQCD =
3
(4)2 3
(3.32)
To the order calculated, one may match the relativistic and non-relativistic functions
to infer that indeed to lowest order in g and v,
QCD
gR
NRQCD
= gR
(3.33)
To match NRQCD to QCD via the renormalisation group behaviour of its couplings and
operators is of course considerably more complicated than to match matrix elements.
Nonetheless, it demonstrates that matching relations survive renormalisation. The renormalised NRQCD coupling is as expected the same in all rgimes, gauge invariant, and
runs simultaneously in all rgimes. The NRQCD function is independent of the gauge
parameter . The Lorentz gauges are a legitimate gauge choice in NRQCD, obeying the
SlavnovTaylor identities. Wave function and vertex renormalisations differ in QCD and
NRQCD, while the functions do not.
In dimensionally regularised NRQCD, the vacuum polarisation receives its sole
contribution from the propagation of on-shell relativistic particles, i.e., of physical soft
or ultrasoft gluons and soft or ultrasoft ghosts and light quarks. Potential gluons, i.e., such
mediating the Coulomb interaction, do not give rise to a gluon renormalisation, nor do they
contribute to any other renormalisation at lowest order in v, (3.19) and Figs. 6, 8 and 9.
This observation applies to all standard gauges, including physical ones like the Coulomb
gauge. It is traced back to the homogeneity of dimensional regularisation (3.10), which in
turn is well known to be connected to a cancellation of ultraviolet and infrared divergences
in scale-less integrals, e.g., [27, p. 172 ff.]. In a cut-off regularisation, the situation is
different as massless tadpoles require both infrared and ultraviolet regularisation resulting
in logarithms, invalidating (3.10) and all diagrammatic rules, e.g., [28, Chapter 18.5]. In
such a regularisation, the Coulomb gluon vacuum polarisation can therefore indeed receive
its main contribution from Coulomb gluons in the intermediate state.
To close this section, mind that this derivation of the NRQCD function is slightly
simpler than its QCD counterpart but may be applied with equal generality. Except in the
determination of the gluon vacuum polarisation, the Lorentz structure of the vertices and
propagators did not enter. The Dirac algebra is not needed except when light fermions
are included, and then the manipulations are straightforward. The vertex corrections in the
Feynman gauge reduce to the computation of one graph: the Abelian vertex which involves
neither Lorentz nor Dirac indices. Non-Abelian vertex corrections are easily extracted, see
Appendix A.2. The presentation demonstrates also intuitively that quarks with mass M
higher than the scale at which the function is computed freeze out and are not to be
included in the light quark number NF of (3.32). In a world with Nf quarks, the QCD
function can hence be found from NRQCD with Nf = NF light quarks and one fictitious
quark with a mass much bigger than any real quark mass.
343
3.7. The Coulomb gauge function

With the gauge independent classification of all one-loop self-energy, vacuum polarisation and vertex correction diagrams completed, it is straightforward to calculate the
function in the Coulomb gauge from the renormalisation of the scalar gluon field. In order to classify the contributing diagrams, one combines the notion that in the Coulomb
gauge, A0 is only potential (Section 2.2) with the knowledge of the topology of the
non-zero one loop diagrams as reported in (3.19), (3.21), (3.22) and Figs. 8 to 10. As
noted in the previous section, the leading order contribution in each self energy, vacuum polarisation or vertex correction graph is obtained when only scalar gluons couple to quarks, and when all gluons are on shell, i.e., either soft or ultrasoft. Because for
the Coulomb gauge, this eliminates any correction which is of the same order as the
quark propagator or the Q A0 Q vertex (Table 1), the soft and potential quark self energy and all vertex corrections, Abelian or non-Abelian, are therefore zero to all orders
in g and v:
Z1Coulomb = 1,
Z2Coulomb = 1.
(3.34)
The coupling constant is hence renormalised only because the potential gluon vacuum
polarisation in the Coulomb gauge is non-zero. A computation of the gluonic part which
makes again use of split dimensional regularisation can be found in [25, Eq. (41)]:
ab, Coulomb
00
(k) =
igR2 2 ab 2 11
N [] + finite.
k
(4)2
3
(3.35)
The ghost part is zero because ghosts do not propagate. The contribution for light fermions
is gauge independent and can hence be extracted from the Lorentz gauge result (3.26). One
arrives finally at

gR2 2 2
11
Coulomb
NF
N [].
(3.36)
=1
Z3
(4)2 3
3
Not surprisingly, the NRQCD function is extracted as (3.32) using (3.31). This particular
part of renormalising NRQCD in the Coulomb gauge may appear simpler than in the
Lorentz gauges, but other parts suffer severely from technical problems like the nontransversality of the gluon propagator.
4. Conclusions and outlook

After presenting the extension to NRQCD of a recently proposed explicit velocity
power counting scheme for a non-relativistic toy field theory [15] following Beneke and
Smirnovs threshold expansion [16], this article presented the computation of the NRQCD
function to lowest order in g and v in the Lorentz gauges and in the Coulomb gauge. It
endorsed the relevance of a new quark representative and a new gluon representative in the
344
soft scaling rgime E |p| Mv in which quarks are static and gluons on shell. HQET
becomes a sub-set of NRQCD. The identification of three different rgimes of scale for
on-shell particles from the poles of the NRQCD propagators leads in a natural way to this
rgime, in addition to the well-known potential one with on-shell quarks and instantaneous
gluons mediating the Coulomb binding [14] and the ultrasoft one with bremsstrahlung
gluons [19]. Neither of the five fields in the three rgimes should be thought of as physical
particles. Rather, they represent the true quark and gluon in the respective rgimes as
the infrared-relevant degrees of freedom. None of the rgimes overlap. Using a rescaling
technique [14,18,19], Section 2 proposed an NRQCD Lagrangean which leads to the
correct behaviour of scattering and production amplitudes [15]. It establishes explicit
velocity power counting, preserved to all orders in perturbation theory once dimensional
regularisation is chosen to complete the theory. The reason is the non-commutativity
of the expansion in small parameters with dimensionally regularised integrals and the
homogeneity of dimensional regularisation.
There is an intimate relation between the recently discovered threshold expansion [16]
and this version of NRQCD which uses dimensional regularisation and has been developed
over the past years [14,15,18,19]. This was demonstrated at the outset of the computation
of the function in Section 3.1 and has already led to fruitful exchange. As noted by
Beneke and Smirnov [16], threshold expansion provides an efficient way to derive the
NRQCD Lagrangean (1.1) from QCD, deepening our understanding of effective field
theories: The coefficients ci , di , ei are obtained by computing QCD diagrams and subdiagrams in the hard rgime, q M. Threshold expansion also shows how to systematise
identifying the different kinematic rgimes of NRQCD. Indeed, NRQCD was incomplete
without the soft rgime and its corresponding degrees of freedom as pointed out by Beneke
and Smirnov [16]. One might therefore see threshold expansion and NRQCD as two
sides of the same coin. The one expands loop integrals to establish a power counting
for Feynman diagrams, while the other uses rescaling properties to classify the relative
strengths of all vertices, allowing for intuitive interpretations of the derived vertex power
counting rules on the level of the Lagrangean (see the end of Section 2.3). An effective field
theory formulation can also be applied more easily to bound state problems. Finally, while
threshold expansion has not yet been proven to be valid on a formal level, the existence of
NRQCD as an effective field theory is guaranteed by the renormalisation group, e.g., [26,
Chapter 8].
The calculation of the NRQCD function in the Lorentz gauges to one loop order in
Section 3 albeit its result was easily anticipated was non-trivial as it demonstrated
that the power counting proposed is consistent also after renormalisation. Because of the
splitting of quark and gluon fields in different representatives in the various rgimes, this
might be thought not to be straightforward. It also resolved a puzzle about the running
of the coupling of gluons mediating the Coulomb interaction. The rescaling and power
counting rules are gauge independent for a wide class of gauges including all standard
gauges. Bare graphs and their corrections are of the same power in v. The soft rgime is
necessary not only to render the same function as in QCD, but even to attain a result
which is independent of the gauge parameter at all and in which the renormalisation
345
constants obey the SlavnovTaylor identities. In the Coulomb gauge calculation of the
function, the quark self energy and the vertex correction of the vertex coupling a quark
minimally to a scalar gluon is easily shown to be zero to all orders in g and v. In the
Coulomb gauge, the vacuum polarisation is therefore the only non-trivial renormalisation
necessary for the function.
The derivation was greatly simplified by diagrammatic rules (Section 3.2) which follow from the homogeneity property of dimensional regularisation and allow one to
systematically recognise the majority of NRQCD graphs as zero to all orders in v
just by drawing them, independently of the gauge and of details of the vertices involved. The rules apply equally well to threshold expansion and will prove fruitful in
more rigorous investigations, also into NRQCD. Here, they were already used to prove
scale conservation and the decoupling of the ultrasoft, heavy quark in Section 3.3.
All topologies of one loop corrections to vertices involving one gluon and one quark
were also classified, and the power counting of the leading order contributions determined.
Dimensional regularisation in NRQCD has many advantages over other regularisation
schemes in which scale-less integrals are non-zero. It preserves the symmetries and
guarantees the tree level power counting to be corrected at most by logarithms. It is
well known that cut-off regularisation violates gauge symmetry and wants removal of
power divergences in the renormalisation process in order to conserve the tree level
power counting to all orders. When massless tadpoles do not vanish, the number of
diagrams to be computed in order to obtain even a simple vertex correction was seen
to increase from one in dimensional regularisation to three or more. In general, it is
not clear whether scale conservation holds in cut-off regularisation: The three different
rgimes require the introduction of at least three artificial cut-off scales to separate them
and to cure ultraviolet divergences. Furthermore, the ultrasoft quark does not decouple,
adding further diagrams. In the end, the answers of all well-defined regularisation schemes
do agree, but dimensional regularisation provides an elegant way which furthermore
often leads to analytic expressions. A disadvantage of dimensional regularisation is that
in its standard formulation, it cannot be used to explore the non-perturbative sector of
NRQCD.
Finally, NRQCD with dimensional regularisation shows how to establish a power
counting in any effective field theory with several low energy scales: The theory is
first divided into one sector containing the particles relativistic at a given heavy scale
and another sector containing the non-relativistic particles. In NRQCD, the former are
light quarks, gluons and ghosts, the latter the heavy quark. The effect of all physics at
larger scales is absorbed into the coefficients of the Lagrangean. Then, one identifies the
combinations of scales in which particles become on shell by looking at the denominators
of the various propagators. This gives the scaling rgimes. Finally, the Lagrangean is
rescaled to dimensionless fields in each rgime to exhibit the vertex and loop power
counting rules. A priori, all couplings obeying scale conservation are allowed. The
diagrammatic rules developed in Section 3.2 still hold. Non-relativistic effective nuclear
field theory is one obvious application with three low energy scales: the pion mass
346
and production threshold, and the anomalously large scattering length of the two body
system. 5
Among the points to be addressed further is an extension of the diagrammatic rules.
An investigation of the influence of soft quarks and gluons on bound state calculations
in NRQED and NRQCD is also important because as seen at the end of Section 2.4
their contribution at O(g 4 ) and higher becomes stronger than retardation effects. To
investigate the non-perturbative sector of NRQCD and to establish its equivalence to QCD
in the continuum is another formidable task.
Acknowledgements
I am indebted to M. Luke for initial discussions about the problem of the function in
power counted NRQCD out of which this article grew, and to M. Burkardt, J. Gasser,
G.P. Lepage, H. Reinhardt, A. Pineda, M.J. Savage and J. Soto for valuable further
exchanges and discussions. The work was supported in part by a Department of Energy
grant DE-FG03-97ER41014.
Appendix A. Some details on split dimensional regularisation

A.1. Useful split dimensionally regularised integrals
This appendix presents some useful formulae for non-covariant loop integrals using
split dimensional regularisation as introduced by Leibbrandt and Williams [25], see also
the appendix in [15]. In its results, split dimensional regularisation agrees with other
methods to compute loop integrals in non-covariant gauges, such as the non-principal
value prescription [30], but two features make it especially attractive: It treats the temporal
and spatial components of the loop integrations on an equal footing, and no recipes are
necessary to deal with, e.g., pinch singularities. Rather, it uses the fact that, like in ordinary
integration, the axioms of dimensional regularisation [26, Chapter 4.1] allow to split the
integration into two separate integrals:
Z
Z
d k0 dd k
dd k
=
.
(A.1)
(2)d
(2) (2)d
Where applicable, split and ordinary dimensional regularisation of covariant integrals must
hence agree once the limit 1 is taken with d still arbitrary.
For the quark self energy, the integral
Z
1
1
dd q
(A.2)
d
2
(2) (q + i) q0 + a + i
5 Following this article, Mehen and Stewart investigated the soft rgime in effective nuclear theory [29].
347
is most easily computed by combining denominators as in HQET,

Z
Z
d
(+1)
dd q 2
.
q + 2(q0 + a) + i
d
(2)
(A.3)
Now, the integrand is in a standard covariant form and can be computed using Minkowski
space methods with the result

Z
2i d2 [2 + 1 d]
1
1
dd q
=
d
(2)d (q 2 + i) q0 + a + i
(4) 2 []
(2a + i)d(2+1) ei( 2 ) .
d
(A.4)
By differentiating with respect to a, the following formula useful for the computation of
the vertex corrections is derived:

Z
2i d2 [2 + d]
1
1
dd q
=
d
(2)d (q 2 + i) (q0 + a + i)
(4) 2 [] []
d
(2a + i)d2 ei( 2 ) .
(A.5)
Another generalisation is easiest computed by first combining the denominators linear in

q0 using Feynman parameters,
1
1
=
q0 + a + i q0 + b + i
Z1
dx
0
1
,
[q0 + xa + (1 x)b + i]2
(A.6)
and employing (A.5):

Z
1
1
1
dd q
(2)d (q 2 + i) q0 + a + i q0 + b + i

2i d2 [2 + 1 d] (2a + i)d21 (2b + i)d21 i( d )
e 2
.
=
d
ab
(4) 2 []
(A.7)
Note that in all three integrals (A.4), (A.5), (A.7), a shift of q by an arbitrary value leaves
the result unaffected. A scale in the momentum integration is only induced by the scales a,
b of the energy integration.
A.2. Computation of exemplary vertex corrections
In order to illustrate the NRQCD power counting scheme further, this appendix
outlines the computation of characteristic Abelian and non-Abelian vertex corrections in
Section 3.6.
The Feynman rules give for the Abelian correction to the Qs As,0 Qs
348
Z
:
3 3 b a c
(igR ) t t t

q02
dd q i bc
1 (1 ) 2
(2)d q 2
q
As usual, t b t a t b = (CF
by rewriting it as
N a
2 )t .
i
i
.
T + q0 T 0 + q0
(A.8)
The integral containing q02 in the numerator is simplified
q02
T T0
T
T0
=
1
+
(A.9)
(T + q0 )(T 0 + q0 )
(T + q0 )(T 0 + q0 ) (T + q0 ) (T 0 + q0 )
and noting that the first integral is zero as it is without scale (3.10). The remaining integrals
are using (A.4), (A.7)

igR3 3
N a 3
t
[] + finite,
(A.10)
C
F
2
2
8 2
from which (3.28) is read up. The Abelian correction to Qp Ap,0 Qp ,
:
Z
(igR )3 3 t b t a t c

q02
dd q i bc
i
1
(1
)
d
2
2
(2)
q
q T + q0
2
i
p2
2M
T 0 + q0
p0 2
2M
, (A.11)
20
p
p
and T 0 T 0 2M
in (A.10). Again, Abelian is
is inferred by replacing T T 2M
extracted as quoted in (3.28). Note that if one were to employ the equations of motion
for the outgoing particles the integral would appear to vanish as no scale is present. This
would mean that there is no Abelian vertex correction to the potential (or ultrasoft) scalar
gluon vertex. In that case, the result for the NRQCD function is not even gauge invariant.
Since the diagram is the leading order contribution to the correction of the Qp Ap,0 Qp
vertex (Fig. 8), the application of the equations of motion is not legitimate.
As noted in Section 3.6, there are two Abelian corrections to Qs As,0 Qp . The leading
one is

Z
q02 i
dd q i bc
i
.
1 (1 ) 2
d
2
(2)
q
q q0 T + q0
(A.12)
The integration (A.4), (A.7) gives the expected Abelian contribution (3.28) to the vertex
normalisation (3.27).
349
The vertex correction already discussed in connection with the cutting rule, (3.13), is
according to Section 3.6 of order v 1 and hence not the leading contribution. Because T 0
p0 2
2
2M Mv and T |p| Mv, one indeed finds
:

q02 i
i
dd q i bc
1
(1
)
(2)d q 2
q 2 T T 0 + q0 p 0 2
2M

p0 2
0
3
3
T 2M
ig
N a 3
[]
+ finite = P.C.(v)!
= R 2 CF
t
8
2
2
T
Z
(A.13)
The diagram will therefore only contribute to the renormalisation of the multipole
expansion. Moreover, because it is sub-leading, one may use inside the integral the
p2
, to set the outgoing
equations of motion at leading order in the power counting, T = 2M
potential quark on its mass-shell. The integral vanishes then, as no scale is present,
see (3.13) and (A.13).
The cutting rule suggests that the Qs Au,0 Qs correction should be related to the soft
quark self energy:
Z

q02
i
dd q i bc
i
.
1
(1
)
(2)d q 2
q 2 T + q0 T + q0
(A.14)
The dot at the cut-off vertex stands again as a reminder that the structure of the vertex will
s (T , p) (3.21).
enter here, and one finds the vertex correction to be proportional to T
Using (A.5), the vertex correction is again (3.28).
Finally, two examples for the straightforward but tedious non-Abelian vertex corrections
of Fig. 9 are given. First, one may calculate the non-Abelian correction to Qs Au,0Qs
Z
:
(igR )2 gR 3 f abc t b t c
i
iG 0 (q) iG 0 (q)
dd q
(2)d T + q0
q2
q2
[2q0 g + q g0 + q g0 ].
(A.15)
The Feynman gauge term, being independent of (1 ), is zero as expected. For this
diagram, even the terms proportional to (1 )2 vanish, which can be shown to be finite
in general and hence not to contribute to the vertex normalisation. In (A.15), only the
350
contribution leading in v is considered. Using the well known relation if abc t a t b = N2 t a

for SU(N ), this can be collapsed down by rewriting numerators analogously to (A.9) to

Z
dd q
1
1 T2
1
.
(A.16)
gR3 3 Nt a (1 )T
(2)d T + q0 q 4 q 2
One extracts the pole part of the non-Abelian vertex correction as (3.29) from the result
obtained using (A.4):
igR3 3 N a 3(1 )
t
[] + finite.
(4)2 2
2
(A.17)
The leading piece of the non-Abelian correction to Qp Ap,0 Qp is extracted as

Z
: (igR )2 gR 3 f abc t b t c
dd q i iG 0 (q) iG 0 (q + 1E)
(2)d q0
q2
(q + 1E)2

2q0 g + (1E q) g0 (q + 21E) g0 ,
(A.18)
where 1E = (0, p0 p). The term proportional to (1 ) contains the ultraviolet

singularity and can be written as
Z

dd q 1
1
3 3
a
(A.19)
q 2 q 2 2E q .
gR Nt (1 )
(2)d q 4 (q + 1E)2 0
Standard formulae for dimensional regularisation yield as divergent part (3.29). The term
proportional to (1 )2 is ultraviolet finite and can be reduced to the integral
Z
N
1
dd q q02
gR3 3 t a (1 )2 1E 2
d
4
2
(2) q (q + 1E)4
=
igR3 3 N a (1 )2
t
.
2
(4)2 2
(A.20)
References
[1] W.E. Caswell, G.P. Lepage, Phys. Lett. B 167 (1986) 437.
[2] G.T. Bodwin, E. Braaten, G.P. Lepage, Phys. Rev. D 51 (1995) 1125; Phys. Rev. D 55 (1997)
5853.
[3] M. Nio, T. Kinoshita, Phys. Rev. D 55 (1997) 7267.
[4] A.H. Hoang, P. Labelle, S.M. Zebarjad, Phys. Rev. Lett. 79 (1997) 3387.
[5] A.H. Hoang, T. Teubner, Phys. Rev. D 58 (1998) 114023.
[6] A. Pineda, J. Soto, Phys. Rev. D 59 (1999) 016005.
[7] C.T.H. Davies, NRQCD Collaboration, Nucl. Phys. 63 (1998) 320.
[8] H. Matsufuru et al., Nucl. Phys. 63 (1998) 368.
[9] A. Czarnecki, K. Melnikov, Phys. Rev. Lett. 80 (1998) 2531.
[10] M. Jezabek et al., Phys. Rev. D 58 (1998) 14006.
[11] E. Braaten, hep-ph/9702225.
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
351
N. Brambilla, A. Vairo, Nucl. Phys. B (Proc. Suppl.) 74 (1999) 201.

A.V. Manohar, Phys. Rev. D 56 (1997) 230.
M. Luke, A.V. Manohar, Phys. Rev. D 55 (1997) 4129.
H.W. Griehammer, Phys. Rev. D 58 (1998) 094027.
M. Beneke, V.A. Smirnov, Nucl. Phys. B 522 (1998) 321.
C. Bauer, A.V. Manohar, Phys. Rev. D 57 (1998) 337.
M. Luke, M.J. Savage, Phys. Rev. D 57 (1998) 413.
B. Grinstein, I.Z. Rothstein, Phys. Rev. D 57 (1998) 78.
H.W. Griehammer, Proceedings of the Workshop, Caltech, February 2627, 1998, in: R. Seki,
U. van Kolck, M.J. Savage (Eds.), Nuclear Physics With Effective Field Theories, World
Scientific, Singapore, 1998, p. 229.
G.P. Lepage et al., Phys. Rev. D 46 (1992) 4052.
P. Labelle, Phys. Rev. D 58 (1998) 093013.
N. Isgur, M.B. Wise, Phys. Lett. B 232 (1989) 113.
N. Isgur, M.B. Wise, Phys. Lett. B 237 (1990) 527.
G. Leibbrandt, J. Williams, Nucl. Phys. B 475 (1996) 469.
J. Collins, Renormalization, Cambridge Monographs on Mathematical Physics, CUP, Cambridge, 1984.
T. Muta, Foundations of Quantum Chromodynamics, Lecture Notes in Physics, Vol. 5, World
Scientific, Singapore, 1987.
T.D. Lee, Particle Physics and Introduction to Field Theory, Contemporary Concepts in Physics,
Vol. 1, Harwood Academic Publishers, Chur, 1988, revised and updated first edition.
T. Mehen, I.W. Stewart, CALT-68-2210; nucl-th/9901064.
K.-C. Lee, S.-L. Nyeo, J. Math. Phys. 35 (1994) 2210.
P. Ramond, Field Theory: A Modern Primer, Frontiers in Physics, Vol. 74, Addison Wesley,
Redwood City, 1990.

Reheating and supersymmetric flat-direction

baryogenesis
Rouzbeh Allahverdi a , Bruce A. Campbell a , John Ellis b
a Department of Physics, University of Alberta, Edmonton, Alberta, Canada T6G 2J1
b Theory Division, CERN, CH-1211 Geneva 23, Switzerland
Received 17 January 2000; accepted 22 February 2000
Abstract
We re-examine AffleckDine baryo/leptogenesis from the oscillation of condensates along flat
directions of the supersymmetric standard model, which attained large vevs at the end of the
inflationary epoch. The key observation is that superpotential interactions couple the flat directions
to other fields, which acquire masses induced by the flat-direction vev that may be sufficiently small
for them to be kinematically accessible to inflaton decay. The resulting plasma of inflaton decay
products then may act on the flat directions via these superpotential Yukawa couplings, inducing
thermal masses and supersymmetry-breaking A terms. In such cases the flat directions start their
oscillations at an earlier time than usually estimated. The oscillations are also terminated earlier, due
to evaporation of the flat direction condensate produced by its interaction with the plasma of inflaton
decay products. In these cases we find that estimates for the resulting baryon/lepton asymmetry
of the universe are substantially altered. We identify scenarios for the Yukawa couplings to the flat
directions, and the order and mass scale of higher-dimensional superpotential interactions that set the
initial flat direction vev, that might lead to acceptable baryo/leptogenesis. 2000 Elsevier Science
B.V. All rights reserved.
PACS: 11.30.Fs; 11.30.Pb; 12.60.Jv; 04.65
Keywords: Supersymmetry; Baryogenesis; Reheating; Cosmology
1. Introduction
Initially, one of the most attractive features of Grand Unified Theories (GUTs) (for
a review, see [1]) was the prospect that they might provide an explanation [2] for the
matterantimatter asymmetry of the Universe, via their new interactions that violate baryon
and/or lepton number. Subsequently, it has been realized that, even in the Standard Model,
at the non-perturbative level there are sphaleron interactions that violate both baryon
and lepton number. This discovery has given rise to new scenarios for baryogenesis,
at the electroweak phase transition [3] or via leptogenesis followed by sphaleron
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 1 2 4 - 3
356
R. Allahverdi et al. / Nuclear Physics B 579 (2000) 355375
reprocessing [4]. Supersymmetric extensions of the Standard Model offer yet more
scenarios for baryogenesis. For example, they may facilitate electroweak baryogenesis by
permitting a first-order electroweak phase transition despite the constraints imposed by
LEP [5]. There is also the possibility that they may contain perturbative interactions that
violate baryon and/or lepton number via a breakdown of R parity, which under certain
circumstances [6] can induce baryogenesis.
However, perhaps the most attractive mechanism offered by supersymmetry is that
proposed by Affleck and Dine, according to which [7] a condensate of a combination of
squark and/or slepton fields may have formed during an inflationary epoch (for reviews,
see [8,9]) in the early universe, causing the vacuum to carry a large net baryon and/or
lepton number, which is then transferred to matter particles when the condensate eventually
decays. We recall that the condensate forms along some flat direction of the effective
potential of the theory, which we take to be the Minimal Supersymmetric extension of the
Standard Model (MSSM) at low energies. In the conventional approach to AffleckDine
baryogenesis, the condensate is essentially static until a relatively late cosmological epoch,
when it starts to oscillate. In turn, the termination of the period of oscillation has been
calculated in terms of the magnitudes of the soft supersymmetry-breaking terms present
in the effective potential, which become significant only at low temperatures, and of the
thermalization effects of inflaton decay [10].
The purpose of this paper is to re-examine this AffleckDine mechanism by incorporating a more complete treatment of the reheating of the universe after the inflationary epoch.
We argue that the flat directions are in general coupled to other fields that are kinematically
accessible to inflaton decay. These fields therefore have non-trivial statistical densities, and
become thermalized. The couplings of these densities to the flat directions induce effective
supersymmetry-breaking masses and A terms for the erstwhile flat fields. As a result, the
flat directions start oscillating earlier than previously estimated. Subsequently, the oscillations also terminate earlier, as the flat-direction condensate interacts with the plasma of
inflaton decay products and evaporates. The bottom line is that previous estimates of the
resulting baryon/lepton asymmetry of the universe may be substantially altered, and we
estimate some orders of magnitude for different representative parameter choices.
2. Flat directions
The D-flat directions of the MSSM are classified by gauge-invariant monomials in the
fields of the theory. These monomials have been classified in [11], and, for directions
which are also F -flat for renormalizable standard model superpotential interactions, the
dimension of the non-renormalizable term in the superpotential which first lifts the
respective D-flat direction has also been derived. Hereafter, we consider only those Dflat directions which are not lifted by renormalizable superpotential interactions. These
correspond to 14 independent monomials, and each monomial represents a complex D-flat
direction: one vev magnitude and one phase (all fields in the monomial have the same vev).
Since the monomials are gauge-invariant, appropriate gauge transformations generated by
357
non-diagonal generators can be used to remove that part of the D-term contribution to the
potential which comes from the non-diagonal generators. Also, any relative phase among
the fields in the monomial can be rotated away by those gauge transformations which are
generated by diagonal generators. There remains only one overall phase, i.e., the phase
of the flat direction, which can be absorbed by redefinition of the scalar fields. Note that
the D-term and F -term parts of the scalar potential are invariant under such a redefinition
(which is equivalent to a U (1) symmetry transformation) while the soft-breaking terms
and fermionic Yukawa terms generally are not. So we can always arrange the vev of the
fields in the monomial to be initially along the real axis. We also note that such a non-zero
vev breaks spontaneously the MSSM gauge group.
As an explicit example, consider the simplest case, which is the H u L flat direction. If the
T3 = 12 component of H u and the T3 = 12 component of L have the same vev, then all the
D terms from both diagonal and non-diagonal generators of the MSSM are zero. The nondiagonal ones are identically zero and the equality of the vev makes the diagonal ones zero
as well. These vevs can then be chosen along the real axis as noted above. There are eight
real degrees of freedom in the H u and L doublets. Two of them comprise the flat direction
and another three are Goldstone bosons eaten by the gauge fields of the spontaneouslybroken symmetries. The remaining three are physical scalars which are coupled to the flat
direction, and are massive due to its vev.
Now that all fields in the monomial have the same vev and are real, by an orthogonal
transformation we can go to a new basis where there is only one direction with a non-zero
vev. Let us label this direction and the orthogonal directions generically as , therefore
R 6= 0 while I = R = I = 0. For the specific H u L example, these are the following
combinations after the Goldstone bosons are absorbed by the Higgs mechanism:
2R = (H1 )R + (L2 )R ,
2I = (H1 )I + (L2 )I ,
21 = (H1 )R (L2 )R ,
22 = (H2 )R (L1 )I ,
23 = (H2 )I + (L1 )R .
The D terms from the T3 and U (1)Y generators give terms g 2 2 1 2 (up to numerical
factors) in the potential, whilst those from T1 and T2 give g 2 2 2 2 and g 2 2 3 2 terms
(up to numerical factors). It is a generic feature that all fields entering in the flat direction
monomial which are left after the Higgs mechanism (except the linear combination which
receives the vev after diagonalization) have masses of order g due to their D-term
couplings to the flat-direction vev.
We now consider supergravity effects, both in minimal models with soft-breaking terms
at the tree level, and in no-scale model (for a review, see [12]), where such terms are absent
at tree-level but arise from quantum corrections [1316]. The superpotential consists of the
tree-level MSSM terms and a series of non-renormalizable terms of successively higher
dimension, which are induced in the effective theory by the dynamics of whatever is the
underlying more fundamental theory. Without imposing R parity (or any other symmetry)
358
all gauge-invariant terms of higher dimension would exist in the superpotential. We may,
however, also wish to impose R parity on the higher-dimensional terms, as we have
done on the renormalizable interactions, to prevent substantial R-parity violation being
fed down from high scales by the renormalization-group running of the soft mass terms
[17]. If we assume that R parity is a discrete gauge symmetry of the theory, then it would
be respected by all gauge-invariant superpotential terms of arbitrary dimension. Relevant
higher-dimensional superpotential terms which lift the flat direction are of the form:
W n
n
,
nM n3
(1)
where n is a number of order one and M is a large mass scale, e.g., the GUT or Planck
scale.
During inflation, supersymmetry is strongly broken by the non-zero energy of the
vacuum. In minimal models this is transferred to the observable sector through the Khler
potential at tree level [18], while in no-scale models (for a review, see [12]) this happens at
the one-loop level [19]. Inflation then induces [18] the soft-breaking terms
CI HI2 ||2 + an HI
n
+ h.c.,
nM n3
(2)
where CI and a are numbers depending on the sector in which the inflaton lies, and HI
is the Hubble constant during inflation. We shall assume here that CI is positive and not
unnaturally small [18,19]. In the presence of the A term, the potential along the angular
direction has the form cos (n + a ), where a is the phase of a. Due to its negative masssquared, the flat direction rolls down towards one of the discrete minima at n +a = and
1
CI
CI
HI M n3 ) n2 , and quickly settles at one of the minima ( (n1)
is O(1)).
|| = ( (n1)
n
n
Therefore, at the end of inflation can be at any of the above-mentioned minima.
In the absence of thermal effects, would track the instantaneous minimum ||
1
(H M n3 ) n2 from the end of inflation until the time when H ' m3/2 , where m3/2
1 TeV is the low-energy supersymmetry-breaking scale [18]. At H ' m3/2 the low-energy
soft terms
m23/2 ||2 + An m3/2
n
+ h.c.
nM n3
(3)
would take over, with the mass-squared of becoming positive, and would then start
its oscillations. Also, the minima along the angular direction would then move in a nonadiabatic way, due generally to different phases for A and a. As a result, starts its free
1
oscillations around the origin with an initial vev osc (m3/2 M n3 ) n2 and frequency
m3/2 and, at the same time, the torque exerted on it causes motion along the angular
direction. In the case that the flat direction carries a baryon (lepton) number this will lead
R
I
to a baryon (lepton) asymmetry nB [7] given by nB = R
t I t . At m3/2 t 1 the
upper bound on nB [7] may be written

1
osc n3
3
osc
(4)
nB 2
M
m3/2 t 2
359
which, after transition to a radiation-dominated universe, results in an nsB that remains

constant as long as there is no further entropy release.
As we will see in subsequent sections, thermal effects of inflaton decay products with
superpotential couplings to the flat direction can fundamentally alter the dynamics of
the flat direction oscillation, and necessitate revision of the estimates for the resulting
baryon/lepton asymmetries produced.
3. Flat-direction superpotential couplings and finite temperature effects

As we have seen, the flat direction has couplings of the form g 2 2 2 to the fields
which are in the monomial that represents it. Besides these D-term couplings, it also has
F -term couplings to other fields which are not present in the monomial. These come
from renormalizable superpotential Yukawas, and have the form 1
W h
(5)
which results in a term h2 ||2 ||2 in the scalar potential. Again for illustration, consider the
H u L flat direction: H u has Yukawa couplings to left-handed and right-handed (s)quarks
while L has Yukawa couplings to H d and right-handed (s)leptons.
In the class of models that we consider, the inflaton is assumed to be in a sector which
is coupled to ordinary matter by interactions of gravitational strength only. In this case,
the inflaton decay always occurs in the perturbative regime and we need not worry about
2 , where
parametric-resonance decay effects [20]. The inflaton decay rate is d m3 /MPl
13
m is the inflaton mass and m 6 10 GeV from the COBE data on the CMBR anisotropy
[21]. Efficient inflaton decay occurs at the time when H ' d and the effective reheat
temperature at that time will be TR (d MPl )1/2 . For m 1013 GeV we get TR
1010 GeV, which is in the allowed range to avoid the gravitino problem (for a review,
see [2224]).
The crucial point to note is that, although inflaton decay effectively completes much
later than the start of its oscillation, nonetheless decay occurs throughout this period. In
2 )1/4 (assuming instant thermalization:
fact, a dilute plasma with temperature T . (H d MPl
we address thermalization below) is present from the first several oscillations, until the
effective completion of the inflaton decay [25]. It is easily seen that it has the highest
instantaneous temperature at the earliest time, which can reach T 6 1013 GeV. This plasma,
however, carries a relatively small fraction of the cosmic energy density, with the bulk still
in inflaton oscillations. The dilution of relics produced from this plasma by the entropy
release from the subsequent decay of the bulk of the inflaton energy is the reason that it
does not lead to gravitino overproduction. It is important to note that the energy density in
the plasma may be comparable to the energy density stored in the condensate along a flat
1 We note that for F -flat directions of the renormalizable piece of the superpotential, which are only lifted
by higher-dimensional nonrenormalizable terms, cannot have such superpotential couplings to fields which
appear in the monomial.
360
direction. As a result, the thermal effects from the plasma may affect the dynamics of flat
direction evolution which, as we see below, occurs in many cases.
All fields with mass less than T , and gauge interactions with the plasma particles, can
reach thermal equilibrium with the plasma. Those fields which are coupled to the flat
direction have generically large masses in the presence of its vev, and might not be excited
thermally. These include the fields which are gauge-coupled to and have a mass g (up
to numerical factors of O(1)) and many of the fields which have superpotential couplings
to , and hence have a mass h (also up to numerical factors of O(1)). For g > T or
h > T , the former or the latter are not in thermal equilibrium, respectively. We recall
that, in the presence of Hubble-induced soft-breaking terms, the minimum of the potential
1
for the flat direction determines that (H M n3 ) n2 , and the plasma temperature is
2 )1/4 . So a field with a coupling h to the flat direction can be in thermal
T (H d MPl
equilibrium provided that h 6 T , which implies that
2(n3)
dn2 MPl
(6)
h4(n2) M 4(n3)
and similarly for the gauge coupling g.
The back-reaction effect of the plasma of quanta of this field will then induce a masssquared +h2 T 2 for the flat direction to which it is coupled. If this exceeds the negative
Hubble-induced mass-squared H 2 , the flat direction starts its oscillation. This happens
for hT > H , i.e., for
H 6n 6
2
H 3 6 h4 d MPl
(7)
and similarly for back-reaction from plasma fields with gauge coupling g to the flat
direction. Therefore, a flat direction will start its oscillations if both of the above conditions
are satisfied simultaneously. We note that the finite-temperature effects of the plasma can
lead to a much earlier oscillatory regime for the flat direction, i.e., when H m3/2 .
It is clear that, in order for a plasma of the quanta of a field to be produced, the coupling
of that field to a flat direction should not be so large that its induced mass prevents its
thermal excitation. On the other hand, in order for its thermal plasma to have a significant
reaction back on the flat direction, its coupling to the flat direction should not be so small
that the thermal mass-squared induced for the flat direction will be smaller than the Hubbleinduced contribution. Therefore, to have significant thermal effects, we need couplings of
intermediate strength in order to have both conditions simultaneously satisfied. For the
fields which have D-term couplings of gauge strength g to flat directions, this is usually
not the case: As will be seen shortly, in most cases their couplings are too large to satisfy the
equilibrium condition. For the fields which have F -term couplings of Yukawa strength h
to the flat direction, the existence of significant thermal effects depends on the value of h,
as well as on the initial value of the flat-direction vev , which in turn depends on the mass
scale and the dimension of the higher-dimensional operator which lifts the flatness.
To organize our discussion then, we first assess the typical values of h to be expected for
couplings to the flat directions. For these typical values, we then estimate the importance
of thermal effects on vevs determined by higher-dimensional operators ranging over the
361
various different dimensions that can lift the flat direction, for both the case of lifting by
the GUT scale: O(1016 ) GeV, and by the Planck scale: O(1019) GeV.
We now list the Yukawa couplings of the MSSM. For low tan , the ratio of H u and H d
vevs, we have
hu1 104 ,
hd1 105 ,
hl1 106 ,
hu2 102 ,
hd2 103 ,
hl2 103 ,
hu3 1,
hd3 102 ,
hl3 102 ,
(8)
whilst the hu s and hd s tend to be more similar for high tan . The only Yukawa couplings
which are significantly different from O(102 ) are hu1 , hd1 , hl1 , and hu3 . Only flat directions
which include only the left- and right-handed up squark, the left- and right-handed down
squark, the left- and right-handed selectron and the left-handed sneutrino will have an h
significantly less than O(102 ).
For low tan , any flat direction which includes right-handed top squarks has a Yukawa
coupling of O(1) to some s, too large for those s to be in thermal equilibrium, given
the expected range of flat direction vevs . The left-handed squarks are coupled to both
H u and H d , so any flat direction which includes a left-handed top squark has a Yukawa
coupling of order 102 to H d as well. For high tan , any flat direction which includes the
left- or right-handed top or bottom squarks has a Yukawa coupling of order 1 since the top
and bottom Yukawas are of the same order. In general, any flat direction which consists
only of the above-mentioned scalars has a Yukawa coupling of order 1 to some s and/or
a Yukawa coupling significantly less than O(102 ) to other s.
Among all flat directions which are not lifted by the renormalizable superpotential terms,
there is only one which allows such a flavor choice: uude with one u in the third generation
and all other scalars in the first generation (i.e., tude). This exceptional flat direction still
has a coupling of O(104 ) to some fields, since it includes the right-handed up squark.
Taking into account all flavor choices for all flat directions which are not lifted at the
renormalizable superpotential level, we can use h ' 102 for the coupling of a generic flat
directions to fields. For the above-mentioned exceptional case we shall use h ' 104 .
So, for our discussion of the dynamics of flat direction oscillations we will consider
three representative cases. We will analyze the dynamics when inflaton decay plasmons
are coupled to the flat direction by: gauge couplings with coupling g ' 101 , generic
Yukawa couplings of order h ' 102 , or suppressed Yukawa couplings of order h ' 104 .
Consideration of these cases should allow us to explore the generic range of physical effects
that arise in flat direction oscillations, from a plasma of inflaton decay products.
We now undertake a detailed analysis to determine in which cases a plasma of inflaton
decay products can be produced, and can initiate the flat-direction oscillations by the
reaction they induce on the flat direction. Whether this occurs or not depends on the vev of
the flat direction, and the strength of the coupling of the plasma quanta to the flat direction.
The initial vev of the flat direction is set by both the underlying scale of the physics of the
higher-dimensional operators that lift the flat direction, and, for a given flat direction, by
the dimension of the gauge-invariant operator of lowest dimension which can be induced
by the underlying dynamics to lift the flat direction.
362
In order to categorize systematically the various cases which arise, we organize them
as follows. First, we divide them into two cases, depending on whether the underlying
scale of the new physics responsible for the higher-dimensional operators which lift the
flat direction and stabilize the vev at the end of inflation are GUT-scale: O(1016) GeV,
or Planck-scale: O(1019) GeV. Each of these cases is subdivided according to whether
the coupling between the flat direction and the inflaton decay products is of gauge
strength (g ' 101 ), standard superpotential Yukawa strength (h ' 102 ), or exceptional
suppressed Yukawa strength (h ' 104 ). As noted above, this covers the generic range
of couplings exhibited by fields in flat directions in the supersymmetric standard model.
Finally, each of these cases is subdivided and tabulated according to the dimension of the
operator that stabilizes the flat-direction vev, setting (given the possibilities listed above
for the underlying scale of the new physics responsible for the operators) the initial vev
of the flat direction. These higher-dimensional operators are listed by the order of the
monomial in the superfields which appears in the superpotential and is responsible for
the operator. We tabulate against the order of the higher-dimensional superpotential term
the following quantities (in Planck units 2 ): the Hubble constant H , the temperature T , and
the value of the flat-direction vev at the onset of oscillations, as well as the combination
hT 2 gT 2
H ( H for the case of the gauge coupling) which will be useful when we discuss the
produced baryon asymmetry in the next section. We also explain the reasons for the values
of the entries appearing, in the light of the two necessary conditions introduced above
for inducing the flat-direction oscillations by plasma effects, i.e., that on the one hand the
mass of the plasmon induced by the coupling to the flat direction is small enough that it
can be populated in the thermal bath from inflaton decay, and, on the other hand, that the
coupling is large enough for back-reaction effects from the plasma to lift the flat direction
sufficiently to start oscillation despite the effects of the Hubble-induced mass.
First, let us consider the case that the scale of the new physics that induces operators
that stabilize the flat direction is of order the GUT scale: O(1016) GeV. We then subdivide
this case according to the strength of the coupling of the inflaton decay products to the
flat direction. To start, we consider the gauge-coupled case with g = 101 . In this case, it
is only for initial flat-direction vevs fixed by either quartic or quintic higher-dimensional
terms in the superpotential that the plasma effects can accelerate the onset of flat direction
oscillation, with the results shown in Table 1. Physically, for superpotential monomials
of sixth order or higher, the initial flat-direction vev is sufficiently large that the mass
generated by its gauge coupling to the prospective inflaton decay products is large enough
to prevent them from being kinematically accessible for thermal excitation. In the case of
a quintic superpotential monomial this is also initially the case, and it is only after Hubble
expansion has reduced , and hence the induced plasmon mass, that thermalized products
of inflaton decay can back-react to induce flat-direction oscillation. However, this only
occurs for H < 1016 , by which time the low-energy soft supersymmetry breaking has
already initiated flat-direction oscillation.
2 From now on, we express some dimensionful quantities in Planck units.
363
Table 1
GUT scale M = 1016 GeV, gauge coupling g = 101
n
gT 2
H
4
5
108
1018
1013/2
109
1011/2
108
101/2
107
Table 2
GUT scale M = 1016 GeV, standard Yukawa coupling h = 102
n
hT 2
H
4
5
6
7
8
1026/3
1026/3
109
1028/3
1034/3
1020/3
1020/3
1027/4
1041/6
1022/3
1035/6
1044/9
1064/15
1064/15
1079/18
102/3
1016/9
102
1031/15
1017/18
For the GUT case M = 1016 GeV with generic Yukawa coupling h = 102 , we have the
results shown in Table 2 for lifting of the flat direction by monomials of the orders listed. In
the cases that the order of the monomial is four or five we have no difficulty satisfying the
condition that h 6 T , so that they are (thermally) populated in the inflaton decay plasma.
For monomials of order six, seven or eight, the induced mass of the prospective plasmon
is, in fact, of the same order or slightly larger than the instantaneous effective temperature.
So thermally they are present, albeit now with some Boltzmann suppression. Moreover, we
also note that these induced masses are less than the mass of the decaying inflaton, and so
they will be produced in the cascade of inflaton decay products, though, as noted above,
after complete thermalization they will be subject to some Boltzmann suppression. In all
cases the value of the Hubble constant at the onset of oscillation will be determined by the
second condition (hT > H ), which requires that the back-reaction-induced mass overcome
the Hubble-induced mass to initiate oscillation. By comparing the results of the tables for
g = 101 and h = 102 , we note that for a general flat direction with h = 102 which
is lifted at the n = 4 superpotential level, the values at the onset of oscillations should be
taken from the gauge analysis. The reason is that, in this case, the back-reaction of the
inflaton decay products which have gauge coupling to the flat direction act at an earlier
time than the back-reaction of those decay products which have Yukawa couplings to it.
For M = 1016 GeV, h = 104 , as a function of the order of the superpotential monomial
lifting the flat direction we have the results shown in Table 3. For these cases, the flatdirection-induced mass is always less than the instantaneous temperature, due to the weak
coupling of the flat direction to the plasmons. The only non-trivial condition now is the
second one (hT > H ), which determines how long one must wait before the Hubbleinduced mass is sufficiently reduced that the back-reaction-induced flat-direction mass can
364
Table 3
GUT scale M = 1016 GeV, exceptional Yukawa coupling h = 104
n
hT 2
H
4
5
6
7
8
9
1034/3
1034/3
1034/3
1034/3
1034/3
1034/3
1022/3
1022/3
1022/3
1022/3
1022/3
1022/3
1043/6
1052/9
1061/12
1014/3
1079/18
1088/21
101/2
1014/9
109/4
108/3
1053/18
1022/7
Table 4
Planck scale M = 1019 GeV, gauge coupling g = 101
n
gT 2
H
4
5
1014
1042
108
1015
107
1014
104
1025
overcome it to initiate oscillation. This fixes the value of H at the onset of oscillation.
Comparing the results of the tables for g = 101 and h = 104 , we note that for an
exceptional flat direction with h = 104 which is lifted at the n = 4 superpotential level,
the values at the onset of oscillations should also be taken from the gauge analysis.
We now turn to the case that the underlying scale of new physics responsible for
generating the higher-dimensional operators is the Planck scale. This means that the values
of the flat direction vevs after inflation will be larger, raising the mass of prospective
plasmons to which they couple, and making it harder to satisfy the constraint that these
putative plasmons be generated thermally, or even be kinematically accessible to inflaton
decay.
For M = 1019 GeV, g = 101 , we have significant effects only for flat directions lifted
by superpotential terms arising from quartic or quintic monomials. In all other cases
(n > 6) the flat direction vev is so large that quanta gauge-coupled to it receive sufficiently
large masses that they can not be thermally populated at the instantaneous temperature
of the inflaton decay products. For the two non-trivial cases we have the results shown
in Table 4. Only in the n = 4 case can we produce thermally a number of plasma quanta
sufficient to induce enough mass for the flat direction to initiate its oscillation at an earlier
time. In the n = 5 case, back-reaction from the plasma of inflaton decay products only
manages to induce flat-direction oscillation after H 1018 , by which time the lowenergy soft supersymmetry breaking has already acted to start the oscillation and also the
inflaton decay has been completed.
For M = 1019 GeV and h = 102 , we again have a case where flat directions lifted by
superpotential monomials of order six or higher result in such a large flat-direction vev that
365
Table 5
Planck scale M = 1019 GeV, standard Yukawa coupling h = 102
n
hT 2
H
4
5
1010
1030
107
1012
105
106
101
1014
Table 6
Planck scale M = 1019 GeV, exceptional Yukawa coupling h = 104
n
hT 2
H
4
5
6
1034/3
1034/3
1038/3
1022/3
1022/3
1023/3
1017/3
1034/9
1019/6
105/3
1032/9
107/2
quanta coupled to it receive too large a mass for them to be thermally excited in the plasma
of inflaton decay products. For the cases of quartic or quintic superpotential monomials,
we have the results shown in Table 5. We again find that only in the n = 4 case can thermal
effects actually induce sufficient mass for the flat direction to initiate oscillation earlier. In
the n = 5 case, back-reaction from the plasma of inflaton decay products only manages
to induce flat-direction oscillation after the low-energy soft supersymmetry breaking has
already done so, and the inflaton decay has been completed. By comparing the results of
the tables for g = 101 and h = 102 , we note that, for a generic flat direction, i.e., one
with h = 102 , the initial values at the onset of oscillations should be taken from the latter.
The reason is that, in this case, the back-reaction of the inflaton decay products which have
Yukawa couplings to the flat direction act at an earlier time than the back-reaction of those
decay products with a gauge coupling to it.
Finally, in the case M = 1019 GeV, h = 104 , for flat directions lifted by monomials
higher than sixth order the resulting flat-direction vevs are sufficiently large that quanta
coupled to it with this coupling are too massive to be excited at the instantaneous
temperature of the inflaton decay products. So the nontrivial cases are those in Table 6.
For n = 6, we marginally satisfy the requirement that hT ' , necessary for thermal
production of quanta coupled to the flat direction, while for n = 4 and n = 5 we do so
comfortably. The second condition, that hT > H for effective back-reaction, then serves
to determine the value of H at the onset of the thermally-induced oscillation. By comparing
the results of the tables for g = 101 and h = 104 , we note that when the exceptional flat
direction with h = 104 is lifted at the n = 4 superpotential level, the values at the onset of
oscillations should be taken from the latter. This is because the back-reaction of the inflaton
decay products with Yukawa coupling to the flat direction act at an earlier time than the
back-reaction of those decay products with gauge couplings to it.
366
We noted above that thermal effects from the plasma can be important up to h ' T or
even somewhat higher. For less than this, they change the convexity of the effective
potential in the direction at much earlier times, inducing the onset of flat-direction
1
1
oscillations. We should note that since H n2 and T H 4 , then decreases at the
same rate as, or more slowly than, T for 7 6 n 6 9. This means that if h T right after
the end of inflation, it will remain so for later times as well. Therefore, in the 7 6 n 6 9
cases for M = 1019 GeV, the Hubble-induced negative mass-squared is dominant and
will not be lifted until H ' 1016 , if h T at H ' 106 .
In sum, we conclude that for M = MGUT , a general flat direction with h ' 102 starts
oscillating at H 1016 in the 4 6 n 6 8 cases. For the exceptional one with h ' 104
it is true in the n = 9 case as well. For M = MPlanck in the denominator, only in the n = 4
case do oscillations of a general flat direction start at H 1016 . In the 5 6 n 6 9 cases,
the flat direction is protected from thermal effects because its large vev induces such a
large mass for fields coupled to it that they cannot be thermally excited in the plasma of
inflaton decay products. For the exceptional flat direction with h ' 104 , this protection is
weaker because of the smaller Yukawa coupling to (which therefore are lighter and can
be excited in thermal equilibrium) and, as a result, oscillations start at H 1016 in the
n = 5, 6 cases also
We need to elaborate on the implicit assumption that the s (s) are effectively thermal
upon production. In the model that we study, the inflaton decays in the perturbative regime,
and the decay products have a momentum less than, or comparable to, the inflaton mass
m 106 . The s (s) which are produced in two-body decays have a momentum of
order m. 3 It can easily be seen that the temperature at which oscillations start (assuming
thermal equilibrium) is 107 in all the above cases. Since the momentum of produced
particles is greater than the the average thermal momentum, the dominant process to reach
equilibrium is through the decay of s (s) to other particles with smaller momenta.
However, the momentum of s (s) is very close to the average thermal momentum.
Since thermalization does not change the energy density in the plasma, the number density
of s (s) is also close to its thermal distribution. Therefore, the plasma-induced massn
n
squared h2 E (g 2 E ) from s (s) is of the same order as h2 T 2 (g 2 T 2 ).
4. Thermal A terms and baryo/leptogenesis

Motion along the angular direction is required for the build-up of a baryon or lepton
asymmetry. This is possible if a torque is exerted on or, equivalently, if is not in one of
the discrete minima along the angular direction, when it starts oscillating. These discrete
minima are due to the A term part of the potential. Before the start of oscillations, the
Hubble-induced A terms are dominant, and the locations of the minima are determined by
them. During inflation, rolls down towards one of these minima and rapidly settles there.
3 The s generically have larger -induced masses than do the s, so their production may be delayed until
Hubble-dilutes to a smaller value.
367
After inflation it tracks that minimum and there is no motion along the angular direction
[18]. What is necessary then is a non-adiabatic change in the location of the minima, such
that at the onset of oscillations is no longer in a minimum along the angular direction.
In the absence of thermal effects, would start its angular motion (as well as its linear
oscillations) at H ' m3/2 . This occurs as a result of uncorrelated phases of the A terms
induced by the Hubble expansion and low-energy supersymmetry breaking. At this time,
the latter takes over from the former, and will in general no longer be in a minimum along
the angular direction. This will lead to the generation of a baryon or lepton asymmetry if
carries a non-zero number of either [18].
As we have seen above, due to thermal effects, in many cases the flat directions start
oscillating at much larger H . At this time the Hubble-induced A terms are still much larger
than the low-energy ones from hidden-sector supersymmetry breaking. In order to have
angular motion for , another A term of size comparable to the Hubble-induced one, but
with uncorrelated phase, is required. Since it is finite-temperature effects from the plasma
that produce a mass-squared which dominates the Hubble-induced one, one might expect
that the same effects also produce an A term which dominates the Hubble-induced A term.
This is the only new effect that could produce such an A term with uncorrelated phase, as
the thermal plasma is the only difference from the standard scenario.
The simplest such thermal A terms arise at tree-level from cross terms from the
following two terms in superpotential
n
nM n3
which results in the contribution
h + n
hn 2 n1 M n3 + h.c.
(9)
(10)
in the scalar potential. In thermal equilibrium, h 2 i can be approximated by T 2 and

therefore the thermal A term is of order
T 2 n1
.
(11)
M n3
There is another thermal A term that arises from one-loop diagrams with gauginos and
fermionic partners of . It results in a contribution, in the thermal bath, of order:

gT 2 T n
.
(12)
n
4 M n3
hn
We have checked that, for the parameter range of interest for this process, this has the same
order of magnitude as the tree-level A term. In the following, we use the tree-level term for
our estimates.
2
The ratio of the thermal A term to the Hubble-induced one is hT
H . It is clear from
the results summarized in the tables that, at the onset of oscillations, the thermal A
term is weaker than the Hubble-induced one in all cases. Therefore, at this time, the
minimum along the angular direction is slightly shifted, the curvature at the minimum
is still determined by the Hubble-induced A term, and the force in the angular direction
368

2 n2
is of order n hTM n3
. The ratio of the thermal A term to the Hubble-induced one grows,
2
however, as hT
H increases in time. Therefore, in what follows we keep both A terms in the
equation of motion of .
We consider the case where oscillations start because of the back-reaction of the fields,
as is the most common case. The masses of R and I are then of order hT . 4 The equation
of motion for the flat direction is then:
+ 3H + h2 T 2 + (n 1)hn
T 2 n2
H n1
+
A
n
M n3
M n3
||2(n2)
= 0.
(13)
M 2(n3)
At this time, the universe is matter-dominated, by the oscillating inflaton field, and
2 )1/2 t 1/2 for H > 1018 . After re-scaling
thus H = 3t2 . Also, T 2 = (H d MPl
n3
1/(n2)
/n )
and t Hosc t, where Hosc is the Hubble constant at the onset
(Hosc M
of oscillations, we get the following equations of motion for the real and imaginary
components of :
+ (n 1)2n

2
R
||n2
R + R + a 1/2 + b 1/2 cos (n 2) +
t
t
t

||n1
cos (n 1) + (n 1)||2(n2) R = 0,
+A
t

2
I
||n2
I + I + a 1/2 b 1/2 sin (n 2) +
t
t
t

||n1
sin (n 1) + (n 1)||2(n2) I = 0.
A
t
2
2 /H 2 , T
Here a h2 Tosc
osc = (Hosc d MPl )
osc
oscillations,
Hosc M n3 (n3)/(n2)
b (n 1)an
1/2
(14)
is the plasma temperature at the onset of
,
M n3
and = O(1) is the relative phase between the thermal and Hubble-induced A terms.
The first two terms and the superpotential term in each of these equations are the same as
in the equations derived in [18], but there are some important differences. First of all, the
flat-direction mass-squared is not the (constant) low-energy value m23/2 , but the thermal
mass which is redshifted as t 1/2 . Also, the Hubble-induced A term with coefficient H
appears instead of the low-energy one with coefficient m3/2 , which is negligible for H
1016 . This explains the 1t factor in front of the Hubble-induced A term. Finally, there
is another A term, the thermal one, which is also redshifted as t 1/2 , because of its T 2
dependence.
At the onset of oscillations, t = ti = 23 and is in one of the minima which are
determined by the Hubble soft terms. Therefore, ||i , which was (Hosc M n3 /n )1/(n2)
4 We use the estimate gT for the plasma-induced masses if oscillations start because of the back-reaction of the
fields.
369
Table 7
The value of nsB for flat directions which undergo plasma-induced oscillations
n
4
5
6
7
8
9
M = 1016 GeV
M = 1019 GeV
h = 102
h = 104
h = 102
h = 104
< 1011
< 1011
1011
4 1011
1010
No plasma effect
< 1011
3 1011
4 1011
1010
3 1010
5 1010
< 1011
No plasma effect
No plasma effect
No plasma effect
No plasma effect
No plasma effect
3 1011
3 109
107
No plasma effect
No plasma effect
No plasma effect
before re-scaling, is scaled to ||i = 1, and also i = n . We have solved these equations
R
I
numerically for i = 0, A = (n 1) and n = 1, and calculated nB = R
t I I . We
find that, among the cases listed in the above-mentioned tables, only for the following ones
do we get an nsB of order 1011 or larger, before any subsequent dilution after reheating.
In Table 7 we see that, in some cases, nsB is near the observed value of 5 1010 .
However, in the most general case, when the standard model gauge group is the only
symmetry group, these viable flat directions constitute only a small subset of all flat
directions. We also see that nsB is larger for the exceptional flat directions, when M =
1019 GeV, and for flat directions which are lifted by terms of higher order n. This is easily
understandable, as for larger M and n, and for smaller h, plasma-induced oscillations start
later and closer to the efficient reheat epoch H = 1018 . Larger M and n lead to a larger
vev for the flat direction and, therefore, the condition h 6 T will be satisfied at a later
time. A smaller value of h, on the other hand, implies that the condition hT > H will be
satisfied at a later time. Later oscillations mean less dilution of the generated lepton/baryon
asymmetry by the plasma of inflaton decay products (we recall that s T 3 is redshifted
only as t 3/4 for H > 1018 ).
Now we comment how our results may be affected by changes in the model-dependent
constants involved in the calculations: the reheat temperature TR (or equivalently the
CI
which appears in the expression for the flatinflaton decay rate), and the constant (n1)
n
direction vev. There are two concerns in this regard. First, whether the two conditions for
plasma-induced oscillations still result in a consistent value for Hosc which is greater
than 1016 , and, secondly, what is the corresponding change in the estimated value for nsB .
CI
' 1. If we assume instead that
In our calculations, we have used TR ' 109 and (n1)
n
CI
' 101 , it turns out that for all cases except the marginal ones
TR ' 1010 and (n1)
n
(the n = 6, 7, 8 cases for h = 102 and M = 1016 GeV, and the n = 6 case for h = 104
and M = 1019 GeV), plasma effects still trigger the oscillations for H > 1016 , though
at a somewhat smaller value of H . Moreover, the value of nsB remains the same within
an order of magnitude. Therefore, the plasma-induced oscillations of the (non-marginal)
flat directions, and the resulting value of nsB , are rather insensitive to the exact order of
CI
, at least for our purposes.
magnitude of TR and (n1)
n
370
5. Evaporation of the flat direction

Now let us find the time when the condensates are knocked out of the zero mode
by the thermal bath. For the evaporation to happen, it is necessary that the thermal bath
includes those particles which are coupled to . Then, two conditions should be satisfied:
first, the scattering rate of off the thermal bath must be sufficient for equilibration, and
secondly, the energy density in the bath must be greater than that in the condensate. The flat
direction has couplings both of Yukawa strength h to s and of gauge strength g to s.
The conditions for thermal production of s and s are h 6 T and g 6 T , respectively.
Since h < g, the s will come to thermal equilibrium at an earlier time. On the other hand,
the scattering rate of off the thermal s is scatt. h4 T while the rate for scattering of
off the thermal s is scatt. g 4 T . Therefore, s are produced earlier but in general have
a smaller scattering rate. The competition between the s and s, and between the ratio
of the energy density of the flat direction to the energy density in the plasma will determine
whether and how the flat direction evaporates.
First we consider those flat directions which have plasma-induced oscillations. If
oscillations start due to the back-reaction of s (which is the situation for most cases)
scatt. h4 T . For a general flat direction with h ' 102 , this is comparable to H at
H ' 1017 , while for the exceptional flat direction with h ' 104 this occurs at a much
smaller H . However, after starts its oscillation, it is redshifted as t 7/8 while T is
redshifted as t 1/4 . This implies that g
T decreases rapidly and soon the s will be in
thermal equilibrium. The rate for scattering of off thermal s is scatt. g 4 T and
scatt. > H at H 6 1012 . The energy density in the condensate at the onset of oscillations
is h2 2 T 2 6 T 4 (recall that h 6 T at this time). The ratio of the two energy densities is
further redshifted as t 5/4 (for H > 1018 ) which ensures the second necessary condition
for the evaporation of condensate, i.e., that the plasma energy density is dominant over the
energy density in the condensate. It can easily be checked that the condensate evaporates
at H 1018 , before the inflaton decay is completed. 5
In those cases in which the plasma effects do not lead to an early oscillation of the flat
direction, oscillations start at H ' 1016 , when the low-energy supersymmetry breaking
takes over the Hubble-induced one. It is important to find the time when the condensate will
evaporate in these cases too. For such flat directions, the ratio of the baryon number density
to the condensate density is of order one [18]. Therefore, if the condensate dominates the
energy density of the universe before evaporation, the resulting nsB will also be of order
one. Some regulating mechanism is then needed in order to obtain the value for successful
big bang nucleosynthesis: nsB 1010 [26].
Now consider a general flat direction with h ' 102 . As we showed, in the 5 6 n
6 9 cases for M = 1019 GeV, and the n = 9 case for M = 1016 GeV, plasma effects are
not important and the flat direction starts oscillating at H ' 1016 . By H ' 1018 the
inflaton has efficiently decayed and has been redshifted by a factor of 102 . From then
5 If oscillations start due to the back-reaction of s, from the beginning g 6 T and
4
scatt. g T .
Therefore, there is no need to wait for further redshift of and again the condensate evaporates at H 1018 .
371
Table 8
Viability of scenarios with generic Yukawa coupling h = 102
M = 1016 GeV
4
5
6
7
8
9
M = 1019 GeV
Early oscillation
Evaporation
Early oscillation
Evaporation
Marginal
Marginal
Marginal
Marginal
on, the universe is radiation-dominated, so t 3/4 and T t 1/2 . Therefore, the energy
density in the condensate is redshifted as t 3/2 whilst the energy density in radiation is
redshifted as t 2 . If the condensate does not evaporate (or decay) until very late times, its
energy density dominates that of the radiation and universe will again be matter-dominated.
At the beginning of oscillations, i.e., at H ' 1016 , has the largest vev in the n = 9
case for M = 1019 GeV, which is ' 1016/7 . At H ' 1018 this is redshifted to '
1030/7 which still leaves h > T , so plasmons with this Yukawa coupling to the flat
direction cannot be produced. However, since redshifts more rapidly than T , eventually
h becomes of order T , after a time such that
T ' 10101/7,
' 1087/7 .
(15)
It is easily seen that at this time the energy density in the condensate and in the radiation
are of the same order. Moreover, scatt. 108 T H and the condensate evaporates
promptly. This case is marginal as the condensate almost dominates the energy density
of the universe at evaporation.
In the 5 6 n 6 8 cases for M = 1019 GeV and the n = 9 case for M = 1016 GeV the
vev is considerably smaller and the energy density in radiation is even more dominant.
Therefore, a general flat direction with h ' 102 will evaporate before dominating the
energy density of the universe. We summarize the situation for a general flat direction with
h ' 102 , regarding both the early, i.e., plasma-induced, oscillation, and evaporation, in
Table 8.
For the exceptional flat direction with h ' 104 the situation is different. Here plasma
effects are not important in the 7 6 n 6 9 cases for M = 1019 GeV. In the n = 9 case the
condition for thermal production of s, h = T gives
T ' 1073/7 ,
' 1045/7
(16)
which means we do not need as much redshift to reduce , so s are produced earlier
and at a higher temperature. However, scatt. 1016 T , which is much smaller than H
at this time. Therefore, the condensate cannot evaporate by scattering off the s. It is
easily seen that hT > m3/2 ' 1016 when h = T . This implies that the mass and energy
density of the flat direction are hT and h2 2 T 2 , respectively, upon thermal production of
372
Table 9
Viability of scenarios with exceptional Yukawa coupling h = 104
M = 1016 GeV
M = 1019 GeV
Early oscillation
Evaporation
Early oscillation
Evaporation
4
5
6
7
8
9
Marginal
s, and the energy density in the flat direction and the thermal bath are comparable. As
long as hT > m3/2 , and T are both redshifted as t 1/2 . During this interval T remains
constant and the flat direction and plasma energy densities remain comparable. Later, when
T < 1012 we have hT < m3/2 , and the energy density in the condensate is m23/2 2 and
begins to dominate the thermal energy density. At some point, g < T and s can be
produced thermally. The scattering rate of the condensate off the s is scatt. 104 T
which is clearly at equilibrium. However, the energy density in the condensate is now
overwhelmingly dominant and evaporation does not occur. For the n = 7, 8 cases the
situation is similar and the condensate does not evaporate. The summary for the exceptional
flat direction, regarding both the early, i.e., plasma-induced, oscillation, and evaporation,
is illustrated in Table 9.
In summary: a general flat direction, i.e., with h ' 102 , which does not have plasmainduced early oscillation, does not come to dominate the energy density of the universe
(the n = 9 case for M = 1019 GeV is marginal). For the exceptional flat direction, i.e.,
with h ' 104 , the situation is different and it dominates the energy density of the universe
before decay.
6. Discussion
We have found that all flat directions, except those which are lifted by nonrenormalizable
superpotential terms of high dimension and with a large mass scale in the denominator, start
oscillating at early times due to plasma effects. For a general flat direction with h ' 102
these are the n = 4 case for M = 1019 GeV and the 4 6 n 6 8 cases for M = 1016 GeV
(with the 6 6 n 6 8 cases being marginal and sensitive to model-dependent parameters).
For the exceptional flat direction with h ' 104 these are the 4 6 n 6 6 cases for M = 1019
GeV (with the n = 6 case being marginal and sensitive to model-dependent parameters)
and all n for M = 1016 GeV. In these cases it is difficult to achieve efficient baryon
asymmetry generation by the oscillation of the condensate along the flat direction. We
showed that a general flat direction, i.e., one with h ' 102 , which is not lifted by thermal
effects, still evaporates before dominating the energy density of the universe. This is not
373
important for baryogenesis, however, and the resulting dilution by the thermal bath can be
used to regulate the nsB which is initially of order one. On the other hand, the exceptional
flat direction, i.e., one with h ' 104 , which is not lifted by plasma effects, dominates the
energy density of the universe before its decay.
For models with supersymmetry breaking via low-energy gauge mediation, on the other
hand, the evaporation of the condensate has yet another implication. In such models there
is a candidate for cold dark matter, the so called Q-ball [27]. In order to have stable Q-balls
as dark matter candidates, some flat directions must dominate the energy density of the
universe. This means that any flat direction which is evaporated by the thermal bath cannot
be used to form a Q-ball.
Now the question is which flat directions are lifted by n > 4 terms. A look at [11] reveals
that only 18 out of 295 directions which are D- and F -flat at the renormalizable level in
the MSSM are not lifted at the n = 4 level. Even a smaller subset of only 2 flat directions
are not lifted at the n = 6 level. If nonrenormalizable terms with n = 4 and n = 5 are not
forbidden by imposing other symmetries, only a very few flat directions in the MSSM can
be used for baryogenesis and even fewer for Q-ball formation (regardless of the mass scale
in the denominator or Yukawa couplings of these flat directions). This is if all higher-order
terms which respect gauge symmetry exist in the superpotential. With other symmetries
(discrete or continuous) imposed on the model, a specific flat direction will, in general, be
lifted at a higher level. The initial vev of can then be larger and and quanta may
not be produced thermally, and the standard treatment of the AffleckDine baryogenesis
may be valid. Model-dependent analysis is needed to identify at which level a given flat
direction is actually lifted, in a given model.
Finally, an interesting possibility is the parametric-resonance decay of a supersymmetric
flat direction to the fields to which it is gauge-coupled. The occurrence and implications
of a potential parametric resonance are more pronounced for those flat directions which
start their oscillations at H ' 1016 , as in the standard scenario. They have an incredibly
large 6 q = ( 2mg3/2 )2 which could be as large as O(1020 ) (the parameter q determines
the strength of resonance [20]). Explosive resonance decay could also prevent these flat
directions from dominating the energy density of the universe. However, the situation is too
complicated to allow simple estimates based on the results of parametric-resonance decay
of a real scalar field. First of all, the renormalizable part of the scalar potential (including
the D-term part which is responsible for parametric-resonance decay to s) is fully known
and very complicated. Moreover, the flat direction itself is a complex scalar field. This may
result in out-of-phase oscillations in the imaginary part of the flat direction, as well as in
other scalar fields which are coupled to the same , which can then substantially alter the
outcome of simple parametric resonance [28].
6 For those flat directions which have plasma-induced oscillations, m

3/2 is replaced by hT or gT , leading to a
considerably smaller q.
374
7. Conclusion
In conclusion, we have seen that many of the MSSM flat directions may start their
oscillations differently than in the standard scenario, where the low-energy supersymmetry
breaking determines the onset of oscillations. The two key ingredients for such a different
behaviour are: superpotential Yukawa couplings of the flat directions to other fields, and the
thermal plasma from partial inflaton decay, whose instantaneous temperature is higher than
the reheat temperature. Together, these lead to an earlier start of the oscillations. On the
one hand, the masses of those fields which are coupled to the flat direction that are induced
by the flat-direction vev are then small enough to be kinematically accessible to inflaton
decay and, on the other hand, induce large enough thermal masses for flat directions from
the back-reaction of those fields to overcome the negative Hubble-induced mass-squared
of the flat directions. Subsequently, thermal masses and A terms may be responsible for
baryo/leptogenesis, but typically result in an insufficient baryon/lepton asymmetry of the
universe. The oscillations are also terminated earlier, due to evaporation of the flat direction
through its interactions with the thermal plasma. It was also shown that even for many flat
directions whose oscillations are not initiated by plasma effects, these effects cause them
to evaporate before dominating the energy density of the universe.
Acknowledgement
The work of RA and BAC was supported in part by the Natural Sciences and Engineering
Research Council of Canada, and they would also like to thank the CERN Theory Division
for kind hospitality during part of this research.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
G.G. Ross, Grand Unified Theories, BenjaminCummings, 1985.

A.D. Sakharov, JETP Lett. 5 (1967) 24.
V. Kuzmin, V. Rubakov, M. Shaposhnikov, Phys. Lett. B 155 (1985) 36.
M. Fukugita, T. Yanagida, Phys. Lett. B 174 (1986) 45.
M. Carena, M. Quiros, C.E. Wagner, Nucl. Phys. B 524 (1998) 3.
B.A. Campbell, S. Davidson, J. Ellis, K.A. Olive, Phys. Lett. B 297 (1992) 118.
I. Affleck, M. Dine, Nucl. Phys. B 249 (1985) 361.
A.D. Linde, Particle Physics and Inflationary Cosmology, Harwood, 1990.
K.A. Olive, Phys. Rep. C 190 (1990) 307.
J. Ellis, K. Enqvist, D.V. Nanopoulos, K.A. Olive, Phys. Lett. B 191 (1987) 343.
T. Gherghetta, C. Kolda, S. Martin, Nucl. Phys. B 468 (1996) 37.
A.B. Lahanas, D.V. Nanopoulos, Phys. Rept. 145 (1987) 1.
G.A. Diamandis, J. Ellis, A.B. Lahanas, D.V. Nanopoulos, Phys. Lett. B 173 (1986) 303.
M.K. Gaillard, V. Jain, Phys. Rev. D 49 (1994) 1951.
M.K. Gaillard, V. Jain, K. Saririan, Phys. Lett. B 387 (1996) 520.
M.K. Gaillard, V. Jain, K. Saririan, Phys. Rev. D 55 (1997) 833.
R. Allahverdi, B.A. Campbell, K.A. Olive, Phys. Lett. B 341 (1994) 166.
M. Dine, L. Randall, S. Thomas, Nucl. Phys. B 458 (1996) 291.
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
375
M.K. Gaillard, H. Murayama, K.A. Olive, Phys. Lett. B 355 (1995) 71.
L. Kofman, A.D. Linde, A.A. Starobinsky, Phys. Rev. D 56 (1997) 3258.
B.A. Campbell, S. Davidson, K.A. Olive, Nucl. Phys. B 399 (1993) 111.
J. Ellis, J.E. Kim, D.V. Nanopoulos, Phys. Lett. B 145 (1984) 181.
M. Kawasaki, T. Moroi, Prog. Theor. Phys. 93 (1995) 879.
S. Sarkar, Rept. Prog. Phys. 59 (1996) 1493.
E. Kolb, M. Turner, The Early Universe, AddisonWesley, 1990, p. 280.
B.A. Campbell, M.K. Gaillard, H. Murayama, K.A. Olive, Nucl. Phys. B 538 (1999) 351.
A. Kusenko, M. Shaposhnikov, Phys. Lett. B 418 (1998) 46.
R. Allahverdi, R.H.A. David Shaw, B.A. Campbell, hep-ph/9909256, to be published in Phys.
Lett. B.

Scalar field theory limits of bosonic string

amplitudes
Alberto Frizzo a , Lorenzo Magnea a, , Rodolfo Russo b
a Dipartimento di Fisica Teorica, Universit di Torino and INFN, Sezione di Torino,
Via P. Giuria 1, I10125 Torino, Italy

b Institut de Physique, Universit de Neuchtel Rue A.-L. Breguet 1, CH-2000 Neuchtel, Switzerland
Received 17 January 2000; revised 17 March 2000; accepted 23 March 2000
Abstract
We describe in detail the techniques needed to compute scattering amplitudes for colored scalars
from the infinite tension limit of bosonic string theory, up to two loops. These techniques apply both
to cubic and quartic interactions, and to planar as well as non-planar diagrams. The resulting field
theories are naturally defined in the spacetime dimension in which they are renormalizable. With
a careful analysis of string moduli space in the Schottky representation we determine the region of
integration for the moduli, which plays a crucial role in the derivation of the correct combinatorial
and color factors for all diagrams. 2000 Elsevier Science B.V. All rights reserved.
PACS: 11.25.Db; 11.90.+t
Keywords: Perturbation theory; Bosonic strings; Scalar fields
1. Introduction
It is well known that all perturbative states in string theory have a squared mass
proportional to the string tension T . Thus, in the low-energy regime (or zero-slope limit,
0 = 1/(2T ) 0), the heavy string states become infinitely massive and decouple, while
the light states survive and their dynamics can be effectively described by an ordinary field
theory. It was understood, already in the old days of dual models, that one can define this
point-like limit in different ways, so that it is possible to recover different field theories. In
the first application of this idea [1], the tree-level amplitudes of a scalar field theory with
cubic interactions were derived from the corresponding amplitudes among scalar string
states; it was then shown that, if massless spin-1 states are selected, in the low-energy limit
one can reproduce the tree diagrams of YangMills theory [2], while, if closed string are
considered, one obtains the amplitudes of Einsteins gravity [3,4].
magnea@to.infn.it
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 2 0 0 - 5
380
A. Frizzo et al. / Nuclear Physics B 579 (2000) 379410
In more recent years further steps were taken, and string techniques were actually
exploited as a simplifying tool that can substitute Feynman diagrams for the explicit
calculation of scattering amplitudes and other quantities of interest in field theory. For
example, effective actions and threshold effects of interest for string unification were
computed in [57]; string-inspired techniques were applied to the evaluation of oneloop QCD scattering amplitudes [812] and renormalization constants [1315]; graviton
scattering amplitudes were computed and their relation to gauge amplitudes explored
[16,17]; progress was made towards the extension of the method to more than one loop
[1823], to amplitudes with external fermions [24], and to off-shell amplitudes [25,26].
String techniques also served to stimulate the development of new techniques in field
theory, that preserve some of the nice features of the string formalism [2732].
At first sight, the use of extended object for constructing particle scattering amplitudes
may appear as an unnecessary detour, since one is interested only in the zero-slope limit.
However, as is now well understood, this procedure presents many useful features. For
instance, string amplitudes are naturally written in a way that takes maximal advantage of
gauge invariance, and the color decomposition is automatically performed. At higher order
in the perturbative expansion, further advantages become apparent: the loop momentum
integrals are already performed, so that helicity methods can be readily employed, and
the result for a set of Feynman diagrams of a given topology is presented directly as
a Schwinger-parameter integral. Moreover, one does not find the large proliferation of
diagrams characteristic of field theories, which makes it extremely difficult to perform high
order calculations. In the case of closed strings one gets only one diagram at each order,
while in the open string the number of diagrams remains small. Finally, in the case of
bosonic strings, the expressions of scattering amplitudes and of the measure of integration
on moduli space are known explicitly for an arbitrary perturbative order; in the sewing
procedure [33,34], they can be obtained from tree level diagrams by identifying pairs of
external legs with an appropriate propagator.
On general grounds, a striking difference between field-theoretic and string-derived
amplitudes is the degree of correlation in the calculation of different amplitudes in different
field theories. In field theory, different amplitudes are largely independent of each other,
and thus in each computation one has to start from the same basic ingredients, the Feynman
rules; also, results obtained in one theory can rarely be exploited in computations for a
different theory: in fact, introducing some modification in the defining Feynman rules, all
subsequent results are affected. In the string approach, the situation is different: the basic
ingredients of all calculations always arise from quantities defined on the string worldsheet, and thus are not dependent on the specific definition of the field theory limit one
may consider. In particular, the building blocks are the measure of integration over string
moduli and the Laplacian Green function on the string world-sheet, essentially the twopoint correlator of two-dimensional scalar fields. This means that a large fraction of the
calculation is done in a general framework, without specifying which states will be selected
by the field theory limit. All these results can then be exploited for deriving different
amplitudes in a given field theory, or even for calculations applying to different field
theories. For instance, the number of external particles does not play such a dramatic role in
381
determining the complexity of the calculation. As we will show explicitly in the two-loop
case, one can learn a lot of information about the shape of the relevant string world-sheets
by studying simple vacuum bubble diagrams. The results obtained in this way will not be
modified by the insertion of external legs. Similarly, there are relations among calculations
in different theories; in fact, one chooses a specific field theory only when one selects,
in the string master formula, the contributions of a specific state. Technically this step
simply amounts to a Taylor expansion of all functions appearing in the string amplitude,
keeping the appropriate powers of the variables describing the string world-sheet. Clearly,
the building blocks of the calculation are not modified by changing the string state one
focuses on; also, the overall normalization and the integration region over string moduli
are fixed once and for all. In general, this unifying way of treating amplitudes of different
theories brings many simplifications; for instance, the tensor algebra associated with the
propagation of spin-1 (or spin-2) particles is bypassed, and the computational complexity
of these amplitudes is almost reduced to that of scalar amplitudes.
Despite all the advantages just described, so far the technique has been fully applied only
to massive scalars and to massless gauge bosons and gravitons, and has been completely
successful only at one loop. There are several technical reasons for this limitation: the
extension to fermions requires in principle the use of superstring amplitudes, and the
multiloop technology in that case has not yet been completely developed; the extension to
massive particles with spin is in principle possible, however one should realize that string
theory is clearly ill-equipped to reproduce field theories with several mass scales, since
all scales at the string level are proportional to the string tension T ; finally, the extension
to two and more loops has proved harder than expected, because it requires a detailed
understanding of multiloop string moduli space, whose corners contributing to the field
theory limit manage to reconstruct the particle amplitudes in a highly non-trivial way. The
present paper is a step towards the solution of the problems connected with the application
of string techniques to multiloop diagrams. In particular, we study the two-loop open string
moduli space, determine the correct integration region over moduli and punctures in the
field theory limit, and show how different corners of moduli space cooperate to reconstruct
individual Feynman diagrams, with the correct symmetry and color factors.
In this paper we focus on the study of scalar interactions. This means that both for
external and for propagating states we will select the contributions coming from the
open string tachyon, a Lorentz scalar taking values in the adjoint representation of the
ChanPaton group, which we shall take to be U (N). In Section 2 we will introduce
the technical tools needed for the computation of bosonic string amplitudes, and the
Schottky parametrization of the string world-sheet. In Section 3 we will show how one can
define two different point-particle limits, by matching in different ways the field coupling
constant g, which is kept fixed when 0 0, to the unique string coupling gS . The
two matchings lead to a cubic scalar interaction in d = 6 2, and to a quartic scalar
interaction in d = 4 2, respectively. Each field theory naturally arises in the space
time dimension in which it is renormalizable, essentially because the string scale in the
intermediate stages plays the role of a renormalization scale, and disappears from the
matching conditions when the field coupling becomes dimensionless. As a first check,
382
some tree-level amplitudes are derived from the string master formula. In Section 4 we
turn to one-loop diagrams, which are fairly straightforward to handle, and thus serve as a
useful preliminary to two-loop calculations. We give explicit expressions for multileg oneparticle-irreducible diagrams both for cubic and quartic interactions, and we show how
non-planar color structures (subleading in the large-N limit) are correctly reproduced in
the string framework. Finally, in Section 5, we turn to the more challenging problem of
two-loop diagrams. Different calculations are presented, up to the four-point amplitude,
both for cubic and quartic interactions. We present a detailed analysis of the two-loop
moduli space in the field theory limit, which leads us to recover the correct results including
the normalization factors. 1 Indeed, it is clear that for these more complicated diagrams
the color decomposition and the combinatoric coefficients are obtained more easily from
the string approach than from usual field theory techniques. In Section 6 we present our
conclusions, and our current assessment of the status of our method.
2. Multiloop scalar amplitudes in string theory

As we have already anticipated, in bosonic string theory it is possible to write in a
compact form a generic loop amplitude among string states, with an arbitrary number of
loops and external legs. This formula can be immediately specialized to the case where
all external states are scalars. The full, normalized, h-loop scattering amplitude [19] of M
tachyons with momenta p1 , . . . , pM , can be written as
A(h)
M (p1 , . . . , pM ) =
h+1
Y
r=1
Tr a1 aNr
2M gSM+2h2
(2)dh
(2 0 )(Md2M2d)/4
"
#2 0 pi pj
Y exp Gr(h)
(z
,
z
)
i
j
,r
i
j
[dm]M
q
.
h
Vi0 (0)Vj0 (0)
i<j
(2.1)
Here gS is the dimensionless string coupling constant, and the product of traces is the
appropriate U (N) ChanPaton factor for a generic h-loop diagram with h + 1 boundaries
(h)
labelled by the index r. In the planar case it becomes simply N h Tr(a1 aM ). Gri ,rj is
the correlator of two world-sheet bosons located at zi on the boundary labelled ri , and at
zj on the boundary rj , while [dm]M
h is the measure of integration over moduli space for an
open Riemann surface with h loops and M punctures. Notice that since we consider U (N)
as a gauge group, we have to take into account only string amplitudes with oriented worldsheets; thus non-planar diagrams arise only when loops are formed by sewing together two
non-consecutive punctures. Here we will not be interested in the exact expression of the
geometric objects appearing in Eq. (2.1), which can be found in [36]; rather, we will focus
on their general features, in order to emphasize the properties which play a crucial role in
the field theory limit. For a more complete presentation of the mathematical tools we will
1 A different method to identify the regions of moduli space corresponding to two-loop quartic interactions,
and to compute the corresponding amplitudes, has recently been introduced in Ref. [35].
383
briefly describe here, we refer to Ref. [36]. As is well known, the Schottky parametrization
is particularly suited for the study of the field theory limit, so we will work always within
this framework. In Fig. 3, in Section 4, and Fig. 6, in Section 5, we present the one- and the
two-loop string world-sheets in the Schottky representation. 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 the same radius and with 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 radius of the circles, which fix,
respectively, the position 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 denoted by and ( = 1, . . . , h), while
the width of the holes is determined by the third parameter characterizing the projective
transformation, the multiplier k . It is possible to use the projective invariance of string
theory to fix the location of up to three of the punctures zi , or of the fixed points, but
one cannot fix the multipliers, which in fact drive the field theory limit. In general, it is
convenient to fix, say, two s and one , except in the one-loop case, where one has only
two fixed points; in this case one also specifies the position of one of the punctures. In the
field theory limit, the surface must degenerate into a graph, and all massive string modes
must decouple. One can show in general [18] that the relevant region of string moduli space
is the region k 0, and furthermore the Taylor expansion of the integrand of the string
amplitude in powers of the multipliers corresponds to a sum over the mass levels of the
states circulating in the loops. Thus, for the field theory limit of scalar amplitudes, we can
always ignore all higher powers in the multipliers. In this limit, the integration measure in
Eq. (2.1) reads
!

h
M
Y
Y
dk d d
dzi
M
d/2
, (2.2)
[det(i )]
[dm]h = (a , b , c )
k2 ( )2
Vi0 (0)
=1
i=1
where the factor (a , b , c ) is the FaddeevPopov determinant associated with the

fixing of the overall projective invariance, and is the period matrix of the surface,
whose explicit expression at one and two loops will be given in Eqs. (4.6) and (5.7) below.
Note that all the dependence on the external states is concentrated in the last term. The
factors Vi0 (0) originate from the need to introduce local coordinates on the surface, Vi (z),
around each puncture, in order to perform the sewing procedure. Before discussing their
role, let us introduce explicit expressions for the Green functions we will need. Also in
this case, we report here just the leading term in the Taylor expansion in the multipliers
(see [36] for the complete string expressions). At one loop, if the two punctures are on the
same boundary, one finds [15]
G (1) (zi , zj ) = log |zi zj | +
1
zi
log2 ,
2 log k
zj
otherwise one must use the non-planar Green function
(2.3)
384

(1)
GNP (zi , zj ) = log |zi + zj | +
1
|zi |
log2
.
2 log k
|zj |
(2.4)
Notice that in Eqs. (2.3) and (2.4) we have already chosen the projective gauge = 0 and
, so that only the multiplier k appears explicitly. At two loops the bosonic Green
function becomes a little more complicated because one has to deal with a non-trivial
dependence on the moduli of the two holes. In the planar case
1
1
G (2) (z1 , z2 ) = log |z1 z2 | + log k1 log k2 log2 S
2

log2 T log k2 + log2 U log k1 2 log T log U log S .
(2.5)
The Green function now depends on two different multipliers and on four fixed points
through the anharmonic ratios
(1 2 )(1 2 )
,
(1 2 )(1 2 )
(z2 1 )(z1 1 )
,
T=
(z2 1 )(z1 1 )
(z2 2 )(z1 2 )
.
U=
(z1 2 )(z2 2 )
S=
(2.6)
As was already noticed in [15], at one loop, these Green functions do not have the
expected periodicity properties. This is not really surprising, since it is known that the
factor exp[G(zi , zj )] appearing in the master equation (2.1) has conformal weight (1/2,
1/2) in the two variables (zi , zj ) and not zero. Thus to give a global definition to Eq. (2.1),
one should multiply it by a function of conformal weight (1/2, 1/2). This suggests that one
can recover a well behaved geometric object if the local coordinates Vi0 (0) compensate for
this problem by having conformal weight 1. A natural choice is to define the Vi0 (0) by
using the inverse of the Abelian differentials, which are the only globally defined objects
having conformal weight one. By following this idea one recovers at one loop the choice
made in [15] where Vi0 (0) = 1 (zi ) = zi , since, in this case, there is a unique Abelian
differential, (z) = 1/z. At two loops, one is lead to identify the inverse of Vi0 (0) with a
linear combination of the two differentials 1 (zi ) and 2 (zi ); to fix the normalization, we
will follow Ref. [21], and require that this linear combination be normalized to one when
one integrates it around the field theory propagator on which the leg is inserted. This is
sufficient to fix the local coordinates for the purpose of the scalar field theory limit.
The string expressions reported here contain just the leading order in the multipliers;
moreover, as we anticipated in the introduction, in order to derive the field theory limit
of Eq. (2.1), one has to expand the integrand also in powers of all the other variables.
The logarithmic terms, however, are non-analytic and must have a special status. In fact, it
turns out that they measure the length of field theory propagators in units of 0 , and so are
directly related to the Schwinger proper times of the limiting field theory. This means that,
technically, the zero-slope limit has to be taken after introducing the field theory variables
that have to be kept fixed: the string coupling constant has to be translated to the appropriate
field theory coupling, while the logarithmic terms in the integrand must be interpreted
385
in terms of proper times, in general as ln (x) t/ 0 . The exact form of this change of
integration variables depends on the particular corner of moduli space considered, as we
will see in detail in the following sections.
Although we gave explicit expressions for all functions entering in Eq. (2.1), our master
formula is still a formal expression, since we have not yet specified the exact region of
integration of the various parameters. We do not attempt to solve this problem in its most
general form; however, in the following sections, we will determine the correct region of
integration, at least in the field theory limit, for the one- and two-loop diagrams. Here we
just anticipate the basic idea. One starts by considering the vacuum amplitudes, where it
is possible to focus only on the world-sheet shape, without having to consider external
punctures. In this way one determines the region of integrations over the fixed points, by
requiring that the surface never become singular; on the other hand, the integration over the
multipliers is fixed by symmetry arguments and is chosen in order to avoid double counting
of equivalent configurations. When external legs are added, this setup is not essentially
modified. From this analysis of the string world-sheet one can draw a clear representation
of the various boundaries of the surface, and the punctures can take all possible values on
these boundaries.
3. Tree-level matching conditions

We begin our analysis of the field theory limit by establishing the relationships between
the string coupling and the couplings of the cubic and quartic theories we want to
reproduce. This is done, as in effective field theory, by computing the simplest amplitudes
both with strings and fields, and matching the results. The string amplitude is, of course,
uniquely defined: different matchings correspond to different ways of taking the infinite
tension limit, and they lead to different field theories in different dimensions. Having
established the connection between the couplings, we go on to describe the computation of
simple tree diagrams, with up to six external legs.
The on-shell, tree-level, color-ordered, M-point scalar amplitude in bosonic string
theory is readily derived from Eq. (2.1), by choosing as Green function simply G(zi , zj ) =
ln |zi zj |. The result is the correctly normalized KobaNielsen amplitude,
AM (p1 , . . . , pM ) = Tr(a1 aM )2M gSM2 (2 0 )(Md2M2d)/4
Z Y
M
Y
0
dzi (za , zb , zc ) (zi zj )2 pi pj .

(0)
i=1
(3.1)
i<j
Here the punctures zi are ordered on a circle, as in the trace, and (za , zb , zc ) is the
FaddeevPopov determinant arising from the fixing of projective invariance,

(za , zb , zc ) = za za(0) zb zb(0) zc zc(0)
(za zb )(za zc )(zb zc ),
(3.2)
386
where za,b,c are three arbitrarily chosen punctures whose location is fixed. In the present
section we will always choose z1 , z2 = 1 and zM = 0, so that all remaining integrals
range between 0 and 1.
With these choices, the 3-point amplitude is simply
A3 (p1 , p2 , p3 ) = Tr(a1 a2 a3 )8gS (2 0 )(d6)/4,
(0)
(3.3)
whereas the 4-point amplitude (contributing to the Veneziano formula) is given by

a1 a2 a3 a4
2
0 (d4)/2
A(0)
4 (p1 , p2 , p3 , p4 ) = Tr( )16gS (2 )
Z1
dz z2 p3 p4 (1 z)2 p2 p3 .
(3.4)
Since we wish to consider a slightly more complicated example, we also give the
expression for the 6-point function,
A(0)
6 (p1 , . . . , p6 )
= Tr(a1 a6 )64gS4 (2 0 )(d3)
Z1
Zz3
dz3
Zz4
dz4

0
0
0
dz5 (1 z3 )2 p2 p3 (1 z4 )2 p2 p4 (1 z5 )2 p2 p5
(z3 z4 )2 p3 p4 (z3 z5 )2 p3 p5 (z4 z5 )2 p4 p5

2 0 p3 p6 2 0 p4 p6 2 0 p5 p6
z4
z5
.
z3
(3.5)
Turning to field theory, let us first consider a scalar U (N) theory defined by the
Lagrangian 2

2
2 2
3
L3 = Tr m + g3 ,
(3.6)
3
where = is a scalar field taking values in the adjoint representation of U (N),
and our generators are normalized by Tr( ) = /2. Feynman rules for scalar U (N)
theories and several useful formulas for color structures are collected in the appendix. The
color-ordered 3-point amplitude (defined as i times the relevant Feynman diagram) in
this theory is simply
a1 a2 a3
A(0)
3 (p1 , p2 , p3 ) = 2g3 Tr( ).
(3.7)
Comparison with Eq. (3.3) yields the matching condition

g3 = 4gS (2 0 )(d6)/4,
(3.8)
2 Note that here we use a convention slightly different from the one employed in Ref. [19]; in particular we
choose a coupling constant g3 which is four times bigger than the one of [19], in order to reduce the number of
factors of two in the amplitudes.
387
already derived in [19]. Note that the coupling is dimensionless in d = 6, as it must. We can
now use Eq. (3.8) to compute higher order scattering amplitudes in the theory defined by
Eq. (3.6), using the string master formula. As a first simple example, consider the 4-point
amplitude, which becomes
a1
a4
2 0
A(0)
4 (p1 , . . . , p4 ) = Tr( )2g3
Z1
dz z2 p3 p4 (1 z)2 p2 p3 .
(3.9)
To get a sensible zero slope limit we must extract from the integral a factor ( 0 )1 . This can
be done by focusing on the potentially singular regions z 0 and z 1, corresponding
to s- and t-channel exchange, respectively (the u-channel diagram cannot contribute to
this color structure). According to the general discussion of Section 2, and considering the
region z 0, we do this by setting z = exp(tz / 0 ), with tz finite and 0 0. Neglecting
O( 0 ) corrections, the change of variables yields
a1
A(0)
4 (p1 , . . . , p4 ) = Tr(
a4
)2g32

tz
0
dtz exp 0 1 + 2 p3 p4
= 2g32 Tr(a1
)
a4
1
,
(p3 + p4 )2 + m2
(3.10)
where we made use of the mass-shell condition m2 = 1/ 0 . Taking into account the
similar contribution arising from the region z 1, we get the complete answer for this
color ordering,

1
1
2
a1
a4
(p
,
.
.
.
,
p
)
=
2g
Tr(
)
+
, (3.11)
A(0)
1
4
3
4
(p3 + p4 )2 + m2 (p2 + p3 )2 + m2
which exactly matches the correct result in field theory, provided the string metric is used.
It is clear that the different diagrams contributing to a given color ordering arise from
different corners of string moduli space, and they are easily identified by the pole structure
of their propagators. To illustrate this in a slightly less trivial configuration, let us consider
the 6-point amplitude in Eq. (3.5), and let us attempt to separate the contributions to two
different given diagrams, say those portrayed in Fig. 1. Using the matching condition,
Eq. (3.8), one easily sees that the string amplitude is proportional to ( 0 )3 , so that we need
to take a limit in all three integration variables, which corresponds to the extraction of three
Fig. 1. Two tree-level six-point diagrams with cubic vertices.
388
propagator poles. If we want to specify the region in moduli space relevant for diagram A,
we must qualitatively have z5 very close to z6 = 0, and z3 very close to z4 . Moreover the
three pairs of variables (z1 , z2 ), (z3 , z4 ) and (z5 , z6 ) have to be kept widely separated from
each other. The desired change of variables is
0
z3 = et3 / ,
z3 z4 = et4 / ,
z5 = et5 / .
(3.12)
In terms of these proper times, and neglecting terms suppressed by powers of 0 , the
amplitude reads
A(0)
6 (p1 , . . . , p6 )

t4
0
dt3 dt4 dt5 exp 0 1 + 2 p3 p4
= Tr(
t3
t3
0

t3
t5
exp 0 1 + 2 0 p5 p6 exp 0 1 + 2 0 p3 p5 + 2 0 p3 p6

+ 2 0 p4 p5 + 2 0 p4 p6 . (3.13)
a1
a6
)2g34
One easily sees that the contribution of this region to the 6-point amplitude is
a1
a6
4
A(0)
6 = Tr( )2g3
1
1
1
,
2
2
2
2
(p1 + p2 ) + m (p3 + p4 ) + m (p5 + p6 )2 + m2
(3.14)
precisely the desired result.

For the other diagram of Fig. 1, one has to consider a different change of variables; in
particular, since now we do not want to group external particles in pairs, all zi s must
be taken widely separated. Thus the corresponding ordered proper times are simply
defined as ti = 0 ln zi . Following the same steps just described, it is the easy to check
that the expected result for diagram B is obtained. We note in passing that finding the
numerical coefficient of a given color ordering for a given Feynman diagram may be a
rather cumbersome task with the conventional Feynman rules, whereas it is immediate
here.
Let us now turn our attention to quartic interactions. We want to reconstruct the
amplitudes derived from the Lagrangian

(3.15)
L4 = Tr m2 2 + g4 4 ,
which, in particular, yields the color-ordered vertex
(0)
A4 (p1 , p2 , p3 , p4 ) = 4g4 Tr(a1 a2 a3 a4 ).
(3.16)
The starting point is now Eq. (3.4), where however in this case we do not need to generate
any extra powers of 0 , as was done to go from Eq. (3.9) to Eq. (3.10). Here the overall
dimensionality is correct, so all we need to do is take the 0 0 limit and integrate over z
without introducing any weight in special corners of moduli space. In other words, the
quartic vertex arises from integration over finite regions of moduli space, and not from its
389
boundaries (regions of infinitesimal size in the field theory limit). As 0 0, the integrand
of Eq. (3.4) simply becomes 1, so we get
A4 (p1 , p2 , p3 , p4 ) = 16gS2 (2 0 )(d4)/2 Tr(a1 a2 a3 a4 ).
(0)
(3.17)
Comparison with Eq. (3.16) yields the matching condition

g4 = 4gS2 (2 0 )(d4)/2.
(3.18)
The same matching condition might have been obtained in a different way, by first
considering the 3 diagrams contributing to A(0)
4 , and then explicitly deleting the internal
propagators by setting their proper times to 0, inserting in turn a (tz / 0 ) and a ((t
tz )/ 0 ). These functions should however be regularized, since they are located at the
boundaries of the integration region. In such circumstances it is natural to weigh each
function with a factor 1/2, and this reproduces the matching in Eq. (3.18). This second
method is closer in spirit to the techniques of [35].
To show that this procedure can be generalized to higher order amplitudes, let us
consider also for 4 a particular diagram contributing to the 6-point amplitude. Note that
there are 105 diagrams contributing to the 3 6-point amplitude, but only 10 with quartic
vertices. We consider, as an example, the diagram depicted in Fig. 2. Using the matching
condition Eq. (3.18) in the string amplitude, Eq. (3.5), one sees that the amplitude has
an overall factor of 0 . Thus we need to take precisely one singular limit in one of the
integration variables, corresponding to the extraction of a single propagator pole. For the
particular diagram at hand, qualitatively, we would like to keep z4 and z5 close to z6 = 0,
while leaving z3 close to 1. This can be achieved by the change of variables
z3 = x,
z4 = et4 / ,
z5 = yet4 / ,
(3.19)
with no proper times associated with x and y. Neglecting as usual terms suppressed by
powers of 0 , we get
a1
A(0)
6 (p1 , . . . , p6 ) = Tr(
Z1
a6
)8g42
Z1
dx
Z
dy
dt4
0

t4
exp 0 2 + 2 0 p4 p5 + 2 0 p4 p6 + 2 0 p5 p6
1
.
(3.20)
= Tr(a1 a6 )8g42
(p4 + p5 + p6 )2 + m2
Fig. 2. A sample tree-level six-point diagram with quartic vertices.
390
Once again, the computation of the color factor of the corresponding Feynman diagram
yields the same result.
4. One-loop diagrams
The multitachyon planar one-loop amplitude derived from Eq. (2.1) can be written as
(1)
1
2M gSM (2 0 )(Md2M2d)/4
(4)d/2
zZM1

Z1
Z1
Zz2
dk
log k d/2
dz2 dz3
dzM
2
k2
AM (p1 , . . . , pM ) = N Tr(a1 aM )
M
Y
"
Y exp G (1) (zi , zj )

q
Vi0 (0) Vj0 (0)
#2 0 pi pj
V 0 (0)
i=1 i
i<j
(4.1)
where we neglected O(k) terms in the measure of integration that will not contribute to
the field theory limit, and we have introduced the local coordinates Vi (z), according to the
general discussion of Section 2. Projective invariance has been used to choose the fixed
points of the single Schottky generator as = 0 and , and to fix z1 = 1. In this
configuration the world-sheet of the string (an annulus) can be represented as in Fig. 3.
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 0 = 1/ k, with the restriction on the ordering implied by
the color trace. This would complicate the
limit, since there
calculation of the field theory
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 thestring integrand is
modular invariant, which
in particular implies that the interval [1, 1/ k ] can be mapped
onto the interval [k, k]. In fact, defining the effective one-loop Green function by
1
1
log Vi0 (0) log Vj0 (0),
(4.2)
2
2
and choosing Vi0 (0) = zi , according to our general discussion, it is easy to check that the
effective Green function at the string level is a function only of the ratio ij = zi /zj , and
satisfies
G(1) (zi , zj ; k) = G (1) (zi , zj )
G(1) (j i , k) = G(1) (ij , k),

G(1) (kj i , k) = G(1) (ij , k).
Fig. 3. The annulus in the Schottky representation.
(4.3)
391
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. (4.1). In the field theory (k 0) limit the effective Green function has the
form
r
r
zj
zi
zi
1
(1)
log2 ,
(4.4)
+
G (zi , zj ; k) = log
zj
zi
2 log k
zj
for zi > zj .
The generalization of Eq. (4.1) to the case of non-planar diagrams is known [37], and
easily understood. A non planar diagram has punctures on both boundaries, so the factor of
N = Tr 1 is replaced by the trace of the ChanPaton
factors corresponding to the punctures
on the second boundary (the interval [1/ k, k ] in Fig. 3). The region of integration
of the corresponding zi is an ordered region on the negative real axis. The string Green
function involving two punctures on the negative real axis is precisely the same as the one
discussed above, since it is a function only of the ratio ij , which remains positive when
both zs change sign. The only subtlety involves the Green function connecting punctures
on different boundaries. In this case the terms in the Green function arising from loop
momentum integration behave differently, and one should choose Vi0 (0) = |zi |, so that the
Green function remains real, and does not have any singularity when |zi | |zj |. In the
field theory limit one finds simply
s
s
!
|zj |
1
|zi |
|zi |
(1)
+
+
log2
,
(4.5)
GNP (zi , zj ; k) = log
|zj |
|zi |
2 log k
|zj |
where the two zs have opposite signs.
4.1. One-loop cubic interactions
Armed with the appropriate string technology, let us examine how one-loop scalar
diagrams with cubic interactions emerge from Eq. (4.1). Following [19], we begin by
considering the general configuration of a one-loop one-particle-irreducible diagram with
n external legs, depicted in Fig. 4. Using the matching condition Eq. (3.8), we find
(1)
g3n
(2 0 )d/2 ( 0 )n
(4)d/2
zZn1

Z1
Z1
Zz2
dk
dz3
dzn
log k d/2 dz2
2
z2
z3
zn
k2
0
k
k
k
"
#
X

0
(1)
exp
2 pi pj G (zi , zj ) .
An,1P I (p1 , . . . , pn ) = N Tr(a1 an )
(4.6)
i<j
As usual, the field theory limit is governed by the multiplier k, which must be taken to
be exponentially suppressed as 0 0, so that the length of the string loop may become
392
Fig. 4. Multileg one-loop one-particle-irreducible diagram in a 3 theory.
infinite in units of 0 . The tachyon double pole is regulated, as described in Ref. [19], by
setting dk/k 2 = exp(m2 0 log k) dk/k. The appropriate change of variables is then
0
k = eT / ,
zi = eti / .
(4.7)
All puncture coordinates must become exponentially small as 0 0, so that the correct
power of 0 may be generated. Defining the Feynman parameters as xi = tn+1i /T , for i =
1, . . . , n 1, we find an expression which may easily be compared with the field-theoretic
result,
(1)
An,1P I (p1 , . . . , pn ) = N Tr(a1 an )

Z
dT T
0
g3n
(4)d/2
n1d/2 m2 T
Z1
xZn2
dx2
dx1
0
"
Zx1
0
#
X

exp T
pi pj xij (1 xij ) ,
dxn1
0
(4.8)
i<j
where xij = xi xj . We note in passing that deriving the coefficient of the leading color
trace from the Feynman rules is not trivial for the general diagram with n legs. In fact,
the above could be considered as a simple proof that the coefficient is N for any n. The
expression for the integrand as a function of the parameters xi also appears automatically
in the most symmetric form; note that this is the correct form for arbitrary values of the
external momenta pi , on- or off-shell.
Before going on to perform a similar calculation for quartic interactions, it is worthwhile
to pause to consider two instructive special cases of Eq. (4.8), and a slight generalization of
it in the case of the four-point function. First of all we would like to point out that Eq. (4.8)
holds in field theory for all n, including n = 2, because of what appears at first sight as a
fortunate coincidence: in fact in field theory the color factor for the two-point function is
twice the one appearing in Eq. (4.8), essentially because Tr(a b ) = Tr(b a ). However
this factor of 2 is compensated by a symmetry factor of 1/2, which is only present for the
two-point amplitude. String theory takes into account these two facts simultaneously and
automatically.
Another unexpected feature of the two-point function in field theory is the fact that,
if one allows the external scalars to take values in the U (1) factor of U (N), that is if
393
one allows the indices a, b to take the value 0, one finds that the color factor of the
corresponding diagram doubles. This is easily seen from Eqs. (A.10) and (A.12), in the
appendix. This fact is unexplained from the point of view of Feynman diagrams, but it
has a natural explanation in string theory. In fact, if in string theory we allow the external
legs to be color singlets, the amplitude receives a contribution from a new diagram, the one
with the two legs inserted on different world-sheet boundaries. This diagram, which is nonplanar from the point of view of string theory, contributes at the same order in N because
with a correctly normalized U (1) generator (see the appendix) on finds that Tr(20 )Tr(1) =
(Tr(0 ))2 . Furthermore, it is easy to check that the functional form of this new diagram
is precisely the same as that of the original diagram, with the same overall factor and
integration region. This is the first simple example of the correct handling of non-planar
contribution in string theory.
We conclude this section by giving a further non-trivial example of a non-planar contribution to a one-loop amplitude. We consider the four-point function, which in field theory yields contributions proportional to double color traces, such as Tr(a1 a4 )Tr(a2 a3 ).
These double trace contributions naturally arise in string theory from the simultaneous insertion of punctures on the two different string boundaries. In field theory, on the other
hand, these terms receive contributions from different Feynman diagrams. Choosing for
example the cyclic order (1, 2, 3, 4) for the external legs, the complete color factor in field
theory is given by

C1234 = N Tr(a1 a2 a3 a4 ) + Tr(a4 a3 a2 a1 )
+ 2Tr(a1 a2 )Tr(a3 a4 ) + 2Tr(a1 a3 )Tr(a2 a4 )
+ 2Tr(a1 a4 )Tr(a2 a3 ),
(4.9)
so that the coefficient of the chosen term is 2. One should keep in mind, however, that
the particular color structure Tr(a1 a4 )Tr(a2 a3 ) receives contributions also from two
other distinct Feynman diagrams, corresponding to the non-cyclic permutations of the
external particles in the original one (in the present case, the orderings (1, 3, 2, 4) and
(1, 4, 2, 3)). String theory must, and does, assemble the contributions of the different
diagrams to the chosen color structure into a single string configuration. This can be
verified by using the non-planar Green function Eq. (4.5), and considering all possible
ways of inserting the punctures on the boundaries. Since we wish to place the puncture z4
on the same boundary as z1 = 1, it must lie on the positive real axis in Fig. 3. As explained
above, this leads to the integration region k 6 z4 6 1. For the other two punctures, on
the negative axis, the integration region is 1 6 {z2 , z3 } 6 k, with no restriction on
the relative ordering of z2 and z3 . Changing variables to x2 = z2 and x3 = z3 , we
have once again four punctures on the positive real axis, which can be placed in the
interval [k, z1 = 1] in six different orderings. Note that the six orderings are distinguishable
in the field theory limit, because the logarithmic term in the generic Green function
has a different field theory limit depending on the ordering, log(zi zj ) log(zi ),
for zi > zj . An explicit calculation shows that the orderings x2 6 x3 6 z4 and z4 6 x3
6 x2 conspire to reconstruct the contribution to the chosen trace of the field theory diagram
with cyclic ordering (1, 2, 3, 4), with the correct overall factor of 2. Similarly, the orderings
394
x2 6 z4 6 x3 and x3 6 z4 6 x2 reconstruct the contribution of the diagram with cyclic

ordering (1, 2, 4, 3), while the remaining two orderings give the last diagram. Once again,
the building blocks of the final result are assembled in novel and non-trivial way.
4.2. One-loop quartic interactions
As in the previous section, it is possible to give a general expression for the oneloop, color-ordered, one-particle-irreducible diagram with n external legs, but only quartic
vertices, shown in Fig. 5. Using the matching condition Eq. (3.18) in Eq. (4.1), we get an
expression very similar to Eq. (4.6),
n/2
g4
(2 0 )n/2d/2
(4)d/2
zZn1

Z1
Z1
Zz2
dk
dz3
dzn
log k d/2 dz2
2
z2
z3
zn
k2
0
k
k
k
"
#
X

0
(1)
exp
2 pi pj G (zi , zj ) .
(1)
An,1P I (p1 , . . . , pn ) = N Tr(a1 an )
(4.10)
i<j
It is clear, however, that in this case we must introduce proper times for only one half of
the integration variables, in order to compensate for the overall factor of ( 0 )n/2 . To obtain
the configuration shown in Fig. 5, the appropriate change of variables is
0
zi = eti / ,
(i odd)
zi = yi zi1 ,
(i even)
k=e
T / 0
(4.11)
with no proper time associated to the variables yi and, as usual, z1 = 1. Using Eq. (4.11)
and neglecting terms suppressed as 0 0, one gets
(1)
An,1P I (p1 , . . . , pn )
(2 g4 )n/2
(4)d/2
Z
Z1
ZT
Ztn3
Z1
dy2
dyn
d/2 m2 T
dT T
e
dt3 dtn1
y2
yn
0
0
0
0
0
"

#
X
(tj ti )2
exp
(pi + pi+1 ) (pj + pj +1 ) tj ti
.
T
= N Tr(a1 an )
(4.12)
i<j odd
The integrals in dyi , ranging over finite regions of moduli space, are logarithmically
divergent and need to be regularized. This divergence is actually a further manifestation
of the tachyonic nature of the bosonic string ground state, and must be dealt with by
the same method used to handle the double pole dk/k 2 : one must substitute yi1
395
Fig. 5. Multileg one-loop one-particle-irreducible diagram in a 4 theory.
exp(m2 0 log yi ). In this case, however, there is no proper time associated with yi so the
exponential factors are suppressed as 0 0 and must be neglected. The product of all
dyi integrals then yields simply a factor of unity. This procedure can be further justified
by noting that the factors of yi1 arise from the factors (Vi0 (0))1 in Eq. (4.1), which are
characteristic of tachyon propagation and are absent, for example, in gluon amplitudes.
In fact Eq. (4.1) can be consistently continued to an arbitrary value of the string intercept
a = m2 0 6= 1, as was done in [38]. In that case one finds that the measure of integration
Q
Q
changes as dk/(k 2 i Vi0 (0)) (dk/k)(k i Vi0 (0))a , indicating that the singularities
generated by the projective transformations Vi0 (0) are of the same type as the double pole
in the multiplier k.
With this regularization and changing variables to xi = 1 ti /T , we find the simple
formula
A(1)
n,1P I (p1 , . . . , pn )
n/2
g
= 2 N Tr( ) 4 d/2
(4)
xZn3
Z
Z1
Zx3
(n2d)/2 m2 T
dT T
e
dx3 dx5
dxn1
n/2
a1
"
exp T
an
(pi + pi+1 ) (pj + pj +1 ) xij (1 xij )
#
(4.13)
i<j odd
Once again, this matches the result arising from the Feynman diagram in Fig. 5, with the
restriction that paired external legs must not be of the same color. This provides a simple
proof of the fact that for such diagrams the coefficient of the leading color trace is 2n/2 N .
We conclude this section by noting a special case that arises in the computation of these
loop amplitudes, when two consecutive external particles (2n 1, 2n) annihilate in two
colorless states running in the loop. Let us, for instance, focus on the special case of
Eq. (4.13) where n = 2; this of course represents a tadpole diagram. For this diagram
Eq. (4.13) yields
(1)
A2,1P I (p1 , p2 ) = 2N
g4
Tr( )
(4)d/2
a b
dT T d/2 em T ,
2
(4.14)
396
which is one half of the result obtained in field theory. This discrepancy can be understood
by observing that the color factor in field theory is of the form
Cab = 8N Tr(a b ) = dab d + 2da db .
(4.15)
The first contraction of d-tensors, which contributes one half of the total result, represents
an anomalous color flow corresponding to a 3 tadpole diagram, in which the colored
scalars a and b = a annihilate into a color singlet state which then self-interacts. Clearly in
this channel the full U (N) color flow is prohibited, and only the U (1) factor contributes.
This term is missed by Eq. (4.13), but can be reproduced by first generating a one-particlereducible 3 diagram and then deleting the zero-momentum propagator, thus attaching the
loop to the external legs. A similar peculiarity will arise when we will consider the twoloop 4 vacuum bubbles. Also in that case a tadpole-like configuration forcing the color
flow to be restricted to U (1) is present. This kind of term is always related to corners of
moduli space corresponding to one-particle-reducible diagrams, and this is signaled by a
color factor which displays a combinations of d symbols typical of these diagrams.
5. Two-loop diagrams
Generalizing the approach described in the previous section to the two-loop case is
not a straightforward task. First, the explicit expressions of the measure and of the other
geometrical objects present in Eq. (2.1) are more complex, so that the computation of
physically interesting quantities (such as YangMills amplitudes) has to be performed
by means of a computer program. Second, there are also conceptual novelties, and the
procedure described in the previous section cannot be applied directly to a two-loop
calculation. Many of the new features are related to the fact that now the string world-sheet
is a two-annulus and some of the simplifying choices that are usually made at one-loop are
not possible any more. For example, in Eq. (4.1) the fixed points played no role, because
they could be gauge-fixed to zero and infinity. In two-loop calculations, on the other hand,
the shape of the string world-sheet can vary in a non-trivial manner. In fact, the measure
Eq. (2.2) crucially depends on at least one of the fixed points, which means that the relative
position of the two holes cannot be fixed. Thus, for a better understanding of the new
geometrical features, it is worthwhile to start the study of two-loop string amplitudes by
considering in detail the Schottky parametrization, as it arises from the sewing procedure
leading to Eq. (2.1). Basically, the two-loop surface can be constructed starting at one loop
and identifying two external legs. This is done by cutting away from the one-loop string
world-sheet two circles, and identifying their boundaries. If one chooses to sew together
the puncture fixed at z1 = 1 with one of the other legs on the same boundary (i.e., with
zi > 0 in order to construct a planar diagram), one obtains the two-loop surface depicted
in Fig. 6.
We fix projective invariance by choosing 2 = 0, 2 and 1 = 1. Then the positions
and radii of the circles in Fig. 6 are completely determined as functions of the multipliers
k1 , k2 and of the fixed point 1 ; in fact, following [19], one can verify that
397
Fig. 6. In the Schottky parametrization, the two-annulus corresponds to the part of the upper-half
plane which is inside the big circle passing through A0 and B 0 , and which is outside the circles K1 ,
K10 and K2 .
B = A =
k2 ,
1 + 1 k1
,
D0 =
1 + k1
1 k1
,
C=
1 k1
1 1 k1
,
C0 =
1 k1
1 + k1
,
D=
1 + k1
1
B 0 = A0 = .
k2
(5.1)
(5.2)
One can check that the points A, B, C and D are identified with A0 , B 0 , C 0 and D 0 ,
respectively, under the action of the two generators of the two-loop Schottky group, i.e.,
0 . By cutting
the projective transformations mapping the circles K1,2 into their images K1,2
0
open a two-annulus, it is possible to map the two segments (AA ) and (DD 0 ) onto the
inner boundaries of the world-sheet, which is natural since their length depends only on k1
and k2 . Then the union of (BC) and (C 0 B 0 ) represents the external boundary. Note that,
in order to avoid a degenerate surface, the various identified circles should not overlap.
This simple consideration gives a first constraint on the region of integration of the string
moduli. In fact, by requiring that B does not touch C one obtains
p
p
p
(5.3)
> k1 + k2 k1 k2 ,
while the same requirement on the segment (DD 0 ) leads to
1 k1
(1 1 ),
D0 D =
1 + k1
(5.4)
so that < 1. Now the interpretation of the three moduli, k1 , k2 and 1 , is particularly
simple. In fact, k2 is the radius of the circle K2 , while the radii of K1 and K10 are equal
and depend on k1
p 1 1
RK1 = k1
.
(5.5)
1 k1
Furthermore, 1 turns out to be inside K1 , while the point 1 = 1 is inside K10 . Therefore,
in this configuration, the circle K10 is fixed while K1 can move, depending on the value
of 1 . In particular, if the point D 0 is very close to D, 1 is almost equal to 1, while if C is
near to B, then 1 is of order of k1 + k2 . In other words, it is possible to interpret 1

as the distance between the two loops. When 1 1, one may expect the field theory
limit (at least for cubic interactions) to yield reducible diagrams with the two loops widely
separated, while when 1 k1 + k2 0 one should obtain the irreducible diagrams

with the two loops attached to each other.
Another observation relevant for deriving the region of integration of world-sheet moduli
is that in the explicit form of all geometrical objects (for instance, the Green function or
the measure) the multipliers k1 and k2 appear symmetrically, reflecting the equivalence of
the two loops. Therefore, in order to avoid double counting of equivalent configurations,
398
one can order the multipliers, by choosing for example k2 < k1 . Note that the multipliers
will always be associated with field theory proper times, so in the zero-slope limit their
ordering should always be interpreted as strong ordering.
A final remark has to be made about the possibility of taking the attractive fixed point
bigger in modulus than the repulsive one . As was shown in [39], in the closed string case
these configurations are related to the one with || < | | by the residual part of the modular
invariance group which survives in the field theory limit. In the open string case there is
no modular invariance, but these surfaces should describe the propagation of closed string
states, and they should not be included in our analysis. Thus, for our purposes, we will
always restrict || to be less than | |. Summarizing, we have found for k1 , k2 and 1 , in
the field theory limit, the same region of integration derived by Roland [39] for the closed
string, i.e., 0 6 k2 < k1 < |1 | 6 1. We are now ready to move on to the evaluation of
two-loop diagrams, starting with the simplest ones to verify our assumptions.
5.1. Vacuum bubbles
Let us start by briefly describing the simplest amplitude, the two-loop vacuum bubble
with cubic interaction. In this case (M = 0 and h = 2), with the projective gauge choice and
integration region described above, and using the matching condition for cubic interaction,
Eq. (3.8), Eq. (2.1) simply becomes
(2)
A0
N 3 g32
(2 0 )3d
=
(4)d 16

Z1
0
d1
(1 1 )2
1
log k1 log k2 log2 1
4
Z1
0
d/2
dk1
k12
Zk1
0
dk2
k22
(5.6)
In Eq. (5.6) only the determinant of the period matrix is needed; at leading order in the two
multipliers, it is

1
(5.7)
log k1 log k2 log2 .
2
4
Note that, as expected, the two multipliers k1 and k2 play the same role and all the
expressions are symmetrical in the exchange of k1 and k2 .
As discussed in the previous section, we expect a contribution to the field theory result
from the limit 1 1 (together with k1 , k2 0); in this case, the appropriate change of
variables is
det(i ) =
t1 = 0 log k1 ,
t2 = 0 log k2 ,
t3 = 0 log(1 1 ).
(5.8)
Introducing the mass m2 to regulate quadratic poles in the usual way, and neglecting terms
suppressed as 0 0, Eq. (5.6) in this region yields
(2)
A0,red
N 3 g32
=
(4)d 2
Z
dt3
Zt2
dt2
dt1 em
2 (t
1 +t2 +t3 )
(t1 t2 )d/2 .
(5.9)
399
Since Eq. (5.9) is symmetrical in t1 and t2 , it is possible to perform the integration over t1
and t2 independently from 0 to by introducing a factor of 1/2. In this way one obtains
the same result of the reducible vacuum bubble of the 3 field theory defined by Eq. (3.6),
including the correct normalization. Again, the factor of 1/4 in the normalization of this
diagram in field theory is a combination of color and symmetry factors, which are unified
in the present approach.
The second expected contribution comes from the limit 1 0 and should give the
irreducible vacuum diagram. In this case each loop is made of two different propagators,
and this suggests a different identification between field theory proper times and string
variables. In fact, the experience acquired at one loop leads us to expect that the multipliers
ki should be associated with the length of an entire loop. Thus we set
k2
k1
0
0
0
= et1 / ,
q2 =
= et2 / ,
q3 = 1 = et3 / .
(5.10)
q1 =
1
1
With this choice, Eq. (5.6) becomes
(2)
A0,irr
N 3 g32
=
(4)d 2
Zt3
dt3
Zt2
dt2
(t1 t2 + t1 t3 + t2 t3 )
0
d/2
dt1 em
2 (t
1 +t2 +t3 )
(5.11)
Since Eq. (5.11) is completely symmetrical, we can introduce a factor of 1/3! and perform
the three integrals independently from 0 to . In this way one correctly reproduces the
irreducible vacuum bubble of our 3 theory. Note that by using a single starting formula,
Eq. (5.6), we have been able to obtain two diagrams which have a different weight. While
in field theory this relative factor of 3 between the two vacuum bubble amplitudes is due
to the combination of different combinatorial and color factors, in the string approach this
relative normalization appears because Eq. (5.6) has different symmetry properties in the
two regions of moduli space that yield the two vacuum bubbles.
It is interesting to see what happens if one considers also non-planar string worldsheets. From Fig. 6, one can see that a non-planar surface may arise if the circle K1 is
centered on the negative axis. The identifications established in the planar case still hold
and by following them it is easy to realize that the surface has only one boundary. When
approaches to zero, the calculations follows exactly the same pattern we have just seen and
the result is again Eq. (5.11), but with a factor of N only, instead of N 3 , since here we have
only one boundary. If however one tries to mimic the limit which gave the reducible bubble,
that is || 1, one sees that there is no singularity in Eq. (5.6). In fact, now approaches
1 and it is not possible to associate a Schwinger proper time to the combination (1 ):
this corner of moduli space gives a vanishing contribution showing that only the irreducible
bubble receives subleading corrections in N . By using the equations given in appendix, it
is easy to recover the same result from a field theory analysis.
We conclude our discussion of two-loop vacuum bubbles by briefly considering the
single diagram arising in the case of quartic interaction. Here we see at work the
mechanism that was suggested for the one-loop 4 tadpole, in Section 4. In fact, the field
theory color factor of the vacuum bubble is of the form
400
C = d
d + 2 d d

= 2N 3 + 2N N 2 + 1 ,
(5.12)
where we emphasized the different origin of the various factors of N . The symmetry factor,
on the other hand, is 1/8. Using string theory, and the matching condition Eq. (3.18),
one sees that the overall normalization of this diagram in the planar case is N 3 g4 . Now,
however, one must introduce proper times only for the two multipliers, since there are only
two propagators in the diagram. Thus the overall symmetry factor for the completion of the
integration region is 1/2 instead of 1/6, for both color flow patterns. If we proposed to add
them, it would appear that the string gives a result too big by a factor of two. The solution
to this puzzle lies in the observation that in both configurations one needs to integrate over
a finite range in , extending to 1, as in Eq. (5.6). One must regulate the singularity as
1, without having a proper time associated with 1 . Singularities of this kind were
studied in Refs. [15,22], and the correct prescription (a function regularization) turns
out to be that these integrals yield precisely a factor of 1/2. Notice that the non-planar
contribution can arise only from the irreducible diagram, which justifies the fact that the
coefficient of N in Eq. (5.12) is 1/2 of the coefficient of N 3 .
5.2. Two-point amplitudes
In this section, scalar amplitudes with two external states are considered; in particular,
the irreducible diagrams of Fig. 7 are derived directly from Eq. (2.1), without using the
simplified procedure described in [19]. Instead, we follow the procedure just employed
for the vacuum bubbles and, for each diagram, we look at the appropriate corners of the
integration region of the various string variables. The new ingredients needed in Eq. (2.1)
in presence of external particles are the two-loop bosonic Green function and, if one wants
to extrapolate the result off-shell, the new expressions of the local coordinates Vi0 (0). It
was shown in [19,20] that from the Green function in Eq. (2.5) the general structure of
two-loop 3 diagrams can be recovered. Moreover Roland and Sato showed in [20] that
the string master formula (2.1) reduces to the particle theory amplitudes in the world-line
approach, even in the multiloop case. However, in those papers there is no derivation for the
region of integration over the punctures zi ; this should be given in terms of the parameters
determining the shape of two-loop world-sheet. Lacking this information, it is difficult
to fix the correct normalizations of the various diagrams, since different corners of the
region of integration over the punctures can contribute to the same field theory diagram.
We will now show that, in order to determine the integration region for the punctures zi , it
Fig. 7. Irreducible two-loop diagrams contributing to the two-point function of the 3 theory.
401
is sufficient to use the simple geometrical description of Fig. 6, and the analysis of vacuum
amplitudes outlined in the previous sections. In fact, as discussed in the introduction,
a general feature of the string approach is that the calculations of diagrams with different
number of external legs are closely related to each other. For instance, it is clear from
the analysis of vacuum bubbles that, in order to construct irreducible diagrams, the limit
1 0 must be considered. At this point one can freeze the world-sheet shape and put the
punctures in all possible configurations on the three boundaries, remembering that, since
we are interested in planar amplitudes, both states must lie on the same boundary of the
two-annulus.
Let us start by considering the case in which the punctures are inserted on the internal
boundary represented by the segment (AA0 ). They should then be integrated in the interval

[1/ k2 , k2 ]. Analogously, if they are on the other internal boundary, the region of
integration is the interval [D, D 0 ], while for the contributions of the external boundary each
zi must be allowed to range between B and C, and between C 0 and B 0 . In the configuration
with small multipliers (qi 0) the world-sheet becomes a graph, and each boundary
degenerates into the union of two distinct field theory propagators. It is interesting to note
that it is possible to identify the specific corners of moduli space associated with each
first quantized field theory diagram, already at the string level. In fact, if z [A0 , A], the
contributions obtained correspond to a diagram with the external particles always attached
to the first loop; but it turns out that they are emitted from the propagator shared by the
two loops if z [1, 1 ], while the intervals z [1 , A] and z [A0 , 1] correspond
to emission from the propagator not shared with the second loop.
In order to show that this identification is correct, let us calculate the two-loop diagram
with one external state in the region [1, 1 ] and the other in the region [A0 , 1] that
should contribute to the first diagram of Fig. 7. As for the local coordinates, here we impose
again that Vi0 (0) = |zi |, since the punctures are on a boundary that, from the point of view
of the Schottky parametrization, is identical to the one encountered in the one-loop case. Of
course, other choices lead to the correct result [19,21] and only a YangMills calculation,
where higher-order terms in qi are relevant, can discriminate among the various options. In
this region, the variable U of Eq. (2.6) can be approximated as U |z1 |1 . Thus from
Eq. (2.1) one can derive a first contribution to the diagram in Fig. 7, by treating the
quadratic poles of the qi variables in the usual way, and by introducing proper times ti
as in Eq. (5.10). One finds
N 2 g34 ab
A(2)
a (p1 , p2 ) =
(4)d 4
Zt2
dt2
m2 (t1 +t2 +t3 )
dt1
0
d/2
1
2 (t2 +t3 )
Zt1
dt3
0
Zt3
dtz2
dtz1
0

exp p1 p2 Ga (t1 , t2 , t3 , tz1 , tz2 ) ,
(5.13)
where the Green function for this diagram is given by

Ga (t1 , t2 , t3 , tz1 , tz2 ) = tz1 + tz2 1 (t1 + t2 )tz22 + (t1 + t3 )tz21 + 2tz1 tz2 t1 , (5.14)
and where = (t1 t2 + t1 t3 + t2 t3 ). The Schwinger proper times tzi are related, as expected,
to the positions of the punctures by tzi = 0 log |zi |. It can be checked that the integrand
402
of Eq. (5.13) is exactly what one expects from a field theory calculation, but the region
of integration over proper times appears strange. In fact, the whole expression is not
symmetric in the exchange of any two proper times, unlike the case of vacuum diagrams,
Eq. (5.11). Thus, it is not possible to immediately complete the integration region of t1 and
t3 extending it to infinity; moreover, from the field theory analysis one would expect the
proper time tz1 to vary between 0 and t2 , while the string result covers only part of this
region.
These problems are treated by taking into account also the other regions of the
integration over the punctures that contribute to the field theory diagram of Fig. 6. For
instance, another configuration that should be considered is z2 [, k2 ] with z1

remaining in the same interval as before. In this case U can be approximated as U /|z2 |,
and from Eq. (2.1) one gets
N 2 g34 ab
(2)
Ab =
(4)d 4
Z
dt2
0
d/2
Zt2
dt1
0
1
2 (t2 +t3 )
Zt1
dt3
0
Zt3
dtz1
t3
dtz2 em
2 (t
1 +t2 +t3 )
exp p1 p2 Gb (t1 , t2 , t3 , tz1 , tz2 ) ,
(5.15)
where

Gb (t1 , t2 , t3 , tz1 , tz2 ) = tz1 tz2 1 (t2 + t3 )(t3 tz2 )2

+ (t1 + t3 )(tz1 tz2 )2 2(tz2 tz1 )(tz2 t3 )t3 .
(5.16)
These results seem completely different from Eq. (5.13), and in fact it is difficult to relate
them to the diagram of Fig. 6, since the Green function is not the one expected. However,
this happens just because Eq. (5.15) is not expressed in terms of the most convenient
variables; in fact, by changing tz1 t2 + t3 tz1 the integrand in Eq. (5.15) becomes
exactly that of Eq. (5.13) and the region of integration over tz1 becomes the interval
[ 12 (t2 + t3 ), t2 ]. At this point the two contributions can be summed by simply extending
the integration of tz1 from 0 to t2 and, as far as the punctures are concerned, the expected
field theory result is obtained. Note that the integrand of Eq. (5.13) is symmetric under the
simultaneous exchanges tz1 tz2 t2 t3 , as implied by the world-sheet representation of
these regions of integration (see Fig. 8). This consideration suggests that the symmetry in
the exchange among t1 , t2 and t3 must be obtained only when the external states are put
on the other two boundaries. In fact, when the two punctures are on the boundary (D, D0 )
Fig. 8. The world-sheet representation of the two-loop contribution derived in the text.
403
the final result is exactly that of Eq. (5.13), but with the roles of t2 and t1 exchanged.
Some changes should however be introduced into the procedure, since this boundary is
not in the usual one-loop form (that is with the multipliers at 0 and ). First, the local
coordinates should be expressed in terms of the abelian differential of this loop; second, the
variables zi are not the ones that are directly related to the Schwinger proper times. In fact,
the minimum value of z in this configuration is D, Eq. (5.1), and not k1 , as one should
expect from the symmetry with the case just considered. In this case the string variable
directly related to the Schwinger proper time is

z
,
(5.17)
x=
z1
which ranges in the interval [ 1k , k1 ], as expected by symmetry. With these changes,

1
the analysis of the integration on the second boundary is completely similar to the one
discussed above.
The contributions coming from the configurations with the punctures on the external
boundary can be calculated along the same line. For the local coordinate one can choose
the most symmetric combination of the two Abelian differentials, since this boundary
surrounds both loops
Vi0 (0)1 = 1 + 2 .
(5.18)
The complete result will now be the combination of four regions, corresponding to z1
B, B 0 and z2 C, C 0 . The four regions combine into a single integral, as for the two
contributions (5.13) and (5.15), and give the same result as Eq. (5.13) with t3 and t1
exchanged.
Taking into account all boundaries, one gets a completely symmetric expression, that
should be further multiplied by a factor of two, since in all the computations presented
the role of z1 and z2 can be exchanged, as was discussed after Eq. (2.2). At this point one
can perform the integration over t1 , t2 and t3 independently, introducing, a factor of 1/3!
that cancels the factor of two coming from the z1 z2 symmetry, and the triple counting
arising from the three different first quantized diagrams. Thus the complete contribution
for the field theory configuration of Fig. 6 is simply
A(2)
a (p1 , p2 ) =
N 2 g34 ab
(4)d 4
Z
dt1
0
d/2
Z
dt2
Zt2
dt3
Zt3
dtz1
dtz2 em
2 (t
1 +t2 +t3 )

exp p1 p2 Ga (t1 , t2 , t3 , tz1 , tz2 ) ,
(5.19)
the expected result.

By using the identification among field theory propagators and regions of integration,
it is possible to derive also, in a similar way, the other irreducible two-point 3 diagram
in Fig. 7. This diagram has different symmetries, which implies, in field theory, that its
normalization differs from the one of Eq. (5.19) by a factor of two. Without entering the
details of the calculations, it is easy to see the string origin of this difference. In fact,
as should be clear from Fig. 9, in this case it is possible to consider two world-sheet
404
Fig. 9. Two different world-sheet configurations that contribute to the same first quantized diagram
configurations for each first quantized diagram, besides the usual exchange between the
external legs z1 and z2 ; this fact is eventually responsible for the different normalization.
The result is
N 2 g34 ab
A(2)
c (p1 , p2 ) =
(4)d 2
Z
dt1
Z
dt2
Zt2
dt3
Ztz1
dtz1
dtz2 em
2 (t
1 +t2 +t3 )

d/2 exp p1 p2 Gc (t1 , t2 , t3 , tz1 , tz2 ) ,
(5.20)
where

Gc (t1 , t2 , t3 , tz1 , tz2 ) = 1 (tz1 tz2 ) (t1 + t3 )(tz1 tz2 ) .
(5.21)
This detailed analysis shows that, in the scalar case, the final form of the whole calculation
can be simply determined by picking one of the various corners of integrations that
are relevant for each diagram. All the other contributions only transform the regions of
integration into the expected ones. This step can also be done by hand, by counting the
number of configurations contributing to the diagram, and dividing it by the factor of 1/3!
needed to perform the integration independently. However, if the symmetrization is carried
out explicitly, it gives a non-trivial check on the correctness of the result, that can be useful
when computing more complicated amplitudes.
We turn very briefly to the case of quartic interactions, by considering the diagram in
Fig. 10. The simplest way to get the correct result for this diagram is perhaps to use the
technique that was briefly mentioned in Section 3, when the 4 matching was performed
by inserting functions for the proper times that must vanish in order to obtain quartic
vertices. Choosing for example the boundary [A, A0 ] in Fig. 6, one sees that the punctures
must be integrated between 1/ k1 and k1 . After splitting the integration region into
field theory propagators, as was done to obtain the 3 diagrams, we can simply insert the
Fig. 10. Two-loop two-point diagram in the 4 theory.
405
appropriate functions, say ((tz1 t2 )/ 0 ), with strength 1/2 since they are all located at
the boundaries of the integration regions. Four ways of inserting two such functions yield
an overall factor of unity. Including the other boundaries in the same way to complete the
integration region yields
A(2)
2
g4
=
N 2 ab
(4)d
Z
dt1
Z
dt2
dt3 d/2 em
2 (t
2
1 +t2 +t3 )+p t1 t2 t3 /
(5.22)
which is the leading color structure of the field theory diagram we wanted to reproduce.
5.3. Four-point amplitudes
As a last non-trivial check of our technique for 3 amplitudes, we have computed
the two-loop four-point diagram depicted in Fig. 11. To extract this diagram from the
general expression of the string master formula Eq. (2.1), we can proceed in two steps:
first, since we are interested in an irreducible configuration, we have to consider the limit
1 0 and introduce the change of variables Eq. (5.10), which generates the structure
of the irreducible two-loop vacuum bubble; then we have to consider all possible ways of
inserting the four punctures. Two conditions must be respected: the cyclic order is fixed,
say to (1, 2, 3, 4), and the pairs of external legs (4, 1) and (2, 3) must attach to different
propagators. For example, considering the boundary corresponding to the interval [A0 , A]
in Fig. 6, we can place 1 and 2 inside the interval [1, 1 ] and 2 and 3 outside, or vice
versa. Similar reasonings apply to the other two boundaries. Thus, introducing proper times
for all the zi variables and summing up all the contributions, we obtain the expression
(2)
A4 (p1 , . . . , p4 )
g6
N2
Tr(a1 a4 ) 3 d
=
2
(4)
Zt3
Z
dtz1
dtz4
0
Z
dtz2
Z
dt1
0
tz2
Zt1
tz1
Z
dt2
dt3
0
dtz3 em
2 (t
1 +t2 +t3 )
d/2

exp p1 p2 Ga (t2 , t1 , t3 , tz2 , tz1 ) + p1 p3 Ga (t2 , t1 , t3 , tz3 , tz1 )
+ p2 p4 Ga (t2 , t1 , t3 , tz2 , tz4 ) + p3 p4 Ga (t2 , t1 , t3 , tz3 , tz4 )

+ p2 p3 Gc (t2 , t1 , t3 , tz2 , tz3 ) + p1 p4 Gc (t2 , t3 , t1 , tz1 , tz4 ) ,
Fig. 11. Two-loop four-point diagram in the 3 theory.
(5.23)
406
where the functions Ga and Gc have been defined in Eqs. (5.14) and (5.21), respectively.
This result is exactly the leading color term of the field theory diagram of Fig. 11, as
expected.
6. Concluding remarks
In this paper we analyze in detail the method to define and compute field theory limits
of string amplitudes and study in particular the case of scalar states. Conceptually, the
target field theory is identified by isolating in the string amplitude the quantities that have
to be kept fixed when 0 goes to zero. This step is essentially determined by looking at
the simplest diagram, as was done in Section 3; in this way one can establish a mapping
between gS and the coupling constant of the field theory one wants to reproduce. Having
done this, it is possible to introduce immediately the dimensional Schwinger parameters ti
which measure the length of the various propagators in units of 0 . This operation absorbs
the residual dependence on the string tension contained in the overall normalization, so that
it becomes possible to take the limit 0 0, keeping the field coupling constant and proper
times ti fixed. The mapping between string logarithms and Schwinger parameters is usually
rather intuitive. As was stressed through the various calculations, it is quite easy to follow,
from a geometrical point of view, how the string world-sheet degenerates into a graph and
how to relate each corner of the integration region to a specific Feynman diagram.
In all our calculations, we have always found a precise matching between the string
results and the field theory Feynman diagrams. We think that this work shows that string
techniques are by now mature, and can be confidently used to simplify the computation of
many quantities in field theory. The results obtained in this way can be firmly trusted; in
fact, the derivation of the combinatorial factors is easier and the tedious algebra of color
decomposition is avoided. Moreover, during the calculation itself, one has the possibility to
perform novel consistency checks. For instance, in two-loop calculations one has to obtain
the same result from very different regions of the string moduli space. We have shown that
this is necessary in order to reconstruct the region of integration over the Schwinger proper
times that is expected from field theory.
The generalization of this approach to the physically interesting case of YangMills
theory is, of course, computationally more demanding. One has to deal with derivatives
of the Green function, and all stringy quantities have to be expanded up to first order
in the multipliers, since the spin-1 particle is not the ground state of the bosonic theory.
Furthermore, besides this technical problem, there is also a conceptual difference. As
we discussed, the three two-loop moduli k1 , k2 and are on the same footing from the
field theory point of view; in fact, they are all associated to 3 propagators. Their role
in the Schottky parametrization is, however, quite different, since only the ks are really
multipliers, while is a fixed point. This asymmetry is not evident in the study of 3
diagrams and one obtains directly from the string expression the correct results, with the
expected field theory symmetry. For instance, Eq. (5.20) is invariant under the exchange of
t1 and t3 . However, it is already clear from the study of YangMills vacuum bubbles [22],
407
that the different origin of the various parameters in the Schottky description plays a nontrivial role in this more complicated case. We think that the study of the world-sheet
geometry will provide other useful information for the derivation of pure glue amplitudes
from the string master formula, and we hope that our analysis is a further step in this
direction.
Acknowledgements
We would like to thank R. Marotta and F. Pezzella for presenting to us their results about
4 theory before publication, and for many instructive discussions. We wish also to thank
P. Di Vecchia, A. Lerda and S. Sciuto for many useful discussions and suggestions. This
work has been partially supported by the EEC under TMR contract ERBFMRX-CT960045 and by the Fond National Suisse.
Appendix
We collect in this appendix the Feynman rules for the scalar theories we are considering,
our conventions and several useful formulas concerning the computation of color factors.
The Feynman rule for the cubic scalar vertex described by the Lagrangian in Eq. (3.6) is
simply
V = i g3 d ,
(A.1)
where d is the completely symmetric U (N) color tensor, described below. Similarly,
for the quartic interaction in Eq. (3.15), the rule is

(A.2)
V = i g4 d d + d d + d d .
Notice that in field theory we use the standard metric (+, , , ), whereas string theory
is naturally formulated in the metric with opposite sign.
To derive the U (N) color algebra, it is useful to start from SU (N) matrices, and then
complement them with the diagonal U (1) generator. In the following we will denote U (N)
indices with greek letters, {, , . . .}, and SU (N) indices with latin ones, {a, b, . . .}, so
that, say, = {0, a} if we assign the value 0 to the U (1) index. Throughout the paper, most
of the calculations have been performed with external particles restricted to the SU (N)
subgroup, unless explicitly stated. We normalize our generators as
1
Tr( ) = .
2
With this normalization, the SU (N) generators satisfy

1 1
c
ab 1 + (dabc + i fabc ) ,
a b =
2 N

1 i k
1 i k
i
a k
,
(a ) j ( ) l =
2 l j N j l
(A.3)
(A.4)
408
where fabc are the SU (N) structure constants, while dabc is the completely symmetric
SU(N) color tensor, satisfying
N2 4
ab ,
N
N 2 12
f ge
dabc .
dbg dc =
2N
dacd dbcd =
daef
(A.5)
To promote the above equations to U (N) we must add a correctly normalized U (1)
generator, which can be taken proportional to the identity matrix,
1
0 = 1.
2N
(A.6)
The anticommutation relations for the U (N) generators can then be summarized by
{ , } = d ,
provided one defines
r
2
ab ,
dab0 =
N
da00 = 0,
r
2
.
d000 =
N
(A.7)
(A.8)
= 0, extending to U (N) the

Note that this implies d0 = 2/N , as well as da
corresponding SU (N) property. Using Eq. (A.9), as well as the fact that fab0 = 0, one
can generalize Eqs. (A.4) and (A.5) to
1
= (d + if ) ,
2

1
( )ij ( )kl = li jk ,
2
(A.9)
and
da db = N ab ,
N
(A.10)
da db dc = dabc .
2
It is worth noticing however that Eq. (A.10) does not smoothly generalize to the case in
which one or more of the external indices take their values in the U (1) subgroup. For
example one finds
d0 d0 = 2N,
d d = N(N 2 + 1),
d0 db
d0 d0
dc
= N d0bc ,
d0 = 2N d000.
(A.11)
409
Finally, a useful formula to connect between standard Feynman rules and color ordered
ones is
d = 2 Tr( { , }).
(A.12)
The analogous formula for the structure constants fabc serves the same purpose in QCD.
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]
J. Scherk, Nucl. Phys. B 31 (1971) 222.

A. Neveu, J. Scherk, Nucl. Phys. B 36 (1972) 55.
T. Yoneya, Nuovo Cimento Lett. 8 (1973) 951.
J. Scherk, J.H. Schwarz, Nucl. Phys. B 81 (1974) 118.
R.R. Metsaev, A.A. Tseytlin, Nucl. Phys. B 298 (1988) 109.
V.S. Kaplunovsky, Nucl. Phys. B 307 (1988) 145, hep-th/9205068.
V.S. Kaplunovsky, Nucl. Phys. B 382 (1992) 436, hep-th/9205070.
Z. Bern, D.A. Kosower, Phys. Rev. Lett. 66 (1991) 1669.
Z. Bern, D.A. Kosower, Nucl. Phys. B 379 (1992) 451.
Z. Bern, Phys. Lett. B 296 (1992) 85.
Z. Bern, L. Dixon, D.A. Kosower, Phys. Rev. Lett. 70 (1993) 2677, hep-ph/9302280.
Z. Bern, L. Dixon, D.A. Kosower, Ann. Rev. Nucl. Part. Sci. 46 (1996) 109, hep-ph/9602280.
Z. Bern, D.A. Kosower, Phys. Rev. D 38 (1988) 1888.
P. Di Vecchia, A. Lerda, L. Magnea, R. Marotta, Phys. Lett. B 351 (1995) 445, hep-th/9502156.
P. Di Vecchia, L. Magnea, A. Lerda, R. Russo, R. Marotta, Nucl. Phys. B 469 (1996) 235,
hep-th/9601143.
Z. Bern, D.C. Dunbar, T. Shimada, Phys. Lett. B 312 (1993) 277, hep-th/9307001.
Z. Bern, L. Dixon, D.C. Dunbar, M. Perelstein, J.S. Rozowsky, Nucl. Phys. B 530 (1998) 401,
hep-th/9802162.
K. Roland, Phys. Lett. B 289 (1992) 148.
P. Di Vecchia, L. Magnea, A. Lerda, R. Marotta, R. Russo, Phys. Lett. B 388 (1996) 65, hepth/9607141.
K. Roland, H. Sato, Nucl. Phys. B 480 (1996) 99, hep-th/9604152.
K. Roland, H. Sato, Nucl. Phys. B 515 (1998) 488, hep-th/9709019.
L. Magnea, R. Russo, in: Proc. DIS 97, Chicago, USA, 1997, J. Repond, D. Krakauer (Eds.),
AIP Conf. Proc., Vol. 407, 1997, p. 913, hep-ph/9706396; in: Proc. Beyond the Standard
Model V, Balholm, Norway, 1997, G. Eigen, P. Osland, B. Stugu (Eds.), AIP Conf. Proc.,
Vol. 415, 1997, p. 347, hep-ph/9708471.
Z. Bern, J.S. Rozowsky, B. Yan, Phys. Lett. B 401 (1997) 273, hep-ph/9702424; in: Proc.
DIS 97, Chicago, USA, 1997, J. Repond, D. Krakauer (Eds.), AIP Conf. Proc., Vol. 407, 1997,
p. 908, hep-ph/9706392.
A. Pasquinucci, K. Roland, Nucl. Phys. B 485 (1997) 241, hep-th/9608022.
L. Cappiello, R. Marotta, R. Pettorino, F. Pezzella, Mod. Phys. Lett. A 13 (1998) 2433, hepth/9804032; Mod. Phys. Lett. A 13 (1998) 2845, hep-th/9808164.
A. Liccardo, R. Marotta, F. Pezzella, Mod. Phys. Lett. A 14 (1999) 799, hep-th/9903027.
Z. Bern, D.C. Dunbar, Nucl. Phys. B 379 (1992) 562.
M.J. Strassler, Nucl. Phys. B 385 (1992) 145, hep-ph/9205205.
M.G. Schmidt, C. Schubert, Phys. Lett. B 331 (1994) 69, hep-th/9403158.
C. Schubert, Acta Phys. Pol. B 27 (1996) 3965, hep-th/9610108.
H. Sato, M.G. Schmidt, Nucl. Phys. B 524 (1998) 742, hep-th/9802127.
H. Sato, M.G. Schmidt, Nucl. Phys. B 560 (1999) 551, hep-th/9812229.
410
[33] P. Di Vecchia, M. Frau, A. Lerda, S. Sciuto, Nucl. Phys B 298 (1988) 526.
[34] P. Di Vecchia, Multiloop amplitudes in string theory, Theor. Phys. (1992) 16, and references
therein.
[35] R. Marotta, F. Pezzella, hep-th/9912158.
[36] P. Di Vecchia, F. Pezzella, M. Frau, K. Hornfeck, A. Lerda, S. Sciuto, Nucl. Phys. B 322 (1989)
317.
[37] M.B. Green, J.H. Schwarz, E. Witten, Superstring Theory, Cambridge University Press, 1987.
[38] V. Alessandrini, D. Amati, M. Le Bellac, D. Olive, Phys. Rep. 1 (1971) 269.
[39] K. Roland, Multiloop amplitudes in pure gauge theories: the superstring approach SISSA/ISAS
131-93-EP (1993).

On stable sector in supermembrane matrix model

I.Ya. Arefeva a,1 , A.S. Koshelev b,2 , P.B. Medvedev c,3
a Steklov Mathematical Institute, Gubkin st. 8, Moscow, 117966, Russia
b Institute of Theoretical and Experimental Physics, B. Cheremushkinskaya st. 25, Moscow, 117218, Russia
c Physical Department, Moscow State University, Moscow, 119899, Russia
Received 29 December 1999; revised 7 March 2000; accepted 27 March 2000
Abstract
We study the spectrum of SU(2) SO(2) matrix supersymmetric quantum mechanics. We use
angular coordinates that allow us to find an explicit solution of the Gauss law constrains and single
out the quantum number n (the Lorentz angular momentum). Energy levels are four-fold degenerate
with respect to n and are labeled by nq , the largest n in a quartet. The Schrdinger equation is reduced
to two different systems of two-dimensional partial differential equations. The choice of a system is
governed by nq . We present the asymptotic solutions for the systems deriving thereby the asymptotic
formula for the spectrum. Odd nq are forbidden, for even nq the spectrum has a continuous part as
well as a discrete one, meanwhile for half-integer nq the spectrum is purely discrete. Taking halfinteger nq one can cure the model from instability caused by the presence of continuous spectrum.
2000 Elsevier Science B.V. All rights reserved.
1. Introduction
Supersymmetric quantum mechanics describing an SO(d) invariant interaction of d
SU(n)-matrices has attracted a lot of attention in the last two decades.
The original interest to this model was caused by its description of 1+0 reduction of (1+
d)-dimensional supersymmetric SU(n) gauge theory. A reduction of non-supersymmetric
d = 3 YangMills theory to a mechanical model was considered in the pioneering papers
[13]. It was noticed that already for the simplest case of SU(2) gauge group one gets
a rather complicated mechanical system with 9 degrees of freedom [4]. This mechanical
system was investigated only within the special ansatz. Within one of them the model
is reduced to a model with two degrees of freedom which is still rather non-trivial and
1 arefeva@genesis.mi.ras.ru
2 medvedev@heron.itep.ru
3 kas@depni.npi.msu.su
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 2 0 5 - 4
412
I.Ya. Arefeva et al. / Nuclear Physics B 579 (2000) 411436
exhibits a chaotic behaviour [13,5,6] in the classical case. Later on, this model 4 has been
investigated in the quantum case. It was also proved that this system has a discrete spectrum
[7,8]. This spectrum possesses a very interesting property it is in some sense a direct
product of two harmonic oscillators [9]. The possible applications for realistic models see
[10,11].
In the end of 80th the interest to the matrix quantum mechanics was inspired by an
observation that it describes a regularized membrane theory in d + 2 spacetime dimensions
[1215]. The membrane theory was supposed to be reached in the limit of large n. Since the
membranes were considered within the 11-dimensional supergravity the supersymmetric
version of SU(n) SO(9) quantum mechanics has to be examined. In contrast to the
bosonic matrix models where the spectrum is purely discrete, in supermembrane matrix
models a continuous spectrum, filling the positive half of the real line was detected [15].
This fact was considered as a manifestation of the instability of the supermembrane against
deformations into stringlike configurations.
Let us also note that more early, in the beginning of the 80s supersymmetric quantum
mechanics (SQM) was proposed as a model for better understanding of supersymmetry
breaking [16]. In this context SQM was considered in [1719]. In particular in [17] the
N = 4 and the N = 16 quantum mechanics (now days known as SU(2) SO(3) and
SU(n) SO(9), respectively) were proposed.
A renovation of interest to a supersymmetric matrix quantum mechanics in the last
three years was motivated by its relation to a description of the dynamics of D0 branes
in superstring theory [20,21]. Moreover, this model in the large n limit pretends to the
role of M-theory [22]. This conjecture has stipulated the recent study of SU(n) SO(d)
supersymmetric quantum mechanics [2325]. Within M-theory there is a very important
question of the existence of normalized eigenfunctions with zero energy, since the zero
modes represent the graviton multiplet of eleven dimensional supergravity. This problem
has attracted attention since the first paper where the model was introduced. One expected
that SU(n) SO(d) supersymmetric quantum mechanics has one normalized zero-mode
for d = 9, and has none for d = 2, 3, 5 (only in these dimensions a supersymmetric model
can be formulated) [2337].
The case of large n is rather involved. The simplest case is the SU(2) one. In this context
the SU(2) quantum mechanics was investigated in [2325]. By using gauge invariant
variables the asymptotic form of the ground state in the SU(2) quantum mechanics
was found in [32] and by generalization of this method for the case of arbitrary n
was found in [34]. The case of arbitrary n was also considered in [31] and n = 3
was considered in [37]. To investigate the zero-mode problem the BornOppenheimer
approximation was applied in [2325,32] (see [32] for a discussion of the status of the
BornOppenheimer approximation). This method allows to find asymptotic behaviour near
infinity. Therefore, if there are no any singularities at a finite region, one can expect an
existence/nonexistence of normalized zero-mode from asymptotic behaviour. An effective
4 Nowadays, the supersymmetric version of this model is known as a toy model and it is often considered to
check and clarify new methods.
413
tool to study the zero-mode problem is an investigation of a system of first order differential
equations caused by the supersymmetry of a desired zero mode. The BornOppenheimer
approximation is also applicable to the study of the first order differential equations.
Recently the authors of [36] have performed an investigation of zero-mode problem using
the first order differential equation. The asymptotic behaviour found in [36] supports a
common belief about the existence of normalized zero-mode in d = 9 and nonexistence in
other dimensions.
Let us make few comments about chaotic behaviour of matrix models. About classical
dynamics of two-dimensional model [13] see [38,39]. Classical dynamics in bosonic
membrane matrix model was investigated and a chaotic behaviour was demonstrated. Later
on, in SU(2) SO(2) matrix model a classical chaos-order transition was found [40,41].
For the Lorentz momentum N small enough (even for small coupling constant) the system
exhibits a chaotic behaviour, for N large enough the system is regular. Up to now the
question of similar transition in quantum case remains open. 0 -corrections to the Yang
Mills approximation of D-particle dynamics were studied in [42] where a stabilization of
the classical trajectories was shown.
The main task of this paper is to find the whole spectrum of SU(2) SO(2)
supersymmetric matrix quantum mechanics. In the previous investigations of this model
the following particular results about the spectrum were obtained. Continuous spectrum
was observed [15] and the nonexistence of normalized zero mode has been proved [27].
The character of spectrum plays an important role in stability/instability of the system.
Let us remind that according to the commonly accepted opinion this model is unstable.
The potential instability of (super) matrix quantum mechanics is evident from classical
consideration. Namely, the potential of matrix models has valleys through which a part
of coordinates can escape to infinity without increasing the energy. For the bosonic
case the classical instability is cured by quantum fluctuations due to which the flat
directions become closed by confined potentials, so that finite energy wave functions
fall off rapidly and spectrum is purely discrete [7,9], that provides stability. The lost of
stability in supersymmetric case is caused by additional contributions to the potential
coming from the fermionic degrees of freedom which cancel the bosonic ones. As
a result of this cancellation wave functions are no more confined and the spectrum
becomes continuous [15]. This cancellation takes place on special states, that means a
coexistence of the continuous spectrum and the discrete one. To specify states for which a
cancellation/non cancellation takes place it is convenient to arrange the states into quartets
enumerated by a number nq . Upon quantization nq can be integer or half-integer. Our
analysis of SU(2) SO(2) supersymmetric quantum mechanics shows that there is a
cancellation in the sector with even nq and there is no cancellation for half-integer nq
(states with odd nq are forbidden). As a result, there is only a discrete spectrum in the
half-integer nq sector and the model is stable. The lowest energy in this sector is positive,
that means the supersymmetry is broken in this sector.
Our main tool in the detailed study of the spectral problem of SU(2) SO(2)
supersymmetric quantum mechanics is the proper coordinates, four angles 1 ,2 , , and
two radii f and g. These coordinates have been already used in [40]. In these coordinates
414
we will find an explicit solution to the Gauss law constrains and single out quantum number
n. Due to this parametrization the spectral problem for a supersymmetric version of the
model with 6 degrees of freedom will be reduced to a pair of systems of partial differential
equations in f and g. The choice of a system is dictated by the value of nq . For halfinteger nq we have just one equation on one function of f and g. For even nq we have
three equations on three functions of f and g.
We present asymptotic solutions of these two systems deriving thereby the asymptotic
formula for the spectrum. For half-integer nq we deal with differential operator corresponding to standard potential problem in quantum mechanics and character of the spectrum can
be understood from the form of potential. Since we have a confining potential the spectrum is discrete. We present an asymptotic solution of the Schrdinger equation and corresponding formula for the spectrum. For even nq we have a matrix second order differential
operator. Just for this system a vanishing of the bosonic confining potential takes place on
special states and the matrix differential operator has the continuous spectrum. To get it
we find the asymptotic solution of the system of three equations. Besides the continuous
spectrum this operator possesses the discrete spectrum. States with integer (half-integer) n
n = 0).
belong to the stable sector if they satisfy the constraint Qn = 0 (Q
The paper is organized as follows. In Section 2 we explore the algebraic structure of
SU(2) SO(2) supersymmetric quantum mechanics in context of energy level degeneracy.
In Section 3 we specify the angular SU(2) SO(2) parametrization. A special attention
is spared to generalized periodicity. In Section 4 we solve the constrains in the angular
parametrization and present the spectral problem as two sets of two-dimensional partial
differential equations. In Section 5 the spectrum of corresponding differential operators is
examined. We present asymptotic solutions of the Schrdinger equation and corresponding
formula for the spectrum.
2. Algebraic structure of SU (2)SO(2) supersymmetric quantum mechanics

2.1. The Hamiltonian and superalgebra
We consider the system described by the Hamiltonian
H = HB + HF ,
2

gs 2
1
1 + 22 +
1 2 ,
HB =
2
2gs

i abc a
i
1 + i2a b c abc 1a i2a b c ,
HF =
2
2
(1)
and constrained by the Gauss law j a | i = 0. ia , jb and a , b are canonically

conjugated pairs with
a b
a b
= ab .
(2)
i , j = i ab ij ,
The Hamiltonian (1) is invariant under SU(2) rotations and redundant Lorentz rotation
generated by
j a = la + sa ,
415
l a = abc ib ic ,
s a = iabc b c ,
1
NB = 1a 2a 2a 1a ,
NF = a a ,
N = NB + NF ,
2
respectively. The commutation relations for the currents read:
a
a b
j , N = 0.
j , j = iabc j c ,
There are two supercharges:

gs
i
a abc 1b 2c ,
Q = a 1a i2a +
2gs
2

gs
i
a abc 1b 2c
Q = a 1a + i2a
2gs
2
and

Q, Q = H.
The supercharges commute with the Hamiltonian up to the generator of SU(2) rotations
vanishing on the physical states.
The remaining commutators with Q and Q are the following:
Q 2 = 0,
(3)
Q2 = 0,

a
a
j = 0,
Q, j = Q,

1

1
N, Q = Q,
N, H = 0.
(4)
N, Q = Q,
2
2
Note, that the algebra (3)(4) is not affected by a shift of N by a constant. As compare
with [36] our N is shifted on 1/4.
2.2. Energy level degeneration
2.2.1. The algebra
On the physical states (j a | i = 0) there are three operators commuting with H : Q,
Q and N . This provides the degeneration of energy levels. In this section we discuss the
irreducible representations vE of the algebra (3)(4) for a non-zero energy E.
Lemma. Let | i vE with E 6= 0 be an eigenvector of H and N , then Q| i = 0 or
i = 0.
Q|
i 6= 0 then one has two additional eigenvectors of
Proof. Suppose that Q| i 6= 0 and Q|
N with the same energy:

(5)
N Q| i = n 12 Q| i ,

1
i .
i = n+
Q|
(6)
N Q|
2
One can check, that the operator

1
K = H N + Q, Q
4
416
is one more Casimir of the algebra (3)(4). By a simple calculation one finds:

1
1
i,
i = E n 14 | i + QQ|
K| i = E n + 14 | i QQ|
2
2

K Q| i = E n 14 Q| i ,

i .
i = E n + 1 Q|
K Q|
4
(7)
(8)
For vE to be irreducible the value of the Casimir on | i should be equal to its value on
i. From Eqs. (7)(8) we see that this requires 5
Q| i and Q|
Q| i = 0
i = 0.
or Q|
and Q act as raising and lowering operators (Eqs. (5), (6)) one can introduce a
Since Q
label: vE,n where n is the value of N on the vector | i ker Q.

Q
n1/2
0.
So, we find that vE,n is a vector space spanned by two eigenvectors of N : |1 i and |2 i
such that
1 i = 0 = Q|2 i
Q|
and
N|1 i = n|1 i,
N|2 i = n
1
2
|2 i.
2.2.2. Discrete symmetry

In addition to the algebra (3)(4) the Hamiltonian (1) admits one more symmetry. The
bosonic part of the Hamiltonian is invariant under the discrete transformation
1 2 .
(9)
To extend this symmetry for the supersymmetric case we put:

,
(10)
The invariance of the Hamiltonian and of the commutation relations gives the following
restrictions on , :
2 = 2 = i,
= 1.
One more restriction:

= i
comes from imposing the condition for transformation law of supercharges Q, Q to be
homogeneous.
The system of equations for and has a solution (up to an overall sign):
= ei/4 ,
= ei/4 .
Denote the operator generating this symmetry by P , then we have

5 Remind that if Q| i = 0 then | i = Q| i with | i = 1 Q|
i.
E
[P , H ] = 0,
[P , j ] = 0,
P Q QP = 0,
P Q QP = 0,
3
P 2 = 1.
{P , N} = P ,
2
417
(11)
(12)
Note, that P acts on the spinor monomials without permutations of spinor factors,
because these permutations are incompatible with the invariance of the anticommutation
relations (2).
n
n + 3/2
Q Q
n 1/2
Q Q
p
n + 2
As P vE,n = vE,2n (action of P -operator on the states will be specified below) the
irreducible representations of the enlarged algebra (3)(4) and (11)(12) are direct sums:
VE,nq = vE,n + vE,2n , where nq is the maximal number of n and 2 n. To avoid
doublecounting one has to restrict the range for n as to exclude n < n+2 which gives n >
1. In the exceptional case n = 1: vE,1 = vE,1+2 , i.e., for n = 1 one has a two-dimensional
instead of a four-dimensional irreducible representation. The further specification of the
range will be given below.
2.3. Action of P -operator on the states
A general wave function is a vector of the eight-dimensional fermionic Fock space
based on the fermionic vacuum |0i: a |0i = 0. It is useful to arrange the fermionic states
according to their parity as follows:
| i = 0 + 1 2 3 + 2 3 1 + 3 1 2

+ 0 1 2 3 + 1 1 + 2 2 + 3 3 |0i,
(13)
where 0 , a , 0 and a are some functions of E1 , E2 . The Fock space can also be created
by operators a acting on the dual vacuum
|0) = 1 2 3 |0i.
To define the action of P on | i one has to specify the action of P on the vacuum |0i
(or on |0)). The unique choice that gives non-degenerate P is:
P |0i = |0),
(14)
where is some numerical factor to be fixed by the constraint P 2 = 1.

It is instructive to represent | i as a column of s with the ordering dictated by
Eq. (13). Within this notation one can write:
418
0
1
3
2
2
P | i = P
0
1

2

3
(1,2)
0 ( 3 )

= P
1
( 2 )
2
( 2 )
3 ( 2 )
(2,1)
0
1
3
2 3
=
0
1

2

3
(15)
(1,2)
where labels (1, 2), (2, 1) indicate the order of bosonic arguments for s.
The constraint P 2 = 1 results in 2 3 = 1 and gives, according to our choice of 6 :
= ei/8 .
In a matrix form the action of the operator P on the fermionic degrees of freedom can
be represented as follows
1
0
0
0
2
0
0
0
0
0
0
0
0
0
.
P =
3 0 0 0
0
0 0
0
0 0
0
0
0 0
2.4. Gauss law and N| i = n| i in components
The Gauss law
j a | i = 0
fixes the SU(2) transformation properties of the component wave functions a , a . These
are given by:
l a | i = s a | i.
The spin operator sE acts on fermionic Fock states and its action can be easily calculated to
give

a b
a
l , = i abc c ,
l , 0 = 0,
a

a b
l , 0 = 0,
l , = i abc c .
Hence, we conclude that: (1) 0 and 0 are SU(2) singlets, (2) E = ( 1 , 2 , 3 ) and
E = ( 1 , 2 , 3 ) are SU(2) triplets. One can check that
lE2 0 = lE2 0 = 0,
6 Starting from the dual vacuum |0) as P |0) = |0i one finds = ei/8 . It is a matter of simple algebra to
check that for these and P is correctly defined, i.e., it does not depend on the choice of the cyclic vector, |0i
or |0), in the fermionic Fock space.
i
E
lE2 , E = 2,
419
i
E = 2
E
lE2 ,
as it must be for the scalar and vector representation. The explicit solution of (16) will be
given below after the parametrization of the configuration space (E1 , E2 ) will be specified.
In the following we shall exploit the eigenstates of N . It is instructive to separate the
fermion number operator NF . Let: N| i = n| i then NB | i = (n NF )| i gives
NB 0 = n0 ,
NB 0 = n

3
2
E
NB E = (n 1),
E
E
NB = n 12 .
0 ,
(16)
2.5. Schrdinger equation in components

The fermionic Fock space decomposition (13) for | i provides a matrix representation
for Schrdinger equation H | i = E| i. Taking | i as a column (see (15)) one deduces
H in a block-diagonal form:
HB i 1 i 2 i 3
1
HB
0
0
0
0
i 2 0 HB
0 HB
i 3 0
H =
,
HB i 1 i 2 i 3
0
i 1 HB
0
0
0
HB
0
i 2
0
0
HB
i 3
where
a = 1a + i2a ,
a = 1a i2a .
(17)
E are not
On physical states (jE| i = 0) the component equations for SU(2) triplet E ()
independent. For instance, the last equation for the upper block is (equation number four):
i 3 0 + HB 3 = E 3 .
(18)
l1
to (18):

il 0 + l 1 HB 3 = El 1 3 ,
Let us apply
1
then by using the commutation relations (16) and [l a , b ] = iabc c one gets the third
Schrdinger equation
i 2 0 + HB 2 = E 2
and so on.
Therefore, we are left with two sets of equations:

E = E0 ,
HB 0 + i(E)
HB 3 i 3 0 = E 3 ,
and
(
E = E ,
HB 0 i(E )
0
3
3
HB + i 0 = E 3 ,
with 1,2 ( 1,2 ) expressed in terms of 3 ( 3 ).
(19)
(20)
420
3. SU (2)SO(2) parametrization
3.1. Parametrization
We parametrize the configuration space {ia } by using the new coordinates {f, g, , 1 ,
2 , } as follows:

1
1 = U + 1 f cos 2 g sin U,
2

1
2 = U + 1 f sin + 2 g cos U .
2
Here a
=
1

0 1
,
1 0

=
2

0 i
,
i 0
(21)

=
3
1
0

0
,
1
(22)
are Pauli matrices and the SU(2) matrix U (1 , 2 , ) is given by

i
i
i
3
1
3
1 exp
exp
2 .
U (1 , , 2 ) = exp
2
2
2
The range for the new coordinates is 0 6 1 , 2 < 2 , < < 0, 0 6 < /2, f, g > 0.
The SU(2) angular variables 1 , , 2 are similar to the Euler angles. The SU(2)
adjoins i can be viewed as the R 3 vectors Ei . Denote eEx , eEy , eEz the unit vectors of the
fixed coordinate system. eEn = eEz (E1 E2 ) is a vector along the knot-line (the line of
intersection of the (E1 , E2 ) and (x, y) planes). The first rotation on 2 , around the z-axis
matches eEn with xE . After this rotation the vector eE3 = E1 E2 falls into the (y, z) plane,
with (Ee3 )y < 0. The second rotation around the x-axis in clockwise direction on the angle
Fig. 1. Angular parametrization.
421
, matches eE3 with zE that gives < < 0. These two rotations place the pair E1 , E2 into
the (x, y) plane and 6 (E1 , E2 ) < ((Ee3 )z is positive).
To fix the third angle 1 , let us take a pair (E1 , E2 ) in (x, y) plane with |E1 | = r1 , |E2 | =
r2 and 6 (E1 , E2 ) = and examine its orbit under the action of the rotations around the zaxis. This orbit is the set {(r1 cos , r1 sin ), (r2 cos( + ), r2 sin( + )} with 0 6 < 2 .
We call a pair (E1 , E2 ) as SO(2)-orthogonal if
11 12 + 21 22 = 0,
(23)
here i1 = (Ei )x , i2 = (Ei )y . The SO(2)-orthogonality condition for the points of the orbit
reads:
tan 2 =
r22 sin 2
r12
+ r22 cos 2
Hence, at any orbit there are four SO(2)-orthogonal pairs (E1 , E2 ). One of these pairs has
E1 lying in the fourth quadrant, i.e., 11 > 0, 12 6 0. This pair we call the special one.
For any pair (E1 , E2 ) we define 1 as the angle of rotations around the zE-axis that matches
(E1 , E2 ), with the special pair. The range for 1 is obviously 0 6 1 < 2 .
To parametrize the special pair (E1 , E2 ) note that the numbers ia form also a pair of
SO(2) (Lorentz) orthogonal vectors: u1 = {11 , 21 } and u2 = {12 , 22 }, so that arg u2 =
arg u1 + /2. The remaining coordinates {f, g, } are defined to be the polar coordinates
for u1 and u2 :
11 = f cos ,
21 = f sin ,
12 = g sin ,
22 = g cos .
As the pair is the special one the ranges for {f, g, } are 0 6 < /2 and f, g > 0.
To summarize, we have proved that by the SU(2) rotation defined by the Euler-like
angles 1 , 2 and any pair (E1 , E2 ) is matched with the special pair which is parametrized
by f, g, . The special case of an angular momentum parametrization was proposed in [10,
11].
Fig. 2. SU(2) and SO(2).
422
3.2. Generalized periodicity

To extend the ranges for the angular variables i up to R 1 axis we employ the usual
2 periodicity conditions for i . As to the angle with 0 6 < /2 the situation is more
subtle.
Let us take a point with SO(2) angle coordinate outside the range: (f, g, + /2, 1) (2
and are irrelevant for the subsequent discussion). By means of (21) this set parametrize
the pair
1,2 = U + (1 )1,2 U (1 ),
U (1 ) = e(i/2)13 ,
where
1 = 1 f sin + 2 g cos ,
2 = 1 f cos 2 g sin .
The pair (E1 , E2 ) is SO(2)-orthogonal but not special. What are the true coordinates for
1,2 ?
In complex notation (17) these read:
i 1
+

i 1
e2
0
0
e2
i(+ 2 )
e
(
f
+
i
g)
=
1
2
i
i
0 e 2 1
0 e 2 1

0
ei(1 2 ) (f + g)
i
=e
0
ei(1 + 2 ) (f g)

0
ei(1 2 ) (f + g)
.
= ei i(1 )
2 (g f )
0
e
Hence, the following points must be identified

f, g, + 2 , 1 , 2 ,
= g, f, , 1 2 , 2 , .
(24)
Below we shall call (24) the generalized periodicity condition in .

3.3. P symmetry in SU(2) SO(2) parametrization
On bosonic degrees of freedom P operator acts as E1 E2 . Here we describe this action
in terms of the new coordinates. Let us denote E10 = E2 , E20 = E1 , then P Ei = Ei0 . We also
shall indicate by primes all the objects referred to the pair (E10 , E20 ).
By definition one has eE30 = Ee3 and eEn0 = Een hence, 20 = + 2 . After 2 rotation eE30
lies in the (x, y) plane with (Ee30 )z = (Ee3 )z and (Ee30 )y = (Ee3 )y . This yields: 0 = .
Let us compare the results of two first rotations for both the pairs (E1 , E2 ) and (E10 , E20 )
that lie in (x, y) plane (we still denote the resulting pairs as (E1 , E2 ), (E10 , E20 ) ). By means
of a simple algebra one finds that for 20 and 0 given above:
1 0 0 !
0 1 0
0 0 1
1
0
0 !
0 cos sin
0 sin cos
cos 2 sin 2 0 !
sin 2 cos 2 0
0
0 1
1
0
0 !
0
0 cos sin 0
0 sin 0 cos 0
423
cos 20 sin 20 0 !
sin 20 cos 20 0 ,
0
0 1
i.e., (E10 , E20 ) and (E1 , E2 ), differs by the rotation on around the y-axis. Thus on the (x, y)
plane one has:
101 = 21 ,
102 = 22 ,
201 = 11 ,
202 = 12 .
One can rewrite these relations in polar coordinates (we use complex notation):
if E1 = r1 ei ,
then E10 = r2 ei(),
E2 = r2 ei ,
E20 = r1 ei() .
Let 1 be the angle that matches (E1 , E2 ) with the special pair (r1 ei(+1 ) , r2 ei(+1) ).
The pair (r2 ei(+1 ) , r1 ei(+1 ) ) is also a special one and

r2 ei(+1) , r1 ei(+1 ) = ei(1 +) r2 ei() , r1 ei()
that yields 10 = 1 = 1 + due to periodicity in 1 . The f 0 , g 0 and 0 are easily
found to be: f 0 = f , g 0 = g and 0 = /2 .
Collecting all the pieces together we find the action of P in new coordinates:
2 + 2 ,
,
1 1 ,
/2 ,
f f,
g g.
(25)
4. Schrdinger equation in SU (2)SO(2) parametrization

4.1. N = n and Gauss law an explicit solution
The SU(2) SO(2) variables provide the explicit solution for wave function | i
dependence on angular variables.
For NB one deduces
NB = i ,
i.e., NB eigenfunctions are the plane waves eik . Applying periodicity condition (24) four
times we find that wave functions components have to be 2 periodic so k must be an
integer. Thus, from Eqs. (16) it follows that n can be integer or half-integer and
0
0
0E
E
,
(26)
for
half-integer
n
:
for integer n : n
.
n
0
0
E
0E
In the following we shall concentrate on the case of integer n, the half-integer case will be
sometimes discussed for the sake of completeness.
424
The components of SU(2) generators l a depend only on Euler angles:

sin 2
+
cos
l 1 = i cos 2
,
sin 1
2

cos 2
cos
,
l 2 = i sin 2
sin 1
2
,
l 3 = i
2
2

1
2
cos
2

2
+
2
cos
.
(lE)2 = 2
sin sin 12 22
1 2
(27)
(28)
These expressions coincide with the usual Euler angles parametrization for SO(3)
generators [43] up to some modification of variables.
For 0 ( 0 ) Eq. (16) shows that 0 does not depend on (1 , 2 , ) and has the form:
for integer n :
for half-integer n :
0 = ein F0 (f, g),

3
e0 (f, g).
0 = ei(n 2 ) F
From (27) one finds 3 ( 3 ) to depend only on and 1 . Thus, taking 3 (f, g, , , 1 )
P
= F (f, g, )(, 1 ) from (28) one has:

2
1 2
cos
+
+
= 2.
(29)
2
sin sin2 12
The solutions of (29) are spherical functions Y1,m with m = 0, +1, 1. Taking for Y1,m the
explicit expressions we get:

for integer n:
3 = ei(n1) F 0 cos + F + sin ei1 + F sin ei1 , (30)

e+ sin ei1 + F
e sin ei1 , (31)
e0 cos + F
for half-integer n: 3 = ei(n1/2) F
where the harmonics F 0 , F + , F are functions of f and g only.
E which are restored by acting of l a on 3
We list the remaining components of E ()
( 3 ). It is useful to slightly modify the basis by passing to the eigenvectors of l 3 (s 3 ):
| i = 0 + ( 1 + i 2 ) 3 + + ( 1 i 2 ) 3 + 3 1 2

+ 0 + ( 1 + i 2 ) + + ( 1 i 2 ) + 3 3 |0i.
In this basis are achieved by using the raising and lowering operators l = l 1 il 2 as
i
i
+ = l+ 3,
= l 3,
2
2
similarly for the tildes. Finally, we have for integer n:

1
+ = ei(n1) ei2 F 0 sin F + (cos + 1)ei1 F (cos 1)ei1 ,
2

1
= ei(n1) ei2 F 0 sin F + (cos 1)ei1 F (cos + 1)ei1 ,
2
for half-integer n:
(32)
(33)
425

i
e+ (cos + 1)ei1 F
e (cos 1)ei1 ,
e0 sin F
+ = ei(n1/2) ei2 F
2

i i(n1/2) i2 e0
e+ (cos 1)ei1 F
e (cos + 1)ei1 .
e
F sin F
= e
2
Further restrictions and symmetry properties of F s come from the generalized -periodicity (24). We shall examine the case of integer n, half-integer case is similar.
Applying (24) twice we obtain restrictions dependent on n:
even n: F 0 = 0,
odd n: F0 = F + = F = 0.
Applying (24) once we get that only the following functions F are consistence with the
general periodicity:
n = 4k:
F 0 = 0,
F0 (f, g) = F0 (g, f ),
F + (f, g) = F + (g, f ),
F (f, g) = F (g, f ),
n = 4k + 2:
(34)
F 0 = 0,
F0 (f, g) = F0 (g, f ),
F + (f, g) = F + (g, f ),
F (f, g) = F (g, f ),
n = 4k + 1:
F0 = F + = F = 0,
F 0 (f, g) = F 0 (g, f ),
n = 4k + 3:
(35)
(36)
F0 = F + = F = 0,
F 0 (f, g) = F 0 (g, f ).
(37)
4.2. Hamiltonian
We deduce the explicit expression for HB in new variables as follows. At first, we
calculate the metric tensor g : ab ij dia djb = g dx dx with xE = (f, g, , 1 , 2 , ).
At second, we obtain the Hamiltonian as the LaplaceBeltrami operator for the metric g
plus the potential term
1 2 2
f g .
8gs
This straightforward but rather tedious calculation was performed with partial use of
Maple. The result is the following:
426

2
1

A
HB = gs 1 + D 2 (lE)2 +
+
4a
c
1
a
(b)2 2

B 2
b sin 21
b cos 21 2
cos
2
a sin
1 2
a2
2 a 2 b2 12

b sin 21
1 2
2
(1
+
cos
+
)
2
cos
a ,
+
2
2
1
2
8gs
2a sin
(38)
here 1 is Laplacian in f , g and:

J
1 J
+
,
D=
J f f
g g
b = f 2 g2 ,
a = fg,
c = f 2 + g2 ,
J = ab,
A = f sin 1 + g cos 1 ,
2
B = f 6 sin2 1 3f 4 g 2 sin2 1 3f 2 g 4 cos2 1 + g 6 cos2 1 .

Note that there are terms with the first order derivatives /f and /g collected in
D. Although the coefficients for these terms are functions of f , g only, these terms are
absent in the usual quantization of the two-dimensional toy model [13], as in the pure
bosonic case as in the fermionic one 7 . These first order derivatives in couple with the zoo
of angular dependent terms in (38) radically modify the potential.
4.3. Schrdinger equation for harmonics
Now we substitute the harmonic expansion for the wave function n into the system
(19) with the explicit Hamiltonian HB given in (38). The result is:
For n odd (only F 0 6= 0) we get one PDE in f and g

c
a2 0
2 c
F = EF 0 ;
(39)
gs 1 + D (n 1) 2 2 F 0 +
b
a
8gs
For n even we get a system of three PDE in f and g

1 2
f +g
f g
c
a F0 F + + F = EF0 ,
gs 1 + D n2 2 F0 +
8gs
b
2
2

4 + g 4 )c
4a
(f
b
c
2
+
F 2F
gs 1 + D (n 1) 2 (n 1) 2
b
b
2a 2 b2
2a
f +g
a2 +
F F0 = EF + ,
8gs
2 2

4a (f 4 + g 4 )c
b +
c
F
F
gs 1 + D (n 1)2 2 + (n 1) 2
b
b
2a 2 b2
2a 2
+
f g
a2
F + F0 = EF .
8gs
2 2
(40)
(41)
(42)
7 The appearance of the first order derivatives upon the correct quantization was at first observed in the
pioneering paper [4].
427
5. The spectrum
Remind, that for any E 6= 0 eigenfunctions are combined into the quartets VE,nq with
quantum numbers nq , nq 1/2, nq + 2, nq + 3/2. To examine the spectrum it is
sufficient to search for any one of these four functions constrained by the additional
Next we search for differential equations
constraints coming from supercharges Q and Q.
for these constraints.
5.1. Supercharges
In the basis (13) the supercharges Q and Q take the form of anti-block-diagonal
matrices:

0 Q
0 Q+
,
Q =
,
(43)
Q=
Q
0
Q+
0
where asterisk stands for Hermitian conjugation.
The 4 4 blocks are calculated to give:
0
gs 1
gs 2
gs 3
g 1
i 3
i 2
0
s
1
gs
gs
,
Q =
i 1
2 gs 2 ig 3
0
gs
s
gs 3 igs 2 igs 1
0
i 1
gs
i 1
1
gs
Q+ =
i
2
2
gs
i 3
gs
i 2
gs
gs 3
gs 3
gs 2
gs 1
i 3
gs
gs 2
1
gs
here we used the complex notation (17), E = E 1 iE 2 and

E = E1 E2 .
For integer n (see Eq. (26)):

0
E

0
0E
= 0 are reduced to
and equations Q = 0, Q
!

E
gs (E )
1
0
Q E =
gs
E 0 + igs
E E
2
!
gs div E
1
= 0.
=
gs grad 0 + igs
E E
2
!

i
E
E )
1 gs (
0
Q+ E =
E 0 gs E E
2
gs
(44)
428
1
=
2
E
i (
gs E )
i
gs E 0 gs
!
E
rot
= 0.
(45)
5.2. Supercharges in SU(2) SO(2) parametrization

At first we do not specify the parity of n and employ the harmonics expansion (Eqs. (30),
(32), (33)) for n . The nondifferential terms in (44), (45) can be easily calculated to give:
fg
vE,
2

fg 0
F ,
E E = ei(n1)
2

fg i1 +
E ,
e F uE + ei1 F w
E E = ei(n1)
2
where we have introduced three vectors:
E=
uE = {cos 2 + i cos sin 2 ,

vE = {sin sin 2 ,
sin 2 i cos cos 2 ,
sin cos 2 ,
w
E = {cos 2 i cos sin 2 ,
(46)
(47)
i sin },
cos },
sin 2 + i cos cos 2 ,
i sin }.
The explicit expressions for E E , are too bulky and are listed in Appendix A. Here we
quote the results deduced by means of the computer calculations:

1
E ,
E 0 = ei(n1) ei1 Rn+ F0 uE + ei1 Rn F0 w

i 2

i(n2)
(f g)
(f + g)
F + Rn+ +
F ,
Rn
E E = 2 e
fg
fg

1
+
R2n
F + + R2n
F
E + LF 0 ,
E E = i2 2 ein
fg
where
Rn+

=
n
+
+
,
f
g f + g
Rn

=
+
f
g f g
(48)
(49)
(50)
and L is an irrelevant differential operator.

Remind, that the harmonics content of the wave function n differs with respect to the
parity of n:
for odd n:
for even n:
F0 = F + = F = 0,
F 0 = 0.
At the same time it is seen from Eqs. (47), (49) that all the terms in Eq. (44) for Q are
expressed solely in terms of F and F0 . Hence, for
odd n:
Qn 0.
(51)
429
n 6= 0, i.e., for
On the other hand, it is known that H n = En and Qn = 0 implies Q
odd n:
n 6= 0.
Q
The case of even n is more involved. By using the above formulae we get
p

fg
= 0 2gs R + F + + R F +
F0 = 0
even n: Q
2n
2n
2 gs
and
even n:
(52)
(53)
p
fg +
2gs Rn F0 + F = 0,
gs
p
fg
2gs Rn F0 + F = 0,
Q = 0
gs

(f g)
(f + g)
+
+
F
F = 0.
R
n
n
fg
fg
One can express F + and F from the first and the second equations respectively. The third
equation of the system above is nothing but the consistency condition for the first and the
second ones.
Now we are in position to examine the restrictions on VE,nq coming from Qn = 0 and
Qn = 0. As it was shown above n can be integer or half-integer. We start with the case
of integer nq . (We omit the subindex q below.) The content of VEn is the following: n ,
n = 0 and Q
n+2 = 0. Both n and n + 2 have the
n 1 , n+2 and n+ 3 with Q
2
2
n=0
same parity so according to (52) odd n are forbidden. For n even the constraint Q
results in PDE (53). We shall examine this case in Section 5.4.
Next we turn to a half-integer n = n0 + 12 . VEn0 + 1 is spanned by n0 + 1 , n0 , n0 + 3 and
2
2
2
n0 +1 with Qn0 = 0, Qn0 +1 = 0. This time one of two numbers either n0 or n0 + 1
will be odd. Let, for instance, n0 to be odd. As follows from (51) Qn0 = 0 and the only
equation to be solved is the Schrdinger equation (39). It will be examined in the next
section.
5.3. Half-integer nq
In this section we examine the solution of Eq. (39)

f 2g2 0
g2 + f 2
g2 + f 2
F = EF 0
F0 +
gs 1 + D 2 2 (n 1)2 2
2
2
8gs
g f
(f g )
by using the BornOppenheimer method in the region f g, as it is adopted in current
literature. This procedure yields an approximate formulae for the
and wave
R spectrum
0
0
function. Note, that in the current variables k k is proportional to F F J df dg.
To eliminate the first derivatives term D we search for the wave function in the form
1
F 0 = 0
J
that yields the equation
430

2
2
g2 + f 2
1 (f 2 + g 2 )3
2 g +f
gs 1 +
(n
1)
0
4 f 2 g 2 (f 2 g 2 )2
g2 f 2
(f 2 g 2 )2
+
f 2 g2 0
= E 0
8gs
and
(54)
Z
k k = C
0 0 df dg.
We start with introducing an extra parameter a as follows: f f/a 2 and g ag and

expanding over the powers of a. Separating the variables as 0 = m (g|f )k (f ) we get
the leading term as

3 1
2
2 2
+
b
g
m = m m ,
2+
4 g2
g
where b = f/(2 2gs ).

The discrete spectrum is given by m = 4b(m + 1), where m = 0, 1, 2, . . ., and m is
expressed in terms of Laguerre polynomials
1
1
2
m = (1)m bg 2 e 2 bg L1m (bg 2 ).
g
The next to leading term in the a-expansion yields the equation for (we omit the
index k)

f
2
(m + 1) = E,
2 +4
f
2 2gs
2gs
where we temporary suppressed the term 1/f 2 . The change of variables f = x + E 2(m+1)
yields an equation for Airy function

2
2(m + 1)
x
= 0.
gs
x 2
The solution in terms of y = ( 2(m + 1)/gs )1/3 x is

q

2 3/2
y K
for y > 0,
3 1/3 3 y
=

|y| J
2
2
3/2 + J
3/2
for y < 0.
1/3 3 |y|
1/3 3 |y|
3
Imposing a boundary condition (0) = 0 we get

J1/3 (k ) + J1/3 (k ) = 0
where
2 gs
E 3/2 .
k =
3 m+1
By using

N () sin
J () cos
J () + J () = 2 cos
2
2
2
431
and asymptotic of J () and N() for large

s

2
cos
,
J ()
2
4
s

2
sin
,
N ()
2
4
we get
3
k = k + , k = 0, 1, 2, . . . .
4
Thus, the result is

3 (m + 1)(k + 34 ) 2/3
Em,k =
gs
2
and
m,k cm,k fg 3/2 e 2 fg L1m (fg 2 )
3 K1/3 2 a 1/2 3 ,
3

J
2 1/2 3
2 1/2 3
a +J
a ,
1/3 3
where
1
,
=
2 2gs
1/3 3
2(m + 1)
a=
,
gs
f > a1 Em,k ,
f < a1 Em,k ,
(55)

1/2

1

= f Em,k .
a
The first correction to the energy is given by

R
V
df
.
1E = R
df
In our case V = [(n 1)2 1/4]/f 2 and the direct calculation gives

4 (n 1)2 14
3 2/3 (m + 1)(k + 34 ) 2/3
1+
.
E=
gs
3 (k + 34 )2
2
5.4. Even nq and continuous spectrum
Here we apply the BornOppenheimer method in the region f g to the system (40)
(42). Our aim is to employ the toy-model-like ansatz for the wave function and justify the
appearance of the continuous spectrum [15].
For f g the leading part in the system (40)(42) reads
1 2e
f +
f
e0 ,
a F0
F +
F = EF
8gs2
2gs
2gs

a2
1
f e
1
F0 = EF + ,
1 + D 2 F+ + 2F+ + 2F
2gs
2g
8gs
2g
e0 +
(1 + D)F
(56)
(57)
432

1
a2
1
f e
1 + D 2 F + 2F + 2F+ +
(58)
F0 = EF ,
2gs
2g
8gs
2g
e0 .
where we rescaled F0 = 2 F
+
Taking F = F we find that (57) and (58) coincide. The reduced system reads:
e0 +
(1 + D)F
a2 e
f
e0 ,
F0 F + = E F
8gs2
gs
(1 + D)F + +
a2 +
f e
F
F0 = EF + .
8gs2
2gs
(59)
(60)
e0 .
It is obvious that (59) and (60) will coincide as well if one put F + = 1 F
2
e0 0 / j yields D 1/(4g 2 ) and we are left

In the asymptotic region the rescaling F
with the equation

1
f
1
0 = E 0 .
1 + 2 0 + 2 f 2 g 2 0
4g
8gs
2gs
Recall, that equation

1
1
+ 2 m + 2 f 2 g 2 m = m m
2
g
4g
8gs
has normalizable solutions for m = g2f

(m + 1/2), where 1/2 is very important (not 1 !).
s
The next step is to solve the equation

2
+
O(1/f
)
= E.
f 2
We confirm that for
m = 0 the linear term from the energy m is fully cancelled by the
linear potential f/( 2gs ) thus producing the continuous spectrum.
6. Discussion and conclusion

The results obtained in Section 5 are collected in the table presented in Fig. 3. States
from the quartets are situated along the horizontal lines. Eigenvalues of N for the states in
a quartet nq are nq , nq 1/2 and nq + 2, ng + 3/2. Each item of the quartet can be
obtained from any other one by the action of supercharges or the permutation operator P .
Our results show that odd nq are forbidden, for even nq the spectrum has a continuous part
as well as a discrete one, meanwhile for half-integer nq the spectrum is purely discrete.
Fig. 3 demonstrates that there is a possibility to put a superselection rule to exclude
the presence of the continuous spectrum. Namely, taking any state from quartet with halfinteger nq one gets with guarantee the state of discrete spectrum. One can also specify the
purely discrete sector by using the supercharges, i.e., the states with integer n satisfy the
n = 0.
constraint Qn = 0, meanwhile states with half-integer n satisfy Q
Besides the continuous spectrum SU(2) SO(2) matrix supersymmetric quantum mechanics possesses one more distinguish feature as compared with the bosonic counterpart.
433
= 0, filled circles correspond to Q = 0;

Fig. 3. Structure of the spectrum: stars correspond to Q
domains bounded by dots correspond to the presence of the continuous spectrum; shared domains
correspond to excluding values of n and nq . Quartets are located along the horizontal lines.
This concerns the behaviour of the wave functions in half-integer nq sector, where spectrum is purely discrete. The expression (55) for the wave function illuminates this difference. It is not a difficult task to realize that in the pure bosonic case the maximum of the
wave function appears at the minimum of the classical potential (fg)2 , i.e., at the bottom
of the valley f = 0 as it should be by the semiclassical reasons. For the supersymmetric
case the picture is quite different, the classical equilibrium line have nothing to do with the
maximum of the wave function (55), this time the wave function tends to zero as f 0.
The origin of this transformation is the reflecting wall like 1/f 2 created by the activation
of the SU(2) angular degrees of freedom for the triplet (l = 1) solution.
Let us note that we have obtained only the asymptotic formula for spectrum. Numeric
investigations (see [44] and references therein) allow to trust the asymptotic formula. It
would be interesting to perform numerical calculations also in our case. It would be also
interesting to study 0 -corrections [42] to the spectrum.
It was argued [40] that the holographic feature of the matrix theory can be related
with the repulsive feature of energy eigenvalues in the quantum chaotic system. Relation
between chaos and holography has been discussed recently in [45]. Quantum chaos in
supersymmetric matrix quantum mechanics will be a subject of future investigations.
Acknowledgments
We would like to thank L.O. Chekhov, B.V. Medvedev, A.K. Pogrebkov, O.A. Rytchkov,
N.A. Slavnov and I.V. Volovich for useful discussions. This work was supported in part by
434
RFFI grant 99-01-00166 and by grant for the leading scientific schools 96-15-96208. I.A.
is also supported by INTAS grant 96-0698.
Appendix A
Derivatives in new coordinates are

(f sin 1 ig cos 1 )
2 i(+2 ) sin (f cos 1 + ig sin 1 )
e
+
=
+
4
fg
fg
2
(f 2 g sin 1 + f 3 cos sin 1 + ig 3 cos cos 1 + ifg 2 cos 1 )

fg(f 2 g 2 )
1
(g sin 1 + f cos sin 1 + if cos 1 + ig cos 1 cos )
f 2 g2

(cos cos 1 i sin 1 )

,
(61)
(i cos sin 1 cos 1 )
f
g

(f sin 1 ig cos 1 )
2 i(2 ) sin (f cos 1 + ig sin 1 )
e
+
=
4
fg
fg
2
(f 2 g sin 1 + f 3 cos sin 1 + ig 3 cos cos 1 ifg 2 cos 1 )

fg(f 2 g 2 )
1
(g sin 1 f cos sin 1 + if cos 1 ig cos 1 cos )
+
f 2 g2

(cos cos 1 + i sin 1 )

,
(62)
(i cos sin 1 + cos 1 )
f
g

(f sin 1 + ig cos 1 )
2 i(2 ) sin (f cos 1 ig sin 1 )
e
+
=
4
fg
fg
2
(f 2 g sin 1 + f 3 cos sin 1 ig 3 cos cos 1 + ifg 2 cos 1 )

fg(f 2 g 2 )
1
(g sin 1 + f cos sin 1 + if cos 1 ig cos 1 cos )
f 2 g2

+ ( cos cos 1 + i sin 1 )

,
(63)
+ (i cos sin 1 cos 1 )
f
g

(f sin 1 + ig cos 1 )
2 i(+2 ) sin (f cos 1 ig sin 1 )
e
+
=
4
fg
fg
2
(f 2 g sin 1 f 3 cos sin 1 + ig 3 cos cos 1 + ifg 2 cos 1 )

fg(f 2 g 2 )
1
(g sin 1 + f cos sin 1 if cos 1 ig cos 1 cos )
f 2 g2

(cos cos 1 + i sin 1 )

,
(64)
+ (i cos sin 1 + cos 1 )
f
g
+
435

cos (if sin 1 + g cos 1 )
2 i cos (if cos 1 g sin 1 )

e
+
=
2
fg
fg sin
2
3
(f 2 g cos 1 + if 3 cos2 sin 1 g 3 cos2 cos 1 ifg 2 sin 1 )
fg sin (f 2 g 2 )
1

sin (g cos 1 + if sin 1 )
sin sin 1
+ sin cos 1
, (65)
+
f
g
f 2 g2

cos (if sin 1 g cos 1 )
2 i cos (if cos 1 + g sin 1 )
e
+
=
3
2
fg
fg sin
2

(f 2 g cos 1 + if 3 cos2 sin 1 + g 3 cos2 cos 1 ifg 2 sin 1 )
1
fg sin (f 2 g 2 )

sin (g cos 1 + if sin 1 )
+ sin sin 1
+ sin cos 1
. (66)
+
f
g
f 2 g2
References
[1] S.G. Matinian, G.K. Savvidy, N.G. Ter-Arutunian-Savvidi, Classical YangMills mechanics.
Nonlinear color oscillations, JETP 53 (1981) 421.
[2] S.G. Matinian, G.K. Savvidy et al., Stochasticity of classical YangMills mechanics and its
elimination by Higgs mechanism, JETP Lett. 34 (1981) 590.
[3] H.M. Asatrian, G.K. Savvidy, Configuration manifold of YangMills classical mechanics, Phys.
Lett. A 99 (1983) 290.
[4] G.K. Savvidy, YangMills quantum mechanics, Phys. Lett. B 159 (1985) 325.
[5] B.V. Chirikov, D.L. Shepelyansky, Stochastic oscillation of classical YangMills fields, JETP
Lett. 34 (1981) 163.
[6] E.S. Nikolaevsky, L.N. Shchur, Nonintegrability of the classical YangMills fields, JETP
Lett. 36 (1982) 218.
[7] M. Lscher, Some analytic results concerning the mass spectrum of YangMills gauge theories
on a torus, Nucl. Phys. B 219 (1983) 233.
[8] B. Simon, Some quantum operators with discrete spectrum but classically continuous spectrum,
Ann. Phys. 146 (1983) 209.
[9] B.V. Medvedev, Dynamical stochasticity and quantization, Teor. Mat. Fiz. 60 (2) (1984) 224.
[10] A.M. Badalyan, Two and three matrix models with SU(2) internal symmetry, Yad. Phys. 39
(1984) 947.
[11] Ya.A. Simonov, QCD Hamiltonian in the polar representation, Yad. Phys. 41 (1985) 1311.
[12] J. Goldstone, unpublished.
[13] J. Hoppe, Quantum theory of a massless relativistic surface, Ph.D. Thesis, 1982; in: Proceedings
of the Workshop on Constraints Theory and Relativistic Dynamics, MIT, World Scientific, 1987.
[14] B. de Wit, J. Hoppe, H. Nicolai, On the quantum mechanics of supermembranes, Nucl. Phys.
B 305 (1988) 545.
[15] B. de Wit, M. Lscher, H. Nicolai, The supermembrane is unstable, Nucl. Phys. B 320 (1989)
135.
[16] E. Witten, Dynamical breaking of supersymmetry, Nucl. Phys. B 188 (1981) 513.
[17] M. Claudson, M. Halpern, Supersymmetric ground state wave functions, Nucl. Phys. B 250
(1985) 689.
436
[18] R. Flume, On quantum mechanics with extended supersymmetry and nonabelian gauge
constraints, Ann. Phys. 164 (1985) 189.
[19] M. Baake, P. Reinicke, V. Rittenberg, Fierz identities for real Clifford algebras and the number
of supercharges, J. Math. Phys. 26 (1985) 1070.
[20] C.M. Hull, P.K. Townsend, Unity of superstring dualities, Nucl. Phys. B 438 (1995) 109.
[21] E. Witten, String theory dynamics in various dimensions, Nucl. Phys. B 443 (1995) 85.
[22] T. Banks, W. Fischler, S.H. Shenker, L. Susskind, M theory as a matrix model: a conjecture,
Phys. Rev. D 55 (1997) 5112; hep-th/9610043.
[23] U.H. Danielsson, G. Ferretti, B. Sundborg, D-particle dynamics and bound states, Int. J. Mod.
Phys. A 11 (1996) 5463; hep-th/9603081.
[24] D. Kabat, P. Pouliot, A comment on zero-brane quantum mechanics, Phys. Rev. Lett. 77 (1996)
1004; hep-th/9603127.
[25] M. Douglas, D. Kabat, P. Pouliot, S. Shenker, D-branes and short distances in string theory,
Nucl. Phys. B 485 (1997) 85.
[26] J. Hoppe, On the construction of zero energy states in supersymmetric matrix models I, II, III,
hep-th/9709132; hep-th/9709217; hep-th/9711033.
[27] J. Frhlich, J. Hoppe, On zero-mass ground states in super-membrane matrix models, Commun.
Math. Phys. 191 (1998) 613; hep-th/9701119.
[28] P. Yi, Witten index and threshold bound states of D-branes, Nucl. Phys. B 505 (1997) 307;
hep-th/9704098.
[29] S. Sethi, M. Stern, D-brane bound state redux, Commun. Math. Phys. 194 (1998) 675; hepth/9705046.
[30] M. Porrati, A. Rozenberg, Bound states at threshold in supersymmetric quantum mechanics,
Nucl. Phys. B 515 (1998) 184; hep-th/9708119.
[31] J. Hoppe, S.-T. Yau, Absence of zero energy states in reduced SU(N) 3d supersymmetric Yang
Mills theory, hep-th/9711169.
[32] M.B. Halpern, C. Schwartz, Asymptotic search for ground states of SU(2) matrix theory, Int. J.
Mod. Phys. A 13 (1998) 4367; hep-th/9712133.
[33] J. Hoppe, S.-T. Yau, Absence of zero energy states in the simplest d = 3 (d = 5?) matrix models,
hep-th/9806152.
[34] A. Konechny, On asymptotic Hamiltonian for SU(N) matrix theory, JHEP (1998) 9810; hepth/9805046.
[35] G.M. Graf, J. Hoppe, Asymptotic ground state for 10-dimensional reduced supersymmetric
SU(2) YangMills theory, hep-th/9805080.
[36] J. Frhlich, G.M. Graf, D. Hasler, J. Hoppe, S.-T. Yau, Asymptotic form of zero energy wave
functions in supersymmetric matrix models, hep-th/9904182.
[37] M. Bordemann, J. Hoppe, R. Suter, Zero energy states for SU(N): A simple exercise in group
theory?, hep-th/9909191.
[38] B.V. Medvedev, Dinamical stochasticity and integrals of motion, Teor. Mat. Fiz. 79 (1989) 618.
[39] B.V. Medvedev, Hamiltonian and commutation relations, Teor. Mat. Fiz., to appear.
[40] I.Ya. Arefeva, P.B. Medvedev, O.A. Rytchkov, I.V. Volovich, Chaos in M(atrix) theory, hepth/9710032.
[41] I.Ya. Arefeva, A.S. Koshelev, P.B. Medvedev, Chaosorder transition in matrix theory, Mod.
Phys. Lett. A 13 (1998) 2481; hep-th/9804021.
[42] I. Arefeva, G. Ferretti, A. Koshelev, Taming the non-Abelian BornInfeld action, Mod. Phys.
Lett. A 13 (1998) 2399; hep-th/9804018.
[43] I.M. Gelfand et al., Representations of Rotation and Lorentz Group and Their Applications,
FizMatGiz, Moscow, 1958 (in Russian).
[44] L. Salasnich, Classical and quantum perturbation theory for two non-resonant oscillators with
quartic interaction, quant-ph/9803069.
[45] G.t Hooft, Quantum gravity as a dissipative deterministic system, gr-qc/9903084.

An algebraic bootstrap for dimensionally reduced

quantum gravity
M. Niedermaier a, , H. Samtleben b,1
a Department of Physics, 100 Allen Hall, University of Pittsburgh, Pittsburgh, PA 15260, USA
b Laboratoire de Physique Thorique de l Ecole Normale Suprieure 24, Rue Lhomond,
75231 Paris Cdex 05, France 2
Received 21 December 1999; accepted 27 March 2000
Abstract
Cylindrical gravitational waves of Einstein gravity are described by an integrable system (Ernst
system) whose quantization is a long standing problem. We propose to bootstrap the quantum theory
along the following lines: the quantum theory is described in terms of matrix elements, e.g., of the
metric operator between spectral-transformed multi-vielbein configurations. These matrix elements
are computed exactly as solutions of a recursive system of functional equations, which in turn is
derived from an underlying quadratic algebra. The Poisson algebra emerging in its classical limit
links the spectral-transformed vielbein and the nonlocal conserved charges and can be derived from
first principles within the Ernst system.
Among the noteworthy features of the quantum theory are: (i) the issue of (non-)renormalizability
is sidestepped and (ii) there is an apparently unavoidable spontaneous breakdown of the SL(2, R)
symmetry that is a remnant of the 4D diffeomorphism invariance in the compactified dimensions.
2000 Elsevier Science B.V. All rights reserved.
1. Introduction
Attempts to construct a quantum theory of gravity based on a functional integral
formulation have so far been unsuccessful. Initially this was thought to be a breakdown
only of its perturbative expansion. Meanwhile various reasonably looking discretized
versions of the functional integral have (in all likelihood) failed to produce a continuum
limit of the desired form. This may indicate the necessity to incorporate specific forms of
matter, it may indicate a failure of the functional integral approach, or the analogy to the
nie@prospero.phyast.pitt.edu
1 henning@lpt.ens.fr
2 UMR 8548: Unit Mixte de Recherche du Centre National de la Recherche Scientifique et de lEcole Normale
Suprieure.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 2 0 7 - 8
438
M. Niedermaier, H. Samtleben / Nuclear Physics B 579 (2000) 437491
quantization of conventional field theories may just be physically misleading altogether.

In any case it seems desirable to locate the source of the problem more clearly by
studying model situations sufficiently complex to make (non-)renormalizability an issue,
but tractable enough to be mathematically controllable.
An intriguing such system is the Ernst system, an infinite-dimensional subsector (the two
Killing vector field reduction) of the full phase space of general relativity; see [16] and
references therein. Together with its abelian truncation it has become a prominent testing
ground to explore certain quantum issues of gravity; see, e.g., [713]. The reduced phase
space is equivalent to that of a two-dimensional diffeomorphism invariant field theory. The
latter couples 2D gravity via a dilaton field to a 2D matter system equivalent to the
noncompact O(1, 2) nonlinear sigma-model. This means, the matter degrees of freedom
are mappings n: H2 from the 2D spacetime manifold to the hyperboloid H2 =
{n = (n0 , n1 , n2 ) R1,2 | n n = (n0 )2 (n1 )2 (n2 )2 = 1, n0 > 0}. If h denotes
the (Lorentzian) metric on and R (2) (h) its scalar curvature, the action of the twodimensional system is given by

Z
1
(1.1)
S = d2 x h R (2) (h) + h n n ,
2
where is the bilinear form on R1,2 . The relation to the coset action principles
usually employed in the literature and that to the original Ernst variables is described in
Appendix A.
Depending on the signature of , this sector physically describes either stationary
axisymmetric solutions or gravitational waves with additional symmetries. The latter case
comprises depending on the norm of the Gowdy universes, colliding plane
wave solutions, and (the case considered here) cylindrical gravitational waves. The original
EinsteinRosen waves form a collinearly polarized subsector, they have n1 0, and are
thus described by the Abelian O(1, 1) subtheory of (1.1).
At first sight, the action (1.1) seems perfectly amenable to conventional quantum field
theoretical techniques. Upon closer inspection however, one is quickly led to address the
following questions:
(i) Can one expect 2D conformal invariance to be unbroken?
(ii) What is the physics of the flat space sigma-model with target space H2 ?
(iii) Is the theory (1.1) renormalizable in perturbation theory?
Let us briefly comment on these issues:
(i) Flat space nonlinear sigma-models are known to exhibit dynamical mass generation, destroying the 2D conformal invariance of the classical theory. In the gauge h =
e2 (where is the flat 2D Minkowski metric) the classical system (1.1) likewise
exhibits a conformal symmetry, and off-hand there is no reason why it should not again be
broken in the quantum theory. Rather, there are strong indications for a dynamical breaking
of the conformal symmetry from other 2D quantum gravity models [14,15]. In any case, it
seems advisable not to employ 2D conformal invariance as a guiding principle to construct
the quantum theory.
439
(ii) Although sigma-models with a noncompact target space have been studied for some
time [1621], even basic qualitative features are unknown. Perturbative renormalizability
in the presence of an infrared cutoff should largely parallel that of the compact case [22].
However the free field Fock space carrying a nonlinear realization of the O(1, N) group
action has indefinite metric [21] and the projection onto a physical state space is not fully
understood. Likewise the extent to which the results [23,24] on the infrared finiteness carry
over has to be examined. One might hope a non-perturbative construction to be feasible
via the lattice approach. Specifically, the results of a lowest order large N analysis of the
O(1, N) models [1618] suggest that a non-trivial continuum limit theory might emerge
when using spacelike hyperbolic variables n n = 1/g0 , g0 > 0, and sending the bare
coupling constant g0 to zero. (Classically this is the region where the Hamiltonian density
fails to be positive semi-definite.) Relying on these indications one would further guess that
the resulting QFT is massive, has unbroken O(1, N) symmetry, a positive definite physical
Hilbert space, and a unitary S-matrix. However this scenario has not been corroborated so
far.
(iii) Here, renormalizability should mean in particular that the target space geometry
H2 is left intact by the renormalization process. In conformal gauge one may take
as a loop counting parameter, in which case the model is 1-loop renormalizable in the
background field expansion [25]. For higher loops however one cannot expect off-hand
that the results for the generalized Riemannian sigma-models [26,27] used in the context
of string theory, employing a much weaker notion of renormalizability, will carry over. Let
us emphasize that the answer is not automatic even if one takes the (ultraviolet and infrared)
renormalizability of the non-compact sigma-models in flat space for granted. One way to
see this is to fix the 2D diffeomorphism invariance in (1.1) completely and to identify
with one of the coordinates on (this corresponds to the 4D Weyl coordinates). Then
(1.1) becomes a flat space action, though with an explicit coordinate dependence which
in particular destroys Poincar invariance. Clearly the presumed renormalizability of the
Poincar invariant flat space theory is not very indicative for the behavior of the other
system.
In summary none of the conventional quantum field theoretical techniques presents itself
to construct a quantum theory for (1.1). We propose therefore to bootstrap the quantum
theory from structures linked to its classical integrability. In upshot the quantum theory is
described in terms of matrix elements, e.g., of the metric operator between certain spectraltransformed multi-vielbein configurations. These matrix elements are described exactly as
solutions of a recursive system of functional equations, which in turn are derived from an
underlying dynamical algebra. The Poisson algebra emerging in its classical limit links the
spectral-transformed vielbein and the non-local conserved charges and can be derived from
first principles within the Ernst system. Schematically one can summarize the approach as
follows:
Dynamical algebra
Functional equations
Exact matrix elements
for sequences of
in (renormalized)
meromorphic functions
quantum theory
440
We first briefly describe the ingredients of the above scheme and then comment on why we
expect it to yield a viable quantum theory for the Ernst system.
The data for the dynamical algebra D are: A solution R of the YangBaxter equation,
a real parameter , and a choice of -operation. There are two sets of generators T ( )ba
and Wa ( ), where C, and the indices a, b refer to a basis in a finite-dimensional vector
space. The algebra D = D(R, , ) associated with the data is basically the most general
simple associative algebra, where simple means that all ideals have been divided out.
In the case at hand the data are as follows: R is the rational sl2 R-matrix multiplied with
a scalar function that ensures a suitable unitarity and crossing condition. The parameter
vanishes, which corresponds to the case where the algebra has an enlarged center. The
latter roughly speaking ensures that the quantum theory has the same number of dynamical
degrees of freedom as the classical theory. For any other value of this correspondence
would be violated and degrees of freedom transversal to the reduced phase space would
become dynamical (without describing a consistent bigger portion of the full phase space).
The possible -structures turn out to fall into several equivalence classes; the proper one is
selected by matching it against that of the classical Poisson algebra.
Once the correct dynamical algebra D has been identified one considers linear
functionals D 3 X h|X|i C over D, where the vectors h| and |i are
characterized by the conditions: T + ( )ba |i = +ba |i and h|T ( )ba = ba h|. The
are numerical matrices subject to certain consistency conditions, and in general
h|i = 0. Given such a functional one can introduce the sequence of functions
faN a1 (N , . . . , 1 ) = h|WaN (N ) Wa1 (1 )|i,
N > 1.
(1.2)
The relations of the dynamical algebra then imply that this sequence satisfies a recursive
system of functional equations, where the consistency of the underlying algebra ensures the
consistency of the functional equations. Conversely the original functional over D provides
an abstract solution to the functional equations. The functional equations can be grouped
into two sets (I) and (II). The set (I) characterizes the functions (1.2) for fixed N, the
second set (II) prescribes how the solutions of (I) are arranged into sequences. Essentially
(II) stipulates that the functions (1.2) have simple poles whenever two -variables differ
2 / l , l being the
by a fixed purely imaginary number (which in physical units equals ilpl
z pl
Planck length and lz the unit length along the symmetry axis) and that the residues at these
poles are linked to a function (1.2) in N 2 variables.
Having arrived at the functional equations one can in principle forget about their
derivation and take them as the starting point. The aim then is to construct explicit
solutions in the form of sequences of meromorphic functions f (N) = faN a1 (N , . . . , 1 ).
To understand their physical interpretation the analogy to the form factor approach [28]
to relativistic integrable QFTs is useful. For these systems a similar interplay between
a dynamical algebra and a system of functional equations (the so-called form factor
equations) exists [29,30]. The solutions f (N) in that case describe the form factors of
the QFT, i.e., matrix elements of some local operator between the physical vacuum and
an asymptotic multi-particle state. In the case at hand of course a conventional quantum
field theoretical framework is not available; in particular scattering states in the usual sense
441
are unlikely to exist. Nevertheless the functions f (N) can still be interpreted as the matrix
elements of some operator between a ground state and a multi-Wa ( ) configuration.
From the analysis of the semi-classical limit one finds that the Wa ( ) are the quantum
counterparts of a spectral-transformed vielbein variable. Moreover the analogy to the CP1
system (cf. Appendix A) suggests to view them as confined degrees of freedom rather than
generators of scattering states. Similarly as in the form factor approach the identification
of the operator whose matrix elements are obtained in that way requires external input. Of
course identification here to some extent just means making contact to a conventional
quantum field theoretical formulation where the operator in question is constructed as a
composite operator from a set of fundamental field operators. As explained before such a
more conventional formulation is presently not available for the Ernst system, so that the
identification of the operator underlying a sequence f (N) in this sense has to be left for
future work.
Note that at no stage any renormalization procedure entered. This is a known, yet
striking, feature of the form factor approach. Mathematically it can be understood in terms
of the extreme rigidity of the underlying dynamical algebra, which simply does not allow
for interesting continuous automorphisms that could account for a renormalization process.
The same is true in the present setting and suggests that the functions f (N) indeed are exact
matrix elements that do not require renormalization.
Physically our most important finding is an apparently unavoidable spontaneous
breakdown of the global SL(2, R) invariance that is a remnant of the original 4D
diffeomorphism invariance in the Killing coordinates. Technically this emerges because
the consistency conditions on the matrices mentioned before Eq. (1.2) do not admit a
SL(2, R) invariant solution. As a consequence the functions f (N) are invariant only under
a maximal compact SO(2) subgroup. Despite the trivial technical origin the symmetry
breaking is a genuine dynamical feature intimately linked to the structure of the dynamical
algebra; in particular it disappears in the semi-classical limit.
As explained before we also regard it as more likely than not, that the 2D conformal
invariance of the classical theory (1.1) will be broken in the quantum theory. A fortiori
then also the diffeomorphism invariance in the two non-Killing coordinates of the Ernst
system will be lost. Together both remnants of the original 4D diffeomorphism invariance
(i.e., that it the Killing and in the non-Killing coordinates) appear to be broken in the
quantum theory for dynamical reasons. Being a field theoretical phenomenon that does not
have a counterpart in systems with finitely many degrees of freedom, the result may well
have significance beyond the symmetry-reduced theory.
The article is organized as follows. In the next section, the dynamical algebra D is
introduced and the functional equations (I), (II) for the matrix elements (1.2) are derived.
In Section 3 we discuss the semi-classical limit of the construction and explain the relation
to the phase space of the classical Ernst system. Also various aspects of the symmetry
breaking are detailed. A compilation of useful action principles for the matter sigmamodel is deferred to Appendix A. In Section 4 a solution technique for the functional
equations (I), (II) is described. We adapt techniques from the algebraic Bethe ansatz which
are summarized in Appendix B. In particular, the apparently new concept of sequential
442
Bethe roots and Bethe vectors is introduced. Finally a list of explicit solutions for the
functions (1.2) with N 6 4 is collected in Appendix C.
2. Dynamical algebra and functional equations
Here we describe the dynamical algebra and derive the functional equations (I), (II) for
the objects (1.2). The discussion is naturally organized into two steps. First an algebra DI is
introduced giving rise to the functional equations (I). Then DI is shown to still contain twosided ideals; the factor algebra obtained by dividing out these ideals is the full dynamical
algebra D and gives rise to the additional functional equations (II). We begin by describing
DI and initially keep the data (R, , ) generic.
2.1. W -extended Yangian doubles
The algebras DI are centrally extended Yangian doubles DY (R) with generators
T ( )ba supplemented by generators Wa ( ). We write WY(R, , ), where R is a is a
solution of the YangBaxter equation satisfying unitarity and crossing. Further is a
real parameter and refers to a choice of -operation. Lower indices refer to a basis in
a finite-dimensional vector space V ; upper indices refer to the dual basis, where indices
are raised and lowered by means of the constant charge conjugation matrix Cab and its
inverse C ba , associated with the given R-matrix. For the moment we only need the data R
and ; the possible -operations will be discussed below. To any R-matrix and parameter
one can assign an associative algebra WY with unity by postulating the following exchange
relations among its generators:
cd
c
d mn
(12 ) T (1 )na T (2 )m
Rmn
b = T (2 )n T (1 )m Rab (12 ),
cd
c +
d mn
(12 ) T + (1 )na T (2 )m
Rmn
i h /),
b = T (2 )n T (1 )m Rab (12 + 2i h
n
Cmn T ( )m
)b = Cab ,
a T ( i h
C mn T ( )am T ( + i h )bn = C ab ,
dc
(12 )Wc (2 )T (1 )ed ,
T (1 )ea Wb (2 ) = Rab
dc
(12 + i2h i h /)Wc (2 )T + (1 )ed ,
T + (1 )ea Wb (2 ) = Rab
dc
(12) Wc (2 ) Wd (1 ), Re 12 6= 0,
Wa (1 )Wb (2 ) = Rab
(T1)
(T2)
(TW)
(WW)
with 12 := 1 2 . The parameter h is included for later convenience; it can be given any
non-zero value by a rescaling of the variables. For convenience we also assume that real
boosts in the variables are unitarily implemented, i.e., eiK X( )eiK = X( + ), with
R, for any generator X( ) of the algebra. For completeness we also note the precise
form of the YangBaxter equation
kp
ji
pi
kj
nm
nm
(12 )Rnc (13)Rmp (23 ) = Rbc
(23)Ram (13 )Rpn (12),
Rab
and the conditions of unitarity and real analyticity

cd
mn
cd
cd
( )Rnm
( ) = ad bc ,
Rab ( ) = Rab
( ).
Rab
(2.1)
(2.2)
443
We add a few remarks. The algebra DY (R) is a well-known structure. For = 2

it can be viewed as a presentation of the quantum double of some underlying infinitedimensional Hopf algebra [31]. The (TW) and (WW) relations are then characteristic for
the intertwining operators between quantum double modules [32,33]. Particular cases are
in which case the parameter can
the Yangian double or the quantum double of Uq (g),
be related to the central extension via c = 2i(1 /2). Here we do not make use of
the co-algebra structure and always treat as a real numerical parameter. The case of the
critical level with enlarged center [34,35] in our conventions corresponds to = 0; for the
sl2 Yangian case it will be studied in detail below.
In preparation, we introduce the following quadratic element which turns to play a
decisive role in the WY algebras
Dab ( ) := Ccd T ( + i h )ca T + ( )db ,
(2.3)
It enjoys the following exchange relations

mn
mn
(21 + i h / 2i h )Dcn (1 )Wm (2 ) = Rac
(12 + i h )Wm (2 )Dnb (1 ),
Rab
mn
mn
(21)Dcn (1 )T + (2 )dm = Rac
(12 i h + i h /)T + (2 )dm Dnb (1 ),
Rab
mn
mn
(21 + i h / 2i h )Dcn (1 )T (2 )dm = Rac
(12 + i h )T (2 )dm Dnb (1 ),
Rab
mn
lk
(21)Rmc
(21 + i h / i h )Dnk (1 )Dld (2 )
Rab
mn
kl
(12 + i h / i h )Dal (2 ) Dkm (1 ).
= Rcd (12)Rbn
(2.4a)
(2.4b)
(2.4c)
(2.4d)
In particular, it follows from these relations that the (generalized) quantum current D( ) =
C ab Dab ( ) lies in the center of the algebra WY(R, 0, ), for any R-matrix obeying a
standard crossing relation with a symmetric charge conjugation matrix Cab . Much of the
construction described in the rest of the paper could therefore be transferred to a fairly
general class of infinite-dimensional quantum algebras.
With the application to the quantum Ernst system in mind however we specialize
already at this point to the sl2 Yangian R-matrix [3638]. Its charge conjugation matrix
is anti-symmetric which enforces a slight modification of the above scheme. The relevant
R-matrix is then given by

i h
cd
c d
d c
+
, a, b, . . . = 1, 2,
(2.5)
Rab ( ) = r( )
i h a b i h a b
where

ch
2h + i sh 2h 2 + 2i h 2i h
r( ) =

,
ch
12 2ih 2ih
2h i sh 2h
satisfies

i h
1 i h 2
i h
r( ) = 1
+
+ O (i h / )3 .
r( )r( i h ) = 1 ,
2 8
Further r(0) = 1 and r(i h + ) = /i h , r(i h + ) = i h /, for 0. For later
reference we list the main properties of the R-matrix (2.5). In addition to the YangBaxter
equations (2.1) and (2.2) one has a sign modified crossing invariance
0
dc
( ) = Caa 0 C dd Rda0 bc (i h ),
Rab
with Cab = iab .
(2.6)
444
The R-matrix (2.5) has no poles in the strip 0 6 Im 6 i h but a simple zero at = i h /2.
At = 0, i h it becomes a permutation matrix and a projector, respectively,
cd
(0) = ad bc ,
Rab
cd
(i h ) = ac bd + ad bc = Cab C cd .
Rab
(2.7)
The semi-classical expansion is

cd
( ) = ac bd
Rab
2

i h cd
i h
ab
ad bc 58 ac bd + O (i h / )3 ,
(2.8)
cd
= ad bc 12 ac bd .
with ab
Next we determine the possible -operations of the WY-algebras with the R-matrix (2.5).
Starting with a general linear ansatz one finds
0
Wa ( ) = Faa Wa 0 ( + i h ),
0
T + ( )ba = Faa Ebb0 T ( + i h / i h )ba 0 ,

0
T ( )ba = (E 1 )bb0 (F 1 )aa T + ( + i h / i h )ba 0 ,
(2.9)
where E, F are GL(2, C) matrices satisfying

EE = 1,
{1},
F F = 1,
det E det F = 1.
(2.10)
Any solution of (2.10) yields a consistent -operation (2.9), for any value of . We omitted
a trivial overall shift by a purely imaginary number in the arguments on the right-hand side
of (2.9). We also omitted scalar prefactors on the right-hand side which can be removed by
a rescaling of the generators. The operator Dab ( ) is basically hermitian with respect to
any of the -operations (2.9), (2.10)
Dab ( ) = Dba ( + i h / 2i h ),
D( ) = D( + i h / 2i h ),
where D( ) := C ab Dab ( ).
(2.11)
Clearly, most of these -operations will be equivalent in being related by an automorphism

of the WY-algebra. Such automorphisms are provided by SL(2, C) basis transformations
0
0
0
Wa ( ) faa Wa 0 ( ), T ( )ba faa gbb0 T ( )ba 0 , f, g SL(2, C), under which the
-structure (2.9) transforms covariantly as
E 7 g 1 Eg ,
7 ,
F 7 f Ff 1 .
(2.12)
It is not hard to classify the inequivalent -structures in the general case. With regard
to the Ernst system however we restrict attention to real linear transformations and thus
require the matrices E, F to be real. This leaves four cases for the possible -operations,
corresponding to the sign choices sign(det F ) = sign(det E) and {1}. Consider first
F : if det F = 1 then F = A 1 , where A SO(2) and j , j = 1, 2, 3, are the Pauli
matrices. If det F = 1 then F = 1. In the former case one can achieve F = 3 by a
similarity transformation; in the latter case one can take F = 1, because the sign can be
absorbed either into E or into a rescaling of the generators. It turns out that det F = 1 is
the case relevant for the Ernst system, so for brevity we consider the possible Es only
445
for det E = 1. The general solution of E 2 = 1, E SL(2, R), then is readily worked
out. For = 1 it leaves only E = 1, for = 1 one finds a two-parameter family of
Es; by a similarity transformation each of its members can be mapped onto E = i 2 . In
summary, we always take F = 1 in (2.9), which leaves only two inequivalent -structures
implemented by real E matrices, namely
= 1; E = 1
and = 1; E = i 2 .
(2.13)
Note that with the second choice the SL(2, R) basis transformations acting on the upper
index are restricted to the SO(2) SL(2, R) subgroup leaving E fixed.
From now on DI will denote the algebra WY(R, 0, ) with the R-matrix (2.5), the
parameter = 0, and the -operation (2.9) with F = 1 and E given by one of the matrices
in (2.13). The case with = 1 will turn out to be the one relevant for the Ernst system.
Often we shall treat the = 1 case as well in order to emphasize the crucial differences
entailed by the seemingly minor flip. For convenient reference let us note explicitly
Wa ( ) = Wa ( + i h ),
T + ( )ba = Ebb0 T ( i h )ba ,
T ( )ba = Ebb0 T + ( i h )ba ,
(2.14)
with E as above, as the -operation of DI . SL(2, R) transformations acting on the lower

index are -automorphisms of DI , and similarly SO(2) rotations acting on the upper index.
Generic SL(2, R) transformations acting on the upper index in contrast are automorphisms
but do not preserve the -structure. Rather the matrix E transforms covariantly as E
g 1 Eg, g SL(2, R).
In addition DI admits some simple -automorphisms given by -dependent rescalings
of the generators. Explicitly
Wa ( ) ( )Wa ( ),
T ( )ba ( )T ( )ba ,
(2.15)
are -automorphisms of DI provided the scalar functions ( ), ( ) obey

( ) = ( + i h ),
( ) = ( + i h ),
( ) ( i h ) = 1.
The last equation in particular entails that
( )
(2.16)
are 2i h periodic functions.
2.2. Diagonalizing the center at the critical level = 0

For = 0 the quantum current is central. Explicitly

[D(1 ), Wa (2 )] = 0 = D(1 ), T (2 )ba , Re 21 6= 0.
(2.17)
The second equation is well known [34]. The first one follows similarly from (2.4), which
also explains the origin of the CDD-like sinh-prefactor in (2.5). 3 Since D( ) is central
it is natural to search for representations of DI on which D( ) acts like a multiple of the
3 However, the relations (2.17) do not imply that the antisymmetric part of D ( ) decouples algebraically.
ab
Defining Mab ( ) := 2i (Dab ( ) + Dba ( )), the relations (2.4d) (at = 0) do not hold with Dab ( ) replaced by
Mab ( ). We shall see in Section 3 how to separate the symmetric part of Dab ( ) in the classical limit.
446
unit operator. The Fock space representations of the Yangian double at the critical level
inherited from a free field realization [35,39] do not have this property. Experience from
other contexts suggests to search for appropriate representations in terms of functionals
over the algebra DI .
Specifically we consider vector functionals (called T -invariant)
DI 3 X hXi = h|X|i,
(2.18)
built from a pair of vectors |i and h| satisfying

T + ( )ba |i = + ( )ba |i,
h|T ( )ba = ( )ba h|,
(2.19)
( )
are numerical matrices which according to (T1), (T2) carry one-dimensional

where
representations of the Yangian algebra Y (sl 2 ), respectively. This implies ( ) =
( ) with 2i h -periodic scalar functions ( ) and constant matrices , satisfying
the relations ( ) ( i h ) = 1 and Ccd ca db = Cab . It is natural to supplement the
conditions on ( ) by the first condition in (2.16). A rescaling (2.15) of the T generators
then allows one to dispense of the -dependence of the matrices in (2.19). Henceforth
we shall use constant matrices. Hermiticity h (X)i = hXi then imposes the conditions
b
+b
0
0
a = +ba Ebb0 .
(2.20)
a = ba Ebb0 ,
We shall mainly need the combination
+n kb
ab := Cmn m
a kC ,
satisfying
b
b
+n
= C bk Cmn m
a = 1 a ,
k a,

det = 1.
ab + ab = aa 1,
1
b
a
(2.21)
Observe that the value (2.18) of the central element D( ) is given by the trace aa and
reinforces the distinct features of the = 1 and = 1 involutions in (2.9):
h|D( )|i = aa h|i,

2, if = 1 and real,
a
a =
0,
if = 1 and real.
(2.22)
Natural choices are: = 1 for = 1, and = 1, + = E for = 1, in which case

ab = ab and ab = Eab , respectively.
Clearly any T -invariant functional (2.18) is uniquely determined by its values on strings
of W -generators, for which we introduce some extra notation
fA ( ) := faN a1 (N , . . . , 1 ) := hWaN (N ) Wa1 (1 )i,
(2.23)
where Re ij 6= 0, i 6= j , = (N , . . . , 1 ), A = (aN , . . . , a1 ). Sometimes also the

shorthand f (N) for the value of h i on a string of N W -generators will be used. With
these definitions one computes
hWaN (N ) Wak+1 (k+1 )D(0 )Wak (k ) Wa1 (1 )i
=T
(0 | )B
A hWbN (N ) Wb1 (1 )i,
(2.24)
447
where T (0 | ), = (N , . . . , 1 ), is basically the familiar transfer matrix

a b
B
T (0 | )B
| )A ,
A = b Ta (0 + i h
1
Tab (0 |N , . . . , 1 )baNN b
a1
(2.25)
cN bN1
c2 b1
:= RcbbN NaN (N,0 )RcN1
aN1 (N1,0 ) Ra a1 (1,0 ).
Implicit in (2.25) are two important features reflecting the fact that D(0 ) is central:
the right-hand side of (2.25) is k independent and for fixed CN and varying 0
C the T (0 | ) form a one-parameter family of commuting matrices. Hence they can be
simultaneously diagonalized and on the eigenvectors D(0 ) will act like a multiple of the
unit operator. We are thus lead to restrict attention to those functionals (2.18) (or later a
subset thereof) for which the functions (2.23) obey
T (0 | )B
A fB ( ) = (0 | )fA ( ).
(2.26)
We note the following hermiticity properties of the T -matrices and their eigenvalues:
T
B T

T (0 | )B
+ i h AT ,
A = T 0 2i h

(2.27)
(0 | ) = 0 2i h | T + i h ,
which follow from (2.11), (2.24) and the general hermiticity condition h (X)i = hXi . The
notation is T = (1 , . . . , N ), AT = (a1 , . . . , aN ), etc. A further important property of the
eigenvalues (0 | ) of (2.26) is
(k i h | ) (k 2i h | ) = 1,
k = 1, . . . , N.
(2.28)
To derive this, consider the matrix T (0 | )T (0 i h | ), which describes the action of

D(0 )D(0 i h ) within the matrix elements (2.23). Then
B

B
, k = 1, . . . , N,
(2.29)
T (0 | )T (0 i h | ) A = i h = A
0
as may be verified by direct computation from (2.25).

The matrix ab in (2.25) can be thought of as describing the deviation from the SL(2, R)
symmetry. In particular T (0 | ) is invariant only under the subgroup of SL(2, R) matrices
obeying
ac bc = cb ca .
(2.30)
For = 1 one may take = 1 and the condition is empty. For = 1 the solutions
of (2.30) can be seen to generate a maximal compact subgroup SO(2) SL(2, R). This
holds irrespective of any further constraints on the matrix , but for the reasons outlined
in Section 2.3 below we take to be real. Of course still has to obey the constraints
in (2.21) and the most general real solution to them may be parameterized as
!
sh ch
, 0 6= R, R.
(2.31)
(, ) :=
1 ch sh
An SL(2, R) basis transformation in (2.24) maps in (2.25) onto g g 1 , g SL(2, R).
The transformed matrix still obeys the constraints in (2.21) and hence can be
parameterized as in (2.31), however with different values for and . The matrices
448
therefore define a conjugacy class in SL(2, R). The transfer matrix T and its eigenvalues
depend on the representative g g 1 , while the eigenvalues are class-functions, i.e., depend
only on the conjugacy class. In particular the eigenvalues of itself are i, independent of
and . For any fixed matrix the solutions of (2.30) then generate the missing compact
conjugacy class of SL(2, R). Explicitly the solutions of (2.30) are given by
ba () = ab cos + (, )ba sin ,
for = 1, 0 6 < 2.
(2.32)
As anticipated, they generate an SO(2) subgroup of SL(2, R). Furthermore, they satisfy
C bm nm ()Cna = ba (). The invariance of T (0 | ) is expressed by
b1
bN
c1
cN
B
T (0 | )C
A c1 () cN () = a1 () aN ()T (0 | )C .
(2.33)
In particular, (2.33) allows one to break up the eigenvalue problem (2.26) into subsectors
of fixed SO(2) charge; cf. Appendix B.
The diagonalization (2.26) of T of course is the object of the Bethe ansatz. We shall
be interested in solutions which are in addition equivariant with respect to the usual
representation of the permutation group SN on the space of tensor-valued functions.
Whence we require

1
(2.34)
fA ( ) = Ls ( )B
A fB s , s SN .
It suffices to specify the action of the generators s1 , . . . , sN1 of SN :
sj (N , . . . , 1 ) = (N , . . . , j , j +1 , . . . , 1 ),
Lsj ( )B
A
1 6 j 6 N 1,
bj+2 bj bj+1
bj1
aj+2
Raj+1 aj (j +1,j )aj1
= abNN
ab11 ,
(2.35)
and the product

Lss 0 ( ) = Ls ( )Ls 0 s 1 ,
s, s 0 SN .
(2.36)
Not surprisingly, the system (2.26) is compatible with (2.34) provided the eigenvalues
(0 | ) are completely symmetric in = (N , . . . , 1 ), which we henceforth assume to be
the case. It suffices to verify the asserted compatibility for the generators of SN ; this in turn
follows from the identities
0
B
B
B
Lsj (sj )A
A T (0 | )A0 Lsj ( )B 0 = T (0 |sj )A ,
1 6 j 6 N 1.
(2.37)
2.3. Further properties of the eigenvectors

The joint solutions (2.26) and (2.34) enjoy a number of other remarkable properties. First
they are also solutions to an asymptotic form of the deformed KnizhnikZamolodchikov
equations (KZE) [28,40], where in the present conventions 2i(1 /2) parameterizes
the level. The critical level = 0 corresponds to the limiting case where these equations
degenerate into an eigenvalue problem for mutually commuting matrices Qk , namely (see,
e.g., [41])
Qk ( )B
A fB ( ) = qk ( )fA ( ),
(2.38)
449
with
b b
N
d c
k+1
k1
1
k
Qk ( )B
A := c Tak (k |N , . . . , k+1 )aN ak+1 Td (k |k1 , . . . , 1 )ak1 a1 .
To see the relation to (2.26) note

B

T (0 | )B
A = i h = Qk ( )A .
0
(2.39)
(2.40)
Hence (2.38) is a consequence of (2.26) with qk ( ) = (k i h | ). The converse is also

true as we show in Appendix B.2. The T eigenvalue problem (2.26) and the seemingly
weaker Qk eigenvalue problem (2.38) therefore are equivalent for = 0.
For 6= 0 the deformed KZE has been found to have an algebraic counterpart [30]:
supplementing the relations of WY (R) by
+
n kl
Wa ( + i h /) = Cmn T ( + i h /)m
/ i h )l C ,
a Wk ( )T ( + i h
(2.41)
the matrix elements (2.23) automatically solve the deformed KZE with level i(2 /),
or equivalently a cyclic equation with shift parameter . Moreover the relation (2.41)
endows the algebra with a modular structure characteristic for a quantum system at finite
temperature 1/. Here we show that much of this structure survives in the limit 0.
We begin by showing that the joint solutions of (2.26) and (2.34) also enjoy the following
cyclic property

1
(2.42)
fA ( ) = ( )B
A fB .
Here is the Coxeter element = sN1 s1 SN , acting by cyclic permutation
(N , . . . , 1 ) = (1 , N , . . . , 2 ) on elements of CN and
b bN
ab12 .
( )B
| )aN1 aN1
A := (N 2i h
(2.43)
Of course one could also have used (2.34) to obtain the relation fA ( ) = L ( )B
A fA
( 1 ), where L ( ) is the representation matrix of SN defined by (2.35), (2.36).
Consistency is ensured by the fact that

1
B
1
(2.44)
L ( )B
A fB = ( )A fB ,
on the joint solutions of (2.26) and (2.34). In other words, although ( ) and L ( ) are
distinct as matrices, they act in the same way on the solutions of (2.26) and (2.34). This
follows from
fA ( )
(2.28)
(2.25)
(N 2i h | )T (N i h | )B
A fB ( )
b
1
(N 2i h | )adN TdbN (N |N1 , . . . , 1 )aN1
f ( )
N1 a1 B

(2.35), (2.36)
B
=
(N 2i h | )adN L 1 1 a a d fB ( )
N1
1

(2.43)
C
1
=
( )B
A L 1 B fC ( ),
and together with (2.34) we obtain (2.42) and (2.44). One may check that consistently the
matrix T has the cyclic property
450
1 bN1 b1 d c

1 bN

T (0 | )B
A = T 0 aN1 a1 c aN
d
1 B 0 1 B
A0
= ( )A T 0 A0 ( )B 0 .
(2.45)
From here one can show that the cyclic equation (2.42) is in fact equivalent to the Qk
eigenvalue problem: specializing (2.45) to 0 = k i h yields cyclicity relations for the
Qk matrices and their eigenvalues

bN1 b1 d c
1 bN
,
Qk1 ( )B
A = Qk ( )aN1 a1 c aN
d
qk1 ( ) = (k i h | ) = qk ( ).
(2.46)
In particular modulo the exchange relations (2.34) the eigenvalue equation (2.38) for
k = N, say, entails all others. On the other hand from the computation before (2.45) one
also sees that the cyclic equation (2.42) and the QN eigenvalue equation are equivalent.
Combining both facts shows the asserted equivalence of (2.38) and (2.42).
A further property of the Qk matrices and their eigenvalues is unravelled by iterating the
cyclic equation (2.42)
N
Y
qk ( )fA ( ) = ()N abNN ab11 fB ( ),
k=1
Q1 ( ) QN ( )
B
= ()N
A
N
Y
ajj ,
(2.47)
j =1
where the second equation follows from the first one together with the fact that for generic
s the matrices Qk have maximal rank. From (2.28) one finds qk ( )qk ( i h ) = . Finally
the qk ( ) have two properties that are readily seen only in the Bethe ansatz construction
relegated to Appendix B: for real they are pure phases and the phase is a gradient; see
also [41,42]. Explicitly, qk ( ) = qk ( )1 and
qk ( ) = eik 1() ,
with k 1( ) = ( Nk ), k = 1, . . . , N,
(2.48)
where ( ) = ( ) := i ln (N i h | ). The fact that the qk ( ) are pure phases is

linked to having in (2.21) chosen to be real.
Real -matrices are also natural because they allow one to introduce a quadratic form
on the space of solutions to (2.26), (2.34). Consider the following quadratic form on the
space of V N -valued functions
Z
(2.49)
hf, gi = dN fA ( ) C AB gB ( ).
It is manifestly Hermitian hf, gi = hg, f i and SL(2, R) invariant. Equations (2.34), (2.38),
and (2.42) give rise to symmetry operations which are unitary with respect to h , i in the
sense that
hf, gi = hf 0 , g 0 i
for

1
fA0 ( ) = Ls ( )B
A fB s ,
s SN ,
fA0 ( ) = qk ( )1 Qk ( )B
A fB ( ),

0
B
1
fA ( ) = ( )A fB .
451
(2.50a)
(2.50b)
(2.50c)
!
Hence, for real ab , h , i induces a quadratic form h , isol on the space of functions f 0 = f ,
which are the joint solutions of (2.34), (2.26) (2.34), (2.38) (2.34), (2.42).
Now let us return to the algebraic description. Since for = 0 (and only then)
the KZE (2.38) and the cyclic equation (2.42) are consequences of the T -eigenvalue
problem (2.26) and (2.34) one expects that in the algebra DI extra algebraic relations hold
which imply (2.38), (2.42) for the matrix elements (2.23). This is indeed the case and the
relevant relations are
+
n kl
Cmn T ( )m
)l C = Wa ( )D( i h ),
a Wk ( )T ( i h
D( 2i h )D( i h )Wa ( ) = Wa ( ).
(2.51a)
(2.51b)
The first relation can be verified simply by pushing T to the right using (TW).
Equation (2.51) then is required for compatibility with the -operations (2.9). Indeed,
applying to (2.51a) yields
+
n kl
Cmn T ( i h )m
)a C = D( 2i h )Wa ( ).
k Wl ( )T ( 2i h
(2.52)
Employing associativity and (T2) one recovers (2.51a) if (2.51b) holds. Inserting
now (2.51a) into the kth position of a matrix element (2.23) precisely produces the kth
KZE equation (2.38). Using the (WW) relations any of them can be seen to be equivalent
to the cyclic equation (2.42). Finally the algebraic consistency relation (2.51b) amounts
to (2.29) when used within the matrix elements (2.23).
Let us summarize the construction until here. For = 0, the quantum current lies in the
center of the algebra DI . Searching for T -invariant functionals (2.18), (2.19) on which it
acts like a multiple of the unit operator, leads to the eigenvalue problem (2.26). We were
interested in those solutions which also enjoy the R-matrix exchange relations (2.34) and
found that they have a number of remarkable bonus properties. Most notably, they satisfy
the asymptotic form (2.38) of the deformed KZE, which we showed to be equivalent to the
cyclic equation (2.42). Moreover on the space of permutation equivariant functions (2.34)
the three requirements: T eigenvalue problem (2.26), Qk eigenvalue problem, and the
cyclic condition (2.42) are all equivalent. The actual solution, e.g., of the T eigenvalue
problem (2.26), (2.34) is relegated to Appendix B. All this was for a fixed number N
of variables j . Next we show that the eigenvectors and eigenvalues for different N can
naturally be arranged into sequences.
2.4. Sequences of eigenvectors
Apart from the nontrivial center, the algebra DI = WY(R, 0, ) contains further twosided ideals which should be divided out. The following relations (R) can be checked to
452
arise in this way and we suspect them to be the only ones 4

Wa ( + i h ) Wb ( ) = ( )Cab D( i h ),
C Wa ( i h ) Wb ( ) = ( i h )D( 2i h ),
ab
( )
(R)
= ( ).
Under the action of the -automorphism

where the function ( ) satisfies
(2.15) ( ) changes according to
( )( + i h )
.
(2.53)
( ) ( )
( ) + ( + i h )
We will later comment on specific choices for ( ). The algebra DI where in addition the
relations (R) are imposed, is our complete dynamical algebra D.
The operator product Wa (1 )Wb (2 ) turns out to be singular as 21 i h , with a first
order pole. The in (R) indicates a normal product defined roughly by taking the residue
at the pole. A more precise definition will be given below in terms of its action within the
T -invariant functionals. For the moment we are only interested in the algebraic properties
of the relations (R).
A stronger, uncontracted version of the second relation is
Wa ( i h ) Wb ( ) = ( i h )Dab ( 2i h ).
(2.54)
Let us momentarily denote these relations by (R1), (R2) and (R3), respectively. Recall that
the exchange relations (TW) are valid also for Re 12 = 0, while for the (WW) relations
these points are a priori excluded. Consider the following formal extension of (WW) to
12 = i h ,
cd
(i h )Wd ( ) Wc ( + i h ),
Wa ( + i h ) Wb ( ) = Rab
(2.55)
The extra minus sign of (2.55) with respect to (WW) is in accordance with the residue
interpretation of the -product. Then:
(R2) H (R3)
by means of (2.51), (TW),
(R1) (R2), (R3)
by means of (2.55).
(2.56)
The first implication can be seen by starting from Wa ( i h )Wb ( ), replacing Wa ( i h )

n
in favor of D( 2i h )1 Cmn T ( i h )m
)T + ( 2i h )l C kl , then using (TW)
a Wk ( i h
+
to move T to the right and finally applying (R2). Thus (R2) and (R3) are equivalent in
D and both are readily seen to be formally equivalent to (R1) once one is allowed to
use (2.55). In order to avoid potential troubles with the formal identity (2.55), however, we
postulate both equations (R) independently, keeping in mind that they are formally related
by (2.55).
We can now restore topological concepts by calling a T -invariant functional (2.18)
analytic if:
(i) The dependence of the values f (N) on the parameters N , . . . , 1 is locally analytic,
possibly with branch points but without cuts.
4 It appears to be a general rule that reducibility of a tensor product of fundamental representations is always
caused by a pole in the R-matrix [43,44]. According to (TW) the Wa play the role of intertwiners between two
such representations.
453
(ii) For X, Y DI the expectation value hXWa (1 )Wb (2 )Y i has a simple pole at 12 =
i h with residue (2 )hXCab D(2 i h )Y i; similarly hXC ab Wa (1 )Wb (2 )Y i has a simple
pole at 12 = i h with residue (2 i h )hXD(2 2i h )Y i. In particular this defines the
product in (R).
From now on we assume all functionals (2.18) to be analytic in this sense. The relations (R)
then imply the N N 2 recursive relations (II) given below for the functions (2.23).
They link the eigenfunctions and eigenvalues of the eigenvalue problem (2.26), (2.34) in
N and in N 2 variables.
In summary we arrive at the following system of functional equations:
(I)
T (0 | )B
A fB ( ) = (0 | )fA ( ),
(a)
fA ( ) = Racdk+1 ak (k+1,k )faN dca1 (N , . . . , k , k+1 , . . . , 1 ),
(b)
where N is fixed, k = 1, . . . , N1, and T is the transfer matrix (2.25). The solutions of the
functional equations (I) for N and N 2 are linked by
Resk+1 =k +i h fA ( ) = (k ) (k i h |pk )Cak+1 ak fpk A (pk ),
(II)
Resk+1 =k i h C
ak+1 ak

(0 | )
fA ( ) = (k i h ) (k 2i h |pk )fpk A (pk ),
k+1 =k i h
= (0 |pk ).
(a)
(b)
(c)
Here k = 1, . . . , N 1, and we adopted the notation pk = (N , . . . , k+2 , k1 , . . . , 1 ),

pk A = (aN , . . . , ak+2 , ak1 , . . . , a1 ). Eq. (IIc) arises as consistency condition for (IIa,b)
whenever their right hand sides are nonvanishing, using the properties

pk B

Cbk+1 bk T (0 | )B
A k+1 =k +i h = Cak+1 ak T (0 |pk )pk A ,

p B
bk+1 bk

T (0 |pk )pkk A ,
(2.59)
C ak+1 ak T (0 | )B
A = i h = C
k+1
of the transfer matrix. Observe that specializing here 0 to k i h yields relations

reciprocal to (2.40). Similarly on the level of the eigenvalues (IIc) for 0 = k i h gives

(2.60)
qk ( ) = +i h = (k i h |pk ),
k+1 k
(2.61)
qk+1 ( ) = i h = (k 2i h |pk ).
k+1
The restrictions of (0 | ) to 0 = k i h and to k+1 = k i h commute. For

completeness let us also note the recursive equation implied by (2.54)

(2.62)
Resk+1 =k i h fA ( ) = 12 (k i h )Cbk+1 bk Qk+1 ( )B
A = i h fpk B (pk ).
k+1
Clearly, viewed as a projection, N N 2, the recursive operation defined through (II)

must have a large kernel. Simply counting the dimensions one expects 2N eigenvectors on
the left-hand side of (II) to be mapped onto only 2N2 on the right-hand side. Conversely
given a solution of (I) in N 2 variables it is non-trivial that Eqs. (II) can be integrated
to a function of N variables within the class of solutions of (I). The explanation why this is
possible stems from the underlying algebraic framework.
Recall the notation D for the algebra DI = WY(R, 0, ) supplemented by the relations (R). As in Section 2.2 we can consider T -invariant functionals, now over the algebra D, i.e.,
454
h i : D C,
X hXi,

b
b
XT + (0 )ba = +ba hXi,
T (0 )a X = a hXi,
(2.63)
keeping all other notations. Again, such a functional will be completely determined by its
values on strings of W -generators. By construction, it gives rise to a solution of the system
of functional equations (I), (II) with the identification fA ( ) := hWA ( )i, and vice versa.
Symbolically we arrive at a one-to-one correspondence:
sequence of
T -invariant functional
solutions of (I), (II)

over D
3. Semi-classical limit and phase space of the Ernst system
Here we study the semi-classical limit of the regular part of the dynamical algebra and
show that essentially the same Poisson algebra describes the classical phase space of the
Ernst system. In particular this lends a physical interpretation to the variables T ( )ba and
Wa ( ).
3.1. Semi-classical limit of the dynamical algebra
We begin with the classical limit of the algebra DI and assume that semi-classically the
operator products in DI behave like a Moyal product, i.e.,

i h cl cl
X , Y + O h 2 ,
XY = Xcl Y cl
2
where Xcl , Y cl are the corresponding functions on phase space. Usually we shall drop the
superscript cl when no confusion is possible. Upon expansion in h the exchange relations
(T1), (TW), (WW) then provide the symplectic structure of a classical Poisson algebra:

1
ef
f
dc
T (1 )ea T (2 )b ,
T (1 )de T (2 )cf ab ef
T (1 )da , T (2 )cb =
12

+
1
ef
f
dc +
T (1 )ea T (2 )b ,
T + (1 )de T (2 )cf ab ef
T (1 )da , T (2 )cb =
12

1
f
f
cd
T (1 )c Wd (2 )ab
,
T (1 )a , Wb (2 ) =
12

1
cd
Wc (1 )Wd (2 )ab
,
Wa (1 ), Wb (2 ) =
12
(3.1a)
(3.1b)
(3.1c)
(3.1d)
cd from (2.8). In particular the operators T ( )b have turned into classical 2 2

with ab
a
matrices which due to (T2) have unit determinant
Ccd T ( )ca T ( )db = Cab .

The -structure (2.14) translates into the classical hermiticity relations
b
+ b
0
0
T ( )a = Ebb0 T + ( )ba ,
T ( )a = Ebb0 T ( )ba ,

Wa ( ) = Wa ( ).
(3.2)
(3.3)
455
The classical analogue of the matrix Dab ( ) from (2.3) is given by

Dab ( ) = Ccd T ( )ca T + ( )db ,
with [Dab ( )] = Dba ( ).
(3.4)
Separating the symmetric Hermitian and the antisymmetric part for = 1 yields
1
Dab ( ) =: iMab ( ) Cab D( ), D( ) = C ab Dab ( ),
2
1
(3.5)
det Mab ( ) + D( )2 = 1 = det Dab ( ).
4
Since for = 1 D( ) is purely imaginary, one has in particular det M( ) > 1.
We proceed by showing that the antisymmetric part of Dab ( ) (and hence also det Mab )
is not a dynamical degree of freedom of the Poisson algebra (3.1)(3.3). One indication
is the fact that it Poisson-commutes with the generators T ( )ba and Wa ( ). More
specifically, there exists an automorphism of the Poisson algebra such that D( ) is mapped
onto a prescribed numerical constant, e.g., D( ) 0 for = 1, yielding det M( ) = 1.
To find the automorphism consider first the following rotation:
T + ( )ba y++ T + ( )ba + y+ T ( )ba ,
T ( )ba y+ T + ( )ba + y T ( )ba ,
Wa ( ) Wa ( ),
(3.6)
which obviously is a homomorphism of the Poisson algebra (3.1). Note, that the parameters
y may depend on here. If we further specify them to be of the form
cos
,
y = q
1 12 D( ) sin(2 )
y = q
sin
1 12 D( ) sin(2 )
with = ( ), the map (3.6) extends to an automorphism of the full structure (3.1),
(3.2). The condition
+ ( ) = ( ) ,
(3.7)
finally ensures compatibility with hermiticity (3.3). For = 1 thus + = =: must

be real-valued, as is D( ). For = 1 both D( ) and + = =: i, R are
purely imaginary. Under this automorphism the antisymmetric part of the matrix Dab ( )
transforms as
D( ) D( ) :=
D( ) D( ) :=
D( ) 2 sin(2)
1 12 D( ) sin(2)
D( ) 2i sinh(2)
1 2i D( ) sinh(2)
for = 1,
for = 1.
(3.8)
For = 1 there are two disjoint orbits of D( ) R under the Poisson automorphism (3.6),
the interval [2, 2] and its complement. In particular, the fixpoints of (3.8) at D = 2
are fake and correspond to a non-invertible map (3.6). For = 1, on the other hand, the
Poisson automorphism (3.6) acts transitively on D( ) iR. This means starting from any
non-zero value of D( ), this automorphism can be used to define new generators of the
456
Poisson algebra (3.1)(3.3) for which the new D( ) D( ) vanishes. This fact ensures
that one can always work with symmetric SL(2, R) matrices
Mab ( ) := iDab ( ) 1 arcsinh( 1 iD()) = Mba ( ).
2
(3.9)
Further one can assume M( ) to have positive trace. This is because a vanishing
trace would contradict the positive determinant, so that Tr M( ) must be either positive
or negative. Since T ( )ba T ( )ba , Wa ( ) Wa ( ) is a -automorphism of the
Poisson algebra (3.1)(3.3) one can take Tr M( ) > 0. The Poisson brackets of M( )
follow from the classical limit of (2.4d)

Mab (1 ), Mcd (2 )

1
mn
=
mn Mmb (1 )Mnd (2 ) + Mam (1 )Mcn (2 )bd
(3.10)
12 ac

1
mn
mn
Mnd (2 ) + Mmb (1 )ad
Mcn (2 ) .
Mam (1 )bc
+
12
The Poisson algebra (3.10) turns out to provide a direct link to the phase space of the Ernst
system which will be detailed in Section 3.2.
We focused on the regular part DI of the dynamical algebra here because the various
operations invoked: taking the semi-classical limit, taking the residue at 12 = i h
and applying the automorphism (3.6) are mutually non-commuting. In particular taking
the semi-classical limit of the relation (R) would require further specifications. However
since we introduced topological concepts only on the level of the matrix elements, not for
the algebra, it is convenient to discuss the semi-classical limit of the recursive structure
directly on the level of the functional equations (I), (II) and their solutions; cf. Section 4.
3.2. Phase space of the classical Ernst system
The classical phase space of the Ernst system can be described in various ways. A nonredundant parameterization is in terms of gauge invariant symmetric SL(2, R) matrices
Mab ( ), which can be viewed as the scattering data from each of which a classical
solution can be reconstructed. These matrices can be shown, starting from the canonical
Poisson brackets associated with the action (1.1), to carry the Poisson structure (3.10) [45].
Hence in this non-redundant parameterization there is a direct correspondence between the
phase space of the Ernst system and the subsector (3.9), (3.10) of the Poisson manifold
emerging in the classical limit of the dynamical algebra D. The deformation parameter
2 / l , where l is the Planck length and l the unit length along the
h comes out to be lpl
z
pl
z
cylindrical symmetry axis.
In the quantum algebra we saw in Section 2.2 that the antisymmetric part of Dab ( )
does not decouple algebraically. This enforced to work with the bigger algebra generated
by T ( )ba , Wa ( ), and to implement the decoupling in terms of an eigenvalue problem.
We now show that a Poisson algebra essentially equivalent to (3.1) also naturally emerges
in the Ernst system. Moreover this lends a physical interpretation to the variables T ( )ba
and Wa ( ) in (3.1).
457
We begin by recalling that the scalar sector of the Ernst system is given by an SL(2, R)
valued matrix Vam which essentially contains the vierbein components of the Killing
directions. The model is invariant under global SL(2, R) and local SO(2) transformations 5
Vam (x) 7 gab Vbn (x)hm
n (x),
with gab SL(2, R), hm

n (x) SO(2).
(3.11)
This invariance has its roots in the four-dimensional theory, SL(2, R) descending from
linear diffeomorphisms in the Killing coordinates, SO(2) being a remnant of the
corresponding part of the local Lorentz group. The bilinear combination
Mab (x) = Vam (x)Vbn (x)mn ,
(3.12)
is invariant under local SO(2) transformations and corresponds to the metric components
in the compactified dimensions. Note that the symmetric mn symbol appearing here is
invariant only under the SO(2) subgroup of SL(2).
The dynamics of the Ernst system is captured by a Lax pair [3,4] whose spectral
parameter in contrast to the flat space integrable systems depends explicitly on
the spacetime coordinates, see [6] for a review. For definiteness, we focus of the case of
cyclindrical gravitational waves. In particular the worldsheet then has Lorentzian signature
and can be covered by coordinates x = (t, r), t R and r > 0. For the description of the
linear system, light-cone coordinates x = t r are most convenient. It is then given by
b )L (x; ),
V(x; ) = V(x;
b
(3.13)
with
1
P (x).
1
Here, Q and P are the compact and noncompact components of the sl(2, R)-valued
current V 1 V lying in so(2) and its orthogonal complement, respectively. The spectral
parameter is given by the following explicit function
p

1
(3.14)
(x; ) = t ( t)2 r 2 ,
r
of the 2D coordinates x = (t, r) and a constant which may be understood as the
underlying constant spectral parameter of (3.13).
The associated monodromy matrices U (r, r 0 , t | ) are obtained in the usual way as path
ordered exponentials of the Lax connection

V 1 t, r; (t, r; ) b
(3.15)
V t, r 0 ; (t, r 0 ; )
U (r, r 0 , t | ) := b
L (x; ) = Q (x) +
Zr 0
= P exp

dz L1 t, z; (t, z; ) ,
which are unique functionals of the connection L = 12 (L0 L1 ). The integrand in (3.15)
lives on the twofold covering of the complex -plane where we choose the branch cut
5 We use abstract index notation in this section to indicate the transformation behavior of the objects: indices
a, b, . . . from the beginning of the alphabet refer to covariance under SL(2, R), whereas indices k, l, . . . from the
middle of the alphabet refer to covariance under local SO(2) rotations.
458
of (3.14) varying on the real -axis while z runs from r to r 0 . The monodromy matrix
/ R. Under (3.11) it transforms as
U (r, r 0 , t | ) hence is well defined for
U (r, r 0 , t | )nm 7 h1 (t, r)km U (r, r 0 , t | )lk h(t, r 0 )nl ,
(3.16)
in particular, it is invariant under the global SL(2, R) transformations.

Assuming regularity of the currents at the spatial boundaries and time-independence at
spatial infinity corresponding to the classical sector of gravitational waves with regular
Ernst potential on the symmetry axis we define the following objects for real values of
the parameter :
T + ( )ba := i lim V0 (t)na Unk (0, , t | + i)(V )lc Ckl C cb ,
0
(3.17a)
T ( )ba := lim V0 (t)na Unk (0, , t | i)(V )lc kl C cb ,

0
n m
Wm
a ( ) := V0 (t)a Un (0, | t|, t | ),
(3.17b)
with
V0 (t) := V(0, t),
V := V(, t) = V().
The T ( ) are conserved and the W ( ) are conserved up to a local gauge transformation
t T ( )ba = 0,
n
W a ( )Q (t, t)m
n , for > t,
(
)
=
t W m
a
W na ( )Q+ (t, t )m
n , for < t.
(3.18)
The fact that the last equation is not 2D covariant is due to the explicit appearance of the
time t in the integration boundary of W m
a ( ). Observe however that according to Eq. (3.18)
(
)
is
continuous
in since Q1 (t, r =0) = 0, as follows from the
the time derivative of W m
a
field equations derived from (1.1) and regularity of the vielbein Vam on the symmetry axis.
As indicated we shall usually suppress the time-dependence of Wam ( ) in the notation.
Restoring it momentarily Wa ( ) ; Wa (t, ) one finds for the time evolution (e.g., for
> t)
W (t, ) = W (0, )e
i(t,) 2
i
(t, ) =
2
Zt

ds Tr Q+ (s, s) 2 ,
(3.19)
where 2 is the Pauli matrix and (t, ) is real for real . Note, that in contrast to W (t, ),
the function (t, ) is defined by integrating over a null line in spacetime, and may hence
not be considered as canonical object on a fixed time slice. In particular, this makes it
difficult to compute its Poisson brackets.
The matrices T ( ) satisfy (3.2) whereas the W ( ) obey
Cmn Wam ( )Wbn ( ) = Cab ,
C ab Wam ( )Wbn ( ) = C mn .
Under the symmetry transformations (3.11), the matrices (3.17) behave as

b
T ( )ba 7 gad T ( )cd g 1 c ,
d n
m
Wm
a ( ) 7 ga W d ( )hn (t, |
t|).
(3.20)
(3.21)
459
Clearly any gauge invariant quantity build from these monodromy matrices will be time
independent. A gauge invariant object of particular interest is the bilinear combination
Mab ( ) = iT ( )ca T + ( )db Cdc = Wam ( )Wbn ( )mn .
(3.22)
The second equality follows from the definition (3.17) in the limit t and shows that
the matrix Mab is symmetric in the indices a, b, and has positive trace. In particular, for
the T ( )ba defined by (3.17), the combination D( ) = C ab Ccd T ( )ca T + ( )db vanishes
automatically. The decomposition of M( ) into the product of T corresponds to the
RiemannHilbert decomposition; the matrices T ( ) are holomorphic in the upper,
respectively, lower half of the complex -plane. It may further be shown that
Mab ( R) = Mab (t = , r = 0),
(3.23)
i.e., this matrix coincides with the physical scalar fields on the axis r = 0 [5,45]. From
the viewpoint of the inverse scattering transform, Eq. (3.23) is a striking result. Usually,
the scattering data associated with a given solution live at timelike infinity and have no
direct relation to the original field variables. In contrast, for the Ernst system, Eq. (3.23)
means that the scattering data live on the symmetry axis r = 0 and are directly related
to the original vielbein variables. The second Gauss-like decomposition of M into the
bilinear product of W s in (3.22) therefore corresponds to the decomposition of the metric
into a spectral-transformed vielbein. As anticipated by the notation, the matrices (3.22)
can be identified with (3.9) and the decomposition in (3.22) may be viewed as the classical
analogue of (2.54).
The fact that the scattering data have an interpretation in terms of the original field
variables imposes an interesting causality constraint, which also clarifies the structure of
the monodromy matrices Wam ( ): monodromy matrices of the form (3.17b) (path ordered
integrals over finite space intervals without any specification of conditions on the physical
fields at the boundary of the interval) do not arise in the usual flat space integrable
systems. The raison dtre in the Ernst system is the spacetime dependence of the spectral
parameter (3.14). The Lax connection (3.13) degenerates at certain points in spacetime
though the physical currents remain regular. This happens at the spatial boundaries
r = 0, , but also, curiously, at r = | t|, a point by no means distinguished in the
physical spacetime. However this point does have special significance for the causal past
of the point (, 0) on the symmetry axis. For given t0 and (with t0 < , say) consider the
intersection of the causal past of (, 0) with the t = t0 surface, i.e., the interval [0, t0 ].
According to causality in the (t, r) Lorentzian space, the vielbein on the symmetry axis at
time t = should be a functional of the initial data on the interval [0, t0 ] only, which
due to the range of integration in (3.17b) it indeed is; cf. Fig. 1.
We proceed by studying the Hermitian structure and the Poisson brackets of the
monodromy matrices (3.17). The hermiticity U (r, r 0 | ) = U (r, r 0 | ) implies
b
db
(3.24)
T ( )a = iT ( )ca M
cd C ,
m
m
Wa ( ) = Wa ( ),
460
Fig. 1. The spectral transformed vielbein Wam ().

n
m
where the constant matrix M
cd (V )c (V )d mn defines a positive definite bilinear
form on SL(2, R). Due to the explicit appearance of M
ab in (3.24), this hermitian structure
is not invariant but transforms covariantly under (3.21). With
cb
Eab = iM
ac C ,
we recover (3.3), i.e., the classical limit of (2.9). Fixing the SL(2, R) symmetry (3.21) by
m
2
setting (V )m
a = a corresponds to the choice E = i in (2.13). Having done so, one is
still left with SL(2, R) basis transformations acting on the lower index
T ( )ba 7 gad T ( )bd ,
d m
Wm
a ( ) 7 ga W d ( ).
(3.25)
The linear transformations (3.25) provide a -automorphism of the Poisson algebra even
after (3.21) has been eaten up by fixing the -structure (e.g., M
ac = ac ) and even
m
when working with SO(2) gauge fixed Wa ( ) matrices. We shall return to the distinction
between (3.21) and (3.25) in Section 3.3.
The Poisson brackets of the monodromy matrices T coincide with (3.1a,b), where
again the coordinate dependence of the spectral parameter plays a crucial role in the
computation [45]. Evaluating the general formula from [45], one similarly obtains for the
matrices Wam ( )

T (1 )ea , Wbm (2 )
1
cd
T (1 )ec Wdm (2 )ab
=
12
1
W l (2 )T (1 )ec Uak (2 | 1 )(U 1 )cn (2 | 1 )Ekn Elm ,

(3.26a)
212 b

m
Wa (1 ), Wbn (2 )
1 m
cd
W (1 )Wdn (2 )ab
=
12 c
(12 ) l
j
W (2 )Wcm (1 )Uak (2 | 1 )(U 1 )cj (2 | 1 )Ek Eln

212 b
(21 ) k
j
W (1 )Wcn (2 )Ubl (1 | 2 )(U 1 )cj (1 | 2 )Ekm El ,

(3.26b)
212 a
461
n = C kn = (i 2 )n here playing
with the step function ( ) = sign( ), and the matrix Em
mk
m
the role of an SO(2) invariant tensor. Further
k
Uak (2 | 1 ) := V0 (t)m
a Um (0, |2 t|, t | 1 ),
(U 1 )ak (2 | 1 ) := Ukm (|2 t|, 0, t | 1 )V0 (t)nb Cmn C ba .

Evidently the Poisson structure is not closed but contains the transition matrices U (2 | 1 ),
etc. on the right-hand side. Nevertheless the Jacobi identities can be verified, and (3.26)
defines a consistent Poisson structure on the phase space of the Ernst system. As remarked
earlier, the combination D( ) = C ab Ccd T ( )ca T + ( )db here vanishes automatically.
Consistency thus requires that D( ) also Poisson commutes with Wam ( ) (with respect
to the brackets (3.26)), which indeed can be verified to be the case. Moreover it can be
shown that (3.26) induces the Poisson structure (3.10) for the gauge invariant phase space
functions Mab ( ), i.e., the same as the somewhat simpler Poisson structure (3.1) obtained
from the classical limit of our dynamical algebra. Thus, up to redundancies, (3.26) can be
regarded as equivalent to (3.1c,d) in the sense that both induce the same structure on the
space of objects invariant under the SO(2) gauge transformations in (3.11). 6 The extra
U -terms in (3.26) can be viewed as being a remnant of the gauge dynamics (3.19). The
problem is that (3.26) can not readily be rewritten as a Poisson bracket structure on the
initial data Wam (0, ). This is because although the time evolution (3.19) can be viewed
as a gauge transformation, the transformation is a non-local functional of the dynamical
variable Q+ (x), so that at some point non-equal-time Poisson brackets would have to be
evaluated. In principle, however, we view the Wa ( ) in (3.1) as being gauge fixed or gauge
invariant and time-independent versions of a linear combination of Wa1 ( ) and Wa2 ( ). We
have not been able so far to properly map the Poisson brackets (3.26) onto (3.1) for such
a combination. Our main argument that it should be possible is, that on gauge invariant
objects like M( ) both induce the same Poisson structure.
One can also check that the counting of degrees of freedom works out: there are two
types of redundancies in (3.26). First (3.26) is invariant under the symmetry transformations (3.21). In particular the local gauge transformations Wam ( ) Wan ( )hm
n (t, | t|)
effectively remove one degree of freedom. In addition (3.26) has a one-dimensional Poisson center generated by the determinant det Wam ( ). Thus (3.26) contains only two physical
degrees of freedom for the Wam ( ) fields, just as (3.1c,d).
Taking the equivalence of (3.26) and (3.1c,d) for granted, the recursive relations (R)
can be viewed as a quantum implementation of the identity (3.22) and the determinant
condition (3.20). To see this set

(3.27)
Wa ( ) = Wam ( ) 1 m ,
where Ad : SL(2, R) SU(1, 1) is the isomorphism (A.11) or (B.2). Then the Wa ( )
are complex fields transforming as Wa ( ) Wa ( )ei(t,|t |) under SO(2) gauge
transformations with h = cos 1 + sin E. Further, they obey
6 A similar point of view is, e.g., usually adopted to study the symplectic structures on the moduli space of flat
connections on Riemann surfaces [46].
462

1
Mab ( ) Cab ,
(3.28)
2
using the definition (3.22) and the determinant condition (3.20). In the quantum theory (3.28) will turn into a singular operator product. Parallel to (R) one can stipulate
Wa ( )Wb ( ) =
i
(3.29)
Wa ( i h ) Wb ( ) = Dab ( 2i h ),
2
while all products of W + with itself and of W with itself are supposed to vanish.
For simplicity we set = i here and only noted the counterpart of (R2), i.e., (2.54); the
interplay with the other versions is analogous to (R1)(R3). The obvious -operation is
Wa ( ) = Wa ( + i h ). Finally, consider the linear combinations
Va ( ) = ei() Wa+ ( ) + ei() Wa ( ),
(3.30)
with parameter ( ) = (t, ), obeying ( ) = ( + i h ). If in addition we take ( )

to be i h -periodic, the products (3.29) etc. for Wa ( ) imply (R) for Va ( ) with = i. If
the W in (3.30) are rescaled by ( ), obeying ( ) = ( i h ), the same holds with
( ) = i( )( + i h ), a form we shall use later. In principle we could have developed
the entire formalism of Section 2 for an enlarged quantum algebra with a pair of Wa ( )
generators and exchange relations like those in DI for both of them. The advantage of
working with the linear combinations (3.30) is that the recursive relations are simpler
because the same rule applies to each pair of V -generators. We do not expect substantial
difficulties in deriving a system of functional equations analogous to (I), (II) for matrix
elements of mixed strings of W generators. Observe however that according to (3.29) the
recursive structure in any case determines only the propagation of the gauge invariant
parts. For example, matrix elements with only W + s would only be constrained by the
analogue of the functional equations (I).
As remarked before we prefer to analyze the semi-classical limit directly on the level
of the functional equations (I), (II) and their solutions. Observe however that comparing
the formal h 0 limit of (3.29) with (3.28) and (3.5) one can match the expressions by
taking D( ) = 2i. Clearly the Poisson algebra (3.26) admits an automorphism analogous
to (3.6). Starting from D = 0 one can achieve D = 2i by taking sh 2 = 1. The
conjecture that the Poisson structures (3.1) and (3.26) are fully equivalent, translates into
one concerning the status of the field ( ). As long as it is regarded as an independent
parameter, the linear combination (3.30) will obey Poisson brackets of the form (3.1),
modified by the extra U -terms. By allowing ( ) to become dynamical (as in (3.19)) one
may hope to render the Va ( ) time independent and either gauge fixed or gauge invariant.
At the same time the U -terms should disappear, making the correspondence Va ( ) ;
Wa ( ) precise.
Summarizing, in the classical limit the dynamical algebra DI gives rise to a phase
space parametrized by symmetric SL(2, R) matrices M equipped with the Poisson
structure (3.10) and the two alternative RiemannHilbert and Gauss decompositions into
matrices T and W , respectively. The resulting Poisson algebra (3.1)(3.3) essentially
coincides with that deduced from the fundamental Poisson brackets in the classical Ernst
system; admittedly yet with some loose ends.
463
3.3. Symmetry breaking

In retrospect we can now also highlight some aspects of our quantum formulation. As
noted in Section 2.1, the antisymmetric part of Dab ( ) does not decouple algebraically
within DI . Thus in order to avoid that the quantum theory has an extra dynamical operator
field D( ), one is forced to go to the critical level = 0, where D( ) becomes central.
However an automorphism of the form (3.6), (3.8) no longer exists in the quantum theory;
the spectrum of D( ) on the state space described by the matrix elements (2.23) is a
characteristic feature of the system. Remarkably the state space exhibits a (spontaneous)
breakdown of the classical SL(2, R) invariance that is a is a remnant of the original four
dimensional diffeomorphism invariance in the Killing coordinates: as we have seen before
for = 1 the eigenvalue problem (2.26) is invariant only under the SO(2) subgroup,
while the algebra itself and the spectrum of D( ) still are fully SL(2, R) invariant. This
suggests that, in contrast to the prevalent assumptions in many approaches to quantum
gravity, diffeomorphism invariance is not sacrosanct; it might be broken for dynamical
reasons.
In view of the possible implications it seems worthwhile to critically reexamine the line
of argument and to see whether, in the context of the present framework, the conclusion
can be avoided. To address the issue, it is convenient to introduce shorthands for the
different SL(2, R) actions involved. Let SL2D be the SL(2, R) action (3.21) inherited
via (3.11) from the linear diffeomorphisms in the Killing coordinates. Let SL2L be the
SL(2, R) action (3.25) acting by basis transformations on the lower index. Finally denote
0
by SL2U the SL(2, R) action Wam ( ) Wam ( ), T ( )ba gbb0 T ( )ba . The latter two
have obvious counterparts in DI and its classical limit, the SL2D action then corresponds
to the diagonal action diag (SL2L SL2U ). On Dab ( ) both SL2D and SL2L act as
Dab ( ) gac gbd Dcd ( ),
(3.31)
and similarly for Mab ( ). Thus if SL2L , and hence (3.31) is broken, also the invariance
under SL2D must be violated. On the other hand, we saw in Section 2 that on the matrix
elements (2.23) the SL2L invariance is indeed broken down to its SO(2) subgroup.
To confirm this result let us re-examine the underlying assumptions. Technically, the
result only hinges on the fact that for = 1 the hermiticity conditions (2.20) do not
permit a solution yielding 1. One can also convince oneself that simple modifications
(restoring a -dependence in , use of a graded hermiticity requirement, etc.) either do
not affect the above feature or are at odds with the classical limit. In particular, allowing
to become complex (and hence to have a nonvanishing trace by (2.21)) does not help.
The solutions of (2.30) would then be given by real linear combinations of + =
2 Tr 1 and 1, and would still generate not more than a SO(2) subgroup of SL(2, R).
Ultimately, the reason why the seemingly kinematical hermiticity requirement can have
such a severe impact on the choice of the relevant representations is, that the algebra D and
its -structure already contain most of the dynamical information about the system.
The close relation to the hermiticity requirement also explains why a group averaging
procedure does not provide a way out. Since the eigenvalues in (2.26) are SL(2, R)
464
invariant, one might hope that by averaging over the conjugacy class (2.31) the invariance
of T and its eigenvectors can be restored. Leaving aside the arbitrariness stemming from
the non-uniqueness of a regularized invariant measure on SL(2, R), the main problem with
this proposal is that the only SL(2, R) invariant outcome of the averaging procedure in T
can be a av 1. This however violates the hermiticity condition, and one is lead back to
the discussion of the preceding paragraph.
One might also suspect that the lack of SL(2, R) invariance in (2.26) is already present
on the level of the (classical or quantum) algebra through the lack of SL(2, R) invariance
of the -structure (3.3) or (2.9) with respect to the upper index. Since the definition of
D( ) involves a contraction over an upper index pair, one might argue, that the violation
of SL(2, R) invariance already enters at this point. However, this is not the case. Though
not invariant, the -structure (3.3) is SL(2, R) covariant. An SL2U rotation of the form
0
Wa ( ) Wa ( ), T ( )ba gbb0 T ( )ba provides an automorphism of the algebra D
under which the -structure (2.9) transforms covariantly as in (2.12). But in the hermiticity
equations (2.11) the matrix E drops out; both the operator D( ) and its hermiticity
condition (2.11) are invariant under both the SL2U and the SL2L actions of the SL(2, R)
algebra. However its matrix elements (2.23) are not! It is clear that the phenomenon
disappears in the classical limit because then D( ) can be set to zero. In summary we
conclude that the classical SL2L symmetry, and by (3.31) therefore the SL2D invariance,
being a remnant of the 4D diffeomorphisms in the Killing coordinates, is spontaneously
broken in the quantum theory.
Because of the coupling to gravity, no conflict with the ColemanMerminWagner
theorem [47,48] arises. In lattice formulations usually also the compactness of the global
symmetry group is assumed in order to ensure the existence of the regularized functional
integral. If, as in Colemans version [48], the existence of the quantum field theory and
the Noether current is postulated, the result should hold also for non-compact groups.
However the Poincar invariance and the cluster property are essential for the argument.
For the Ernst system the former is manifestly absent and the latter would at least require are
interpretation. In the statistical mechanics context, one way to look at the MerminWagner
theorem is as a tug of war between entropy and energy. For a flat space sigma-model
in 2D the entropy always wins, forcing the system to remain in a disordered state even
at low temperatures. From this perspective the above result indicates that the coupling to
gravity changes this entropyenergy balance such that a breaking of the SL(2, R) symmetry
becomes possible.
Concerning the worldsheet diffeomorphism invariance, it is clear that if 2D conformal
invariance is broken, so is invariance under diffeomorphisms. Classically the constraints
induced by fixing the conformal gauge h = e2 in the action (1.1) are
1
T = 2 + n n,
2
(3.32)
where = 12 ln |+ | is a conformal scalar. Their Poisson brackets form two

commuting copies of the VirasoroWitt algebra. Further (3.32) can be checked to generate
infinitesimal conformal transformations on gauge invariant objects, otherwise an additional
465
SO(2) gauge transformation is induced. In particular, for the conserved charges (3.17) and
Weyl coordinates = r one has {T (t, r), T ( )ba } = 0, and modulo some technical
subtleties a similar equation holds for Wam ( ). Classically the (weakly) vanishing of the
constraints is compatible with the equation of motion for . The latter can be integrated
to express in terms of and P . Off-shell the essential dynamical features of should
be captured by the quantum counterparts of its Poisson brackets with T ( )ba and Wam ( ).
The latter turn out to be remarkably simple

(t, 0), Wam ( ) = Wam ( ),
(3.33)
(t, 0), T ( )ba = T ( ),
assuming that (t, r) vanishes for r . The obvious quantum counterparts are
eiK T ( )ba eiK = T ( + )ba ,
eiK Wa ( )eiK = Wa ( + ).
(3.34)
That is, translations in are unitarily implemented, and the generator K is the quantum
counterpart of the conformal factor in the 2D metric.
4. Exact matrix elements

We return now to the functional equations (I), (II) of Section 2. As outlined, its solutions
are conjectured to describe exact matrix elements in the quantum theory, without the need
for any renormalization. We begin by describing a solution algorithm for (I), (II).
4.1. Solution algorithm
Before turning to the solution procedure, let us note some simple structural features
of (I) and (II). Clearly a solution of (I) is determined only up to multiplication by a scalar
function completely symmetric in N , . . . , 1 . The recursive equations (II) cut down this
ambiguity to scalar, symmetric functions solving
P ( )|k+1 =k i h = P (pk ).
(4.1)
In other words, if (f (N) )N>N0 (with N even or odd depending on the starting member) is
a sequence solving (I), (II) and (P (N) )N>N0 is a sequence solving (4.1), then the sequence
obtained by pointwise multiplication, (P (N) f (N) )N>N0 , again is a solution of (I), (II).
According to (IIc), the eigenvalues sequences are an example of a sequence (4.1) but there
P
sj /h , with s odd, or any smooth scalar function
are many others, e.g., power sums N
j =1 e
of N + + 1 2(t1 + + t ), where t are the Bethe roots of Appendix B. Usually one
will be interested in solutions of (I), (II) which are minimal in the sense that one cannot
naturally split off a solution of (4.1). Apart from these obvious ambiguities we expect
that associated with each starting member f (N0 ) there is basically only a single sequence
solving (I), (II).
To actually find solutions of (I), (II) we make an ansatz of the form
Y (kl )
,
(4.2)
fA ( ) = fA ( )c(N)( )
kl i h
k>l
466
where c(N) ( ) absorbs the ( ) dependence and ( ) satisfies

( ) = r( )( ),
( )( i h ) = 1.
(4.3)
The rationale for (4.3) is that the first relation effectively replaces (Ib) by exchange relations
with a rational R-matrix, the second one turns out to achieve the same for the recursive
relations (IIa,b). The solution of (4.3) analytic in the strip h /2 < Im < h /2 is given by
( Z
)
i
t
dt h(t)
exp
sin (i h + 2 ) ,
( ) = tanh
2h
2
t ch 2t
2h
where h(t) = 2
0
t
/2
+ et
e
.
(4.4)
1 + et
The functional equations (4.3) are readily verified by means of the integral representation
)
( Z
dt
(4.5)
h(t) sin t .
r( ) = exp i
h
t
0
In addition ( ) has the following properties: it has a simple pole at = i h with residue
h 0 , where 0 := i(i h ) = lim0 /(h /) 1.54678. The only real zero is at = 0.
Further it obeys ( ) = ( ) and ( ) i for , cf. Fig. 2.
It remains to specify the prefactors c(N) ( ) in (4.2), supposed to absorb the ( )
dependence and overall constants. For definiteness we assume ( ) to be of the form
( ) = i ( )( + i h ),
(4.6)
where ( ) satisfies ( ) = ( + i h ). The overall i in (4.6) takes care of the = 1

hermiticity property. Comparing with (2.53) one sees that (4.6) amounts to fixing the
normalization of the T (i.e., ( ) 1) while still allowing for automorphisms (2.15)
with a nontrivial ( ). The restriction to ( ) = 1 is not indispensable; however since
their h 0 limits can only be 1 anyhow, not much generality is lost. Conversely any
prescribed ( ) enjoying suitable analyticity and fall-off properties can be written in the
factorized form (4.6). With ( ) of the form (4.6) we then take
Fig. 2. Real (solid) and imaginary (dashed) part of ln[( hu )/ tanh u2 ] for real u.
(N)
( ) = c
(N0 )
1
(N ) (N0 +1 )
0
(NN0 )/2
,
N > N0 .
467
(4.7)
Finally we enter with the ansatz (4.2) into the functional equations (I) and (II). The Eqs. (I)
translate into
T (0 | )B
A fB ( ) = (0 | )fA ( ),
cd
fA ( ) = R a a (k+1,k )faN dca1 (sk ),
(4.8a)
(4.8b)
k+1 k
where

)cd := 1 ac d i h ad c ,
R(
ab
b
b
+ i h
N
Y
(j 0 2i h )r(i h j 0 )T (0 | ).
T (0 | ) :=
(4.9)
j =1
The redefinition of T renders the components of T (0 | ) polynomial in the j 0 i h . The

associated eigenvalues (0 | ) turn out to be rational functions of j , 0 and the Bethe
roots, cf. (B.18). The solutions f in N and N 2 variables are now linked by
Y

(j k + i h )Cak+1 ak fpk A (pk ), (4.10a)
fA ( ) = +i h = (k i h |pk )
k+1
ak+1 ak

fA ( )
k+1 =k i h
j 6=k+1,k
= 2 (k 2i h |pk )
(j k + 2i h )fpk A (pk ),

(0 | ) = +i h = k0 (k0 2i h ) (0 |pk ),
k+1 k
= (k0 i h )(k0 3i h ) (0 |pk ),
(0 | )
k+1 =k i h
(4.10b)
j 6=k+1,k
(4.10c)
where again (4.10c) applies only when the right-hand sides of (4.10a,b) are nonvanishing.
Clearly any one of the Eqs. (4.10c) implies the other, we noted them both for symmetry.
The system of Eqs. (4.8), (4.10) for f is now quasirational in the sense that all ingredients
are rational functions in j , 0 and the Bethe roots, while the Bethe roots themselves are
algebraic functions of the j .
The proper hermiticity requirements are

fA ( ) = fAT T + i h ,

(4.11)
fe;n 1 ( ) = fe;1 n T + i h .
The first one is just the general hermiticity requirement expressed in terms of f. The
second one is the transcription into the charged basis introduced in Appendix B.1. In
brief one can switch to a basis in V N on which the matrices and in (2.31), (2.32) are
diagonalized. The components of f (N) in the new basis are denoted by fe;N 1 ( ), where
j {1} and e = N + + 1 . Under a U (1) ' SO(2) rotation (2.32) they transform as
fe ( ) eie fe ( ), i.e., with charge e. It is easy to see that the functional equations (I), (II),
or (4.8), (4.10), then split up into decoupled sectors of fixed charge e {N, N 2, . . . , N},
making the charged basis particularly useful for their analysis. In group theoretical terms
the basis transformation is related to the isomorphism SL(2, R) SU(1, 1).
468
Let us now address the problem of solving the functional equations (4.8), (4.10),
in the charged basis. As indicated Eq. (4.8a) can be solved by means of the Bethe
ansatz; some details are provided in Appendix B. In upshot an eigenvector we;N 1 ( )
is constructed by introducing a set of auxiliary variables t1 , . . . , t , = (N e)/2, such
that a candidate eigenvector we;N 1 (t| ) = we;N 1 (t1 , . . . , t |N , . . . , 1 ) turns into a
proper eigenvector, provided the parameters are turned into judiciously chosen functions
of the j s, i.e.,
we;N 1 ( ) = we;N 1 (t ( )| ),
t ( ): solution of BAE (B.19).
(4.12)
Here BAE are the Bethe ansatz equations whose solutions are (complicated) algebraic
functions t ( ), = 1, . . . , , completely symmetric in the j s. The candidate eigenvector can be chosen to be polynomial in the j and the auxiliary parameters t . A Bethe
eigenvector we ( ) solving (4.8a) will not necessarily obey (4.8b). By (B.31) we know
however that
Y
i
we;N 1 ( ),
(4.13)
fe;N 1 ( ) = e ( )
kl + i h
k>l
solves both (4.8a) and (4.8b). Here e ( ) is a completely symmetric function in j not
constrained by (4.8a,b) in any way.
In general a solution of (4.8a,b) will not satisfy (4.10). The generic expression (4.13)
however contains two pieces of information left unspecified so far. First the choice of a
specific Bethe eigenvector and second the choice of a specific symmetric function e ( ).
The obvious way to proceed is to try adjusting these two ingredients such that also the
recursive equations (4.10) are satisfied. As explained in Appendix B the choice of the
proper Bethe eigenvector amountsto a choice of the proper Bethe root. For the charge
N
independent eigenvectors with distinct eigenvalues
e sector one typically expects
(though there may be some degeneracies). We expect that typically one and only one of
them in addition satisfies the recursive equations (4.10a,b) in which case the associated
eigenvalue satisfies (4.10c). Moreover the sequential eigenvector and eigenvalue can
already be identified on the level of the Bethe roots. We call a solution of the BAE (B.19)
in N variables j a sequential -tuple of Bethe roots, if it is real for real j and satisfies

t ( ) = +i h = t (pk ), = 1, . . . , 1,
k
k+1

i h
(4.14)
t ( ) = i h = k , k = 1, . . . , N 1,
k
k+1
2
where on the right-hand side of the first equation a sequential ( 1)-tuple in N 2
variables occurs. Observe that the combination e = N 2 is invariant under N 7
N 2, 7 1; as indicated it can be identified with the conserved U (1) ' SO(2)
(N)
charge carried by fe . In other words the sequential solutions to the BAE can naturally
be arranged into sequences of fixed charge e where consecutive members are linked
by (4.14). In Appendix B we show that such sequential Bethe roots exist and that the
eigenvalues (0 | ) associated with a sequence satisfy the recursive equations (IIc). Upon
restriction to k+1 = k i h the corresponding eigenvectors should then always become
proportional to the right-hand side of (4.10a,b).
469
Having fixed the Bethe eigenvector in (4.13) to be one associated with a sequential
Bethe root, the only freedom left resides in the symmetric function e ( ). The aim now
is to adjust it such that Eqs. (4.10a,b) hold identically. Since in the N N 2 recursion
step the Bethe roots t , < , enter via (0 |pk ), it is natural to require that e ( ) is
a rational function in N , . . . , 1 and t1 , . . . , t1 , invariant under shifts j j + const.
Schematically,
e ( ) = e ( ) :
symmetric, shift invariant,

rational in N , . . . , 1 and t1 , . . . , t1 .
(4.15)
We have verified these features for N 6 4 by explicit computation and are confident that
they are generic.
In summary we arrive at the following solution procedure for the recursive functional
equations (I), (II) or their quasirational form (4.8), (4.10):
(a) For given N and given SO(2) charge e = N 2, N 4, . . . , N + 2 compute the
trial
eigenvector we (t| ) via (B.17). It contains = 12 (N e) Bethe parameters and has
N
independent components.
(b) Verify that for any Bethe root satisfying (4.14) the restrictions of fe ( ) in (4.13) to
k+1 = k i h are proportional to the right-hand sides of (4.10a,b). Multiply by a scalar
function e ( ) of the type (4.15) such that (4.10a,b) holds.
(c) When feasible compute the sequential -tuple of Bethe roots explicitly. Repeat the
procedure for N 7 N + 2.
The explicit form of the sequential Bethe roots will usually be available only for small N;
its existence is ensured by the results in Appendix B. Together (a)(c) produces a solution
to the quasirational system (4.8), (4.10). Inserting into (4.2) then yields a solution of the
original system of functional equations (I), (II). For claritys sake let us emphasize that
we have not proven step (b) to work always. However we verified it on sufficiently nontrivial examples to conjecture with some confidence that it does. A plausibility argument
of course stems from the very algebraic construction used to derive (I), (II). A T -invariant
functional is an N-independent object, once it exists at all it will automatically produce a
non-terminating sequence of functions f (N) solving (I), (II). The case e = N is excluded
in (a) because those solutions of (4.8) necessarily vanish upon restriction to k+1 = k i h .
However they will naturally serve as a starting member f (N0 ) to a sequence. The above
procedure then yields semi-infinite sequences (f (N) )N>|e| , which we expect to be basically
uniquely determined by their starting member f (|e|) . To illustrate the scheme, we have
collected the first few members of the charge e = 0, 1, 2 sequences in Appendix C.
4.2. Semi-classical limit
Let us now consider the classical limit of these solutions. Recalling from (4.2), (4.13)
Y i(kl )
,
(4.16)
fe;N 1 ( ) = c(N) ( )e ( )we;N 1 ( )
2 + h 2
k>l kl
this amounts to studying the h 0 limit of the various ingredients. For the transcendental
function ( ) in (4.4) one simply has (cf. Fig. 2)
470
( ) = i + O(h ).
(4.17)
For the scalar prefactor e ( ) the explicit results of Appendix C suggest that
e ( ) = ecl ( ) + O(h ),
(4.18)
where ecl ( ) is a ratio of homogeneous polynomials in N , . . . , 1 . However provided

e ( ) depends on the Bethe roots (which will generically be the case) the corresponding
factor will no longer be symmetric in N , . . . , 1 . This feature is related to the curious
behavior of the Bethe roots in the h 0 limit discussed in Appendix B.4. Recall for
real N , . . . , 1 the sequential Bethe roots are real-valued and completely symmetric in all
variables. For h 0 however one has

(4.19)
t ( ) = j () + O h 2 , = 1, . . . , ,
where the index j () {1, . . . , n} of the preferred variable j () is determined by the
choice of branch in the Bethe root and the relative size of the variables. Further j () 6=
j () for 6= . The Bethe vector (B.17) contains an explicit power of h = h (Ne)/2
(since the matrix operator B(t| ) is O(h )). Taking out this explicit power we define
cl
( ) = lim h we;N 1 ( ).
we;
N 1
(4.20)
h 0
cl
( ) has only one nonvanishing component
As shown in Appendix B.4, we;
N 1
cl
( ) = 0 unless (N , . . . , 1 ) = (N , . . . , 1 ),
we;
N 1
(4.21)
where (N , . . . , 1 ) {}N is a particular sign configuration of charge e. The sign pattern

(N , . . . , 1 ) of the nonvanishing component is determined by the asymptotics (4.19) of the
Bethe roots t1 , . . . , t as follows: let I = {j (1), . . . , j ()} {1, . . . , N} be the subset of
indices appearing on the right-hand side of (4.19). Then

1,
if j
/ I ,
(4.22)
j =
1, if j I .
Viewed as a scalar function of the s the component we;N 1 ( ) is a homogeneous
polynomial of degree (N 1), however again no longer a symmetric one.
cl
( ) in
Combining (4.16)(4.22) we conclude that for N > N0 the leading term fe;
N 1
the h -expansion of fe;N 1 ( ) is given by
Y 1
,
2
k>l kl

cl
( ) + O h +1 ,
fe;N 1 ( ) = h fe;
N 1
cl
cl
( ) = de(N) ( )ecl ( )we;
( )
fe;
N 1
N 1
where
(4.23)
where = (N e)/2 as before and

de(N) ( ) = ()N(N1)/2c(N)( )cl .
(4.24)
Here we assumed that ( ) in (4.6) has a regular h 0 limit cl ( ) in terms of which the
limit c(N) ( )cl = limh 0 c(N) ( ) is defined. As remarked before a fixed relative size of the
471
variables N , . . . , 1 is implicit in (4.23), say N > > 1 . The results for other orderings
then are compatible with the classical limit of the exchange relations in (I), i.e.,
cl
cl
(N , . . . , 1 ) = fe;
(N , . . . , k , k+1 , . . . , 1 ).
fe;
N 1
N k k+1 1
(4.25)
cl
( ) is separately symmetric in all j with j = 1 and all j with
In particular fe;
N 1
j = 1. As described before they are also eigenvectors of the semi-classical transfer
matrix (B.39) with eigenvalues (B.40).
It is then natural to ask whether the f cl also obey a classical counterpart of the recursive
relations (II). This is not automatic since one cannot a priori expect the operation of taking
the residue at k+1 = k i h to commute with the limit h 0. Examples where they dont
commute are the function (k+1,k ), the eigenvalues of the transfer matrix, or the Bethe
roots. Nevertheless for the final solutions it turns out that both operations do commute, up
to a numerical factor Z:

cl
( ),
(4.26)
lim h Resk+1 =k +i h fe;N 1 ( ) = Z Resk+1 =k fe;
N 1
h 0
and similarly for the appropriate residue at k+1 = k i h . The derivation of (4.26) is
deferred to Appendix B. In principle the proportionality constant could depend on the
solution considered. The explicit checks of (4.26) on the N 6 4 solutions however suggest
that it is universal and given by: Z = ()N 0 /2, 0 1.54678. Using (4.26) in (IIa,b) one
finds that both reduce to a single recurrent relation for f cl
cl
Z Resk+1 =k fe;
( )
N 1
cl
cl
(k )Ck+1 k fe;
(pk )
N k+2 k1 1
X
j 6=k+1,k
!
j
.
kj
(4.27)
The last factor on the right-hand side equals the leading term, restricted to k+1 = k and
k+1 + k = 0, in the h expansion (B.40) of the transfer matrix eigenvalues. This is the
consistency condition on (4.27) analogous to (IIc). ((IIc) itself cannot naively be expanded
in powers of h,
due to the non-commutativity of the two operations mentioned earlier.)
In summary, the solutions of the functional equations (I), (II) admit a consistent semiclassical expansion. The leading term (4.23) of this expansion has, for a given ordering of
N , . . . , 1 , only one nonvanishing component; different orderings being related by (4.25).
Further these leading terms are themselves linked by the recurrence relation (4.27). Of
course it is tempting to ask whether the leading terms have a direct interpretation in the
classical theory. This is beyond the scope of the present paper; however the vertex operator
formalism of [49,50] should be the appropriate framework to address the issue.
5. Conclusions
Motivated by the fact that none of the conventional field theoretical techniques presents
itself to develop a quantum theory for the Ernst system, we proposed to bootstrap the
quantum theory from structures linked to its classical integrability. Starting from a few
472
reasonable assumptions on the nature of the quantum counterparts of these integrable

structures, a very rigid computational framework emerged. The eventual outcome are
meromorphic functions f (N) ( ), conjectured to describe exact matrix elements in the
quantum theory, without the need for any renormalization.
On a technical level, our main result is the system of functional equations (I), (II) for the
functions f (N) ( ). One of the equations is a standard eigenvalue problem for the transfer
matrix, which is why the functions f (N) ( ) can be viewed as Bethe eigenvectors, however
very special ones. The special feature is that they are members of semi-infinite sequences in
a similar way the form factors of an integrable QFT are; consecutive members being linked
by recursive relations. The solution procedure of the functional equations for ordinary
form factors [28] ( = 2 ) and for replica deformed ones [30] ( generic) can be viewed
as selecting those special solutions of the deformed KZE enjoying such extra recursive
relations. Employing the integral representation for the latter [4042], the integrands can
be seen to obey recursive relations similar to our = 0 ones [51]. This might indicate that
the = 0 system of functional equations found here can serve as a master system, from
whose solutions the solutions of the 6= 0 systems can be obtained by a universal integral
transformation.
Physically, the most interesting finding is the spontaneous breakdown of the SL(2, R)
symmetry that is a remnant of the 4D diffeomorphism invariance in the compactified
dimensions. The matrix elements f (N) ( ) and hence the state space generated by the
physical operators supposed to underly them are covariant only under the SO(2) subgroup,
despite the fact that the algebraic framework is fully SL(2, R) covariant. Clearly in a next
step one should try to gain a better understanding of the field theoretical meaning of the
matrix elements f (N) ( ) and how physically interesting quantities can be computed in
terms of them. To this end it would be important to develop, at least to some extent,
a more conventional field theoretical formulation. A combination of perturbative and semiclassical techniques should be adequate for this purpose.
Non-perturbatively one might aim at developing a dynamical triangulations approach,
using the exact results proposed here, or quantities computed thereof, as a guideline.
In particular, it would be important to understand the statistical mechanics origin of the
dynamical breaking of the SL(2, R) symmetry, e.g., as a tug of war between energy and
entropy as in conventional spin models. The mechanism may well contain clues for the
breaking of diffeomorphism invariance beyond the symmetry reduced theory.
Acknowledgements
We wish to thank H. DeVega and D. Korotkin for valuable discussions. The work of
M.N. was supported by NSF grant 97-22097; that of H.S. by EU contract ERBFMRXCT96-0012.
473
Appendix A. Action principles for coset sigma models

As outlined in the introduction the 2D matter sector of the Ernst system is a nonlinear
sigma-model whose target space is a (noncompact) homogeneous space. Such sigmamodels are known as coset sigma models, they are classically integrable and several action
principles and parameterizations turn out to be useful. Although in the bulk of the paper
no direct use of these actions is made, both the parameterizations employed and the origin
of the (gauge) symmetries is best understood in the terms of these action principles. For
simplicity we omit the coupling to 2D gravity here, each of the coset sigma-models can be
coupled to gravity as in (1.1).
Coset sigma-models describe the dynamics of generalized harmonic maps from a
2-dimensional spacetime to a homogeneous space of the form G/H , where G is a
(simple) matrix Lie group and H a maximal subgroup of G. For the purposes of this
appendix, we take to be 2-dimensional. Minkowski space with signature (+, ). There
are two useful action principles for these coset sigma-models. A gauge theoretical one
Z

1
d2 x Tr D V V 1 D VV 1 ,
(A.1)
S[V, Q] =
2
where V is a group-valued field transforming as V Vh under an H -valued gauge
transformation, and Q is the associated connection ensuring that D V = V
VQ transforms covariantly. Clearly the gauge symmetry removes dim H degrees of
freedom leaving dim G/H physical ones. Alternatively, one can choose a non-redundant
parameterization of the coset space by matrices M G obeying a suitable quadratic
constraint M0 (M) = 1, where 0 is an involutive outer automorphism of G. The
subgroup H can then be characterized as being fixed by a related involutive automorphism
of G, given by (g) = g01 0 (g)g0 , for some fixed g0 G which likewise satisfies
g0 0 (g0 ) = 1. Explicitly, the matrices M can be constructed as M = Vg01 0 (V 1 );
they are gauge invariant and parameterize the coset space as V runs through G. Using
MM 1 = 2D VV 1 , the action (A.1) becomes
Z

1
(A.2)
d2 x Tr MM 1 MM 1 , M0 (M) = 1.
S[M] =
8
Specifically we are interested in the case G/H = SL(2, R)/SO(2) ' SU(1, 1)/U(1).
The theory with a compact target space SU(2)/U(1) can be seen to have two useful
reformulations. One as a U(1) gauge theory on the projective space PC2 =: CP1 , known
as the CP1 model. The second one parameterizes the coset matrices M in (A.2) by real
3-dimensional unit vectors and yields the S2 Heisenberg spin-model. (The latter is also
known as the O(3) nonlinear sigma-model, though O(3)/O(2) would be the proper coset
notation.) The aim in the following is to derive similar reformulations for the noncompact
SU(1, 1)/U(1) model.
We begin with the counterpart of the CP1 formulation and choose an explicit
parameterization of the SU(1, 1) matrix fields. Since V SU(1, 1) if g 3 g = 3 one
has
474

V=

h=
z1 z2
z2 z1

|z1 |2 |z2 |2 = 1,
ei 0
0 ei

,
R.
(A.3)
Here and below j , j = 1, 2, 3, are the Pauli matrices. The automorphism in the case at
hand is given by (g) = 3 g 3 , g SU(1, 1), i.e., g0 = i 3 and 0 = id in the general
framework outlined before. On the component fields acts as (z1 ) = z1 , (z2 ) = z2 .
The condition h H ' U(1) iff (h) = h then yields the above parameterization of h.
The gauge transformations V Vh simply amount to z1 z1 ei , z2 z2 ei and A
A + , where Q = iA 3 . Inserting this parameterization of V and Q into the
action (A.1) one finds
Z

z z = 1,
S[z, A] = d2 x z + iA z ( z iA z),

z1

(A.4)
, w z = w1 z1 w2 z2 .
where z = z1 , z2 , z =
z2
This is a U(1) gauge theory on the Lorentzian projective space PC1,1 , and is a noncompact analogue of the CP1 model. The SU(1, 1)/U(1) coset theory can therefore be
viewed as a matrix version of (A.4).
The second reformulation of the SU(1, 1)/U(1) theory starts from the coset action (A.2)
and yields a hyperbolic counterpart of the S2 spin-model. In upshot it simply amounts
to parameterizing the matrices M by elements of the 2-dimensional hyperboloid H2 ,
replacing the sphere S2 in the compact case. Following the general construction we first
compute the gauge invariant matrix

i(|z1 |2 + |z2 |2 )
2iz2z1
, M 2 = 1.
(A.5)
M = Vi 3 V 1 =
2iz2 z1
i(|z1 |2 + |z2 |2 )
Elements Y of the Lie algebra su(1, 1) are characterized by 3 Y = Y 3 . A convenient
basis is 0 = i 3 , 1 = 1 , 2 = 2 . For any real triplet n = (n0 , n1 , n2 ) the matrix
P j j
0 2
1 2
2 2
j n defines an U (1, 1) matrix which squares to the multiple (n ) + (n ) + (n )
of the unit matrix. The relations
X
nj j inj = z j z, j = 0, 1, 2,
(A.6)
M=
j
(where the is that of (A.4)) then provide an isomorphism

SU(1, 1)/U(1) H2 ,

2
2
2
H2 = n R1,2 n n = n0 n1 n2 = 1, n0 > 0 ,
(A.7)
where R1,2 is the ambient Lorentzian vector space of signature (+, , ). Substituting (A.6) into the coset action (A.2) one obtains
Z
1
(A.8)
d2 x n n, n n = 1,
S[n] =
4
475
where the bilinear form is that of R1,2 . An alternative way of arriving at (A.8) would
have been to vary (A.4) with respect to A and substitute the (algebraic) equation of motion
back into the action. We preferred the above route in order to have the link (A.6) to the coset
matrices M at our disposal. The action (A.8) is also convenient to verify that the energy
density of the system is positive definite
1
H = [0 n 0 n + 1 n 1 n] > 0.
2
(A.9)
To see (A.9) one differentiates the constraint and uses Schwarz inequality
!
2
X

2
2
1
0 2
j
j
n n < n1 + n2 , = 0, 1.
n = 0 2
(n )
(A.10)
j =1
It is essential here that the hyperboloid is timelike, for a spacelike hyperboloid the energy
density would be indefinite.
In the above we started with the complex SU(1, 1)/U(1) coset space because it is the
description most convenient in the bulk of the paper. The dimensional reduction of the 4D
EinsteinHilbert action however initially yields a SL(2, R)/SO(2) matter sigma-model [2].
The relation is given by the isomorphism

1
1 i
,
Ad : SL(2, R) SU(1, 1), =
2 1 i

z1 = 12 [(a + d) i(b c)],
a b
z1 z2
(A.11)
,
z2 z1
c d
z2 = 12 [(a d) i(b + c)].
For the H2 variables one gets
n0 = 12 (a 2 + b2 + c2 + d 2 ),
n1 = (ac + bd),
n2 = 12 (a 2 + b2 c2 d 2 ).
(A.12)
The automorphism for SL(2, R) is inner and is given by (g) = (g T )1 = (i 2 )1 g(i 2 ),

i.e., by (a) = d , (b) = c on the above components. It evidently selects the proper
SO(2) subgroup which is mapped onto the diagonal SU(1, 1) matrices in (A.3). On the
gauge invariant H2 variables (in either the SL(2, R) or the SU(1, 1) version) acts as
(n0 ) = n0 , (n1 ) = n1 , (n2 ) = n2 , which maps H2 onto itself and has only the trivial fixed point n = (1, 0, 0). In the coset space SL(2, R)/SO(2) one can pick representatives
given by upper triangular SL(2, R) matrices. A conventional parameterization is

1/2
B1/2
(A.13)
, 1/2 > 0, B R.
V=
0
1/2
The combination E = + iB then is the Ernst potential used in the general relativity
literature. In this parameterization the hyperbolic spins read
n0 =
2 + B 2 + 1
,
2
n1 =
B
,
n2 =
2 + B 2 1
.
2
(A.14)
476
In particular the EinsteinRosen waves have B 0, i.e., n1 0, and are thus described by a
O(1, 1) matter sigma-model. For the gauge invariant M matrix in the SL(2, R) description
one obtains

n1
n2 + n0
,
(A.15)
M = Vi 2 V 1 = VV T i 2 =
n2 n0 n1
related to the SU(1, 1) version (A.5), (A.6) by Ad, as it should. Contact to the
0
parameterization used, e.g., in (3.23) is made by lowering one index, Mab = iMab Cb0 b ,
etc.
In the case of the compact coset space SU(2)/U(1) the counterpart of the action (A.4)
yields the classical CP1 model, the counterpart of (A.8) yields the O(3) nonlinear sigmamodel. Heuristically one can then think of the n-fields as mesonic bound states of the
quark-doublets z, via nj = z j z. The analogy is correct in the sense that in the quantum
field theory only the n-field generates scattering states.
Appendix B. Diagonalization of T
Here we describe the solution of the eigenvalue problem (2.26), (2.34) for case = 1,
where only the SO(2) SL(2, R) invariance remains. The remaining abelian symmetry is
nevertheless useful in that it still allows one to decompose the full problem into pieces of
lower dimensionality.
B.1. Decomposition into charge e sectors
To this end we switch to a basis in V N diagonalizing . With the choice = (, )
in (2.31) the eigenvalues are i and we simply label the components with respect to the
new basis by the sign of the corresponding eigenvalue, i.e.,
wN 1 ( ) = aNn a11 waN a1 ( ),
j {}.
(B.1)
For the inverse matrix we use the index staggering ( 1 )a , such that in =
a ab ( 1 )b the nonvanishing components of are ++ = i = . Explicitly, one
finds for > 0

1
ch (i + sh )
i 0
. (B.2)
, with = a =
1 =
0 i
2 ch ch (i + sh )
a . Clearly, the same transformation also diagonalizes
It obeys det = i and (a ) =
the matrix () in (2.32), i.e., () = e 2 ( + ) . The charge conjugation matrix

becomes C = . For simplicity we shall refer to the basis (B.2) as the charged
basis (regarding the phase ei as associated with a U (1) charge) and to the original basis
as the real basis. We write e for the U (1) charge of wN 1 ( ), i.e., e = N + + 1 is
the number of + minus the number of in a multi-index.
In group theoretical terms Ad yields a two-parameter family of automorphisms
SL(2, R) SU(1, 1) generalizing (A.11). It suffices to verify this on the level of the Lie
i
477
algebras. Let 0 = i 3 , 1 = 1 , 2 = 2 denote the basis of su(1, 1) used before. The

matrices 1 j , j = 0, 1, 2, then are readily checked to be real and trace-free, and hence
can serve as a (non-standard) basis of sl(2, R).
Returning to the eigenvalue problem (2.26) it follows from (2.33) that
1
T (0 | )NN
1 = 0,
unless N + + 1 = N + + 1 .
(B.3)
In the charged basis (2.26) therefore decomposes into decoupled sectors of dimension

N
N
1
(B.4)
=
= me (N), := (N e).
me (N) =
2
N
Explicitly we write
Te (0 | )we ( ) = e (0 | )we ( ),
e = N, N 2, . . . , N + 2, N,
(B.5)
where we ( ) is a column vector of length me (N). For convenience we split off here a scalar
factor
T (0 | ) =
N
Y
r(j 0 i h )
T (0 | ),
j 0 2i h
(B.6)
j =1
which renders the components of T (0 | ) polynomial in the j 0 i h . Similarly the

KZE eigenvalue problem (2.38) splits up into decoupled sectors with fixed charge.
The eigenvalues qe;k ( ) in the charge e sector are phases and in addition obey
QN
kN ) = ()N i e , which follows from (2.47). Again it is convenient to split
k=1 qN (
off a scalar function from the Qk ( ) matrices and write
Y
Y
k ( ),
r(j k )Q
T (k i h | ) = h
(j k i h )Q k ( ).
(B.7)
Qk ( ) = i
j 6=k
j 6=k
The reduced eigenvalues in the charge e sector defined by

Q e;k ( )we ( ) = qe;k ( )we ( ),
i.e., qe;k ( ) =
1Y
1
e (k i h | ), (B.8)
j k i h
h
j 6=k
will then satisfy

N
Y
qe;k ( ) = i eN .
(B.9)
k=1
e;k
The T eigenvalue problem (B.5) is in fact equivalent to the seemingly weaker Q
eigenvalue problem (B.8). To see this note that T (0 | ) is a polynomial in 0 of degree

N 1, due to the tracelessness of . The second relation in (B.7) therefore amounts to
N linear equations for the N matrix-valued coefficients of this polynomial. Provided all
variables j are distinct these equations turn out to be independent, so that the T (0 | )
matrix is uniquely determined by the Q k ( ) matrices. By expanding the T eigenvalue
problem into powers of 0 one sees that also the eigenvalues (0 | ) must be polynomials
478
of degree N 1 in 0 . By the same token they will be uniquely determined by their values
at N points, i.e., by the eigenvalues qe;k ( ). Explicitly one finds
e (0 | ) = h
N
X
qe;j ( )
j =1
Y (0k + i h )(kj i h )
.
j k
(B.10)
k6=j
Finally let us note a number of useful involution properties linking the spectrum in the
charge e and the charge e sector. Since
N 1
1
T (0 | )NN
1 = T (0 | )N 1 ,
(B.11)
one infers:
If e (0 | ) Spec Te (0 | ) then e (0 | ) Spec Te (0 | ),
If qe;k ( ) Spec Qe;k ( ) then qe;k ( ) Spec Qe;k ( ).
(B.12)
In particular for charge e = 0 this means all eigenvalues come in pairs differing only by a
sign. For e 6= 0 the sign flip e 7 e amounts to 7 N so that by (B.4) one expects,
at least for generic rapidities, the eigenvalues to be in 11 correspondence. Under complex
conjugation one has [wN 1 ( )] = w1 N ( T + i h ) and a counterpart of the first
equation in (2.27). Combined with (B.11) this implies
If e (0 | ) Spec Te (0 | ) then e (0 2i h | + i h ) Spec Te (0 | ).
For the corresponding eigenvectors one can choose normalizations such that

[wN 1 ( )] = w1 N T + i h ,
(B.13)
(B.14)
which also matches the properties of the Bethe ansatz vectors (B.17) below.
B.2. Bethe ansatz equations
The aim in the following is to compute the eigenvalues and eigenvectors in (B.5)
explicitly. For small N this can be done by brute force but for generic N it is useful to
parameterize the solutions in terms of the roots of the Bethe equations. The literature on
the Bethe ansatz is enormous, some guidance can be obtained from the book [52] and, e.g.,
the following papers [5356]. Transferred to the present context the construction can be
outlined as follows: denote by
+
+
(B.15)
A := 1 aN 1 a ( 1 )+
a1 ,
N1
a cyclic vector on which the operator T from (2.25) in the charged basis acts as follows
a ab ( 1 )c Tbc (0 + i h | )B
A B
Y
i (0j + 2i h )A
.
Y
=
0
i (0j + i h )A
(B.16)
Here, Tab (0 | )B
A denotes the monodromy matrix (2.25) with the same prefactor taken out
as in (B.6). Following the Bethe ansatz procedure we generate candidate eigenstates from
479
c
by the repeated action of B(t| ) := +a ab ( 1 )
| ). The matrix operators
c Tb (t + i h
B(t| ) are commuting for different values of t and each B(t | ) lowers the SO(2) charge
of a candidate eigenstate of T by the two units. The candidate eigenstates can be made
proper eigenstates by turning the parameters t into judiciously chosen functions of the j .
In upshot one obtains eigenvectors
we ( ) =
B(t | ),
:= 12 (N e),
(B.17)
=1
with eigenvalues
Q
2i h ) (0 t + i h /2)
Q
e (0 | ) =
(0 t + 3i h /2)
Q
Q
i j (j 0 i h ) (0 t + 5i h /2)
Q
,
/2)
(0 t + 3i h
i
j (j 0
(B.18)
where the Bethe roots t are solutions of the following Bethe ansatz equations (BAE)
N
Y t t i h
Y
j t i h /2
=
,
j t + i h /2
t t + i h
= 1, . . . , .
(B.19)
6=
j =1
The only modification of the BAE as compared to the standard case = 1 is the sign on
the r.h.s. which comes from the ratio of the eigenvalues of . These equations ensure that
e (0 | ) is indeed a polynomial in 0 of degree N 1, as anticipated in Section B.1:
e (0 | ) = (1)N1 h 0N1 (N 2)
h
X
(j 3i h /2)
(1)N1 h 0N2 (N 2 1)
(B.20)
i
X
(t 3i h /2) +
+2
=:
N1
X
0 e;p ( ),
(B.21)
p=0
where the coefficients obey e;p ( ) = e;p ( + 3i h ). For e = N no Bethe roots are
required and (B.17), (B.18) should be interpreted as wN;A ( ) = A and
Y
Y
(B.22)
N (0 | ) = i (j 0 2i h ) i (j 0 i h ),
j
from which one computes qN;k = 1, k = 1, . . . , N. Since for e = N the eigenvalue

problem (B.5) is one-dimensional it follows from (B.12) that N (0 | ) = N (0 | ) and
qN;k = 1. In the expressions (B.18) for the eigenvalues coming out of the Bethe ansatz
however this is not obvious as now N Bethe roots are required, rather than none as in the
charge N sector. More generally the charge e and charge e sector enter asymmetrically
in the Bethe ansatz construction, though by (B.12) they are practically identical.
480
The eigenvalues qe;k ( ) of the asymptotic qKZE equations are accordingly given by
Y k t i h /2
, |e| 6 N 2.
(B.23)
qe;k ( ) =
qN;k = 1,
t + i h /2
k
The explicit form (B.23) allows to directly check many of the properties which we have
derived in the main text. Inspection of the BAE (B.19) shows that for real j the Bethe roots
appear in (possibly degenerate) complex conjugate pairs (t , t ). The qe;k ( ) therefore
indeed are pure phases, consistent with (2.48). Assuming the Bethe roots to be completely
symmetric functions of the j Eq. (B.23) also makes manifest the cyclic property (2.46) of
the qe;k ( ). For the product of the eigenvalues qe;k ( ) we obtain with (B.19) and (B.23):
Y t t i h
Y
= i Ne ,
qe;k ( ) = ()
(B.24)
t t + i h
6=
k
confirming (B.9) and thus (2.47).
Finally the logarithmic derivative of the BAE yields
i h
(k t i h /2)(k t + i h /2)
X
X 2i h k (t t )
i h k t
,
+
=
2
2
(j t i h /2)(j t + i h /2)
(t t ) (i h )
6=
(B.25)
which may be used to prove that

k ln qe;l ( ) = l ln qe;k ( ),
(B.26)
as claimed in (2.48) and in accordance with [41,42].

Let us also briefly comment on the solutions to the BAE. In analogy to the homogeneous
case (all j equal [56]) one expects that only the solutions with t 6= t , 6= , are relevant.
Assuming that t t 6= 0, i h the Eqs. (B.19) can be rewritten in polynomial form.
Specifically they constitute a system of polynomial equations of degree + N 1
for the unknowns t N1 (N + + 1 ), = 1, . . . , , whose coefficients are symmetric
polynomials in j = j N1 (N + + 1 ), j = 1, . . . , N. The point of adding and
subtracting the center of mass term is that the j s are boost invariant. In particular it
follows that
1
(B.27)
t (N + + 1 )
N
are boost invariant, i.e., are (completely symmetric) functions of the differences j k only.
It may be instructive to exemplify the construction for the simplest case N = 2:
e = 2 : Bethe roots:
q2;2 = q2;1 = 1,
2 (0 |2, 1 ) = h (20 + 10 3i h ),
q
h
i
2 ,
e = 0 : Bethe roots: t = 12 1 + 2 h 2 + 12
q
2 ,
0 (0 |2, 1 ) = h h 2 + 21
481
i h + 12
= q0;1(1 , 2 ),
i h 12
q
h
i
2
e = 2 : Bethe roots: t1 = 12 1 + 2 + h 2 + 12
q
h
i
2
t2 = 12 1 + 2 h 2 + 12
q0;2(2 , 1 ) = i
2 (0 |2 , 1 ) = h (20 + 10 3i h ),
q2;2 = q2;1 = 1.
(B.28)
Let us now return to the eigenvectors (B.17). Clearly any eigenvector is only determined up
to multiplication by an arbitrary scalar function, or the corresponding linear combinations
in the case with degeneracies. The Bethe eigenvectors (B.17) will in general not obey the
exchange relations

fe;n 1 ( ) = R kk+1k+1
k (k+1,k )fe;n 1 (sk ),
+ i h
i h
)+
)+
)++
, R(
, R(
,
(B.29)
R(
++ =
+ =
+ =
+ i h
+ i h
+ i h
going along with the redefined monodromy matrix T (0 | ). However, it is not difficult to
modify them so that they do. Due to the symmetry e (0 |s ) = e (0 | ), s SN , a joint
solution of (B.5), (B.29) can be obtained simply by symmetrizing with the R-matrix:
let
SN 3 s L s ( ) be the representation of the permutation group analogous to (2.35), (2.36),
The identity (2.37) remains valid with L replaced by L and
just with R replaced by R.
implies that

1 X
1
1
Ls ( )NN
fe;N 1 ( )
1 we;N 1 s ,
N!
(B.30)
sSN
is an eigenvector obeying also (B.29). Here indicates that one is still free to multiply
by a completely symmetric function in j without affecting (B.5), (B.29). Observe that
and
since we (s ) = we (t (s )|s ) = we (t ( )|s ) the operations: R-symmetrization
inserting the Bethe roots t = t ( ) commute. It is clearly convenient to first perform
the symmetrization and then insert the Bethe roots. Taking once more advantage of the
equivariance properties analogous to (2.37) one finds that the symmetrization just results
in a scalar prefactor. Explicitly, for any given Bethe eigenvector (B.17) the product
Y
i
we;N 1 ( ),
(B.31)
fe;N 1 ( )
kl + i h
k>l
solves both (B.5) and (B.29). It is manifestly polynomial in the Bethe roots and rational in
the j s.
B.3. Sequential Bethe roots
Let us examine the behavior of the Bethe ansatz equations and their solutions under
pinching k+1 k i h of the insertions N , . . . , 1 . The relations (II) imply that the
SO(2) charge of the eigenvectors is conserved under k+1 k i h , i.e.,
N N 2,
e e,
1.
(B.32)
482
This suggests that the Bethe roots describing these (special) eigenvectors might likewise
be related. Indeed the BAE (B.19) are consistent with the following N N 2 reduction
of its solutions
i h
,
2
t ( )|k+1 =k i h = t (pk ),
t ( )|k+1 =k i h = k
for < .
(B.33)
Since the Bethe roots are symmetric in all j it suffices to verify (B.33) for the k+1 =
0
= k0 + i h reduction
k + i h case. The other then formally follows from applying the k+1
0
to = (N , . . . , k+1 , k i h , k1 , . . . , 1 ). It is easy to verify that with (B.33) the
BAE (B.19) for < reduce to the BAE with N 2 insertions for the t (pk ). The
equation for = is a bit more subtle and requires to specify the limit in which the
pinched configuration is approached. Entering with the ansatz
t ( ) = k +
i h
+ /Z( ) + o(),
2
for k+1 = k + i h + ,
(B.34)
into the = BAE one obtains for N > 2

Y
Y t k 3i h /2
j k i h
i h
/Z
=
t k + i h /2
/Z
i h
j k
<
j 6=k,k+1
Y
= (1 Z)
j k i h
.
j k
j 6=k,k+1
(B.35)
This can be taken to define Z = Z( ) in (B.34), showing the consistency of the reduction
rule for t ( ) as 0.
Of course, not every solution of the BAE will satisfy (B.33), in fact the vast majority
will not. The argument shows however that under the same genericity assumptions under
which solutions exist at all, there also exists at each recursion step N 2 7 N at least one
-tuple of Bethe roots enjoying the property (B.33). In addition (B.33) is compatible with
the following reality condition
t ( ) = t ( ),
= 1, . . . , .
(B.36)
Here we refer to the observation after (B.23) that the solutions of the BAE come in
pairs (t ( ) , t ( )), where in general t ( ) 6= t ( ). We call a solution of the BAE
a sequential tuple of Bethe roots, if all roots are distinct, real in the sense of (B.36), and
satisfy (B.33). 7
To justify the terminology let us consider the behavior of the eigenvalues (0 | ) under
k+1 k + i h . From (B.33) one finds
k+1 k +i h
e (0 | ) ik0 (k0 2i h )
(0 t + 3i h /2)1
<
7 The concept appears to be new. The only article we are aware of where a pinching of inhomogeneities is
considered, is [53].
"
(j 0 2i h )
j 6=k,k+1
(0 t + i h /2)
<
(j 0 i h )
j 6=k,k+1
#
(0 t + 5i h /2)
<
k0 (k0 2i h )e (0 |pk ),
483
(B.37)
and similarly for k+1 = k i h . Hence for any -tuple of Bethe roots satisfying (B.33)
the associated eigenvalues satisfy the recursive relation (IIc) from Section 2.4
e (0 | )|k+1 =k i h = e (0 |pk ).
(B.38)
One can also easily verify (2.61), i.e., the fact that the limits 0 k i h and k+1
k + i h of (0 | ) commute.
B.4. Semi-classical limit
The semi-classical limit of the transfer matrix T (2.25) follows directly from (2.8):
!
X k

1 X k X H k
B
2
+ (i h )
+
(B.39)
+ O h 3 ,
T (0 | )A = i h
2
0k
2
0k
0k
k
with
bN
bk
b1
(k )B
A = a N a k a 1 ,
X kl (k + l )
,
Hk =
kl
l6=k
bN
bk bl
b1
(kl )B
A = a N a k a l a 1 ,
cd from (2.8). This expansion is valid either in the sense of a formal power series in
and ab
h or, with a numerical h , in the region Im 0k h , Im lk h , l 6= k, in order to prevent
a mixing of different powers of h . The absence of a term of order h 0 in (B.39) is due to
the tracelessness (2.22) of and distinguishes this case from the usual situation = 1
(see, e.g., [54,57]). The same fact implies k2 = 1 and furthermore that the operators k
and Hamiltonians Hk form a family of mutually commuting operators. (In fact the Hk can
be viewed as the Hamiltonians of an Abelian SO(2) KnizhnikZamolodchikov system.)
Simultaneous diagonalization of the k yields eigenvectors with only one nonvanishing
P
component wN 1 , (N , . . . , 1 ) {}N , in the charged basis,
j j = e. On these
eigenvectors the Hk already act diagonally. Thus the first terms in the semi-classical
expansion of the eigenvalues are given by:
!
X k
X k
X k
2 1
+ i h
(B.40)
+ O(h )3 .
(0 | ) = h
2
0k
2
0k
0k
kl
k6=l,
k
k =l
This phenomenon can also be understood in terms of the Bethe ansatz. Examination of the
explicit solutions of the Bethe roots for N = 2, 3 indicates that the symmetry in N , . . . , 1
gets lost in the limit h 0 and that they typically behave like

t ( ) = j () + (i h )2 s ( ) + O h 3 ,
(B.41)
484
for some j () {1, . . . , N} with j () 6= j () for 6= . This curious behavior is directly

linked to the seemingly innocent sign flip in the Bethe ansatz equations (B.19). Indeed,
entering with (B.41) into the BAE and matching coefficients in powers of h one finds at
O(h )
1 X
1X
1
1
.
(B.42)
s ( ) =
4
j j () 2
j () j ()
6=
j 6=j ()
Generally one can show that the Bethe roots admit a power series expansion in h (in the
region Im kl h , k 6= l) whose coefficients are uniquely determined by the assignment
j () in (B.41).
Expanding the BA expression for the eigenvalues (B.18) one obtains
!
X 1
X 1
2
+ O(h 2 ),
e (0 | ) = h
j 0
t
j
!
X 1
X 1
(B.43)
+ O(h 2 ).
= h
j 0
j 0
j I
/
j I
In the second line we inserted (B.41) and denoted by I = {j (1), . . . , j ()} {1, . . . , N}
the subset of j s appearing on the right-hand side of (B.41). Comparing now with the
/ I and j = 1 if j I . A similar
result (B.40) we conclude that j = 1 if j
computation then yields
qe;k ( ) = k + O(h ),
(B.44)
which one can also check to be consistent with (B.10).
Appendix C. Explicit solutions for N 6 4

Here we illustrate the solution procedure for the functional equations (I), (II) outlined
in Section 4 and list the first few members of the charge e = 0, 1, 2 sequences. The
eigenvectors will be given in the charged basis (B.1); we denote by fe;n 1 ( ), j {}
the Nth member of the charge e sequence in this basis. Taking advantage of the duality
described in Appendix B one can restrict attention to positive charges.
Let us begin with the charge e = 1 sector. For N = 1 one will naturally take f1; = f1;
be prescribed non-zero constants which serve as the starting member of the sequence. For
later convenience we take f1; = +, and c(1) = 1. To determine the N = 3 member we
follow the procedure (a)(c) described in Section 4. The components of the Bethe trial
vector (B.17) are
w1;++ ( ) = h u3 u2 ,
w1;++ ( ) = h u3 (u1 i h ),
w1;++ ( ) = h (u2 i h )(u1 i h ),
(C.1)
485
where uj = j t + i h /2 and t is the Bethe parameter. The ansatz (4.13) reads

f1;3 2 1 ( ) = 1 ( )
i3
w1;3 2 1 ( ).
(21 + i h )(31 + i h )(32 + i h )
For step (b) one first verifies that the consistency conditions

f1;3 2 1 ( ) = +i h ;t = +i h /2 C3 2 f1;1 ,
2
3 2
C 3 2 f1;3 2 1 ( ) = i h ;t = i h /2 f1;1 ,
3
(C.2)
(C.3)
(and a similar pair for k = 1) are obeyed. Following (4.15) the symmetric function 1 ( )
searched for to turn (4.10) into identities should be a rational, boost invariant function
of the j . Naturally one will select the one with the smallest possible numerator and
denominator degrees. Since (0 |1 ) = h this fixes
f1;3 2 1 ( ) = 2i
32 + 22 + 12 3 2 3 1 2 1 + 3h 2
w1;3 2 1 ( ).
(21 + i h )(31 + i h )(32 + i h )
(C.4)
So far only the existence of the sequential Bethe root, i.e., its defining properties (4.14)
have been used. For N = 3 one can still find it explicitly, a presentation valid for R3 is
1
1
(1 + 2 + 3 ) is 1/3
3
6

i
+ s 1/3 9h 2 + 4(1213 + 23 21 + 31 32 ) ,
6
s := 4i(13 + 23 )(12 + 32 )(21 + 31)
3

+ 9h 2 + 4(12 13 + 23 21 + 31 32 )
1/2
16(13 + 23)2 (12 + 32 )2 (21 + 31 )2
,
t ( ) =
(C.5)
where the expression under the square root is positive for all R3 . Further s 1/3 is defined
to be the cube root of s that is real for 3 = 12 (1 + 2 ) (and cyclic) and equals the positive
square root of [9h 2 + 4(1213 + 23 21 + 31 32 )]. With this choice one has

1/3 2
= 9h + 4(12 13 + 23 21 + 31 32 ) s 1/3 , for R3 ,
(C.6)
s
and t ( ) is indeed real for R3 . It is also instructive to study the branch points of this
Bethe root. They are located at the zeros of the square root in (C.5) and have no intersection
with the strip |Im ij | < h . E.g., as a function of 3 the Bethe root has 4 branch points of
order 2 such that moving 3 around two of them interchanges the two non-sequential Bethe
roots whereas the other two separate the sequential Bethe root from the non-sequential
ones. Under pinching 2 1 + i h the latter two vanish at complex infinity which is just
in agreement with the desired behavior (4.14).
Having illustrated the procedure for the charge e = 1 case we now just present the results
for the e = 2 and e = 0 series. For e = 2 we take f++ ( ) = i(21 i h ) with c(2) = 1 as
the starting member. The n = 4 member is conveniently expressed in terms of the Bethe
trial vectors, which for N = 4, e = 2 read
486
w2;+++ ( ) = hu
4 u3 u2 ,
w2;+++ ( ) = hu
4 u3 (u1 i h ),
w2;+++ ( ) = hu
4 (u2 i h )(u1 i h ),
w2;+++ ( ) = h(u
3 i h )(u2 i h )(u1 i h ),
(C.7)
with uj = j t + i h /2. The symmetric rational function 2 ( ) is conveniently described

in terms of a basis of boost invariant symmetric polynomials
2 = 4 3 + 4 2 + 4 1 + 3 2 + 3 1 + 2 1 ,
3 = 4 3 2 + 4 3 1 + 4 2 1 + 3 2 1 ,
4 = 4 3 2 1 ,
P
where j = j 14 k k . Explicitly it is given by

2 ( ) = 163 124 + 22 8h 2 2 + 7h 4 ,
and satisfies

2 ( )
4 =3 +i h
(C.8)

2
= 2(31 + 32 + i h ) 21
+ h 2 (32 + 2i h )
(31 + 2i h )(32 i h )(31 i h ).
(C.9)
The final result is

f2;4 3 2 1 ( ) = 2 ( )
Y
k>l
i
w2;4 3 2 1 ( ).
kl + i h
(C.10)
For the e = 0 sequence one has two options, it can start at N0 = 0 or at N0 = 2. Of course
already the N = 2 members will be different and accordingly two distinct sequences will
emerge. For the N0 = 0 series one naturally takes f = f = h with c(0) = 1 as the starting
member. The next member of the series is then given by
2i
h u2 ,
21 + i h
2i
h (u1 i h ).
f0;+ ( ) =
21 + i h
f0;+ ( ) =
(C.11)
Alternatively one can consider an e = 0 series starting at N0 = 2. An appropriate starting

member then is
f0;+ ( ) = i(21 i h )h u2 ,
f0;+ ( ) = i(21 i h )h (u1 i h ),
(C.12)
and we take c(2) = 1. Equivalently this amounts to having 0 (2 , 1 ) = 2 for the N0 = 0

2 + h2 for the N = 2 series.
series and 0 (2 , 1 ) = 21
0
To describe the N = 4 members of both sequences we again first note the the Bethe trial
vectors. For N = 4, e = 0 there are two Bethe parameters t1 , t2 . We set uj := j t1 +i h /2,
vj := j t2 + i h /2, in terms of which the Bethe trial vectors come out to be
487

w0;++ ( ) = h 2 v4 v3 u4 u3 u2 v1 + u1 v2 i h (u1 + v2 ) h 2 ,
w0;++ ( ) = h 2 v4 u4 u3 u2 (v2 i h )v1 + (u2 i h )(u1 i h )(v3 i h )v2
h 4 v4 u4 u3 (u1 i h ),

w0;++ ( ) = h 2 v4 (v1 i h )u4 (u1 i h ) u3 v2 + u2 v3 i h (u2 + v3 ) h 2 ,
w0;++ ( ) = h 2 (v3 i h )(v2 i h )v1 u4 u3 u2
+ h 2 (v4 i h )v3 v2 (u3 i h )(u2 i h )(u1 i h )
h 4 (v3 i h )u4 u3 (u1 i h ) h 4 v2 u4 (u2 i h )(u1 i h ),
w0;++ ( ) = h 2 (v1 i h )(u1 i h )
(v3 i h )v2 u4 u3 + (v4 i h )v3 (u3 i h )(u2 i h )
h 4 (v1 i h )(u1 i h )u4 (u2 i h ),

w0;++ ( ) = h 2 (v2 i h )(v1 i h )(u2 i h )(u1 i h )

u4 v3 + u3 v4 i h (u3 + v4 ) h 2 .
As required they are invariant under uj vj and enjoy the property (B.14).
The symmetric multiplier functions 0 ( ) are now given by the product of a symmetric
polynomial v0 ( ) and a factor u0 ( ) depending on the first Bethe root. Explicitly

(C.13)
N0 = 0 series: 0 ( ) = u0 ( ) 124 + 22 8h 2 2 + 7h 4 ,

2
3
2 2
N0 = 2 series: 0 ( ) = u0 ( ) 164 2 183 42 + 18h 2

40h 2 4 24h 4 2 + 10h 6 ,
where

2 U+ + U
1
i(U+ U ) + (U+ + U ) 4t1 (4 + 3 + 2 + 1 ) ,
u0 ( ) =
2h
h U+ U
4
Y
(j t1 i h /2).
(C.14)
with U =
j =1
u0 ( ) is completely symmetric, boost invariant and real for real s. Using

q
U+ U
2 + h2 ,
=
3
+
i
h
21
2i h
3
1
2
U+ + U 4 =3 +i h
one verifies

u0 ( )
q
2 + h2
16 21
q
q
=

,
4 =3 +i h
2 + h2 i h + + 2 + h2
3i h + 32 + 31 21
32
31
21
and further (4.10). Finally the N = 4 member of the two e = 0 series is given by
Y
i
w0;4 3 2 1 ( ),
f0;4 3 2 1 ( ) = 0 ( )
kl + i h
k>l
with 0 ( ) given in (C.13).
(C.15)
488
The semi-classical limit of these N 6 4 solutions is readily taken, and one can verify the
general pattern described in Section 4.2. Specifically let us verify the semi-classical residue
Eq. (4.27), and along the way determine the constant Z in (4.26). It is convenient to work
with the reduced functions f. So, in a first step we note the counterparts of Eqs. (4.23)
(4.27) in terms of f. One finds
Y i
cl
cl
( ) = ecl ( )we;
( )
, where
fe;
N 1
N 1
kl
k>l

cl
( ) + O h +1 .
(C.16)
fe;N 1 ( ) = h fe;
N 1
If we assume analogously to (4.26)

cl
( ) = +i h .
lim h fe;N 1 ( )|k+1 =k +i h = Z fe;
N 1
k
k+1
h 0
and similarly for k+1 = k i h , the recursive equations (4.10) turn into

cl
( ) =
Z fe;
N 1
k
k+1
!
Y
X j
cl
2
= Ck+1 k fe;
(pk )
kl
.
N k+2 k1 1
kj
l6=k+1,k
(C.17)
(C.18)
j 6=k+1,k
On the other hand f and f are related by (4.2), while f cl and fcl are related by
Y i
cl
cl
( ) = fe;
( )de(N) ( )
.
fe;
N 1
N 1
kl
(C.19)
k>l
Matching (4.26) against (C.17) one finds
Z = ()N1 0 Z,
0 1.54678.
To this end we first note

It remains to verify (C.18) and to determine Z.
i
h

,
lim e ( )|k+1 =k i h = lim e ( )
h 0
h0
k+1 =k
(C.20)
(C.21)
which for the e ( ) involving Bethe roots is note quite automatic. It follows however from
the observation that the classical limit of the t (B.41) and the pinching operation (B.33)
commute in the relevant situations: (C.21) is only relevant when the right-hand side
of (C.18) is non-vanishing; that is when k+1 6= k , and when a branch of the Bethe roots
is selected by having all but k+1 real, say. With these specifications one can choose a
labeling of the Bethe roots such that j () 6= k, k + 1 for all < . Indeed, since either
k I or k + 1 I , only one of the corresponding s appears on the right-hand side
of (B.41), which one can label to be j () . This ensures the asserted commutativity with
the pinching operation (B.33).
A similar argument can then be applied to the remainder fe ( )/e ( ). The components
of the Bethe vectors are symmetric polynomials in the Bethe roots, and after canceling
Q
common terms against the k>l 1/kl numerator, the operation to be performed on the lefthand side of (C.17) is known to have a regular limit. The result must thus be proportional
to the right-hand side. A proportionality constant different from 1 can arise as a remnant
489
of the before-mentioned cancellations. In principle, the constant Z could depend on the

solution considered. However the explicit evaluation for the N 6 4 solutions suggests the
universal value Z = 1/2.
To see this, note, e.g.,

1cl ( ) = = 2(21)2 ,
3 2
2cl ( ) = = 2(13 + 23)(32 31 21 )2 ,
4 3
cl
0 ( ) = = 4(21)3 (32 )2 , N0 = 2 series,
4 3
(C.22)
0cl ( ) = = 421(32)2 , N0 = 0 series.
4
From here one can readily verify (C.18) with Z = 1/2.

References
[1] F. Ernst, New formulation of the axially symmetric gravitational field problem, Phys. Rev. 167
(1968) 1175.
[2] R. Geroch, A method for generating solutions of Einsteins equations, J. Math. Phys. 12 (1971)
918.
[3] V. Belinskii, V. Zakharov, Integration of the Einstein equations by means of the inverse
scattering problem technique and construction of exact soliton solutions, Sov. Phys. JETP 48
(1978) 985.
[4] D. Maison, Are the stationary, axially symmetric Einstein equations completely integrable?,
Phys. Rev. Lett. 41 (1978) 521.
[5] P. Breitenlohner, D. Maison, On the Geroch group, Ann. Inst. H. Poincar. Phys. Thor. 46
(1987) 215.
[6] H. Nicolai, Two-dimensional gravities and supergravities as integrable systems, in: H. Mitter,
H. Gausterer (Eds.), Recent Aspects of Quantum Fields, Springer-Verlag, Berlin, 1991.
[7] K. Kuchar, Canonical quantization of cylindrical gravitational waves, Phys. Rev. D 4 (1971)
955.
[8] A. Ashtekar, M. Pierri, Probing quantum gravity through exactly soluble midisuperspaces, 1,
J. Math. Phys. 37 (1996) 6250.
[9] A. Ashtekar, Large quantum gravity effects: unforeseen limitations of the classical theory, Phys.
Rev. Lett. 77 (1996) 4864.
[10] D. Korotkin, H. Nicolai, Isomonodromic quantization of dimensionally reduced gravity, Nucl.
Phys. B 475 (1996) 397.
[11] G.A. Mena Marugn, Canonical quantization of the Gowdy model, Phys. Rev. D 56 (1997) 908.
[12] J. Cruz, A. Mikovic, J. Navarro-Salas, Free field realization of cylindrically symmetric Einstein
gravity, Phys. Lett. B 437 (1998) 273.
[13] D. Korotkin, H. Samtleben, Canonical quantization of cylindrical gravitational waves with two
polarizations, Phys. Rev. Lett. 80 (1998) 14.
[14] I. Klebanov, I. Kogan, A. Polyakov, Gravitational dressing of renormalization group, Phys. Rev.
Lett. 71 (1993) 3243.
[15] J. Ambjrn, K. Ghoroku, 2d quantum gravity coupled to renormalizable matter fields, Int. J.
Mod. Phys. A 9 (1994) 5689.
[16] M. Gomes, Y.K. Ha, Noncompact sigma model and dynamical mass generation, Phys.
Lett. 145B (1984) 235.
[17] Y.K. Ha, Noncompact symmetries in field theories with indefinite metric, Nucl. Phys. B 256
(1985) 687.
490
[18] T. Morozumi, S. Nojiri, An analysis of noncompact nonlinear sigma models, Prog. Theor.
Phys. 75 (1986) 677.
[19] M. Gomes, Y.K. Ha, Dynamical gauge boson in SU(N, 1)-type models, Phys. Rev. Lett. 58
(1987) 2390.
[20] S.A. Brunini, M. Gomes, A.J. da Silva, Remarks on noncompact sigma models, Phys. Rev. D 38
(1988) 706.
[21] J.W. van Holten, Quantum noncompact sigma models, J. Math. Phys. 28 (1987) 1420.
[22] J. Zinn-Justin, Quantum Field Theory and Critical Phenomena, Clarendon, Oxford, UK, 1989.
[23] F. David, Cancellations of infrared divergences in the two-dimensional nonlinear sigma models,
Commun. Math. Phys. 81 (1981) 149.
[24] S. Elitzur, The applicability of perturbation expansion to two-dimensional Goldstone systems,
Nucl. Phys. B 212 (1983) 501.
[25] B. de Wit, M.T. Grisaru, H. Nicolai, E. Rabinovici, Two loop finiteness of d = 2 supergravity,
Phys. Lett. B 286 (1992) 78.
[26] D.H. Friedan, Nonlinear models in 2 + dimensions, Ann. Phys. 163 (1985) 318.
[27] C.G. Callan, D. Friedan, E.J. Martinec, M.J. Perry, Strings in background fields, Nucl.
Phys. B 262 (1985) 593.
[28] F.A. Smirnov, Form factors in completely integrable models of quantum field theory, in:
Advanced series in Mathematical Physics, World Scientific, Singapore, 1992.
[29] M.R. Niedermaier, An algebraic approach to form-factors, Nucl. Phys. B 440 (1995) 603.
[30] M.R. Niedermaier, Form-factors, thermal states and modular structures, Nucl. Phys. B 519
(1998) 517.
[31] V. Drinfeld, Quantum groups, in: Proc. Int. Congress Math., Berkeley, 1986, AMS, Providence, RI, 1986, pp. 798820.
[32] A. LeClair, F.A. Smirnov, Infinite quantum group symmetry of fields in massive 2-D quantum
field theory, Int. J. Mod. Phys. A 7 (1992) 2997.
[33] D. Bernard, A. LeClair, The quantum double in integrable quantum field theory, Nucl.
Phys. B 399 (1993) 709.
[34] N. Reshetikhin, M. Semenov-Tian-Shansky, Central extensions of quantum current groups, Lett.
Math. Phys. 19 (1990) 133.
[35] E. Frenkel, N. Reshetikhin, Quantum affine algebras and deformation of Virasoro and Walgebras, Commun. Math. Phys. 178 (1996) 237.
[36] E. Sklyanin, On the complete integrability of the LandauLifshitz equation, preprint LOMI E3-79, Leningrad, 1979.
[37] L. Faddeev, E. Sklyanin, L. Takhtajan, Quantum inverse scattering method, Theor. Math.
Phys. 40 (1979) 194.
[38] V. Drinfeld, Hopf algebras and the quantum YangBaxter equation, Sov. Math. Dokl. 32 (1985)
254.
[39] H. Konno, Free field representation of level Yangian double DY(sl(2))() and deformation of
Wakimoto modules, Lett. Math. Phys. 40 (1997) 321.
[40] I. Frenkel, N. Reshetikhin, Quantum affine algebras and holonomic difference equations,
Commun. Math. Phys. 146 (1992) 1.
[41] V. Tarasov, A. Varchenko, Asymptotic solutions to the quantized KnizhnikZamolodchikov
equation and Bethe vectors, in: Mathematics in St. Petersburg, AMS, 1996, pp. 235273.
[42] V. Tarasov, A. Varchenko, Geometry of q-hypergeometric functions as a bridge between
Yangians and quantum affine algebras, Inventiones Mathematicae 128 (1997) 501.
[43] V. Chari, A. Pressley, A Guide to Quantum Groups, Cambridge University Press, Cambridge,
1994.
[44] E. Frenkel, E. Mukhin, Combinatorics of q-characters of finite-dimensional representations of
quantum affine algebras, preprint math. QA/9911112.
491
[45] D. Korotkin, H. Samtleben, Yangian symmetry in integrable quantum gravity, Nucl. Phys. B 527
(1998) 657.
[46] V. Fock, A. Rosly, Poisson structures on moduli of flat connections on Riemann surfaces and
r-matrices, preprint ITEP 72-92, Moscow, 1992.
[47] N.D. Mermin, H. Wagner, Absence of ferromagnetism or antiferromagnetism in onedimensional or two-dimensional isotropic Heisenberg models, Phys. Rev. Lett. 17 (1966) 1133.
[48] S. Coleman, There are no Goldstone bosons in two dimensions, Commun. Math. Phys. 31
(1973) 259.
[49] D. Bernard, B. Julia, Twisted self-duality of dimensionally reduced gravity and vertex operators,
Nucl. Phys. B 547 (1999) 427.
[50] D. Bernard, N. Regnault, Vertex operator solutions of 2d dimensionally reduced gravity, preprint
SPhT-99-017, solv-int/9902017.
[51] M. Pillin, Replica deformation of the su(2) invariant Thirring model via solutions of the qKZ
equation, preprint KCL-MTH-99-28, hep-th/9907147.
[52] V.E. Korepin, N.M. Bogoliubov, A.G. Izergin, Quantum Inverse Scattering Method and
Correlation Functions, Cambridge University Press, Cambridge, 1993.
[53] A.N. Kirillov, N.Y. Reshetikhin, The Yangians, Bethe ansatz and combinatorics, Lett. Math.
Phys. 12 (1986) 199.
[54] H. Babujian, Off-shell Bethe ansatz equations and N -point correlators in the SU(2) WZNW
theory, J. Phys. A 26 (1993) 6981.
[55] J.M. Maillet, J.S. de Santos, Drinfeld twists and algebraic Bethe ansatz, Comm. Math.
Phys. 210 (2000) 177.
[56] L.D. Faddeev, Algebraic aspects of the Bethe ansatz, Int. J. Mod. Phys. 10 (1995) 1845.
[57] B. Feigin, E. Frenkel, N. Reshetikhin, Gaudin model, Bethe ansatz and critical level, Commun.
Math. Phys. 166 (1994) 27.

Two-loop YangMills theory in the world-line

formalism and an EulerHeisenberg type action
Haru-Tada Sato 1,2 , Michael G. Schmidt 3 , Claus Zahlten 4
Institut fr Theoretische Physik, Universitt Heidelberg, Philosophenweg 16, D-69120 Heidelberg, Germany
Received 14 March 2000; accepted 5 April 2000
Abstract
Within the framework of the world-line formalism we write down in detail a two-loop Euler
Heisenberg type action for gluon loops in YangMills theory and discuss its divergence structure.
We exactly perform all the world-line moduli integrals at two loops by inserting a mass parameter,
and then extract divergent coefficients to be renormalized. 2000 Elsevier Science B.V. All rights
reserved.
PACS: 11.15.Bt; 11.55.-m; 11.90.+t
Keywords: World-line formalism; BernKosower rules; YangMills theory; EulerHeisenberg action; Two-loop
integrals
1. Introduction
The BernKosower method is described as a set of simple rules to obtain gluon scattering
amplitudes at one loop, and it is known to improve the computational efficiency over the
current Feynman diagram technique [1]. Those rules are derived from a string theory in the
limit where the inverse string tension vanishes, as a consequence of the idea that a string
world-sheet degenerates into a desired particle diagram at a singular point on the boundary
of moduli space [25]. The integration over moduli space naturally covers all necessary
Feynman diagrams appearing in field theory, and we hence have a compact master formula
for particle scattering amplitudes. Thus the most conspicuous point in this formalism is
that the diagram summation is already finished in the formula without introducing the loop
integral and the Dirac trace for a given scattering [6,7]. This idea is also applied to graviton
scattering [8,9].
1 sato@thphys.uni-heidelberg.de
2 Present address: Theory group, KEK (Tanashi), Midori-machi 3-2-1, 188-8501 Tokyo, Japan.
3 m.g.schmidt@thphys.uni-heidelberg.de
4 zahlten@thphys.uni-heidelberg.de
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 2 2 0 - 0
H.-T. Sato et al. / Nuclear Physics B 579 (2000) 492524
493
The discovery of the BernKosower rules has also stimulated investigations for a new
mathematical structure of quantum field theory; how to reflect the string-like structure into
field theory as such. The first rederivation of the BernKosower rules was accomplished
by Strassler in Ref. [10], where the background field method, the proper time method and
the path integral method for a first quantized (0 + 1)-dimensional field theory (world-line
formulation) are well combined [11]. There are many other fruitful examples along this
stream [12,13] (see Section 1 of Ref. [14] for updated references), and these examples are
the strong incentives to study the world-line formalism from the theoretical point of view,
especially from the viewpoint of its higher loop extensions [1517]. The present paper also
discusses two-loop YangMills theory in the world-line formalism with the aim to develop
techniques for higher loops.
This paper is a continuation of the previous work [14], where the effective action of
YangMills theory at the two-loop order is derived based on the world-line formalism;
also developed there is a certain technique, which generates multiloop generalizations of
the one-loop trace-log (determinant) formula. However, these arguments are still inside
the shell of formal arguments, since we are left with the problem of how to deal with the
multi-integrals of world-line moduli parameters. The moduli integrals of a higher genus
world-sheet are too complicated to perform, while one might naturally expect that this
situation would be improved in the field theory limit. Although we, of course, have an
option to computerize these complicated integrals, there are still difficulties, for example
in three loop QED integrals [18].
It is certainly valuable to analyze two-loop integrals of YangMills theory in the worldline formalism in particular if many outer particles or an EulerHeisenberg type constant
field are involved. It might also hint to the world-line moduli integrations at higher loop
orders.
In this paper, after a detailed derivation of the two-loop EulerHeisenberg action with
gluon loops in a pseudo-abelian gauge field background, we shall present some technical
issues of how to deal with the world-line moduli integrals in the gluon effective action
at the second order of the Taylor expansion in terms of external background fields. This
analysis is also essential to examine the divergence structure related to a wave function
and gauge fixing parameter renormalization. We only discuss the gluon loop part, since the
ghost loop part is rather simple and can be dealt with in the same way as the gluon loop
case.
When we perform the integrals, we insert a mass parameter in order to regularize
divergences. Generally speaking, massive propagators in the Feynman rule method are
difficult to integrate in an analytic way. Contrastingly in our formalism, we shall go through
the entire procedure analytically, and all the results will be written in hypergeometric
functions. This is certainly an intriguing point of this paper.
This paper is organized as follows. In Section 2, we write down our starting formulae
for the gluon loop effective action at two loops. We slightly modify a few notations
from the previous presentation [14] through the path and proper time inversions presented
in Appendix A. In Section 3, a derivation of the one-loop -function coefficient serves
as an example for how calculations in our formalism can be simplified by specializing
494
to the pseudo-abelian case. In Section 4, we apply the pseudo-abelian technique to the

calculation of the gluon effective action presented in Section 2. Here, we only perform
the world-line path integral parts. This section is a completion of the parts outlined in the
previous paper [14], and the details of the computation are contained in Appendix B. In
Section 5, we further study how to integrate the world-line moduli integral parts, which
are the final integrations to obtain a fully integrated form of EulerHeisenberg type action.
For simplicity, we only consider the gluon kinetic term, through the Taylor expansions
concerning the external field strength. The Taylor coefficients are the functions of worldline moduli parameters, and we show that these coefficients can be integrated; the details
are in Appendices C and D. Appendix E is the Feynman diagram analysis to be compared
with our results.
2. Two-loop effective action

Let us first review in brief the world-line representation of the two-loop effective action
in YangMills theory [14]. In this paper we only discuss the gluon loop part. It is given by
[A] = I1 [A] + I2 [A],
(2.1)
where (I1 = 1 + 2 , I2 = 3(2) in the previous paper)

1
I1 [A] =
8
ZS
dS
Z
d
w(T3Z)=x( )
[Dw]T3
[Dx]S
dT3
w(0)=x(0)

ae
(w (0) x (0))w (T3 )W[]
[x; S, , 0]

ae
+ x (0)w (T3 )W[] [x; S, , 0] Wea [w; T3 , 0],
(2.2)
and
1
I2 [A] =
4
I
[Dx1 ]T1
dT1 dT2
[Dx2 ]T2 4 x1 (0) x2 (0)

TrC a W [x1; T1 , 0] TrC a W [x2 ; T2, 0] ,
with the following compact notations:
Z
Z
RT 2
[Dx]T F [x] = Dx e1/4 0 x d F [x],
(ZT3
Wea [w; T3 , 0] = P exp
for any functional F [x],
d M [w]
(2.5)
(ZS
"
(2.4)
)ea
ae
[x; S, , 0] = TrC
W
(2.3)
)
d M [x]
P exp
(Z
) #
P exp
d M [x]
, (2.6)
e
495
Fig. 1. (a) The loop type parametrization. (b) The symmetric parametrization.
and
M [x]
ab

1

c
0
c
0
0
= 2i F x( ) A x( ) 0 x ( )
(c )ab .
2
0 =
(2.7)
Here we have slightly changed the notations used in the previous paper [14]; (i) the
previous definition of M is associated with D = iA, while the present M is associated
with D = + iA, (ii) the path ordering directions are modified to be the standard one, and
some related formulae are listed in Appendix A. As in the previous paper, we always use
Euclidean spacetime conventions.
It is convenient to have another representation for I1 [A], based on the symmetric
parametrization in Fig. 1(b), which treats the individual gluon lines in a more equal way,
thus allowing a greater class of transformations by inverting and relabelling the gluon paths
and leading to significant simplifications in the concrete calculations. The expression (2.2)
is based on the loop type parametrization in Fig. 1(a). The transformation rules between (a)
and (b) are known [1517,19], and we thus have the following symmetric representation
for I1 [A]:
" 3 xk (TZk )=y2
#
Z
Z
Z
Y
1
D
D
dT1 dT2 dT3 d y1 d y2
[Dxk ]Tk
I1 [A] =
8
h
k=1 x (0)=y
k
1

1
ae
e[]
x2 , x1 ; T2 , 0, T1 , 0
x3 (0) x1 (0) x3 (T3 )W
i ea
1
e ae
+ x 1 (0)x3 (T3 )W
[] x2 , x1 ; T2 , 0, T1 , 0 W [x3 ; T3 , 0],
(2.8)
where
T 1 = ,
and
T 2 = S ,

1
ae
e
x2 , x1 ; T2 , 0, T1 , 0
W
(ZT2
"
(2.9)

d M x21
= TrC P exp
a
(ZT1
) #
P exp
d M [x1 ]
.
e
(2.10)
Note that the transition from loop type to symmetric parametrization requires both splitting
the loop path into two parts and inverting one of them (which we denote by x21 ) to
496
achieve all three paths to start at y1 and to end at y2 . In contrast to the naive expectation,
this suggests that in a general background, one may not just write down the product of
three propagators starting and ending at identical points to represent a loop with inserted
propagator. See Appendix A for a detailed definition of the notation x21 and some
comments on path inversion.
After finishing the next section, we shall discuss the world-line path integrals for xk in
Section 4, and the integrals of the world-line moduli parts (proper times) Tk in Section 5.
3. The pseudo-abelian case

For the rest of this paper, we confine ourselves to the pseudo-abelian su(2) with constant
field strength. Thus we assume
Aa (x) = A (x)na ,
with na na = 1 and constant n,
(3.1)
i.e., the color dependence of the non-abelian gauge fields is supposed to be factored out
in form of a constant unit vector in color space. Within these settings calculations are
simplified considerably, though non-abelian results can still be reproduced as shall be seen
below. As an example and to introduce some notations, we here show a brief sketch of how
our formalism works within the calculation of one-loop -function coefficients.
The assumption (3.1) leads to similar decompositions for the field strength
a
(x) = F (x)na
F
with F (x) = A (x) A (x)
(3.2)
and the matrix M [x]

M [x] = M [x] (nc c ) M [x] T ,
(3.3)
where M [x] is the Lorentz matrix defined by

1
M [x] = 2i F (x) A (x)x 1L
2
(3.4)
and 1L denotes the unit Lorentz matrix (in the Euclidean space).
So far we have not used our additional assumption of a constant field strength, nor have
we fixed the gauge for the external gauge fields Aa . If we take into account the constancy
of the field strength, we may choose
1
(3.5)
A (x) = x F ,
2
henceforth expecting M [x] to be of the form

1
(3.6)
M [x] = 2i F x F x 1L ,
4
rather than (3.4). In addition, it is convenient to define the integrated matrices (omitting the
index ), and we can simply write
ZT
ZT
M [x] d = M[x] T ,
with M[x] =
M [x] d.
0
(3.7)
497
It is the benefit of confining ourselves to the pseudo-abelian case, that the M [x]
matrices for different values of the parameter become commuting quantities

(3.8)
M [x], M 0 [x] = 0.
Thus we are allowed to drop path ordering from all of our expressions. This leads to the
following decomposition
P exp{M T } = 1L I + sinh{M} T + cosh{M} T+ ,
(3.9)
in terms of the su(2) matrices

T = nc c ,
T+ = (T )2 ,
I = 1C T+ ,
where 1C = diag(1, 1, 1).
(3.10)
Now, using the properties

TrC T = 0,
CA TrC T+ = 2,
TrC I = 1
(3.11)
the one-loop effective action for the gluon loop is calculated as follows [20]:
1
1-loop
[A] =
G
1
2
0
Z
dT m2 T
e
T
!aa
ZT
M [x] d
[Dx]T P exp
dT m2 T
e
T
I
[Dx]T D + CA TrL (cosh M)
1
=
4
dT m2 T
e
T
[Dx]T D + CA TrL eM
+ (F F ), (3.12)
where we have introduced the gluon mass term em T for regularization. The second
contribution with F replaced by F counts for a factor of two, thus with the one-loop
path integral normalization
)
(
Z

ZT
I
2

1
1/2 sin F T
D/2
dD x0 (3.13)
d x + 2ixF x = (4T )
detL
Dx exp
4
FT
2
we are led to
1-loop
[A] =
G
Z
1
2
dT T 1D/2 em T
2(4)D/2
0
Z

1/2 sin F T
dD x0 .
D + CA TrL e2i F T detL
FT
(3.14)
For now we are interested in the two-point function only, i.e., in the second functional
derivative of the effective action with respect to Aa . To this end, we only need the second
order term of an expansion of (3.14) in terms of F . Omitting constant and higher order
terms we find
498
1-loop
G
[A] =
Z

Z
CA
D
1D/2 m2 T
2
dT T
e
dD x0 TrL F 2 + .
2(4)D/2 12
0
(3.15)
Setting D = 4 2 and performing the T integration leads to the following pole structure
in :
Z

g 2 CA 10
1
1-loop
[A] = 0 2
dD x0 F F + O 0 + , (3.16)
G
(4)
3
4
where we have revived the gauge coupling g and where g0 is the dimensionless coupling
constant defined by g = g0 . Finally we calculate the functional derivative and transform
into momentum space: using

Z

1
(3.17)
dD x0 F F ab k 2 k k
4
we read off
G ab
=

g02 CA ab 10
k 2 k k + O 0 .
2
(4)
3
(3.18)
Similarly, the one-loop contribution from a ghost loop can be calculated: as can be
deduced from Ref. [14], the ghost one-loop action is given by changing the overall
normalization in (3.12) from 1/2 to 1, and only employing the Lorentz scalar term
e instead of using M . The corresponding pseudoin (2.7); i.e., define iAc x c M
f is defined analogically to the gluon loop case (q.v., Eqs. (3.6) and
abelian quantity M
(3.7)). Thus we find the one-loop ghost action
1-loop
F P [A] =
dT m2 T
e
T
ZT
[Dx]T P exp
e [x] d
M
Z
=
!aa
dT m2 T
e
T

f
[Dx]T 1 + CA cosh M
1
=
2
dT m2 T
e
T
f
[Dx]T 1 + CA eM + (F F ).
(3.19)
Again taking into account the F term by a factor of two and using (3.13), we arrive at
1-loop
F P [A] =
1
(4)D/2
Z
dT T
0
1D/2 m2 T
Z

1/2 sin F T
dD x0 .
1 + CA detL
FT
(3.20)
Expanding this expression in the same way as done in (3.15), we derive
1-loop
F P [A] = +
CA
12(4)D/2
Z
dT T
1D/2 m2 T
499
dD x0 TrL F 2 +
(3.21)
Z

g02 CA 1
1
D
d x0 F F + O 0 + ,
=
2
(4) 3
4
and hence
F P ab
=

g02 CA ab 1
k 2 k k + O 0 .
2
(4)
3
(3.22)
(3.23)
Gathering Eqs. (3.18) and (3.23), the correct (one-loop) -function coefficient 11/3 is
reproduced.
4. Two-loop EulerHeisenberg formulas

Now, let us consider the extension of the above calculations to the two-loop case.
We deal with the symmetric representations (2.3) and (2.8). In the two-loop case, as
understood from (2.7), we have to keep in mind that the sign of the xF x term changes
due to the derivative, if we invert the path. For example in (2.10), one should notice (see
also Appendix A) that

1

ab
1 0
c
( )
(c )ab
x21 ( 0 ) Ac x21 ( 0 ) 0 x2
M x21 = 2i F
2
0

=
1

c
0
c
0
0
= 2i F x2 ( ) + A x2 ( ) 0 x2 ( )
(c )ab
2
0
=T2
ba
= MT2 [x2 ] .
(4.1)
Reflecting this fact, it is rather convenient to introduce the signature index (= 1, 0) on
the Lorentz matrix M:
def
M()
k =
ZTk

1
d 2i ||F xk F x k 1L ,
4
(4.2)
where the k stands for the line labels 1, 2 and 3. With this matrix notation (4.2), the general
form for the color matrix part of the action (2.8) is written in the form

eae x 1 , x1 ; T2, 0, T1 , 0 Wea [x3 ; T3, 0]
W
2
h

(+)
(+)
e
i
ea
(4.3)
= TrC a exp M()
2 T exp M1 T exp M3 T .
After using the expansion (cf. Eq. (3.9))
()
()

()
exp Mk T = 1L I + sinh Mk T + cosh Mk T+ ,
we perform the color traces applying the following formulae:
(4.4)
500

TrC a T e T I ea = 2,

TrC a T e I Tea = 2,

TrC a Ie T Tea = 2,
(4.5)
where the 3rd formula follows from the 2nd one with the properties
(T )T = T ,
I T = I.
(4.6)
For any other combinations of A, B and C chosen out of {I, T , T+ }, the following formula
applies
TrC (a Ae B)C ea = 0.
(4.7)
Thus the quantity (4.3) is calculated as follows:

eae x 1 , x1 ; T2, 0, T1 , 0 Wea [x3 ; T3, 0]
W
2

(+)
()
(+)
()
= 2 cosh M2 cosh M3 + sinh M2 sinh M3

(+)
(+)
(+)
(+)
+ 2 cosh M1 cosh M3 sinh M1 sinh M3

()
(+)
()
(+)
+ 2 cosh M1 cosh M2 + sinh M1 sinh M2
()
(+)
(0)
= exp M1 exp M2 exp M3

(+)
+ exp M(+)
exp M(0)
1
2 exp M3
(+)
()
(0)
+ exp M1 exp M2 exp M3 + (F F ).
(4.8)
(4.9)
Now, as sketched in Ref. [14], performing the trivial integral (the 1st term) in Eq. (4.2)
(ZTk
)

1 2
()
d xk
exp Mk
exp
4
0
1
= exp{2i||Tk F } exp
4
ZTk
d
x k2
+ 2i()xk F xk
)
,
(4.10)
and introducing the quantities

1X
=
4
3
(1 ,2 ,3 )
ZTk

d x k2 + 2ik xk F xk ,
a = 1, 0,
k=1 0
we have the following formula:

)
(
3 ZTk

1X
2 e ae
d xk W x21 , x1 ; T2, 0, T1 , 0 Wea [x3 ; T3, 0]
exp
4
k=1 0
= exp{2iT2 F } exp{2iT3F } eS
(0,,+)
(4.11)
+ exp{2iT1 F } exp{2iT3F } eS
(+,0,)
+ exp{2iT1 F } exp{2iT2F } eS
(+,,0)
501
+ (F F ).
(4.12)
It is worth noticing here that the interaction terms xk F xk defined on the three different lines
possess different values; this fact is related to the su(2) structure abc . In the following,
we calculate I1 and I2 separately, since these two quantities involve different world-line
topology.
4.1. The I1 [A] part
Applying the formula (4.12) to Eq. (2.8), we obtain the following expressions (for the
convenience of presentation, we split I1 [A] into two quantities depending on whether x1 x3
or x3 x3 correlations):
I1 [A] = 1 [A] + 2 [A],
(4.13)
where
1
1 [A] =
8
h
"
Z
D
d y1
dT1 dT2 dT3

0
d y2
3
Y
xk (T
Zk )=y2
k=1 x (0)=y
k
1

i S (0,,+)
e
e2iT2 F e2iT3 F 1L TrL e2iT3 F
h
+ e2iT2 F e2iT1 F 1L TrL e2iT1 F

h
+ e
2i(T1 +T3 )F
1L TrL e
i
2i(T1 +T3 )F
eS
i
(+,,0)
S (+,0,)
+ (F F ),
(4.14)
2 [A] =
" 3
Z
Z
Z
Y
1
D
D
dT1 dT2 dT3 d y1 d y2
h
Dxk x3 (0)x3 (T3 )
xk (T
Zk )=y2
Dxk x1 (0)x3 (T3 )
k=1 x (0)=y
k
1

(0,,+)
2i sin 2F (T2 + T3 ) e2i(T2 T3 )F + e2iT2 F TrL e2iT3 F eS

(+,,0)
+ 2i sin 2F (T1 + T2 ) e2i(T2 T1 )F + e2iT2 F TrL e2iT1 F eS
i

(+,0,)
+ 2 cos 2F (T1 + T3 ) + e2i(T3 T1 )F + 1L TrL e2i(T1 +T3 )F eS
+ (F F ).
(4.15)
Then we perform the path integrals of the form

xa ( )xb ( 0 ) ( , , )
1 2 3
" 3 xk (TZk )=y2
#
Z
Z
Y
(1 ,2 ,3 )
= dD y1 dD y2
Dxk xa ( )xb ( 0 )eS
,
k=1 x (0)=y
k
1
(4.16)
502
and this yields

xa ( )xb ( 0 ) (
1 ,2 ,3 )
ab
= N (1 ,2 ,3 ) 0 G
(, 0 ; 1, 2 , 3 ),
where
N
(1 ,2 ,3 )
= (4)
1/2
detL
3
X
!
l F cot l F Tl
l=1
3
Y
D/2
Tk
(4.17)
1/2
detL
k=1
sin k F Tk
k F Tk
Z
dD x0 ,
(4.18)
and 5
ab
G
(, 0 ; 1 , 2 , 3 )
= ab Ga (, 0 ) + 2

"
3
X
#1 !
k F cot(k F Tk )
k=1
e2ia F
1
1
e2ia F Ta 1 2

0
e2ib F
1 1
e2ib F Tb 1 2
(4.19)
with
Ga (, 0 )

( 0 )2
a
0
0
G
(,
)
=
B
Ta

=
1
F
1
ia F Ta GaB (, 0 )
a
0
e
+
i
F
G
(,
,
a
B
Ta
2F 2 sin(F Ta )
a = 0,
a 6= 0.
(4.20)
Inserting each value of (4.17) at (, 0 ) = (0, T3 ) into Eqs. (4.14) and (4.15), we therefore
obtain (the details are presented in Appendix B)
1
I1 [A] = (4)D
2

TrL

1/2
dT1 dT2 dT3 detL

0
F2
F
F 2 T3
2 sin F T1 cos 2F (T1 + 2T2 )
F sin F T2
2 sin F (T1 + T2 ) cos F (2T1 + 3T2 )
+ {1 2 cos 2F (T1 + T2 )} sin F T2 cos F (T1 T2 )
5 In Eq. (4.19), one may replace Ga (, 0 ) Ga (, 0 ) Ga (, 0) Ga (0, 0 ) as seen in [14], however
ab .
our final results do not change because we only need the derivatives 0 G
503
F
4 sin F T1 sin F T2 sin 2F (T1 + T2 ) 2 sin F T1 cos F (2T1 + 3T2 )
F

2 sin F T2 cos F (T1 2T2 ) sin F (T1 + T2 ) cos 2F (T1 T2 )
+
F 2 T3
sin F (T1 + T2 ) cos F (2T1 + T2 )
F sin F T2

sin F T1 cos 2F (T1 + T2 )
+ TrL

F
3 sin F T1 cos F (2T1 + T2 ) + cos 2F T1 sin F (T1 + T2 )
F
TrL (cos 2F T2 )

F 2 T3
sin F T2 cos F (T1 T2 ) + cos F T2 sin F (T1 + T2 )
+ TrL
F sin F T2

F
sin F T1 cos F T2
sin F T1 cos 2F T2 +
F

TrL cos 2F (T1 + T2 )

+ (T2 )2(1 D)TrL (cos 2F T1 ) + (T3 )TrL cos 2F (T1 T2 )
Z
dD x0 ,
(4.21)
(T3 )TrL (cos 2F T1 ) TrL (cos 2F T2 )
+
where
F = sin F T1 sin F T2 + F T3 sin F (T1 + T2 ).
(4.22)
4.2. The I2 [A] part

The computation of the other quantity I2 [A] is similar to the above calculations, however
the topology of the world-line diagram is different in this case. Let us start with the
following expression. First, Eq. (2.3) with (3.7) inserted becomes
1
I2 [A] =
4
I
dT1 dT2
I
[Dx1 ]T1
[Dx2 ]T2 x1 (0) x2 (0)

(+)

a

TrC a exp M(+)
1 T TrC exp M2 T .
(4.23)
With the expansion (4.4) and the properties

TrC a T+ = TrC a I = 0,

TrC a T = nb TrC a b = nb 2 ab = 2na ,
(4.25)
we have the formula

()
()

TrC a exp Mk T = 2na sinh Mk .
(4.26)
Remembering the relation na na = 1, we then derive from (4.23)
(4.24)
504
Z
I2 [A] =
I
[Dx1 ]T1
dT1 dT2
[Dx2 ]T2

(+)
(+)
x1 (0) x2 (0) TrL sinh M1 sinh M2
Z

1
dT1 dT2 N (+,+) TrL e2i(T1 +T2 )F N (+,) TrL e2i(T1 T2 )F
=
4
0
+ (F F ),
(4.27)
where
Z
N (1 ,2 ) =
"
Z
dD y1
2
Y
dD y2
xk (T
Zk )=y2
Dxk D (y1 y2 )eS(1 ,2 ) ,
(4.28)
k=1 x (0)=y
k
1
with
1X
=
4
2
(1 ,2 )
ZTk

d x k2 + 2ik xk F xk .
(4.29)
k=1 0
The normalizations N (1 ,2 ) satisfy the following properties:

N (+,+) = N (+,) = N (,+) = N (,) ,
(4.30)
since, in the present case, the inversions of paths (i.e., the changes of s signs) do not
change the value of the path integral (4.28): note that all the initial and ending points of
two closed loops are identical. The (F F ) terms in (4.27) lead again to a factor of
two, and thus we have
1
I2 [A] =
2

dT1 dT2 N (+,) TrL e2i(T1 +T2 )F e2i(T1 T2 )F .
(4.31)
The quantity (4.28) can be evaluated as follows. Recalling the relation

D/2

1
1
exp
(y1 y2 )2 ,
D (y1 y2 ) = lim
T3 0 4T3
4T3
(4.32)
we convert the -function to the following path integral form:

"
x3 (T
Z3 )=y2
1
Dx3 exp
4
(y1 y2 ) = lim
D
T3 0
x32 ( ) d
and this leads to the relations

N
= lim N
T3 0
(1 ,2 ,0)
= (4)
We therefore have the expression
(4.33)
x3 (0)=y1
(1 ,2 )
ZT3
1/2
detL
sin F T1 sin F T2
F2
Z
dD x0 .
(4.34)
1
I2 [A] = (4)D
2
1/2
dT1 dT2 detL
sin F T1 sin F T2
F2
505
TrL cos 2F (T1 + T2 ) TrL cos 2F (T1 T2 )
Z
dD x0 .
(4.35)
It is interesting that the normalization N (1 ,2 ) can still be obtained as a singular limit (the
T3 0 limit) from Eq. (4.18), although I2 is not a singular part of I1 (cf. (4.21)) [14].
5. The world-line moduli integrals
In this section, we study the divergence structure of the EulerHeisenberg-type action
derived in the previous sections. In the world-line formalism, divergences come out
explicitly after performing the proper time integrals (the Ta integrals; a = 1, 2, 3 in the
present case), and hence we have to examine how to perform these integrals before
discussing renormalizations. However, generally speaking, multi-integrations are difficult
to perform in an analytic way, and hence we here consider the simplest case corresponding
to gluon two-point function parts.
Let us consider the Taylor expansions of [A] concerning F (omitting constant terms)
in the same way as done in the one-loop case. Expanding I1 and I2 up to the second order
of F , we extract the following quantities from Eqs. (4.21) and (4.35):

Z
g04 4
93
D
d x0 F F 30C1 + 42C2 + 48C3 + C4 2C5 10C6
I1 [A] =
2
(4)42

57
17
17C1 + 29C2 + C3 + 32C4 C6
2
3

2
(5.1)
2C1 + 2C2 + 5C3 + 6C4 C6 2 + ,
3
Z
g 4 4
(5.2)
I2 [A] = 0 42 dD x0 F F (4)C5 + ,
(4)
with
Z
2
C1 = dT1 dT2 dT3 4 T14 T2 em (T1 +T2 +T3 ) ,
0
Z
C2 =
0
Z
C3 =
dT1 dT2 dT3 4 T13 T22 em
2 (T +T +T )
1
2
3
dT1 dT2 dT3 4 T13 T2 T3 em
2 (T +T +T )
1
2
3
Z
C4 =
0
dT1 dT2 dT3 4 T12 T22 T3 em
2 (T +T +T )
1
2
3
506
Z
C5 =
0
Z
C6 =
dT1 dT2 dT3 4 T13 T23 (T3 )em
2 (T +T +T )
1
2
3
dT1 dT2 dT3 4 T14 T22 (T3 )em
2 (T +T +T )
1
2
3
(5.3)
where we have put D = 4 2 and = T1 T2 + T2 T3 + T3 T1 . Also, the damping mass

2
factor em (T1 +T2 +T3 ) is inserted in each Ci in the same way as in the one-loop calculation
(Section 3). In the meantime, we shall introduce the notation 0 = {; > 1} for C1 and C6 ,
in order not to mix it up with the (usual) infinitesimal parameter > 0. This description
is indispensable for the convergency of the C1 and C6 integrals, although we shall set
0 = after all, expecting the analytic continuation (see also Section 8.1.2 in [21]). The
0 divergences are related to the divergences from the artificial mass term insertion, since
C1 and C6 contain tadpole contributions, which vanish in the m 0 limit (in the sense of
dimensional regularization).
In C5 and C6 , all the integrals are easy to perform, and hence we simply write down
C5 = m2
C6 = m
2

2 2
2 (),
(5.4)
0
( + 1) ( 1).
(5.5)
The rest of Ci are computed in detail in Appendix C, and we only show the results as
follows:
2 (2)
2 1
B(0 1, + 2) + m2
C1 = m2
64 ( + 3) ( 3)
3

4B 2, 12 F 2, 3 + , 52 , 14 3B 3, 12 F 3, 3 + , 72 , 14
2 (2 + 1) h

B , 12 3F2 1, 2 + 1, ; 1, + 12 ; 14
4
+ m2
(3 )( 2)

i
3
1
(5.6)
4 B + 1, 2 3F2 1, 2 + 1, + 1; 1, + 32 , 14 ,
2 1

1
7 1
C2 = m2
64 (3 + ) ( 3)B 3, 2 F 3, 3 + ; 2 ; 4
2 (2)

B , 12 3F2 2, 1, ; 2, + 12 ; 14 ,
4
(5.7)
+ m2
3
2 1 (4 + ) ( 2)

1
5 1
C3 = C4 + m2
16 ( + 2)( + 3) B 2, 2 3F2 4 , 2 + , 2; 3 , 2 ; 4
2

(2 )
B , 12 3F2 2, , 2, ; 1, + 12 ; 14 , (5.8)
4 (2)
+ m2
(4 )

(4
+
)
( 2)
2
1
1
7 1
C4 = m2
32 ( + 2)( + 3) B 3, 2 3F2 4 , 2 + , 3; 3 , 2 ; 4
2 1

(2 )
1
+ m2
2 4 (2) (4 ) B + 1, 2

(5.9)
3F2 2, 2, + 1; 1, + 32 ; 14 .
507
We can rewrite these expressions in terms of the hypergeometric function F 2F1 only,
and the results and their derivations are presented in Appendix C (see (C.20), (C.21), (C.38)
and (C.39)). Since we could not find any convenient transformation formula from 3F2 to
2F1 in the literature, we have established a transformation technique in Appendix C. (A
more concise explanation can be found in Appendix D.)
Let us consider the expressions (C.20), (C.21), (C.38) and (C.39). We now perform
the Taylor expansions of the hypergeometric functions around = 0, in order to see the
divergence structures of the coefficients Ci . Here, we are only interested in 1/ terms in
the sense of the MS scheme, and the hypergeometric functions which contribute to the
desired pole terms are only generated from the following expansion:

F , a; + b; 14 = F , 0; ; 14

(5.10)
+ aF (0,1,0) , 0; ; 14 + bF (0,0,1) , 0; ; 14 + O 2 ,
where
def
F (n,m,l) (, ; ; z) = n m l F (, ; ; z).
(5.11)
These differential coefficients (for 6= 0) are evaluated by

F (0,0,1)(, 0; ; z) = 0,
F (0,1,0)(, 0; ; z) = F (1,0,0)(0, ; ; z) = z 3F2 (1, 1, + 1; 2, + 1; z).
F (, 0; ; z) = 1,
(5.12)
(5.13)
Here we again encounter the generalized hypergeometric function 3F2 , however in the
present case, it can be reduced to the ordinary hypergeometric function 2F1 through the
following formula (derived in Appendix D):
1X
F (k, ; ; z),
n
n
3F2 (1, , n + 1; 2, ; z) =
n > 1, |z| < 1, <( ) > 0.
(5.14)
n > 1, |z| < 1, <( ) > 0,
(5.15)
k=1
Combining (5.13) and (5.14), we have

F (0,1,0)(n, 0; ; z) =
n
zX
F (k, 1; + 1; z),
k=1
and thus

az X
F (k, 1; + 1; z) + O 2 ,
F (n, a; + b; z) = 1 +
k=1
n > 1, |z| < 1, <( ) > 0.
(5.16)
Owing to this formula, all coefficients in front of 1/ can be written in terms of 2F1 and the
gamma functions. After some algebra, we obtain

m 1
1
5 m 1
1
1
0
0
+ O(1),
C2 = 2 +
+ O(1),
C1 = 2 + +
9
3
18
3
6
6

m 1
1
1
1
+
+ O(1),
C40 =
+ O(1),
C30 =
2
12
72
6
12
508
C50 =
1
2m
+ O(1),
2
1
C60 = + O(1),
where the overall factor (42 )2 seen in (5.1) and (5.2) is absorbed in Ci ; i.e.,
2
Ci0 = 42 Ci ,
(5.17)
(5.18)
and we have defined

m = E + ln
m2
,
42
E = Euler const.
Therefore from Eqs. (5.1), (5.2) and (5.17), we obtain

Z

4g04 4
11
1
1
D
0
8
F
+
x
F
d
I1 [A] =
m
0
+ O ,
4
2
(4)
2
4
Z

4

4g0
8m
4
1
dD x0 F F + O 0 ,
2+
I2 [A] =
(4)4
(5.19)
(5.20)
(5.21)
and due to Eq. (2.1) the renormalization part of the effective action (purely gluon parts) at
the 2nd order in F in our regularization is found to be
Z

4g04
11
1
D
d
F
x
F
(5.22)
+ O 0 .
[A] =
0
4
2
4
(4)
Finally, let us put a comment on what our results imply. Picking up C5 and C6 from
Eqs. (5.1) and (5.2), and using (3.17), we extract the following quantities corresponding to
self-energy parts:

4 ab 2
k k k 6C50 ,
4
(4)

4 ab 2
= g04
k k k 10C60 .
4
(4)
4
5 ab
= g0
(5.23)
6 ab
(5.24)
As briefly shown in Appendix E, the 5 and 6 exactly coincide with the Feynman
diagram results, if the coefficients of gluon kinetic terms are evaluated in the region close to
the light cone k 2 0; in other words, if k 2 is much smaller than the mass parameter m2 .
(Note that CA = 2 in the su(2) case.) Thus, it is very natural to expect that the other
coefficients C1 , C2 , C3 and C4 should possess the same meaning.
6. Conclusions and discussions

In this paper, we have explicitly calculated the gluon parts of the two-loop Euler
Heisenberg actions, which are organized at the level of an implicit formulation in the
previous paper [14]. The present results are still preliminary to reach a clear physical
quantity such as -functions, however this paper is an important step toward a full-fledged
extension of the world-line formalism to two-loop YangMills theories. One of the main
obstacles for this aim is the problem of how to integrate the proper time variables at
higher loop calculations. Also, in the sense of field theory limit of string theory, this is
509
an important problem: at the level of string theory, it is recognized as the problem how to
perform the moduli integrals on a multiloop world-sheet.
We have performed the path integrals in Section 4, applying the world-line Green
functions to all combinations of su(2)-like charges (the signatures; q.v. (4.11),
(4.12), (4.14) and (4.15)). Then extracting the parts corresponding to the wave function
renormalization, we have been able to perform the world-line moduli integrals to reveal
the divergence structure in Section 5. In the standard method, it is difficult to perform
the loop integrals containing a mass parameter, while in our case, all the integrations
are carried out in Appendix C, giving rise to (generalized) hypergeometric functions as a
result. This is certainly a significant point from a theoretical viewpoint, and hence we have
mainly focused on the technical issue concerning the integrations. We should also note
that the pseudo-abelian technique has worked out both at one- and two-loop levels, with
reproducing the Feynman diagram results of Appendix E. We expect that our integration
method will straightforwardly apply to higher order terms in F as well.
Although we have verified that our results contain correct Feynman diagram contributions, the followings should further be investigated as a next step toward our goal: the
coincidence between the present results and those by the Feynman diagram method can
only be understood in the region close to the light cone k 2 0 (k 2 m2 ), where k are
the external gluon momenta. On the other hand, as seen in Section 3, we do not have the
restriction on k 2 at the one-loop level, in order to extract the -function. Similarly we shall
encounter the same difference in the ghost loop calculations, and should clarify the reason
for this kind of discrepancy. Related to this issue, another question is whether or not we can
evaluate the pole structures of Ci for m = 0 (or m2 k 2 ). As inferred from Eqs. (E.14)
and (E.16), the integrals Ci might depend on the region of either k 2 m2 or not. However
the present calculations do not indicate such a dependence, simply because Ci are not the
Fourier modes of gluon two-point function. To clarify this point, one should compute the
correlator of two gluon vertex operators (of bosonic field representation) along the outline
of Appendix B in Ref. [14]; in this case, the formulation should be extended to the super
world-line formalism in order to optimize the inclusion of the four point interactions involving external legs. Anyway, in order to find the correct -function coefficient at two
loops, we also have to add the contributions including the counter terms generated from
one-loop divergences (in the massive formulation).
Following Ref. [25] a gauge symmetry breaking IR gluon mass m2 (introduced there
only as a device to separate IR and UV divergences) requires a further counter term
1
Zx m2 Aa Aa ,
2
with Zx =
g 2 CA
16 2
(6.1)
and CA = 2 in our case, cancelling m2 dependent singularities in the MS dimensional

regularization scheme. Insertion of this counter term gluon mass into the one-loop
contribution of order F 2 as found in Eq. (3.15), i.e., calculating with m2 m2 + Zx m2
and expanding to first order in m2 /, thus using

g 2 m2 CA
m2 T
m2 T
T ,
(6.2)
e
1
e
16 2
510
yields, e.g., a singular two-loop F 2 -contribution:

Z
Z

2 2
CA
D
1D/2 m2 T g m CA
2
T
dD x0 TrL F 2
dT
T
e
ct =
2(4)D/2 12
16 2
0
Z

g 4 CA2 10
1
D
0
d
F
=
x
F
0
+ O .
4
4
(4) 3
(6.3)
However in the background formalism there should be further counter terms including
quantum as well as background fields. This has to be further analyzed in order to reproduce
the usual -function coefficient.
This paper concerns a theoretical interest, and is an important part of the ongoing longterm effort to find a way to getting over difficulties in the current calculation methods for
higher loop amplitudes in field theory. Since the whole computation process in the worldline formalism looks completely different from the standard field theory calculations, the
above questions are the milestones in the future study, and should be solved in order to
make future practical applications successful.
Acknowledgement
We would like to thank M. Jamin for helpful discussions concerning infrared regularization.
Appendix A. List of path reversal formulae

This appendix is a brief note on the reversals of path ordering and proper time directions.
The standard definition of the path ordering (the normal type) is
(Z
Z1
d M [x] =
P exp
Z
X
n=0
Zn1
d2
d1
dn M1 [x] Mn [x],
(A.1)
and the anti-path ordering (used in [14] for a certain reason) is
(Z
)
Zn1
Z1
Z
X
d M [x] =
d1 d2
dn Mn [x] M1 [x].
P exp
n=0
(A.2)
We here assume that the M in the above two definitions are the same objects.
The relations between the path and the anti-path ordering formulae are given by
(ZT
ae
[x; T , 0] = P exp
W
)ae
d M [x]
(ZT
)ea
T
= P exp
d M [x]
,
(A.3)
(Z
"
a
ae
W
[x; S, , 0] = TrC
(ZS
)
T
P exp
d M [x]
d MT [x]
P exp
511
(A.4)
where the M T represents the transposition on both the color and Lorentz spaces.
Note that the transposition makes an additional minus sign in front of the Ax term in M .
If we change the sign of gauge coupling g, this additional sign drops out, and the starting
formulae presented in Section 2 follows from the previous paper directly.
The other useful observation is related to path inversion: as can be seen from Eq. (2.7)
by distinguishing and 0 carefully (see Eq. (4.1)), the following relation holds:
ab
ba
(A.5)
M [x 1 ] = M + [x] ,
where x 1 denotes the inverted path x, each of them defined by
x : [ , ] R 4 ,
x
: [ , ] R ,
4
7 x( ),
7 x
(A.6)
( ) := x( + ),
(A.7)
respectively. As a consequence of Eq. (A.5) we have

)T
(Z
d M [x]
P exp
(Z
)
1
= P exp
d M x
,
(A.8)
where again T means transposition in both color as well as Lorentz space.

Describing path inversion by the x 1 symbol is a useful tool in performing world-line
calculations (rather than changing the direction of or other possibilities), because it nicely
fits together with the following identities for world-line path integrals: 6
x(S)=y
Z 2
x(S)=y
Z 1

DxF x 1
DxF [x] =
x(0)=y1
(A.9)
x(0)=y2
and consequently
I
x(S)=y
Z
Z
DxF [x] =
DxF [x] =
d y
x(0)=y
x(S)=y
Z
Dx F [x 1 ] =
d y
DxF [x 1 ],
x(0)=y
(A.10)
where F is an arbitrary functional and the integrands F [x 1 ] are to be understood as
follows: for any path x integrated over within the path integral the corresponding path x 1
is to be constructed (in thoughts) and the functional is to be evaluated for this path x 1 .
Thus, in the path integral, x 1 depends on x. Note that we also have
6 These identities can easily be verified from the path integral discretization.
512

x(S)=y
Z 2
x(S)=y
Z 2
[Dx]S F [x] =
x(0)=y1
Dx e
RS
0
d 14 x 2
x(S)=y
Z 1
x(0)=y1
RS
0
d 14 (x 1 )2

F x 1
x(0)=y2
x(S)=y
Z 1
Dx e
Dx e
F [x] =
RS
0
d 14 x 2
F x
x(0)=y2
x(S)=y
Z 1

[Dx]S F x 1 , (A.11)
=
x(0)=y2
i.e., our usual bracket notation for the free path integral part is not affected.
Using these relations and identities one can easily verify that the gluon propagator in a
background, given by the world-line representation [20]
(ZT
)ab
x(T
Z
Z)=x1
ab
(x1 , x2 ) =
[Dx]T P exp
dT
0
x(0)=x2
d M [x]
0
(A.12)
satisfies the following property:

ba
ab
(x1 , x2 ) = (x2 , x1 ).
(A.13)
Appendix B. The derivation of I1 [A]

In this appendix, we show some details for the computation in Section 4.1. First, we
need various values of the quantity (4.17) at (, 0 ) = (0, T3 ) for a, b = 1, 3, and those are
given by the following. The necessary derivatives of the Green functions are

2T2 F 2 ei F (T3 T1 )
13
(, 0 ; +, 0, ) =0, 0 =T =
, (B.1)
0 G
3
sin F T1 sin F T3 + F T2 sin F (T1 + T3 )

2F sin F T2 ei F T3
13
(, 0 ; 0, , +) =0, 0 =T =
, (B.2)
0 G
3

2F sin F T2 ei F T1
13
0

, (B.3)
0 G (, ; +, , 0) =0, 0 =T =
3

2T2 F 2 ei2F T3 sin F T1 cosecF T3
33
0

0 G (, ; +, 0, ) =0, 0 =T =
3

F
+ 2 1L (T3 )
ei F T3
,
(B.4)
sin F T3

2T1 F 2 ei2F T3 sin F T2 cosecF T3
33
0

0 G (, ; 0, , +) =0, 0 =T =
3

F
+ 2 1L (T3 )
ei F T3
,
(B.5)
sin F T3

2T31 sin F T1 sin F T2
33
(, 0 ; +, , 0) =0, 0 =T =
0 G
3

1
+ 2 (T3 )
(B.6)
(1L ) .
T3
513
The normalization constants are given by

Z

F2
1/2
(+,,0)
D
= (4) detL
dD x0 , (B.7)
N
and the other values can be obtained by exchanging the labels a on a and Ta
simultaneously; for example,

(B.8)
N (+,0,) = N (+,,0) T T .
2
Note also
N (+,+,0) = N (,,0) = N (,,0) .
(B.9)
Now, plugging in (B.1)(B.6) and the corresponding normalizations, for each line of
Eqs. (4.14) and (4.15) we get an expression of the following structure:

N (1 ,2 ,3 ) TrL (power series in F ) + TrL (power series in F )

(B.10)
TrL (power series in F ) .
Because of the antisymmetry of F , only even powers of F contribute to the traces.
Furthermore we have the property
N (1 ,2 ,3 ) (F ) = N (1 ,2 ,3 ) (F ),
(B.11)
and thus (B.10) is invariant with respect to F F . Therefore the F F terms in

Eqs. (4.14) and (4.15) just give a factor of two.
Using this fact, the expressions (B.1)(B.6) and the corresponding normalizations we
obtain from Eqs. (4.14) and (4.15) straightforwardly:
2
Z
1
1/2 F
D
dT1 dT2 dT3 detL
2 [A] = (4)
(2)
2
F
0
(
F 2 T2
2 cos F (T1 T3 ) cos 2F (T1 + T3 ) + cos F (T3 T1 )
TrL
(2)
F

+ cos F (T1 T3 )TrL cos 2F (T1 + T3 )

+ 2TrL
F sin F T1
2 sin F T3 sin 2F (T1 + T3 ) cos F (2T1 T3 )
(2)
F
)

+ cos F (2T1 + T3 )TrL cos 2F T3
dD x0 ,
(B.12)
and
1
1 [A] = (4)D
2
Z
0

1/2
dT1 dT2 dT3 detL
F2
(2)
F
514
TrL
F
2 cos F (2T1 + 3T3 ) + cos F (2T1 + T3 )TrL cos 2F T3
sin F T3

+ 2 cos F T3 TrL cos 2F (T1 + T3 )

+ TrL
sin F T1 sin F T3
(2)
F T2
1
T2
cos 2F (T3 T1 ) cos 2F T3 TrL cos 2F T1

+ TrL
F 2 T2 sin F T1
(2)
F sin F T3
2 cos 2F (T1 + 2T3 ) cos 2F (T1 + T3 )TrL cos 2F T3
cos 2F T3 TrL cos 2F (T1 + T3 )
+ (T3 )2(1 D)TrL cos 2F T1

)

+ (T2 ) TrL cos 2F (T1 T3 ) TrL (cos 2F T1 ) TrL (cos 2F T3 )
Z
dD x0 ,
(B.13)
with
(2)
F = sin F T1 sin F T3 + F T2 sin F (T1 + T3 ).
(B.14)
The sum of (B.12) and (B.13) gives I1 [A] by definition, and is certainly equivalent to
Eq. (4.21). In order to obtain Eq. (4.21) itself, we should further take the following
modification into account. Exchanging T2 and T3 , and using the relations followed
from (B.14)
F
F 2 T2 sin F (T1 + T3 )
F
= (2) sin F T1 +
,
(2)
sin F T3
sin
F
T
3
F
F
(B.15)
sin F T1 sin F T3
F
1
= (2) sin F (T1 + T3 ),

(2)
T2
F T2
F
(B.16)
we finally arrive at the full expression shown in Eq. (4.21).
Appendix C. Computational details of Ci

In this appendix, we show the details of how to perform all the integrals in C1 , C2 , C3
and C4 . Let us first perform the T3 integrals in Ci . Applying the following formula to the
T3 parts in (5.3)
Z
0
ept (1 + at) dt = p1 a ep/a (1 ; p/a),
a>0
(C.1)
515
and then transforming the T1 and T2 integrals with

Z
Z
dT1 dT2 f (T1 , T2 ) =
Z1
dT T

du f T (1 u), T u ,
(C.2)
we obtain the followings for i = 1 and 2 (further using m2 T = t)

Ci = m

2 2
Z1
dt

du fi (u)t 2+ et +t u(1u) 3; tu(1 u) ,
i = 1, 2,
(C.3)
and for i = 3 and 4

Ci = m

2 2
Z1
dt

du fi (u)t et +t u(1u) (3 )t 3; tu(1 u)
+ t u(1 u) 3; tu(1 u) + t u(1 u) ( 3; ) ,
(C.4)

= tu(1 u)
(C.5)
and
fi (u) = u
i2a
(1 u)
a+5i
i1
a=
2

(C.6)
G
with Gauss integer symbol [ ]G . In the following, we evaluate (C.3) and (C.4) separately
because we shall proceed on different technique and formulae.
C.1 C1 and C2
We now consider the t integration in (C.3). In the first place, it can be integrated in terms
of the formula (6.455.1 in [24]):
Z
0
t 1 ept (, t) dt =

( + )
p
F
1,
+
;
+
1;
,
( + p)+
+p
<( + p), < , <( + ) > 0.
(C.7)
Then applying the formula (9.131.2 in [24]),

F (, ; ; z)
( + ) ( )
(1 z) F ( , ; + 1; 1 z)
=
() ()
( ) ( )
F (, ; + + 1; 1 z),
(C.8)
+
( ) ( )
we have
516
Ci = m

2 2
Z1
du fi (u) ( + 3) ( 3)F 4 , 3 + ; 4 ; u(1 u)
+ m

2 2
Z1
0

3 (2)
F 2, 1; 2; u(1 u) .
du fi (u) u(1 u)
3
(C.9)
In order to perform the u integrations, we consider the following define u+ (u ) for larger
(smaller) values of u, and apply
p

1
(C.10)
u = 1 1 y .
2
With this change of variable, the following formula holds for an arbitrary function H (u)
Z1
1
du fi (u)H u(1 u) =
4
Z1
dy
gi (y)H (y/4),
1y
where we have defined gi (y) for all i:

1+ 1y
1 1y
+ fi
,
gi (y) fi
2
2
(C.11)
(C.12)
and we hence have

1
1
y(4 3y),
g2 (y) = y 2 ,
(C.13)
16
16
1
1
g4 (y) = y 2 .
(C.14)
g3 (y) = y(2 y),
8
8
At a glance, one may realize that the y integration can be performed by the formula
(7.512.12 in [24])
g1 (y) =
Z1
(1 t)1 t 1 F (, ; ; zt) dt = B(, ) 3 F2 (, , ; , + ; z),
0
< , < > 0, |z| < 1
(C.15)
and this is exactly the way we obtain the generalized hypergeometric 3F2 expressions (5.6)
and (5.7).
In order to further derive hypergeometric 2F1 expressions, we rather notice the special
case = in (C.15). Before applying (C.15), we adjust auxiliary variables of 2F1 until the
special case is applicable, with using the formula (9.137.18 in [24])
F (, ; ; z) =

F (, ; + 1; z) + F ( + 1, ; + 1; z).
(C.16)
Before writing down the final results for C1 and C2 , we here remark on the way how C1
contains the 0 dependence. Using the following formula (9.137.11 in [24]) with = 0
F (, ; ; z) F (, + 1; ; z) + zF ( + 1, + 1; + 1; z) = 0,
(C.17)
and then decomposing the second term in (C.9)

2 y
y
y
=1+
,
F 2 + 1, 1; 1;
F 2, 1; 2;
4
2 4
4
517
(C.18)
we extract the following integral from C1 :

l1 = 4 (2)
def
Z1
y 2(4 3y)(1 y)1/2 dy
Z1
= (2)
u2 (1 u)+1 du = (2)B(0 1, + 2).
(C.19)
Evaluating the l1 contribution separately, we obtain the results written in terms of 2 F1 only:
2 (2)
2 1
B(0 1, + 2) + m2
C1 = m2
64 ( + 3) ( 3)
3

4B 2, 12 F 2, 3 + ; 52 ; 14 3B 3, 12 F 3, 3 + ; 72 ; 14
(2 + 1)B(, 12 )
+ (m2 )2 4e
(3 )( 2)( 1)

( 2)F 1, 2 + 1; + 12 ; 14 + F 2, 2 + 1; + 12 ; 14

2 e 3 (2 + 1)B( + 1, 12 ) 2

F 1, 2 + 1; + 32 ; 14
4
m2
4
(3 )( 2)

2
2( 2)
F 2, 2 + 1; + 32 ; 14 +
F 3, 2 + 1; + 32 ; 14 ,(C.20)
+
( 1)
( 1)

2 2 1
1
7 1
C2 = m
64 (3 + ) ( 3)B 3, 2 F 3, 3 + ; 2 ; 4
1

2 2 (2)B(, 2 )
4
F 1, 2; + 12 ; 14
+ m
1

2
2
1 1
1 1
F 2, 2; + 2 ; 4 +
F 3, 2; + 2 ; 4 .
(C.21)
2
(2 )(3 )
C.2. C3 and C4
In this case, we split Ci ; i = 3, 4 as
2

(3 )R1 + R2 + R3 ,
Ci = m2
(C.22)
where
Z1
R1 =
Z
du fi (u)

t 1+ et +t u(1u) 3; tu(1 u) dt,
Z1
R2 =
Z
du fi (u)u(1 u)
(C.23)

t 2+ et +t u(1u) 3; tu(1 u) dt,
(C.24)
518
Z1
R3 =
Z
du fi (u)u(1 u)
t 2+ et +t u(1u)
( 3; ) dt.
(C.25)
First, we integrate R2 using (C.7) in the same way as done in (C.3):

(2)
R2 =
+3
Z1

2

fi (u) u(1 u)
F 1, 2; + 4; 1 u(1 u) du.
(C.26)
Second, we notice that the incomplete gamma function is related to the Whittaker function:
(, z) = z(1)/2ez/2 W(1)/2,/2(z),
(C.27)
and W, (z) satisfy

W, (z) = W, (z),

z
W, (z) W+1, (z),
zz W, (z) =
2

1
1/2
W1, (z).
W, (z) = z W1/2,+1/2(z) +
2
(C.28)
(C.29)
(C.30)
Applying the formula (C.27) and its derivative to R1 and R3 , one can prove the relation
R3 =
4
1
R1 R2 + R4 ,
2
2
(C.31)
where
Z1
R4 =

(2)2
du fi (u) u(1 u)
t 3/2 et +t u(1u)/2 W(4)/2,(3)/2( ) dt.
(C.32)
We here remove the derivative W from the r.h.s. of Eq. (C.32), making use of the
relation (C.29):
1
4
R1
R4 = R2
2
2
Z1

(4)/2
du fi (u) u(1 u)
t 2 1 et +t u(1u)/2W(2)/2,(3)/2( ) dt.
(C.33)
Adjusting the indices on W on the r.h.s. of (C.33) in terms of the recursion relation (C.30)
(with = ( 2)/2, = ( 3)/2), we then apply the formula
Z
0
t W, (at)ept dt =
( + + 32 ) ( + 32 )a +1/2
( + 2)(p + a/2)++3/2
519

1
2p a
3
,
F + + , + ; + 2;
2
2
2p + a
3
|< | > 0
2
and the r.h.s. of (C.33) becomes
< +
4
(2) ( + 2)
1
R2
R1 (3 )R1
2
2
( + 3)
1
Z

2

F 2, 1; + 3; 1 u(1 u) .
fi (u) u(1 u)
(C.34)
(C.35)
Substituting this expression for R4 in Eq. (C.31), which should then be inserted into
Eq. (C.22), and thereby eliminating R1 , we calculate Ci as follows:
Ci = m

2 2
Z1

2
du fi (u) u(1 u)
(2)

1
1
F 1, 2; + 4; 1 u(1 u) +
F 1, 2; + 3; 1 u(1 u)
3+
2+

2
Z1
y
y
(m2 )2 (2)
dy,
gi (y)(1 y)1/2
F 2, 2; + 4; 1
=
4(2 + )(3 + )
4
4

(C.36)
where we have used the transformations (C.11) and (C.16) at the second equality. We now
transform the argument of the hypergeometric function from 1 y/4 to y/4 through the
formula (C.8), and thus giving rise to F (4 , 2 + ; 3 ; y/4) and F (2, 2; 1; y/4).
(If we apply (C.15) at this stage, we obtain the generalized hypergeometric 3F2 expressions
(5.8) and (5.9).)
For the purpose to reduce 3F2 to 2F1 , we apply (C.16) twice to the latter F (two of
them explained right above), and apply the following formula to the former F (setting
= = 3 , = 2 + ):
F (, + 1; ; z) +
F (, ; ; z).
(C.37)
After that, we are able to integrate them by using (C.15) (with = ), and the results are
therefore
F ( + 1, ; ; z) =
2 1 (4 + ) ( 2)

B 2, 12
C3 = m2
16 ( + 2)( + 3)(3 )

(2 + )F 2, 3 + ; 52 ; 14 + (1 2)F 2, 2 + ; 52 ; 14
+ m2
2
(2)
(2 ) B( 12 , )
(4 ) ( 1)4
520

( 3)F 2, 2; + 12 ; 14 + 2F 3, 2; + 12 ; 14 C4 ,
2 1 (4 + ) ( 2)

B 3, 12
C4 = m2
32 ( + 2)( + 3)(3 )

(2 + )F 3, 3 + ; 72 ; 14 + (1 2)F 3, 2 + ; 72 ; 14
2
(2 ) B( 12 , + 1)
(2)
+ m2
(4 ) 2( 1)4

( 3)( 2)F 2, 2; + 32 ; 14 + 4( 3)F 3, 2; + 32 ; 14

+ 6F 4, 2; + 32 ; 14 .
(C.38)
(C.39)
Appendix D. The proof of Eq. (5.14)

We show the outline of how to prove the formula (5.14). Basically we follow the
same technique as we used in Appendix C, with using (C.15) and (C.16) for the aim of
rearranging 3F2 into 2F1 . We evaluate the l.h.s. of Eq. (C.15) in two ways: one is the direct
result (the r.h.s. of the formula), and the other is a combination with (C.16).
On the one hand, setting = n in (C.15), we have
Z1
(1 t)1 t 1 F (, ; n; zt) dt = B(, )3F2 (, , ; n, + ; z),
0
< , < > 0, |z| < 1.
(D.1)
On the other hand, using (C.16) n > 1 times, the integrand in Eq. (D.1) can be expanded as

r
n
n
Y
k XY +p1 n
F ( + r, ; ; zt),
(D.2)
(1 t)1 t 1
k
p r
k=1
r=0 p=1
Q
where we define 0p=1 1. Then applying (C.15) (with the case = ) to this expression,
the l.h.s. of (D.1) becomes

r
n
n
Y
k XY +p1 n
F ( + r, ; + ; z).
(D.3)
B(, )
k
p r
k=1
r=0 p=1
We thus have the equality

n, + ; z)

r
n
k XY +p1 n
3F2 (, , ;
n
Y
k=1
r=0 p=1
p r
Putting = 1, = n + 2, and using the identity

r
Y
n
p
= 1,
n+1p r
p=1
F ( + r, ; + ; z).
(D.4)
(D.5)
521
we therefore have proven the formula (5.14):

1 X
F (r + 1, ; + ; z),
n+1
n
3F2 (1, , n + 2; 2, + ; z) =
(D.6)
r=0
which holds for <(), <() > 0, |z| < 1 and n > 0 (we have proven it for n > 1, but the
n = 0 case is trivial).
Appendix E. Feynman diagram results

In this appendix, we present some Feynman diagram calculations in reference to the
results obtained by our method. We follow the same notations as Refs. [22,23] in the
Minkowski space, however use the massive propagator in the (background) Feynman
gauge:
a
= (i)
ab g
.
k 2 m2 + i
We only deal with the parts which contain the eight-figure vacuum diagram. When the
massive propagator is introduced, the tadpole contributions remain (see the diagrams (a)
and (b) in Fig. 2). After some calculations, the tadpole diagram (a) reads
ab
CA2 ab
(m2 )1 (3 2) ( 1)
(4)2

8 k 2 g k k J1 + D(k k J1 + 4k J2 + 4J3 ) ,
T(a) = g 4
where
(E.1)
1
dD p
,
D
2
(2) i ((p + k) m2 )(p2 m2 )2
Z
p
dD p
,
J2 =
D
2
(2) i ((p + k) m2 )(p2 m2 )2
Z
p p
dD p
,
J3 =
(2)D i ((p + k)2 m2 )(p2 m2 )2
J1 =
and these are equal to
Fig. 2. The eight-figure diagrams contained in the coefficients C5 and C6 .
(E.2)
(E.3)
(E.4)
522
( + 1)
( + 1)
j1 ,
J2 =
k j2 ,
2
(4)
(4)2

( + 1) 1
1
g j3 k k j2 ,
J3 =
(4)2 2
2
J1 =
(E.5)
with
Z1
j1 =
x m2 k 2 (x x 2 )
1
dx,
(E.6)
Z1
j2 =
x(1 x) m2 k 2 x x 2
1
dx,
(E.7)
Z1
j3 =
x m2 k 2 x x 2
1 2 1 2
m
k (2j2 j1 ).
2
2
(E.8)
Thus Eq. (E.1) becomes

1
CA2 ab
(3 2) ( 1) ( + 1)
m2
42
(4)

D 2
2
m
+ k g k k (D 8)j1 2Dj2 , ,
and the other tadpole diagram (b) is calculated as

(a) ab
= g4
12
CA2 ab
g (3 2)D ( 1) ().
m2
42
(4)
As a result, the sum of these tadpole contributions takes the transversal form:
ab
T(b) = g 4
(E.9)
(E.10)
(a) ab
(b) ab
+ T
1
C 2 ab
= g 4 A 42 m2
(3 2) ( 1) ( + 1)
(4)
T ab
= T

k 2 g k k (D 8)j1 2Dj2 .
When k 2 0 (|k 2 | m2 ), the quantities ji behave as

1
1

1
1
1
,
j 2 = m2
,
j 3 = m2 ,
j 1 = m2
2
6
2
and when m2 0, they reduce to
1
1
1
B(, ),
j2 = k 2
B(1 , 1 ),
j1 = k 2
2

1
j3 = k 2 B(1 , 1 ).
2
Therefore in these limits Eq. (E.11) behaves
( 10
2 ab

C
, k2 0
4 A
2
g
k
k
k
+ O(1).
T ab
= g0
(4)4
0,
m=0
(E.11)
(E.12)
(E.13)
(E.14)
523
Next, the diagram (c) amounts to

ab
(e) = 6g04

CA2 ab 2
k g k k exp 2 ln 42 2 ()(2j3 )2 ,
4
(4)
(E.15)
where j3 is given by (E.8). Substituting Eqs. (E.12) and (E.13) for this j3 in each limit, we
derive the following limits of (E.15):
6
12
k2 0
2 + m ,
2
ab

C
ab
(e)
4 A
2
+ O(1), (E.16)
k g k k
= g0
(4)4
6 24 + 12 , m = 0
2
where m is given by (5.19), and
= E + ln
k 2
.
42
(E.17)
We therefore conclude that our results coincide with the Feynman diagram calculations in
the situation k 2 0 (m 6= 0) (q.v., (5.23) and (5.24)):

CA2 ab 2
k g k k (10C60 ) = 6 ab
,
4
(4)

C 2 ab
= g04 A 4 k 2 g k k (6C50 ) = 5 ab
.
(4)
4
T ab
= g0
ab
(e)
(E.18)
(E.19)
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
Z. Bern, D.A. Kosower, Nucl. Phys. B 379 (1992) 451.

R.R. Metsaev, A.A. Tseytlin, Nucl. Phys. B 298 (1988) 109.
Z. Bern, Phys. Lett. B 296 (1992) 85.
P. Di Vecchia, A. Lerda, L. Magnea, R. Marotta, Phys. Lett. B 351 (1995) 445.
P. Di Vecchia, A. Lerda, L. Magnea, R. Marotta, R. Russo, Nucl. Phys. B 469 (1996) 235.
Z. Bern, D.C. Dunbar, Nucl. Phys. B 379 (1992) 562.
Z. Bern, L. Dixon, D.A. Kosower, Phys. Rev. Lett. 70 (1993) 2677; Nucl. Phys. B 412 (1994)
751.
Z. Bern, D.C. Dunbar, T. Shimada, Phys. Lett. B 312 (1993) 277.
D.C. Dunbar, P.S. Norridge, Nucl. Phys. B 433 (1995) 181.
M.J. Strassler, Nucl. Phys. B 385 (1992) 145.
A.M. Polyakov, Gauge Fields and Strings, Harwood, 1987.
C. Schubert, Acta Phys. Polon. B 27 (1996) 3965, print-97-273.
M.G. Schmidt, C. Schubert, lapth-703-98, hep-th/9810161.
H.-T. Sato, M.G. Schmidt, Nucl. Phys. B 560 (1999) 551.
M.G. Schmidt, C. Schubert, Phys. Lett. B 331 (1994) 69; Phys. Rev. D 53 (1996) 2150.
K. Roland, H.-T. Sato, Nucl. Phys. B 480 (1996) 99; Nucl. Phys. B 515 (1998) 488.
H.-T. Sato, M.G. Schmidt, Nucl. Phys. 524 (1998) 742.
D. Fliegner, M.G. Schmidt, C. Schubert, Nucl. Phys. (Proc. Suppl.) 51C (1996) 174.
H.-T. Sato, Phys. Lett. B 371 (1996) 270.
M. Reuter, M.G. Schmidt, C. Schubert, Ann. Phys. (NY) 259 (1997) 313.
C. Itzykson, J.B. Zuber, Quantum Field Theory, McGraw-Hill, 1980.
L.F. Abbott, Nucl. Phys. B 185 (1981) 189, Acta Phys. Polon. B 13 (1982) 33.
524
[23] D.M. Capper, A. MacLean, Nucl. Phys. B 203 (1982) 413.

[24] I.S. Gradshteyn, I.M. Ryzhik, Table of Integrals, Series and Products, Academic Press.
[25] K. Chetyrkin, M. Misiak, M. Mnz, Nucl. Phys. B 518 (1998) 473.

A note on relation between holographic RG

equation and Polchinskis RG equation
Miao Li a,b,
a Institute of Theoretical Physics, Academia Sinica, Beijing 100080, China
b Department of Physics, National Taiwan University, Taipei 106, Taiwan
Received 21 February 2000; accepted 27 March 2000
Abstract
We clarify the relation between the recently formulated holographic renormalization group
equation and Polchinskis exact renormalization group equation. 2000 Elsevier Science B.V. All
rights reserved.
PACS: 11.10.Hi; 04.50.+h
The holographic renormalization group flow has been clarified in the context of
AdS/CFT correspondence [1,2] as well as in the context of open string versus closed
string [3]. (For earlier attempts in this, see [411], also see [12].) It was noticed in these
papers that the holographic RG equation arising either from the HamiltonJacobi theory of
supergravity or from world-sheet considerations has a strong resemblance to Polchinskis
exact RG equation [13].
Apparently the two equations are different. We aim in this short note to clarify the
relation between them. The effective action in the AdS/CFT correspondence [1416] is
defined as a functional of coupling constants, while the effective action of Polchinski is
a functional of the fundamental fields. Thus the holographic RG equation is naturally a
differential equation in coupling coupling constants, and the Polchinski RG equation is
one in fundamental fields.
For simplicity and without loss of generality, we will consider the field theory of a single
Hermitean matrix. We start with a single matrix model in 0 dimension to illustrate some of
our ideas. Although in the 0-dimensional matrix model there is no infinity to remove, one
still can design an artificial RG flow by introducing a cut-off in the quadratic term in the
action
1
(1)
S0 = NK(t) tr 2 ,
2
mli@phys.ntu.edu.tw
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 2 0 1 - 7
526
M. Li / Nuclear Physics B 579 (2000) 525532
where is a Hermitean matrix, K(t) is the cut-off propagator, depending on t = ln a,

a is the running cut-off. Unlike in a genuine field theory, where once an interaction
term is introduced in the action, many other terms will be generated with a nontrivial
K(t) in order to keep the physics invariant under changing t. In our case, there is
much freedom in satisfying the RG flow, as we shall explain later. To mimic the N = 4
super YangMills theory, we introduce the interaction part as a sum of single trace
operators
X
n (t) tr n ,
(2)
S1 = N
n>3
where again a factor N is introduced to follow the usual large N field theory convention.
With this convention, the effective action as a functional of n defined by
Z
S((t ),t )
= [d] eS0+S1 ,
(3)
e
has the usual genus expansion
X
N 22h Fh ,
S(n ) =
(4)
and the connected two point functions of operators tr n in the leading order is proportional
to N 0 . We pause to emphasize that it is crucial to introduce only single trace operators. In
the AdS/CFT correspondence, a single trace operator is related to a field in SUGRA or
string theory, the role of which is played by n in our toy model. Multitrace operators are
related to multiparticle states. The RG equation as formulated in [1,2] has to do with only
single trace operators.
To ensure the effective action S be independent of t, S1 must satisfy a differential
equation [13]. As we shall see, this differential equation can not be satisfied by our model
in which S1 contains only single trace operators. Thus we need to relax this equation to be
the one valid only when taken average in the path integral, namely

2 S1
ij,lk S1 S1
ij,lk
+ K2
+ K3 = 0,
(5)
t S1 + K1
ij lk
ij lk
where
R
[d]O eS0 +S1
.
hOi = R
[d] eS0+S1
We will determine K1 , K2 , K3 momentarily. Note that the constant term K3 can be

removed by a shift of S1 , and this shift can be absorbed into the definition for the measure
of . Thus in the following we will ignore this term.
For completeness, we will derive Eq. (5). We start with

Z
1
S
2
(6)
t e = [d] t K tr + t S1 eS0 +S1 = 0.
2
Use (5) to replace t S1 in (6). Next, use the fact
527
S1 S1 S0 +S1
e
ij lk

Z
ij,lk S1
= [d]K1
NKkl +
eS0 +S1
ij
lk

Z
2 S1
ij,lk S1
ij,lk
= [d] NKK1
kl K1
eS0 +S1 ,
ij
ij lk
ij,lk
[d]K1
ij,lk
ij,lk
we see that if we choose K2 = K1 , then the second derivatives of S1 cancel. Applying

the same trick to the first term in the last line of the above equation

Z
ij,lk
eS0 +S1
[d]NKK1 kl NKj i +
ij
Z
ij,lk
ij,j i S0 +S1
= [d] N 2 K 2 K1 j i kl NK1
.
e
Now the first term in the above can be used to cancel the first term in (6) if
ij,lk
K1
1 1
N t K 1 ik j l .
2
(7)
And the inhomogeneous term is removed by choosing K3 = 12 N 2 t ln K. However, as

we remarked before, this term can be absorbed into a redefinition of the measure and
henceforth we will ignore it. To summarize, we have derived the following equation

1
2 S1
S1 S1
+
= 0.
(8)
t S1 + N 1 t K 1
2
ij j i
ij j i
As we advertised, this is the weak form of Polchinskis equation.
The above equation is not valid if the average symbol is removed. (We call this equation
the strong form of Polchinski equation.) To see this, we compute
X
S1 S1
= N2
gn tr n ,
(9)
ij j i
n>4
where
gn =
l(n + 2 l)l n+2l ,
(10)
and
X
2 S1
=
NGmn tr m tr n
ij j i
(11)
Gmn = (m + n + 2)m+n+2 .
(12)
with
If the original Polchinski equation applies, then t S1 contains only single trace operators,
and can be used to balance the single trace operators in (9). However, (11) contains double
trace operators, and can not be balanced in Polchinski equation, in the large N limit, since
these operators are new independent operators. In order to solve Polchinski equation, we
528
need to introduce in S1 double trace operators. This in turn generates triple trace operators
in 2 S1 , etc. Thus in order for the Polchinski equation to hold, all multiple trace operators
must be introduced. With a little thought, it is easy to realize that this conclusion holds for
any matrix model, including N = 4 SYM. We thus learn that it is impossible satisfy the
strong form of Polchinski equation without introducing multiple trace operators.
On the other hand, there is no problem to satisfy the weak form of Polchinskis equation,
Eq. (8). It simply generates a first order differential equations for n (t). Also, as we shall
see shortly, the term 2 S1 in (8) is the same order as the term S1 S1 , in the large N
limit. This is quite different from the speculation of [1,2], where it is conjectured that the
holographic RG equation is just the strong form of Polchinski equation, if so, then 2 S1 is
suppressed by 1/N 2 .
Define the beta function
dn
,
(13)
n () =
dt
then
S
,
(14)
ht S1 i = n Nhtr n i = n
n
where we suppressed summation over n. Use (9),

S
S1 S1
.
= Ngn
ij j i
n
(15)
Use (11),

2
2 S1
eS
= N 1 Gmn eS
ij j i
m n

2S
S S
1
= N Gmn
+
.
m n m n
Substituting (13), (15) and (16) into (8), we find

1
1
2S
S
S S
+ N 2 t K 1 Gmn
+
= 0.
n + t K 1 gn
2
n 2
m n m n
(16)
(17)
This equation is almost the same as the holographic RG equation, say as presented in [3].
The RG equation in [3] is derived for the leading order in the large N limit. To compare
with that, we use the genus expansion of (4) to derive in the leading order

1
F0 F0
F0 1
1
+ t K 1 Gmn
= 0.
(18)
n + t K gn
2
n 2
m n
Two crucial points deserve mentioning explicitly. unlike one would naively think, the
term S1 S1 in the weak form of Polchinski equation is not identified with SS in the
holographic RG equation. Rather, it generates only the form S, a correction to the beta
function in (17). On the other hand the term 2 S1 in the weak form of Polchinski equation
generates both the terms SS and 2 S in the holographic RG equation, with the term
2 S subleading to SS in the large N limit. Clearly, the former is identified with the
529
disconnected part of two point functions, as also used in [3], while the latter is identified
with connected part of two point functions. Clearly, both of these terms appear only in
a large N theory, since they come from the double traced operators in the Polchinski
equation. With a single scalar field, one has only the S term. The resulting RG equation
can not be interpreted as coming from the HamiltonJacobi equation of a gravity theory.
In the AdS/CFT correspondence, the beta functions are determined by the local
part of the effective action S. Denote this local part by Sloc . It contains kinetic term
1 mn
2 G t m t n , so the beta function is given by
m = Gmn
Sloc
,
n
(19)
where Sloc is a functional of m (t) which is determined by the initial value problem.
However, in our toy model there is no such a kinetic term in the extra dimension t, the
reason is quite simple: the effective action (3) is defined already as a functional of initial
values m (t). We do not know how to define an off-shell action which can be expressed
as an integral over the whole range of t. In fact, the beta functions are not completely
determined by demanding RG invariance of the effective action only.
The 0-dimensional one-matrix model can be solved completely in the large N limit [17].
For instance, one can derive the SchwingerDyson equation in this limit. Apparently given
a finite set of functions {n (t)}, one of them is determined by the rest by requiring RG
invariance. Our above discussions serve only for the purpose of deriving the holographic
RG equation from the weak Polchinski equation. Eq. (17) has a flavor of the string equation
and Virasoro constraints in one-matrix model. Incidentally these equations can be derived
by an action principle [18]. It may be worthwhile to pursue along this direction.
The beta-functions are completely determined in a field theory. The new ingredient here
is the requirement of the cut-off independent correlation functions. It is straightforward to
generalize the above consideration to one matrix model in D-dimensional spacetime. We
use euclidean signature. Now the regularized kinetic action is
Z
1
(20)
S0 = N dD x dD y K(x y, t) tr i (x)i (y),
2
where K(x y, t) is the cut-off inverse propagator. When t , it tends to a delta
function. Since we are dealing with a field theory, in order to have a RG invariant partition
function, it is not enough to have a local interaction action S1 . Assume the inverse of
K(x y, t) exist (so that the kinetic term is not degenerate), denote this inverse by
K 1 (x y):
Z
(21)
dD z K 1 (x z)K(z y) = D (x y).
The weak form of Polchinski equation reads

Z
S1
2 S1
S1
1
D
D
+
= 0,
d x d y K1 (x y)
t S1 + N
(x)ij (y)j i (x)ij (y)j i
(22)
530
where
1
K1 (x y) = 1
2
dD x 0 dD y 0 K 1 (x x 0 )t K(x 0 y 0 )K 1 (y 0 y).
(23)
In order to satisfy the above equation, S1 must contain all the nonlocal terms. If we start
with a local action
XZ
dD x n (x) tr n (x),
(24)
S1 = N
then
X
S1
S1
= N2
m (x)n (y) tr m (x) n (y),
(x)ij (y)j i
(25)
infinitely many nonlocal terms are generated. The above can be expanded in derivatives
of . On the other hand,
X
2 S1
=
N D (x y)(m + n + 2) tr m (x) tr n (y),
(26)
(x)ij (y)j i
yielding a contact term. This contributes to the holographic RG equation a term

Z
X
S
2S
S
2
+
(m + n + 2) m+n+2 (x)
. (27)
N K1 (0)
m (x) n (x) m (x)n (x)
This is good news, since in the holographic RG equation, this is indeed a contact term. It
originates from the fact that m (x) is a field in the AdS space, thus has a local quadratic
term in the effective action. Denote the Fourier transform of K(x y) by K(p), then
K 1 (p) 0 when |p| , and the coefficient K1 (0) in (27) is given by
Z
1
1
(28)
dD p 2 t K 1 (p),
K1 (0) =
2
p
which is certainly convergent.
To have a closed form of RG equation, we thus introduce all possible single trace
operators into the interaction part S1 . A generic operator is
tr i1 in j1 jm .
Denote such a generic operator by OI (x), and the corresponding coupling by I (x). Now
go through the above steps, we will arrive at the following holographic RG equation

Z
S
S
2S
S
+
G
+
= 0, (29)
dD x I + g I
IJ
I (x)
I (x) J (x) I (x) J (x)
where all the components of the metric GI J are proportional to an integral of p2 t K 1 (p)
weighted by a polynomial of p. So for each component of GI J to be well-defined, the
cut-off propagator K 1 (p) must fall off more rapidly than any negative power of p for
large |p|. Note again that the correction to the beta function, g I , comes from the part
S1 S1 in the weak Polchinski equation. To compare with [3], we can identify I + g I
with the beta function defined on the world-sheet, where the cut-off is defined on the worldsheet. As already pointed out in [3], there is a UV/UV relation between string world-sheet
531
physics and spacetime physics. The two cut-offs are not identical, thus the two definitions
of the beta functions are not the same. Once again, our RG equation (29) is valid for any N .
Use the genus expansion (4), one recovers the RG equation of [13] in the large N limit.
The subleading term 2 S in (29) is to be interpreted as coming from quantum corrections
in the AdS/CFT context.
We also want to emphasize the fact that our RG equation (29) involves a single integral.
This is also true for the equation derived in [3]. In order to remove this integral to obtain a
local form, we need to introduce position-dependent cut-off. This is also related to general
covariance in AdS/CFT. We leave a detailed discussion of this to another work.
More equations can be derived by demanding the renormalized correlation functions to
be independent of the cut-off. These equations are just CallanSymanzik equations. They
can be derived from the local form of the RG equation [1,2], but not from (29). We can
derive them in the matrix field theory by generalizing the steps leading to (29). These
equations put together will give a closed system of equations for S and I . We are not sure
whether these beta functions can be written in a form (19). Note that the metric GI J in (19)
on the moduli space is the same as in (29), and is already determined in deriving (29). It
would be highly nontrivial if all these beta functions are given by a single functional Sloc .
Maybe this is the most crucial criteria for a holographic theory, and is generically violated
by an arbitrary matrix field theory such as the single scalar field theory.
It remains to generalize our construction to N = 4 super YangMills theory. Although
we do not see essential difficulty in doing this, we need to resolve the problem of
introducing a gauge invariant cut-off. A naive cut-off will not work, since this is not
compatible with local gauge transformation which mixes all energy scales. Another way to
see this is through the naive cut-off YangMills action
Z
dD x tr F (x)F (y)K(x y, t),
it is certainly not gauge invariant. It has been suggested to use Wilson loop as gauge
invariant variable to overcome this difficulty [19], and this line of approach was followed
up in [20]. However, the existence of AdS/CFT correspondence indicates that a gauge
invariant cut-off exists for local gauge invariant variables. We suspect that stochastic
quantization [21] may be one way to gauge invariantly regulate YangMills theory. And
the stochastic time has been interpreted as the RG scale recently in [22,23].
The discussion presented here may be viewed as a zero-slope limit of approach of [3].
However, the relation between string world-sheet and large N diagrams need to be further
clarified. Also, our approach does not has the drawback of assuming perturbation theory as
in [3]. The open/closed string duality need to be understood. One particularly nice example
was discussed in [24].
Acknowledgments
This work was supported by a grant of NSC and by a Hundred People Project grant
of Academia Sinica. I thank M. Yu for several useful conversations, and T. Yoneya for
comments on the manuscript.
532
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]
J. de Boer, E. Verlinde, H. Verlinde, On the holographic renormalization group, hep-th/9912012.

E. Verlinde, H. Verlinde, Gravity and the cosmological constant, hep-th/9912018.
J. Khoury, H. Verlinde, On open/closed string duality, hep-th/0001056.
E.T. Akhmedov, A remark on the AdS/CFT correspondence and the renormalization group flow,
Phys. Lett. B 442 (1998) 152, hep-th/9806217.
E. Alvarez, C. Gmez, Geometric holography, the renormalization group and the c-theorem,
Nucl. Phys. B 541 (1999) 441, hep-th/9807226.
D.Z. Freedman, S.S. Gubser, K. Pilch, N.P. Warner, Renormalization group flows from
holography-supersymmetry and a c-theorem, hep-th/9906194.
L. Girardello, M. Petrini, M. Porrati, A. Zaffaroni, The supergravity dual of N = 1 super Yang
Mills theory, hep-th/9909047; Novel local CFT and exact results on perturbations of N = 4
super YangMills from AdS dynamics, hep-th/9810126.
M. Porrati, A. Starinets, RG fixed points in supergravity duals of 4-d field theory and
asymptotically AdS spaces, Phys. Lett. B 454 (1999) 77, hep-th/9903241.
V. Balasubramanian, P. Kraus, Space-time and the holographic renormalization group, Phys.
Rev. Lett. 83 (1999) 3605, hep-th/9903190.
K. Skenderis, P.K. Townsend, Gravitational stability and renormalization group flow, hepth/9909070.
O. DeWolfe, D.Z. Freedman, S.S. Gubser, A. Karch, Modeling the fifth dimension with scalars
and gravity, hep-th/9909134.
C. Schmidhuber, AdS flows and Weyl gravity, hep-th/9912155.
J. Polchinski, Renormalization and effective Lagrangians, Nucl. Phys. B 231 (1984) 269.
J. Maldacena, The large N limit of superconformal field theories and supergravity, Adv. Theor.
Math. Phys. 2 (1998) 231, hep-th/9711200.
E. Witten, Anti-de Sitter space and holography, Adv. Theor. Math. Phys. 2 (1998) 253, hepth/9802150.
S. Gubser, I. Klebanov, A. Polyakov, Gauge theory correlators from noncritical string theory,
Phys. Lett. B 428 (1998) 105, hep-th/9802109.
E. Brezin, S. Wadia, The Large N Expansion in Quantum Field Theory and Statistical Physics:
From Spin Systems to Two-dimensional Gravity, World Scientific, Singapore, 1993.
T. Yoneya, Action principle, Virasoro structure and analyticity in nonperturbative twodimensional gravity, Int. J. Mod. Phys. A 7 (1992) 4015.
T.R. Morris, A manifestly gauge invariant exact renormalization group, hep-th/9810104.
S. Hirano, Exact renormalization group and loop equation, hep-th/9910256.
G. Parisi, Y.S. Wu, Sci. Sin. 24 (1981) 484.
V. Periwal, String field theory Hamiltonian from YangMills theories, hep-th/9906052.
G. Lifschytz, V. Periwal, Dynamical truncation of the string spectrum at finite N , hepth/9909152.
M. Li, Y.S. Wu, Holography and noncommutative YangMills, hep-th/9909085.
Nuclear Physics B 579 [FS] (2000) 535560

Replica-deformation of the SU(2)-invariant

Thirring model via solutions of the qKZ equation
Mathias Pillin ,1
Department of Mathematics, Kings College London, Strand, London WC2R 2LS, UK
Received 16 March 2000; revised 12 April 2000; accepted 26 April 2000
Abstract
The response of an integrable QFT under variation of the Unruh temperature has recently been
shown to be computable from an S-matrix preserving (replica) deformation of the form factor
approach. We show that replica-deformed form factors of the SU(2)-invariant Thirring model can
be found among the solutions of the rational sl2 -type quantum KnizhnikZamolodchikov equation
at generic level. We show that modulo conserved charge solutions the deformed form factors are in
one-to-one correspondence to the ones at level zero and use this to conjecture the deformed form
factors of the Noether current in our model. 2000 Elsevier Science B.V. All rights reserved.
PACS: 11.55.Ds; 11.10.Kk
Keywords: Form factors; Replica; Deformation; qKZ equation
Dedicated to the memory of Mosh Flato
1. Introduction
The form factor approach to massive integrable quantum field theories (QFTs) [9,20]
provides a powerful tool to compute exact matrix elements of local operators from the
knowledge of the (factorized) S-matrix. To be precise, a form factor in this context is
the matrix element of some local operator between the physical vacuum and a multiparticle scattering state. These form factors contain all the information about the QFT as
the Wightman functions can in principle be (re-)constructed in terms of them by saturating
with a complete set of scattering states. Alternatively each local operator can be set into
correspondence to a sequence of form factors; a feature that has been used to classify the
full content of local operators in certain models, see, e.g., [19,22].
Current address: ETAS, PTS-A, Borsigstr. 10, D-70469 Stuttgart, Germany.
1 map@mth.kcl.ac.uk
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 2 5 6 - X
536
M. Pillin / Nuclear Physics B 579 [FS] (2000) 535560
For an integrable QFT the form factors can be characterized axiomatically as tensorvalued meromorphic functions obeying a recursive set of functional equations, the socalled form factor equations [20].
As this will be of importance later we anticipate that one of these equations (the cyclic
form factor equation), see Eq. (8) below, is closely related to the BisignanoWichmann
Unruh thermalisation phenomenon [1,26]. In particular the cyclicity parameter = 2 is
both physically and mathematically linked to the Unruh temperature TU = 1/2 (in natural
units), seemingly of a very different origin. This connection that has been used to formulate
the cyclic form factor equation without reference to bootstrap analyticity properties and to
generalize it to non-integrable QFTs [16]. This observation was then the starting point for
the replica-deformation of integrable QFTs in [14] to which we return below.
Among the most prominent non-scalar massive integrable models which have been
solved by the form factor method without thermalisation (i.e., at TU = 1/2 ) are the sineGordon model and the SU(2)-invariant Thirring model [20,21]. We will refer to the latter,
which will concern us in this paper, simply as the Thirring model (even though this clashes
a bit with standard usage).
On the other hand, in an a priori independent direction of research it was shown in a
pioneering work [4] that the classical KnizhnikZamolodchikov differential equation (for
a survey see [3]) which arises in conformal field theory admits a quantum deformation. This
deformation amounts in a consistent system of holonomic difference equation for a tensor
valued meromorphic function, which is referred to as deformed or quantum Knizhnik
Zamolodchikov (qKZ) equation. The authors of [4] already realised that a subset of the
standard form factor equations [20] provide a very specific example of a qKZ equation.
Much work has been done during recent years, to elaborate this connection, see [7,11,13]
and references therein.
In the seminal paper [24] the complete set of solutions of the qKZ equation for a rational
sl2 -type R-matrix has been found. In this case one considers a function (z1 , . . . , zn )
taking values in a tensor product of sl2 -modules V1 Vn , and subject to a system of
difference equations with shift parameter p
(z1 , . . . , zm + p, . . . , zn )
= Rm,m1 (zm zm1 + p) Rm,1 (zm z1 + p)
Rm,n (zm zn ) Rm,m+1 (zm zm+1 ) (z1 , . . . , zm , . . . , zn ).
(1)
Here, Ri,j (z) End(Vi Vj ) is the rational sl2 R-matrix. The more general case of this
equation involving an extra parameter as studied in [24] will not be of importance in this
paper.
In a nut-shell the solutions of (1) are found as generalised hypergeometric integrals.
Subsequently, the construction of [24] has been applied to irreducible representations of sl2
in [12]. It is this case which is relevant here and which will provide our main mathematical
tool.
Returning to form factors, it has been shown in [13] that Smirnovs solutions of the form
factor equations in the Thirring model [20,21] can indeed be extracted from the general
solutions of (1). In particular, Smirnovs results follow if we take only two-dimensional
537
fundamental representations V of sl2 (obviously corresponding to the physical states in

the model) and set p = 2i. This particular choice is referred to as level zero solution of
the qKZ equation. In other words, the form factors of the Thirring model can be identified
with specific solutions of the qKZ equation at level zero.
The physics problem we wish to solve is to construct the replica-deformed form
factors in the sense of [14] for the SU(2)-invariant massive Thirring model.
We shall review some aspects of the replica-deformation in Section 2. Here it may
be sufficient to remark that in view of the aforementioned relation to the Bisignano
WichmannUnruh thermalization, taking p 6= 2i in the qKZ equations amounts to
studying the continuation of the original QFT to some off-critical Unruh temperature.
The quest for such a continuation naturally arises in the broader context of quantum
gravity, see, e.g., [2,5], in particular one is interested in the response of a QFT under an
infinitesimal variation of the Unruh temperature. Of course the problem in the first place
consists in defining the continuation of the QFT in question. For the partition function
Callan and Wilczek [2] proposed a replica prescription to define its continuation. In [14]
more generally the invariance of the S-matrix has been proposed as the defining criterion.
Specifically for an integrable QFT this criterion turned out to uniquely determine the proper
deformation of the form factor equations. In particular there is still a cyclic equation whose
cyclicity parameter p 6= 2i physically corresponds to an off-critical Unruh temperature
and which is equivalent to the qKZ equation at generic (non-zero) level. The main
difference to the ordinary form factor equations lies in the residue equations which
prescribe how solutions with different number of variables are arranged into sequences.
For models with a diagonal S-matrix the qKZ is easy to implement and some replicadeformed form factors have been computed in [14,18]. However, the complete set has not
yet been obtained for any model.
As it was mentioned above the standard form factors of the Thirring model can be found
among the solutions of (1) at level zero. It will be the key point of this paper to show
that the physical problem of construction the replica relatives of them can be solved by
means of the solutions of (1) at generic level. To be a bit more precise we have to arrange
solutions of (1) in sequences such that functions (z1 , . . . , zn ) with different numbers of
arguments are linked by the form factor equations to be described in Section 2.
In particular if the local operator whose form factors are being considered has isospin j ,
n
Vj ,
then in group theoretical terms its n-particle form factor is an intertwiner V1/2
where Vj is the irreducible sl2 -module of isospin j . In other words one has a one-to-one
correspondence
sequence of level zero qKZ solutions
local operator of isospin j
in SU(2) Thirring model

intertwining V n Vj , n > n0
(2)
Here n0 is the starting member of a sequence and the non-vanishing members of a

sequence have all either n odd or n even. For simplicity we shall refer to such a sequence
as an isospin j sequence. The aim in the bulk of the paper is to achieve something similar
for the replica-deformed form factors. Since the replica-deformed system is no longer an
ordinary Poincar invariant QFT, the bootstrap viewpoint becomes even more important.
538
That is the collection of deformed form factors is supposed to define the replica-deformed
system.
Our main result can then be summarized as follows: the right hand side of the
correspondence (2) admits a unique replica-deformation, provided on both sides the
equivalence classes modulo pointwise multiplication with a conserved charge eigenvalue
(i.e., a regular solution of the form factor equations with trivial S-matrix) are being
considered. Symbolically we can write
sequence of level zero
sequence of level p 2i 6= 0
(3)
qKZ solutions of isospin j
qKZ solutions of isospin j
modulo conserved charge eigenvalue
modulo conserved charge eigenvalue
The structure of the conserved charge eigenvalues in both cases is very different and so
is the kinematical arena [14]. However, in view of (3) the structure of the state space in
both theories is very similar as will be shown in Section 5, much in the spirit of the replica
idea [2,14].
The outline of the paper is as follows. In the next section we briefly review the replicadeformation of massive integrable field theories and describe the deformed form factor
equations. In Section 3 we summarise the essentials of the bootstrap description of the
Thirring model and prepare the deformed minimal form factors [18]. In Section 4 we
rewrite the solutions of the rational sl2 -type qKZ equation of [12,24] in a way that facilates
the comparison with the (undeformed) form factors of the Thirring model. It turns out that
the solutions can be parameterised by a polynomial similar to Smirnovs form factors. We
proceed by generalising the completeness proof of the level zero solutions [23] to generic
level. It is essentially this completeness result that underlies the correspondence (3). In
Section 6 we give the result for the deformed form factors, and first show that they satisfy
a subset of the modified form factor equations of [14]. Here we take advantage of the
fact that the replica deformation leaves the bootstrap S-matrix, and hence the structure
of the algebraic Bethe vectors, unaffected. Then we derive conditions on the polynomial
entering the qKZ solutions under which the corresponding form factors in addition satisfy
the modified kinematical residue equations. Finally, using (2) and (3) we give explicit
expressions of some form factors which we conjecture to be the replica-deformed form
factors of the Noether current in the SU(2) Thirring model. The last section is left to a
discussion of the results.
2. Replica deformation of the form factor approach

In this section we review the results of [1416] and state the modified form factor equations. The starting point is the BisignanoWichmannUnruh thermalisation phenomenon
stating that the vacuum of a Minkowski space QFT looks like a thermal state of inverse
temperature = 2 with respect to the Killing time of the Rindler wedge [1,26]. This fact
can be encoded symbolically in the following formula, where Oi (xi ), i = 1, . . . , n, denotes
some local operator with support inside the Rindler wedge W

h0|O1 (x1 ) On (xn )|0i = trHW exp(2K)O1 (x1 ) On (xn ) .
539
(4)
Here K stands for the generator of Lorentz boosts in W . The trace, if it existed, would
be over the state space HW of the Rindler space QFT. The latter, however, has never been
defined for an interacting QFT, and since e2K is an unbounded operator on the original
state space the trace is likely to be meaningless. Eq. (4) should thus be taken only as a
mnemonic.
Suppose now that we want to study the same QFT with the parameter shifted away
from 2 , a problem which naturally arises in the broader context of quantum gravity, see,
e.g., [2,5]. Formally this amounts to replace
/2
= exp(K),
(5)
exp(2K) exp(2K)
in (4) while keeping everything else fixed. In the case of an integrable QFT we can require
specifically that the factorized S-matrix of the model is unchanged upon (5). This can
be viewed as a concrete realisation of the replica idea of Callan and Wilczek [2]. In
particular, if one could make sense out of the deformed correlator (4), (5), its response
under an infinitesimal variation of the Unruh temperature could be obtained simply by
differentiation

trHW exp(2K)O1 (x1 ) On (xn )

.
(6)
=2
The crux of the problem of course is to define the deformed correlators consistent with
the replica idea (5) and in a way that avoids mathematical ambiguities. For example naively
regularizing the trace in (4) would most likely lead to enormous technical problems with
renormalizability since the translation invariance of the original QFT is lost for 6= 2 .
The solution proposed in [14] is to implement the replica idea on the level of form factors.
A form factor in this context is the matrix element of a local operator O(x) between the
vacuum and an n-particle scattering state. The scattering states depend on rapidities zi and
are created from the vacuum by means of FaddevZamolodchikov-type operators Aai (zi ),
where the index ai refers to the charge of a state. The form factor is then
Fa1 ...an (z1 , . . . , zn )

= h0|O(x)|z1 , . . . , zn ia1 ...an trHW e2K O(x)Aa1 (z1 ) Aan (zn ) ,
(7)
where we are again using our trace-mnemonic (4). The trace on the right-hand side here is
now even more problematic than in (4) because the As are operators acting on the physical
Hilbert space of the original QFT, not on HW . (In some approaches to the qKZ equation
similarly looking traces over vertex operators acting on the Fock spaces of auxiliary free
bosons are employed. These traces are, of course both mathematically and conceptually
distinct from the ones indicated in this paper.) Nevertheless, the form factors do enjoy a
certain cyclicity property as if they were traces, a fact that has been derived from general
quantum field theoretical principles and linked to the thermalisation phenomenon (4)
in [16]. Evidently the replica idea (5) can be applied to (7) as well, and here it can be
made operational. This is because for the replica deformed form factors a modified system
of form factor equations exists. It is uniquely determined by (5) and the requirement that
540
the S-matrix in unaffected by the deformation. For fixed n the first two equations are as
follows [14]:
Fa1 ,a2 ...an (z1 + i, z2 , . . . , zn ) = Fa2 ...an ,a1 (z2 , . . . , zn , z1 ),
(8)
0
an1
an0
0
Fa1 ,...an ,an1 (z1 , . . . , zn , zn1 ) = San1
an (zn1 zn )Fa1 ,...,an1
an0 (z1 , . . . , zn1 , zn ).
cd (z) is the factorised scattering operator. If this scattering operator has something
Here Sab
to do with an R-matrix, it is not difficult to see that these two equations are equivalent to a
qKZ equation (1) at generic level if we identify p and in an obvious way by p = i.
For = 2 we recover the level zero situation or the standard Watson equations for form
factors [20]. The parameter is a phase, which may be different from = 1 for certain
local operators of the model, see [20,21].
The particular solutions of (8) searched for are according to [14] supposed to have simple
poles at zn zn1 = i mod i. Thus in essence we are looking for a certain subset of
solutions of the qKZ equations. The residues at these poles are determined by the following
equations which link solutions of (8) with n and n 2 variables
Fa1 ,...an1 ,an (z1 , . . . , zn1 , zn ) = iCan1 an Fa1 ,...an2 (z1 , . . . , zn2 ).
(9)
C an1 an Fa1 ,...an1 ,an (z1 , . . . , zn1 , zn ) = iFa1 ,...an2 (z1 , . . . , zn2 ).
res
zn =zn1 +i
res
zn =zn1 i
Here Cab is the charge conjugation matrix and the normalisation is such that the inner
product of two one-particle states is a hz1 |z2 ib = 2ab (z1 z2 ).
Note that the residue equation for zn = zn1 + i is similar to the condition satisfied
by traces of vertex operators in lattice models [7,8]. Without the second equation (9)
however the recursion n 2 n would be highly ambiguous. Taking advantage of the
qKZ equation to implement the analytic continuation zn zn + i the second equation (9)
can be rewritten as follows
res
zn =zn1 i+i
Fa1 ...an (z1 , . . . , zn1 , zn )

c
n1 n2
= Fb1 ...bn2 (z1 , . . . , zn2 )Can c Sccb1 a11 (zn1 z1 ) San1
an2 (zn1 zn2 ). (10)
This form turns out to be more convenient later. Observe that (10) looks like the second
term in the ordinary = 2 residue equation. For 2 the poles at zn = zn1 + i and
zn = zn1 i + i merge and the residues add up, so that the ordinary residue equation
of [20] is recovered.
Eqs. (8) and (9) are the replica deformed form factor equations. As explained in the
introduction for a given S-matrix we take their solutions to define the replica deformed
system. The variation of of away from 2 then has wide ranging physical consequences.
First of all the standard translation invariance is broken while Lorentz symmetry is
maintained. Quite generally the kinematical properties of the 6= 2 models are drastically
changed. For example it has been shown in [14] that the mass eigenvalues of asymptotic
states are altered in a way such that it costs more energy to boost two particles relative to
each other than in the = 2 case. Moreover, under certain circumstances not the entire
541
momentum phase space is accessible to the particles. This phenomenon is close in spirit to
t Hoofts picture of scattering states subject to quantum gravitational transmutation [6].
3. The SU(2)-invariant Thirring model

The aim in the rest of the paper is to study the replica-deformation of the (1 + 1)dimensional SU(2)-Thirring model in terms of its form factors. In this section we prepare
the basic ingredients for its bootstrap description. For the sake of orientation however let
us also display the classical Lagrangian. It is given in terms of a two-component spinor ,
valued in the fundamental representation of SU(2)

a
a .
g
(11)
L = i
Here a are the Pauli matrices. The QFT associated with this model exhibits dynamical
mass generation and is believed to be asymptotically free. The physical particles of the
theory are an SU(2) doublet of massive kinks. The index {} refers to either of these
kinks.
Moreover it is known that the model is integrable, and the factorised S-matrix is given
as a function of the kink-rapidity z by the following expression
0 0
S1122 (z) = S0 (z)
0 0
0 0
z11 22 + h12 21
,
z+h
h = i.
(12)
We use the abbreviation h = i as indicated in (12) from now on whenever suitable in

accordance with [13]. The scalar part S0 of the S-matrix is given by
S0 (z) =
( 12
( 12 +
z
z
2h ) ( 2h )
.
z
z
2h ) ( 2h )
(13)
Without S0 the expression in (12) coincides with the rational sl2 -R-matrix in the defining
representation [12,24].
We also note that the charge conjugation matrix associated with (11) which enters the
form factor equations (9) and (10) is given as follows
C = i 2 .
(14)
The ordinary form factors associated with this S-matrix can be found in [20]. They
have two features one expects to be present also in the deformed case. First the tensor
structure and second the appearance of a product of so-called minimal form factors in the
solutions. Concerning the tensor structure we recall that an n-particle form factor (7) can
be indexed by the kinks and anti-kinks of the scattering state and hence takes values in an
n-fold tensor product V n of fundamental representations of SU(2). On physical grounds
one is interested in solutions which ultimately can be interpreted as the form factors of
a local operator with a definite isospin j . Mathematically speaking, these operators are
intertwiners V n Vj to the irreducible representation of SU(2) of isospin j . The space
of these intertwiners has exactly the same dimension as the subspace of singular vectors
542
in a given weight space (V n )l V n , which we will denote by (V n )l . Let us define

P
these spaces. The Pauli matrices in the tensor product space are given by a = ni=1 ia .
We can then define the weight subspaces

(15)
(V n )l = v V n | 3 v = (n 2l)v = 2j v .
sing
As it was said before we can restrict our attention to solutions (z1 , . . . , zn ) of the
sing
qKZ equation valued in the space of singular vectors in (V n )l , i.e., to those satisfying
+
(z1 , . . . , zn ) = 0. This is related to an unbroken SU(2)-symmetry. Then 2l 6 n and all
other components in an irreducible SU(2)-multiplet of isospin j = n/2 l can be obtained
by acting with on the singular vector.
To make the notation comparable to [12,13] we map the indices a1 , . . . , an of the form
factor (7) to a set M as follows. Let M = {m1 < m2 < < ml } {1, 2, . . . , n}, such
that the number of elements of M is #M = l. In other words this set labels the positions at
which the form factor (7) has kinks of type in the scattering state, i.e., M = {i|i = }.
Hence, we can identify vectors in V n as
vM := v1 vn .
(16)
Naturally we shall employ the weight decomposition (15) and the labelling (16) also in
the deformed case.
Further we prepare here the deformed counterpart of the minimal form factor. This is
(up to factors coming from representation theory and the pole structure) the two particle
form factor as explained in [18]. The deformed minimal form factor f (z) is a solution of
the functional equations
f (z) = S0 (z)f (z),
f (z + i) = f (z).
(17)
The solution to these equations were found in [18] in terms of Barnes Digamma functions.
Let complex numbers 1 , 2 be such that Re i > 0 for i = 1, 2. We define the Digamma
function [8,18] by the Hankel contour integral as
Z
1
exp(xt)(log(t) + ) dt
,
(18)
log 2 (x|1 , 2 ) =
Q2
2i
t
i=1 (1 exp(i t))
where denotes the Euler constant. Note that the periods 1 , 2 enter symmetrically
in (18). The Barnes periodicity of this function is due to the relation
1
2 (x + 1 |1 , 2 )
=
,
2 (x|1 , 2 )
1 (x|2 )
where 1 can be identified with the usual function as
1/2+x/2
(x/2 )/ 2.
1 (x|2 ) = 2
(19)
(20)
The relevant solutions of (17) are then given by

2 ( iz|2, )2( + + iz|2, )
.
(21)
2 (iz|2, )2( + iz|2, )
Note that apart from the higher periodicity the structure of (21) is similar to the one of the
scalar S-matrix (13). This feature was observed to be common for a large class of integrable
f (z) =
543
field theories in [18]. The minimal form factor (21) was given an interpretation in terms
of central extensions of Yangian doubles in [10]. As remarked before we can identify the
inverse Unruh temperature with the shift parameter p of the qKZ equation (1) via
p = i.
(22)
We find it convenient to use either of these parameters throughout the paper, keeping the
above identification in mind. For = 2 one recovers using (18) from (21) up to trivial
factors the minimal form factor (z) used by Smirnov in [20,21].
For later use we conclude this section with the following identities among the minimal
form factors
1
,
1 (iz|)1( + iz|)
S0 (z)
.
f (z)f (z + h + i) =
1 (iz|)1( iz|)
f (z)f (z h) =
(23)
4. Solutions of the rational qKZ equation

It was mentioned in the previous section that up to the scalar factor the S-matrix
of the Thirring model is just the rational sl2 -R-matrix in the defining representation. This
means that the first two functional equations (8) to be satisfied by the form factor (7) can,
up to scalar factors which turn out to be the minimal form factors be identified with
the solutions of the qKZ equations with rational R-matrix as given in [12,24]. Obviously
our case is particularly simple in this respect because we have to work with only one type
of representation of sl2 .
Below we first recall the solutions of [12,24] at arbitrary level and then proceed by
rewriting them in a form suitable for our purposes. The new form readily allows one to
make contact to the level zero solutions [13,20,21] in the limit p 2i, and secondly
allows for a transparent procedure to single out those qKZ solutions which in addition
satisfy the modified kinematical residue equations (9) and (10).
We start by defining the rational hypergeometric space of [24] which will carry the
information on the number and position of the kinks and anti-kinks of the form factor (7).
This space consists of rational functions with at most simple poles and a certain asymptotic
behaviour [24]. We define
l
Y ta zk +
Y
1
.
(24)
gM (t1 , . . . , tl ) =
ta zma
ta zk
a=1
16k<ma
Let f = f (t1 , . . . , tl ) be a function. Let be an element of the symmetric group Sl .

Following [12,24] we define a special symmetrisation in that for a simple transposition
(a, a + 1) Sl we set
[f ](a,a+1)(t1 , . . . , ta , ta+1 , . . . , tl ) := f (t1 , . . . , ta+1 , ta , . . . , tl )
ta ta+1 + h
.
ta ta+1 h
(25)
544
Using this symmetrisation one can now introduce the following basis in the rational
hypergeometric space
X
(26)
gM (t1 , . . . , tl ).
w M (t1 , . . . , tl ) =
Sl
It is clear that in order to make our results comparable with [20,21] we have to get rid of
the unusual symmetrisation occurring in (26).
An important object in the theory qKZ equations is the phase function. This object
provides the link between the analytic structure of the solutions and the geometry of local
systems [3,24]. In the conventions of [24] the phase function is given by:
b 1 , . . . , tl ; z1 , . . . , zn ) =
(t
n
l Y
Y
a zj ; )
(t
a=1 j =1
a tb ; h),
(t
(27)
16a<b6l
where
((x + )/p)
.
(x;
) :=
((x )/p)
(28)
We refrain from describing the connection coefficients arising from (27), details can be
found in [24].
The third and last entity required to describe the solution space of the rational qKZ
equation is the trigonometric hypergeometric space Fq . The functions in this space are of
the form
P (1 , . . . , l ; 1 , . . . , n )
Y
16a<b6l
n
l Y
Y
exp(i(zj ta )/p)
sin((ta zj )/p)
a=1 j =1
sin((ta tb )/p)
.
sin((ta tb h)/p)
(29)
Here we have set a = exp(2ita /p) and j = exp(2izj /p). P is a polynomial with
complex coefficients and symmetric in the variables t1 , . . . , tn . According to [24] we can
restrict the partial degree in these variables in order to match the dimension of the space
of such polynomials P to the dimension of the required representation space of sl2 . We
are not going to describe this here but we will show explicitly in the next section how the
completeness of the qKZ solutions in the case of interest for this paper is directly encoded
into the properties of P . Moreover, the local operators in the model can be classified in
terms of this object as will be explained in Sections 6 and 7.
With these preparations at hand we can now write down the space of solutions of the
qKZ equation for a function with values in V n in terms of an integral representation. Let
W = WP be a function from the space Fq defined in (29). The integrand is given by
b 1 , . . . , tl )WP (t1 , . . . , tl ),
I (w M , WP ) = w M (t1 , . . . , tl )(t
(30)
where we omitted the dependence of I (w M , WP ) on z1 , . . . , zn . The solution of the

equation (1) with values in representations indicated in the previous section is then given
by:
W (z1 , . . . , zn ) =
X Z
#M=lC
545
dt1 dtl I (w M , WP )vM .
(31)
Cl
We specify the integration cycles Cl below after having rewritten this solution in a more
convenient form. It has been shown in [12,24] that W is a solution to the qKZ equation.
We will indicate how to check this after having introduced a basis in the solution space
which is more appropriate to the aim of this paper. We stress again that the information
on the space of representations of sl2 is contained only in the function w M . Physically
speaking this means that once we have fixed the pattern of kinks and anti-kinks in the
candidate form factor, the only freedom of choice resides in the polynomial P .
Eventually we are interested in constructing the form factors of the replica-deformed
SU(2)-invariant Thirring model. To this end it is useful to rewrite the solution spaces (31)
in a form which allows a direct comparison with the standard level zero form factors as
they have been given in [13,20,21]. In particular we need to get rid of the non-standard
symmetrising operation (25) for the rational weight function (26). However, the alterations
to be performed do not change (up to an overall normalisation) the solution (31).
We begin by replacing some objects appearing in (31) as follows:
ta zj ta zj + h,
a, j ;
h/2.
(32)
This replacement will be in effect from now on.

Next, we replace a part of the phase function (27) which does depend only on differences
of the variables t as follows

h (ta tb )
1
(33)
(ta tb h) sin (ta tb h).

ta tb h
p
p
p
As it is implicit already in [13] we can use this to replace the non-standard symmetrisation (25) by a standard one
Y

(ta tb h) = Asym gM (t1 , . . . , tl ) , (34)
wM (t1 , . . . , tl ) := w M (t1 , . . . , tl )
16a<b6l
where
gM (t1 , . . . , tl ) = g M (t1 , . . . , tl )
(ta tb h),
(35)
16a<b6l
and Asym denotes the standard antisymmetrisation with respect to the variables t1 , . . . , tl .
This means that upon performing the steps outlined above, we can work with the same
rational weight function as in the level zero case [13]. The same feature has been used
in [11] for determinant formulas in the case of the trigonometric qKZ equation.
As will be explained below, a natural consequence of this is that the (algebraic) Bethe
vectors in the -deformed case will be the same as in the standard level zero case.
In addition we can also cancel the pure t-dependent part in the denominator of (29). This
has in particular the remarkable consequence that the space Fq introduced in (29) almost
coincides with the corresponding space introduced in [13,23] in the level zero case. We
shall exploit this fact below.
546
Next we reorganise the remaining parts of the integrand in (31). This will be done in
a way to point out the similarity between (31) and trace formulas of vertex operators as
developed in [7,8,10]. To this end we first define the analogue of the phase function used
by Smirnov [20,21]

p 3h/2 y
h/2 + y
i
(36)
exp y .
(y) :=
p
p
p
Note that here we put in the exponential factor from (29) just as a matter of convenience.
Then we define an odd function necessary to guarantee the complete symmetry of the
integrand with respect to t

h+y
hy
1
y .
(37)
sin
(y) := 3
p
p
p
p
For later use let us note the identity
(y)(y + h) =
exp(2i(y + h/2)/p)
.
(y + 3h/2)(y + h/2) sin p (y + h/2) sin p (y + 5h/2)
(38)
Observe that this expression (up the the exponential factor) tends to the corresponding
identity for the phase functions in [20,21] in the case 2 .
Having now collected all the data we can write down the solution of the qKZ equation
in a new form, which is, however, up to a normalisation identical with (31). Since now the
only degrees of freedom of the solution reside in the polynomial P of (29) we prefer to
label the solution by this object
Z
X Z
wM (t1 , . . . , tl )P (1 , . . . , l ; 1, . . . , n )
P (z1 , . . . , zn ) =
#M=l C
n
l Y
Y
Cl
(ta zj )
a=1 j =1
(ta tb )dt1 dtl vM .
(39)
16a<b6l
The singular hyperplanes of the integrand in (39) are

ta = zj h/2 + Z60 p,
ta = zj + p 3h/2 + Z>0 p,
ta = zj 3h/2,
depending on M,
(40)
ta = tb h + Z60 p,
ta = tb + h + Z>0 p.
Note that the comment in the third line means that this pole may occur only in the basis
function wM and hence its presence depends on the indexing set M.
This now allows us to specify the integration contours as follows. For any a = 1, . . . , l
we let the integration contour Ca separate:
(i) the hyperplanes of the first line from the ones of the second and third line in (40),
(ii) the hyperplanes of the fourth line from the ones in the fifth line of (40).
547
This choice is convenient for our purposes, but as it has been indicated in [12,24] the
solution space is by analytical continuation actually independent of the particular choice
of the integration contour.
We are now going to check that (39) is a solution of the rational qKZ equation. Of course
this is clear by construction [24]. However, this gives us the opportunity to introduce some
new objects on the way and to illustrate nicely their use, as this will be of importance later.
A fundamental role in [24] is played by discrete derivates with respect to the variables t
and z, respectively. We describe these derivatives within the modified form of the solution
space (39). Therefore we introduce shift operators:
Zj f (t1 , . . . , tl ; z1 , . . . , zn ) := l+j (t, z)f (t1 , . . . , tl ; z1 , . . . , zj + p, . . . , zn ),
Qa f (t1 , . . . , tl ; z1 , . . . , zn ) := a (t, z)f (t1 , . . . , ta + p, . . . , tl ; z1 , . . . , zn ),
(41)
where
l+j (t, z) =
a (t, z) =
l
Y
ta zj + 3h/2 p
,
ta zj + h/2 p
a=1
l
Y
b=1,
b6=a
n
ta tb + h Y ta zj + h/2
,
ta tb + p h
ta zj + 3h/2
(42)
j =1
are the connection coefficients of [12,24] adjusted to the choice of the solution (39).
With these objects we can define discrete partial derivatives, using the convention that
Da for a = 1, . . . , l acts on variable ta while Dl+j for j = 1, . . . , n acts on variable zj
Da = 1 Qa ,
Dl+j = 1 Zj .
(43)
The proof that (39) is a solution of the rational qKZ equation is then parallel to the one
given in [13] for the level zero case. We state the necessary steps here because we will need
this result later for the solution of the first two form factor equations (8).
0 0
0 0
Following (12) and (13) let R(z)11 22 = S(z)11 22 S0 (z)1 , and f(a)
1 ,...,n for a = 1, . . . , l be
a family of functions which can be expressed in terms of the rational hypergeometric basis
functions (34). The qKZ equation for functions of the form (39) valued in V n is then
equivalent to the following equations on the rational hypergeometric basis functions (34),
taking into account the definition of the set M at the end of Section 3.
We state the first one in z1 only, but its generality is obvious
Z1 w1 ,2 ,...,n ( ; z1 , z2 , . . . , zn ) w2 ,...,n ,1 ( ; z2 , . . . , zn , z1 )
l
X
Da f(a)
.
=
1 ,...,n
(44)
a=1
The second equation refers to the exchange of two z variables

w1 ,...,i+1 ,i ,...,n ( ; z1 , . . . , zi+1 , zi , . . . , zn )
0
X
0 i+1
Rii i+1
(zi zi+1 )w1 ,...,i0 0 ,...,n ( ; z1 , . . . , zi , zi+1 , . . . , zn ).
=
10 ,20 =
We then have the following
i+1
(45)
548
Theorem 1. The function (39) valued in the tensor product V n is a solution of the
rational sl2 qKZ equation.
Proof. For the proof it is sufficient to recall that the hypergeometric integral of [12,24]
as defined in (39) has exact hypergeometric forms (cf. [24] Lemma 2.21), which are
essentially given by the derivatives Da defined in (43). 2
5. Completeness of the solutions
Here we prove the completeness of the solutions (39) of the qKZ equation. On the basis
of form factors in the sine-Gordon model the completeness of the solutions was shown
by Smirnov [22] in establishing a bilinear relation among the solutions of the form factor
space. Later is was shown by Tarasov [23] that (at least for the rational qKZ equation)
similar relations can be extracted from the solution spaces of the qKZ equation directly.
In [23] the fact was used that for the form of the qKZ equation (1) studied in this paper,
there exist zero solutions. In our language this means that there exist polynomials P in (39)
such that the integral vanishes automatically. This guarantees that the dimension of the
solution space matches the dimension of (V n )sing rather than that of V n . For a detailed
presentation of this topic, see [24].
In this section we are going to derive a completeness relation for the solutions of the
qKZ equation at generic level, similar to the one in the level zero case [22,23]. Since we
modelled our solution parallel to the level zero case [13,23] we can expect that the proofs
here will be quite similar to the ones given in [13,23]. The main difference is due to the
more complicated pole structure of the integrand (40).
We begin by noting the dimension of the
space Fq introduced in (29); it is finitedimensional with dim Fq = dim(V n )l = nl .
Let us first consider the case l = 1 in the integral (39), which means in particular that
there are no -functions present. Insert for the polynomial P either of the two choices
Ps1 (t1 ) = e2ihn/p
n
Y
j =1
Ps2 (t1 ) =
n
Y
j =1
exp
(t1 zj + 3h/2),
p
or
(46)
exp (t1 zj + h/2).

p
If we ignore for the moment that the partial degree in exp(2it1 /p) is not as it may have
been assumed in (29) it is not difficult to check that the first expression cancels all possible
poles appearing in the second line of (40). Closing the contour C1 in (39) such that it
encircles the poles in question, we realise that the integral vanishes.
The same kind of argument applies to the second choice in (46). Here we close the
contour C1 such that it contains all possibles poles appearing in the first line of (40).
Hence we have established that for l = 1
Ps1 = 0 = Ps2 .
(47)
549
This result is the generalisation of Lemma 5.3 in [13] to the generic level case.
Let us proceed to the case of l = 2. According to [23] we set
(1) (t) = Ps1 (t) Ps2 (t).
(48)
From the partial degree point of view this is a proper candidate for the polynomial P . It is
clear from (47) that we also have
(1) = 0.
(49)
Let us give an interpretation of this object. According to [24] it is possible to construct basis
functions similar to (34) for the trigonometric hypergeometric space as defined on (29).
If we take the definition (2.26) of [24] and rewrite the basis according to our form of the
solutions (39) of the qKZ equation, we find that these basis functions for l = 1 and variables
z1 , . . . , zn can be recast as a polynomial
Y

e2i(t zj +)/p 1
P (t) = exp 4i(n )/p
16j <
2i(t zj )/p

1 .
(50)
<j 6n
Note that we still keep the convention (32). We can formally extend the definition (50) also
to the case = 0.
Then it is straightforward to check that
n
Y
n

X
Pj (t).
e2i(t zj +)/p 1 = P0 (t) + e4i/p 1
j =1
(51)
j =1
The trigonometric hypergeometric basis functions for l = 2 in their polynomial form

(appropriate to the conventions laid out in the previous section) can for < be
constructed from (50) as

1

Asym P (t1 )P (t2 ) sin (t1 t2 h) .

(52)
P (t1 , t2 ) = sin (t1 t2 )
p
p
Having seen the structure of the basis functions, we can now write down a polynomial (2)

1
(2) (t1 , t2 ) = (1) (t1 ) (1) (t2 ) sin (t1 t2 )

p

sin (t1 t2 h) sin (t2 t1 h) .

(53)
p
p
Taking into account the pole structure (40) it is possible along the lines in the l = 1 case to
check that for l = 2 we have
(2) = 0.
(54)
The functions (1) and (2) are the analogs of the kernel functions at level zero as
introduced by Tarasov in [23]. Therefore the construction of the graded spaces in Section 7
of [23] can be applied to the case of generic level as well.
550
This means that we have now found the necessary set of kernel polynomials for the
solutions of the qKZ equation. The function (2) is the analog of what was found in [22]
as a consequence of the deformed Riemann bilinear identity at level zero.
We say that the arguments z1 , . . . , zn are in generic position, if the singular hyperplanes
in (40) do not coincide. For the case when they do, see Section 6.
The main point now is that the kernel solutions (49) and (54) reduce the actual dimension
sing
of the parameter space from dim(V n )l to dim(V n )l . This can be established using
arguments absolutely parallel to the ones employed in the proof of Theorems 4.3 and 4.4
in [23]. This then gives rise to the following
Theorem 2. For generic values of the arguments z1 , . . . , zn , the solutions (39) span the
sing
space (V n )l .
Remark. This result is crucial for the correspondence (3) between the ordinary and the
deformed form factors. We shall return to it after Theorem 4.
6. The replica-deformed form factors

In this section we define the replica-deformed form factors and determine under which
conditions on the objects P they satisfy the modified form factor equations of Section 2.
According to what has been said in Section 3 we can encode the index structure of the form
factor (7) as
F1 ,...,n (z1 , . . . , zn ) = FM (z1 , . . . , zn ).
(55)
P (z , . . . , z ) the summand corresponding to the basis element w in the

We denote by IM
1
n
M
expression for the solution P of the qKZ equation, as stated in (39).
Recall that the form factors of the standard Thirring model [20,21] are (cf. [13]) special
solutions of the rational sl2 qKZ equation at level zero multiplied by a scalar function. We
will now show that the form factors in the replica-deformed Thirring model in the sense
of [14] naturally arise from the solutions of the rational sl2 qKZ equation at generic level.
Specifically we will show that for a suitable choice of the objects P the functions
Y
cl
P
f (zi zj ) IM
(z1 , . . . , zn ),
(56)
FM (z1 , . . . , zn ) =
(2i)l
n>i>j >1
are solutions of the deformed form factor equations (8)(10). Note that we take 2l 6 n,
later we shall see that j = n/2 l can be identified with the isospin of the underlying local
operator. For the constant cl we take:
l
(57)
cl = 2h(3h/2) (h/p)()/ eih/(2p) .
The fact that (56) is a solution to the first two form factor equations (8) can be proved
without problems. Due to its importance we state this fact as the following
551
Theorem 3. The function FM (z1 , . . . , zn ) as defined in (56) satisfies the form factor
equations (8).
Proof. Take the identities (17) for the minimal form factor f (z) to realize that the factor
Q
n>i>j >1 f (zi zj ) saturates the scalar part of the S-matrix (12) in the form factor
equations (8). The theorem in Section 4 together with (44) and (45) establishes the
result. 2
Remark. If we stick to the definition of the form factor given in (56), which means that we
keep the the requirements set up earlier on the polynomial P in (29), then FM satisfies (8)
with = 1. There are, however, situations where we have to weaken the requirements on P
depending on the local operator we are going to describe with this object. For example in
the next section we will see an example, where we relax the p-periodicity in the variables
z of P . In this case may be different from unity but still satisfies || = 1.
We now come to the part in the form factor business going beyond the qKZ equation.
Namely, we have to single out those solutions described in Theorem 3, which do satisfy
the equations (9).
Our strategy to prove this will be similar to the one used by Smirnov [20,21]. In our
case this means the following. In [20,21] the source for the poles zi = zj h were taken
to reside entirely in the algebraic Bethe vectors M . Additional contributions arising from
the residue evaluation of the integrals (39) at level zero, the origin of which is the pinching
of integration contours, are removed by a trick.
The algebraic Bethe vectors M were shown in [13] (Lemma 6.4) to descend directly
from the rational basis functions wM . Now we have seen that at generic level by (34) we
can work with the same basis functions as in the level zero case.
This essentially reflects the replica-understanding of [2] in varying the Unruh temperature in an integrable quantum field theory as sketched in Section 2. There we have seen
that this deformation changes many aspect of the physics but preserves the S-matrix of the
usual model. And as the S-matrix is unchanged, the algebraic Bethe ansatz [21] remains, of
course, unchanged. We could now outline along the lines of [13] how to construct the Bethe
vectors from the the solution of the qKZ equation (39). However, this is quite technical and
in addition we would like to show that the kinematical residue equations (9) and (10) can
be solved using the basis functions wM directly and using the singular hyperplanes of the
hypergeometric integral (40) only (mainly because the other way is by Smirnovs work
very well understood).
For that purpose we simply state the now obvious but physically important
Lemma 1. The replica deformation of the SU(2)-invariant Thirring model has no effect
on the algebraic Bethe ansatz.
We can now proceed to find solutions of the kinematical equations (9) and (10). We first
collect some results on the residues of the functions wM (t1 , . . . , tl ; z1 , . . . , zn ) as defined
552
P below. For the proofs to

in (34) as they will be needed to evaluate some of the integrals IM
be given below it is sufficient to consider only poles in the variables zn1 and zn . Therefore
we prefer not to write down most of the formulas in the sequel in full generality, but rather
stick to the sufficient case.
For both n 1, n
/ M or n 1, n M we find that
res
tl =zn 3h/2
wM (t1 , . . . , tl ) = 0 =
res
tl =zn1 3h/2
wM (t1 , . . . , tl ).
(58)
The next case is when n M but n 1

/ M. We set N = M \ {n}
res
tl =zn 3h/2
wM (t1 , . . . , tl ; z1 , . . . , zn )|zn =zn1 h
= 2wN (t1 , . . . , tl1 ; z1 , . . . , zn2 )
n2
Y
k=1
res
tl =zn 3h/2
l1
zn1 zk 2h Y
(ta zn1 + 3h/2),
zn1 zk h
a=1
wM (t1 , . . . , tl ; z1 , . . . , zn )|zn =zn1 +h = 0.
(59)
If we take the residue with respect to zn1 rather than zn , we get:

res
tl =zn1 3h/2
wM (t1 , . . . , tl ; z1 , . . . , zn )|zn =zn1 h
= wN (t1 , . . . , tl1 ; z1 , . . . , zn2 )
n2
Y
k=1
res
tl =zn1 3h/2
l1
zn1 zk h Y
(ta zn1 + h/2),
zn1 zk
a=1
wM (t1 , . . . , tl ; z1 , . . . , zn )|zn =zn1 +h
res
tl =zn1 3h/2
wM (t1 , . . . , tl ; z1 , . . . , zn )|zn =zn1 h .
(60)
At last we treat the case of n

/ M but n 1 M. Set N 0 = M \ {n 1}. Here we find that
restl =zn 3h/2 wM (t1 , . . . , tl ) = 0, while
res
tl =zn1 3h/2
wM (t1 , . . . , tl ; z1 , . . . , zn )
= w (t1 , . . . , tl1 ; z1 , . . . , zn2 )

N0
n2
Y
k=1
l1
zn1 zk h Y
(ta zn1 + h/2),
zn1 zk
(61)
a=1
which is obviously independent of the pole structure.

/M
Lemma 2. The form factor (56) has a KR pole zn = zn1 h if n 1 M and n
and arises from the singular hyperplanes ta = zn1 3h/2.
/ M and n M and arises
The form factor (56) has a KR pole zn = zn1 + h if n 1
from the singular hyperplanes ta = zn 3h/2.
There are no KR poles if both n 1, n M or n 1, n
/ M.
Proof. In (40) we have collected the singular hyperplanes of the integral (39). We could
think of evaluating the integrals by means of the Leray residue method. Having done this
one realises that the source for poles of the kind zn = zn1 h or zn = zn1 + h can only
reside in expressions arising from the phase function (36). Now a kinematical pole of the
553
first type appears if we take ta = zn1 3h/2. This singularity shows up if it is present in
wM (34) and hence if n 1 M. We have to check that if this is the case the presence of
n in M leads to a vanishing contribution. This is easy to see because, e.g.,

wM z =z h = 0.
(62)
res
tl =zn1 3h/2
n1
The proof of the second statement is similar. The last statement is then a consequence
of (58) and the first two statements in Lemma 2.
We can now check under which conditions of P the form factor satisfies the kinematical
residue equations (9) and (10) in the thermalised case.
First of all it is clear from Lemma 2 that if both n 1, n M or n 1, n
/ M the
residue of FM at the points in question vanishes. This means that in this case FM satisfies
the equations (9) and (10) since the charge conjugation matrix C vanishes here.
Next, we mention that due to (40) for the other cases discussed in Lemma 2, which
means (n1 , n ) = (, +) or = (+, ) we do not encounter the pinching of integration
contours which occurs in the case of standard form factors in the model [21].
The verification of the residue equation (9) for zn = zn1 h is more or less
straightforward. We use the proposition of this section and evaluate the multiple integral
P at t = z
IM
l
n1 3h/2. This pole is present for (n1 , n ) = (, +) and for (n1 , n ) =
(+, ). Then we take the residue of the remaining expression at the point in question and
use the identities (38) and (23).
Let us state that we set = 0 if (n1 , n ) = (, +) and = 1 if (n1 , n ) = (+, ),
and that from now on we write the parameter P in (56) as Pl,n in order to indicate its
dependence on the variables t and z. The result of the procedure outlined before is then
FM (z1 , . . . , zn )
Y
f (zi zj )
= (1)
res
zn =zn1 h
n2>i>j >1
dt1
C1
Cl1
n2
l1
YY
dtl1 wN (t1 , . . . , tl1 ; z1 , . . . , zn2 )Pl,n | tl =zn1 3h/2

zn =zn1 h
(ta zj )
a=1 j =1
l1
Y
a=1
cl1
(2i)l1
l1
Y
(ta tb )
a<b
n2
Y
ei(zn1 zj 3h/2)/p
j =1
e2i(ta zn1 +h/2)/p
sin p (ta zn1 + h/2) sin p (ta zn1 + 5h/2)
(63)
This in turn means that FM satisfies the first kind of kinematical residue equations if

l1
Y

e2i(ta zn1 +h/2)/p
Pl,n
tl =zn1 3h/2
sin p (ta zn1 + h/2) sin p (ta zn1 + 5h/2)
a=1
zn =zn1 h
n2
Y
i(zn1 zj 3h/2)/p
j =1
Pl1,n2 .
(64)
554
The means that this is an equality up to a phase, which is permitted by the requirements
on the form factors.
To check the second type of kinematical poles at zn = zn1 + h, which is at least not
explicitly necessary in the level zero case, we take a different strategy. Rather than to
take the residue directly at this point we perform an analytic continuation and evaluate
the residue at zn = zn1 + p h. The fact that the corresponding residue equations are
equivalent is shown in cite [14]. In addition it may be nice to see how the S-matrix factors
in (10) come about.
We use the property (44) which was used in the construction of solutions of the qKZ
equations in Section 3. Up to total derivatives we have
w1 ,...,n1 ,n ( ; z1 , . . . , zn1 , zn ) Z1 wn ,1 ,...,n1 ( ; zn , z1 , . . . , zn1 ).
(65)
Here Z1 acts on the first index n associated with the variable zn .

Now we move zn back to the place n 1 using the R-matrix property of the rational
hypergeometric basis functions according to (45). Hence we get up to total derivatives
w1 ,...,n1 ,n ( ; z1 , . . . , zn1 , zn )
0
n2
n2
(t, z)R11n1 (z1 zn p)R2212 (z2 zn p) Rn2
n3 (zn2 zn p)
w0 ,...,0
n2 ,n2 ,n1
( ; z1 , z2 , . . . , zn + p, zn1 ).
(66)
P with this replacement in mind at t = z + p

Hence we can now evaluate the integral IM
l
n
3h/2 much in the same way as above.
The result then is
res
zn =zn1 +hp
FM (z1 , . . . , zn )
Y
Y
cl1
S(zn1 zi )
l1
(2i)
i=1
n2>i>j >1

Z
Z

dtl1 wN (t1 , . . . , tl1 ; z1 , . . . , zn2 )Pl,n
dt1
= (1)+1
C1
tl =zn +p3h/2
zn =zn1 +hp
Cl1
l1 n2
Y
Y
(ta zj )
a=1 j =1
sin
l1
Y
a<b
l1
Y
a=1
n2
f (zi zj )
p (ta
(ta tb )
n2
Y
ei(zn1 zj h/2)/p
j =1
e2i(ta zn1 h/2)/p

.
zn1 h/2) sin p (ta zn1 + 3h/2)
(67)
Hence the kinematical residue equation (10) is satisfied if

l1
Y

e2i(ta zn1 h/2)/p
Pl,n
tl =zn +p3h/2
sin p (ta zn1 h/2) sin p (ta zn1 + 3h/2)
zn =zn1 +hp
n2
Y
j =1
a=1
ei(zn1 zj h/2)/p Pl1,n2 .
(68)
555
We are going to discuss some important solutions of the equations (64) and (68) in the
following section.
Obviously, the value of n/2 l an invariant under the kinematical residue equations.
This invariant can be interpreted as the isospin of form factor sequence. This means that
by (64) and (68) we get sequences of solutions of all form factor equations for each value
of the isospin.
Let us collect the main results in a theorem.
Theorem 4. There exist infinite sequences
Y
cl
P
f (zi zj ) IMl,n (z1 , . . . , zn ),
FM (z1 , . . . , zn ) =
l
(2i)
(69)
n>i>j >1
of solutions of the qKZ equation with the R-matrix being the S-matrix (12) of the Thirring
model. Each sequence is labelled by the invariant n/2 l. If in addition Pl,n satisfies the
recursion relations (64) and (68) then the members of each sequence are linked by (9) and
hence provide replica-deformed form factors of local operators in the Thirring model with
isospin n/2 l.
Of course every sequence (69) can be multiplied pointwise with a sequence Q(z1 , . . . , zn )
without changing the properties described, provided Q(z1 , . . . , zn ) is completely symmetric, i-periodic in all arguments and satisfies
Q(z1 , . . . , zn )|zn =zn1 i = Q(z1 , . . . , zn2 ).
(70)
Physically the solutions of (70) correspond to eigenvalues of a local conserved charge.

As shown in [14] the structure of these eigenvalues for 6= 2 is quite different from
that in the undeformed case. Also the eigenvalues in both cases are not automatically in
one-to-one correspondence, though they can be made so by imposing suitable minimality
conditions.
In general we have to add that in order to comply with the general requirements on the
function space Fq introduced in Section 4 and with the form factor equations we shall
require that functions Q satisfying (70) do at least not lead to additional kinematical poles
in the sense of (9) and (10).
Alternatively such sequences arise as ambiguities in the solution of the recursive
equations for the polynomials Pl,n . This means that if we have found a particular solution
el,n (t; z) =
Pl,n of Eqs. (64) and (68) we may multiply this solution with another function Q
e
Ql,n (t1 , . . . , tl ; z1 , . . . , zn ) subject to the two conditions

el1,n2 (t; z),
el,n (t; z) tl =zn1 3h/2 = Q
Q
zn =zn1 h

el1,n2 (t; z).
el,n (t; z) tl =zn +p3h/2 = Q
Q
(71)
zn =zn1 +hp
el,n
e satisfies the (64) and (68) trivially. Hence, the expression Pl,n Q
This means that Q
is then automatically a solution to (64) and (68) as well. However, we have to make sure
that this procedure takes place within the space Fq . If we would not take care about this
556
then we easily lose the property that the form factors as defined in (56) with an insertion
el,n in the hypergeometric integral I P still is a solution to the first
of the expression Pl,n Q
M
two form factor (and hence qKZ) equations, which is of course an essential property.
The equations (71) admit the following two classes of solutions 2 which, however, are
not necessarily inside the space Fq .
The first one is:

el,n (t; z) = q z1 + + zn 2(t1 + + tn ) + 2l i ,
(72)
Q
where q is a largely arbitrary scalar function. A further type of solutions is given by the
following off-critical Bethe eigenvalues:
el,n (t; z) =
Q
n
Y
iS0 (zj i)
j =1
)
l
l
n
Y
ta + 3i/2 Y ta + 7i/2 Y zj i
+
,
ta + 5i/2
ta + 5i/2
zj 2i
a=1
a=1
j =1
(73)
(
where is a spectral parameter.

In any case, modulo their respective ambiguities, the completeness result for Section 5
(together with its undeformed counterpart) now implies the announced correspondence (3):
the form factor sequences in the replica-deformed model are essentially in one-to-one
correspondence to their undeformed QFT counterparts. In particular this allows one to
identify the replica-copy of the form factors of a local operators. As an illustration we
present the replica-deformed form factors of the Noether current in the next section.
7. Deformed form factors of the Noether current

As it was mentioned in the introduction, the form factor approach enables to classify
the full space of local operators of a given model. This arises from the fact that there exist
kernel solutions to the equations (64) and (68). This feature is illustrated in [20] where
the local fields in the sine-Gordon model have been counted, and in [19] where the full
space of local operators in the sinh-Gordon model has been found. In the latter model
the local operators in one superselection sector were identified with the integer powers
of the elementary field appearing in the Lagrangian together with an exponential field. It
should be possible to mimic the analysis of the local fields of [22] in the present case as
well. However, this is beyond the scope of the present paper. Instead we present in this
section solutions to (64) and (68) which in the limit = 2 turn into the form factors
corresponding to the SU(2) Noether current in the Thirring model. These form factors are
well understood in the standard case and it has been shown in [13] how these form factors
arise from the qKZ solutions at level zero and how they come about from the trace formulae
of vertex operators in the Yangian double of [10].
2 I thank the referee for asking me to add this.
557
Let ja be the Noether current associated with the Lagrangian (11); is the Lorentz index
and a the SU(2)-index as introduced in Section 3. It is convenient to switch to lightcone
coordinates both in Minkowski space and in internal space, i.e.,
ja = j0a j1a ,
j = j1 ij2 ,
= .
(74)
Then we can introduce an index = , 3 labelling the components of the lightcone

currents as j . A solution of the equations (64) and (68) is given by
i(2ln/2+ )
= dl,n
e
Pl,n
l
Y
a=1
Pn
2i(l+ )ta /p
j=1 zj /p
Y
16a<b6l

sin (ta tb h) sin (ta tb + h) ,

p
p
is
where the constant dl,n

i
n 2l 2( + 1) l .
dl,n = exp
p
(75)
(76)
Comparing this with the level zero counterpart [13,21] we find that this choice of Pl,n
corresponds to the form factors of the current j . The form factors of the other components
of j can obviously be obtained by applying the operator + once or twice to the
corresponding form factor.
A closer look on the solution (75) shows that the corresponding form factor does satisfy
the first two of the form factor equations (8) with a factor 6= 1. But still we have || = 1
as it should be. The reasons why this feature appears is discussed in detail in [21].
Of course, in order to establish that these solutions do really correspond to the replicadeformed form factors of the currents we have to prove several conditions like in the level
zero case, see Section 6 of [21]. The most important of these conditions is the current
conservation condition ja = 0.
To verify this in the deformed framework involves the knowledge of the deformed
eigenvalues of the lightcone momenta. We remind that the kinematics in the deformed
model is quite different from the usual model as was outlined in Section 2. Even though
these eigenvalues have been characterised and computed for some low values of the particle
number n in [14], a general formula is not known at present. This certainly deserves a
further study also in the light of the characterisation of other operators in the replica-copy
of the Thirring model as well.
Therefore, we can at present only conjecture that the form factors characterised by the
solutions (75) do correspond to the Noether-current operators in the replica-copy of the
Thirring model. Nevertheless we can give an argument to support our conjecture. Let P
be the eigenvalues of the lightcone momenta on an n-particle state in the 6= 2 copy of
the Thirring model according to [14]. We denote by (z1 , . . . , zn ) the functions which
we conjecture to be the form factors of the currents j .
The explicit difference of + and has two different sources. The first one is simple
and resides in the purely z-dependent part of (75). This bit is not affected by the integration.
The second source comes from the -dependent part in the first product term in (75).
558
This term is trivial in the sense that it is responsible neither for poles nor for zeros of the
integrand. In addition it is p-periodic.
Let us introduce a function Q = Q(z1 , . . . , zn ) which would ideally be a solution to (70).
We use this function together with the eigenvalues of the lightcone momenta to define the
following auxiliary objects:
J+ := P+ Q1/2 + ,
J := P Q1/2 .
(77)
Then we can write the current conservation in the form

P+ J + P J+ = P+ P Q1/2 (+ Q ) = 0,
(78)
provided an object Q exists. This existence, however, is guaranteed by the following

argument. We have mentioned above that we have two sources of possible differences
between + and . It is obvious that the first difference, which is purely z-dependent
can be absorbed without any difficulty in Q. Then, the hypergeometric integrals which
define can be understood in terms of a multidimensional residue evaluation which
leads to infinite series expressions for . Using the property of the regular spacing of
the poles (40) and what has been said before about the properties of the second source
it is clear that there exists an object Q which in the infinite series makes the expression
+ Q termwise vanish.
8. Discussion
In this paper we have been trying to illustrate the natural connection between rational
solutions of the qKZ equation and the replica-deformed form factors in the SU(2)-invariant
Thirring model. Here we took up two independent directions of recent research.
The first one is Niedermaiers formalism showing that every integrable massive quantum
field theory in 1 + 1 dimensions admits an S-matrix preserving deformation, which can be
interpreted as providing a replica-copy of the original QFT with an off-critical Unruh
temperature. It was shown in [14] that this replica-copy can be described exactly in terms
of its form factors, the latter being solutions of a a modification of the standard form factor
equations [9,20]. The present paper is the first in which these equations have been solved
for a particular model in full generality.
As a technical tool we took advantage of another recent development, namely the study
of the solutions of the rational sl2 qKZ equation [12,24]. As one could suspect by looking
at the first set of modified form factor equations (8), their solution in the Thirring model
case should somehow reside in the space of solutions of the qKZ equation for a particular
representation of sl2 . The question then is whether among these solutions we can find some
satisfying the deformed kinematical residue equations (9) as well.
The main purpose of this work was to show that this is indeed possible. In outline we
rewrote the qKZ solutions of [12,24] at generic level in a way faciliating the comparison
with the standard level zero situation. We argued that this form of the solution naturally
shares the property that the algebraic Bethe vectors remain unchanged under the replica
559
thermalisation. This fact could have been expected since the modified form factor equations
use the usual scattering matrix. We then supplemented our qKZ solutions with the minimal
form factor to define an object (56) having only the polynomial P unspecified. Using
the pole structure of the qKZ solutions we derived conditions on P from the modified
kinematical residue equations and solved these conditions to conjecture the deformed form
factors of the Noether-current in the SU(2) Thirring model.
To summarise we have shown that the Unruh-thermalised form factors arise naturally
from qKZ solutions at generic level as the standard form factors do at level zero.
We may also anticipate that the 0 limit of our construction bears an unexpected
relation to dimensionally reduced quantum gravity [17].
It should now be possible to derive integral representations of the form factors for the
thermalised sine-Gordon model using the results on trigonometric qKZ solutions as derived
in [25]. It would be particularly interesting to construct the thermalised form factors of the
breather sectors. This would involve also the construction of modified fusing relations for
the form factors. The latter have not been supplied in [14]. In addition solving this problem
would also naturally lead to an integral representation of the thermalised form factors in
the sinh-Gordon model, a problem which has only partially been solved in [14].
Acknowledgement
I am grateful to M. Niedermaier for many discussions on the replica-deformed form
factor equations and to M. Flohr for conversations on hypergeometric integrals. Thanks go
also to A. Nakayashiki for a useful correspondence on the work [13].
This work was supported by EPSRC (grant GR/L26216).
References
[1] J. Bisognano, E. Wichmann, J. Math. Phys. 16 (1975) 985; J. Math. Phys. 17 (1976) 303.
[2] C. Callan, F. Wilczek, Phys. Lett. B 333 (1994) 55.
[3] P.I. Etingof, I.B. Frenkel, A.A. Kirillov, Lectures on Representation Theory and Knizhnik
Zamolodchikov Equations, Am. Math. Soc., 1998.
[4] I.B. Frenkel, N.Yu. Reshetikhin, Commun. Math. Phys. 146 (1992) 1.
[5] V.P. Frolov, D.V. Fursaev, A.I. Zelnikov, Phys. Rev. D 54 (1996) 2711.
[6] G. t Hooft, Class. Quant. Grav. 16 (1999) 395.
[7] M. Jimbo, T. Miwa, Algebraic Analysis of Solvable Lattice Models, CBMS 85, Am. Math.
Soc., 1995.
[8] M. Jimbo, T. Miwa, J. Phys. A 29 (1996) 2923.
[9] M. Karowski, P. Weisz, Nucl. Phys. B 139 (1978) 455.
[10] S. Khoroshkin, D. Lebedev, S. Pakuliak, Lett. Math. Phys. 41 (1997) 31.
[11] T. Miwa, Y. Takeyama, Determinant formula for the solutions of the quantum Knizhnik
Zamolodchikov equation with |q| = 1, preprint math.QA/9812096.
[12] E. Mukhin, A. Varchenko, The quantized KnizhnikZamolodchikov equation in tensor products
of irreducible sl2 -modules, preprint q-alg/9709026.
[13] A. Nakayashiki, S. Pakuliak, V. Tarasov, On solutions of the KZ and qKZ equations at level
zero, preprint q-alg/9712002.
560
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
M. Niedermaier, Nucl. Phys. B 535 (1998) 621.

M. Niedermaier, Nucl. Phys. B 519 (1998) 517.
M. Niedermaier, Commun. Math. Phys. 196 (1998) 411.
M. Niedermaier, H. Samtleben, private communication.
M. Pillin, Phys. Lett. B 448 (1999) 227.
M. Pillin, Int. J. Mod. Phys. A 13 (1998) 4469.
F.A. Smirnov, Form Factors in Completely Integrable Models of Quantum Field Theory, Adv.
Series in Math. Phys. 14, World Scientific, Singapore, 1992.
F.A. Smirnov, Lectures on integrable massive models of quantum field theory, in: M.-L. Ge,
B.-H. Zhao (Eds.), Nankai Lectures on Math. Phys., World Scientific, Singapore, 1990.
F.A. Smirnov, Nucl. Phys. B 435 (1995) 807.
V. Tarasov, Completeness of the hypergeometric solutions of the qKZ equation at level zero,
preprint MPI 98-87, math.QA/9811048.
V. Tarasov, A. Varchenko, Invent. Math. 128 (1997) 501; e-print q-alg/9604011.
V. Tarasov, A. Varchenko, Astrisque 246 (1997); e-print q-alg/9703044.
W. Unruh, Phys. Rev. D 14 (1976) 870.

Structure constants for the D-series Virasoro

minimal models
Ingo Runkel
Mathematics Department, Kings College London, Strand, London WC2R 2LS, UK
Received 7 September 1999; revised 27 October 1999; accepted 2 November 1999
Abstract
In this paper expressions are given for the bulk and boundary structure constants of D-series
Virasoro minimal models on the upper half plane. It is the continuation of an earlier work on the
A-series. The solution for the boundary theory is found first and then extended to the bulk. The
modular invariant bulk field content is recovered as the maximal set of bulk fields consistent with
the boundary theory. It is found that the structure constants are unique up to redefinition of the fields
and in the chosen normalisation exhibit a manifest Z2 -symmetry associated to the D-diagram. The
solution has been subjected to random numerical tests against the constraints it has to fulfill. 2000
Elsevier Science B.V. All rights reserved.
PACS: 11.25.Hf; 03.65.Fd
Keywords: Conformal field theory; Boundary problems; Minimal models; Structure constants; Sewing
constraints
1. Introduction
One of the aims of studying quantum field theories is to compute correlation functions
of the fields in the theory. In two-dimensional conformal field theory one uses the presence
of an infinite amount of symmetries to achieve this goal.
We will consider a special subclass of conformal field theories, called minimal models.
One of their properties is that the correlators fulfill certain linear differential equations
(see [1] or, e.g., [2] for details). One can find a distinguished set of solutions to these
differential equations, called conformal blocks. A correlator can then be expressed as a
bilinear combination of these blocks. To know the coefficients in this bilinear combination
a set of complex numbers, the structure constants, are needed. Structure constants appear
in the short distance expansion (or operator product expansion, OPE) of certain fields in
the conformal field theory, called primary fields.
ingo@mth.kcl.ac.uk
PII: S 0 5 5 0 - 3 2 1 3 ( 9 9 ) 0 0 7 0 7 - 5
562
I. Runkel / Nuclear Physics B 579 [FS] (2000) 561589
In this paper we consider only Virasoro minimal models. One approach to calculate all
structure constants is to start in a situation without boundaries. In this case only bulk fields
are present and one can solve the bulk theory alone by determining the conformal blocks
and a consistent set of structure constants for the bulk fields (see [3] for the spinless case).
It is possible to classify all modular invariant minimal models without boundaries [4] with
the result that they fall into an ADE pattern.
Starting from a well defined bulk theory one can proceed and introduce a boundary
into the system. As it turns out the differential equations that the correlators fulfill stay the
same, but now the correlator is a linear, not a bilinear, combination of the conformal blocks
mentioned above (see [5]) and furthermore two new sets of structure constants appear: in
addition to the bulk structure constants that determine the OPE of two bulk primary fields,
there are the bulkboundary couplings that describe the expansion of a bulk field in terms
of boundary fields, and the boundary structure constants for the OPE of two boundary
fields.
One can now classify all conformally invariant boundary conditions and the boundary
fields that live on these boundaries or interpolate between two adjacent boundary
conditions (see [610]). Again one finds an ADE pattern for all possible boundary
theories [9,10], plus tadpole diagrams, which have been discarded as the resulting theory
does not match with any of the modular invariant bulk theories.
All three sets of structure constants can in principle be worked out from six consistency
conditions that arise from taking different limits in certain correlators and equating the
different OPEs in these limits. The constraints that arise in this way are called sewing
constraints (see [11] and for an extension to non-orientable surfaces [7]). A more general
discussion of the sewing constraints in presence of nontrivial multiplicities and their
relation to [12] can be found in [13].
Above we have started from a bulk theory, which can be solved consistently, and then
introduced a boundary. Thus the bulk structure constants are known first and then a solution
for the other two sets of structure constants consistent with these has to be obtained.
Whereas the bulk structure constants for diagonal and nondiagonal minimal models can
be found in the literature [3,1416], these calculations have not been carried out for the
boundary structure constants and only for a subset of the bulkboundary couplings (see
[610]).
In this paper and a previous work [17] the missing structure constants for the A- and
D-series are calculated by taking the opposite path. The starting point now is a theory
with only boundary fields. One chooses an ADE type boundary field content and tries
to determine a consistent set of boundary structure constants. Once this is achieved the
boundary theory is extended to the bulk and it is only at this point that bulk fields are
introduced. It is interesting that one finds that the maximal bulk field content consistent
with the boundary theory coincides with the modular invariant one given in [4]. The bulk
boundary couplings and the bulk structure constants can now be obtained from the sewing
constraints in a straight forward way. The results are explicit expressions for all three sets
of structure constants in terms of so called fusion (F-) matrices (see Appendix A for details
and references). The expressions are free of sign ambiguities (e.g., they do not contain any
563
square roots) so that they are well suited for numerical computations.
The Deven minimal models have a larger symmetry algebra (a W-algebra, see [18,19]
and references therein) that contains the Virasoro algebra as subalgebra. In principle one
could understand these conformal field theories in terms of the W-algebra, with W-algebra
chiral blocks and F-matrices. But since these CFTs are still finitely reducible with respect
to the Virasoro algebra alone (i.e., only a finite number of irreducible highest weight
representations of the Virasoro algebra occur), it is possible and indeed simple to use only
the Virasoro symmetry.
The fact that one does not use the fully extended chiral algebra and the possible presence
of multiple copies of primary fields means that the sewing constraints in Ref. [11] must be
rederived, which has been done. The results are presented in Section 2 and an explicit
example is provided in Appendix A.
Both, in the A- and the D-series, the sewing constraints are overdetermined and only a
subset is used to find the structure constants. It has to some extend been tested numerically
(but not proven analytically) that the given structure constants solve the full set of
constraints. It is shown that, given the boundary field content, any solution can be brought
to a standard form by redefining the fields. In other words, under gauge transformations
the space of solutions to the sewing constraints consists of only one orbit.
The formulae for all bulk and boundary structure constants are given in Eqs. (43)
(47) and (53)(55). A short description of how to implement them can be found in the
conclusion. The derivation of these expressions is organised as follows: in Section 2 the
notation for the various fields and structure constants is introduced and the rederived
sewing constraints for genus zero surfaces are stated in a suitable form. Section 3 gives
a short description of the procedure introduced in [9,10] for the construction of cylinder
partition functions. This defines the boundary field content which is used as input in the
construction of the solution to the sewing constraints. The boundary structure constants
are found in Section 4. Finally the extension to the bulk theory is carried out in Section 5.
Here the maximal bulk field content consistent with the boundary theory is determined and
it is observed to be modular invariant. The remaining structure constants, i.e., the bulk
boundary couplings and the bulk structure constants are calculated and expressions for
the vacuum expectation value of the unit disc are obtained. A choice of basis is presented
which renders all structure constants real. Finally it is observed that a subset of all structure
constants is left invariant under a large number of distinct Z2 -actions.
2. Genus zero sewing constraints

Unless otherwise mentioned, throughout this paper the boundary conformal field theory
will be considered on the upper half plane (UHP) with the boundary given by the real line.
In a general Virasoro minimal model there can be several primary bulk or boundary
fields transforming in the same highest weight representation(s) of the Virasoro algebra.
Therefore we cannot label the fields solely by their representations. Virasoro highest
weight representations are labelled by small letters i, j , k and to take care of fields with
564
multiplicities these carry an additional Greek index i , j , k . Boundary conditions are

labelled by small letters a, b, c or also x, y, z.
It is always possible to make the (bulk- and boundary-) two-point functions diagonal by
an appropriate choice of labels and normalisation of the fields: hi (x)j (y)i=i,j ,
f (x, y).
A bulk field i transforms in the tensor product of two Virasoro highest weight
representations i . The two representations have conformal weights hi , h or, for better
(ab)
readability, hi , h i . A boundary field k lives between boundary conditions a and b and
transforms in one Virasoro representation k of conformal weight hk .
The vacuum expectation value of the UHP with boundary condition a imposed on the
(a)
real line will be denoted with h1iUHP . The vacuum expectation value of the full complex
plane with no boundaries present will be denoted with h1i or h0|0i.
The leading terms in the bulkbulk, bulkboundary and boundaryboundary operator
product expansions of primary fields are, in this order:
X

Ci j k (z w)hk hi hj (z w)
hk hi hj
i (z)j (w) =
k,

k (w) + , |z|>|w|,
X
(aa)
(a)
Bi k (2y)hk hi hi k (x) + ,
i (x + iy) =
k,
(x)j(bc)
(y) =
i(ab)
(abc)k
C i j
(1)
y > 0,

(x y)hk hi hj k(ac)
(y) + ,
x > y.
(2)
(3)
k,
The omissions stand for an infinite sum of descendants of the primary field in question.
Eqs. (1)(3) define the three sets of structure constants which are necessary to compute
the correlation functions of the minimal model under consideration: the bulk structure
constants Ci j k with three bulk fields i , j , k , the bulkboundary couplings (a)Bi k
with boundary condition a, bulk field i and boundary field k and the boundary structure
(abc)k
constants C i j with boundary conditions a, b, c and boundary fields i , j , k .

By taking different limits in correlation functions one obtains a set of constraints, called
sewing constraints. For orientable genus zero surfaces it is enough to consider the following
four constraints [11]:
For four boundary fields i , j , k , ` and boundary conditions a, b, c, d:
X
(bcd)q
(a)
(ada)1
C j k C (abd)`
i q C ` ` h1iUHP
XX
p

j k
(abc)p
(cda)p
(a)
C i j C k ` C (aca)1
h1i
F
.
pq
p p
UHP
i `
For two boundary fields p , q and one bulk field i :

X
(abb)q
(a)
(b)
Bi ` C p ` C (aba)1
q q h1iUHP
(4)
XX

(aaa)1
(a)
Bi k C p q C k k h1iUHP
(a)
(aba)k
565
ei(2hm 2hi hp hq + 2 (hk +h` )) Fkm

q
i
Fm`
.
i p
p q
(5)
For two bulk fields i , j and one boundary field k :

X
(a)
Ci j m (a)Bm k C (aaa)1
k k h1iUHP
XX
p,q
(a)
Bi
p (a)
Bj
(aaa)k
(a)
C p q C (aaa)1
k k h1iUHP
X
r
ei 2 (hk +hp hq 2hr +hm hm hi +hi +hj +hj )
Fqr

k
r

Fpm
Fr m
.
p j
i j
m k
(6)
For four bulk fields i , j , k , ` :

X
Ci k q Cj ` q Cq q 1 h1i
XX
p,p

Ci j p Ck ` p Cp p 1 h1i
i(hi h i +h` h ` hp +h p hq +h q )

Fpq

j `
`
Fp q
.
i k
k
(7)
These constraints have been rederived using notation and techniques for calculating with
conformal blocks presented in [12]. Eqs. (4)(7) require a specific renormalisation of the
F-matrices. This normalisation, together with the explicit derivation of the constraint (6) as
an example are given in Appendix A.
The form of the constraints given here is slightly different from [11]. First of all only
the Virasoro symmetry was used in their derivation and not the maximally extended
chiral algebra. Secondly slightly different limits in the correlators were chosen so that
the expectation value of the identity field cancels from all expressions. Furthermore the
present notation takes into account fields with multiplicities (see also [13]). Leaving out
the sums over multiplicties and the multiplicity indices, (4)(7) can be rearranged to match
the corresponding equations in [11].
3. Cylinder partition function

First a note about a convention used throughout this paper. To take care of the
redundancy in the labelling of representations i with Kac-labels (r, s) we will only consider
pairs where r is odd. The set of entries in the Kac-table with r odd is in one-to-one
correspondence to Virasoro highest weight representations of the given central charge.
566
Before we can begin to solve (4)(7) we need to know which fields are present in
the theory. Since our starting point is going to be the boundary theory the first thing to
determine is the boundary field content on the upper half plane or, equivalently, the cylinder
partition function.
In Ref. [9,10] a method for the construction of the cylinder partition functions associated
to a pair of Lie-algebras An , G is given. Here we give a quick summary of this method.
3.1. General construction
First choose an odd number p and a number q coprime to p. We will construct the
cylinder partition function of a minimal model with the following central charge:
c=16
(p q)2
.
pq
(8)
Let A be the adjacency matrix of the Dynkin diagram associated to the Lie-algebra Ap1
and G be the adjacency matrix for a Lie-algebra with Coxeter number q.
For X = A or X = G define the matrix valued functions Vn (X) recursively via
Vn (X) = V2 (X)Vn1 (X) Vn2 (X),
V1 (X) = id and V2 (X) = X.
(9)
The Vn are called fused adjacency matrices and form a representation of the Verlinde fusion
algebra.
Let be an odd node of A and be any node of G. Then a = (, ) labels the possible
Then the partition function of a
boundary conditions. Let a = (, ) and b = (,
).
cylinder of circumference T and length L with boundary conditions a and b is given by:
X
Vr (A) Vs (G) r,s (q), q = exp(T /L).

(10)
Za|b =
r=1,...,p1,
s=1,...,q1
When we take the identity r,s = pr,qs for Virasoro characters into account we can
rewrite (10) in the following unique way:
X
nia b i (q).
(11)
Za|b =
i=(r odd,s)
In Refs. [9,10] the genus one sewing constraint for a cylinder with no field insertions was
analysed with the result that G has to be the adjacency matrix of an ADE type Dynkin
diagram or of a tadpole diagram.
The numbers nia b can be interpreted as the number of times the representation i occurs
in between the boundary conditions a and b and they thus describe the field content on the
boundary (in this case the real line).
3.2. Example: the A-series
For the pair of Lie algebras (Ap1 , Aq1 ) one finds nia b = Nia b , i.e., the field content
is just given by the Verlinde fusion numbers [6]. Distinct boundary conditions are given by
pairings of an odd node in the first A-diagram with any node in the second:
1 2s
si
s
i
p1
s s
i
1 i
si
s2 si
q1
s i
i
s
567
o
.
(12)
There is a distinguished boundary condition, which we will call 1-boundary, that

corresponds to the first node in each A-diagram, i.e., the (1, 1)-node.
There are no fields with multiplicity and there is a unique field i = a between the a- and
1-boundary: a i 1 i = a. The formula nia b = Nia b can be understood as saying
that the representations i that can live between the a- and b-boundary are exactly those
occurring in the fusion of the representations a and b: a a 1 b b a i b .
3.3. The D-series
Here q has to be an even number and the boundary conditions are given by pairings of
an odd node in the A-diagram with any node in the D-diagram:
)
(
s q
i
1 2s
si
s
i
p1
s s
i
1 i
s2 si
si
si
@
s q +1
@i
(13)
In particular the total number of boundary conditions is 14 (p 1)(q + 2). Again the
boundary condition associated with the first node of each diagram, i.e., the (1, 1)-node,
will get a special name. It will be called so that it is not confused with the 1-boundary in
the A-series. The -boundary will play a role similar to that of the 1-boundary in [17].
When looking at the D-series boundary field content one observes the following: the
boundary conditions can be organised in two categories. In the first case, which we will
denote as i-type boundaries, a boundary condition x = (, ) is associated with an odd
node of the A-diagram and any node of the D-diagram except for the two at the split
end. The second case we will call n-type. An n-type boundary a is associated with any
odd A-node and one of the two end nodes in the D-diagram.
i-type:
s si
s i
i
s
si ,
@
@s
n-type:
s
i
s
.
@
s
@i
(14)
The names i-type and n-type stand for invariant and non-invariant and are related to the
Z2 -symmetry of the D-diagram.
Carrying through the procedure outlined in Section 3.1 one finds that an n-type boundary
, denoted by au (unique). It has
a = (, ) has exactly one field living on a
representation labels (, q/2).
with KacAn i-type boundary x = (, ) has two fields fields living on x
labels (, ) and (, q ) which we will denote by xe and xo (x even and x odd). The
labelling is arbitrary, but we choose to always give the identity field on the -boundary an
even label.
are given
Note that since the Kac-labels of the representations that live on x
in terms of the node labels of the boundary condition x = (, ) some confusion might
arise. But from the context it will be clear which one of the two interpretations makes
sense.
568
For definiteness we will fix a specific even/odd labelling for boundary fields. It will
turn out later that the structure constants involving only i-type boundaries have an explicit
Z2 -symmetry that sends even fields to themselves and odd fields to minus themselves.
This symmetry is not assumed, but a consequence of the constraints and the gauge we
choose. Following the arguments in [24] we would like the ground state, i.e., the state
of lowest conformal weight, for each pair of boundary conditions to be invariant under
with x = (, ) an i-type
that symmetry. This can be achieved for all pairs x
boundary by assigning e/o-labels to the two representations living between these boundary
conditions in the following way:

xe =
(, ),
< p/2,
(, q ), > p/2,

xo =
(, q ), < p/2,
(, ),
> p/2.
(15)
As discussed in [24] the physical motivation for this choice comes from relating the Z2 action to the effect of a disorder line stretching from boundary to boundary on a cylinder
and trying to interpret the resulting amplitude as a partition function. It is also found there
that in general it is not possible to fix a labelling s.t. the ground state is invariant for all
possible pairs of boundary conditions.
The conformal weight of a highest weight representation with Kac-labels (r, s) is given
by hr,s = (1/4pq)((qr ps)2 (p q)2 ). Thus with definition (15) we find:

p
q
.
h(xo ) h(xe ) =
2
2
(16)
In particular the -boundary itself is of i-type and it has two fields living on it, the
identity e = 1 and the field o with h(o ) = (p/2 1)(q/2 1). We see that h(o ) is
integer for q/2 odd and half-integer for q/2 even.
We will now proceed to assign e/o/u-labels to all boundary fields, not just the ones
adjacent to the -boundary, in the following way: any field adjacent to an n-type boundary
will get the label u. In this case no multiplicities do occur. In the case x ` y where
both x and y are of i-type, the possible representations ` are those that occur in the fusion
of xe , ye (which is the same set as in the fusion of xo , yo ) or xe , yo (which is the same as
xo , ye ). If ` occurs in the xe , ye -fusion it gets an e-label, i.e., = e and if it occurs in the xe ,
yo -fusion an o-label = o. If ` occurs in both fusions, this representation has multiplicity
two and the two corresponding fields have labels è and ò .
The field content between different boundary conditions is summed up in Table 1 listing
the numbers nia b from (11) for the different cases (here the representation ` has Kac-labels
` = (r` , s` ) and the boundary conditions a, b have labels (a , a ) and (b , b )).
Notice the double role of the e/o/u-indices: When x is a boundary condition, then xe , xo
or xu denote the even/odd/unique field living between the x- and -boundary. In particular
xe and xo denote different representations. If ` is a representation, then è , ò or ù all
denote the same representation and distinguish fields with multiplicities by labelling one
as even and the other as odd.
569
Table 1
Field content and labels between the two types of boundary conditions
Boundary types
`-labels
ii
e, o
ni
nn
Field content
n`x y = N`xe ye + N`xe yo = N`xo yo + N`xo ye
nà y = Nàu ye = Nàu yo
a 6= b : nà b = Nàu bu , if s` 3 mod 4
and nà b = 0 otherwise
a = b : nà a = Nàu au , if s` 1 mod 4
and nà b = 0 otherwise
3.4. Ordering of boundary conditions

The constraint equations (4)(7) allow for a large amount of gauge freedom of the
structure constants. Here we find a particular set of structure constants and show that
any solution of the constraints can be transformed into this set by a regauging. For
the construction of the structure constants presented here an ordering of the boundary
Then we define:
conditions has to be introduced. Let x = (, ) and y = (,
).
x <y
or ( = and < ).
( < )
(17)
4. Boundary structure constants

In this section we will construct a set of boundary structure constants that solve the
sewing constraint (4) under the condition that there exists a solution at all (for the given
field content). The line of argument in this section is as follows: we assume that there
exists a solution to (4) for the D-series field content. Then we use the property of (4) that
a change of basis in the set of primary boundary fields maps a solution of (4) to a new
solution. By redefining the boundary fields one by one this freedom is exploited to adjust
a subset of boundary structure constants to the values we want. This subset together with
(4) then fixes the value of a general boundary structure constant. Altogether the statement
is that any solution can via change of basis be brought to the form presented in the end of
this section.
As input for the boundary solution we use
(a) the D-series field content obtained in Section 3.3;
(x)i(ba)
(y)i are nonzero, i.e.,
(b) the assumption that all two-point functions hi(ab)
(aba)1
C i i 6=0 for all a, b, i that are allowed by (a).

An interpretation of assumption (b) is that a zero two-point function implies that the field
in question does not occur in our model and hence we are looking at a field content different
from what we demanded in (a). This can be seen as follows: suppose C (aba)1
i i = 0 for some
. Any correlator can

choice of a, b, i . Consider a correlator which contains the field i(ab)
570
be expressed as a sum of conformal blocks with coefficients given as products of structure

constants. It is always possible to take a limit of this correlator where all bulk fields are
are taken together.
taken to the boundary and then all boundary fields except for i(ab)
(x)i(ba)
(y)i. Thus in this limit
In the end we are left with the two-point function hi(ab)
the coefficient in front of each conformal block contains the factor C (aba)1
i i and hence the
(ab)
correlator is identically zero. Therefore any correlator involving i vanishes and this
field can as well be removed from the model, so that we effectively have a field content
different from (a).
The actual calculation of the boundary structure constants is a slightly lengthy case
by case study and is presented only for completeness. The results are summerised in
Section 4.5.
4.1. Nonzero boundary structure constants
First we investigate the consequence of (b), i.e., that all two-point functions are nonzero.
Eq. (4) can be rewritten so that the two-point structure constants cancel. To do so we set
` = 1 and obtain the three-point identity:
j
k
a i b c a
(bca)i
(aba)1
(abc)k
(aca)1
C j k C i i = C i j C k k .
(18)
Applying this to the r.h.s. of (4) we get:

j
k
a i b c d ` a
(bcd)q
(abd)`
C j k C i q

(abc)p
(acd)`
C i j C p k Fpq
p,

j k
.
i `
(19)
Note that there is no freedom the rescale the identity field 1 on any boundary. Its
(xxy)j
(xy)
(xy)
normalisation is already fixed by the condition that 1 i = i , i.e., that C 1i =
(xyy)j
i,j , and similar for C i 1 .

Define an operation on the labels e, o, u as follows: xe = xo , xo = xe and xu = xu ,
i.e., in terms of Kac-labels: (r, s) = (r, q s). Take o xi x xi o with
q = 1 and x any boundary condition. The sum on the r.h.s. of (19) reduces to p = xi
because the representations xi and o can fuse only to xi . We get:

o o
(x)1
(x)xi (x)o
6= 0.
(20)
C xi xi = C o xi C xi xi Fxi 1
xi xi
The l.h.s. is nonzero due to assumption (b) and hence all terms on the r.h.s. have to be
nonzero for a solution with properties (a) and (b).
In the same way, for an n-type boundary a and an arbitrary boundary x we can consider
xi x ku a ku x xi , again with q = 1. The sum reduces to p = au and we are
left with:

xi xi
(xa)au (ax)xi
6= 0.
(21)
=
C
C
F
C (xax)1
au 1
xi ku
au ku
ku ku
k k
Again all terms on the r.h.s. have to be nonzero.
571
4.2. Mixed boundaries

Define the boundary condition label as = (1, q/2), i.e., the first node of the
A-diagram and the upper end node of the D-diagram in (13). Consider (19) with
u o o u and q = 1. The sum on the r.h.s. reduces to p = u and
we are left with:

o o
()u 2
=
C
F
.
(22)
C ()1
u 1
u o
o o
u u
Using (18) in the form u o u
()u
implies C u o
()
= C o u u. We start by
()
rescaling o such that:

u
= 1.
C ()
o u
(23)
From (22) it now follows that

C ()1
o o
= A,
where A = Fu 1

u u
,
o o
(24)
(x)
(x)x
defining the constant A. We use the freedom to rescale xo to fix C o xo e = A for any
i-type boundary x.
Let a be an n-type boundary. Taking o u ù a au and q = au the sum
reduces to p = u and we see, using (23):

o au
(a)au (a)au
(a)au
.
(25)
C u ù C o au = C u ù Fu au
u `
The representation ù lives on
and the fusion u , au ù exists by
(a)a
C u ù u
is nonzero (see (21)) and the F-matrix element

construction of the field content.
has to be independant of the specific choice of ù we make. We denote this F-matrix
element with Ba :

o au
(26)
, where ù a .
Ba = Fu au
u `
u
Comparing to (25) we see C (a)a
o au = Ba . It is also useful to define a constant Cx for any
boundary x as follows:

o o
o o
,
x of n-type: Cx = Fxu 1
.
(27)
x of i-type: Cx = Fxe 1
xo xo
xu xu
From (19) with o o au a au , q = au and p = 1 we get the following

identity for any n-type boundary a:
Ca Ba =
A
.
Ba
(28)
(xy)
Let x and y be two boundaries s.t. one is of n-type. Then by rescaling ù for x 6=
(xy)y
and ` 6= 1 we fix C x ù
= 1 where , stand for the labels e or u, as applicable for the
572
present boundary conditions. Let x be an i-type boundary. We can use (19) to calculate the
structure constants involving xo instead of xe :

A
Ba
o au
o xe
(xa)au
(ax)xo
,
C au ù =
.
(29)
Fx a
Fa x
C xo ù =
Ba e u xo `
A u o au `
Using (19) and the identities in Appendix B one can verify that the definitions above
(xy)y
(yx)y
give rise to symmetrical structure constants, i.e., C x ù = C ù x where at least one of
x, y is of n-type.
4.3. i-type boundaries
(xy)y
Let x, y be i-type boundaries. Consider the structure constant C x ` where the labels
, , are either e or o (as u cannot occur in this situation). We will now try to regauge
the boundary fields such that the following rule holds:
(xy)y
The structure constant C x ` can be nonzero only if {, , } is one of the
(unordered) sets {e, e, e} or {e, o, o}.
y occurs with multiplicity one. Recall
Suppose that the representation ` on x
from the analysis of the D-series boundary field content in Section (3.3) that then the above
rule is automatically true, since, e.g., if ` gets label e then it occurs in the fusion of xe , ye
or xo , yo but not in the fusion of xe , yo and xo , ye .
If the representation ` occurs with multiplicity two, we can use the freedom to form
linear combinations of the two fields to make the above rule true. This is explained in more
detail below.
(xy)
(xy)
Let ` be a representation that occurs with multiplicity two. Let è and ò be the
two fields. Recall that we normalised the two-point functions in such a way that only
(xy) (yx)
(xy) (yx)
hè è i and hò ò i are nonzero. When taking linear combinations of the primary
fields one has to preserve this condition, or the form of the sewing constraints would
change. One gets the following constraints:
(xy) (yx)
(xy) (yx)
ò è = 0.
(30)
è ò = 0,
Now consider the following change of basis:

(xy)
(yx)

(xy)
è
è
è
a b
a
=
,
=
(xy)
(xy)
(yx)
c d
c
ò new
ò old
ò new
b
d

(yx)
è
(yx)
ò
(31)
old
where the two matrices are invertible. The constraints (30) amount to:
(xyx)1
(xyx)1
a c C è è ,old + bd C ò ò ,old = 0,
(xyx)1
(xyx)1
ca C è è ,old + d b C ò ò ,old = 0.
(32)
We see that we can choose the new basis of (xy) arbitrarily and the above condition fixes
the direction, but not the length of the new basis vectors for (yx) .
This change of basis can be used to bring the following 22-matrix composed of
structure constants to diagonal form:

(xy)ye
(xy)y
C xe è o
C xe è
0
.
(33)
(xy)y
(xy)y
0
C xe ò e C xe ò o
573
(xy)y
We now redefined the structure constants C x ` in such a way that they obey our desired
coupling relations {e, e, e} or {e, o, o}.
(yx)
(yx)
The change of basis fixes the new è and ò up to rescaling. However they do
already obey the odd/even-coupling relation. To see this consider (19) in the form:
xe x è y ò x xe
(xyx)1
C è ò
, q = 1

(xy)y (yx)x
= C xe è e C ye ò e Fye 1

` `
.
xe xe
(34)
(xy)ye
The l.h.s. of this equation is zero and on the r.h.s. both C xe è

are nonzero (this can be seen from evaluating
(yx)xe
Hence C ye ko
From
(B.4):
(xyx)1
C è è
and the F-matrix entry

(xyx)1
6= 0 instead of C è ò = 0 in (34)).
= 0 as we said.
e
C (x)x
o xo
o
= A in the previous section we can determine C (x)x
by using (19) and
o xe
o o xe x xe
o
C (x)x
=
o xe
()1
C o o
(x)x
C o xo e
, q = xo

1
x
F1xo o e =
.
o xe
Cx
(35)
(xy)y
(xy)
(xy)
Next we fix the form of C x ` . For ` 6= 1, x 6= we rescale è and ò s.t.:

if x 6 y,
1,

y
(xy)y
(xy)y
o ye
(36)
C xe ò o = F
C xo è o = Fxe yo o e ,
xo yo x ` , if x > y.
xo `
e
o
= 1 and that for x = the above reduces to
One can verify that for ` = 1 indeed C (xx)x
xo 1
(x)
given before.
the expressions for C
As in the case of mixed boundaries one can now check that the definitions in (36) give
rise to reflection symmetric structure constants, but this time there is an exception for
x = y: in this case some of the structure constants are antisymmetric under reflection. As
opposed to the A-series, for the D-series it is in general impossible to find a gauge in which
all boundary structure constants have reflection symmetry. To see this consider (19) in the
form:
xe x ò x xe o
(xx)x
o
C ò xe o C (x)
xe xo
(xx)x
o
= C xe ò o C (x)
Fxo xo
xo xe

xe o
.
` xe
(37)
Eq. (B.3) forces the F-matrix entry in (37) to square to one. But in general it does take
o
o
and C (x)
, in
the values 1, depending on ò . So no matter how we choose C (x)
xe xo
xo xe
(xx)xo
general we cannot avoid that if C xe ò is symmetric for some values of ò , it will be
antisymmetric for others.
One can verify that (36) implies that for even fields alone we get a solution that resembles
(xy)y
the A-series, i.e., C xe è e = 1, whereas the various structure constants involving odd fields
are:
574

(xy)yo
For x 6 y: C xe ò
= 1,
(xy)yo
C xo è
= Fxe yo

o ye
,
xo `

o yo
= A Fxe ye
Cy .
xo `

y
y
(xy)y
(xy)y
C xo è o = Fxe yo o e ,
For x > y: C xe ò o = Fxo yo o e ,
xe `
xo `

(xy)y
C xo ò e
(xy)ye
C xo ò
= A Cx .
(38)
4.4. General boundary structure constants

Consider (19) in the form
z
x x ir y js z
(xyz)kt
C ir js
q = kt
(xy)y
(yz)z

C x ir C y js
x
F
y k
(xz)z
i
C x kt

z
,
j
(39)
where the set {, t, } has to be one of {e, e, e}, {e, o, o} or {u, , }. Since all constants
that can occur on the r.h.s. have been computed in the previous two sections, the general
boundary structure constants can be obtained from (39).
Suppose all boundaries are of i-type. Then choosing = e and = t reduces the sum
in (39) to = r. In the numerator on the r.h.s. we now see the boundary structure constant
(yz)z
C yr js t , which obeys the even/odd coupling rule as seen in Section 4.3. This implies that
(xyz)k
the even/odd coupling rule extends to all boundary structure constants, i.e., C ir js t can be
nonzero only if {r, s, t} is one of the sets {e, e, e}, {e, o, o} or {u, , }.
We can also investigate the behaviour of the boundary structure constants under
(xy)y
(xyz)k
(zyx)k
reflection, i.e., given C ir js t , what is C js ir t ? Using (18) we check that for C x `
reflection symmetry is equivalent to = 1 in the following equation:
(xy)y
C x `
(x)1
C x x
(y)1
C y y
(yx)x
C y ` .
(40)
As remarked in the previous two sections, using the explicit form of the structure constants
derived there, we see that = 1 in all cases except for one, and that is = o and x = y,
where we can have = 1. This can be made more precise in the light of (37), if we define:

xe o
(41)
and (x, y, ` ) = 1 in all other cases.
(x, x, ò ) = Fxo xo
` xe
Then (40) holds with = (x, y, ì ) for all x, y, ì and we can use this relation to evaluate
the reflection property of the general boundary structure constants obtained in (39). We
get:
(xyz)kt
C ir js
(x, y, ir )(y, z, js )
(zyx)k
C js ir t .
(x, z, kt )
(42)
575
4.5. Collection of results

We found: any solution to the boundary four-point constraint (4) for a given D-series
field content can be brought to the same form by regauging the fields, i.e., if there is a
solution, it is essentially unique.
In the gauge we chose, there are two conditions that have to be satisfied before a given
(xy) (yz)
(xz)
fusion i j k can exist. First the according Verlinde fusion number Nij k has
to be nonzero and second the e/u/o-labels {, , } have to be one of the sets {u, , },
{e, e, e} or {e, o, o} (where stands for any label). The assignment of the e/o/u-labels is
described in Section 3.3. From this even/oddcoupling rule we see immediately that the
boundary structure constants involving only i-type boundaries have a Z2 -symmetry of the
(xy)
(xy)
(xy)
(xy)
form è è and ò ò .
There are two boundary conditions which we have endowed with a special name.
stands for the boundary condition associated to the (1, 1) pair of nodes in the diagram
(13) and for (1, q/2). The boundary structure constants take the following form:
Two-point functions with (the constants A and Cx are defined in (24) and (27)):
C (x)1
xu xu = 1,
C (x)1
xe xe = 1,
C (x)1
xo xo = A Cx .
(43)
For x any type and y of n-type (By is defined in (26)):

(xy)y
C xu ù u
(xy)y
C xe ù u
= 1,
= 1,
(xy)y
C xo ù u

A
o yu
=
Fx y
.
By e u xo `
(44)
For x, y of i-type and x 6 y (the ordering is defined in Section 3.4):

(xy)ye
C xe è
(xy)y
C xo è o
= 1,
= Fxe yo
(xy)yo
C xe ò
o ye
,
xo `
= 1,

(xy)y
C xo ò e
= A Cy Fxe ye

o yo
.
xo `
(45)
All other cases can be obtained from this list by using (40) with = 1 (since the case x = y
is covered in the list), with results listed in (29), (38).
For the general boundary structure constants we distinguish two cases. First, for x, z of
n-type and y of i-type the sum over in (39) has to be carried out and we get:

Bx
xu zu
ye o
zu o
xu zu
(xyz)ku
Fxu yo
Fye zu
Fyo k
+
. (46)
C iu ju = Fye k
i j
i xu
j yo
i j
Bz
In all other cases the sum reduces to one term with the result:
(xy)y
(yz)z

C x ir C y js
x z
(xyz)kt
Fy k
.
C ir js =
(xz)z
i j
C x kt
(47)
The r.h.s. does not depend on the specific choice of , , as long as the combinations
{, r, }, {, s, } and {, t, } are allowed by the even/odd coupling rule. The structure
constant in the denominator is then automatically nonzero (see Section 4.1). The boundary
structure constants are either symmetric or antisymmetric under reflection, the precise
behaviour is given in Eq. (42).
576
Note in particular that if x, y, z are all of n-type or if the fields ir , js , kt are even, the
solution takes the same form as in the A-series (see Eq. (31) in [17]):

x z
x z
(xyz)k
(xyz)k
(48)
C ie je e = Fye k e e , and C iu ju u = Fyu k u u .
i j
i j
In fact this form also holds for all mixed cases with only e/u field labels, except when the
boundaries x, z are of n-type and y is of i-type, when we obtained (46).
5. Extension to bulk theory

The bulkboundary couplings (a)Bi k can be determined by analysing the sewing
(xyx)1
constraint (5) in a rewritten form using the assumption that all C i i are nonzero for
fields present in the model (see Section 4):
X
X
X
1
(abb)q
(aab)q
(a)
(b)
Bi ` C p ` =
Bi k C k p
ei(2hm2hi hp hq + 2 (hk +h` ))
k,
Fkm

q
i
Fm`
.
i p
p q
(49)
Suppose now there is a boundary condition a and a bulk field i s.t. i does not couple
to any boundary field on the boundary a, i.e., (a)Bi k = 0 for all k with nka a 6= 0. Then,
roughly speaking (this is discussed in more detail below), since the r.h.s. of (49) is now
zero, all Bs involving the bulk field i are zero. Since any n-point function has a limit in
which we take all bulk fields to the boundary, and hence in this limit the factors in front of
the conformal blocks contain the product of the Bs of all bulk fields, any n-point function
involving the field i is identically zero. This in turn just means that the bulk field i is not
part of the model and can be left out.
Using this idea we can now determine the maximal bulk field content consistent with
(49). This does not prove that the maximal field content has to be modular invariant, but it
turns out that it is indeed the D-series bulk field content derived in [4].
To make this more precise, the classification of boundary conditions and their field
content in [9,10] establishes a one-to-one correspondence between boundary conditions
and diagonal bulk fields. In this approach specifying the possible boundary conditions is
equivalent to giving the diagonal part of the bulk partition function, and from [4] one knows
which off-diagonal parts have to be added to make it modular invariant. The statement thus
is that, at least of the the A- and D-series, the bulk field content obtained by this procedure
is the maximal consistent one.
5.1. Bulk field content
For an arbitrary boundary x consider (49) with a = , b = x, p = x , q = x . The `
sum reduces to one element and we get:
577
Bi ` C (xx)x
= ()Bi 1 C (x)x
( Fs )|k=1
x `
1x
(x)
+ ()Bi o C (x)x
o x ( Fs )|k=o .
(50)
First note that from an argument similar to (20) and (21) we see that all boundary
structure constants that appear in (50) are nonzero if they are allowed by fusion and the
even/odd coupling rule (see Section 4.1).
Suppose a bulk field i does not couple to the -boundary, i.e., ()Bi 1 = 0 and
()B o = 0. Then the r.h.s. of (50) is identically zero and since the boundary structure
i
constant on the l.h.s. is nonzero it follows that (a)Bi k = 0 for all a, k . As we argued
before this implies that all correlators involving i vanish and the field can be removed
from the bulk theory. Hence for any bulk field i we need at least one of ()Bi 1 and ()Bi o
to be nonzero.
The bulk field i transforms in a tensor product of two representations i . Suppose i
is a field with spin, i.e., i 6= . Then i and cannot fuse to the identity and hence it follows
that ()Bi 1 = 0. So for the field i to exist we need ()Bi o 6= 0. This is only possible if
the fusion i, o exists, i.e., only pairs with Kac-labels i = (r, s) and = (r, q s) are
possible. Now consider (50) with x = , ` = x = o , x = 1. We obtain (compare also
to (16)):
Bi o = ()Bi o e2i(hi hi ) = ()Bi o e2i( 2 r)( 2 s) .
()
(51)
For the exponential to be +1 we need q/2 s to be even (recall that for the D-series p is
always odd and q always even).
Next suppose that the bulk field i is diagonal, i.e., i = . Let i have Kac-labels (r, s).
As before denote by i the representation with Kac-labels (r, q s). Consider (50) with
x = , ` = 1, x = x = o . The F-sum in (49) reduces to m = i and we obtain:
Bi 1 = ()Bi 1 e2i(hi
()
h h )
o
i
= ()Bi 1 eis .
(52)
Hence only diagonal bulk fields with representation i = (r, s) where s is odd are present.
Recall that by convention r is always odd.
Up to now we have determined which pairs of representations are possible in the bulk.
Now we have to determine their multiplicities. The maximal multiplicity a bulk field can
have is equal to the number of boundary fields it couples to on the -boundary. This can
be seen as follows: suppose two bulk fields i , i in the same representation i can due
to the fusion rules only couple to one field on the -boundary. Then we can define
6= 0 and ()B = 0.
two new bulk fields as linear combination of i , i s.t., say, ()Bi,new
i,new
Now the new i field does not couple to the -boundary at all and can be removed from the
theory. A similar argument forbids more than two bulk fields in the same representation that
by fusion can couple to the two fields on the -boundary. The only bulk-representations
that allow coupling to both fields on the -boundary is the pair i i where i has Kac-labels
(r, q/2) and only in this case multiplicity two can occur.
Table 2 summarizes the maximal bulk field content consistent with a D-type cylinder
partition function. Each representation in the table occurs once for each possible fusion.
The only situation in which multiplicity arises is for pairs of representations that can fuse
578
Table 2
Maximal bulk field content for D-series
i, fuse to
Representations
(i, ) with i = = (r, s), where r and s odd
(i, ) with i = (r, s), = (r, qs), where r odd and s q/2 even
to both 1 and o , in which case we have multiplicity two. Note that as mentioned before
this can only happen for a diagonal bulk field with both representations of the form (r, q/2)
and when q/2 is odd, i.e., in Deven models.
Note that the same line of argument also works for the A-series. Here the situation is
much simpler, since there is only one field, i.e., the identity, on the 1-boundary. Hence
only diagonal bulk fields are allowed and only multiplicity one can occur. This is precisely
the A-series modular invariant bulk field content.
5.2. Bulkboundary couplings
As a first step to determine the bulkboundary couplings we will choose a basis of
primary bulk fields s.t. each field couples either to 1 or to o on the -boundary, but not to
both simultaneously.
As we have seen in the previous section, this demand is automatically enforced by
the fusion rules except for diagonal bulk fields in the representation i i where i has
Kac-labels (r, q/2) and we are looking at a Deven model. In this case the corresponding
bulk fields have multiplicity two and similar to the argument for redefining the boundary
structure constants in Section 4.3 we can take linear combinations of the two bulk fields
to obtain the desired coupling rule. We will denote bulk fields that couple to 1 on the
-boundary as even and bulk fields that couple to the boundary field o as odd. The new
bulkboundary couplings then fulfill ()Bie o = 0 for the even bulk fields ie and ()Bio 1 = 0
for the odd fields io .
We are still free to rescale all the bulk fields i . We use this freedom to set (the odd
diagonal fields are normalised differently so that the (x)Bi 1 represent the Pasquier algebra
as seen in Section 5.6):
Bio o =
()
Bi S1 i
,
A S1 1
Bie 1 = ()Bio o =
()
S1 i
S1 1
Bio 1 = ()Bie o = 0,
()
for diagonal odd fields io (i.e., i = ),
ei 2 (hi hi ) ,
for all other cases,
from the coupling rule.
(53)
With this redefinition of the fields the sum on the r.h.s. of (50) reduces to one term. Using
(50) with = and k = the general bulkboundary coupling takes the form:
Bi ` = ()Bi
(x)
exp i 2hm 2hi hx hx + 12 (h + h` )

F m
x
Fm`
i x
1,

1
i
Cx ,
,
B
x x
x
0,
579
= e and = e or u,
= o and = o,
= o and = u,
otherwise,
(54)
where = e if x is an i-type boundary and = u if x is an n-type boundary.

Note in particular that any bulk field can couple to a u-field on an n-type boundary, but
for an i-type boundary an even bulk field can only couple to an even boundary field and
an odd bulk field to an odd boundary field. This indicates that the Z2 -symmetry of the
boundary structure constants for i-type boundaries carries over to the bulk.
5.3. Bulk structure constants
The bulk structure constants can now be obtained from (6). Take a = , the sum on the
l.h.s. then reduces to the term where ()Bm k 6= 0, i.e., k = . On the r.h.s. the sum over
p and q reduces to p = , q = . The r-sum is also reduced to one element. We are
left with:

Ci j m = exp i 2 h + h h + 2(hj hr ) + hm h m hi + h i hj + h j

()B ()B
i
j
u u

r

F
F
F
F
, (55)
u
r
m
rm
()B

i j
m
j
m
where r = j if = e and r = j if = o. The first F-matrix element in (55) implements
an even/odd coupling rule for bulk fields, i.e., the only combinations for {, , } which
can be nonzero are {e, e, e} and {e, o, o}.
This implies that apart from other symmetries the bulk structure constants may have
there is a manifest Z2 -symmetry which sends even bulk fields to themselves and odd one
to minus themselves. Together with Section 4.5 and Eq. (54) we can now conclude that any
correlator with no or only i-type boundaries is manifestly invariant under the Z2 -symmetry
e e and o o applied to bulk and boundary simultaneously.
It is useful to know the bulk two-point functions in the present normalisation of fields.
From the explicit expression in (55) we get:
Ci i 1 =
S1 i
S1 1
Cio io 1 = A Fi1
for diagonal fields i (i.e., i = ),

o o S1 i
1 (1)s(io ) ,

S1
for nondiagonal fields io ,
(56)
where s(i ) = hi h i Z is the spin of the bulk field i .

In [16] it was shown that the minimal model bulk structure constants in the A- and
D-series are related by rational numbers, called relative structure constants. Taking the
explicit expression (55) together with the normalisation (56) we find (numerical) agreement
with these results. Also, for unitary models the signs of the two-point structure constants
580
in (56) are the same as in [16], where they were shown to lead to real bulk structure
constants (in the unitary case).
Note that if all bulk fields are even the solution (55) takes the same form as in the
A-series:

1
j i
me
.
(57)
Cie je = Fm1
i j
Together with Eqs. (48) for the boundary structure constants and (54) for the bulk
boundary couplings this illustrates another interesting point. If one considers the even
fields alone, that is all bulk and boundary fields that are invariant under the Z2 -symmetry
mentioned above, they form a subalgebra (as the coupling e, e o is not allowed) and the
structure constants are identical to the A-series in the following way:
Consider the A-series boundary theory associated to the pair of diagrams (Ap1 , Aq1 ).
In this theory consider only the boundary conditions (, ) with < p/2 and < q/2 or
> p/2 and > q/2 (compare to (15)) and the bulk fields i i where i has Kac-labels
(r, s) with both r and s odd. Then this is a closed subset of fields of the A-series theory
and in the normalisation chosen its structure constants coincide with those of the invariant
fields of the D-series theory (Ap1 , Dq/2+1 ). This correspondence seems natural form the
point of view that the invariant part of the D-diagram is an A-diagram.
5.4. Real structure constants
In the normalisation chosen in this paper the F-matrices are all real. This implies first of
all that the boundary structure constants as given in Section 4.5 are all real.
For the bulk structure constants we have seen in (57) that they are real as long as only
even fields are involved. If odd fields are present consider (55) in the case Cio jo me . The
phase factors cancel and we are left with a real expression. All other cases are real as well,
as can for example be seen from the three-point function Ci j k Ck k 1 = Cj k i Ci i 1
together with the fact that all two-point structure constants (56) are real. Thus in the present
normalisation all bulk structure constants are real, both for unitary and non-unitary models.
For the bulkboundary couplings in (54) we can apply the identity (B.8) to obtain the
result:

(x)
B(i ) ` = (x)B( i) ` .
(58)
This means that complex conjugation relates the bulkboundary coupling for a bulk field of
spin s to that of the field of spin s. In particular all bulkboundary couplings for diagonal
bulk fields are real.
It is in general not possible to choose a basis of primary fields s.t. all structure constants
are real. To see this we construct a gauge invariant expression from the sewing constraints
which cannot be fulfilled by real structure constants. Consider for example two diagonal
bulk fields i and m and a field j with spin one. Take (6) with k = 1 and a = . This
forces p = q = o and r = . We get:

Ci j m ()Bm 1 = ei(/2) Fo m

i () o () o ()1
Bj C o o .
Bi
i j
581
(59)
The F-matrix entry is real and will in general be nonzero, provided that all fusions are
allowed. As discussed in Section 5.1 the spin one field j has to couple to o (i.e.,
()B o 6= 0) and we can find a diagonal field i that couples to . The boundary twoj
o
point function has to be nonzero by assumption (b) in Section 4. Hence the r.h.s. is nonzero
and thus some of the D-series structure constants will have nonzero imaginary part if they
solve the sewing constraints.
We can however consider the following non-primary basis of bulk fields to obtain real
structure constants: diagonal (even or odd) bulk fields stay as they are and for a pair s ,
s of nondiagonal bulk fields with spins s we define a new set of fields as follows:
1
r = (s + s ),
2
i =
1
(s s ).
2i
(60)
The new fields r and i are no longer primary, but one finds that the coefficients
describing their behaviour under arbitrary conformal mappings are all real. Since the OPE
is determined from the transformation behaviour of the fields one expects the coefficients
appearing in the OPE to be real as well.
One can see explicitly from (58) that all bulkboundary couplings (x)B r and (x)B i
are real and one can verify numerically that all bulk structure constants are real in the new
basis as well. This holds for both unitary and non-unitary models.
5.5. Discrete symmetries of the structure constants
Denote the set of bulk and boundary structure constants given in (43)(47) and (53)(55)
with A. We see that A depends explicitly on the assignment of e/o-labels in Section 3.3
and the ordering of the boundary conditions in Section 3.4.
Denote the set of structure constants obtained by the same procedure, but with a different
initial assignment of e/o-labels in Section 3.3 as B. It was shown in this paper that any
solution to the sewing constraints with the D-series field content can be brought to the
form A. In particular there is a change of basis in the set of primary fields that maps B to A.
For the primary fields in A we considered a Z2 -action that takes even fields to themselves
and odd fields to minus themselves. This action leaves the structure constants involving no
or only i-type boundaries invariant.
However a different initial e/o-labelling leads to a different action of Z2 on the boundary
fields. The Z2 -action of B does via the change of basis induce a Z2 -action on A which is
different from the original one. It follows that each distinct e/o-assignment results in a
distinct Z2 -action on A.
Going back to Section 3.3 we see that for each i-type boundary x 6= we are free
x we call xe and which xo . The
to choose which of the two fields living on
specific assignment made in Section 3.3 was motivated on physical grounds.
There are altogether N = 2(p1)(q2)/41 distinct assignments. Thus among other
symmetries the structure constants may have, there are N Z2 s acting on the set of primary
582
fields that leave the subset of the structure constants in A that involve no or only i-type
boundary conditions invariant. One can convince oneself that the Z2 s all act in the same
way on the bulk fields but can be distinguished from their action on the boundary fields.
The collection of Z2 s just presented does not necessarily exhaust all discrete symmetries
as can be seen for example in the three-states Potts model which has an additional Z3 symmetry of whose Z2 subgroups only one is contained in the above list.
5.6. Disc partition function and Pasquier algebra
Up two now all constraints on the structure constants were derived in such a way that
(x)
the vacuum expectation values h1iUHP cancelled. We will next derive some constraints
for the vacuum expectation values of the unit disc h1i(x)
D . Consider the two-point function
(xy) (yx)
hi i i on the unit disc. Taking different limits leads to the constraint:
(xyx)1
(x)
(yxy)1
(y)
C i i h1iD = C i i h1iD .
(61)
(y)
The vacuum expectation values h1i(x)

D , h1iD can be interpreted as the partition function of
the unit disc with different boundary conditions. This expression thus relates the partition
functions of different boundary conditions in the same geometry. The relation (61) remains
true for different (genus zero) geometries, but the actual values of the partition functions
do depend on the shape of the system (see, e.g., [25]). For example the partition function
c/6 , where c is the central charge [25].
for a disc of radius r is given by h1i(x)
D r
Substituting the explicit boundary two-point functions into (61) we arrive at the
following expressions (let x, y be i-type boundaries and a, b be n-type boundaries):
(y)
h1i
h1i(x)
D
= yDe ,
S1 xe
S1
h1i(x)
h1i(a)
D
D
=
2
,
x
S1 e
S1 au
h1i(a)
h1i(b)
D
D
=
.
a
S1 u
S1 bu
(62)
In [9,10] the matrices ni that descibe the boundary field content are simultaneously
diagonalised in the following way:
nix y =
X Si j
x,j (y,j ) ,
j
S
1
j diag
(63)
where the sum runs over all (even or odd) diagonal bulk fields and the vectors x form a
complete orthonormal basis.
Replicating the line of argument in Section 8 of [17] in the present notation we get the
following expression for the matrices ni :
nix y =
X Si j
S1 j
(y)
(x)
(x)
Bj 1 h1iD (y)Bj 1 h1iD .

j
1
S
Cj j h0|0i
j diag 1
(64)
Using (56) and normalising the bulk vacuum:

h0|0i = S1 1
(65)
we find the following relation between the bulkboundary couplings and the vectors x :
x,j = (x)Bj 1 h1i(x)

D.
583
(66)
By evaluating (66) for j = 1 and recalling that (x)B1 1 = 1 we get:

h1i(x)
D = x,1 .
(67)
Furthermore we see that the the bulkboundary couplings to the identity on the boundary
are given by a ratio of s [9,10]:
Bj 1 =
(x)
x,j
x,1
(68)
In the present normalisation the Bs form one-dimensional representations of an algebra,

a fact that was used to classify possible boundary conditions in [610]. The algebra was
identified as the Pasquier algebra [26] in [9,10]:
X
(x)
Bi 1(x)Bj 1 =
Mi j k (x)Bk 1 ,
(69)
k
where Mi j k = x x,i x,j (x,k ) /x,1 .

From the relations (67), (68) and (62) together with the constraint that the vectors x
form an orthonormal basis we get the following expression:
2 S1 xe , if x is of i-type,
(x)
1
x,1 =
(70)
x,i = Bi x,1 ,
1
S1 xu , if x is of n-type.
2
Since the bulkboundary couplings are known from (54) one can now check that the s
so defined are all real and verify numerically that they fulfill (63), with nix y on the l.h.s.
taken from Section 3.3.
6. Conclusion
The aim of this paper was to find all structure constants of the D-series minimal model
boundary conformal field theory and to point out some of their properties.
The structure constants can be calculated in several steps. The basic ingredient are the
fusion matrices, as all structure constants are expressed in terms of these. A procedure
to obtain the fusion matrices recursively is given in [17], together with references to the
original literature. The matrices can also be extracted by rescaling explicit expressions
found, e.g., in [2022] to match the normalisation fixed in Appendix A. As a first step
the D-series field content of the boundary theory is computed, as described in Section 3.3.
Next all boundary structure constants are obtained from the results collected in Section 4.5.
The boundary theory is now fully determined. The extension from the boundary to the
bulk recovers the familiar D-series bulk field content as the maximal set of bulk fields
consistent with the boundary theory (see Section 5.1). Last, the remaining two sets of
structure constants, the bulkboundary couplings and the bulk structure constants are given
in (53), (54) and (55).
584

(xyz)t
The boundary structure constants C r s and the bulk structure constants Ci j k

obtained by this procedure are all real, both for unitary and non-unitary minimal models.
However some of the bulkboundary couplings (x)Bi ` are imaginary. There is in general
no basis of primary fields such that all structure constants real. In Section 5.4 a non-primary
basis of fields was presented in which all three sets of structure constants are real.
The solution described above was obtained by deriving a necessary form of the structure
constants. It was shown that any solution to the sewing constraints (4)(7) for the Dseries field content can be brought to this form. So just as for the A-series it follows
that the structure constants are unique up to redefinition of the fields. It was however not
analytically shown that the given structure constants do indeed solve the full set of genus
zero sewing constraints, since they are overdetermined and only a subset was used for the
actual solution. But numerical tests carried out for several minimal models where each of
Eqs. (4)(7) was checked repeatedly with a different random selection of fields show no
contradiction.
It was also found that a subset of the D-series structure constants have a manifest Z2 symmetry: if no or only i-type boundaries are present in a correlator it is invariant under the
change eveneven and odd odd. This symmetry can be associated to the Z2 symmetry of the D-diagram in the sense that the boundary conditions fall in two classes:
the i-type boundaries are associated to nodes left invariant by that symmetry, and the ntype boundaries to nodes which are not. We have also seen that the even fields alone form
a closed subalgebra with structure constants equal to the corresponding ones of an A-type
theory.
In deriving the structure constants we have used only the Virasoro algebra, and not
considered the fact that it could be part of a larger symmetry algebra, as is the case for
Deven models. From the present point of view, curious properties of the solution such as
the boundary field content, the even/odd coupling rule or numerical coincidences in the
structure constants, are obtained as results of the calculation, but lack an explanation in
terms of the additional structure present in the theory. One can use the maximally extended
chiral algebra as a starting point for the analysis of a boundary conformal field theory (see,
e.g., [27] and references therein) and it would be interesting to interpret the explicit results
presented here in this more general framework.
In Section 5.5 it was pointed out that an initial freedom in the labelling of the boundary
fields actually implies that the just mentioned Z2 is part of a collection of 2(p1)(q2)/41
Z2 s, each acting in a distinct way on the set of primary boundary fields. This action leaves
structure constants with no or only i-type boundaries invariant.
Finally the vacuum expectation values h1i(x)
D of the unit disc with different boundary
conditions were computed and it was pointed out that in the present normalistaion the bulk
boundary couplings (x)Bi 1 form one-dimensional representations of the Pasquier algebra.
To make contact with different conventions in the literature the bulk and boundary fields
may have to be redefined. The normalisations chosen in this paper can be obtained from
(56) in terms of the bulk two-point functions, or from (53) in terms of the bulkboundary
couplings to the identity field on the boundary.
585
Acknowledgements
This work was supported by the EPSRC, the DAAD and Kings College London. I am
especially grateful to my supervisor G.M.T. Watts for suggesting the project, encouragement, criticism and helpful conversations. Furthermore I wish to thank V.B. Petkova, P. Ruelle and J.-B. Zuber for useful discussions.
Appendix A. Derivation of the five-point constraint

This appendix was included to illustrate how the notation and techniques for calculating
with conformal blocks described in [12] were used to rederive the sewing constraints in
[11].
The sewing constraints arise from taking two different limits in a given correlator where
the fields approach each other or the boundary and demanding that the two sets of structure
constants and conformal blocks give rise to the same function.
Conformal blocks associated to different asymptotic regiems can be transformed into
each other using the fusion- or F-matrices. The F-matrices are defined as follows:
j
q
j
k
p
i
(z)
`
(w)
X
q

Fpq
j k
i `
k
(z w)
`
(w)
(A.1)
The conformal blocks appearing in (A.1) solve the linear differential equations associated
with the (chiral) correlator hi|j (z)k (w)|li.
The normalisation convention for conformal blocks chosen in this paper is (the dots
stand for a regular expression in w):
j
k
p
i
(1)
`
(w)
= whp hk h` (1 + ),
j
q
k
(1 w)
`
(w)
= (1 w)hq hj hk (1 + ).
(A.2)
The example we consider is the derivation of the constraint resulting from taking
different limits in the correlator involving two bulk fields i , j and one boundary field
on the upper half plane with the boundary condition labelled a.
k(aa)
586
In the first limit we take the two bulk fields to the boundary and are left with a threepoint function on the boundary. In the second limit we start by taking the OPE of the two
bulk fields and then take the remaining bulk field to the boundary. Let z = xz + iyz and
w = xw + iyw . The asymptotic behaviour in the two limiting cases is then given by:

(aa)
k (x)
i (z, z )j (w, w)

XX
(aaa)k
(aaa)1
(a)
(a)
Bi p (a)Bj q C p q C k k h1iUHP
yz 0,
yw 0
p,q
(2yz )hp hi hi (2yw )hq hj hj (xz xw )hk hp hq

(xz x)hq hp hk (xw x)hp hk hq

XX
(a)
m (a)
k (aaa)1
Ci j
Bm C k k h1iUHP
zw,
z w
m,m
hm hi hj (2yw )hk hm hm (xw x)2hk .

(z w)hm hi hj (z w)
(A.3)
On the other hand the correlator can be expressed as a linear combination of conformal
blocks. We use two sets of conformal blocks to express the correlator in two different
ways, one associated to each asymptotic behavour in (A.3). We obtain the following two
linear combinations:

(aa)
k (x)
i (z, z )j (w, w)
i
cp,q
w w
p,q
z z
w
i
j
m
zw
z w
w w
dm,m
m,m
0
w
0
x
(A.4)
Taking the limits calculated in (A.3) for the exact expressions in terms of conformal blocks
(A.4) relates the coefficients cp,q and dm,m to the products of structure constants obtained
by applying the OPE. The precise relation is:
X
(aaa)k (aaa)1
(a)
(a)
Bi p (a)Bj q C p q C k k h1iUHP ,
cp,q ei 2 (hp +hq hi hi hj hj ) =
,
dm,m e
i 2 (hk hm h m )
(a)
k k h1iUHP .
(A.5)
587
The phase factors originat from relating z z = 2iyz to 2yz , etc. The sums over ,
and in (A.5) take care of fields with multiplicites. Fields that transform in the same
representation of the Virasoro algebra show the same asymptotic behaviour and cannot be
discriminated by conformal blocks. Hence their structure constants occur as a sum in front
of the according conformal block.
The two sets of conformal blocks in (A.4) are related by a basis transformation. This
transformation can be carried out in several steps making use of two basic moves of
braiding and fusion implemented by the B- and F-matrix (see [12] for details). One possible
way to perform the basis transformation is as follows:
i
z z
w w

Fqr
r,m
Fqr
r
z
0
x
j
zw
z w
k
k
r
w

k () j

B
Fr m
p j pm i r
m k
i
p
z
i
i
z
k () j
B
p j pm i r
r,m,m
X j 1 0
=
F
p k qr
r
i(hi +hr hp hm )
0
x
Fqr
r,m,m

k
r

Fpm
Fr m
p j
i j
m k
i
j
m
zw
k
w w
z w
0
w
0
x
(A.6)
588
Putting (A.4)(A.6) together we recover the sewing constraint (6):

X
(a)
k k h1iUHP
XX
p,q
(a)
Bi
p (a)
Bj
(aaa)k (aaa)1
(a)
C p q C k k h1iUHP
X
r
ei 2 (hk +hp hq 2hr +hm hm hi +hi +hj +hj )

Fqr

k
r

Fpm
Fr m
.
p j
i j
m k
(A.7)
The summation range for r is principally over all entries in the Kac-table (quotiened by
Z2 ) but the F-matrix entries will only be nonzero if the following three fusions are allowed:
j
p
(A.8)
Appendix B. Some F-matrix identities

The following constitutes a collection of F-matrix identies for Virasoro minimal models
used in the paper. They are either directly taken from [12] or special cases thereof. Note that
for the S-matrix we have S1 i = S1 i where the -operation was defined as i = (r, q s)
if i has Kac-labels (r, s).

j l
() j k
i(hi +h` hp hq )
Bpq
Fpq
=e
,
(B.1)
i `
i k

Fpq

j k
i `
` i
= Fpq
= Fpq
,
i `
j k
k j

Fpr

F1k

b c
d c
Frq
= p,q ,
a d
a b

S1 1 S1 k
i j
i i
Fk1
,
= i
i j
j j
S1 S1 j
(B.2)

xi o
b o
Fxi b
= 1,
a b
a xi
(B.3)

xe o
o o
Fxo 1
= 1,
xe o
xe xe
(B.4)

Fb xi

F1xo

S1 k
i i
k k
o o

Fxo 1
= j Fj 1
,
= Fxe 1 o o ,
xe xe
xo xo
j j
i i
S1

S1 1
i i
o o
F11
F11
= i,
= 1,
o o
i i
S1
Fk1

Fpi

j k
i i
i j
k k
Fn1
Fp1
= Fnk
,
n `
``
`p
` `
(B.5)
(B.6)
(B.7)

i p
q p
Fs`
i q
i i
s

X
1
p i
p q
ei(hp hq 2hi +2hm + 2 (hk +h` )) Fkm

Fm`
.
=
q i
i i
m
ei(hp +hq +2hi 2hs 2 (hk +h` )) Fks
589
(B.8)
References
[1] A.A. Belavin, A.M. Polyakov, A.B. Zamolodchikov, Nucl. Phys. B 241 (1984) 333380.
[2] Ph. Di Francesco, P. Mathieu, D. Snchal, Conformal Field Theory, Springer, 1998.
[3] Vl.S. Dotsenko, V.A. Fateev, Nucl. Phys. B 240 (1984) 312348; Nucl. Phys. B 251 (1985)
691734.
[4] A. Capelli, C. Itzykson, J.-B. Zuber, Nucl. Phys. B 280 (1987) 445465.
[5] J.L. Cardy, Nucl. Phys. B 240 (1984) 514532.
[6] J.L. Cardy, D.C. Lewellen, Phys. Lett. B 259 (1991) 274278.
[7] G. Pradisi, A. Sagnotti, Ya.S. Stanev, Phys. Lett. B 356 (1995) 230238; Phys. Lett. B 381
(1996) 97104.
[8] J. Fuchs, C. Schweigert, Phys. Lett. B 414 (1997) 251259.
[9] R.E. Behrend, P.A. Pearce, J.-B. Zuber, J. Phys. A 31 (1998) L763L770.
[10] R.E. Behrend, P.A. Pearce, V.B. Petkova, J.-B. Zuber, Phys. Lett. B 444 (1998) 163166.
[11] D.C. Lewellen, Nucl. Phys. B 372 (1992) 654682.
[12] G. Moore, N. Seiberg, Lectures on RCFT, Physics, Geometry, and Topology, Plenum Press,
New York, 1990.
[13] R.E. Behrend, P.A. Pearce, V.B. Petkova, J.-B. Zuber, hep-th/9908036.
[14] J. Fuchs, A. Klemm, Ann. Phys. 194 (1989) 303.
[15] V.B. Petkova, Phys. Lett. B 225 (1989) 357362.
[16] V.B. Petkova, J.-B. Zuber, Nucl. Phys. B 438 (1995) 347372.
[17] I. Runkel, Nucl. Phys. B 549 (1999) 563578.
[18] W. Eholzer, M. Flohr, A. Honecker, R. Hubel, W. Nahm, R. Varnhagen, Nucl. Phys. B 383
(1992) 249290.
[19] P. Bouwknegt, K. Schoutens, Phys. Rept. 223 (1993) 183276.
[20] L. Alvarez-Gaum, C. Gomez, G. Sierra, Phys. Lett. B 220 (1989) 142152.
[21] G. Felder, J. Frhlich, G. Keller, Comm. Math. Phys. 124 (1989) 647664.
[22] P. Furlan, A.Ch. Ganchev, V.B. Petkova, Int. J. Mod. Phys. A 5 (1990) 27212735.
[23] J.L. Cardy, Nucl. Phys. B 324 (1989) 581596.
[24] P. Ruelle, hep-th/9904100.
[25] J.L. Cardy, I. Peschel, Nucl. Phys. B 300 (1988) 377399.
[26] V. Pasquier, J. Phys. A 20 (1987) 57075733.
[27] J. Fuchs, C. Schweigert, hep-th/9902132, hep-th/9908025.

The AharonovBohm problem revisited

Yoichiro Nambu
The Enrico Fermi Institute, University of Chicago, USA
Received 10 January 2000; accepted 27 April 2000
Abstract
Explicit solutions are constructed, and their properties investigated, for a nonrelativistic charged
particle in two dimensions in the presence of an arbitrary number of nonquantized magnetic vortices
in free space as well as in a uniform magnetic field. After eliminating the gauge potential, the vortices
are represented as branch points in one of the complex coordinates. Multivortex solutions in free
space can be obtained only if the vortices are treated as dynamical objects. But all the solutions
suffer from some unphysical properties. The formulas may be generalized to describe a system of
anyons. 2000 Elsevier Science B.V. All rights reserved.
PACS: 03.65.Bz; 04.50.+h; 05.30.Pr; 11.10.Kk; 11.27.+d
Keywords: AharonovBohm; Magnetic vortex; Gauge field; Nontrivial topology; Anyon
1. Introduction
In a classic paper, Aharonov and Bohm [1] have pointed out that a locally trivial vector
potential of a gauge field can lead to observable effects in a nontrivial topology through
the phase of the wave function of a charged particle. This has spawned many important
theoretical concepts in later developments in gauge theories, like instanton, monopole,
and confinement by monopole condensation. On the experimental side, confirmation
of the original AharonovBohm (AB) effect due to a nonquantized magnetic flux
tube has been somewhat controversial because of the difficulty of setting up ideal
conditions, but there have been several experiments, notably those by Tonomura et al. [2]
showing the shifts in diffraction fringes due to magnetic vortices, as expected by the
AB effect. (See [12] for a general review of the problem.) The present work was
motivated by the theoretical question: what would be the properties of a medium filled
with many nonquantized magnetic vortices? Assume, for simplicity, that the vortices are
all parallel so that problem can be reduced to that of an electrically charged particle
in a 2-plane pierced by pointlike magnetic vortices. Then, for example, what would
be the behavior of a charged particle in a lattice of such vortices? What would be
its energy spectrum? Would the vortices cause more drastic effects than a phase shift
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 2 5 8 - 3
Y. Nambu / Nuclear Physics B 579 [FS] (2000) 590616
591
in the wave function? Although the semiclassical arguments suffice to discuss and
analyze most situations realistically, it may require a more rigorous treatment of the
Schroedinger equation to answer these questions. This turns out to be a highly nontrivial
problem. To our knowledge, the eigenfunctions of the Schroedinger equation in the
presence of two or more vortices have not been explicitly constructed. A qualitative
discussion of special cases was made by Peshkin et al. [3]. The path integral formulation
of the problem [47] provides good general insight, but it is not very helpful for
our purposes. Some other theoretical problems related to the present one have been
discussed by Aharonov and Casher [10], Lewis [17] Dubrovin and Novikov [11], and
Jackiw [14].
We develop here a method of solving the many-vortex problem by adapting the work of
Sommerfeld [8,9]. He solved the problem of diffraction of light by a semi-infinite wall in
two dimensions by regarding the wall as a branch cut in a two-sheeted Riemann surface.
The incoming plane wave in the physical plane dives into the unphysical second sheet as it
hits the cut, while the reflected wave emerges from the second sheet through the cut. The
solutions were constructed as a contour integral of a kernel in a complex angular variable
over the two sheets.
The AB problem can be posed in a similar fashion. The pure gauge vector potential
of a vortex is singular at the vortex site. In the complex variables z = x + iy it behaves
like (/2)(i/z, i/z), where 2 is the vortex strength. After removing the potential by a
gauge transformation, we get a wave function that satisfies a free Schroedinger equation,
but it is singular in the sense that the function acquires a phase = exp(2i) after
encircling the vortex once, so the wave function must vanish at the vortex if is noninteger. Regarded as a function in the complex 2-plane, the problem then reduces to that
of solving the free Schroedinger equation on a Riemann surface where the vortex is a
branch point, and the solution must satisfy the boundary condition that it has a fractional
angular momentum + n around it in such a way that the function vanishes at the
singularity.
The paper is organized as follows. We first establish the basic formalism of complex
integral representation, apply it to rederive the known results for the case of a single vortex
with or without the presence of a uniform magnetic field [1,17]. Then the problem of
many vortices is addressed by two methods which yield different types of solutions. One
of them is to use multiple integral representations, and is applicable only if a magnetic
field is present. The other is to treat the vortices as dynamical objects having properties
dual to the charges. An important general property that emerges is that the nontrivial phase
information around all vortices must be represented as branch points in either the z or the
z coordinate, but not a mixture of them. It means that, since the wave function must vanish
at the vortex sites, the (fractional) angular momentum around each vortex, as opposed
to the intrinsic vortex strength, has the same sign, either positive or negative. Thus for
example, if the two vortex strengths are 1 > 0 and 2 < 0, the solution is constructed
as having positive angular momenta 1 + l1 > 0 and l2 + 2 > 0, li = integer, in the z
variable (and similarly negative angular momenta in z ). Otherwise the solution would have
to be represented by a function in z in a region around one vortex, and by a function in z in
592
another, but it is not possible to match the values of the wave functions and their derivatives
at the interface. 1
The fact that angular momenta around all the vortices must have the same sign is not
intuitively clear, but may be inferred from the zero energy solutions, where the nontrivial
information about all the vortices must be encoded in terms of either analytic or antianalytic functions. It is also understandable from the following consideration. Suppose
1 = 2 , and let them approach each other and collapse to nothing. The wave function
originally vanished at both sites and was finite elsewhere. So it will vanish at the point of
collapse, hence it will be in a p or higher angular momentum state.
These general features lead to some seemingly unphysical consequences, indicating a
pathological nature of multivortex systems as a mathematical problem. For example, the
asymptotic states are not complete: N 1 of the low angular momentum states are missing
in an N -vortex system. This prevents us from constructing a scattering amplitude, even
though we would expect it to look like that of a single vortex with combined strengths.
Furthermore, a lattice medium made up of many vortices tends to expel the charge as if it
was placed in an antimagnetic field, because the angular momentum of the charge at radius
R from the origin increases like R 2 .
In constructing explicit solutions, we also uncover other puzzling features. It is found
that, whereas multivortex solutions in a magnetic field can be found easily, it is impossible
in free space unless they are treated as dynamical objects.
2. Basics
Let a vortex of strength be located at the origin. For an electron of charge -e, the
AharonovBohm gauge potential A and the covariant derivative D in the standard circular
gauge are given by

(Ax , Ay ) = (/2) y/r 2 , x/r 2
= (/2)(y , x ) ln(r),
A = 2 2 (x),

D ( iA) = /x + iy/ 2r 2 , /y ix/ 2r 2 .
(2.1)
For notational convenience, we are taking e = h = 1 so that integer corresponds to a

quantized vortex. In terms of the complex variables z = x + iy and z = x iy, they become
x = (z + z )/2,
y = (z z )/2i,
/z = (1/2)(/x i/y),
1 To see this, consider at zero energy two vortices of opposite signs located at ib. Since any analytic function
of z or of z satisfies the Schroedinger equation, assume: = f1 = (z ib) g(z) in the upper half plane
(y = =z > 0), where g(z) is regular, and = f2 = (z ib)1 g(z) in the lower half plane (y = =z < 0).
Analytically each can be extended to the whole plane. Obviously f1 = f2 on the real line, but their normal
derivatives are opposite and 6= 0. (If the derivatives were zero, the functions would vanish identically.) The same
conclusion is reached if the solution in each half plane is itself a sum of analytic and anti-analytic parts.
593
/ z = (1/2)(/x + i/y),

2 /x 2 + 2 /y 2 = 4(/z)(/ z ),
Az = i/2z,
Az = i/2z,
Dz = z + /(2z),
Dz = z /(2z).
(2.2)
The basic Schroedinger equation reads, in imaginary time (or energy eigenvalue) form,
after redefining mE/2 = k 2 /4 E, k being the momentum,

{Dz , Dz }/2 + = 0

(2.3)
{Dz , Dz }/2 + E = 0 ,
with the condition that is one-valued and finite. (We will not necessarily require
finiteness of derivatives, though.) Here the imaginary time is for later convenience, and
z and z are regarded as independent variables spanning a complex 2-dimensional plane.
The physical space is its subspace where z and z are complex conjugates of each other
(real xy plane). Hereafter we will refer to this situation as on shell. Since Eq. (2.3) has
independent first derivatives in z and z , the actual physical space also includes its tangential
neighborhoods. After solutions have been written down in z and z , however, we can stay
on shell by reverting to x and y. The ordering of Dz and Dz may be ignored since their
potential noncommutativity at their singularities will not arise. From Eq. (2.2) we see that
Eq. (2.3) is invariant under the operation
z z ,
(2.4)
It corresponds to a reflection y y. The interchange of z and z is also effected by

complex conjugation of the wave function which corresponds to time reversal.
We next eliminate from Eq. (2.3) the gauge potential by a singular gauge transformation G:
= G,
G = (z/z)/2 ,
(2.5)
so that now satisfies a free Schroedinger equation:

(z z + t ) = 0

(z z + E) = 0 .
(2.6)
Since the original is one-valued, must be singular in such a way as to cancel the
singularity in the gauge function in Eq. (2.5). This means that must be defined as a wave
function on a Riemann surface with a branch cut running from 0 to +, and on shell there
is a phase change as we go around the origin once:
( = 2) = ( = 0),
(z = z = 0) = 0
= exp(2i),
(2.7)
and subject to conditions of finiteness at infinity on shell. In the following we will mainly
be concerned with the eigenvalue equation. An elementary solution of Eq. (2.6) is
exp(zt E z /t),
(2.8)
594
where t is a complex momentum. The general solution of Eq. (2.6) that satisfies the
boundary conditions may be built up as a superposition of elementary solutions
Z
(2.9)
= exp(zt E z /t)f (t) dt
C
with an appropriate choice of the function f (t) and the integration path C in such a way as
to satisfy Eq. (2.7). We recognize the familiar integral representation of Bessel functions.
Now assume for the time being 0 < < 1, and let
f (t) = t n1 ,
n = 0, 1, 2, . . .
(2.10)
and choose the contour C of integration to be

C : U , (0+) ,
(2.11)
which goes around 0 (say, anticlockwise) starting from, and ending at, infinity in the
direction U = exp(i ) such that <(zt) < 0 in order to keep the integral convergent. This
means that, as the phase of z rotates by 2 , the contour integration will have to make a
counter-rotation by 2 , so the factor f (t) will yield a phase factor e2i as is required
by Eq. (2.7). Furthermore, stays finite on shell when z and z go to infinity, and also at
zero since
Z
(2.12)
t 1n dt = 0, n + > 0.
These properties can be made more explicit by a change of integration variable, t tE/z
(0+)
Z
=z
exp(Et zz/t)t n1 dt
n+
(2.13)
up to a constant factor, and where the contour is now (, (0+)). So Eq. (2.13) gives a
desired set of solutions for any and integer n = 0, 1, 2, . . . that satisfy n + > 0. In the
case n + < 0, we can choose a new contour around 0 in Eq. (2.9):

(2.14)
C 0 = U 0, (0) ,
i.e., one that starts from the origin in the direction U = exp(i ) such that <z/t > 0,
and comes back after encircling it clockwise. This is equivalent to the conjugate form
(z-type) of according to Eq. (2.4), which can be converted to the original contour by
the substitution in Eq. (2.10): t E/t, resulting in a new f (t)0 = t +n1 . Eq. (2.13) then
becomes, up to a constant factor,
= z
(0+)
Z
exp(Et + zz/t)t +n1 dt,
n + < 0.
(2.15)
In terms of Bessel functions, the z- and z -type solutions lead to solutions in the original
gauge:

n + > 0, z-type: = (z/z)n/2 J+n k(zz)1/2 ,

n + < 0, z -type: = (z/z)n/2 J|+n| k(zz)1/2 .
595
(2.16)
Thus the physical states are labeled by an integer angular momentum (in the original
gauge) n = 0, 1, 2, . . . that goes with Bessel functions of order | + n|. The angular
momentum operator L is given by zz z z . Eqs. (2.13) and (2.15) show that the nontrivial
phase information in is carried respectively by z and z . On the other hand, although we
assumed 0 < < 1, Eq. (2.16) can be seen to be valid for general . Applying the operators
z and z on a serves to change n by 1. This may be repeated any number of times as
long as the wave funtion remains finite, thereby generating a complete set of states.
Some remarks are in order here about the problems associated with non-single valued
wave functions and the limitations in taking derivatives. The first problem poses a difficulty
in defining a complete set of states. Although scalar products are independent of the choice
of (unitary) gauges, it does not make sense to expand a function with a branch point at
z = a in terms of functions with a branch point at z = b. Similarly, the meaning of going
to momentum space by Fourier transforms is not clear since it will depend on where cuts
are made. We should, therefore, go back to the original non-singular gauges to build a
Hilbert space of physical states. We may, however, work with non-single valued functions
in enumerating the states since there is a 11 correspondence between the two sets.
The second problem causes difficulty in defining higher powers of momentum as
observables (in any gauge), e.g., in a power series expansion of the finite translation
operator, since they will remain bounded and self-adjoint only up to a certain power. The
problem may be said to be due to our assumption of pointlike vortices, as are some other
problems, to which we will come later.
3. Scattering
Since the solutions Eqs. (2.16) are Bessel functions of non-integer order, we expect that
the scattering amplitude of an electron by the vortex can be obtained by a superposition
of all solutions labeled by n, which would form a complete set. However, since they all
vanish at the origin, they form a complete set only in the punctured plane R 2 0, hence it
would not be possible to construct a plane wave in R 2 with this set of functions. What
is missing is an equivalent of the s wave component that remains finite at the origin,
a situation analogous to the case of hard sphere scattering. But it is also different because
of the long range nature of the gauge potential as in the case of the Coulomb scattering.
A proper scattering amplitude can be constructed in the following way. Assuming 0 <
< 1, consider the integral
Z
exp(zt E z /t)f (t) dt,
=
C1 +C2
f (t) = t /(t tin ),
tin = iE 1/2.
Here tin is the z-momentum of the incoming plane wave moving in the x direction:
(3.1)
596
Fig. 1. Contours for scattering.

in = exp iE 1/2 (z + z ) = exp(ikx).
(3.2)
The contours C1 and C2 are defined as follows (Fig. 1). Denote by R0 the reference
Riemann sheet, and by Rn the sheet reached by going counter-clockwise around the
origin n times. For a given position z = r exp(i ) = rU , C1 starts from U , goes
counter-clockwise around the origin and tin , to end up at U on R1 . C2 starts
from 0 U , goes clockwise around the origin, but avoiding tin , to end up at 0 U
on R1 . These may be denoted respectively as (, (0+, (tin +))) and (0, (0)). Note that
the integral converges for either path because 0 < [] < 1. When the factor 1/(t tin )
is expanded, the two integrals cover all powers of t/t0 , i.e., all the momenta from
to +.
Some properties of Eq. (3.1) are obvious:
(1) The integral converges for z = z = 0, giving = 0.
(2) When is an integer, the two contours can be detached from their limits to form a
closed circle around tin , which then picks up the incoming wave in only, so there
is no scattering.
That the formula gives the correct scattering amplitude can be shown by computing its asymptotic form by the saddle point method. First take the general formula Eq. (2.9). Writing
the integrand of as exp S, its saddle points are determined by the extrema t0 of S(t), and
performing a Gaussian integration around them in the direction of the steepest descent:
S 0 = z + E z /t02 + /t0 1/(t0 tin ) = 0,
S 00 = 2E z /t03 /t02 + 1/(t0 tin )2 .
(3.3)
For large E 1/2|z| = E 1/2 r 1, there are three saddle points on a circle of radius E 1/2 : one
near the pole tin representing the plane wave, and a conjugate pair t+ , t representing radially outgoing and incoming waves respectively. (tin and t do not coincide except when
= 0 or .) For the radial components,
t iE 1/2 exp(i ),

+1
.
eS exp i 2E 1/2 R (2 + 1)/4 i/ 1 ei iE 1/2ei
The Gaussian integration around t yields
(3.4)

+ i exp i kR (2 + 1)/4 + i (/R)1/2 /2

i/2
for + ,
e / sin(/2) ,
iei/2 / cos(/2) , for .
597
(3.5)
The phase factor (1) and an overall sign must be determined according to the sense
of each contour around 0. The contours C1 and C2 are such that when the two contributions are added, the incoming component is canceled, leaving only the outgoing one. After
reverting to the original gauge, the scattering amplitude becomes

(3.6)
scat = sin() exp i kR (2 + 1)/4 / R 1/2 sin(/2) ,
which reproduces the known results [1,1316]. The formula is valid except for a nearforward region where the width of the Gaussian integration (k/R)1/2 becomes larger
than |tin t+| k .
Eq. (3.1) made use of the fact that 0 < < 1 so that the two integrals could have the
same integrand but different contours. In angular momentum, decomposition, they together
cover all values from to , which would be necessary to represent a plane wave.
For general = [] + l, we should substitute its fractional part [] in Eqs. (3.1)(3.5).
The scattering amplitude, Eq. (3.6) will then be modified in addition by a phase factor
exp(il ).
4. Presence of a magnetic field

4.1. Basic formulas
In this section we allow for the presence of a constant magnetic field in addition to the
vortices. The vector potential for a constant magnetic field B = /2 is
Az = z /2,
Az = z/2,
(4.1)
which is to to be added to the AharonovBohm potential. After gauging away the latter,
we have

(4.2)
H = E, H = (z z /2), (z + z/2) /2.
Assuming > 0, make a further transformation
0
H = (E /2),
= exp( zz/2),
0
H = (z z )z .
(4.3)
0
Since z commutes with H , an elementary solution is then

= exp(zt)(t z )E/ 1/2 ,
(4.4)
where t is a complex constant. If there are no vortices, must be single-valued and

finite everywhere, i.e., the second factor must be a polynomial, which leads to the familiar
Landau spectrum:
E/ 1/2 = n = 1, 2, . . .
(4.5)
598
so that
n = exp(tz)(t z )n .
(4.6)
For n > 0, vanishes at z = t, and goes to zero at large |z|. The states are labeled
by the level number n, and a continuous complex parameter t which is the analog of
the continuous real parameter in the Landau gauge. (The true energy is E/4m in our
convention.) Here t/ corresponds to the center of the classical Larmor orbit with angular
momentum n 6 0. Note that z z = t z is a raising operator for the level number n,
hence hti = h z i. We will conveniently refer to this factor as raising factor.
The space of all ns and all complex ts is overcomplete on the physical shell, unlike in
the Landau gauge. (In the circular gauge, the radius |t| and n label the states [17].) But by
expanding Eq. (4.6) in powers of t, or by the Fourier integral
I
(4.7)
n,l = n t l1 dt, l > 0
we get a complete set centered at zero and labeled by two integers n and l, giving a total
angular momentum j = l n. Each solution contains a factor zl times a polynomial of
order n in z and z . Eq. (4.7) allows an intuitive interpretation that these states are an
epicycle-like superposition of Larmor orbits, with orbital angular momentum l, along
another circle of a certain radius (see below). When < 0, clearly we may take the
conjugate solutions to the above z z , , j j , and the energy levels are given
in general by E/| | = n + 1/2. It is also instructive to examine the limit 0, where
we should recover the original zero field results. Since the energy spectrum is proportional
to , however, in general it will be driven to zero in this limit, as may be seen from the
elementary solution Eq. (4.6) which does not have a limit E = n/ except for n = 0,
E 0. But by supplying an additional factor t n , we get
lim exp( zz/2 + zt)(1 z /t)E/ 1/2 = exp(zt z E/t),
(4.8)
which reproduces the exponential factor in free space.

4.2. A vortex in a magnetic field
There are two ways to introduce a vortex of strength . One is to change n to n + in
Eq. (4.6) (provided that it is > 0), and to let it be the gauge transformed . It represents a
vortex located at t/ , and the energy is shifted:
n+,t = exp(zt)(t z )n+ ,
E/ 1/2 = n + > 0.
By expanding Eq. (4.9) in powers of t, or taking the moments
I
n,l = n+,t t l1 dt, 0 6 l 6 l0 = n + []
(4.9)
(4.10)
599
we get l0 states with the same shifted energy. This is a z -type representation. The restriction
l 6 l0 arises from the condition (0) = 0. 2
For l > l0 , the proper formula is given by
Z
n,l = n,t t l1 dt,
C
E/ 1/2 = n,

C = , (0+, z +) ,
(4.11)
which is of the z-type, and a straightforward extension of the zero field case. Apart from
an overall Gaussian factor, it is a polynomial in z. The energy does not get shifted. The
reason why a nonpolynomial raising factor in the integrand is not allowed is that, at large
distances, becomes exp(+ zz/2), as may be seen by finding the saddle points. It is not
possible to avoid the exponential blowup by changing the sign of or energy. The problem
is avoided only if the factor is a polynomial. (According to the previous subsection, it can
also be seen that only unshifted levels have zero field limits.) Obviously these same results
will hold for < 0, for which the conjugate representations are used.
Summarizing, a number l0 = n + [] of states at level n have their energy (more
precisely E/| | 1/2) shifted to n + > 0, and the rest is uninfluenced by the vortex.
Consider in particular the ground states n = 0 with || < 1. If > 0, > 0 (or if < 0,
< 0), i.e., the magnetic field and the vortex are parallel, then the energy of one state is
shifted upwards by . If, on the other hand, the field and the vortex are antiparallel, none
of the ground states can change energy; the energy can only go up, but not down below the
zero point minimum.
The following semiclassical argument explains the difference between shifted and
unshifted levels (see also [17]). The state n,t around t occupies a disk of radius R
(n/ )1/2 . The orbital radius of the epicyle state n,l is r l/E 1/2 l/( n)1/2 . So the
latter state will cover the vortex at the origin only if R < r. As for the asymmetry in the
ground state, write the Hamiltonian as
H = p2 L(/r 2 + ) + (/r + r)2 .
(4.12)
For L = 0, the minimum of the last term is if > 0, and 0 if < 0.
5. The multivortex problem

5.1. Multiple integral representation
At zero energy, the general solution of the Schroedinger equation is either analytic or
antianalytic. The one vortex solution reduces to z or z . More than one vortices
can be easily accommodated by taking products of n such z-type or z -type factors, but we
cannot form general mixed products. When E > 0, the solutions will involve both z and z ,
2 It turns into a z-type representation if we change the sign of , and let the path be C = (, (0+)). The energy
remains positive.
600
but by continuity this property will persist. In the following we assume nonzero energy.
Our goal is to find a solution which has proper monodromy properties at each vortex:
Mi = i , where Mi denotes the monodromy operation going around vortex i, giving a
phase i = exp(ii ). Since, however, the radial second order differential equation obtained
after removing G has singularities only at 0 and , it is not clear how to introduce more
vortex singularities.
So let us first try a perturbative solution in energy for two vortices. Denote their location
and strength by (b1 , 1 ), (b2 , 2 ), and write z b1 = z1 , z b2 = z2 for short: =
P (n)
(z, z ), (0) = z11 z22 . Then
E n = z z (n+1) .
(5.1)
(1) can be obtained by integrating (0) over z from b1 (b2 ), and over z from b2 (b1 ). This
(1) vanishes at both sites, but it does not have the proper monodromy at the second (first)
site. We can make (1) behave correctly at both sites by taking (1) = (z1 )+n1 (z2 )2 +n2 z
with integer ni > 1. (0) is then obtained by differentiating (1) . But to generate a series to
the N th order, we have to start from an N th order term of sufficiently high power ni > N ,
and we lose control of the perturbation expansion.
Next we look for a nonperturtive solution in terms of a double integral representation for
two vortices. The gauge potential and gauge function are respectively a sum and a product
of two components:
Az = 1 /z1 + 2 /z2 ,
G(z) = (z1 /z1 )
1 /2
(z2 /z2 )
Az = A z ,
2 /2
(5.2)
Consider an elementary solution of the Schroedinger equation with parameters t1 , t2 , s1 ,

s2 :

exp z1 t1 + z2 t2 E 1/2 (z1 s1 + z 2 s2 ) f (t1 , t2 , s1 , s2 ),
(t1 + t2 )(s1 + s2 ) = E.
(5.3)
The idea is to integrate this over one of t1 , s1 and one of t2 , s2 independently to satisfy
proper monodromy relations, but the constraint precludes mixed types involving t and s,
and it will be sufficient to examine the z-type only. Since the third variable turns out
superfluous, we are led to the integrand
1 1 2 1
t2
,
(5.4)
exp z1 t1 + z2 t2 (z b)E/(t
1 + t2 ) t1
The integration contours (C1 , C2 ) are formally taken, respectively,
with an arbitrary b.
around 0 to U (1 ) and U (2 ). It is in general not possible, however, to avoid
the vanishing of t1 + t2 by a proper choice of the contours, and this would invalidate the
formula. As C1 or C2 makes a 2 rotation, there always occur crossings of C1 and C2 =
C2 (t2 ), where t1 + t2 = 0, and we have to set up a convention for one of the ts to make
a detour around the other. A change of the crossing point leads to a change in the integral,
and after making a 2 rotation, it will yield an extra piece in beyond a phase factor.
A detailed study shows that the extra piece is a regular function of z so that the monodromy
601
matrices are two-dimensional, but they do not take the desired diagonal form. In other
words, it is not possible to satisfy the two monodromy conditions simultaneously. 3
Later an alternative form of single integral representation for dynamical vortices will be
found to solve the problem. But we will show next that the multiple integral representation
can be used without difficulty if a magnetic field is present, although the solutions obtained
are of the type that do not have zero field limits.
5.2. Solutions in a magnetic field
Unlike the zero field case, it is easy to generalize the z-type representation for unshifted
levels, Eq. (4.11), to N vortices with the multiple integral method. The solution is given by
Z
Z
X

Y

0
t i 1 dti ,
zi ti (T z )n
n = exp( zz/2) exp
T=
ti , i0
= i + li > n,
i = 1, . . . , N.
(5.5)
The factor (T z )n can become zero, but this does not cause the earlier problem since
it is a polynomial in the ti s containing powers up to N . The condition on the i s insures
that the overall power of each ti is < 1. For the ground state n = 0 we have
Y 0

zi i / i0 !,
(5.6)
0 = exp( zz/2)0 .
0 =
The nth level has an additional factor which is a polynomial of order Nn in the zi . They can
be obtained by applying the raising operator z z n times to the ground state {i0 +n},0
so that the condition on the s are satisfied. Thus for this class of solutions the Landau
levels are not affected by the vortices.
5.3. Zero field limit
Earlier we found that the multiple integral representation failed in the zero field case, so
it is an interesting problem to study the limit of letting go to zero, = E/(n + 1/2) 0,
keeping energy fixed, or n . We may evaluate the formula Eq. (5.5) for N vortices in
the large n limit by finding the saddle points with respect to each ti :
X

ti NT
(5.7)
ti = i0 + n zi + n/(T E z /n) T , T =
which is a contradiction for N > 1. A simple dimensional analysis shows in fact that in
this case the integral goes like z(N1)n . For N = 1, on the other hand, a consistent limit
exists as was shown in Eq. (4.8). The origin of the problem is traced back to the fact that
each ti factor in Eq. (4.16) had a power 0 n in order to give negative powers of ti for
all the terms in (T z )n , and this contributed an overall power of z(N1)n , meaning that
the angular momentum grows indefinitely with n.
3 Start from an initial position of z for which C and C may be taken to go to in opposite directions, and
1
2
let C2 lie inside of C1 . Then t1 + t2 6= 0. Denoting by Mi the monodromy operation going around vortex i, we
have the result: M1 = 1 , M2 = 2 + I , where I is a regular function, i.e., Mi I = I . Note [M1 , M2 ] =
(1 2 )I .
602
6. Dynamical vortices
6.1. The chargevortex system
The difficulties of the general N -vortex problem can be resolved if we treat the vortices
themselves as dynamical objects with their own kinetic energies, for reasons which are not
clear. First we note that, when the vortices are made dynamical objects, they are under the
influence of a magnetic (dual) vector potentials produced by the charges which, however,
are of the same form as the electric vector potentials produced by the vortices and acting
on the charges. This is obvious because, for the wave function as a function of the
coordinates of a charge and a vortex, a rotation of the charge around a vortex should be
equivalent to a rotation of the vortex around the charge. (A general description of charges
and vortices as dual objects is given in the appendix.) Thus the gauge transformation G of
Eq. (2.5) will remove the potentials from both the charge and the vortex, and we will end
up with a free Hamiltonian
H = Hz + Hb = z z (1/)b b ,
(6.1)
where the vortex is represented by coordinates b, b and mass (the mass of the charge
= 1). This generalizes to any number of vortices as well as charges if we consider
only chargevortex interactions, and ignore chargecharge and vortexvortex interactions.
Since each vortex has its own strength in general, the possible statistics among them will
not be considered here.
The center of mass commutes with the free Hamiltonian, hence it is a constant of motion
even with the boundary condition imposed by the vortex singularities since they depend on
relative coordinates. So the wave function can be separated into that for a center of mass
motion and that for the relative motion, and we may deal with only the relative part. We
will not do this explicitly because it will complicate the formulation.
Suppose there are N vortices with masses i , i = 1, . . . , N . We relabel the 2N + 2
coordinates of the charge and the dynamical vortices z, z , {bi , bi } uniformly as {ui , u i },
i = 0, . . . , N , and similarly the masses as {i }, 0 = 1. The vortices will eventually be
set to their classical positions bi0 = u0i , i 6= 0. The coordinates relative to them are denoted
with primes: u0i = ui u0i , i > 0. The total Hamiltonian H is the sum of N + 1 individual
ones
X
Hi = H.
(6.2)
Hi = (1/i )ui u i ,
Further define

Ii = exp i ui u 0i /t ,
I=
Ii .
The I s are the heat kernels satisfying the relations

Hi Ii = i ui u i /t 2 + 1/t Ii = t Ii , H I = t I,
(6.3)
t 6= 0.
(6.4)
603
Here t may be regarded either as an imaginary time in a time-dependent Schroedinger

equation, or simply as a parameter to be integrated over. 4 The ansatz for is now,
Z
(6.5)
= exp(Et)I (ui , t) dt,
C
with some contour C. Now apply H to it, making use of Eq. (6.4):
Z
i
h
X
1/i u 0i I ui dt
H = exp(Et) (t I )
Z
X
i
h
= exp(Et) t I + I
ui ui dt
Z
= E + exp(Et)I D dt/t,
X
ui ui .
D tt +
(6.6)
So will be a solution provided that D = 0, i.e., is a homogeneous function of degree

zero. Note that D = 0 amounts to a time-dependent Schroedinger equation for the wave
function I if it (6= 0) is interpreted as the time.
We will choose
= (z/t)
N
Y
(u0 ui )/t
0
i
i0 = i + li > 0.
(6.7)
Thus
(0+)
Z

X
0
0 Y
(z bi )/t i dt/t N+1
exp Et
i ui u 0i /t (z/t)0
= JP 0 +00 +N (kR)(kR)N0
!1/2
N
X
0
i ui u i
.
R=
0 00
(z bi )i ,
(6.8)
0
0
(Trivial numerical factors are ignored). Here (z/t)0 , 00 > 0, is an optional factor
representing a nondynamical vortex at the origin (or a kinematic zero if 00 is an integer.)
There are also conjugate representations obtained by switching each i0 > 0 to some
mi i0 > 0, and similarly for 00 . When no dynamical vortices are present, Eq. (6.8) is
equivalent to Eq. (2.13) or (2.15).
A couple of remarks are in order:
4 With the first interpretation, there are time-dependent solutions for nondynamical vortices of the following
0
Q
form, using only the kernel I0 (z): = I0 (z/t), = i=1,...,N (z/t ivi )i . The vortex coordinates bi = vi t
are moving with velocities vi . But since the bi s are fixed, it does not correspond to physically moving vortices.
(However, this form is of the same type as the function given below.) Solutions with dynamical vortices of the
0
Q
simple form exp(iP X + i P X)

(z bi )i , where P and X refer to the center of mass, are not admissible since
they are not bounded at infinity.
604
a) The above construction of the wave function can be generalized to a system made
P
up of many charges as well as vortices. In this case the sum
ui u 0i in Eq. (6.3) will
extend over the charges as well, and will be a product of those for individual charges. An
appropriate statistics among them may also be taken into account.
b) If the charges and vortices are identified, we will be dealing with a system of
anyons [18]. For anyons with a statistical parameter , let all the coordinates ui , u i be
equivalent, having the same mass, and let the factor be made up of all pairs: =
Q
((ui uk )/t)/ or its conjugate form. We will not discuss the problem of anyons in
this paper.
6.2. Pathology of multivortex solutions
The of Eq. (6.8) obviously has the proper monodromy properties. When the vortex masses i become large, the Bessel function oscillates rapidly with their positions bi
except at their intended classical positions bi0 . Around each bi0 , an uncertainty bi generates an uncertainty R b 0b/R and an uncertainty E/E i bi0 bi /R 2 = R 2 /R 2 in
energy locally, as may be seen by applying H to the integrand of Eq. (6.8). So the uncertainties in both vortex positions and energy are kept small. In other aspects, however, the
above solution is not physically satisfactory, and these all seem related to each other. We
will examine them in the following.
6.2.1. Asymptotic R dependence and effective Hamiltonian

First of all, the solutions of the type given by Eq. (6.8) are not the natural ones when
considered as wave functions for each of the individual particles. Because of the factor
1/t N+1 , they go like 1/R N+1/2 1/|ui |N+1/2 when one of the ui s becomes large,
contrary to the expectation that they should go like 1/|ui |1/2 in the physical 2D space. The
additional power of 1/t which is responsible for this is a quantum effect associated with
the dynamical degrees of freedom of a vortex. If a vortex is to be pinned to a fixed position,
its degrees of freedom will have to be frozen. By the same token, if two vortices are to be
merged, those of the relative motion will have to be suppressed. But the vortex velocities
b /, b /, when acting on I , are independent of , a property that was necessary for
forming the factor . So the derivatives, i.e., the dynamics of the vortices cannot be ignored
at the level of the Schroedinger equation even for large masses. Herein lies the need for
treating the vortices as dynamical, and the difficulties of pinning them down. 5
In general, we can recover the proper 2D asymptotic behavior by superposing those
states that are polarized in the direction of the particular pair of coordinates ui ,
u i , i.e., those consisting of high powers of homogeneous monomials ui /t and u i /t.
may be
Asymptotically, such a state must be an eigenstate of Hz since all the b, bs
dropped. Consider then applying the Greens function
5 However, in the opposite limit of the single charge becoming infinitely heavy, the wave function of the vortices
would reduce to a product of the individual ones as in the case of many charges in the presence of a single static
vortex.
P = 1/(Ez Hz ) = 1/(Ez + z z ),
Ez < E
605
(6.9)
to the of Eq. (6.8). In coordinate space it means propagating (folding) in the z, z

1/2
subspace with the Neumann function (Greens function in free space) N0 (2Ez |z|)
1/2
sin(2Ez |z| /4)/|z|1/2 . P also satisfies the Schroedinger equation. Since the source
is concentrated around the origin like 1/|z|N+1/2 , it is clear that goes like 1/|z|1/2 . 6
Unfortunately, however, this operation does not respect the boundary conditions at the
vortex sites. In order for this to be an exact solution, P must be a Greens function
constructed out of solutions in the presence of fixed vortices. But we have found out that
such solutions do not exist.
A way to deal with this problem in the present case, albeit somewhat artificial, is to
introduce an integer vortex at infinity, b , of strength N . This changes the number of
N+1 away
vortices to N + 1. Then put an extra factor (t/b )N+1 to , and renormalize 1/b
as being effectively a constant. We get
Z

0
0 Y
(z bi )/t i dt/t
= exp R 2 /t + Et (z/t)0
C
0
= JP 0 +0 (kR)(z/R)0
(z bi )/R
0
i
(6.10)
Alternatively, we may define the following effective Hamiltonian which yield the desired
solutions:
X

bi bi Nb b
Heff = Hz + z1 z
X

bi bi N b z .
(6.11)
+ z 1 z
Here also a return vortex of negative signature at infinity, b , b , are inserted in order to
make Heff formally Hermitian. Being linear in the derivatives, we may say that the vortex
coordinates retain only partially their dynamical status. Unlike the original H , however,
Heff is not translationally invariant. The solution for Eq. (6.11) is given by replacing R in
Eq. (6.10) with |z|.
The asymptotic properties of the vortices are still abnormal. When one of them, say the
ith, recedes to infinity, the saddle point value of t will
P0 be |bi |. Since t appears in each
factor of , the wave function goes like 1/|bi |1/2+ , where the sum excludes the ith.
Thus the mere presence of the charge causes a correlation among the vortices.
6.2.2. Completeness
The Bessel function in Eq. (6.8) corresponds to a solution of the wave equation in
a (2N + 2)-dimensional space in the so-called hyperspherical harmonics expansion. In
the absence of the factor , the general solutions in this space are spanned by Bessel
6 In momentum (Fourier) space, P is a simple multiplication by 1/(p p E), and = (p 2 E) if the
z z
factor is ignored. For large |z| = |u0 | |ui |, i > 0, we can see its effect by setting the latter ui s to zero, and
inverting the Fourier form of P back to real space. The vortex coordinates contribute only their phase space.
R
The integrations boil down to P = N /(N 1)! 0E J0 (2p|z|)(Ez p 2 )1 (E p 2 )N p dp, which contains a
piece that behaves J0 (k|z|). (To include the case Ez = E, use (Ez Hz )N .)
606
functions times the (2N + 2)-dimensional spherical harmonics. The latter can be generated
by applying ui and u i , which commute with one another and with H , repeatedly to the
s-wave solution JN /R N . Obviously each operation, respectively, generates a negative or
positive angular momentum around the origin in the 2-dimensional subspaces spanned
by ui , u i . In the integral representation of Eq. (6.8) without , application of ui (u i )
will bring down a factor u i /t (ui /t). Application of a monomial of these operators will
generate a polynomial in the us and us,
and this way a complete set of states labeled by
the N + 1 magnetic quantum numbers of O(2N + 2) and a Casimir invariant (the degree
of the monomial) will be generated. Fractional factors in are of a similar nature, but the
power of angular momentum around each vortex is correlated to be all positive for , or all
negative for its conjugate representation. We cannot apply ui to the above an arbitrary
number of times because the factor will become singular. Hence there arises the question
whether or not the s of Eq. (6.8) and their conjugate counterparts form a complete set of
states by which it is possible to prepare any physically reasonable conditions. Note that
is not an eigenstate of the individual Hi s, although they commute with H . Therefore the
Hi s cannot be diagonalized in the space of these states. Otherwise we would have been
able to find solutions for static vortices.
It is true that, in the neighborhood of a given vortex site, there exists a complete set
of angular momenta, but we have also to look at the asymptotic states. This would be
necessary, for example, in order to construct the scattering amplitude. The i0 and take the
lowest values 0 < [i ] < 1, and its conjugate has the lowest values 0 < [ i ] = 1 [i ] < 1.
P
For large |z| > |bi |, the lowest angular momentum L is the sum [i ]. The conjugate
P
P
states then have [ i ] = 1 [i ], and L = [ i ] = [i ] N . As can be seen from
Eq. (6.8), the for different li with the same L have different polynomials of degree L
in z/t. In other words, when is regarded as a function of z for fixed flux coordinates,
the states are degenerate in L. This classification is valid not only for asymptotic states but
in general. But there is a gap of N between the s and their conjugates. In other words,
for N > 1, there is an interval of N 1 missing angular momenta. (Therefore the problem
does not arise for asymptotic states with respect to the vortex coordinates. Neither does a
vortex with an integer strength cause missing states, since then [] = 0, and its conjugate
also can take the lowest value 0.) For example, if N = 2, 1 < 1, 2 < 1, 1 + 2 < 1,
its conjugate state has 100 = 1 1 , 200 = 1 2 , 100 + 200 = 2 1 1 . In the original
gauge, the former has angular momentum 0, and the latter 2, so the state 1 is missing.
Because of this, the scattering problem cannot be handled. If we simulate the scattering
amplitude for a single vortex case by combining available positive (z-type) and negative
(z-type) angular momenta, we would end up with the expected amplitude for a single vortex
P
of strength of
, minus (or plus the shadows of) the missing partial waves, and these
contain both incoming and outgoing waves. We are thus led to assert that the dynamical
multivortex solutions obtained here retain some pathological properties from a physical
point of view. In order to cure the problem, we could allow powers down to [] 1, which
would make it singular at the vortices, but the wave function remains square-integrable
607
there. 7 Alternatively we could give the vortices both negative powers and a finite size, so
the singular part is matched with their conjugate forms at the boundaries, e.g., (z b)[]1
1[] . But the matching would require the coefficients to depend on the
against (z b)
vortex coordinates, so that the result would not be a solution of H . This would then force
us back to nondynamical vortices, which we have found not possible, and we would have
to be content with some kind of approximate solutions. Such an approximate solution is
suggested by Eq. (6.11):
(0+)
Z
exp(zz/t + Et)(z/t)0
(z bi )/t
0
i
dt/t,
(6.12)
P
which is a solution of Hz only asymptotically. Expanding the factor, = z i (1

P
b /z + ), we see that becomes exact to order 1/z if the origin is chosen so that
P i i
i bi = 0.
It seems uncertain, however, in what sense Eq. (6.12) is a valid approximation in the
non-asymptotic region. From a physical point of view, the multiple integration formulas
of Section 5 might be more natural. They yield saddle point solutions that look physically
more reasonable. On the other hand, they violate the gauge conditions and the singlevaluedness of the wave function in the present zero field case.
6.3. Dynamical solutions in a magnetic field
In the presence of a magnetic field, the problem of missing angular momenta still
persists, but we will not pursue it any further. According to Section 4, the Hamiltonian
for a charge in a magnetic field is (after removing a factor exp(zz) from ) is
Hz = (z z )z ,
(6.13)
and a class of static solutions for unshifted energy levels are given by Eq. (5.5). To
accommodate dynamical vortices, we simply multiply the integrand of Eq. (5.5) by new
factors:
Z

X
Y
0

exp Ei i bi0 /ti t i 1 dti ,
= exp( zz/2) exp
zi ti (T z )n
X
(6.14)
T=
ti , i0 = i + li > n, i = 1, . . . , N
where Ei is the energy of the ith vortex, and is arbitrary. Thus the energies of the charge
and the vortices are individually constant in spite of the correlations among them imposed
by the boundary conditions. This is in contrast to the zero field case where only the total
energy can be fixed. 8 It is then natural that these solutions in a magnetic field do not
have zero-field limits, although it does not explain the reasons for it. The original static
vortex formula, Eq. (5.5) is recovered if the Ei s are set to zero, meaning that it can also be
7 I thank Eguchi for raising the question of allowing negative powers.
R
8 Except for a single chargevortex pair, for which = exp(zt E z /t) exp(bt + E b 0 /t)t 0 1 dt is
z
b
obviously a solution.
608
regarded as a solution for zero energy dynamical vortices. We have not found dynamical
solutions that have a zero-field limit.
6.4. Shifted Landau levels
As was given in Section 4.2, there are also solutions with shifted Landau levels for a
single vortex. Its straight multivortex generalizations do not seem possible, but solutions
do exist if each vortex is placed in a harmonic potential so that its level spacing matches
that of the charge in the magnetic field. The details are found in Appendix B.
7. Large N limit and vortex medium

It was found in Section 6.2 that, unless singular behavior at vortex sites are allowed, the
P
order of the Bessel function representing goes up with the number of vortices as i0 ,
with all s being of the same sign > 0 and adding up. This leads to dramatic effects when
the number becomes very large. Suppose the vortices of equal strength are distributed
with a density 1/a 2 up to a large radius R0 , so that the total number of vortices is N =
(R0 /a)2 . The angular momentum L of the wave function at radius R < R0 (R > R0 )
is (R/a)2 ((R0 /a)2 ). Its azimuthal momentum is L/R = R/a 2 ((R0 /a)2 /R),
so the radial momentum for large R: R 2 E 1, is pR iR/a 2 , and the radial
R
phase is ipR dr (R/a)2 /2 ((R0 /a)2 ln R). For R < R0 , the wave function
will behave like Gaussian or anti-Gaussian. Below we will investigate it in more detail.
Assume that the positions of vortices of strength are effectively fixed to form a regular
square lattice of unit size at sites bi , i = m + in, although the lattice structure is not
essential. The gauge potential Az , Az and the gauge function G are
X
1/(z bi ),
Az = (/2)
X
1/(z bi ),
Az = (/2)
Y
/2
.
(7.1)
G=
(z bi )/(z bi )
As the sum goes to infinity, they fail to converge, hence an extra pure gauge subtraction
terms become necessary (except for i = (0, 0)). The modified forms are then
Az = (/2) (z), Az = (/2)(z),
X

1/(z bi ) + 1/z + z/bi2 ,
(z) =

/2
,
G = (z)/ (z)
Y

(1 z/bi ) exp z/bi + z2 /2bi2 .
(z) = z
i6=0
(7.2)
R
Here and = exp( dz) and are the elliptic functions of Weierstrass with full periods
1, i [19]. Because of the subtractions, they are not periodic but satisfy
(z + m + in) = (z) + (m in),

(z + m + in) = (z) exp m2 + n2 /2 + z(m in) (1)m+n+mn .
609
(7.3)
The leading asymptotic factors are insensitive to the lattice structure, and may be derived
by a continuum approximation. If the vortex density is 1/a 2 ,
Z
X

1/(z bi ) 1/ z r exp(i ) r dr d/a 2
= |z|2 /za 2 = z /a 2 (z).
(7.4)
The contributions to the integral over b from r > |z| vanish because the phases of the
contributions from points at radius R > |z| from the origin cancel. Integrating this from 0
to z along the radius, we get

Y
0

(7.5)
z (z bi )/bi exp (|z|/a)2/2 (z).
These non-analytic expressions are a result of confusing z and the cell coordinates (m, n)a.
We can also use this method to define a cutoff version of and :

z /a,
if |z| < R0 ,
R0 (z/a)
R02 /az = Na/z, if |z| > R0 ;

exp (|z|/a)2/2 , if |z| < R0 ,
(7.6)
R0 (z/a)
exp(N/2)(z/R0 )N , if |z| > R0 .
We can now substitute them in Eq. (6.12), taking the origin of coordinates at the center
of the lattice. 9 Then = R0 /t N . First, the value of at the origin can be exactly
calculated, and yields (0) = 1/N! For the asymptotic behavior, we can use the the saddle
points to get

(7.7)
C(t0 ) exp E|z|2 /t0 + t0 t0N R0 (z/a) ,
where C(t0 ) is the contribution from the Gaussian integral around t0 , and
t0 = iE 1/2|z|,
if N E|z|2 ,

N
1/2
/ iE 1/2|z|
;
exp 2iE 1/2|z| + N/2 z/ iaE 1/2|z|
t0 = N,
if ER02 < E|z|2 < N,
N 1/2 exp(3N/2)(z/aN)N
= exp(N/2)(z/a)N /N! = exp(N/2)(z/a)N (0);
t0 = N,
if 1 < E|z|2 < ER02 ,

|z|2
(0).
exp |z|2 /2 z/|z|
(7.8)
The first form is the true asymptotic limit. The second is valid in the near outside region
|z| > R0 . The third applies to the inside of the vortex medium. These are all consistent with
9 A possible alternative choice for is = (z/a)/ (lt) with some l and the contour running outside of
R0
R0
the vortices: lt > R0 . But this leads to the same asymptotic results as those given below, up to a normalization.
610
the physical argument given at the beginning of the section. 10 From above we see that the
wave function inside the vortex area grows with the distance in an anti-Gaussian way like
exp(|z|2 ) inside the medium, then with a power like |z|N to distances E 1/2 |z| > N
where the true asymptotic behavior sets in. In other words, the wave function is pushed out
of the vortex region and beyond (unless the energy is so large as to skip the intermediate
regimes).
8. Summary and discussion

In terms of physical concepts, the AharonovBohm problem for many vortices is a
straightforward extension of the single vortex case, and is not expected to contain anything
conceptually new. Yet in an attempt to construct explicit solutions, we have found some
pathological features that may seem counter-intuitive and physically unreasonable. In the
idealized AB problem of an infinitely thin vortex, the wave function of a charge in its
vicinity is described by Bessel functions whose order is shifted by the vortex strength from
n to | + n|. Since the latter has to vanish at the vortex, the component which would be an
s-wave in free space is pushed out irrespective of the sign of . This effect will magnify in
a medium made up of vortices.
To handle the general case of many vortices, the gauge potential may be removed by a
singular gauge transformation. The Hamiltonian is reduced to that of a free particle except
for boundary conditions. Around each vortex i the transformed wave function must develop
a branch point in one of the complex coordinates, with a positive fractional power so that
it has the form zi +ni , or z i +ni , times a holonomic function in z and z . In particular,
at zero energy the general solution is an arbitrary sum of a function of z and a function
of z , and each part must carry all the branch points. In other words, the fractional angular
momentum around each vortex, as opposed to the intrinsic vortex strengths i , must all
be either positive for the analytic, and negative for the anti-analytic solution. At nonzero
energies, the wave function of either type contains both variables, but by continuity the
above qualitative features do not change. (This turns out to be true even in a magnetic field
where the spectrum is discrete.) The total angular momentum of a wave function around a
large circle containing a cluster of vortices is then a sum of positive or negative numbers
which will grow with the number of vortices, resulting in the expulsion of a charge from
the interior of the cluster. This is because the wave function has to vanish at each vortex
site, whether the vortices add up to an integer or not.
This conclusion does not require the knowledge of explicit solutions. On the other
hand, the construction of multivortex solutions has turned out to be more complicated.
Although a class of solutions in a magnetic field can be obtained rather easily, in free
space we have to treat the vortices as dynamical objects with their own Hamiltonians.
Besides, all these solutions share the unphysical property that the multivortex solutions are
not asymptotically complete in angular momentum. For N vortices, there are N 1 low
10 In the exterior region, satisfies D = z = t = N , so is a semiclassical solution of H
z
z
t
z
(footnote 5), whereas Dz is not zero in the interior region. (Of course the full D = 0 is always guaranteed.)
611
angular momenta that are missing. Consequently, the wave function of many vortices in
free space do not behave at large distances like that of a single vortex, and a satisfactory
scattering amplitude, i.e., one consisting of a plane wave-like part and an outgoing radial
part, does not exist. The expulsion of a charge from a vortex medium is of the same origin.
To recover those missing states we must allow negative powers down to [||] 1 at the
vortices. An alternative possibility might be to give the vortices a finite size. But then the
wave function fails to satisfy the Hamiltonian for dynamical vortices, and we are forced
back to dealing with static vortices, for which solutions cannot be constructed.
These general results do not seem to be due to our reliance on the singular gauge and
analyticity. To see this, first consider a vortex 1 at the origin in the original gauge so
a solution may be expanded in Bessel functions J|1 +n| of positive fractional order. If
there is a second vortex 2 at point P , they have to be re-expanded in Bessel functions
J|2 +m| around P . But the complete set of eigenfunctions for the latter covers only the
space R 2 P . In fact, the general addition formulas for Bessel functions show that the
re-expansion involves Bessel functions of integer order around the second site. Therefore
we have to start from the beginning with functions that have the proper behavior at both
sites. This situation may be contrasted with that of hard core potentials. There the space is
also punctured, but the matching involves Bessel functions of integer order, and there is no
problem in constructing solutions.
The multivortex problem becomes easier to handle in a magnetic field. A class of static
and dynamical solutions exist. Vortices of finite size can also be accommodated. But these
solutions do not have zero field limits. This raises the possibility that in real situations there
must always be some magnetic field in the medium.
It is noteworthy that some of the Landau states get the energy shifted by the vortices
while others remain unaffected. The effect is asymmetric for the lowest level with respect
to the relative sign of the field and the vortex. These properties may have some interesting
physical consequences. But multivortex solutions with shifted levels do not seem to
exist unless the vortices are put under appropriate potentials. The general mathematical
difficulties and pathologies of the multivortex problem as well as the role of the magnetic
field remain to be understood in physical terms.
Finally there are interesting questions regarding the relevance of the present results to
the problem of anyons. We wish to pursue those later.
Acknowledgement
This work was supported by NSF Grant PHY-99-00194 and the University of Chicago.
I thank Prof. T. Eguchi for continuing interest and critical discussions (which started in
1979), and H. Awata, Y. Hosotani, R. Jackiw, H. Kawai, Y. Ohnuki, M. Peshkin, P. Ramond,
P. Wiegmann and F. Wilczek for valuable comments and information about the literature
at various stages of the work. Part of the work was done at Osaka University and the
International Institute for Advanced Studies. I thank Prof. K. Kikkawa and Prof. T. Sawada
for their hospitality.
612
Appendices
Appendix A. Field-theoretic description of charges and vortices
The vortices can be regarded as dynamic magnetic particles. They interact with the
charges through the AharonovBohm effect, but not directly among themselves. A vortex
is a concentration = Ba 2 of the magnetic field density B over an infinitesimal area a 2 ,
2 /2 of the magnetic field
its energy (per unit z direction) is a part of the energy density Fxy
plus an amount necessary to keep it concentrated [20]. We will assume uniformity and
ignore the dynamics in the z direction. When a vortex moves, the magnetic field inside will
acquire electric field components of order v/c, so (Bz , Ey , Ex ) will be associated with a
moving vortex. The vector potential associated with E will be present in the medium, even
though it may be a shielded potential.
Let A be the usual vector potential, G and F two vector fields, J and K the electric and
magnetic currents in 2 + 1 dimensions, so they have mass dimension 2. J is then not the
usual charge density since the charge is not extended in the z direction, but may be thought
of as a kink soliton z of an extended field . This is the only place where the z direction
enters. Let the Lagrangian L be
X
L=
Gi Fi /2 iG curl A + eA J + gG K + Lmat ,
G1 = D2 ,
G2 = D1 ,
G3 = H 3 ,
F = G,
= diagonal matrix [, , ].
(A.1)
Lmat refers to the kinetic part of matter (charge and vortex) Lagrangian. Euclidean metric
is used for simplicity, with x3 = ix0 , etc. The electric and magnetic charges e and g are
dimensionless; G and F have mass dimensions 1 and 2, respectively. B, H , E, D have
the usual meaning in the Maxwell theory except for dimensionality: the parameters ,
are respectively the analog of magnetic permeability and inverse dielectric constant with
the dimensions of mass. Eq. (A.1) shows that the magnetic vector potential is invariant
under the usual gauge transformation of the electric vector potential A. Varying G and A
as independent fields, we get the field equations
F + i curl A = gK,
i curl G = eJ.
(A.2)
If the magnetic current K is absent, Eq. (A.2) takes the usual Maxwell form. If, on the
other hand, and are set to zero, the equations show a symmetry under duality between
J, A and K, G:
i curl A = gK,
i curl G = eJ,
(A.3)
and an invariance under both an electric and a magnetic gauge transformation:

A A + grad ,
G G + grad .
(A.4)
(Boundary contributions are assumed to vanish under these gauge transformations.) The
left-hand side of Eq. (A.3) are the CoulombLorentz forces acting, respectively, on charge
and vortex, which are seen to come directly from the currents of their opposite numbers.
613
For static electric and magnetic sources located, respectively, at xj , yj and xk , yk ,

Eq. (A.3) yields

A1 , A2 = g(v2 , v1 )/ |v|2 , v = (x xk , y yk ),

G1 , G2 = e(u2 , u1 )/ |u|2 , u = (x xj , y yj ),
A+ = iA1 + A2 = g/z,
A = iA1 + A2 = g/z,
z = v1 + iv2 ,
G+ = iG1 G2 = e/w,
G = iG1 G2 = e/w,
w = u1 + iu2 .
(A.5)
Hence the vector fields eA and gG acting on charge and on vortex are the same, and
proportional to eg, and the same form of gauge transformation removes the potential from
the charge and the vortex Hamiltonian.
If 6= 0, there is no duality symmetry. Conservation of the magnetic charge requires
div G = 0,
(A.6)
which is the same as in the Maxwell case. The magnetic gauge invariance still holds if
the in Eq. (A.4) satisfies
div grad = 0.
(A.7)
By taking the Coulomb gauge 1 A1 + 2 A2 = 0, Eq. (A.3) leads to

2 A3 = ig1 ( K)12 eJ3 ,

1 2 A1 + 1 3 F31 = ig 1 2 K3 + 1 3 K2 ieJ1 ,

1 2 A2 1 3 F32 = ig 1 1 K3 1 3 K1 ieJ2 .
(A.8)
If static, 3 = 0, or if , the contribution to A from K, and one to G from J become

2 A3 = eJ3 ig1 ( K)12 ,
2 A1 = ig2 K3 ,
1
G3 = i
2 A2 = ig1 K3 ,
g( A)12,
G2 = i1 1 A3 .
(A.9)
p
2
2
Since the Greens kernel is ln(r)/2, r = (x + y ), the potentials A1,2 and G1,2 are
then found to be the same as Eq. (A.3):
G1 = i
2 A3 ,
eA3 = ieg ln(r)/,

eA1,2 = gG1,2 = eg(2 , 1 ) ln(r)/.
(A.10)
Eq. (A.8) shows that the magnetic Coulomb potential G3 acting on magnetic charge
gK0 = igK3 is, apart from self interaction, indeed the usual magnetic field G12 = H3 =
B3 /, where is interpreted as an effective magnetic permeability. The term G12 F12 /2 =
H32/2 is the magnetic field energy. When an external magnetic field Hex is imposed, we
should change it to (H + Hex )2 /2, which means that the vortex will feel a magnetic
Coulomb potential Gex . By the same token, if an external electric field is present without
614
suffering shielding, the vortex will feel an electric Lorentz force. On the other hand,
the potential A3 acting on a charge at (x, y) and generated by another charge at 0 is
(ln(r)/2)eJ3 . Since this should be the 3D Coulomb potential 1/(4r), we must
interpret to be z , i.e., the derivative in the hidden z direction at z = 0. 11
Appendix B. Some more formulas for external field problems

a) Harmonic potential
Let the Hamiltonian for a vortex in a harmonic potential be
H h = (1/)b b + 2 b b,
(B.1)
and take its solutions with no radial nodes, given by

= exp 1/2 b b zm (or z m ),
E = (m + 1)/1/2 ,
m = integer > 0.
(B.2)
For the combined system of a charge in a magnetic field > 0 and two vortices in their
respective harmonic potentials, choose their strengths i , i = 1, 2, to satisfy 2i = i .
Then a set of solutions with shifted Landau levels can be constructed as follows.
I

X
1/2
(B.3)
exp zt
i i bi bi (t z i )1 (t z i )2 t l2 dt, l 6 1 , 2 .
The radius of integration around the origin should be < |zi |. For the proof we expand the
factors in two ways:
(t z i )1 (t z i )2
X
=
cm1 m2 t m1 +m2 z 11 m1 z 22 m2
X
m
m
0
=
cm
(t z )1 +2 m1 m2 b1 1 b2 2 .
1 m2
(B.4)
The first expansion shows that each term has the correct monodromy, and the condition
on l insures that only a finite number of terms with positive powers i mi > 0 survive.
Each term in the second expansion (valid for large enough |z|) is an eigenstate of H with
energy (10 + 20 m1 m2 + 1/2) and of the H h s with energy (m1 + m2 + 2), so that
the total energy is always (10 + 20 + 5/2).
Eq. (B.3) can be easily generalized to any number of vortices, and the energy is shifted
by the sum of all vortex strengths. But the solutions are rather artificial and unrealistic in
that (a) the potentials for vortices of different masses have to be adjusted to give the same
level spacing, and (b) the centers of the potentials must be the same, so they cannot be used
to localize the vortices at arbitrary positions.
11 As was mentioned above, the charge is like a kink soliton, or a monopole with a string, that is sitting in the
2D space. If charges were extended in the z direction
like the vortices, the logarithmic potential would arise by
R
integrating the 3D Coulomb potential over z: dz/(r 2 + z2 ) = ln(z + (r 2 + z2 )1/2 /r) (after renormalizing away
infinities). We get back the 3D potential if we take this as the 2D potential and differentiate it by z at z = 0.
615
b) Time-dependent solutions
We can generalize the kernel I of Eq. (6.3) when an external field is present. Make the
ansatz
Z

I = exp zzf (s) g(s),
= I ds, = (z, s) or (z, s)
(B.5)
and require (H + s ) = 0, with s standing for the Euclidean time. In the case of a
magnetic field, H = (z + z )z , > 0, we obtain the equations
f (f + ) + s f = 0,

s + (f + )zz = 0.
s f + s g/g = 0,
(B.6)
The solution for a single static vortex is given by

f = / exp( s) 1 ,
g = 1/ 1 exp( s) ,
= (zf ) .
(B.7)
In the limit 0, f reduces to the original zero field result f = 1/s, g = 1/s. An
eigenfunction with energy E should result if the integrand of Eq. (B.5) is multiplied by
a factor exp(Es). By a change of variable f (s) = t/z, we get
0
exp( s) = 1 z /t,
gds = dt/t,
= t ,
Z
= exp(zzf ) exp(Es)g ds
Z
0
= exp(zt)(1 z /t)E/ t 1 dt,
(B.8)
which is of the form given in Section 4, and shows how the integration path and the
eigenvalues are to be determined.
In the case of a harmonic potential, Eqs. (B.7) and (B.8) are replaced by
f = coth(s),
and
g = / sinh(s),
= (zg)
(B.9)
Z
=
exp(zzf )g exp(Es) ds
I
exp(zt) (t z)/(t + z)
E/
t 2 2 z 2
( 0 1)/2
dt.
(B.10)
The integration is around the pole at z, and the eigenvalues are given by E/ = 1 +
0 + 2n = 1 + + l + 2n, where l is the angular momentum, and n corresponds to the
radial quantum number. Together with their conjugates, they span a complete set of states,
including those of Eq. (B.2).
References
[1] Y. Aharonov, D. Bohm, Phys. Rev. 15 (1959) 485.
[2] A. Tonomura, T. Matsuda, T. Kawasaki, J. Endo, S. Yan, H. Yamada, Phys. Rev. Lett. 56 (1986)
792, and earlier works cited therein.
[3] M. Peshkin, I. Talmi, R.D. Kaye, Ann. Phys. 12 (1961) 426.
616
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
L.S. Schulman, J. Math. Phys. 12 (1971) 304.

A. Inomata, V.A. Singh, J. Math. Phys. 19 (1978) 2318.
A. Inomata, C.C. Bernido, Phys. Lett. A 77 (1980) 394.
P. Stovicek, Phys. Lett. A 142 (1989) 5.
A. Sommerfeld, Math. Ann. 317 (1886); Optics, Academic Press, 1954, p. 249.
W. Pauli, Phys. Rev. 54 (1938) 924.
Y. Aharonov, A. Casher, Phys. Rev. A 19 (1979) 2461.
B.A. Dubrovin, S.P. Novikov, Sov. Phys. JETP 52 (1980) 511.
S. Olariu, I.I. Propescu, Rev. Mod. Phys. 7 (1985) 339.
S. Ruijsenaars, Ann. Phys. (NY) 146 (1983) 1.
R. Jackiw, Ann. Phys. 201 (1990) 83.
S. Sakoda, M. Omote, J. Math. Phys. 38 (1997) 716.
M. Alvarez, Phys. Rev. A 54 (1996) 112.
R.R. Lewis, Phys. Rev. A 28 (1983) 1228.
F. Wilczek, Fractional Statistics and Anyon Superconductivity, World Scientific, 1990.
Whittaker, Watson, A Course on Modern Analysis, Cambridge Univ. Press, 1962.
M. Peshkin, A. Tonomura, The AharonovBohm Effect, Lecture Notes in Physics 340,
Springer-Verlag, 1989.

Large and small density approximations

to the thermodynamic Bethe ansatz
A. Fring 1 , C. Korff 2
Institut fr Theoretische Physik, Freie Universitt Berlin, Arnimallee 14, D-14195 Berlin, Germany
Abstract
We provide analytical solutions to the thermodynamic Bethe ansatz equations in the large and
small density approximations. We extend results previously obtained for leading order behaviour of
the scaling function of affine Toda field theories related to simply laced Lie algebras to the nonsimply laced case. The comparison with semi-classical methods shows perfect agreement for the
simply laced case. We derive the Y-systems for affine Toda field theories with real coupling constant
and employ them to improve the large density approximations. We test the quality of our analysis
(1)
(3)
explicitly for the sinh-Gordon model and the (G2 , D4 )-affine Toda field theory. 2000 Elsevier
PACS: 11.10.Kk; 11.55.Ds; 05.70.Jk; 05.30.-d; 64.60.Fr
1. Introduction
In the context of (1 + 1)-dimensional integrable quantum field theories numerous
methods have been developed to compute various quantities in an exact manner, that is
non-perturbative in the coupling constant. Sometimes it is even possible to perform the
related computations analytically. Often the evaluation within one particular approach
lacks information, typically a constant, which might be supplied by an entirely different
method. Ideally one would like to achieve a situation in which each approach is selfconsistent.
An example for the situation just outlined is given for instance in the form-factor
program [1,2], which allows in principle to compute correlation functions. The lowest
non-vanishing form-factor is not fixed within this approach and is typically obtained from
elsewhere. For instance, the vacuum expectation value of the energymomentum tensor
1 fring@physik.fu-berlin.de
2 korff@physik.fu-berlin.de
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 2 5 0 - 9
618
A. Fring, C. Korff / Nuclear Physics B 579 [FS] (2000) 617631
can be extracted from the thermodynamic Bethe ansatz (TBA) [3,4]. Alternatively, one
can compute correlation functions by perturbing around the conformal field theory [5].
Also in this approach one appeals to the thermodynamic Bethe ansatz for the vacuum
expectation value of the energymomentum tensor and to the Bethe ansatz [6,7] for
the relation between the coupling constant and the masses. The latter correspondence is
also needed in an approach initiated recently in [810], where it was observed that the
ultraviolet asymptotic behaviour for many theories may be well approximated by zeromode dynamics. The considerations in [8] exploit the knowledge of an exact reflection
amplitude, which on one hand results from certain manipulation on the three-point function
of the underlying ultraviolet conformal field theory and on the other hand has a semiclassical counterpart in the related quantum mechanical problem. The question why this
approach allows to compute scaling functions with a high accuracy is still to be settled [11].
The computation of scaling functions by means of the TBA does conceptually not
require any additional input from other methods. However, up-to-now it is only possible
to tackle the problem numerically due to the nonlinear nature of the central equation
involved. Several attempts have been made to formulate analytical approximations. This
is desirable for various reasons, one being that the numerical effort becomes quite
considerable for some models with increasing particle species content. A further reason
is of course that analytical expressions allow to study further the deeper structures
of the theories. For instance, for affine Toda field theories (ATFT) [1214] related to
simply laced Lie algebras some analytical expressions have been provided [1517]. In the
approximation method of [1517] a constant was left undetermined, which can be fixed
in the same spirit as outlined for the other methods in the preceding paragraph, namely
by appealing to another approach. Contrary to the claim in [9], we demonstrate in the
present manuscript that it is possible to fix the constant in this way without approximating
higher order terms. Extending the analysis of [16,17] also to the non-simply laced case
in the present manuscript, we will demonstrate in addition that the constant may also be
well approximated from within the TBA-analysis. Furthermore, we give simple analytical
expressions for improved approximations in the large and small density regime.
Our manuscript is organized as follows: In Section 2 we provide large and small density
approximations for the solutions of the TBA-equations. In Section 3 we assemble the
necessary data for ATFT needed to extend our previous analysis to the non-simply laced
case. We derive universal TBA-equations and Y-systems for all ATFT and show how they
may be utilized to improve on the analytical approximations. We derive the related scaling
function. We test the quality of the various approximations for the explicit example of the
(3)
sinh-Gordon model and the (G(1)
2 , D4 )-ATFT. We state our conclusions in Section 4.
2. Large and small density approximations
2.1. The TBA
The object of investigation of the TBA is a multiparticle system containing n different
particle species with masses mi (1 6 i 6 n) whose dynamical interaction is described
619
by a factorizable diagonal scattering matrix Sij ( ), which is a function of the rapidity

difference . We assume the statistical interaction to be of fermionic type. 3 Adopting
the notation of [17] the thermodynamic Bethe ansatz equations [3,4], which characterize
the thermodynamic equilibrium of such a system, are the n coupled nonlinear integral
equations
n
X

ij Lj ( ).
rmi cosh + ln eLi () 1 =
(1)
j =1
Here the scaling parameter r is given by the inverse temperature times a mass scale. The
convolution of two functions is abbreviated as usual by
Z
1
d 0 f ( 0 )g( 0 ).
(f g)( ) :=
2
The TBA kernel reads
d
(2)
ij ( ) := i ln Sij ( ).
d
The functions Li , which are to be determined as solutions of the TBA-equations (1),
are related to the particle densities ri and the densities of available states hi as Li =
ln(1 + ri /hi ), such that for physical reasons Li > 0. Keeping this definition in mind, we
speak of the large density regime when Li > ln 2 and of the small density regime when
Li < ln 2. It is sometimes useful to express matters in terms the pseudo-energies i ( ) :=
ln[exp(Li ( )) 1].
Having solved the TBA-equations (1) for the L-functions one is in principle in the
position to evaluate the scaling function
Z
n
6r X
mi
d Li ( ) cosh
c(r) = 2
i=1
(3)
which can be interpreted as off-critical effective central charge belonging to the conformal
field theory obtained in the ultraviolet limit, i.e., r 0. It is our goal in this manuscript to
approximate this function in a simple analytical way to high accuracy.
2.2. Approximative analytical solutions
In general it is possible to solve the TBA-equations numerically, where the convergence
of the iterative procedure is guaranteed by means of the Banach fixed point theorem [17].
However, the numerical problem becomes quite complex when one increases the number
of particle species. For this reason, and more important because one would like to gain
a deeper structural insight into the solutions of (1), it is desirable to obtain analytical
solutions to the TBA-equations. Due to the nonlinear nature of (1) only few analytical
solutions are known. Nonetheless, one may obtain approximated analytical solutions when
3 In order to keep the discussion as simple as possible we do not treat general statistics here as, for instance,
Haldane type [18]. Generalizations of our arguments in this sense are straightforward.
620
r tends to zero. For large Li , i.e., for large particle densities, it was shown in [1517]
that the integral equation (1) may be turned into a set of differential equations of infinite
order. Under certain natural assumptions, which are however not satisfied universally for
all models, one may approximate these equations by second order differential equations,
whose solutions are given by
2

cos (i )
arccos(i 2i )
0
.
(4)
for | | 6
Li ( ) = ln
i
2i2 i
The restriction on the range of the rapidity stems from the physical requirement Li > 0.
P (2)
The n constants i = j ij are determined by a power series expansion of the TBAkernel
Z
X
(n)
d ij ( )eit = 2
(i)n ij t n .
(5)
ij (t) :=
e
n=0
The dependence on the scaling parameter r enters through the quantity
.
(6)
i =
2(i ln(r/2))
Here the i , i are constants of integration. There is a very crude lower bound we can put
immediately on i . From the fact that L0i (0) > ln 2, we deduce i > 1/( i ) + ln(r/2).
For particular models we will provide below a rigorous argument which establishes that in
fact they do not depend on the particle type, such that we may replace i and i
. We will also show that they can be fixed by appealing to the semi-classical approach
in [8,10]. In addition, we provide an argument which determines them approximately
from within the TBA-analysis by matching the large and small density regimes. Since
the constant turns out to be model dependent, we will report on it in detail below when we
discuss concrete theories.
The restriction on the range for the rapidities in (4), for which the large density approximation L0i ( ) ceases to be valid, makes it desirable to develop also an approximation for
small densities. For extremely small densities we naturally expect that the solution will
tend to the one for a free theory. Solving (1) for vanishing kernel yields the well-known
solution

f
(7)
Li ( ) = ln 1 + ermi cosh .
Ideally we would like to have expressions for both regions which match at some distinct
f
rapidity value, say im , to be specified below. Since L0i ( ) and Li ( ) become relatively
poor approximations in the transition region between large and small densities, we seek for
improved analytical expressions. This is easily achieved by expanding (1) around the zero
order small density approximations. In this case we obtain the integral representation
!
n
X

f
ij Lj ( ) .
(8)
Lsi ( ) = exp rmi cosh +
j =1
For vanishing ij we may check for consistency and observe that the functions Ls ( )
become the first term of the expansion in (7). One could try to proceed similarly for the
621
large density regime and develop around L0i instead of Li . However, there is an immediate
problem resulting from the restriction on the range of rapidities for the validity of L0i , which
makes it problematic to compute the convolution. We shall therefore proceed in a different
manner for the large density regime and employ Y-systems for this purpose.
In many cases the TBA-equations may be expressed equivalently as a set of functional
relations referred to as Y-systems in the literature [19]. Introducing the quantities Yi =
exp(i ), the determining equations can always be cast into the general form

(9)
Yi ( + ii )Yi ( ii ) = exp gi ( )
with i being some real number and gi ( ) being a function whose precise form depends
on the particular model. We can formally solve the equation by Fourier transformations

1
i ( ) = 2i cosh(/2/i ) ,
(10)
Yi ( ) = exp (gi i )( ) ,
i.e., substituting (10) into the l.h.s. of (9) yields exp(gi ( )). Of course, this identification
is not completely compelling and we could have chosen also a different combination of
Y s. However, in order to be able to evaluate the gi ( ) we require a concrete functional
input for the function Yi ( ) in form of an approximated function. Choosing here the large
density approximation L0 makes the choice for gi ( ) with hindsight somewhat canonical,
since other combinations lead generally to non-physical answers.
We replace now inside the defining relation of gi ( ) the Y s by Yi ( ) exp(L0i ( )) 1.
Analogously to the approximating approach in [1517], we can replace the convolution by
an infinite series of differentials
i ( ) = (gi i )( ) =
i(m)
m=0
dm
gi ( ),
d m
(11)
where the s are defined by the power series expansion

Z
d i ( ) eit = 2
X
m=0
(m) m
(i)m i
t =
X
E2m
(i )2m t 2m .
(2m)!
(12)
m=0
The Em denote the Euler numbers, which enter through the expansion
X x 2m E2m
1
=
.
cosh x
(2m)!
m=0
In accordance with the assumptions of our previous approximations for the solutions of
the TBA-equations in the large density approximation, we can neglect all higher order
derivatives of the L0i ( ). Thus we only keep the zeroth order in (11). From (12) we read off
(0)
the coefficient i = 1/2, such that we obtain a simply expression for an improved large
density approximation

(13)
Lli ( ) = ln 1 + Yil ( ) = ln 1 + exp(gi ( )/2) .
In principle we could proceed similarly for the small density approximation and replace
now Yi ( ) exp(Lsi ( )) 1 in the defining relations for the gi s. However, in this
622
situation we can not neglect the higher order derivatives of the Lsi such that we have to
keep the convolution in (11) and end up with an integral representation instead. We now
wish to match Lsi and Lli in the transition region between the small and large density
approximations at some distinct value of the rapidity, say im . We select this point to be
the value when the function fi ( ) = (6/ 2 )rmi Li ( ) cosh , which is proportional to the
free energy density for a particular particle species, has its maximum in the small density
approximation

d s
fi ( ) = 0.
(14)
d
m
i
In regard to the quantity we wish to compute, the scaling function (3), this is the point
in which we would like to have the highest degree of agreement between the exact and
approximated solution, since this will optimize the outcome for c(r). Having specified the
im , the matching condition provides a simple rational to fix the constant i

Lli im = Lsi im im .
(15)
Clearly, in general we can not solve these equations analytically, but it is a trivial
numerical problem which is by no means comparable with the one of solving (1). Needless
to say that the outcome of (15) is not to be considered as exact, but as our examples
below demonstrate it will lead to rather good approximations. One of the reasons why
this procedure is successful is that Lsi (im ) is still very close to the precise solution, despite
the fact that is at its worst in comparison with the remaining rapidity range.
Combining the improved large and small density approximation we have the following
approximated analytical L-functions for the entire range of the rapidity
l
Li ( ) for | | 6 im
a
(16)
Li ( ) =
Lsi ( ) for | | > im ,
such that the scaling function becomes well approximated by
n Z
X
c(r) '
d fia ( ).
(17)
i=1 0
To develop matters further and report on the quality of L0 , Lf , Ls , Ll we have to specify

a particular theory at this point.
3. Affine Toda field theory

Affine Toda field theories [1214] form a well studied class of relativistic integrable
quantum field theories in 1 + 1 spacetime dimensions. To each of these field theories a
b(`) ) [20] is associated whose structure allows universal
pair of affine Lie algebras (Xn(1) , X
statements concerning its properties, like the S-matrix, the mass spectrum, the fusing rules,
b(`) denotes a twisted affine Lie algebra w.r.t. a Dynkin diagram automorphism
etc. Here X
b(`) is obtained from
of order `. Both algebras are chosen to be dual to each other, i.e., X
623
the non-twisted algebra Xn(1) of rank n by exchanging roots and co-roots. For Xn(1) simply(1)
b(`) , ` = 1, which is reflected in the quantum
laced both algebras coincide, i.e., Xn
=X
theory by a strong-weak self-duality in the coupling constant. Moreover, the mass spectrum
renormalizes by an overall factor and the poles of the S-matrix in the physical sheet do not
depend on the coupling constant. For non-simply laced Lie algebras these features cease to
be valid. The quantum masses are now coupling dependent and flow between the classical
(1)
b(`) in the weak and strong coupling limit, respectively.
masses associated with Xn and X
Consequently, the physical poles of the S-matrix shift depending on the coupling and the
strong-weak self-duality is broken.
3.1. The universal S-matrix
Remarkably, despite these structural differences the S-matrix of ATFT can be cast into a
universal form covering the simply-laced as well as the non-simply laced case [21,22]. For
our purposes the formulation in form of an integral representation is most useful
Z
Sij ( ) = exp
t
dt
ij (t) sinh ,
t
i
(18)
ij (t) = 8 sinh(th ) sinh(tj H t) [K]q(t )q(t

)
1
ij
(19)
Denoting a q-deformed integer n as common by [n]q = (q n q n )/(q 1 q 1 ), we

introduced here a doubly q-deformed version of the Cartan matrix K [2123] of the
non-twisted Lie algebra

(20)
[Kij ]q q = q q ti + q 1 q ti ij [Iij ]q
for the generic deformation parameters q, q.
The incidence matrix Iij = 2ij Kij of the
(1)
Xn related Dynkin diagram is symmetrized by the integers ti , i.e., Iij tj = Ij i ti . With i
being a simple root we fix the length of the long roots to be 2 and choose the convention
ti = ì2 /2. Inside the integral representation (18) we take
q(t) = et h ,
q(t)
= et H ,
with h :=
2B
,
2h
H :=
B
2H
(21)
for the deformation parameters, where 0 6 B 6 2 is the effective coupling constant. We

(1)
further need the Coxeter numbers h, h and the dual Coxeter numbers h , h of Xn and
b(`) . Complete tables of
b(`) , respectively, as well as the `-th Coxeter number H = `h of X
X
these quantities for individual algebras may be found in [20].
The incidence matrix satisfies the relation [21,22]
n
X
[Iij ]q(i)
mj = 2 cosh(h + ti H )mi ,
(22)
j =1
which will turn out to be crucial for the arguments below. We introduced here the imaginary
angles h = ih and H = iH .
624
3.2. The TBA-kernel

From the universal integral representation (18), we can now immediately derive the
Fourier transformed TBA-kernel (5) for ATFT. However, when taking the logarithmic
derivative one has to be careful about interchanging the derivative with the integral, since
these two operations do not commute. Comparison with the block representation of the
S-matrix [22] yields
1
ij ( ) =
2
Z
dt ij (t) exp
t
,
i
(23)
such that the Fourier transformed universal TBA-kernel (5) acquires the form
ij (t) = 2ij ij (t)
e
= 2ij 8 sinh(th ) sinh(tj H ) [K]q(t )q(t
1
ij
(24)
To be able to carry out the discussion of the previous section we require the second order
(2)
coefficient ij in the power series expansion (5). From (24) we read off directly
ij(2) =
2
2
B(2 B)Kij1 tj = B(2 B)(i j ).
hH
hh
(25)
In the latter equality we used the fact that the inverse of the Cartan matrix is related to the
P
fundamental weights as i = j Kij1 j , ti = ì2 /2 and H = `h = `h . This implies on
the other hand that
2
B(2 B)(i )
hh
P
with = i i being the Weyl vector. Therefore,
i =
n
X
i=1
i = B(2 B)
(26)
22
2 (h + 1)
.
=
nB(2
B)
hh
12h
(27)
We used here the Freudenthalde Vries strange formula 2 = (h /12) dim Xn (see, e.g.,
[24]) and the fact that dim Xn(1) = n(h + 1). Thus we have generalized the result of [17] to
the non-simply laced case. Notice that in terms of quantities belonging to the non-twisted
Lie algebra Xn(1) the formula (27) is identical for the simply laced and the non-simply laced
case.
(1)
3.3. Universal TBA-equations and Y-systems

In analogy to the discussion for simply-laced Lie algebras [17], the universal expression
for the kernel (24) can be exploited in order to derive universal TBA-equations for all
ATFT, which may be expressed equivalently as a set of functional relations referred to
as Y-systems. Fourier transforming (1) in a suitable manner and invoking the convolution
625
theorem we can manipulate the TBA-equations by using the expression (24). After Fourier
transforming back we obtain
i +
n
X
ij Lj =
j =1
n
X
ij (j + Lj ).
(28)
j =1
The universal TBA-kernels and are then given by

1
,
i ( ) = 2(h + ti H ) cosh
2(h + ti H )
ij ( ) =
Iij
X

i + i(2k 1 Iij )H ,
(29)
(30)
k=1

ij ( ) = i + (h ti H ) + i (h ti H ) ij .
(31)
The key point here is that the entire mass dependence, which enters through the on-shell
energies mi cosh , has dropped out completely from the equations due to the identity (22).
Noting further that
mj cosh =
[Iij ]q(i)
Iij
X

mj cosh + (2k 1 Iij )H ,
(32)
k=1
we have assembled all ingredients to derive functional relations for the quantities Yi =
exp(i ). For this purpose we may either shift the TBA-equations appropriately in the
complex rapidity plane or use again Fourier transformations, see [17]
Yi ( +h + ti H )Yi ( h ti H )
[1 + Yi ( + h ti H )][1 + Yi ( h + ti H )]
.
= Q
QIij
n
1
j =1 k=1 [1 + Yj ( + (2k 1 Iij )H )]
(33)
These equations are of the general form (9) and specify concretely the quantities and
gi ( ). We recover various particular cases from (33). In case the associated Lie algebra is
simply-laced, we have h + ti H i/ h, h ti H i/ h(1 B) and Iij 0, 1, such
that we recover the relations derived in [17]. As stated therein we obtain the system for
minimal ATFT [19] by taking the limit B i.
The concrete formula for the approximated solution of the Y-systems in the large density
regime, as defined in (13), reads
Yil ( ) =
cos(2i ) + cos(2(h ti H )i )
4i i2
Iij
1/2
n Y
Y
2j j2
.
1
cos2 (j ( + (2m 1 Iij )H )
(34)
j =1 m=1
Exploiting possible periodicities of the functional equations (33) they may be utilized in
the process of obtaining approximated analytical solutions [25]. As we demonstrated they
can also be employed to improve on approximated analytical solution in the large density
626
regime. In the following subsection we supply a further application and use them to put
constraints on the constant of integration i in (6).
3.4. The constants of integration and
There are various constraints we can put on the constants i and i on general grounds,
e.g., the lower bound already mentioned. Having the numerical data at hand we can
use them to approximate the constant. In [17] this was done by matching L0 with the
numerical data at = 0 and a simple analytical approximation was provided num =
ln[B(2 B)21+B(2B)]. Of course, the idea is to become entirely independent of the
numerical analysis. For this reason the argument which led to (15) was given.
When we consider a concrete theory like ATFT, we can exploit its particular structure
and put additional constraints on the constants from general properties. For instance, when
we restrict ourselves to the simply laced case it is obvious to demand that the constants
respect also the strong-weak duality, i.e., i (B) = i (2 B) and i (B) = i (2 B).
Finally we present a brief argument which establishes that the constants i are in fact
independent of the particle type i. We replace for this purpose in the functional relations
(33) the Y-functions by Yil ( ) and consider the equation at = 0, such that
cosh2 [i (h + ti H )] 2i2 i
cosh2 [i (h + ti H )]
Iij
n Y
Y
(cosh2 [i (2m 1 Iij )H ] 2i2i )1/2
.
=
cosh[i (2m 1 Iij )H ]
(35)
j =1 m=1
Keeping in mind that i is a very small quantity in the ultraviolet regime, we expand (35)
up to second order in i , which yields after cancellation
4ti h H =
j2 j
j2
i2 B(2 B) X
B(2 B) X
=
=
K
K
ij 2
ij 2 (j ).
2
hh
hh
i
i
j =1
(36)
j =1
We substituted here the expression (26) for the constants j in the last equality. Using
P
once more the relation i = j Kij1 j , we can evaluate the inner product such that (36)
reduces to
i2 =
n
n X
X
j =1 k=1
Kij
j2
i2
2
Kj1
k k .
(37)
Clearly this equation is satisfied if all the i are identical. From the uniqueness of the
solution of the TBA-equations follows then immediately that we can always take i
. Since the uniqueness is only rigorously established [17] for some of the cases we are
treating here, it is reassuring that we can obtain the same result also directly from (37).
From the fact that the i are real numbers and all entries of the inverse Cartan matrix are
positive follows that i2 = j2 for all i and j . The ambiguity in the sign is irrelevant for the
use in L( ).
627
3.5. The scaling functions

In [17] it was proven that the leading order behaviour of the scaling function is given by

2 B(2 B)(h + 1)
3
=n 1
.
(38)
c(r) ' n
( ln(r/2))2
4h( ln(r/2))2
From our arguments in Section 3.2, which led to the general expression for the constant
in form of (27), follows that in fact this expression holds for all affine Toda field theories
b(`) ). However, strong-weak
related to a dual pair of simple affine Lie algebras (Xn(1) , X
duality is only guaranteed for ` = 1.
Restricting ourselves to the simply laced case, we can view the results of [810] obtained
by means of a semi-classical treatment for the scaling function as complementary to the one
obtained from the TBA-analysis and compare directly with the expression (38). Translating
the quantities in [8,10] to our conventions, i.e., R r, B B/2, we observe that c(r)
becomes a power series expansion in . We also observe that the second order coefficients
precisely coincide in their general form. Comparing the expressions, we may read off
directly
semi
B

4 ( h1 )( B2 1) 2 1
E
= ln
B
B
k ( h1 2h
) (1 + 2h
)
(39)
for all ATFT related to simply laced Lie algebras. 4 Here E denotes Eulers constant
Q
and k = ( li=1 nni i )1/(2h) is a constant which can be computed from the Kac labels ni
of the related Lie algebra. Contrary to the statement made in [9], this identification can
be carried out effortlessly without the need of higher order terms. Recalling the simple
analytical expression num of [17] we may now compare. Fig. 1 demonstrates impressively
that this working hypothesis shows exactly the same qualitative behaviour as semi and also
quantitatively the difference is remarkably small.
To illustrate the quality of our approximate solutions to the TBA-equations, we shall
now work out some explicit examples.
3.6. Explicit examples
To exhibit whether there are any qualitative differences between the simply laced and
non-simply laced case we consider the first examples of these series.
3.6.1. The sinh-Gordon model
The sinh-Gordon model is the easiest example in the simply laced series and therefore
ideally suited as testing ground. The Coxeter number is h = 2 in this case. An efficient way
to approximate the L-functions to a very high accuracy is
4 The expressions in [8] and [10] only coincide if in the former case m = 1 and in the latter m = 1/2. In
addition, we note a missing bracket in equation (6.20) of [8], which is needed for the identification. Replace C
4QC therein.
628
Fig. 1. Numerically fitted constant num versus the constant from the semi-classical approach for the
sinh-Gordon model semi .

cos(2 ) + cosh((1 B))
ln 1 +
for | | 6 m ,
2
a
4
L ( ) =

for | | > m ,
exp rm cosh + Lf ( )
(40)
with
( ) =
4 sin(B/2) cosh
,
cosh 2 cos B
2 B(2 B)
.
8
(41)
The determining equation for the matching point reads

sinh m

rm
sinh(2 m ) + cosh( m ) 0 Lf ( m ) = 0.
2
(42)
For instance, for B = 0.4 this equation yields m = 11.9999 such that the matching
condition (15) gives m = 0.4913. Fig. 2(a) shows that the large and small density
approximation L0 and Lf may be improved in a fairly easy way. In view of the simplicity
of the expression La the agreement with the numerical solution is quite remarkable.
Fig. 2(a) also illustrates that when using the constant semi instead of m the agreement
with the numerical solutions appears slightly better for small rapidities. When we employ
num instead of semi the difference between the two approximated solutions is beyond
resolution. However, as may be deduced from Fig. 2(b), with regard to the computation of
the scaling function the difference between using m instead of semi is almost negligible.
Whereas in the former case the resulting value for the scaling function is slightly below the
correct value, it is slightly above by almost the same amount in the latter case. More on the
approximation of the scaling function in form of (38) may be found in [17].
629
Fig. 2. Various L-functions and free energy densities for the sinh-Gordon model (a), (b) at B = 0.4
(1)
(3)
and r = 105 and (G2 , D4 )-ATFT at B = 0.5 and r = 105 (c), (d).
(1)
(3)
3.6.2. (G2 , D4 )-ATFT

In this case we have h = 6 and H = 12 for the related Coxeter numbers. The two masses
are m1 = m sin((1/6 B/24)) and m2 = m sin((1/3 B/12)). The L-functions are
well approximated by
#
"
s
2
cos(2 ) + cosh 13 B4
2
ln 1 +
1
for | | 6 1m ,
2
2 ( )
a
4
cos
1
L1 ( ) =

f
f
for | | > 1m ,
exp rm1 cosh + 11 L1 + 12 L2 ( )
"
#
s

1
5B
2
1
cos(2
)
+
cosh
3
24
ln 1 +
1
42 2
cos2 (( + kB
k=1
12 ))
m
for | | 6 2 ,
La2 ( ) =

f
f
exp rm2 cosh + 21 L1 + 22 L2 ( )
for | | > 2m ,
with given by (23) and
7 2 B(2 B)
5 2 B(2 B)
2 B(2 B)
,
2 =
,
=
.
(43)
72
8
36
Using now the numerical data L1 (0) = 4.2524 and L2 (0) = 3.67144 as benchmarks, we
compute by matching them with La1 (0) and La2 (0) the constant to = 1.1397 in both cases.
This confirms our general result of Section 3.4. Evaluating the Eqs. (15) and (14) we obtain
for B = 0.5 the matching values for the rapidities 1m = 12.744 and 2m = 12.278 such
1 =
630
that 1m = 1.9539 and 2m = 1.5572. Fig. 2(c) and 2(d) show a good agreement with the
numerical outcome.
The approximated analytical expression for the scaling function reads
c(r) ' 2
7 2 B(2 B)
.
12( ln(r/2))2
(44)
This expressions differs from the one quoted in [17], since in there the sign of some
scattering matrices at zero rapidity was chosen differently.
4. Conclusions
We have demonstrated that it is possible to find simple analytical solutions to the TBAequation in the large and small density regime, which approximate the exact solution to
high accuracy. By matching the two solutions at the point in which the particle density and
the density of available states coincide, it is possible to fix the constant of integration, which
originated in the approximation scheme of [1517] and was left undetermined therein.
Alternatively the constant may be fixed by a direct comparison with a semi-classical
treatment of the problem. It is not necessary for this to proceed to higher order differential
equations as was claimed in [9]. Of course one may proceed further to higher orders,
but since the solutions to the higher order differential equations may only be obtained
approximately one does not gain any further structural insight and moreover one has lost
the virtue of the first order approximation, its simplicity.
We derived the Y-systems for all ATFT and besides demonstrating how they can be
utilized to improve on the large density approximations we also showed how they can be
used to put constraints on the constant of integration.
We have proven that the expression (38) for the scaling function is of a general nature,
i.e., valid for all ATFT. It is desirable to extend the semi-classical analysis [10] also to the
non-simply laced case. This would allow to read off the constant also in that case.
Note added in proof

After the submission of this manuscript a paper appeared [26] which generalizes the
procedure carried out for the simply laced algebras [810] also to the non-simply laced
case. Once again the leading order of the scaling function (38) coincides precisely with our
analysis.
Acknowledgements
The authors are grateful to the Deutsche Forschungsgemeinschaft (Sfb288) for financial
support. We acknowledge constructive conversations with B.J. Schulz at the early stage of
this work and are grateful to V.A. Fateev for kind comments.
631
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
P. Weisz, Phys. Lett. B 67 (1977) 179.

M. Karowski, P. Weisz, Nucl. Phys. B 139 (1978) 445.
Al.B. Zamolodchikov, Nucl. Phys. B 342 (1990) 695.
T.R. Klassen, E. Melzer, Nucl. Phys. B 338 (1990) 485; Nucl. Phys. B 350 (1991) 635.
Al.B. Zamolodchikov, Nucl. Phys. B 348 (1991) 619.
V.A. Fateev, Phys. Lett. B 324 (1994) 45.
Al.B. Zamolodchikov, Int. J. Mod. Phys. A 10 (1995) 1125.
A.B. Zamolodchikov, Al.B. Zamolodchikov, Nucl. Phys. B 477 (1996) 577.
C. Ahn, C. Kim, C. Rim, Nucl. Phys. B 556 (1999) 505.
C. Ahn, V.A. Fateev, C. Kim, C. Rim, B. Yang, Reflection amplitudes of ADE Toda theories
and thermodynamic Bethe ansatz, hep-th/9907072.
Al.B. Zamolodchikov, in: 4-th Bologna Workshop on CFT and Integrable Models, Bologna,
1999.
A.V. Mikhailov, M.A. Olshanetsky, A.M. Perelomov, Commun. Math. Phys. 79 (1981) 473.
G. Wilson, Ergod. Th. Dyn. Syst. 1 (1981) 361.
D.I. Olive, N. Turok, Nucl. Phys. B 257 [FS14] (1985) 277.
Al.B. Zamolodchikov, Resonance factorized scattering and roaming trajectories, ENS-LPS-335,
preprint, 1991.
M.J. Martins, Nucl. Phys. B 394 (1993) 339.
A. Fring, C. Korff, B.J. Schulz, Nucl. Phys. B 549 (1999) 579.
A. Bytsko, A. Fring, Nucl. Phys. B 532 (1998) 588.
Al.B. Zamolodchikov, Phys. Lett. B 253 (1991) 391.
V.G. Kac, Infinite Dimensional Lie Algebras, CUP, Cambridge, 1990.
T. Oota, Nucl. Phys. B 504 (1997) 738.
A. Fring, C. Korff, B.J. Schulz, Nucl. Phys. B 567 (2000) 409.
E. Frenkel, N. Reshetikhin, Deformations of W-algebras associated to simple Lie algebras, qalg/9708006 v2.
P. Goddard, D.I. Olive, Int. J. Mod. Phys. A 1 (1986) 303.
Al.B. Zamolodchikov, Nucl. Phys. B 358 (1991) 497; Nucl. Phys. B 358 (1991) 524; Nucl.
Phys. B 366 (1991) 122.
C. Ahn, P. Baseilhac, V.A. Fateev, C. Kim, C. Rim, Reflection amplitudes in non-simply laced
Toda theories and thermodynamic Bethe ansatz, hep-th/0002213.

Short distance behaviour of correlators in the 2d

Ising model in a magnetic field
M. Caselle a, , P. Grinza a,1 , N. Magnoli b,2
a Dipartimento di Fisica Teorica dellUniversit di Torino and Istituto Nazionale di Fisica Nucleare,
Sezione di Torino, via P. Giuria 1, I-10125 Torino, Italy

b Dipartimento di Fisica, Universit di Genova and Istituto Nazionale di Fisica Nucleare,
Sezione di Genova, via Dodecaneso 33, I-16146 Genova, Italy

Received 21 September 1999; revised 16 February 2000; accepted 9 March 2000
Abstract
We study the h i, h i, hi correlators in the 2d Ising model perturbed by a magnetic field. We
compare the results of a set of high precision Monte Carlo simulations with the predictions of two
different approximations: the Form Factor approach, based on the exact S-matrix description of the
model, and a short distance perturbative expansion around the conformal point. Both methods give
very good results, the first one performs better for distances larger than the correlation length, while
the second one is more precise for distances smaller than the correlation length. In order to improve
this agreement we extend the perturbative analysis to the second order in the derivatives of the OPE
constants. 2000 Elsevier Science B.V. All rights reserved.
PACS: 11.25.Hf; 11.15.Bt; 11.10.-z; 11.10.Gh
Keywords: Ising model; Conformal field theories; Perturbation theory
1. Introduction
In these last years much progress has been done in the study of two-dimensional
statistical systems in the neighbourhood of critical points. In the framework of quantum
field theory these systems can be seen as conformal field theories (CFTs) perturbed by
some relevant operator. Since the seminal work of Belavin et al. [1] we have an almost
complete understanding of CFTs (at least for the minimal models): we have complete lists
of all the operators of the theories and explicit expressions for the correlators. However
much less is known on their relevant perturbations. In some cases it has been possible to
caselle@to.infn.it
1 grinza@to.infn.it
2 magnoli@ge.infn.it
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 1 6 1 - 9
636
M. Caselle et al. / Nuclear Physics B 579 [FS] (2000) 635666
show that these perturbations give rise to integrable models [2,4]. In these cases again we
have a rather precise description of the theory. In particular it is possible to obtain the exact
asymptotic expression for the large distance behaviour of the correlators [5]. From this
information several important results (and in particular all the universal amplitude ratios)
can be obtained.
However in comparing with numerical simulations or with experiments one is often
interested in the short distance behaviour of the correlators (short here means for distances
smaller or equal than the correlation length) and this is not easily accessible in integrable
systems. Moreover integrable perturbations represent only a small subset of the possible
theories. For instance in the case of the Ising model both the purely thermal and the
purely magnetic perturbations are integrable, but for any combination of them the exact
integrability is lost.
For these reasons, besides the S-matrix results, it is important to develop a perturbative
approach well defined in the short distance regime of the theory and such that it does
not rely on the exact integrability of the model. This is however a rather difficult task. In
fact any naive perturbative expansion of the (massless) CFT along a relevant direction, is
affected by infrared divergences (IR) and some non-trivial strategy is needed.
Recently, in [7,8], a new approach has been proposed to overcome this difficulty (see
[920] for relevant related works and preexisting ideas). The method is based on Wilsons
operator product expansion (OPE). Roughly speaking the main idea of this new approach
is that the Wilsons coefficients of the OPE, being well defined at short distance, can be
assumed to have a regular, IR safe, perturbative expansion with respect to the coupling.
For this reason we shall refer to it in the following as the IRS (infrared safe) perturbative
approach.
The main requirement of this IRS approach is the the knowledge of the Wilson
coefficients (and their derivatives with respect to the perturbing coupling). For this reason it
is particularly efficient if applied to perturbations of exactly solved theories like 2d critical
CFTs (but the framework is quite general and in principle could be extended also to higher
dimensions).
The price one has to pay to control in this way the IR divergences is that one needs,
as an external input information, the expectation values of the operators involved in the
expansion. There are at this point two possibilities. The first one is to concentrate only
on observables in which these expectation values exactly cancel. This is a small but very
interesting subset of the informations that we can obtain with the IRS perturbation.
The second possibility is to obtain the desired expectation values with some other
method or extract them from numerical simulations (an interesting numerical approach
to obtain these VEV is based on the truncated conformal space technique, see [21,22] and
references therein).
From this point of view, the IRS approach becomes particularly powerful if applied to
integrable perturbations, since in this case some of the expectation values can be deduced
from the S-matrix of the model.
The last step one has to face in comparing the results of the IRS method with
simulations or experiments is the presence of a nonuniversal normalization factor between
637
the operators in the continuum quantum field theory and their lattice discretizations. These
normalizations (and the related normalization of the coupling of the perturbation) can be
fixed if an exact solution of the lattice model exists at the critical point. Actually much less
is needed. One only needs the exact expression (or even only its large distance asymptotic
form) of a correlator involving the operators in which we are interested. This makes the
Ising model perturbed by a magnetic field a perfect candidate for testing the IRS method. In
fact it is well known that this model is exactly integrable [2,4] and all the amplitude ratios
and expectation values of the primary fields are known. Moreover the Ising model is exactly
solvable at the critical point and the exact expression is known for several correlators [23,
24].
In fact the IRS approach was successfully tested with the magnetic perturbation of the
Ising model in [8]. The aim of this paper is to make a further step in this direction. In
particular we have three goals:
(a) Compare the results of the method with new high precision Monte Carlo simulations
so as to test the range of applicability of the method.
(b) Compare the IRS method with the results obtained in the S-matrix framework with
the so called form factor (FF) approach.
(c) Show that it is possible to extend the analysis of [7,8] to higher orders in the
perturbative coupling and discuss the technical problems that one has to afford
following this route.
In particular we shall study in this paper, as an example, the second order term in
the perturbative expansion of the hi correlator. The reason for this choice is that this
correlator has a very peculiar behaviour since its first order correction turns out to be
exactly zero, thus our second order calculation is mandatory if one is interested to study
the influence of the magnetic field on the simple critical behaviour of the correlator.
This paper is organized as follows. Section 2 is devoted to a general description of the
Ising model in a magnetic field both on the lattice and in the continuum. The aim of this
section is to fix conventions and normalizations which will be useful in the following.
In Section 3 we shall briefly describe the IRS method, while in Section 4 we shall
extend it to second order derivatives of the magnetic field. In Section 5 we shall briefly
describe our Monte Carlo simulation while in Section 6 we shall compare the results of
our simulations with IRS and FF predictions. Finally Section 7 will be devoted to some
concluding remarks. The details of the calculation of the second order derivative of the
Wilson coefficient are collected in the appendix. We have reported in three tables at the
end of the paper a sample of the results of our simulations.
2. Ising model in a magnetic field

The continuum theory, which is the starting point of the IRS expansion is given by the
action:
Z
(1)
A = A0 + h d 2 x (x),
638
where A0 is the action of the conformal field theory which describes the Ising model at the
critical point. Let us start our analysis by looking in detail at this CFT.
2.1. The Ising model at the critical point
The Ising model at the critical point is described by the unitary minimal CFT with central
charge c = 1/2. It contains three conformal families whose primary fields 1, , have
scaling dimensions x = 0, 1/8, 1, respectively. The fusion rule algebra is
[][] = [1],
[ ][] = [ ],
(2)
[ ][ ] = [1] + [].
Once the operator content is known, the only remaining information which is needed to
completely identify the theory are the OPE constants. The OPE algebra is defined as
X {k}
Cij (r){k} (0),
(3)
i (r)j (0) =
{k}
where with the notation {k} we mean that the sum runs over all the fields of the conformal
family [k]. The structure functions Cijk (r) are c-number functions of r which must be single
valued in order to take into account locality. In the large r limit they decay with a power
like behaviour
Cijk (r) |r|
dim(Cijk )
(4)
whose amplitude is given by

k
b kij lim Cijk (r)|r|dim(Cij ) .

C
r
(5)
The actual value of these constants depends on the normalization of the fields, which can
be chosen by fixing the long distance behaviour of, for instance, the and correlators.
In this paper we follow the commonly adopted convention which is: 3
1
, |x| ,
|x|1/4
1
h(x)(0)i = 2 , |x| .
|x|
h (x) (0)i =
(6)
(7)
With these conventions we have, for the structure constants among primary fields
b , = C
b , = 0,
b , = C
C
(8)
b ,1 = C
b 1, = C
b = 1
b 1, = C
C
,1
(9)
and
b , = 1 .
b , = C
C
2
(10)
3 Notice the change of normalization with respect to [8]. The operator of the present paper corresponds to 2
times that of Ref. [8].
639
2.2. The Ising model in a magnetic field

If we switch on the magnetic field h, the structure functions acquire a h dependence so
that we have in general
X {k}
Cij (h, r){k} (0).
(11)
i (r)j (0) =
{k}
Also the mean values of the and operators acquire a dependence on h. Standard
renormalization group arguments allow one to relate this h dependence to the scaling
dimensions of the operators of the theory and lead to the following expressions:
h ih = A h1/15 + ,
(12)
hih = A h
(13)
8/15
+ .
The exact value of the two constants A and A can be found in [25] and [26], respectively,
A =
with
C=
2 C2
2
15 (sin 2
3 + sin 5 + sin 15 )
= 1.27758227 . . . ,

4/5
4 sin 5 ( 15 ) 4 2 ( 34 ) 2 ( 13
16 )
8
( 23 ) ( 15
)
3
( 14 ) 2 ( 16
)
(14)
(15)
and
A = 2.00314 . . . .
(16)
Notice however that these amplitudes are not universal. They depend on the details of
the regularization scheme. Thus some further work is needed to obtain their value on the
lattice.
2.3. The lattice model
The lattice version of the above model is defined by the following partition function:
X (P +H P )
hi,ji i j
i i ,
e
(17)
Z=
i =1
where the notation hi, j i denotes nearest neighbour sites in the lattice which we assume
to be a two-dimensional square lattice of size L. In order to select only the magnetic
perturbation, must be fixed to its critical value:

= c = 12 log 2 + 1 = 0.4406868 . . .
finally by defining hl = c H we find
X P +h P
e c hi,ji i j l i i .
Z=
i =1
(18)
640
In the following we shall denote the lattice discretization of the operators , with the
index l. The magnetization M(h) is defined as usual:

X

1
1
(log Z)
=
i ,
(19)
M(h)
N h
N
l
=c
L2
denotes the number of sites of the lattice. This result suggests the following
where N
definition for the lattice discretization of
1 X
i ,
(20)
l
N
i
so that the mean value of l coincides with M(h):

hl i M(h).
(21)
Similarly, we define the internal energy as:

1 X
1
(log Z)
=
i j .
E(h)
2N h
2N
l
=c
(22)
hi,j i
For the energy operator one must also take into account the presence of an additional
bulk contribution at the critical point. This constant
can be easily evaluated (for instance
by using KramerWannier duality) to be 0 = 1/ 2. This result suggests, for the lattice
discretization of , the following definition
1 X
1
(23)
i j
l
2N
2
hi,j i
so that the mean value of l coincides with the singular part of E(h):
1
hl i E(h) .
2
According to the above discussion we expect:
(24)
hl ih = Al hl
1/15
+ ,
(25)
8/15
hl ih = Al hl
+ ,
(26)
where the lattice amplitudes Al , Al are different from the corresponding amplitudes
evaluated in the continuum.
In order to relate the lattice results with the continuum ones, we must fix the relative
normalizations of versus l , versus l and h versus hl . 4 The simplest way to do this is
to look at the analogous of Eqs. (6), (7) at the critical point (namely for hl = 0) [27]. From
the exact solution of the Ising model [23] we know that
hi j ih=0 =
R2
,
|rij |1/4
(27)
4 This essentially amounts to measure all the quantities in units of the lattice spacing. For this reason we can
fix in the following the lattice spacing to 1 and neglect it.
641
where rij denotes the distance on the lattice between the sites i and j and we know from
[24] that:
0
R2 = e3 (1) 25/24 = 0.70338 . . . .
(28)
By comparing this result with Eq. (6) we find

l = R = 0.83868 . . . .
(29)
From this we can also obtain the normalization of the lattice magnetic field which must
exactly compensate that of the spin operator in the perturbation term h . We find:
hl = (R )1 h = 1.1923 . . . h.
(30)
Combining these two results we obtain the value in lattice units of the constant A
Al = (R )16/15A = 1.058 . . . .
(31)
In the case of the energy operator the connected correlator on the lattice, at hl = 0 and for
any value of , has the following expression [28]:
2

2
(32)
K1 (r) K02 (r) ,
l (0)l (r) c =
where K0 and K1 are modified Bessel functions, is a parameter related to the reduced
temperature, defined as
= 4| c |
(33)
and with the index c we denote the connected correlator (notice that thanks to the definition
(23) no disconnected part must be subtracted at the critical point and the index c becomes
redundant). This expression has a finite value in the 0 limit (namely at the critical
point). In fact the Bessel functions difference can be expanded in the small argument limit
as

K12 (r) K02 (r) =
1
+
(r)2
(34)
thus giving, exactly at the critical point:

l (0)l (r) =
1
R2
.
(r)2 |r|2
(35)
Hence R = 1/ . By comparing this result with Eq. (7) we find

l = R =
(36)
and from this we obtain the expression in lattice units of A

Al = (R )8/15(R )A = 0.58051 . . . .
(37)
642
2.4. Correlators
In the remaining part of this paper we shall be mainly interested in the dependence on
the external magnetic field of the following correlators:

(38)
G, (0) (r) ,

(39)
G, (0)(r) ,

(40)
G, k (0)(r) ,
where k sign(h). We already know the behaviour at the critical point of the first two
of them, which is given by Eqs. (6), (7) in the continuum (or equivalently Eqs. (27),
(35) on the lattice), while the OPE constants reported in Eq. (8) immediately tell us that
h(0) (r)i = 0.
For small values of h we may expect to add to these results correction terms functions of
h and r. However standard renormalization group arguments show that these two variables
are actually related and that there is a natural scaling variable which describes the short
distance expansion of these correlators in a magnetic field which is t |h||r|15/8. In order
to obtain an explicit expansion in powers of t we must absorb the scaling dimensions of
the various operators in the expansion. 5 To this end let us define

(41)
F, (0) (r) |r|1/4,
2
(42)
F, (0)(r) |r| ,
9/8
(43)
F, k (0)(r) |r| ,
where k sign(h).
The powers which appear in the t expansion of the functions F can be immediately
deduced from the analysis of the OPE via the IRS method, that will be described in the
following section.
3. The infrared safe approach

The goal of the method presented in Ref. [7] is to obtain informations about the short
distance behavior of a conformal field theory perturbed by relevant operators.
The general idea behind this approach, (for preexisting ideas see [1317]) is the fact that
Wilson coefficients, being short distance objects, can be taken to have a regular, IR safe,
perturbative expansion with respect to the coupling. This OPE approach leaves unfixed
some constants that parameterize the vacuum expectation values of operators that appear
in conformal field theory.
In [7] it was found that the correlators of the perturbed CFT are given in terms of the
derivatives of the Wilson coefficients (calculated at h = 0 point). To be precise, they appear
in the following way
5 On the contrary, in the large distance regime where one may use the predictions obtained with the form factor
approach, the natural normalization is that of the G, functions defined above.
643
X {k}

Cij (h, r) {k} (0) h
i (r)j (0) h =
=
X
{k}
{k}
{k}
{k}
{k}
Cij (0, r) + h Cij (0, r)h + 12 h2 Cij (0, r)h2

+ {k} (0) h .
It was also shown that a general formula could be written for the nth derivative of the
Wilson coefficients with respect to h. Here we will write only the first and the second order
derivatives for the Wilson coefficients,
*"
! #+
Z 0
X
X
1 b
2
b
h Ca1 a2 hb XR i =
d z z a b
Ca1 a2 b XR
(44)
b
and
X
Z
h2 Cab1 a2 hb XR i =
Z
2
d z
*"
2 0
z z0 a b
d z
Z
h Cab1 a2
!
Cab1 a2 b
#+
XR
b
0

d2 z [z b XR ] .
(45)
The general structure of this formula is a sum of the naive perturbative term plus n (for
the nth order coefficient) infrared counterterms. The asterisk reminds that the sum on the
counterterms is truncated and is performed up to a given infrared dimension (see again [7]).
The construction of the IRS expansion requires two steps.
First, one must select by using the OPE rules which operators can appear in the various
expansions, identify their scaling dimensions, select the dominant ones and give their
expression in terms of the structure constants and of their derivatives.
Second, one must evaluate the derivatives of the structure constants by using Eq. (44)
or (45) to reduce them to suitable integrals over correlators evaluated at the critical
point. This allows in principle to complete the analysis, since the explicit form of all
possible critical correlators is known. However in general these integrals are highly
non-trivial and their evaluation represents the major problem of the whole approach.
These two steps where performed in [8] for all the terms in the expansion involving
at most first order derivatives of the structure constants. This allows to obtain the first
three terms in the expansion of the h i and h i correlators (which are reported for
completeness at the end of this section). On the contrary for the hi in this way one can
only obtain the first two terms of the expansion. Moreover one can verify by an explicit
calculation that the second of them is identically zero ([7]). Thus in the hi correlator,
in order to reach the first non-trivial correction to scaling, it is mandatory to extend the
analysis of [8] and to deal with second order derivatives of the Wilson coefficients. We
shall address this problem in the next section. In particular, in Section 4.1 we shall discuss
the first step of the IRS analysis, and select among the possible candidates the one with the
lowest power of t which, as anticipated, turns out to involve a second order derivative of
a structure constant. Then in Section 4.2 (and in the appendix) we shall explicitly evaluate
this contribution. Let us conclude this section by listing for all the three correlators the first
three terms of the IRS expansion
644
F, = B1 + B2 t 8/15 + B3 t 16/15 + O(t 2 ),

1
2 16/15
3 2
+ B
t
+ B
t + O(t 32/15),
F, = B
F, = B1 t 1/15 + B2 t + B3 t 23/15 + O(t 31/15),
(46)
(47)
(48)
i
are
where the coefficients B
b 1 ,
B1 = C
b ,
B2 = A C
b ,
B3 = A h C
1 =C
b 1 ,
B
2 =A C
B
h b ,
3 = 1 2C
b1
B
2 h ,
b ,
B1 = A C
b 1 ,
B2 = h C
b .
B3 = A h C
b k is the extension
the derivatives of Cikj are given by (44) and the notation h C
i j
of the definition given in Eq. (5) to the derivatives of the Wilson coefficients. The
first order derivatives have been calculated in [8] we report here for completeness their
numerical value. Notice a slight change with respect to [8] due to the different choice of
normalizations for the operator (the of the present paper corresponds to 2 times that
of Ref. [8])
b = 0.40374,
h C
b = 0,
h C
b 1 = 3.29627,
h C
b = 0.90900.
h C
The second order derivative which appears in last one in the hi correlator requires a
more involved calculation which we shall discuss in the next section and in the appendix.
4. Second order corrections

4.1. Dimensional analysis
To estimate the higher order corrections to hi we must analyse two kinds of possible
contributions:
The expectation values of secondary operators multiplied by the Wilson coefficients
and their first derivatives;
The second derivatives of Wilson coefficients.
We would like to understand which is the most important of these terms. In the Ising
model there are two secondary operators at first level, obtained by acting on and with
the Virasoro generator L1 and its hermitian conjugate (the action of L1 on 1 gives 0).
We start by considering
1 L1 L 1
(49)
1 L1 L 1 ,
(50)
and
645
where Lk , L k are Virasoro generators. It is clear that the expectation value of this kind of
operators is zero being total derivatives. So let us go to second level of the algebra. There
are two possible terms:
L21 L 21 ,
L2 L 2 ,
(51)
where is a generic primary field. In this situation we have to consider also the identity
operator.
In the identity sector the only contribution is given by
T T = L2 L 2 1,
(52)
i.e., the energy-momentum tensor. Again a simple dimensional analysis shows that
dim T T = 4,
hT T ih = AT T |h|32/15
(53)
giving
b TT t 32/15 .
AT T C
(54)
It is clear that the terms containing secondary operators (of second level) of and are of
higher order in t and will not be considered here.
A second possible contribution is given by the higher order derivative
*"
! #+
Z 0
Z 0
X
X
2 b
2
2 0
b
h Ca1 a2 hb XR i =
d z
d z z z0 a b
Ca1 a2 b XR
b
Z
h Cab1 a2
b
0

d2 z [z b XR ] .
(55)
Let us fix XR = 1. An elementary computation shows that the series is truncated and only
those operators having xb 6 15
8 appear in it.
It follows that
Z 0
Z 0
Z 0

1
1
=
d2 z
d2 z0 z z0 r 0 C
d2 z hz 0 i,
(56)
+ h C
h2 C
= 0, and we can say that
but h C
Z 0
Z 0

2 1
2
1
d z
d2 z0 z z0 r 0 C
.
h C =
Again from dimensional analysis we get

b 1 = 14
dim h2 C
8
and the contribution to hi of the second order derivative is given by
1 2 1 2
2 h C t .
(57)
(58)
(59)
It is also clear that, being XR = 1 the lowest dimension operator, derivatives of Wilson
coefficients relative to and will give terms with a higher power in t.
Let us write finally the perturbative expansion of this correlator

b 1 + 1 h2 C
b 1 t 2 + O t 32/15 .
(60)
F = C
2
646
4.2. The Wilson derivative

Let us remember that
Z 0
Z 0

1
1
=
d2 z
d2 z0 z z0 r 0 C
,
h2 C
(61)
where h (z1 ) (z2 )(z3 )(z4 )i denotes the correlator at the critical point which can be
written as

|z12 (z32 + z42 ) 2z32 z42 |2
.
(z1 ) (z2 )(z3 )(z4 ) =
4|z42 z32 z41 z31 ||z43|2 |z12 |1/4
(62)
By fixing the values of z1 = z, z2 = w, z4 = 0 and by rescaling r we can choose z3 = 1,

we get
Z 0
Z 0
|z(1 w) + w(1 z)|2
2 1
2
d w
d2 z
+ ,
(63)
h C =
4|w(1 w)z(1 z)||z w|1/4
where the dots indicate the counterterms.
The explicit calculation of this integral can be done using a technique developed by
Mathur [29]. The general idea behind this approach is to factorize the integral in a
holomorphic and antiholomorphic part using Stokes theorem. The calculation is reported
in the appendix. After this calculation, in order to get rid of the infrared cutoff we perform
a Mellin transform of the integral and the infrared counterterm (see [8] for more details).
In this we we end up with a finite result when the infrared cutoff goes to infinity. The final
result is
b 1 = 97.5936 . . . .
h2 C
(64)
Let us stress that the techniques that we have discussed in this section can be extended
to any order in the derivatives of the Wilson coefficients. This is a rather important
observation, since it allows, in principle, to study the IRS corrections, in a consistent way,
to any given order in t.
5. The Monte Carlo simulation

It has been recently shown [30] that in the case of the 2d Ising model in a magnetic field,
algorithms based on the exact (or approximate) diagonalization of the transfer-matrix are
much more effective than standard Monte Carlo simulations. In particular this is true for
all possible observables involved in the large distance behaviour of the model. The only
exception is represented by the short distance behaviour of the pointpoint correlators
which is the subject of the present paper. In fact in order to reach lattices as large as
possible in the transfer-matrix programs discussed in [30] only zero momentum projected
observables could be studied, while we are instead interested in pointpoint correlators.
Moreover, we need to have a window as large as possible between the region (few lattice
spacings) dominated by the lattice artifacts and the correlation length. This windows
647
shrinks to zero in the transfer-matrix approach where only small values of the correlation
length can be studied.
For these reasons we decided to perform our tests with standard Monte Carlo
simulations. We used a SwendsenWang type algorithm, modified so as to take into
account the presence of an external magnetic field. For a detailed description of the
algorithm see for instance [31,32].
5.1. Finite size effects
As a preliminary test we performed a simulation at hl = 4.4069 104 (which
corresponds to H = 0.001) with lattice size L = 128 which exactly coincides with one
of the simulations reported in [31] and found results in complete agreement with those
quoted in [31]. Then for the same value of hl we performed a set of high precision
simulations varying the lattice size so as to check the presence of possible finite size effects.
In particular we compared our estimates of the mean magnetization, susceptibility and
internal energy with the known exact results, extrapolated at the value of hl at which we
performed the simulations. 6
The comparison is reported in Table 1. It turns out that lattice sizes at least larger than
12 times the correlation length are needed to be sure that finite size effects are under
control (with this we mean that the systematic errors induced by the finite size of the
lattice are smaller than the statistical errors of the simulations and can be neglected). A side
consequence of this observation is that the simulations reported in [31] are indeed affected
by rather large finite size effects.
Table 1
Finite size effects at hl = 4.4069 104 . In the first column we report the lattice sizes used in the
simulations, in the remaining three columns the mean values of the magnetization, susceptibility and
internal energy. In the last row we report the exact results obtained by using the known values of the
amplitudes for these quantities
L
120
140
160
200
exact
M
0.63110(16)
0.63247(16)
0.63245(14)
0.63255(11)
0.63260
113(1)
98.4(6)
96.4(4)
95.4(3)
95.7
E
0.71638(3)
0.71645(3)
0.71639(3)
0.71643(3)
0.71642
6 If one is interested in a high precision comparison, also the contribution of secondary fields should be taken
into account in extracting these exact estimates. The amplitude of some of these secondary fields have been
evaluated numerically in [30]. In the case of M and , for the values of hl in which we are interested, the
contributions of the secondary fields are strictly smaller than the statistical errors of the results of our simulations
and hence can be neglected in the comparison. On the contrary for the internal energy it turns out that the
amplitude of the first correction is rather large (see [30] for details) and must be taken into account. In fact,
if we would neglect it, instead of the value reported in Table 1 we would find E = 0.71652 in clear disagreement
with the Monte Carlo results. This represents a non trivial test of the results of [30].
648
It is interesting to notice that the magnetic observables are more affected by finite size
effects than the thermal one. As one can easily expect the largest corrections appear in the
case of the susceptibility.
5.2. The simulation
Once we were sure to have finite size effects under control we performed a set of high
precision simulations of the model for three different values of the magnetic field.
An important quantity to understand the range of validity of the IRS approximation
is the correlation length. Roughly speaking we expect that the IRS results should give a
reasonable approximation for distances equal or smaller than the correlation length, while
above it the form factor approach should give results of better quality. For this reason it is
important to have a good estimate of . This can be easily obtained from the knowledge of
the spectrum of the theory. We find, in lattice units:
8/15
(hl ) = 0.24935 . . .hl
(65)
(see [30] for details on the continuum to lattice conversion of ). In Table 2 we have
reported the expected values of in our cases.
For all the values of hl that we simulated, we studied the three correlators: h (0) (r)i,
h (0)(r)i and h(0)(r)i, for r = 1, . . . , Lmax , where the maximum distance Lmax was
chosen to be roughly twice the correlation length. In this way we can test our results also in
the large distance regime, where predictions from the form factor approach are expected to
give very precise estimates for the correlators. Notice that when studying the large distance
behaviour of correlators one is usually interested in the zero momentum projection of the
connected part of the correlator. On the contrary in the present case we are interested in
the pointpoint correlators without mean value subtraction or zero momentum projections.
This must be taken into account when comparing the data with those obtained with the
form factor approach. Some informations on the simulations are reported in Table 3.
We report an example of our results (for the value hl = 4.4069 104 ) in the first
columns of Tables 7, 8 and 9. The quoted errors have been obtained with a standard
Jacknife method.
Table 2
Values of the correlations lengths for the three choices of hl
hl
4.4069 104
2.2034 104
1.1017 104
15.4
22.4
32.2
649
Table 3
Some informations on the simulations. Lmax denotes the maximum distance at which the correlators
have been evaluated, its value almost coincides with twice the correlation length. L denotes the lattice
size, hl the magnetic field. In the third column we have reported the number of measures while in
the fourth column we have reported the number of SW sweeps which separates two measures
hl
Measures
Sweep/Measures
Lmax
4.4069 104
2.2034 104
1.1017 104
200
300
400
4 105
2 105
1 105
5
5
10
30
45
65
6. Discussion of the results

In Figs. 18 and Tables 7, 8 and 9 we compare our estimates for the correlators with the
IRS and form factor predictions. For completeness we briefly recall here the form factor
results (see [33,34] for details) and give the numerical values (once all the conversion
factors are taken into account) of the constants in the IRS approach.
6.1. Form factor results
The scattering theory which describes the scaling limit of the Ising model in a magnetic
field [4] contains eight different species of self-conjugated particles Aa , a = 1, . . . , 8, with
masses
m2 = 2m1 cos = (1.6180339887 . . .)m1 ,

5
= (1.9890437907 . . .)m1 ,
m3 = 2m1 cos
30
7
= (2.4048671724 . . .)m1 ,
m4 = 2m2 cos
30
2
= (2.9562952015 . . .)m1 ,
m5 = 2m2 cos
15
= (3.2183404585 . . .)m1 ,
m6 = 2m2 cos
30
7
= (3.8911568233 . . .)m1 ,
m7 = 4m2 cos cos
5
30
2
= (4.7833861168 . . .)m1 ,
(66)
m8 = 4m2 cos cos
5
15
m1 (hl ) denotes the overall mass scale and coincides with 1/ , hence its value in lattice
units is
8/15
m1 (hl ) = 4.0104 . . . hl
(67)
From the knowledge of the masses and of the form factors it is possible to obtain a large
distance approximation for the correlators by constructing a spectral sum over a complete
set of intermediate states. We thus find for any pair of local operators 1 and 2 :
650

G1 2 (x) 1 (x)2 (0)
Z
X
=
n=0 > >>
n
1
2
d1
dn
2
2
(1 , . . . , n )Fa1 1,...,an (1 , . . . , n )e|x|

Fa1 2,...,a
n
Pn
k=1 mk cosh k
(68)
where the form factors Fa1 ,...,an (1 , . . . , n ) are the matrix elements of the local operator
(x) on the asymptotic states Aa , i.e., they are defined as
Fa1 ,...,an (1 , . . . , n ) = h0|(0)|Aa1 (1 ) . . . Aan (n )i.
(69)
The important point is that these form factors can be exactly computed in the integrable
models once the S-matrix is known.
It is natural to organize the above expansion setting a reference value mr and keeping
in the spectral sum only the states with a mass smaller than mr . Looking at Eq. (68) we
see that we must expect, as a consequence of this truncation, a systematic error in the
approximation of the order O(emr x ). Up to mr = 2m1 (where m1 is the fundamental mass
of the model) only single particle states survive in the sum and Eq. (68) greatly simplifies.
In particular looking at Eq. (66) we see that only the first three states (which are the only
ones below the pair production threshold) survive. Eq. (68) becomes in this case, (choosing
for instance the spinspin correlator)
!
3
X
|Fi |2
2
K0 (mi r) ,
(70)
G (r) M (hl ) 1 +
i=1
where K0 (x) is the modified Bessel function and Fi denotes the overlap (measured in units
of the magnetization) of the ith state with the operator. M(hl ) denotes the magnetization
and its hl dependence is given in (12), (31).
Similarly the spinenergy and energyenergy correlators are given by
!
3
X
Fi Fi
K0 (mi r) ,
(71)
kG (r) M(hl )E(hl ) 1 +
i=1
!
3
X
|Fi |2
2
K0 (mi r) .
(72)
G (r) E (hl ) 1 +
i=1
Fi
Fi
and
have been evaluated in [33,34]. Their value is reported in
The constants
Tables 4 and 5.
One can systematically improve the approximation by setting higher values of mr . For
instance, for mr = 3 one must keep into account the first five single particle form factors
together with the two-particle ones involving the (1,1), (1,2) and (1,3) pairs, and so on. By
using the results of [33,34] also these multiparticle form factors can be evaluated exactly.
As discussed above, in Eqs. (70), (71) and (72) we expect systematic errors of the order
O(e2m1 r ). For distances larger than the correlation length these deviations are very small
but become increasingly relevant as the correlation length is approached. It would be
important to have an estimate of the magnitude of these corrections. From this point of
Table 4
Overlap amplitudes for the spin operator
F1
F2
F3
F4
F5
F6
F7
F8
=0.64090211
= 0.33867436
=0.18662854
= 0.14277176
= 0.06032607
=0.04338937
= 0.01642569
=0.00303607
651
Table 5
Overlap amplitudes for the energy operator
F1 =3.70658437
F2 = 3.42228876
F3 =2.38433446
F4 = 2.26840624
F5 = 1.21338371
F6 =0.96176431
F7 = 0.45230320
F8 =0.10584899
view the present case is a perfect laboratory since we know that all other possible sources
of systematic errors are under control. We report an example of the results obtained using
Eqs. (70), (71) and (72) (for the value hl = 4.4069 104 ) in the last column of Tables 7,
8 and 9. In order to gain further insight on the convergence of the approximation we have
also evaluated the contribution of the first eight terms of the spectral series (i.e., with mr =
3m1 ). We report in Fig. 10 the result of this analysis, together with those with mr = 2m1
and mr = m1 for comparison, in the particular case of the correlator.
6.2. IRS approach
By using the results of Sections 3 and 4 and the exact knowledge of the constants R and
i
in lattice units. They are reported in Table 6. 7
R one can easily write the constants B
We report an example of our results (obtained by plugging the values of Table 6 in
Eqs. (46), (47) and (48) in the particular case hl = 4.4069 104 ), in the second column
of Tables 7, 8 and 9.
6.3. Comparison with MC results
6.3.1. Lattice artifacts
It is interesting to see that lattice artifacts are confined to a remarkably small region of
few lattice spacings, which shows a negligible dependence on the magnetic field or on the
type of correlator. Since the lattice artifacts decrease so quickly it is rather easy to find
where does the region of applicability of the IRS results starts, by simply looking at the
distance Lmin where for the first time the IRS prediction becomes compatible (within the
errors) with the MC data or, if this never happens, looking at the location of the minimum
difference between IRS predictions and the MC simulations. 8 It turns out that for all
7 In doing this conversion one must also take into account the h factor contained in t. To stress this fact we
introduce in analogy to hl a new scaling variable tl |hl | |r|15/8 . In the coefficients reported in Table 6 the
conversion from t to tl has already been taken into account, hence they refer to the IRS expansion in powers of tl
of the correlators.
8 We may expect that higher orders in the t expansion improve the large distance behaviour of the IRS results,
but they give a negligible contribution around Lmin where the discrepancy between Monte Carlo data and IRS
predictions is completely dominated by lattice artifacts.
652
Table 6
Coefficients of the IRS expansion in lattice units
B1 = 0.7034 . . .
B2 = 0.6414 . . .
B3 =0.3007 . . .
1 = 0.1013 . . .
B
2 = 0
B
3 = 3.4776 . . .
B
B1 = 0.1685 . . .
B2 = 0.7380 . . .
B3 =0.3712 . . .
correlators and hl values Lmin ranges between 7 and 9 lattice spacings. This tells us that,
at least in the case of the Ising model perturbed by a magnetic field, the IRS method has
a large window of applicability, which becomes larger and larger as the critical point is
approached. This is well exemplified by Figs. 6 and 7 where the difference between MC
data and IRS predictions is plotted, for the correlator in the short range region, first in
units of the lattice spacing and then in units of the correlation length. It would be interesting
to test if this behaviour also holds for other models or for different realizations of this one.
6.3.2. hl dependence
In the range Lmin < r < the agreement between IRS predictions and MC results is
always very good. In particular, it seems that the method reaches its better results in the
case of correlator. As expected, the IRS approximation becomes better and better as
we approach the critical point (see Figs. 3, 4 and 5). First because the range of validity
becomes larger and second because the systematic deviations due to the terms neglected
in the expansion, which are proportional to higher powers of hl become less important. In
particular for the correlator at hl = 1.1017 104 there is a wide range (more than
40 lattice spacings) in which the IRS prediction coincides with the MC results within the
errors (see Fig. 8).
6.3.3. IRS versus FF
Looking at Tables 7, 8, 9 and at the Figs. 15 we see that, as expected, the FF approach
performs better than the IRS one for distances larger than the correlation length and that
the opposite is true for distances smaller than . It is interesting to see that for distances of
the order of the correlation length the IRS and FF methods give comparable performances.
Some interesting informations on the systematic errors involved in the two approximations
can be extracted from the data:
(1) While the systematic errors in the IRS approach have a polynomial behaviour in the
distance r (which is contained in tl ), those of the FF approach have an exponential
behaviour. This is clearly visible in Fig. 2 where the deviations are plotted as a
function of r for the correlator G at hl = 4.4069 104 and in Figs. 35 where
653
Table 7
Comparison of the Monte Carlo estimates (second column), the IRS results (third column) and the
form factor (FF) results (fourth column) for the correlator G at hl = 4.4069 104 . In the first
column is reported the distance in lattice units
r
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
MC
IRS
FF
0.71643(2)
0.61138(3)
0.55862(4)
0.52628(5)
0.50401(6)
0.48757(6)
0.47487(7)
0.46475(7)
0.45651(7)
0.44968(8)
0.44393(8)
0.43904(8)
0.43484(9)
0.43120(9)
0.42803(9)
0.42526(10)
0.42281(10)
0.42065(10)
0.41874(10)
0.41702(10)
0.41549(11)
0.41412(11)
0.41290(11)
0.41179(11)
0.41079(11)
0.40988(12)
0.40907(12)
0.40832(12)
0.40765(12)
0.40703(12)
0.71371
0.60871
0.55764
0.52590
0.50385
0.48750
0.47483
0.46472
0.45647
0.44961
0.44383
0.43889
0.43463
0.43093
0.42768
0.42481
0.42226
0.41998
0.41792
0.41606
0.41436
0.41280
0.41137
0.41003
0.40879
0.40762
0.40652
0.40547
0.40447
0.40351
0.59385
0.54528
0.51757
0.49852
0.48425
0.47303
0.46393
0.45638
0.45002
0.44459
0.43991
0.43584
0.43229
0.42916
0.42639
0.42394
0.42175
0.41980
0.41805
0.41648
0.41506
0.41378
0.41263
0.41158
0.41064
0.40978
0.40899
0.40828
0.40763
0.40704
they are plotted for all the correlators and all the values of hl as functions of r/ .
This makes the IRS method still reasonably reliable even for distances twice the
correlation length.
(2) We may obtain a rough estimate of the magnitude of the systematic errors involved
in the IRS approach with the following argument. Since tl |hl ||r|15/8 , looking at
Eq. (65) we see that for distances of the order of the correlation length we have tl
0.06. Depending on the correlator chosen, we would expect deviations of the order
31/15
32/15
i
), O(tl
). Since in general the B
constants are of order unity,
O(tl2 ), O(tl
this amounts to an expected deviation for the F functions of order F 0.004.
This expectation is in good agreement with the values of F obtained by comparing
654
Table 8
r
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
MC
IRS
FF
0.104067(2)
0.029348(2)
0.012327(3)
0.006674(2)
0.004190(2)
0.002879(2)
0.002103(2)
0.001606(2)
0.001268(2)
0.001029(2)
0.000854(2)
0.000720(2)
0.000615(2)
0.000534(2)
0.000469(2)
0.000415(2)
0.000370(2)
0.000334(2)
0.000305(2)
0.000280(2)
0.000258(2)
0.000239(2)
0.000221(2)
0.000205(2)
0.000194(2)
0.000184(2)
0.000174(2)
0.000166(2)
0.000158(2)
0.000152(2)
0.101321
0.025332
0.011262
0.006340
0.004064
0.002829
0.002087
0.001608
0.001281
0.001049
0.000880
0.000753
0.000657
0.000582
0.000524
0.000478
0.000442
0.000414
0.000392
0.000375
0.000362
0.000353
0.000347
0.000344
0.000342
0.000343
0.000345
0.000349
0.000354
0.000360
0.002330
0.001735
0.001399
0.001169
0.000999
0.000867
0.000761
0.000674
0.000601
0.000540
0.000488
0.000443
0.000404
0.000371
0.000341
0.000315
0.000292
0.000272
0.000254
0.000238
0.000224
0.000212
0.000200
0.000190
0.000181
0.000173
0.000165
0.000159
0.000153
0.000147
the MC and IRS estimates at the distance r = . (See for instance the data reported
in Tables 7, 8 and 9. In using these data one must take into account the normalization
between F, and G, functions.)
(3) A similar analysis can be performed in the case of the FF method. In this case we
know that the systematic errors are of order O(emr r ). We shall further discuss them
in Section 6.3.5 below. In the large distance regime r > 1.5 the performances of
the FF approach are very good. For instance, in this region, for the lowest value
of h that we studied: hl = 1.1017 104 the FF predictions for the correlator
coincide with the MC results within the errors (see Fig. 5).
655
Table 9
r
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
MC
IRS
FF
0.104077(19)
0.053137(12)
0.034975(11)
0.026275(11)
0.021240(11)
0.017966(11)
0.015676(11)
0.013990(10)
0.012705(10)
0.011699(10)
0.010891(10)
0.010231(11)
0.009685(11)
0.009228(11)
0.008841(11)
0.008510(11)
0.008224(11)
0.007978(11)
0.007765(10)
0.007577(10)
0.007411(11)
0.007265(11)
0.007134(11)
0.007018(11)
0.006914(11)
0.006822(11)
0.006740(11)
0.006667(11)
0.006600(12)
0.006539(12)
0.101011
0.050882
0.034286
0.026062
0.021181
0.017967
0.015704
0.014032
0.012753
0.011748
0.010942
0.010282
0.009736
0.009277
0.008888
0.008555
0.008268
0.008018
0.007800
0.007608
0.007438
0.007287
0.007152
0.007031
0.006922
0.006823
0.006733
0.006650
0.006575
0.006505
0.025867
0.020745
0.017834
0.015840
0.014353
0.013190
0.012251
0.011476
0.010826
0.010274
0.009800
0.009391
0.009035
0.008723
0.008448
0.008206
0.007991
0.007800
0.007629
0.007476
0.007340
0.007217
0.007106
0.007006
0.006916
0.006834
0.006760
0.006693
0.006632
0.006577
6.3.4. Convergence of the IRS expansion

It is interesting to study the convergence properties of the IRS method, i.e., to see if the
agreement with the Monte Carlo data improves as higher terms are added in the expansion.
h i and h i.
In these two cases the agreement improves as new terms are added in the expansion.
This is clearly visible in Fig. 8 where we have plotted the difference between the
MC data for the correlator h i at hl = 1.1017 104 and the IRS results with one
(pluses), two (crosses) and three (diamonds) terms in the expansion. The analogous
plot for the h i correlator shows exactly the same behaviour.
hi.
656
Fig. 1. Comparison of the Monte Carlo estimates (crosses), the IRS results (asterisks) and the form
factor results (pluses) for the correlator G at hl = 4.4069 104 .
Fig. 2. Differences between MC estimates and IRS results (crosses) and between MC and form factor
results (pluses) for the correlator G at hl = 1.1017 104 . In this figure, and in all the following
ones the errors in the MC estimates are not reported since they are smaller than the symbol size.
657
Fig. 3. Differences between MC estimates and IRS results (crosses for hl = 4.4069 104 , dotted
squares for 2.2034 104 and circles for 1.1017 104 ) and between MC and form factor results
(pluses for hl = 4.4069 104 , diamonds for 2.2034 104 and filled squares for 1.1017 104 )
for the correlator G . Distances are measured in units of the correlation length.
Fig. 4. Same as Fig. 3, but for the G correlator. Notice the different scale on the y axis.
658
Fig. 5. Same as Fig. 3, but for the G correlator.
Fig. 6. Differences between MC estimates and IRS results for hl = 4.4069 104 (pluses),
hl = 2.2034 104 (crosses) and hl = 1.1017 104 (diamonds) for the correlator, in the
short distance region dominated by the lattice artifacts.
659
Fig. 7. Same as in Fig. 6, but with distances measured in units of the correlation length.
Fig. 8. Difference between the MC data for the h i correlator at hl = 1.1017 104 and the IRS
results with one (pluses), two (crosses) and three (diamonds) terms in the expansion.
660
Fig. 9. Difference between the MC data for the hi correlator at hl = 1.1017 104 and the IRS
results with one (pluses) and two (crosses) terms in the expansion.
Fig. 10. Difference between the MC data for the h i correlator at hl = 1.1017 104 and the FF
results with one (pluses), three (crosses) and eight (diamonds) terms in the spectral series.
661
In this case we find exactly the opposite behaviour. As it is shown in Fig. 9 the
new coefficient that we evaluated does not improve the agreement with the Monte
Carlo data which is, by the way, impressively good already with the simple zero order
contribution to the correlator. We see two possible reasons for this, rather unexpected,
behaviour.
(1) As noticed in Section 4, the next term in the perturbative expansion of the
correlator has an exponent 32/15, which is very near to the one that we evaluated.
We cannot evaluate this further contribution since it would require the knowledge
of the expectation value of the T T operator. In principle this term could well
compensate the deviation that we observe.
(2) This behaviour could be an indication of the bad convergence properties of the
IRS method. If this is the case it would be very interesting to test (in view of the
good behaviour of the other two correlators) if this is a peculiar feature of the
hi correlator or a feature of the expansion itself. This issue could be settled in
principle by looking to the higher perturbative terms in the expansions of the two
other correlators. We plan to address this point in a forthcoming paper.
6.3.5. Convergence of the FF expansion
In order to study the convergence of the FF approximation we have compared the Monte
Carlo data in the case of the h i correlator at hl = 1.1017 104 with the result of the FF
approximation truncated at mr = m1 , mr = 2m1 , mr = 3m1 , respectively. This corresponds
to take into account one, three and eight states, respectively, in the spectral sum. The results
of the comparison are reported in Fig. 10. Looking at this figure one may see that as higher
orders are added the approximation smoothly converges to the MC data. By comparing the
three approximations one may get a perception of the convergence rate of the method.
7. Concluding remarks
In this paper we have compared the predictions of the IRS and FF approximations for
the , and correlators with the results of a set of high precision MC simulations
of the 2d Ising model perturbed by a magnetic field. To this end we have extended the IRS
approach to second order derivatives of the structure constants. Our main results are:
Lattice artifacts are confined in a small region of few lattice spacings.
There is a wide region ranging from 79 lattice spacings to the correlation length
in which the MC data are in good agreement with the IRS results.
The agreement improves as the critical point is approached.
For distances smaller than the IRS gives a better approximation than the FF method,
while the opposite is true for distances larger than .
The IRS method can be extended in principle to any order in the derivatives of the
Wilson coefficients, by using the integration method of [29] and the technique of the
Mellin transform. However in the case that we studied, i.e., the hi correlator, the IRS
method turns out to show rather bad convergence properties. It remains an open problem to
662
understand if this is a limit of the method itself or if it is a peculiar feature of the correlator
that we have chosen.
It would be very interesting to extend this analysis to other models in this same
universality class. In particular one could study the model recently introduced in [35,36] for
which an exact bethe ansatz solution, out of the critical point exists. Another interesting
application of the method would be the study of the correlators in the case of the most
general perturbation of the Ising critical point (i.e., a mixed situation with both magnetic
and thermal perturbations). In this case the exact integrability is lost but the IRS method
is still valid and could give important informations on the behaviour of the correlators.
In particular it would allow us to compare our approximation with the interesting results,
directly obtained on the lattice, in [37].
Appendix
b 1 involves the calculation of the integral
The evaluation of the coefficient h2 C
Z
Z
2
e
f r
s
d2 z |z| |1 z| zn (1 z) m |z w| ,(73)
Z = d w |w| |1 w| w (1 w)
where n, m, r, s N and , , , e, f R.
It is useful to introduce the following theorem (see [8,29]). Let us consider an integral
of the form
Z
N
X
f (w)Q f (w ),
(74)
I = d2 w
,=1
N
where {f (w)}N
=1,N , {f (w )}=1,N are two sets of independent functions and Q is a
constant matrix. Let us assume that f (w ) e (f (w)) have the same monodromies, in
particular the two sets of functions f (w) and g (f (w )) must have the same branch
points {wk }m+1
k=0 such that
0 = |w0 | < |w1 | < < |wm | < |wm+1 | =
(75)
and they have to be analytic elsewhere.

If we assume now that the matrix Q is invariant under the monodromy group action
Q = Mkt QMk ,
k,
(76)
where Mk are the monodromy matrices of f and g related to the branch points wk , it
follows that we are able to express I in terms of one-dimensional integrals (see [8,29] for
more details)
m
t
i X (k)
I (1 Mk+1 )1 (1 Mk )1 Q I (k) ,
I=
2
k=1
where t is the transposition and
(77)
Z
I (k)
dw f (w),
I (k)
dw f(w ),
663
(78)
C k
Ck
where Ck (C k ) are counter-clockwise (clockwise) circumferences enclosing all the branch

points of modulus lower than wk , starting at wk+ (infinitesimally over the cut at wk ) and
ending at wk (infinitesimally under the cut at wk ).
Now we are able to evaluate both the z-plane and w-plane integrations of (73) using the
previous lemma.
First, to perform the z-plane integration, we pose
Z
(79)
Iz (w, w ) = d2 z |z| |1 z| zn (1 z) m |z w|
so the z-plane integration involves the following branch points
z0 = 0,
z1 = w,
z2 = 1,
z = .
Thus, the application of (77) gives

(1) (1) (1)
(2) (2) (2)
Iz (w, w ) = 2i I1 T12 I2 + 2i I1 T12 I 2 ,
(80)
(81)
where
(1)
=
T12
ei (ei 1)
(ei(+ ) 1)(ei 1)
(82)
and
ei(+ )(ei 1)
(ei(++ ) 1)(ei(+ ) 1)
are the only non vanishing entries of the matrices
t
T (1) = (1 M2 )1 (1 M1 )1 Q,
t
T (2) = (1 M )1 (1 M2 )1 Q.
(84)
This imply that we have to take in account only the integrals

(/2 + n + 1) ( /2 + 1)
I1(1) = ei 1 w1+/2+ /2+n
(/2 + /2 + 2 + n)

F /2, /2 + n + 1; /2 + /2 + 2 + n; w ,

(/2 + 1) ( /2 + 1)
(1)
I 2 = ei 1 w 1+/2+ /2
(/2 + /2 + 2)

F m /2, /2 + 1; /2 + /2 + 2; w
(85)
(2)
T12 =
(83)
and

(/2 /2 /2 n 1) (/2 + 1)
I1(2) = ei(++ ) 1 ()m
(/2 /2 n)

F /2, /2 /2 /2 n 1; /2 /2 n; w ,
(/2 /2 /2 m 1) (/2 + 1 + m)
I 2(2) = ei(++ ) 1
(/2 /2)

F /2, /2 /2 /2 m 1; /2 /2; w .
(86)
664
Finally, putting all these relations in (81), we can recover the wanted result for Iz (w, w ).
The w-plane integration is very similar to the previous one. Now we have to evaluate
Z
(87)
Z = d2 w |w |e |1 w|f w r (1 w)s Iz (w, w ),
which involves w0 = 0, w1 = 1, w = as branch points. Hence the solution is given by
(77), i.e.,

(88)
Z = 2i I (1) T (1) I (1) ,
where
t
T (1) = (1 M )1 (1 M0 )1 Q.
(89)
The contribution coming from I (1) is

(1)
I1 = eie 1
Z1

dz f1 = eie 1 J1 ,
0
(1)
I2
= e
i(e++ )
Z1

dz f2 = ei(e++ ) 1 J2 ,
0
(1)
I 1 = eie 1
Z1

dz f1 = eie 1 J1 ,
0
(1)
I 2
= e
i(e++ )
Z1

dz f2 = ei(e++ ) 1 J2 ,
(90)
that, in terms of generalized hypergeometric functions, becomes

J1 = B /2 /2 /2 n 1, /2 + 1 B 1 + e/2, 1 + f/2 + s

e
e+f

n 1, , 1 + ;
n, 2 +
+ s; 1 ,
3 F2
2
2
2
2
2

J2 = B 1 + /2, 1 + /2 + n B 2 + e/2 + /2 + /2 + n, 1 + f/2 + s

+
+ n + e/2, 1 + + n; 2 +
+ n, 3 +
3 F2 , 2 +
2
2
2
2

+ +e+f
+ n + s; 1 ,
+
2

J1 = B /2 /2 /2 m 1, /2 + 1 + m B 1 + e/2 + k, 1 + f/2

m 1,
, 1 + e/2 + k;
,2
3 F2
2
2
2

e+f
+ k; 1 ,
+
2

J2 = B 1 + /2, 1 + /2 B 2 + e/2 + /2 + /2 + k, 1 + f/2
665
+ +e
m, 2 +
+ k, 1 + ; 2 +
,3
2
2
2
2

+ +e+f
+ k; 1 .
+
2
3 F2
(91)
Thus the solution has the form

Z = t11 J1 J1 + t12 J1 J2 + t21 J2 J1 + t22 J2 J2 ,
(92)
where the matrix elements tij are the following

t11 = 11 S(e/2)S(/2)S ( + + )/2

S(/2)S(/2)S(f/2)S ( + + e + f + 2 )/2

+ S( /2)S ( + + )/2 S ( + e + f + )/2 S (f + + )/2 ,

t22 = 11 S(/2)S( /2)S ( + e + )/2

S(/2)S(/2)S (e + f )/2 S 1/2( + f + )

+ S( /2)S ( + + )/2 S(f/2)S (e + f + + )/2) ,
t12 = t12 = 11 S(/2)S(/2)S(e/2)S( /2)

S ( + e + )/2 S ( + + )/2 S ( + )/2 ,
with

1 = S ( + )/2

S(/2)S(/2)S (e + f )/2 S ( + + e + f + 2 )/2

+ S( /2)S ( + + )/2 S ( + e + f + )/2 S (e + f + + )/2
and S(x) = sin(x).
For all details on the calculation we refer to [39].
Acknowledgements
We thank F. Gliozzi, R. Guida, M. Hasenbusch and R. Tateo for helpful discussions
and for a critical reading of the paper. This work was partially supported by the European
Commission TMR programme ERBFMRX-CT96-0045.
References
[2] A.B. Zamolodchikov, Al.B. Zamolodchikov, Ann. Phys. 120 (1979) 253.
[3] G. Mussardo, Phys. Rep. 218 (1992) 215.
666
[4] A.B. Zamolodchikov, Adv. Stud. Pure Math. 19 (1989) 641; Int. J. Mod. Phys. A 3 (1988) 743.
[5] F.A. Smirnov, Form Factors in Completely Integrable Models of Quantum Field Theory, World
Scientific, 1992.
[6] J.L. Cardy, G. Mussardo, Nucl. Phys. B [FS] 410 (1993) 451.
[7] R. Guida, N. Magnoli, Nucl. Phys. B 471 (1996) 361.
[8] R. Guida, N. Magnoli, Nucl. Phys. B 483 (1997) 563.
[9] H. Saleur, C. Itzykson, J. Stat. Phys. 48 (1987) 449.
[10] A. Cappelli, J.I. Latorre, Nucl. Phys. B 340 (1990) 659.
[11] A.W.W. Ludwig, J.L. Cardy, Nucl. Phys. B [FS19] 285 (1987) 687718.
[12] Vl.S. Dotsenko, Nucl. Phys. B 314 (1989) 687.
[13] Al.B. Zamolodchikov, Nucl. Phys. B 348 (1991) 619.
[14] H. Sonoda, Nucl. Phys. B 383 (1992) 172; Nucl. Phys. B 394 (1993) 302.
[15] B. Mikhak, A.M. Zarkesh, Nucl. Phys. B [FS] 430 (1994) 656.
[16] K.G. Wilson, Phys. Rev. 179 (1969) 1499.
[17] F.J. Wegner, J. Phys. A 8 (1975) 710.
[18] E. Brzin, J.C. Le Guillou, J. Zinn-Justin, Phys. Rev. Lett. 32 (1974) 473.
[19] E. Brzin, J.C. Le Guillou, J. Zinn-Justin, in: C. Domb, M.S. Green (Eds.), Phase transitions
and critical phenomena, Vol. 6, Academic Press, London, 1976.
[20] M.C. Bergre, F. David, Ann. Phys. 142 (1982) 416.
[21] V.P. Yurov, Al.B. Zamolodchikov, Int. J. Mod. Phys. A 5 (1990) 3221.
[22] R. Guida, N. Magnoli, Phys. Lett. B 411 (1997) 127.
[23] B.M. McCoy, T.T. Wu, The Two-dimensional Ising Model, Harvard University Press, Cambridge, 1973.
[24] T.T. Wu, Phys. Rev. 149 (1966) 380.
[25] V. Fateev, Phys. Lett. B 324 (1994) 45.
[26] V. Fateev, S. Lukyanov, A. Zamolodchikov, Al. Zamolodchikov, Nucl. Phys. B 516 (1998) 652.
[27] Smilga, Phys. Rev. D 55 (1997) 443.
[28] R. Hecht, Phys. Rev. 158 (1967) 557.
[29] S.D. Mathur, Nucl. Phys. B 369 (1992) 433.
[30] M. Caselle, M. Hasenbusch, Nucl. Phys. B 579 (2000) 667, next paper in this issue.
[31] P.G. Lauwers, V. Rittenberg, Phys. Lett. B 233 (1989) 197, and preprint Bonn University
BONN-HE-89-11 (unpublished).
[32] C. Destri, F. Di Renzo, E. Onofri, O. Rossi, G.P. Tecchiolli, Phys. Lett. B 278 (1992) 311.
[33] G. Delfino, G. Mussardo, Nucl. Phys. B 455 (1995) 724.
[34] G. Delfino, P. Simonetti, Phys. Lett. B 383 (1996) 450.
[35] V. Bazhanov, B. Nienhuis, S.O. Warnaar, Phys. Lett. B 322 (1994) 198.
[36] U. Grimm, B. Nienhuis, Phys. Rev. E 55 (1997) 5011.
[37] B.M. McCoy, T.T. Wu, Phys. Rev. D 18 (1978) 1259.
[38] B.M. McCoy, Mu-Lin Yan, Nucl. Phys. B [FS14] 257 (1985) 303339.
[39] P. Grinza, Tesi di Laurea, Torino University, 1998.

Critical amplitudes and mass spectrum of the 2d

Ising model in a magnetic field
M. Caselle a, , M. Hasenbusch b,1
a Dipartimento di Fisica Teorica dellUniversit di Torino, Istituto Nazionale di Fisica Nucleare, Sezione di
Torino, via P.Giuria 1, I-10125 Torino, Italy

b Humboldt Universitt zu Berlin, Institut fr Physik, Invalidenstr. 110, D-10099 Berlin, Germany
Received 9 December 1999; accepted 1 February 2000
Abstract
We compute the spectrum and several critical amplitudes of the two-dimensional Ising model in
a magnetic field with the transfer matrix method. The three lightest masses and their overlaps with
the spin and the energy operators are computed on lattices of a width up to L1 = 21. In extracting
the continuum results we also take into account the corrections to scaling due to irrelevant operators.
In contrast with previous Monte Carlo simulations our final results are in perfect agreement with
the predictions of S-matrix and conformal field theory. We also obtain the amplitudes of some of
the subleading corrections, for which no S-matrix prediction has yet been obtained. 2000 Elsevier
1. Introduction
In these last years there has been much progress in the study of 2d spin models in the
neighbourhood of critical points. The scaling limit of such models is described in general
by the action A obtained by perturbing the conformal field theory (CFT) which describes
the critical point with one (or more) of the relevant operators which appear in the spectrum
of the CFT:
Z
(1)
A = A0 + d 2 x (x),
where A0 is the action of the CFT at the critical point [1] and (x) is the perturbing
operator. Few years ago A. Zamolodchikov in a seminal paper [2] suggested that in some
special cases these perturbed theories are equivalent to relatively simple quantum field
theories [3] whose mass spectrum and S-matrix are explicitly known. Later it was realized
caselle@to.infn.it
1 hasenbus@physik.hu-berlin.de
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 0 7 4 - 2
668
M. Caselle, M. Hasenbusch / Nuclear Physics B 579 [FS] (2000) 667703
that these theories had a deep connection with the Dynkin diagrams of suitable Lie algebras
and, from the exact knowledge of the S-matrix, several other informations, and in particular
some critical amplitudes were obtained (for a review, see for instance [4]). While these
results have formally the status of conjectures, they successfully passed in these last years
so many tests that they are now universally accepted. The most fascinating example of
these S-matrix models is the Ising model perturbed by an external magnetic field, which
is also the model which was originally studied by Zamolodchikov in [2]. This model is
highly non-trivial. Its spectrum contains 8 stable scalar particles, all with different masses.
Both the masses and the entries of the S-matrix are based of the numerology of the E8
exceptional Lie algebra. In particular the ratio between the first two masses is predicted
to be the golden ratio m2 /m1 = 2 cos( 5 ). The simplest realization of this QFT is the
2d Ising model at = c in presence of an external magnetic field h. However there
are several other models which belong to the same universality class. In particular, the
first numerical check of the predictions of [2] was performed on the Ising quantum spin
chain [5] in which the first few states of the spectrum were precisely observed. Another
interesting realization was presented, in [68] where the dilute A3 IRF (Interaction Round
a Face) model was solved exactly and the predicted spectrum of states was found [8].
Despite these successes, little progress has been achieved in testing Zamolodchikovs
proposal directly in the 2d Ising spin model. Even more, it is exactly for this model that
one faces the only existing discrepancy between Zamolodchikovs results and Monte Carlo
simulations.
The Ising spin model in a magnetic field was studied numerically in [9,10]. In both
papers the authors studied the spinspin correlator and did not find the spectrum predicted
by [2]. Their data were compatible with the presence in the spectrum of only the lowest
mass state. The explanation suggested in [9,10] was that probably the higher masses had
a negligible overlap amplitude with the spin operator. However, later, in [11] these overlaps
were evaluated explicitly in the S-matrix framework and turned out to be of the same order
of magnitude as the overlap with the lowest mass state.
In this paper we shall address this problem. We shall show that Zamolodchikovs
proposal (and the calculations of [11]) is correct also in the case of the 2d Ising spin
model and that the apparent disagreement was due to the fact that it is very difficult to
extract a complex spectrum from a multi-exponential fit to the spinspin correlator. We
have been prompted to this explanation by another example that we recently studied, in
which exactly the same phenomenon happens: the 3d Ising model [12]. In this case also,
a multi-exponential fit to the spinspin correlator seems to indicate the presence of a single
state in the spectrum, while, using a suitable variational method and diagonalizing a set of
improved operators one can clearly see the rich spectrum of the model.
While in previous numerical works [9,10] the model was studied by using Monte
Carlo simulations we tried in the present paper a different approach based on the exact
diagonalization of the transfer matrix.
This approach has various advantages: it gives direct access to the mass spectrum of
the model and allows to obtain numerical estimates of various quantities with impressively
small uncertainties. However it has the serious drawback that only transfer matrices of
669
limited size can be handled and it is difficult to extract from them the continuum limit
results in which we are interested. During the last years various strategies have been
elaborated to attack this problem, but all of them are affected by systematic errors whose
size is in general unknown.
In this paper we propose a new approach based on the fact that, by using the exact
solution of the Ising model at the critical point, one can construct very precise expansions
for the scaling functions in powers of the perturbing field. More precisely, thanks to the
knowledge of the spectrum of the model, it is possible to list all the irrelevant fields which
may appear in the effective Hamiltonian and select them on the basis of the symmetry
properties of the observables under study.
Our strategy could be summarized as follows.
Choose a set of values of h for which the correlation length is much smaller than the
maximum lattice size that we can study 2 . Diagonalize the transfer matrix for various
values of the transverse size of the lattice and extract all the observables of interest.
Extrapolate the numbers thus obtained to the thermodynamic limit. Thanks to the very
small correlation length, the finite size behaviour is dominated by a rapidly decreasing
exponential and the thermodynamic limit can be reached with very small uncertainties
(we list in a set of tables at the end of the paper the results that we obtained in this
way).
Construct for each observable the scaling function keeping the first 7 or 8 terms in the
expansion in powers of the perturbing field.
Fit the data with these truncated scaling functions. By varying the the number of
input data and of subleading terms used in the scaling functions we may then obtain
a reliable estimate of the systematic deviations involved in our estimates (see the
discussion in Secttion 6).
Which are the observables of interest mentioned above?
Usually, when looking at the scaling regime of statistical models one can study only
adimensional amplitude ratios which are the only quantities which, thanks to universality,
do not depend on the details of the lattice models, but only on the features of the underlying
QFT. However the Ising model can be solved exactly at the critical point also on the lattice
and explicit expressions for the spinspin and energyenergy correlators are known. This
allows to write the explicit expression in lattice units of the amplitudes evaluated in the
framework of the S-matrix theory. Thus one is able to predict not only adimensional
amplitude ratios but also the values of the critical amplitudes themselves. This greatly
enhances the predictive power of the S-matrix theory and makes much more stringent the
numerical test that we perform.
The final result of our analysis is that all the observables that we can measure perfectly
agree with the S-matrix predictions.
In particular we obtain very precise estimates for the first three masses, for several
critical amplitudes and, what is more important, for the overlap amplitude of the first two
2 In particular we decided to keep the ratio /L < 0.1. This means that we only studied values of h for which
0
the correlation length was smaller than two lattice spacings.
670
masses with the spin and energy operators, a result which had never been obtained before.
We also measure the amplitude of some of the subleading corrections in the scaling
functions, for which no S-matrix prediction exists for the moment. In particular we found
that the amplitude of the corrections due to the energy momentum tensor in translationally
invariant observables is compatible with zero.
This paper is organized as follows. In Section 2 we introduce the model in which we are
interested, collect some known results from S-matrix theory and finally give the translation
in lattice units of the critical amplitudes evaluated in the S-matrix framework. In Section 3
we construct the scaling functions. This only requires the use of very simple and well
known results of Conformal Field Theory. Notwithstanding this it turns out to be a rather
non trivial exercise. Since it could be a result of general utility (it could be extended to
other models for which the CFT solution is known or to other quantities of the Ising model
in a magnetic field that we have not studied in the present paper) instead of simply giving
the results, we derived the scaling functions explicitly and tried to give as much details as
possible. Section 4 is devoted to a description of the transfer matrix method. In Section 5
we deal with the thermodynamic limit while in Section 6 we analyze the transfer matrix
results and give our best estimates for the critical amplitudes in which we are interested.
Section 7 is devoted to some concluding remarks.
To help the reader to reproduce our analysis (or to follow some alternative fitting
procedure) we list in four tables at the end of the paper the data that we obtained with
the transfer matrix approach.
2. Ising model in a magnetic field

In this section we shall review the existing theoretical informations on the Ising model
in a magnetic field. This will require four steps. First (in Section 2.1) we shall define the
lattice version of the model, discuss its action and define the observables in which we shall
be interested in the following. Then (in Section 2.2) we shall turn to the continuum version
of the theory, described in the present case by the action:
Z
(2)
A = A0 + h d 2 x (x),
where (x) is the perturbing operator. In particular we shall discuss, within the framework
of the renormalization group, the expected scaling behaviour of the various quantities of
interest and define the corresponding critical amplitudes. In Section 2.3 we shall use the
knowledge of the S-matrix of the model to obtain the value of some of the amplitudes of
interest by using the Thermodynamic Bethe Ansatz (TBA) and the form factor approach.
Finally in Section 2.4 we shall turn back to the lattice model and show how the continuum
results can be translated in lattice units.
2.1. The lattice model
The lattice version of the Ising model in a magnetic field is defined by the partition
function
Z=
e(
hn,mi n m +H
n n )
671
(3)
i =1
where the field variable n takes the values {1}; n (n0 , n1 ) labels the sites of a square
lattice of size L0 and L1 in the two directions and hn, mi denotes nearest neighbour sites on
the lattice. In our calculations with the transfer matrix method we shall treat asymmetrically
the two directions. We shall denote n0 as the time coordinate and n1 as the space one. The
number of sites of the lattice will be denoted by N L0 L1 . In the thermodynamic limit
both L0 and L1 must go to infinity and only in this limit we may recover the results of the
continuum theory. In our actual calculations with the transfer matrix method we shall study
finite values of L1 and then extrapolate the results to infinity. This extrapolation induces
systematic errors which are the main source of uncertainty of our results, since the rounding
errors in the transfer matrix diagonalization are essentially negligible. In Section 6 below,
we shall discuss these systematic errors and estimate their magnitude.
In order to select only the magnetic perturbation, the coupling must be fixed to its
critical value
1
= c = log ( 2 + 1) = 0.4406868 . . .
2
by defining hl = c H we end up with
P
P
X
ec hn,mi n m +hl n n .
(4)
Z(hl ) =
i =1
hl denotes the lattice discretization of the magnetic field h which appears in the continuum
action Eq. (2). It must be, for symmetry reasons, an odd function of h.
2.1.1. Lattice operators
It is useful to define the lattice analogous of the spin and energy operators of the
continuum theory. They will correspond to linear combinations of the relevant and
irrelevant operators of the continuum theory with suitable symmetry properties with respect
to the Z2 symmetry of the model (odd for the spin operator and even for the energy one).
Near the critical point this linear combination will be dominated by the relevant operator
and the only remaining freedom will be a conversion constant relating the continuum and
lattice versions of the two operators (we shall find this constants in Section 2.5). The
simplest choices for these lattice analogous are:
Spin operator
l (x) x ,
(5)
i.e., the operator which associates to each site of the lattice the value of the spin at
that site.
Energy operator
X

1
y b ,
(6)
l (x) x
4
y n.n. x
where the sum runs over the four nearest neighbour sites y of x. b represents a
constant bulk term which we shall discuss below.
672
The index l indicates that these are the lattice discretizations of the continuous operators.
We shall denote in the following the normalized sum over all the sites of these operators
simply as
1 X
1 X
l (x), l
l (x).
(7)
l
N x
N x
2.1.2. Observables
Free Energy
The free energy is defined as

1
(8)
f (hl ) log Z(hl ) .
N
It is important to stress that f (hl ) is composed by a bulk term fb (hl ) which is
an analytic even function of hl and by a singular part fs (hl ) which contains the
relevant informations on the theory as the critical point is approached. The continuum
theory can give informations only on fs . The value of fb (0) can be obtained from the
exact solution of the lattice model at hl = 0, = c (see [13])
2G 1
+ log 2 = 0.9296953982 . . .,
(9)
fb =
2
where G is the Catalan constant.
Magnetization
The magnetization per site M(hl ) is defined as

1
1 X
i .
(10)
log Z(hl ) =
M(hl )
N hl
N
i
Hence we have
M(hl ) = hl i.
(11)
Magnetic Susceptibility
The magnetic susceptibility is defined as
M(hl )
.
hl
Internal Energy
b l ) as
We define the internal energy density E(h
X

b l) 1
n m .
E(h
2N
(hl )
(12)
(13)
hn,mi
As for the free energy, also in this case one has a bulk analytic contribution Eb (hl )
which is an even function of hl . Let us define b Eb (0). The value of Eb (0) can be
easily evaluated (for instance by using KramersWannier duality) to be b = 1 . Let
2
b l ) b , we have
us define E(hl ) E(h

1 X
1
(14)
n m .
E(hl ) =
2N
2
hn,mi
673
Hence we have
E(hl ) = hl i.
(15)
As usual the internal energy can also be obtained by deriving the free energy with
respect to . However it is important to stress that, due to the magnetic perturbation
(see Eq. (3)) in performing the derivative we also extract from the Boltzmann factor a
term proportional to H l . Hence we have:

b l ) = 1 log Z(hl ) hl l .
E(h
2N
2c
(16)
This observation will play an important role in the following.

2.1.3. Correlators
We are interested in the spinspin and in the energyenergy connected correlators
defined as

(17)
G, (r) l (0)l (r) hl i2 l (0)l (r) c ,

(18)
G, (r) l (0)l (r) hl i2 l (0)l (r) c .
For a nonzero magnetic field these correlators are very complicated, unknown, functions
of h and r, however a good approximation in the large distance regime r is 3

G (r) X |Fi (h)|2
K0 mi (h)r ,
=
2
hl i
(19)
where the sum is over the low laying single particle states of the spectrum, mi (h) denotes
their mass the functions Fi (h) their overlap with the operator. Similarly we have

G (r) X |Fi (h)|2
K0 mi (h)r ,
=
2
hl i
(20)
where the spectrum is the same as for the spinspin correlator but the overlap constants are
different.
A particular role is played by the lowest mass m1 which gives the dominant contribution
in the large distance regime. Its inverse corresponds to the (exponential) correlation length
of the model and sets the scale for all dimensional quantities in the model. In particular
the large distance regime mentioned few lines above means large with respect to the
correlation length.
In the approximation of Eqs. (19) and (20) one is neglecting the cut-type contributions
which appear above the two-particle threshold, i.e., at twice the value of m1 . For this reason
we shall concentrate in the following only on the three first states of the spectrum which
are the only ones which lie below such threshold (see Eq. (40) below).
3 For a discussion of the limits of this approximation and of the corrections which must be taken into account
when the short distance regime is approached see [14].
674
2.1.4. Time slice correlators

It is very useful to study the zero momentum projections of the above defined correlators.
They are commonly named time slice correlators. The magnetization of a time slice is given
by
1 X
(n0 ,n1 ) .
(21)
Sn0
L1 n
1
The time slice correlation function is then defined as

X

hSn0 Sn0 + i hSn0 i2 ,
G0 ( )
(22)
n0
where the index 0 indicates that this is the zero momentum projection of the original
correlator. Starting from Eq. (19) it is easy to show that in the large limit G0 ( ) behaves
as
G0 ( ) X |Fi (h)|2 mi (h)| |
=
e
.
(23)
hl i2
mi (h)L1
i
A similar result, with the obvious modifications, holds also for G0, .
2.2. Critical behaviour
In this section we discuss the critical behaviour of the model by using standard
renormalization group methods, keeping in the expansions only the first order in the
perturbing field. Both the results and the analysis are well known and can be found in
any textbook. We report it here since it will serve us as a starting point for the more refined
analysis which we shall perform in Section 3 below.
2.2.1. Critical indices
The starting point of the renormalization group analysis is the singular part of the
free energy fs (t, h) (where t is the reduced temperature). Standard renormalization group
arguments (see for instance [15]) allow to write fs in terms of a suitable scaling function
:

uh d/yh
ut /ut0

fs (t, h) =
,
(24)
u
|u /u |yt /yh
h0
h0
where ut0 and uh0 are reference scales that depend on the model. uh and ut denote the
scaling variables associated to the magnetic and energy operators respectively and yh , yt
are their RG-exponents. ut and uh do not exactly coincide with t and h but are instead
analytic functions of them. The only constraint is that they must respect the Z2 parity of
t and h. Near the critical point we may suitably rescale so as to identify ut = t and
uh = h. Thus, setting t = 0 we immediately obtain the asymptotic critical behaviour of fs
fs |h|d/yh .
(25)
Taking the derivative with respect to h (or t) and then setting t = 0 we can obtain from
Eq. (24) also the asymptotic critical behaviour of the other observables in which we are
interested
M |h|d/yh 1 ,
675
(26)
|h|
d/yh 2
E |h|
(dyt )/yh
(27)
.
(28)
From the exact solution of the Ising model at the critical point we know that yh =
and yt = 1. Inserting these values in the above expressions we find
16
fs |h| 15 ,
(29)
1
15
M |h| ,
|h|
14
15
15
8
(30)
,
(31)
8
15
E |h| .
(32)
The masses mi have as scaling exponent, as usual, 1/yh , hence

8
mi |h| 15 .
(33)
Finally from the definitions of Eqs. (19) and (20) we see that the overlap amplitudes
behave as adimensional constants.
2.2.2. Critical amplitudes
In order to describe the scaling behaviour of the model we also need to know the
proportionality constants in the above scaling functions. These constants are usually called
critical amplitudes. Using the results collected in Eqs. (29)(32) we have the following
definitions:
Af lim f h 15 ,
AM lim Mh 15 ,
AE lim Eh 15 ,
Ami lim mi h 15 ,
16
h0
h0
AFi lim Fi ,
h0
h0
h0
AFi lim Fi .
14
A lim h 15 ,
h0
(34)
(35)
(36)
h0
Notice for completeness that in the literature (see for instance [16]) the amplitudes A ,
AM and Am1 are usually denoted as
A c ,
1
15
AM Dc
Am1
1
.
c
(37)
We shall show in the next section that all these amplitudes can be exactly evaluated in
the framework of the S-matrix approach. As a preliminary step let us notice that since M
and are obtained as derivatives of f we have
AM =
16
Af ,
15
A =
1
AM .
15
(38)
2.2.3. Universal amplitude ratios

From the above critical amplitudes one can construct universal combinations which do
not depend on the particular realization of the model. For this reason they have been widely
676
studied in the literature. In particular there are two classical amplitude combinations
which involve the critical amplitudes defined above (see for instance [16]). They are:
R Dc B 14 ,
Q2 ( /c )(c /0 ) 4
(39)
where we used the notations of Eq. (37). and 0 denote the critical amplitudes of the
susceptibility and exponential correlation length for h = 0 and a small positive reduced
temperature, while B denotes the critical amplitude of the magnetization for h = 0 and
a small negative reduced temperature. Notice however that, since (as we mentioned in
the introduction) we are able to give the explicit relation between lattice and continuum
expectation values, we are not constrained to study only universal combination but can
determine exactly the various critical amplitudes.
2.3. S-matrix results
In 1989 A. Zamolodchikov [2] suggested that the scaling limit of the Ising Model in a
magnetic field could be described by a a scattering theory which contains eight different
species of self-conjugated particles Aa , a = 1, . . . , 8 with masses
= (1.6180339887 . . .)m1 ,
5
= (1.9890437907 . . .)m1 ,
m3 = 2m1 cos
30
7
= (2.4048671724 . . .)m1 ,
m4 = 2m2 cos
30
2
= (2.9562952015 . . .)m1 ,
(40)
m5 = 2m2 cos
15
= (3.2183404585 . . .)m1 ,
m6 = 2m2 cos
30
7
= (3.8911568233 . . .)m1 ,
m7 = 4m2 cos cos
5
30
2
= (4.7833861168 . . .)m1 ,
m8 = 4m2 cos cos
5
15
where m1 (h) is the lowest mass of the theory. As mentioned above it coincides with the
inverse of the (exponential) correlation length. Few years later, from the knowledge of the
S-matrix of the theory V. Fateev [17] obtained explicit predictions for some of the critical
amplitudes defined above.
In order to evaluate the amplitudes one must first fix the normalization of the operators
involved which can be set, for instance, by fixing the constant in front of the long
distance behaviour of the correlators at the critical point. It is important to make explicit
this normalization choice, since it will allow us, by comparing with the corresponding
correlators in the lattice theory to convert explicitly the continuum results in lattice units.
Following the choice of [17] we assume:
m2 = 2m1 cos

1
,
(x) (0) =
1
|x| 4
|x| ,
(41)
Table 1
Critical amplitudes
Am1
Af
AM
A
AE
Table 2
Classical amplitude ratios
= 4.40490858 . . .
= 1.19773338 . . .
= 1.27758227 . . .
= 0.08517215 . . .
= 2.00314 . . .

1
(x)(0) = 2 ,
|x|
677
R = 6.77828502 . . .
Q2 = 3.23513834 . . .
|x| .
(42)
With these conventions one finds [17]:

Am1 = C,
Af =
(43)
C2
8(sin 2
3 + sin 5 + sin 15 )
(44)
where

4
5
4 sin 5 15 4 2 34 2 13
16
= 4.40490858 . . . .
C=

8
3
23 15
14 2 16
(45)
From Af one immediately obtains AM and A .

The amplitude AE requires a more complicated analysis. Its exact expression has been
obtained only recently in [18]
AE = 2.00314 . . . .
(46)
We summarize in Table 1 these S-matrix predictions for the critical amplitudes.

From these critical amplitudes, and using the values of , B and 0 one immediately
obtains the classical amplitude ratios defined above (see for instance [19]). They are
reported in Table 2.
Finally, the critical overlap amplitudes AFi and AFi were evaluated in [11,20]. They
are reported in Tables 3 and 4.
2.4. Conversion to lattice units
While the values listed in Tables 2, 3 and 4 are universal, the amplitudes listed in Table 1
depend on the details of the regularization scheme. Thus some further work is needed
to obtain their value on the lattice. We shall denote in the following the lattice critical
amplitudes with an index l. Thus, for instance,
1
15
AlM = lim hl ihl

hl 0
(47)
to be compared with the continuum critical amplitude defined in Eq. (34)

1
AM = lim h ih 15 .
h0
(48)
678
Table 3
Critical overlap amplitudes for the spin operator
AF
1
AF
2
AF
3
AF
4
AF
5
AF
6
AF
7
AF
8
= 0.64090211 . . .
= 0.33867436 . . .
= 0.18662854 . . .
= 0.14277176 . . .
= 0.06032607 . . .
= 0.04338937 . . .
= 0.01642569 . . .
= 0.00303607 . . .
Table 4
Critical overlap amplitudes
for the energy operator
AF
1
AF
2
AF
3
AF
4
AF
5
AF
6
AF
7
AF
8
= 3.70658437 . . .
= 3.42228876 . . .
= 2.38433446 . . .
= 2.26840624 . . .
= 1.21338371 . . .
= 0.96176431 . . .
= 0.45230320 . . .
= 0.10584899 . . .
In order to relate the lattice results with the continuum ones we must study the
relationship between the lattice operators and the continuum ones. In general the
lattice operators will be given by the most general combination of continuum operators
compatible with the symmetries of the lattice operator multiplied by the most general
analytic functions of t and h (with a parity which is again constrained by the symmetry of
the operators involved). Thus, for instance, anticipating the discussion that we shall make
in Section 3, we have
l = f0 (t, h) + fi (t, h)i ,
(49)
f0 (t, h)
and fi (t, h) are suitable functions of t and h and with the notation i we
where
denote all the other fields of the theory (both relevant and irrelevant) which respect the
symmetries of the lattice.
A similar relations also holds for the energy operator:
l = g0 (t, h) + gi (t, h)i .
(50)
Finally, also hl is related to the continuum magnetic field h by a relation of the type
hl = b0 (t, h)h,
(51)
where b0 (t, h) must be an even function of h.

At the first order in t and h these combinations greatly simplify and essentially reduce
to a different choice of normalization between the continuum operators and their lattice
analogous:
l R ,
l R ,
hl Rh h
(52)
where R , R and Rh are three constants which correspond to the h 0, t 0 limit of

the f0 , g0 and b0 functions.
If we want to compare the S-matrix results discussed in the previous section with our
lattice results we must fix these normalizations 4 . The simplest way to do this is to look at
the analogous of Eqs. (41, 42) at the critical point (namely for hl = 0) [21] .
4 This essentially amounts to measure all the quantities in units of the lattice spacing. For this reason we can
fix in the following the lattice spacing to 1 and neglect it.
679
In fact, if hl = 0 it is possible to obtain an explicit expression for the spinspin and

energyenergy correlators (for a comprehensive review see for instance [22,23]) directly
on the lattice, for any value of . Choosing in particular = c , and looking at the large
distance behaviour of these lattice correlators we may immediately fix the normalization
constants. Let us look first at R .
We know from [24] that:
hi j ih=0 =
R2
,
|rij |1/4
(53)
where rij denotes the distance on the lattice between the sites i and j and
0
R2 = e3 (1) 25/24 = 0.70338 . . . .
(54)

R = 0.83868 . . . .
(55)
From this we can also obtain the normalization of the lattice magnetic field which must
exactly compensate that of the spin operator in the perturbation term h . We find:
Rh = (R )1 = 1.1923 . . . .
(56)
Combining these two results we obtain the value in lattice units of the constant A
AlM = (R )16/15AM = 1.058 . . . .
(57)
From this one can easily obtain also Alf , Al and Am1 .
Let us look now at R . In the case of the energy operator the connected correlator on the
lattice, at hl = 0 and for any value of has the following expression [25]:
2

2
(58)
K1 (r) K02 (r) ,
l (0)l (r) c =
where K0 and K1 are modified Bessel functions, is a parameter related to the reduced
temperature, defined as
= 4| c |
(59)
and with the index c we denote the connected correlator (notice that thanks to the definition
(14) no disconnected part must be subtracted at the critical point and the index c becomes
redundant). This expression has a finite value in the 0 limit (namely at the critical
point). In fact the Bessel functions difference can be expanded in the small argument limit
as

2
1
+
(60)
K1 (r) K02 (r) =
(r)2
thus giving, exactly at the critical point:

l (0)l (r) =
1
.
(r)2
(61)
680

R =
(62)
and from this we obtain the expression in lattice units of A

Al = (R )8/15(R )A = 0.58051 . . . .
(63)
Our results are summarized in Table 5.

Table 5
Critical amplitudes in lattice units
Alm1 = 4.01039911 . . .
Alf = 0.99279949 . . .
AlM = 1.05898612 . . .
Al = 0.07059907 . . .
AlE = 0.58051 . . .
2.4.1. Alternative derivation of R

In this section we discuss, for completeness, an alternative derivation of R . It can be
used in those cases in which the correlators are not known, but the internal energy is known
on a finite size lattice at the critical point. Then R can be obtained by comparing the finite
size behaviour of the internal energy on the lattice with that predicted by conformal field
theory in the continuum. In the case of the Ising model, thanks to the beautiful work by
Ferdinand and Fisher [13], we know that on a square lattice of size L0 L1 with L0 > L1
with periodic boundary conditions the internal energy must scale as:
hl i =
1
2 ( )3 ( )4 ( )
,
2 ( ) + 3 ( ) + 4 ( ) L1
(64)
0
where i ( ) denotes the ith Jacobi theta function and i L
L1 .
The same behaviour can be studied in the continuum theory, by using CFT techniques.
The result [26] is
hi =
1
1 ( )0
.
2 ( ) + 3 ( ) + 4 ( ) L1
(65)
By using the relation

1 ( )0 = 2 ( )3 ( )4 ( )
(66)
which allows to express the derivative of the 1 ( ) in terms of ordinary theta functions
we see that the two Eqs. (64) and (65) agree only if we choose, as we did in the previous
section, R = 1 .
681
3. Scaling functions
In this section we shall construct the scaling functions for the various quantities in which
we are interested. Our aim is to give the form (i.e., the value of the scaling exponents) of the
first 78 terms of the expansion in powers of h of the scaling functions and at the same time
to identify the operators in the lattice Hamiltonian from which they originate. To this end
we shall first deal in Section 3.1 with the theory at the critical point. We shall in particular
discuss its spectrum, which can be constructed explicitly by using CFT techniques. Next,
in Section 3.2, we discuss in the framework of the renormalization group approach the
origin of the subleading terms in the scaling functions, and show how to obtain their
exponents from the knowledge of the renormalization group eigenvalues yi of the irrelevant
operators. While in general this analysis is only of limited interest since the yi of the
irrelevant operators are unknown, in the present case, thanks to the CFT solution discussed
in Section 3.1, it becomes highly predictive and will allow us to explicitly construct in
Sections 3.3 and 3.4 the scaling functions. In particular in Section 3.3 we shall list all
the irrelevant operators which may appear in the effective Hamiltonian and discuss their
symmetry properties, while in Section 3.4 we shall write the scaling functions and identify
the operators involved in the various scaling terms.
3.1. The Ising model at the critical point
The Ising model at the critical point is described by the unitary minimal model with
central charge c = 1/2 [1]. Its spectrum can be divided into three conformal families
characterized by different transformation properties under the dual and Z2 symmetries
of the model. They are the identity, spin and energy families and are commonly denoted as
[I ], [ ], []. Let us discuss their features in detail.
Primary fields
Each family contains a relevant operator which is called primary field (and gives the
name to the entire family). Their conformal weights are hI = 0, h = 1/16 and h =
1/2 respectively. The relationship between conformal weights and renormalization
group eigenvalues is: y = 2 2h. Hence the relevant operators must have h < 1.
Secondary fields
All the remaining operators of the three families (which are called secondary fields)
are generated from the primary ones by applying the generators Li and L i of the
Virasoro algebra. In the following we shall denote the most general irrelevant field in
the [ ] family (which are odd with respect to the Z2 symmetry) with the notation i
and the most general fields belonging to the energy [] or to the identity [I ] families
(which are Z2 even) with i and i respectively. It can be shown that by applying a
generator of index k: Lk or L k to a field (where = I, , depending on the
case) of conformal weight h we obtain a new operator of weight h = h + k. In
general any combination of Li and L i generators is allowed, and the conformal
weight of the resulting operator will be shifted by the sum of the indices of the
generators used to create it. If we denote by n the sum of the indices of the generators
682
of type Li and with n the sum of those of type L i the conformal weight of the
The corresponding RG eigenvalue will be y =
resulting operator will be h + n + n.
hence all the secondary fields are irrelevant operators.
2 2h n n,
Nonzero spin states
The secondary fields may have a non zero spin, which is given by the difference
n n.
In general one is interested in scalar quantities and hence in the subset of those
irrelevant fields which have n = n.
However on a square lattice the rotational group
is broken to the finite subgroup C4 (cyclic group of order four). Accordingly, only
spin 0, 1, 2, 3 are allowed on the lattice. If an operator of the continuum theory
has spin j N, then its lattice discretization l behaves as a spin j (mod 4) operator
with respect to the C4 subgroup. As a consequence all the operators which in the
continuum limit have spin j = 4N with N non-negative integer can appear in the
lattice discretization of a scalar field. This will play a major role in the following.
Null vectors
Some of the secondary fields disappear from the spectrum due to the null vector
conditions. This happens in particular for one of the two states at level 2 in the
and families and for the unique state at level 1 in the identity family. From each
null state one can generate, by applying the Virasoro operators a whole family of null
states hence at level 2 in the identity family there is only one surviving secondary
field, which can be identified with the stress energy tensor.
Secondary fields generated by L1
Among all the secondary fields a particular role is played by those generated by the
L1 Virasoro generator. L1 is the generator of translations on the lattice and as a
consequence it has zero eigenvalue on translational invariant observables. Another
way to state this results is to notice that L1 can be represented as a total derivative,
and as such it gives zero if applied to an operator which can be obtained as the integral
over the whole lattice of a suitable density (i.e., a translationally invariant operator).
3.2. RG analysis for h 6= 0
We shall discuss the higher order corrections to the RG analysis of Section 2.2 along the
lines of [27], to which we refer for a more detailed discussion. The only improvement that
we make with respect to [27] is in the part devoted to the contribution due to the irrelevant
operators, in which we shall make use of the results discussed in the previous section.
We expect three types of corrections to the asymptotic results reported in Section 2.2:
(a) Analytic corrections
They are due to the fact, already mentioned in Section 2.2, that the actual scaling
variables in the RG approach are not hl and t but uh and ut which are in principle
the most general analytic functions of hl and t which respect the Z2 parity of hl
and t. Let us write the Taylor expansion for uh and ut , keeping only those first few
orders that are needed for our analysis (we use the notations of [27]).

(67)
uh = hl 1 + ch t + dh t 2 + eh h2l + O t 3 ,th2l ,

2
2
3
2
4
4 2 2
(68)
ut = t + bt hl + ct t + dt t + et thl + ft hl + O t ,t hl .
683
The corrections induced by the higher terms in uh and ut are of three types.
The first one is very simple to understand. It is due to the higher powers of
hl contained in uh which lead to corrections to the power behaviours listed in
Section 2.2.1, which are shifted by even integer powers of hl . For instance in the
free energy, as a consequence of the eh h2l term in uh , we expect a correction of
this type:

16
(69)
fs (hl ) = Alf |hl | 15 1 + Alf,3 |hl |2 +
with Alf,3 = eh . The indices f, 3 in Af,3 only denote the fact (that we shall discuss
in detail in the next section) that this term is the third term in the hl expansion of
the scaling function of the singular part of the free energy.
The second type of correction is due to the terms that depend on hl which appear
in ut . Their peculiar feature is that, even if they are originated by analytic terms
in the scaling variables, they lead in general to non analytic contributions in the
scaling functions For instance, as a consequence of the bt h2l term in ut , we find
in the free energy a correction of the type:

16
2 yt
(70)
fs (hl ) = Alf |hl | 15 1 + Alf,2 |hl | yh +
0
yt
(0)
22
with Af,2 = 15
16 (0) and 2 yh = 15 .
The corrections of the third type only appear when studying the internal energy.
They are due to the terms linear in t which are present in uh and ut . The most
important of these contributions is the one due to the ch t term in uh which gives a
8/15
to the dominant scaling behaviour of the internal
correction proportional to hl
energy. We shall discuss these terms in Section 3.4.3 below.
(b) Corrections due to irrelevant operators in the lattice Hamiltonian
These can be treated within the framework of the RG as follows. Let us study as
an example the case of an irrelevant operator belonging to the Identity family. Let
us call the corresponding scaling variable u3 and its RG eigenvalue y3 (since u3 is
irrelevant, y3 < 0). In this case the dependence of u3 on t and hl is 5
u3 = u03 + at + bh2l + .
(71)
Let us for the moment neglect higher order terms and assume u3 = u03 . Then looking
again at the singular part of the free energy we find

t
d/yh
0
|y3 |/yh
, u |hl |
.
fs (t, hl ) = |hl / h0 |
|hl |yt /yh 3
(72)
Since u03 |hl ||y3 |/yh is small as hl 0 it is reasonable to assume that we can expand
fs in a Taylor series of u03 |hl ||y3 |/yh (notice that in Eq. (72) fs is not singular since
it is evaluated at |hl | > 0). Hence we find (setting again t = 0)

(73)
fs = |hl |d/yh a1 + a2 u03 |hl ||y3 |/yh + ,
5 In general for the irrelevant operators there is no need to tune u0 to 0 to approach the critical point. However
3
we shall see below that, for symmetry reasons, u03 = 0 for all the irrelevant operators belonging to the [ ] and []
families.
684
where a1 , a2 , u03 are non-universal constants.

This analysis can be repeated without changes for any new irrelevant operator: u4 ,
y4 and so on. As a last remark, notice that on top of these non analytic corrections
we also expect analytic contributions due to the higher order terms contained in
Eq. (71).
While in general this analysis is only of limited interest since the yi of the irrelevant
operators are usually unknown, in the present case we may identify the irrelevant
operators with the secondary fields discussed in 3.1 and use the corresponding RGexponents as input of our analysis.
(c) Logarithmic corrections
As it is well known, the specific heat of the 2d Ising model at hl = 0 approaches the
critical point with a logarithmic singularity. This means that in the free energy there
must be a term of the type 0 u2t log(ut ). While in general we could expect 0 to
8/15
be a generic function of the ratio ut /uh , the absence of leading log corrections in
M and strongly constraints this function which is usually assumed to be a simple
constant. Notwithstanding this, the presence of terms that depend on hl in ut implies
that log type contributions may appear also in the case t = 0, hl 6= 0 in which we are
interested. These can be easily obtained by inserting Eq. (68) into 0 u2t log(ut ) and
then making the suitable derivatives and limits [27]. In the case of the free energy
one obtains a term proportional to h4l log(hl ) which is too high to be observed in our
fits. However for the internal energy the first contribution is proportional to a smaller
power of hl : h2l log(hl ) and must be taken into account in the scaling function.
3.3. The effective lattice Hamiltonian
Let us call HCFT the Hamiltonian which describes the continuum theory at the critical
point. The perturbed Hamiltonian 6 in the continuum is given by:
H = HCFT + h.
(74)
The aim of this section is to construct the lattice analogous (which we shall call Hlat ) of
H.
Notice that Hlat is different from the microscopic Hamiltonian which appears in the
exponent of Eq. (4). Eq. (4) describes the model at the level of the lattice spacing. We are
instead interested in the large distance effective action which one obtains when the short
range degrees of freedom are integrated out, i.e., after a large enough number of iterations
of the Renormalization Group transformation has been performed. Hlat will contain all the
irrelevant operators which are compatible with the symmetries of the lattice model. In this
section we shall first discuss the relation between the lattice and the continuum operators,
then we shall construct the lattice Hamiltonian in the hl = 0 case and finally we shall
extend our results to the hl 6= 0 case.
6 Here we follow the convention usually adopted in conformal field theory. In the standard notation of classical
statistical mechanics one would denote this quantity Hamiltonian density rather than Hamiltonian.
685
3.3.1. Relations between lattice and continuum operators

The lattice operators are given by the most general combination of continuum operators
compatible with the symmetries of the lattice operator multiplied by the most general
analytic functions of t and hl (with a parity which is again constrained by the symmetry
of the operators involved). In the following, to avoid a too heavy notation, we shall neglect
the t dependence 7 .
For the spin operator we have
l = f0 (hl ) + hl f0 (hl ) + fi (hl )i + hl fi (hl )i + hl fiI (hl )i ,
i N,
(75)
fi (hl ) fi (hl )
and fiI (hl ) are even functions of hl .

where
For the energy operator we have
l = g0 (hl ) + hl g0 (hl ) + hl gi (hl )i + gi (hl )i + h2l giI (hl )i ,
i N,
(76)
where again gi (hl ) gi (hl ) and giI (hl ) are even functions of hl and the h2l term in front of
giI (hl ) is due to the change of sign of the operator under duality transformation at hl = 0
(see the discussion at the beginning of Section 3.3.2).
Among all the possible irrelevant fields only those which respect the lattice symmetries
(i.e., those of spin 0 (mod 4)) are allowed in the sums. At this stage also irrelevant operators
containing L1 or L 1 appear in the sums. It is only when these operators are applied on
translationally invariant states (i.e., on the vacuum) that they disappear. This will happen
for instance when we shall study the mean value of the free energy.
3.3.2. Construction of Hlat (hl = 0)
In this case all the operators belonging to the [ ] family are excluded due to the Z2
symmetry. Also the operators belonging to the [] family are excluded for a more subtle
reason. The Ising model (both on the lattice and in the continuum) is invariant under duality
transformations while the operators belonging to the [] family change sign under duality,
thus they also cannot appear in Hlat (hl = 0). Thus we expect
Hlat = HCFT + u0i i ,
i [I ],
(77)
u0i
are constants. There are however further restrictions:

where the
Hlat is a scalar density, hence only operators with angular momentum j = 4k, k =
0, 1, 2, . . . are allowed.
The operator which acts on the Hilbert space of the theory is the space integral of the
Hamiltonian (density) Hlat . As such it is translational invariant and only operators
which do not contain the generators L1 or L 1 survive the integration of Eq. (77)
over the space.
Let us list in order of increasing conformal weight the first few operators which fulfill all
the constraints:
7 The t dependence in the scaling variables of the irrelevant operators plays a role only in the construction of
the scaling function for the internal energy and we shall resume it in Section 3.4.3 below.
686
1 = L2 L 2 I,
4 = L4 I, 5 = L 4 I,
0 = I,
2 = (L2 )2 I, 3 = (L 2 )2 I,
6 = L3 L 3 I, . . . .
(78)
(79)
Some of these fields have a natural interpretation. 0 gives rise to the bulk contribution in
the free energy (see the discussion of Section 2.1.2). 1 , 2 , 3 are related to the energy
momentum tensor: T T , T 2 , T 2 respectively. All the fields listed above except the identity
and 6 have the same conformal weight h = 4. The corresponding RG eigenvalue is y =
2. The field 6 has conformal weight h6 = 6 and RG eigenvalue y6 = 4.
3.3.3. Extension to hl 6= 0
Mimicking the continuum case we have, also on the lattice,
Hlat (hl ) = Hlat (hl = 0) + hl l .
(80)
Inserting the expression of l of Eq. (75) we find

Hlat (hl ) = HCFT + ui (hl )i ,
(81)
where this time there is no more restriction coming from the Z2 symmetry and duality,
hence i denotes here the most general operator of the spectrum with spin j = 4k, k =
0, 1, . . . . The ui (hl ) are even or odd functions of hl , depending on the parity of i but in
the even sector only for the operators belonging to the identity family limhl 0 ui (hl ) 6= 0
(according to Eq. (71) we have ui (hl = 0) = u0i ). For the operators belonging to the energy
family the first nonzero contribution in the ui (hl ) functions is of order h2l .
Let us list, starting from those with the lowest conformal weight, the new operators
which were not present in Hlat (hl = 0). For future convenience let us separate those which
do not contain L1 , L 1 generators from the remaining ones.
1
(A) States which are not generated by L1 , L
States belonging to [ ]
In the [ ] family the lowest ones are L4 , L 4 and L3 L 3 . In fact L1
disappears for translational invariance and due to the null vector equation the
L2 state which appears at level 2 can always be rewritten as L21 with
suitable coefficients. The conformal weights of L4 and L 4 are h,4 = 4 +
1
1
8 . The corresponding RG eigenvalue is y,4 = 2 8 . The conformal weight of
1
L3 L3 is h,33 = 6 + 8 . The corresponding RG eigenvalue is y,33 = 4 18 .

States belonging to []
The most important contribution from the [] family is the one proportional to
h2l which is responsible for the h2l term which appears in ut as we discussed
in the previous section. Besides this one, the lowest states which appear in the
[] family must be of the type L4 or L 4 . In fact the same mechanism which
allowed us to eliminate the secondary fields of level 2 in the [ ] family also works
for the [] family. On top of this in the [] family a new null vector appears at
level 3, thus allowing us to eliminate also all the fields at this level. Keeping also
into account the fact that the corresponding ui (hl ) functions must start from h2l
we immediately see that all these states have too high powers of hl to contribute
to the scaling function and can be neglected.
687
1 generators
(B) States which contain L1 , L
The lowest states are, in order of increasing weight:
L1 L 1 , whose conformal weight is h,11 = 2 + 18 . The corresponding RG
eigenvalue is y,11 = 18 .
L21 L 21 , whose conformal weight is h,22 = 4 + 18 . The corresponding RG
eigenvalue is y,22 = 2 18 .
L1 L 1 , whose conformal weight is h,11 = 3. The corresponding RG
eigenvalue is y,11 = 1.
3.4. Scaling functions
Using the results of the previous section we are now in the position to write the
expression for the scaling functions keeping all the corrections up to the order h3l .
3.4.1. The free energy
Due to translational invariance, only the secondary fields which are not generated by
L1 , L 1 contribute to the free energy. We find, for the singular part of the lattice free
energy:
16
16
22
30
32
fs (hl ) = Alf |hl | 15 1 + Alf,1 |hl | 15 + Alf,2 |hl | 15 + Alf,3 |hl | 15 + Alf,4 |hl | 15

38
44
+ Alf,5 |hl | 15 + Alf,6 |hl | 15 +
(82)
where Alf,n denotes the amplitude, normalized to the critical amplitude, of the nth
subleading correction.
Let us discuss the origin of the various corrections:
16
Alf,1 |hl | 15
This term is entirely due to the T T , T 2 and T 2 irrelevant fields in the Hamiltonian.
22
Alf,2 |hl | 15
This term is due to the bt h2l term in ut (or, equivalently, to the appearance of a h2
term in the Hamiltonian).
30
Alf,3 |hl | 15
This term is due to the eh h2l term in uh .
32
Alf,4 |hl | 15
This term keeps into account the second term in the Taylor expansion of the T T
like corrections and the contribution of the fields L3 L 3 I and hL4 in the
Hamiltonian.
38
Alf,5 |hl | 15
This is the product of the Af,1 and Af2 corrections.
44
Alf,6 |hl | 15
This is the second term in the Taylor expansion of the h2l correction.
To these terms we must then add the bulk contributions
fb (hl ) = fb + fb,1 h2l + fb,2 h4l + .
(83)
688
We have already noticed in Section 2.1.2 that fb can be obtained from the exact solution
of the Ising model on the lattice at the critical point. Also the next term: fb,1 can be
evaluated (with a precision of ten digits) by noticing that it corresponds to the constant
contribution to the susceptibility at the critical point. This term has been evaluated in [28].
We neglect for the moment this information and keep the fb,1 amplitude in the scaling
function as a free parameter. It is the first subleading term in the scaling function and as
such it can be rather precisely estimated with the fitting procedure that we shall discuss
below. We shall compare our estimates with the expected value in Section 6 and use the
comparison as a test of the reliability of our results.
Combining Eqs. (82) and (83) we find:
16
14
16
22
f (hl ) = fb + Alf |hl | 15 1 + Alf,b |hl | 15 + Alf,1 |hl | 15 + Alf,2 |hl | 15

30
32
38
44
+ Alf,3 |hl | 15 + Alf,4 |hl | 15 + Alf,5 |hl | 15 + Alf,6 |hl | 15 + ,
(84)
where Alf,b is fb,1 /Alf and Alf,6 takes also into account now the contribution of fb,2 .
Deriving this expression with respect to hl we obtain the scaling functions for the
magnetization and the susceptibility 8 .
3.4.2. Mass spectrum
The simplest way to deal with the mass spectrum is to turns out to be fit the square of
the masses. The scaling function is very similar to that which describes the singular part of
the free energy Eq. (82). The only additional terms are due to the secondary fields which
contain L1 L 1 . It turns out that the corresponding scaling dimension exactly match those
which already appear in Eq. (82). In fact
16
hl L1 L 1 gives a contribution which scales with |hl | 15 and its amplitude can be
absorbed in Alf,1 .
32
hl L2 L 2 gives a contribution which scales with |hl | 15 and its amplitude can be
1 1
absorbed in Alf,4 .
38
h2l L1 L 1 gives a contribution which scales with |hl | 15 and its amplitude can be
absorbed in Alf,5 .
Thus the functional form of the scaling function for the masses is exactly the same of
Eq. (82).
2
16
16
22
m2i (hl ) = Almi |hl | 15 1 + Almi ,1 |hl | 15 + Almi ,2 |hl | 15
30
32
+ Almi ,3 |hl | 15 + Almi ,4 |hl | 15

38
44
+ Almi ,5 |hl | 15 + Almi ,6 |hl | 15 + .
(85)
However we shall see below that the presence of these new fields and in particular of
L1 L 1 has very important consequences.
8 In this way we obtain directly the lattice definitions of these two quantities, since we are deriving the lattice
free energy with respect to the lattice magnetic field. There is no need to go through the continuum definition of
the magnetization.
689
3.4.3. Internal energy

We may obtain the internal energy as a derivative with respect to t of the singular
part of the free energy. However in doing this we must resume (as discussed above) the
t dependence in the scaling variables. This leads to some new terms in the scaling function
8
40
(due to the ch t term in uh ), 24
with powers 15
15 and 15 (due to the t terms in scaling
variables of the irrelevant operators). It is nice to see that the presence of these additional
contributions can be understood in another, equivalent, way. Looking at Eq. (16) or (76)
we see that the internal energy on the lattice contains a term of type hl . The powers listed
above are exactly those that we obtain keeping into account the additional hl term in
the scaling function. Keeping also into account the bulk contribution we end up with the
following scaling function. We have:
8
16
E(hl ) = AlE |hl | 15 1 + AlE,1 |hl | 15 + AlE,2 |hl | 15

22
22
+ AlE,b |hl | 15 + AlE,log |hl | 15 log|hl |

24
30
32
+ AlE,3 |hl | 15 + AlE,4 |hl | 15 + AlE,5 |hl | 15

38
40
+ AlE,6 |hl | 15 + AlE,7 |hl | 15 + + ,
(86)
where AlE,b denotes the amplitude of the h2l term in the bulk part of the internal energy,
AlE,log denotes the amplitude of the h2l log |hl | term discussed in Section 3.1 and the bulk
constant term has been already taken into account in the definition of E(h). The first
correction which appears in the internal energy (with amplitude AlE,1 ) is the one with the
lowest power of hl among all the subleading terms of the various scaling functions this.
Its effect on the scaling behaviour of the internal energy is very important and it is easily
observable also in standard Monte Carlo simulations [14].
3.4.4. Overlaps
Also in this case we fitted the square of the overlap constants. The scaling functions
can be obtained with a straightforward application of the arguments discussed above. Also
fields generated by hL1 L 1 must be taken into account. Moreover, for the overlaps with
the internal energy also the h term must be taken into account. We end up with the
following result for the magnetic overlaps:
14
16
22
|Fi (hl )|2 = |AlF |2 1 + AlF ,1 |hl | 15 + AlF ,1 |hl | 15 + AlF ,1 |hl | 15
i
i
i
i

28
30
+ AlF ,1 |hl | 15 + AlF ,1 |hl | 15 + + .
i
(87)
While for the energy overlaps we have

8
16
22
|Fi (hl )|2 = |AlF |2 1 + AlF ,1 |hl | 15 + AlF ,1 |hl | 15 + AlF ,1 |hl | 15
i
i
i
i

24
30
+ AlF ,1 |hl | 15 + AlF ,1 |hl | 15 + + .
i
(88)
690
4. The transfer matrix method

We computed the mass spectrum and observables by numerical diagonalization of the
transfer matrix. The transfer matrix was introduced by Kramers and Wannier [29] in 1941.
For a discussion of the transfer matrix see, e.g., Refs. [30,31]. The starting point is a simple
transformation of the Boltzmann factor
X
X
n m + hl
n = T (u1 , u2 ) T (u2 , u3 ) T (uL0 , u1 )
(89)
exp
hn,mi
dequation where un0 = ((n0 ,1) , (n0 ,2) , . . . , (n0 ,L1 ) ) is the spin configuration on the time
slice n0 . T is given by
T (un0 , un0 +1 ) = V (un0 )1/2 U (un0 , un0 +1 )V (un0 +1 )1/2
with
U (un0 , un0 +1 ) = exp
L1
X
(90)
!
(n0 ,n1 ) (n0 +1,n1 )
(91)
n1 =1
and
V (un0 ) = exp
L1
X
(n0 ,n1 ) (n0 ,n1 +1) + hl
n1 =1
L1
X
!
(n0 ,n1 ) .
The partition function becomes

X
X L
X
X
exp
n m + hl
n = tr T L0 =
i 0 ,
Z=
n 1
hn,mi
(92)
n1
(93)
where T is interpreted as a matrix. The time-slice configurations are the indices of the
matrix. The number of configurations on a time slice is 2L1 . Therefore the transfer matrix
is a 2L1 2L1 matrix. By construction the transfer matrix is positive and symmetric. The
i are the eigenvalues of the transfer matrix.
4.1. Computing observables
Observables that are defined on a single time slice can be easily expressed in the transfer
matrix formalism. Let us discuss as examples the magnetisation and the internal energy

P
P
P
=1 exp
hn,mi n m + hl
n n 1,1
h1,1 i =
Z
P L0
L
0
hi|S|ii
tr ST
= i Pi L
,
(94)
=
0
tr T L0
i i
where S is a diagonal matrix. The values on the diagonal are given by 1,1 on the
configurations. (S(u, u0 ) = (u, u0 ) u(1) where u(1) denotes on the first site of the time
slice). The |ii are normalized eigenvectors of T .
691
In the limit L0 the expression simplifies to

h1,1 i = h0|S|0i,
(95)
where |0i is the eigenvector with the largest eigenvalue.

The energy can be computed in a similar way. The diagonal matrix corresponding to the
energy is given by
E(u, u0 ) = (u, u0 )u(1)u(2) .
(96)
Note that we can only express the product of nearest neighbour spins in this simple form if
both spins belong to the same time slice.
In order to understand the relation of the mass spectrum with the eigenvalue spectrum
of the transfer matrix we have to compute correlation functions with separation in time
direction. The time-slice correlation function Eq. (22) becomes in the limit L0
X
exp mi | | h0|S|iihi|
S|0i
(97)
hS0 S i =
i
with

i
mi = log
0
(98)
P
S is translational invariant (in the space
dequation and e
S = L11 (u, u0 ) n1 un1 . Note that e
direction) and has therefore only overlaps with zero-moment eigenvectors of T .
With Eq. (23) we get
|Fi | =
mi L1
h0|e
S|ii
.
e
h0|S|0i
(99)
An analogous result can be obtained for the energy.

4.2. Computing the eigenvectors and eigenvalues of T
The remaining problem is to compute (numerically) eigenvectors and eigenvalues of
the transfer matrix. Since we are interested in the thermodynamic limit as well as in the
continuum limit we would like to use as large values of L1 as possible. This soon becomes
a very difficult task since the dimension of the transfer matrix increases exponentially
with L1 . The problem slightly simplifies if one is interested in the computation of the the
leading eigenvalues and eigenvectors only, and in these last years various methods have
been developed to address this task (for a comprehensive discussion of existing approaches
see, e.g., Ref. [31] or the appendix of Ref. [32]). In particular there are two approaches
which have shown to be the most effective ones.
The first one reduces the numerical complexity of the problem by writing the transfer
matrix as a product of sparse matrices. See Refs. [31,32].
The second one is to reduce the dimension of the transfer matrix by restricting it to
definite channels.
Since we are only interested in the zero-momentum states of the system we decided to
follow the second approach and to compute the zero-momentum reduced transfer matrix.
692
The zero-momentum reduced transfer matrix acts on the space of equivalence classes of
configurations on slices that transform into each other by translations.
The matrix elements of the reduced transfer-matrix are given by
1/2 X X
T (u, v)
Te(e
u,e
v ) = n(e
u )n(e
v)
ue
u ve
v
= n(e
u )/n(e
v)
1/2 X
T (u, v),
(100)
ve
v
where n(e
u ) is the number of configurations in e
u. For example for L0 = 20 the dimension
of the transfer matrix is reduced from 1048576 to 52488.
Still the matrix is too large to save all elements of the matrix in the memory of the
computer. Therefore we applied an iterative solver and computed the elements of Te
whenever they were needed.
As solver we used a generalized power method as discussed in the appendix of Ref. [32].
The lattice sizes that we could reach in this way were large enough for our purpose, thus
we made no further effort to improve our method and it is well possible that our algorithm
might still not be the optimal one.
We propose here, as a suggestion to the interested reader, some directions in which it
could be improved.
One could try to mix the two strategies mentioned above and try to factorize the
reduced transfer matrix as a product of sparse matrices. However note that the
complexity of the problem increases exponentially with the lattice size. Therefore
even a big improvement in the method would allow just to go up in the maximal L1
by a few sites.
One could study the transfer matrix along the diagonals
of the square lattice. Since
2,
naively
one could increase the
the distance between two points on the
diagonal
is
accessible lattice size by a factor of 2.
5. Thermodynamic limit
In order to take the thermodynamic limit we must know the finite size scaling behaviour
of the various observables as a function of L1 . This is a very interesting subject in itself
and several exact results have been obtained in this context starting from the exact S-matrix
solution and using Thermodynamic Bethe Ansatz (TBA) techniques [17,33].
For instance, it is possible to construct a large L asymptotic expansion for the finite size
scaling (FSS) of the energy levels based only on the knowledge of the exact S-matrix of
the theory [34,35]. Let us look to this FSS behaviour in mode detail.
Let us define 1ma (L) as the deviation of the mass ma of the particle a from its
asymptotic value:
1ma (L) ma (L) ma ().
(101)
Then in the large L limit, the shift (normalized to the lowest mass m1 ) is dominated by an
exponential decrease of the type

0
1 X 2abc
1ma (L)
2
exp (abc L),
m1
8ma
abc
693
(102)
b,c
where the constants abc and abc can be obtained from the S-matrix and the prime in the
sum of Eq. (102) means that the sum must be done only on those combinations of indices
that fulfill the condition: |m2b m2c | < m2a . In particular the abc turn out to be of order one,
so that the FSS corrections are dominated (as one could naively expect) by a decreasing
exponential of the type exp(L1 / ) where the correlation length is the inverse of the
lowest mass of the theory.
In principle we could use our data to test also the TBA predictions for the FSS. However
we preferred to follow a different approach. We chose values of h large enough so as to
fulfill the condition L1 / 1 for the largest values of L1 that we could reach. In this way
we could essentially neglect all the details of the FSS functions and approximate them with
a single exponential (or, in some cases, with a pair of exponentials). In order to study the
FSS functions one should choose smaller values of h. We plan to address this issue in a
forthcoming paper. With our choice of h we drastically simplify the FSS problem, however
nothing is obtained for free. The price we have to pay following this route is that we need
to know several terms in the scaling functions to fit such large values of h. This explains
the major effort that we devoted to this issue in Section 3.
5.1. Numerical extrapolation
According to the above discussion, for the extrapolation of our data to the thermodynamic limit we made no use of the quantitative theoretical results. We made only use of
the qualitative result that the corrections due to the finite L1 vanish exponentially.
We used as ansatz for the extrapolation either
A(L1 ) = A() + c1 exp(L1 /z1 )
(103)
A(L1 ) = A() + c1 exp(L1 /z1 ) + c2 exp(L1 /z2 ),
(104)
or
where A represents any of the quantities that we have studied.

In order to compute the free parameters A(), c1 and z1 or A(), c1 , z1 , c2 and z2
we solved numerically the system of equations that results from the lattice sizes L1,max ,
L1,max 1 and L1,max 2 or L1,max , L1,max 1, L1,max 2, L1,max 3 and L1,max 4.
The error of A() was estimated by comparing results where L1,max is the largest lattice
size that is available and from L01,max = L1,max 1. Mostly ansatz (104) was used to obtain
the final result. In some cases however the numerical accuracy was not sufficient to resolve
the second exponential term. Then the final result was taken from the ansatz (103).
In Table 6 we give, as example, the extrapolation to the thermodynamic limit of the free
energy at hl = 0.075. As input for the extrapolation we used the free energy computed
up to 12 digits. We consider all these digits save of rounding errors. Within the given
precision the free energy has not yet converged at L1 = 21. The single exponential
extrapolation (103) converges (within the given precision) at L1 = 18. For larger lattices
694
Table 6
Extrapolation of the free energy at hl = 0.075 to the thermodynamic limit.
In the first column we give the lattice size L1 . In the second column the
free energy for this lattice size is given. In the third column we present
the extrapolation with a single exponential and in the fourth column the
extrapolation with a double exponential ansatz
L1
f , Eq. (103)
f , Eq. (104)
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
0.993343441146
0.992384038449
0.992160642059
0.992102865951
0.992086845804
0.992082188258
0.992080787928
0.992080356320
0.992080220728
0.992080177480
0.992080163514
0.992080158958
0.992080157458
0.992080156961
0.992080156795
0.992080156739
0.992080156721
0.992080156714
0.992092835647
0.992082710940
0.992080699493
0.992080279123
0.992080185903
0.992080164020
0.992080158619
0.992080157225
0.992080156853
0.992080156752
0.992080156722
0.992080156715
0.992080156712
0.992080156710
0.992080156712
0.992080156710
0.992080279141
0.992080180502
0.992080161487
0.992080157709
0.992080156931
0.992080156758
0.992080156721
0.992080156716
0.992080156709
0.992080156713
0.992080156710
0.992080156710
0.992080156713
the result fluctuates in the last digit due to rounding errors of the input data. The
double exponential extrapolation (103) converges at L1 = 16. As final result for the
thermodynamic limit we quote f (0.075) = 0.99208015671.
6. Analysis of the results

The major problem in extracting the continuum limit results from the data listed in
Tables 1013 is to estimate the systematic errors involved in the truncation of the scaling
functions that we use in the fits. We shall devote the first part of this section to a detailed
description of the procedure that we followed to estimate this uncertainty. We shall give
upper and lower bounds for the critical amplitudes which turn out to be very near to
each other and allow for high precision predictions (in some cases we can fix 5 or even
6 significative digits). We then compare our predictions with the results obtained in the
framework of the S-matrix approach. In all cases we find a perfect agreement within our
bounds. Finally in Section 6.3, we give, assuming as fixed input the S-matrix predictions
for the critical amplitudes, our best estimates for the amplitudes of some of the subleading
terms involved in the fits.
695
Table 7
Fits to the magnetization fulfilling the requirements 14 (see text). In
the first column the best fit results for the critical amplitude (with in
parenthesis the error induced by the systematic errors of the input data), in
the second column the number of parameters in the fit, in the third column
the number of degrees of freedom and in the last column the confidence
level. For each value of Nf we only report the fits with the minimum
and maximum allowed number of d.o.f, since the best fit of Alf changes
monotonically as the data are eliminated from the fit
AlM
1.05898893(196)
1.05899447(58)
1.05898584(74)
1.05898882(22)
1.05898178(156)
1.05898375(8)
1.05898433(8)
1.05898694(18)
1.05898729(13)
Nf
d.o.f.
C.L. (%)
4
4
5
5
6
6
7
8
8
4
6
6
7
6
10
14
15
17
83
50
94
48
98
98
80
99
96
6.1. Systematic errors

In order to estimate the systematic errors involved in our estimates of the critical
amplitudes we performed for each observable several independent fits starting with a fitting
function containing only the dominant scaling dimension and then adding the subleading
fields one by one. For each fitting function we tried first to fit all the exiting data (those
listed in Tables 1013) and then eliminated the data one by one starting from the farthest
from the critical point (i.e., from those with the highest values of hl ). Among the (very
large) set of estimates of the critical amplitudes we selected only those fulfilling the
following requirements:
1. The reduced 2 of the fit must be of order unity 9 . In order to fix precisely a threshold
we required the fit to have a confidence level larger than 30%.
2. The number of degrees of freedom of the fit (i.e., the number of data fitted minus the
number of free parameters in the fitting function) must be larger than 3.
3. For all the subleading fields included in the fitting function, the amplitude estimated
from the fit must be larger than the corresponding errors, otherwise the field is
eliminated from the fit.
4. The amplitudes of the subleading fields (in units of the critical amplitude) must be
such that when multiplied for the corresponding power of hl , (for the largest value of
9 This is a slightly incorrect use of the 2 function since the input data are affected by errors which are of
systematic more than statistic nature. Notice however that we do not use it to determine best fit values for the
observables that we fit (we shall only give upper and lower bounds for them) but only as a tool to eliminate those
situations in which the fitting functions are clearly unable to describe the input data.
696
Table 8
Lower and upper bounds for various critical amplitudes discussed in the text
and, in the last two lines, for the two universal amplitude ratios R and Q2
Observable
Alf
AlM
Al
AlE
Alm1
Alm2
Alm3
|AlF |
Lower bound
Upper bound
Theory
0.9927985
1.058980
0.07055
0.58050
4.01031
6.486
7.91
0.6405
0.9928005
1.058995
0.07072
0.58059
4.01052
6.491
8.02
0.6411
0.9927995. . .
1.058986. . .
0.070599. . .
0.58051. . .
4.01040. . .
6.4890. . .
7.9769. . .
0.6409. . .
|AlF |
3.699
3.714
3.7066. . .
|AlF |
0.3
0.35
0.3387. . .
|AlF |
3.32
3.45
3.4222. . .
R
Q2
6.7774
3.2296
1
2
2
6.7789
3.2374
6.77828. . .
3.23514. . .
hl involved in the fit) must give a contribution much smaller than 1 (in order to fix a
threshold we required it to be strictly smaller than 0.3).
In general only a small number of combinations of data and degrees of freedom fulfills
simultaneously all these requirements. Among all the corresponding estimates of the
critical amplitude we then select the smallest and the largest ones as lower and upper
bounds 10 .
6.2. Critical amplitudes
In Table 7 we report as an example the fits to the magnetization (with the scaling function
obtained by deriving Eq. (84)) fulfilling the above requirements. For each value of Nf we
only report the fits with the minimum and maximum allowed number of d.o.f., since the
best fit result for AlM changes monotonically as the data are eliminated from the fit. This is
a general pattern for all the observables that we studied and greatly simplifies the analysis
of the data. Looking at the table one can see that at least four parameters are needed in
the fit to have a reasonable confidence level, due to the very small error of the data that
we use. In the last line we report the only fit in which all the 25 data reported in Table 10
have been used. It required taking into account the first eight terms of the scaling function.
For Nf > 8, even if we use all the data at our disposal we cannot fulfil requirement 3. It is
interesting to notice that the fits which give the best approximations to the exact value of
10 In making this choice we also keep into account the errors in the estimates induced by the systematic errors
of the input data.
697
AlM are those in which we use the largest possible number of terms of the scaling function.
This is a general pattern for all the observables that we studied. All the fits were performed
using the double precision NAG routine GO2DAF. The bounds that we obtained are listed
in Table 8 together with the S-matrix predictions. From these results we immediately obtain
the upper and lower bounds for the universal amplitude ratios of Table 2. They are reported
in the last two lines of Table 8.
6.3. Subleading operators
In principle we could try to estimate in the fits discussed above also the amplitudes of
the first two or three subleading terms in the scaling functions, however it is clear that the
results that we would obtain would be strongly cross correlated and we would not be able
to give reliable estimates for the corresponding errors (except, at most, for the first one of
them, the next to leading term in the scaling function).
In order to obtain some information on the subleading terms we decided to follow
another route. The results of the previous section strongly support the correctness of the Smatrix predictions. We decided then to assume these predictions as an input of our analysis,
fixing their values in the scaling functions. Then we used the same procedure discussed in
Section 6.1 to identify the amplitude of the first subleading field. Let us look to the various
scaling functions in more detail
6.3.1. Free energy
This is the case for which we have the most precise data. Moreover we may use the data
for the magnetization and the susceptibility as a cross check of our estimates.
Combining all the data at our disposal we end up with a rather precise estimate for Alf,b ,
which turns out to be bounded by:
0.055 < Alf,b < 0.050.
(105)
As mentioned in Section 3.4.1 it is possible to evaluate this amplitude in a completely

different way, by looking at the constant term in the magnetic susceptibility of the model at
the critical point. The comparison between our estimate and the expected value represents
a test of the reliability of our fitting procedure. The expected value of this amplitude [28]
is (in our units)
Alf,b = 0.0524442 . . .
(106)
which is indeed in perfect agreement with our estimates.

We can then use the value of Eq. (106) as a fixed input and try to estimate the
amplitude of the following subleading field which has a very important physical meaning
being the contribution due to the presence of the T T (and related terms) operator in the
lattice Hamiltonian. Remarkably enough, it turns out, by applying the usual analysis, that
the corresponding amplitude Alf,1 is compatible with zero. More precisely we see that,
changing the number of input data and of parameters in the scaling function, the sign
of Alf,1 changes randomly and its modulus is never larger than 104 . The same pattern is
698
Table 9
Lower and upper bounds for the amplitudes of some of the subleading
corrections
0.055 < Alf,b < 0.050
|Alf,1 | < 0.00005
0.020 < Alf,2 < 0.022
0.646 < AlE,1 < 0.644
|AlE,2 | <
0.005
0.21 < Alm ,1 < 0.20

1
0.48 < Alm ,1 < 0.41
2
0.65 < Alm ,1 < 0.50
3
Table 10
Data used in the fits
hl
0.20
0.19
0.18
0.17
0.16
0.15
0.14
0.13
0.12
0.11
0.10
0.09
0.08
0.075
0.066103019026467
0.055085849188723
0.05
0.044068679350978
0.033051509513233
0.03
0.022034339675489
0.02
0.015
0.01
0.0088137358702
1.106272538601(1)
1.096943627061(1)
1.087640179593(1)
1.078363862266(1)
1.069116534998(1)
1.059900287285(1)
1.050717483321(1)
1.041570819851(1)
1.032463401585(1)
1.023398841451(1)
1.014381396853(1)
1.00541615982(1)
0.99650933082(1)
0.99208015671(1)
0.98424336850(1)
0.97462849835(1)
0.97022834(1)
0.96513182856(1)
0.95578360408(2)
0.95322656(1)
0.9466343376(2)
0.94497330(2)
0.9409395(1)
0.936994(1)
0.93607461(2)
0.934113075978(1)
0.931644255995(1)
0.929017517063(1)
0.926215008782(1)
0.923215694344(1)
0.919994540350(1)
0.916521430645(1)
0.91275968274(1)
0.90866397795(1)
0.90417740232(1)
0.89922709483(1)
0.89371763122(1)
0.88752055778(1)
0.88411094491(1)
0.87741739906(1)
0.86771621938(2)
0.86255168(1)
0.8558157835(1)
0.840485633(1)
0.83533709(5)
0.81901353(2)
0.8139196(1)
0.7988985(1)
0.77805(5)
0.771605(1)
0.182495416253(1)
0.179101587939(1)
0.175544472125(1)
0.171809915290(1)
0.167881687799(1)
0.163741028380(1)
0.159366050850(1)
0.154730958303(1)
0.149804982192(1)
0.14455091814(1)
0.13892305302(1)
0.13286414108(1)
0.12630083230(1)
0.1228010112(1)
0.1161548337(1)
0.10703505648(2)
0.10241966(1)
0.096641767(1)
0.084469355(1)
0.0806726(1)
0.0695436(1)
0.0663409(2)
0.057595(1)
0.047045(3)
0.044149(2)
699
Table 11
hl
1/m1
1/m2
1/m3
0.20
0.19
0.18
0.17
0.16
0.15
0.14
0.13
0.12
0.11
0.10
0.09
0.08
0.075
0.066103019026467
0.055085849188723
0.05
0.044068679350978
0.033051509513233
0.03
0.022034339675489
0.59778522553(1)
0.61388448719(1)
0.63134670477(1)
0.65037325706(1)
0.67120940172(1)
0.69415734924(1)
0.71959442645(1)
0.74799884641(1)
0.77998715416(1)
0.81637015277(1)
0.85823913569(5)
0.9071039295(1)
0.965123997(1)
0.998514180(1)
1.067300500(2)
1.17524158(3)
1.237044(1)
1.322589(6)
1.54057(2)
1.6218(2)
1.91(1)
0.37795775263(1)
0.38765653507(1)
0.39818995529(1)
0.40968266918(1)
0.42228634593(5)
0.43618773124(1)
0.45161985381(4)
0.4688779288(2)
0.488342470(1)
0.510513817(1)
0.5360654(1)
0.5659287(6)
0.60144(1)
0.62189(1)
0.66405(5)
0.7305(1)
0.768(1)
0.82(1)
0.310888(1)
0.318578(1)
0.326940(1)
0.336077(2)
0.346115(3)
0.357209(3)
0.369548(4)
0.38338(1)
0.3990(1)
0.4168(1)
0.4374(5)
0.4624(5)
0.492(1)
0.508(1)
0.543(1)
reproduced in the magnetization and in the susceptibility. We summarize these observations

with the following bound
|Alf,1 | < 0.00005.
(107)
This result agrees with the observation concerning the absence of corrections due to
irrelevant operators in the case t 6= 0 and h = 0. (For a thorough discussion of this point
see [36] and references therein.)
If we also assume that Alf,1 = 0 then we may give a reliable estimate for the amplitude
of Alf,2 which turns out to be bounded by:
0.020 < Alf,2 < 0.022.
(108)
This is the highest subleading term that we could study with a reasonable degree of
confidence in our scaling functions.
6.3.2. Internal energy
In the case of the internal energy the first subleading amplitude can be studied with very
8
high confidence since it is associated to a very small exponent: |hl | 15 . The result turns out
to be
0.646 < AlE,1 < 0.644.
(109)
700
Table 12
hl
|F1 |2
|F2 |2
0.20
0.19
0.18
0.17
0.16
0.15
0.14
0.13
0.12
0.11
0.10
0.09
0.08
0.075
0.066103019026467
0.055085849188723
0.05
0.044068679350978
0.033051509513233
0.29041938711(1)
0.29570405694(1)
0.30107729858(1)
0.30653975241(1)
0.31209194307(1)
0.31773424601(1)
0.32346684419(1)
0.3292896717(1)
0.3352023388(1)
0.3412040323(4)
0.3472933781(4)
0.3534682486(5)
0.359725487(1)
0.362883627(1)
0.368548928(2)
0.3756378(4)
0.378934(5)
0.38280(2)
0.3899(1)
0.03800933(1)
0.04039999(1)
0.04291078(1)
0.04554676(1)
0.04831337(1)
0.05121641(1)
0.05426214(1)
0.05745711(3)
0.0608082(1)
0.0643227(2)
0.068008(1)
0.07187(1)
0.0759(1)
0.0780(2)
0.0818(5)
In this case the fits are so constrained that we can study with a rather good degree of
confidence also the next subleading correction, AlE,2 which is very interesting, since it
again contains the T T term discussed above. In agreement with the previous observations
also in this case the amplitude turns out to be compatible with zero. More precisely its sign
changes randomly as the input data are changed in the fits and its modulus can be bounded
by:
|AlE,2 | < 0.005
(110)
which is not as strong as the bound of Eq. (107) but clearly goes in the same direction.
6.3.3. Masses
The most interesting feature of the scaling functions for the masses is that there is no
analytic term and the first subleading contribution Almi ,1 is the exact analogous of the Alf,1
term for the free energy. In this case we find a non zero contribution for Almi ,1 . In particular
we find the following bounds for the three masses that we studied:
0.21 < Alm1 ,1 < 0.20,
(111)
< 0.41,
(112)
< 0.50.
(113)
0.48 < Alm2 ,1

0.65 < Alm3 ,1
In the case of the masses a preferred direction is singled out. Therefore, one has to expect
that there is a finite overlap with the irrelevant operator that breaks the rotational symmetry.
701
Table 13
hl
|F1 |2
|F2 |2
0.20
0.19
0.18
0.17
0.16
0.15
0.14
0.13
0.12
0.11
0.10
0.09
0.08
0.075
0.066103019026467
0.055085849188723
0.05
0.044068679350978
0.033051509513233
11.7000114647(1)
11.8448924368(1)
11.9888157747(1)
12.1315853880(1)
12.2729879270(1)
12.4127893980(1)
12.5507307560(1)
12.6865220830(5)
12.819834783(1)
12.950290902(2)
13.077448185(2)
13.200778543(2)
13.31963596(3)
13.3771415(1)
13.475815(5)
13.59037(3)
13.6398(5)
13.695(1)
13.78(1)
8.0468067(5)
8.3246112(5)
8.6017424(5)
8.8774943(5)
9.1511600(5)
9.422023(1)
9.689348(2)
9.952360(5)
10.21022(1)
10.46202(3)
10.7067(5)
10.943(3)
11.17(1)
11.28(1)
11.46(2)
11.6(5)
Notice that a similar contribution has been observed also in the case of the thermal
perturbation of the Ising model in [37] where the authors studied the breaking of rotational
invariance in the two point correlator (see Section 4.G of [37] for a discussion of this point).
Our results on the amplitude of the subleading corrections are summarized in Table 9.
7. Conclusions
The major goal of this paper was to test the S-matrix description proposed by
Zamolodchikov in [2] for of the 2d Ising model perturbed by a magnetic field. To this
end we developed some tools and obtained some results which are rather interesting in
themselves. In particular
We improved the standard transfer matrix calculations by implementing a zero
momentum projection which allowed us to drastically reduce the dimension of the
matrix.
We discussed in detail the relationship between continuum and lattice observables.
By using CFT results at the critical point we constructed the first 78 terms of the
scaling functions for various quantities on the lattice.
We could obtain in this way very precise numerical estimates for several critical
amplitudes (in some cases with 5 or even 6 significative digits) and in all cases we found a
perfect agreement between S-matrix predictions and lattice results.
702
By assuming the S-matrix predictions as an input of our analysis we could estimate

some of the subleading amplitudes in the scaling functions. In one case the value of
the subleading amplitude was already known and again we found a complete agreement
between theoretical prediction and numerical estimate. For the remaining ones there is up
to our knowledge no theoretical prediction. They are collected in Table 8 and represent
the most interesting outcome of our analysis. We leave them as a challenge for theorists
working in the field.
Among the others, the most surprising result concerns the T T term which turns out
to have a negligible amplitude in the scaling functions of the translationally invariant
observables. It would be nice to understand which is the reason of such behaviour.
Let us conclude by stressing that the techniques that we have developed can be easily
extended to the case in which a combinations of both thermal and magnetic perturbations
is present. In this case the exact integrability is lost and our numerical methods could help
to test new approaches and suggest new ideas.
Acknowledgement
We thank A.B. Zamolodchikov, Al.B. Zamolodchikov, V. Fateev, M. Campostrini,
A. Pelissetto, P. Rossi, E. Vicari and R. Tateo for useful discussions and correspondence
on the subject. In particular we are deeply indebted with A.B. Zamolodchikov for his help
in the construction of the scaling functions discussed in Section 3. This work was partially
supported by the European Commission TMR programme ERBFMRX-CT96-0045.
References
[2] A.B. Zamolodchikov, in: Advanced Studies in Pure Mathematics, Vol. 19, 1989, p. 641; Int. J.
Mod. Phys A 3 (1988) 743.
[3] A.B. Zamolodchikov, Al.B. Zamolodchikov, Ann. Phys. 120 (1979) 253.
[4] G. Mussardo, Phys. Rep. 218 (1992) 215.
[5] M. Hankel, H. Saleur, J. Phys. A 22 (1989) L513.
[6] V. Bazhanov, B. Nienhuis, S.O. Warnaar, Phys. Lett. B 322 (1994) 198.
[7] U. Grimm, B. Nienhuis, Phys. Rev. E 55 (1997) 5011.
[8] M.T. Batchelor, K.A. Seaton, J. Phys. A 30 (1997) L479; Nucl. Phys. B 520 (1999) 697.
[9] P.G. Lauwers, V. Rittenberg, Phys. Lett. B 233 (1989) 197, preprint Bonn University BONNHE-89-11 (unpublished).
[10] C. Destri, F. Di Renzo, E. Onofri, P. Rossi, G.P. Tecchiolli, Phys. Lett. B 278 (1992) 311.
[11] G. Delfino, G. Mussardo, Nucl. Phys. B 455 (1995) 724.
[12] M. Caselle, M. Hasenbusch, P. Provero, Nucl. Phys. B 556 (1999) 575.
[13] A.E. Ferdinand, M.E. Fisher, Phys. Rev. 185 (1969) 832.
[14] M. Caselle, P. Grinza, N. Magnoli, Nucl. Phys. B 579 (2000) 635, preceding paper in this issue.
[15] J. Cardy, Scaling and Renormalization in Statistical Physics, Cambridge University Press, 1996.
[16] V. Privman, P.C. Hohenberg, A. Aharony, Universal critical-point amplitude relations, in: C.
Domb, J.L. Lebowitz (Eds.), Phase Transition and Critical Phenomena, Vol. 14, Academic
Press, 1991.
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
703
V. Fateev, Phys. Lett. B 324 (1994) 45.

V. Fateev, S. Lukyanov, A. Zamolodchikov, Al. Zamolodchikov, Nucl. Phys. B 516 (1998) 652.
G. Delfino, Phys. Lett. B 419 (1998) 291.
G. Delfino, P. Simonetti, Phys. Lett. B 383 (1996) 450.
A.V. Smilga, Phys. Rev. D 55 (1997) 443.
B.M. McCoy, T.T. Wu, The Two Dimensional Ising Model, Harvard Univ. Press, Cambridge,
1973.
B.M. McCoy, in: V.V. Bazhanov, C.J. Burden (Eds.), Statistical Mechanics and Field Theory,
World Scientific, 1995.
T.T. Wu, Phys. Rev. 149 (1966) 380.
R. Hecht, Phys. Rev. 158 (1967) 557.
P. di Francesco, H. Saleur, J-B. Zuber, Nucl. Phys. B 290 (1987) 527.
A. Aharony, M.E. Fisher, Phys. Rev. B 27 (1983) 4394.
X.P. Kong, H. Au-Yang, J.H.H. Perk, Phys. Lett. A 116 (1986) 54.
H.A. Kramers, G.H. Wannier, Phys. Rev. 60 (1941) 252, Phys. Rev. 60 (1941) 263.
W.J. Camp, M.E. Fisher, Phys. Rev. B 6 (1972) 946.
M.P. Nightingale, in: V. Privman (Ed.), Finite Size Scaling and Numerical Simulation of
Statistical Systems, World Scientific, 1990.
H.L. Richards, M.A. Novotny, P.A. Rikvold, Phys. Rev. B 48 (1993) 14584.
Al.B. Zamolodchikov, Nucl. Phys. B 342 (1990) 695; Phys. Lett. B 253 (1991) 391.
M. Henkel, J. Phys. A 24 (1991) L133.
T.R. Klassen, E. Melzer, Nucl. Phys. B 362 (1991) 329.
J. Salas, A.D. Sokal, cond-mat/9904038v1.
M. Campostrini, A. Pelissetto, P. Rossi, E. Vicari, Phys. Rev. E 57 (1998) 184.

Boundary conditions
in rational conformal field theories
Roger E. Behrend a , Paul A. Pearce b , Valentina B. Petkova c,1 ,
Jean-Bernard Zuber d
a C.N. Yang Institute for Theoretical Physics, State University of New York, Stony Brook, NY 11794-3840, USA
b Department of Mathematics and Statistics, University of Melbourne, Parkville, Victoria 3052, Australia
c Arnold Sommerfeld Institute for Mathematical Physics, TU Clausthal, Leibnizstr. 10,
D-38678 Clausthal-Zellerfeld, Germany

d Service de Physique Thorique, CEA-Saclay, 91191 Gif-sur-Yvette Cedex, France
Received 16 August 1999; accepted 15 September 1999
Abstract
We develop further the theory of Rational Conformal Field Theories (RCFTs) on a cylinder with
specified boundary conditions emphasizing the role of a triplet of algebras: the Verlinde, graph fusion
and Pasquier algebras. We show that solving Cardys equation, expressing consistency of a RCFT
on a cylinder, is equivalent to finding integer valued matrix representations of the Verlinde algebra.
These matrices allow us to naturally associate a graph G to each RCFT such that the conformal
boundary conditions are labelled by the nodes of G. This approach is carried to completion for
sl(2) theories leading to complete sets of conformal boundary conditions, their associated cylinder
partition functions and the A-D-E classification. We also review the current status for WZW sl(3)
theories. Finally, a systematic generalization of the formalism of CardyLewellen is developed to
allow for multiplicities arising from more general representations of the Verlinde algebra. We obtain
information on the bulk-boundary coefficients and reproduce the relevant algebraic structures from
the sewing constraints. 2000 Elsevier Science B.V. All rights reserved.
By mistake none of the corrections the authors sent us have been incorporated in the earlier publication of
this paper (Nucl. Phys. B 570 (2000) 525589). Therefore the correct version of this paper will here be reprinted
in full.
1 Permanent address: Institute for Nuclear Research and Nuclear Energy, Tzarigradsko Chaussee 72, 1784
Sofia, Bulgaria.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 2 2 5 - X
708
R.E. Behrend et al. / Nuclear Physics B 579 [FS] (2000) 707773
1. Introduction
1.1. History and motivation
The subject of boundary conformal field theory has a fairly long history. It was born
more than ten years ago, in parallel work on open string theory [18] and on conformal field
theories (CFTs) describing critical systems with boundaries [9]. The work of Cardy [10]
was a landmark, leading to the unification of methods, to the introduction of important
concepts such as boundary conformal fields and to the systematic investigation of their
properties and couplings [11,12]. The subject remained dormant for some time, in spite
of some activity motivated again by string theory [1315] and of beautiful applications
to the Kondo problem [16,17]. Lately, the subject has undergone a revival of interest in
connection with various problems. On the one hand, work on boundary conditions in
integrable field theories and boundary flows [1820] and on quantum impurities [21,22]
motivated a closer look at boundary CFT. On the other hand, within statistical mechanics,
integrable Boltzmann weights satisfying the so-called Boundary YangBaxter Equation
(BYBE) were constructed in lattice models [2325]. Finally new progress in string theory
was another reason to reconsider the problem. Generalizations of D-branes as boundary
conditions in CFT have been studied by several groups [2634].
In the present work, we want to reconsider several issues in the discussion of boundary
conditions in (rational) conformal field theories: what are the general boundary conditions
that may be imposed, what are the structure constants of the bulk and boundary fields in
the presence of these boundary conditions. These are the basic questions that we want to
address. The methods that we use are not essentially new, but are based on the systematic
exploitation of the work of Cardy and Lewellen [1012].
Among the main results of this paper:
We establish a connection between the classification of boundary conditions and the
classification of integer valued representations of the fusion algebra. A preliminary
account of this result was given in [35,36]. In the same vein, we show that it is natural
to associate graphs to these problems. In particular, an ADE classification of boundary
conditions for WessZuminoWitten (WZW) and minimal sl(2) theories emerges in
a natural and simple way. A discussion of the state of the art for sl(3) models is also
included.
We point out the deep connections between the features of conformal field theory in
the bulk and in the presence of boundaries. The classification of the latter has some
bearing on the classification of the bulk properties (modular invariants, etc.). This is
not a new observation. In particular, in string theory many connections are known to
exist between open and closed string sectors, but it seems that the point had not been
stressed enough. A triplet of algebras, specifically the graph fusion algebra and its
dual, the Pasquier algebra, appears naturally in our discussion, along with the Verlinde
algebra.
We reanalyse in a systematic way the couplings (structure constants) of fields in the
presence of boundaries and the equations they satisfy, generalising the formalism
709
of CardyLewellen to accomodate the appearance of nontrivial multiplicities. In the

diagonal cases we find a direct relation between the chiral duality identities of Moore
and Seiberg and the basic sewing relations of the boundary CFT. The main point
is the observation that the bulk boundary coefficients in the diagonal case essentially
coincide with the matrices S(j ) of modular transformations of torus 1-point functions.
In this way the two basic bulk-boundary equations [11,12] are shown to be equivalent
to the torus duality identity of [37,38].
Some more particular results include the extension of Cardys equation to nonspecialized characters, thus lifting an ambiguity in the original derivation, the proof
of uniqueness of boundary conditions for b
sl(2) WZW and minimal models and b
sl(N)1
models, the clarification of the role of the graph algebra and the recovery of this algebra
along with its dual, the Pasquier algebra, from the boundary sewing constraints.
1.2. Background on bulk CFT
In this paper, we are only concerned with Rational Conformal Field Theories (RCFTs).
We first establish notations, etc. In the study of a RCFT, one first specifies a chiral
algebra A. It is the Virasoro algebra or one of its extensions: current algebra, W algebra,
etc. The generators of this algebra will be denoted generically Wn and include the Virasoro
generators Ln . At a given level, the theory is rational, i.e., A has only a finite set I of
irreducible representations Vi , i I. The label i indexes the representation conjugate
to i, and i = 1 refers to the identity (or vacuum) representation. We also suppose that
c
the characters i (q) = trVi q L0 24 of these representations, the matrix S of modular
transformations of the s and the fusion coefficients Nij k of the Vs are all known. The
matrix Sij is symmetric and unitary and satisfies S 2 = C, where C is the conjugation matrix
Cij = j i . The fusion coefficients are assumed to be given in terms of S by the Verlinde
formula [39]
Nij k =
X Sil Sj l S
kl
lI
S1l
(1.1)
an assumption that rules out some cases of RCFTs.

A physical conformal theory is then defined by a collection of bulk and boundary fields
and their 3-point couplings (OPE Coefficients). In particular, the spectrum of bulk fields is
described by the finite set Spec of pairs (j, j) of representations, possibly appearing with
some multiplicities Nj j , of the left and right copies of A, such that the Hilbert space of the
theory on an infinitely long cylinder reads
M
Vj V j ,
(1.2)
H=
(j,j)Spec
with the same multiplicities Nj,j . The modular invariant torus partition function
X

Nj j j (q) j (q)
Ztorus =
j,j
(1.3)
710
is a convenient way to encode this information. The finite subset E of labels of elements of
the spectrum that are leftright symmetric will play a central role in the following
E = {j |(j, j = j ) Spec},
(1.4)
and will be called the set of exponents of the theory. Recall that these exponents may come
with some multiplicities. To distinguish them as different elements of the set E a second
index will be often added, i.e., (j, ) E, for j I.
In terms of all these data, one is in principle able to compute exactly all correlation
functions of the CFT on an arbitrary 2D surface, with or without boundaries [12,37,38,40,
41]. These data, however, are subject to consistency constraints: single-valuedness of npoint functions on the plane, modular invariance of the torus or annulus partition function,
etc., all rooted in the locality properties of the theory. In this paper, we shall reexamine the
conditions that stem from surfaces with boundaries (half-plane or disk, cylinder or annulus)
and explore their consequences.
For later reference, let us also recall that RCFTs fall in two classes. In the first
class (type I), the Hilbert space (1.2) is a diagonal sum of representations of a larger,
extended, algebra A0 A. Accordingly, the partition function (1.3) is a sum of squares
of sums of characters
X X 2

i .
(1.5)
Z=

blocks B iB
The second class (type II) is obtained from the first by letting an automorphism of the
fusion rules of the extended algebra A0 act on the right components, thus resulting in a
non-block-diagonal partition function

X X X
i
j .
(1.6)
Z=
B
iB
j (B)
For example, in the classical case of sl(2) theories, classified by A-D-E Dynkin diagrams,
the A, D2p , E6 and E8 cases are of the first type, whereas the D2p+1 and E7 are obtained
respectively from the A4p1 and D10 cases by a Z2 automorphism of their fusion rules.
We shall see below that the study of boundary conditions on a cylinder has some bearing
on these expressions of torus partition functions.
2. Cardy equation and Verlinde algebra

2.1. Boundary states
As discussed in [42,43], on the boundary of a domain such as the upper half plane or a
semi-infinite cylinder, one must impose a continuity condition of the form

W (z) = W (z)z=z .
(2.1)
T (z) = T (z)z=z ,
While the first of these conditions has the direct physical meaning of the absence of energymomentum flow across the boundary, or the preservation of the real boundary by diffeo-
711
morphisms, the condition(s) on the other W may be generalized to incorporate a possible

gluing automorphism [26,28,29,44]

(2.2)
W (z) = W (z)z=z .
A semi-annular domain in the upper half-plane may be conformally mapped into an
annulus in the complex plane by = exp(2iw/T ), w = L log z. Then as shown by
Ishibashi [2] and Cardy [10], the boundary condition becomes
2 T ( ) = 2 T ( ),
sW W ( ) = ( )sW W ( )
for | | = 1 and | | = e2L/T ,
(2.3)
where sW denotes the spin of W , or more generally

sW W ( ) = ( )sW W ( ).
Through radial quantization, this translates into a condition on boundary states |ai

(2.4)
Wn (1)sW (W n ) a = 0.
This includes in particular the condition that

Ln L n a = 0,
(2.5)
assuming that the automorphism keeps invariant the Virasoro generators. For the central
charge operator we have (k k)|ai

= 0.
Solutions to this linear system are spanned by special states called Ishibashi states [2],
labelled by the finite set E = {j |(j, j = (j )) Spec}, where depends in particular
on . To see this, let us consider first the simpler equation (2.5) in the case when A
is the Virasoro algebra and is trivial. Then observe that one may solve (2.5) in each
component of (1.2) independently, as these spaces are invariant under the action of the two
P
in Vj Vj is in onecopies of A. Now we recall that any state A = n,n an,n |j, ni |j, ni
P
of Vj into Vj .
to-one correspondence with a homomorphism XA = n,n an,n |j, nihj, n|
This uses the scalar product in Vj . Since Ln = Ln for that scalar product, (2.5) implies
that Ln XA = XA Ln , i.e., that XA intertwines the action of Ln in the two representations Vj
and Vj of the Virasoro algebra. As these two representations are irreducible, they must be
equivalent, which by our convention on the labelling of representations, means that j = j.
Thus the only non-vanishing components of A in H are in diagonal products Vj Vj and in
P
each one, XA is proportional to the projector Pj = n |j, nihj, n|. To fix the normalization
we choose XA = Pj and the corresponding Ishibashi state is denoted |j ii. This completes
the proof 2 that there is an independent boundary state |j ii for each element of the set
E = {j |(j, j = j ) Spec}.
The argument is a formal extension of the proof, based on the Schur lemma, of
the existence and uniqueness of an invariant in the tensor product of finite-dimensional
representations. It extends to the odd spin sW case (2.4). We have to use the fact that
2 Many thanks to G. Watts (private communication) to whom we owe this elegant derivation. Some elements
had appeared already in M. Bauers PhD thesis (1989).
712
Wn = (1)sW U 1 Wn U with respect to a bilinear (or Hermitian) form where U is a

unitary (or antiunitary) operator. One exploits the same definition of the homomorphism
XA : Vj 0 Vj , now Vj 0 , Vj being highest weight representations of the chiral algebra
A generated by Wn . However XA corresponds to states in Vj U U Vj 0 , where U is
1
.
a unitary (antiunitary) operator implementing the automorphism (Wn ) = U Wn U
0
Eq. (2.4) leads to Wn XA = XA Wn again with the result j = j and XA = Pj while the
P
Ishibashi states are given by |j ii = n |j, ni U U |j, ni.
The operator U is in general nontrivial, e.g., for the b
sl(N)k WZW theories Wn =
w(W
n ) where w is the horizontal projection of the Chevalley involution of the affine
i ) = ei , for the simple roots
algebra [45], i.e., it is determined by w(e
i ) = f i , w(f
i of sl(N), where e /f are raising/lowering operators respectively, and for the Cartan
generators w(h
i ) = hi , i = 1, 2, . . . , N 1. The pair (j, (j )) characterising E refers
to the eigenvalues of hi on the first term in |j ii . If is the identity, then (j ) = j . If
= w0 , where w0 represents the longest element of the Weyl group, then w0 (j ) = j
and hence (j ) = w0 w(j
) = j . On the other hand (j ) = j for coinciding with the
Chevalley automorphism = w.
In the last two of these examples we can identify V j =
U U Vj with a highest weight module V(j ) , (j ) I. It should be stressed that all these
automorphisms keep invariant the Sugawara Virasoro generators so the condition (2.5)
is also satisfied on the corresponding Ishibashi states.
We shall hereafter drop the explicit dependence on .
In fact we still have to define a norm (or a scalar product) on boundary states, in
particular on the Ishibashi states. We have to face two difficulties. First because of the
infinite dimension of the representation Vj , the most naive norm, proportional to tr Pj ,
would be infinite. The second problem concerns non-unitary representations. In such cases,
the hermitian form on Vj is not positive definite, and we may encounter signs in the norm
of states.
1
i
The first problem requires some regularization of the naive norm. Let q 2 = e be a
1
c
c
We write in
real number, 0 < q < 1. Then hhj |q 2 (L0 +L0 12 ) |j ii = tr Pj q L0 24 = j (q).
general, allowing some multiplicity = 1, . . . , Njj for the representations:

0 0 1 (L +L c )
(2.6)
j , q 2 0 0 12 j, = jj 0 0 j (q).
The norm of |j ii should then be some renormalized version of the q 1 limit of (2.6)
tends to 0. In
[2,10], i.e., of the limit in which q = e2i , the modular transform of q,
unitary theories, a new scalar product on boundary states may be defined according to
1

c
hhj kj 0 0 ii = lim q 24 j 0 , 0 q 2 (L0 +L0 12 ) j,

q1
= jj 0 0 S1j ,
= jj 0 0 lim q 24 j (q)
q0
(2.7)
where we have used the fact that in a unitary theory, the leading character in the q 0
c
limit is that of the identity operator 1 (q) q 24 . Note that Sj 1 is, up to a factor 1/S11 ,
the quantum dimension of the representation j , a positive number. Thus in unitary theories
the normalization chosen for Xj is such that the states |j ii are orthogonal for the scalar
product (2.7), with a square norm equal to S1j . Although in non-unitary theories the
713
limit q 0 in (2.7) does not exist in general, due to the existence of representations of
conformal weight hi < 0 that will dominate that limit, we may still define the norm by the
same formula as (2.7). Alternatively, if j0 denotes the unique representation of smallest
conformal weight hj0 < 0 belonging to E, and ceff := c 24hj0 , then we may define
hhj kj 0 0 ii = jj 0 0
ceff
S1j
lim q 24 j (q).
Sj0 j q0
(2.8)
In all cases we thus have

hhj kj 0 0 ii = jj 0 0 S1j
(2.9)
which is now of indefinite sign. In the sequel, we use more compact notations and the
multiplicity label will be implicit when referring to j E.
The most general boundary state |ai satisfying condition (2.4) must be a linear
combination of these Ishibashi states, which, for later convenience, we write as
|ai =
X aj
p
|j ii.
S1j
j E
(2.10)
We denote by V = {a} the set labelling the boundary states. We assume that an involution
j
j
j
a a in the set V is defined and that a = a = (a ) , where j j is an involution
in E (in general (j, ) (j , ), see Appendix B for examples). We define conjugate
states as
hb| :=
j
X
hhj | p b .
S1j
j E
(2.11)
As explained in [26], this conjugate state may be regarded as resulting from the action of
an antilinear CPT operation. As a consequence

X j j
X aj j
b
hhj kj ii =
a b
(2.12)
hbkai =
S1j
j E
j E
so that the orthonormality of the boundary states is equivalent to that of the s.

In some cases, such as in the computation of partition functions involving the specialised
characters in the next section, it is sufficient to impose only the Virasoro condition (2.5)
on the boundary states. Then the sum in (2.10), when interpreted in terms of Ishibashi
states pertaining to some extended symmetry, may include states |j ii with different .
For example, in the minimal b
sl(2) models when multiplicities occur in E, one can build
the Ishibashi states using the Coulomb gas realisation with A = u(1).
Then there are two

choices of keeping Ln invariant. This allows, in particular, the construction of two
different Ishibashi states with the same value of the scaling dimension, i.e., the explicit
resolution of the degeneracy of states denoted |j, ii above. Such mixtures of Ishibashi
states may be used in determining the boundary states of the non-diagonal (A, Deven )
models.
714
Fig. 1. The two computations of the partition function Zb|a : (a) on the cylinder, between the boundary
states a and b, (b) as a periodic time evolution on the strip, with boundary conditions a and b.
2.2. The Cardy equation

We now consider a conformal field theory on a finite cylinder. Following Cardy, the
partition function may be expressed in two alternative ways. Regarded as resulting from
the evolution of the system between boundary states a and b under the action of the
Hamiltonian on (i.e., the translation operator along) the cylinder, it is
1
c
(2.13)
Zb|a = b q 2 (L0 +L0 12 ) a ,
where q 2 = e2 T describes the aspect-ratio of the cylinder of period T and length L
as in Fig. 1. Decomposing the boundary states on the Ishibashi basis and using (2.6), one
obtains
X j j j (q)
a b
,
(2.14)
Zb|a =
Sj 1
L
j E
where the states |j ii are admissible Ishibashi states of the system, i.e., the label j runs over
the set E.
On the other hand, Zb|a may be regarded as resulting from the periodic time evolution
under the action of the translation operator along the finite width strip in the presence
of boundary conditions a and b. The latter manifest themselves only in the nature of the
Hilbert space Hba and its decomposition into representations of a single chiral algebra:
L
T
Hba = i nib a Vi with non-negative integer multiplicities nib a . If q = e L , Zb|a is a
linear form in the characters
X
i (q)nib a .
(2.15)
Zb|a =
iI
We choose to write the modular transformation of characters in the form i (q) =

P
P
hence j (q)
= i Sj i i (q). Provided that specialized characters i (q)
j Sij j (q),
are considered, this complex conjugation is immaterial, since i (q) = i (q). We shall,
however, make later use of unspecialized characters (Appendix A), for which it does matter.
With this convention, and assuming for the time being the independence of characters, the
two expressions (2.14) and (2.15) are consistent provided
nia b =
715
X Sij j j
a b .
S1j
(2.16)
j E
In the sequel we will refer to this as the Cardy equation. In the left-hand side, we have used
the first of the following symmetries
nia b = ni b a = nib a ,
(2.17)
j
a .
which follow from the properties of the modular matrix and of the coefficients
The boundary states |ai, |bi, are thus such that nia b is a non-negative integer. Uniqueness
of the vacuum implies n1a b 6 1. The Cardy equation (2.16) is a non-linear constraint on
the components of boundary states |ai and |bi on the basis of Ishibashi states. Note also
P
/ E and, except in cases with multiplicities,
that it implies that i nib a Sj i vanishes if j
must factorize into a product of contributions of the a and b boundary states, a nontrivial
constraint. Still, these constraints seem difficult to solve in full generality.
Before we proceed, we have to pause on the question of independence of characters.
In general, it is not true that specialized characters such as those that we have been using
so far, are linearly independent. For instance, complex conjugate representations i and i
give rise to the same character i (q). Unfortunately, little is known about unspecialized
characters for general chiral algebras, beside the case of affine algebras. In Appendix A,
we show that in the case of rational conformal field theories with a current algebra, the
previous discussion may indeed be repeated if the energy momentum tensor of the theory
has been modified in such a way that unspecialized characters appear. Then using the
known modular transformations of the latter [45], one derives (2.16). We shall therefore
assume that (2.16) holds true for general RCFT.
We now return to the Cardy equation (2.16), and assume that we have found an
orthonormal set of boundary states, i.e., satisfying
X j j
a b = ab .
(2.18)
j E
Moreover, we make the stronger assumption that we have found a complete set of such
states, i.e., satisfying
X j j 0
a a = jj 0 .
(2.19)
a
(Note that this implies that the number of these boundary states must be equal to the
cardinality of E.)
Finally we recall that the ratios Sij /S1j , for a fixed j I, form a one-dimensional
representation of the fusion algebra, as a consequence of the Verlinde formula (1.1):
X
Si j
Si1 j Si2 j
=
Ni1 i2 i3 3 .
(2.20)
S1j S1j
S1j
i3 I
It follows from (2.19), (2.20) that the matrices ni , defined by

(ni )a b = nia b ,
i I,
(2.21)
716
also satisfy the (commuting) fusion algebra

X
Ni1 i2 i3 ni3 .
ni1 ni2 =
(2.22)
i3 I
By (2.18), n1 = I , the unit matrix, and by (2.17), ni = nTi .

Conversely, given a set of matrices with non-negative integer elements, satisfying ni =
T
ni , n1 = I and the fusion algebra, they form a commuting set, and thus each ni commutes
with its transpose. These matrices are thus normal matrices that may be diagonalized in
an orthonormal basis. Their eigenvalues are of the form Sij /S1j for some j , and they may
thus be written in the form (2.16). If one pretends to determine the spectrum E from the
ns, one has to impose also that j = 1 appears only once in E, as a manifestation of the
uniqueness of the vacuum.
We thus conclude that the search for orthonormal and complete solutions to the Cardy
equation is equivalent to the search for N valued representations of the fusion algebra
satisfying nTi = ni .
This is the first important result of this paper, already presented succinctly in [36]. The
fact that some solutions to the Cardy equation were associated with representations of
the fusion algebra had been noticed before. In his seminal paper [10], Cardy considered
the case of diagonal theories (for which E = I) and showed that the ni matrices were
nothing other than the fusion matrices Ni , thus obtaining an alternative and more intuitive
derivation of the Verlinde formula. In an antecedent work by Saleur and Bauer [9], other
solutions had been obtained in non-diagonal theories, starting from their lattice realization,
and the fact that these ni coefficients satisfied the fusion rules had been emphasized in [46].
More recently Pradisi, Sagnotti and Stanev [14,15,34] proposed a different argument to
the same effect, where a notion of completeness of boundary conditions is also playing a
crucial role.
Solutions such that all matrices ni may be written as ni = (n1 )i (n2 )i after the same
suitable permutation of rows and columns can be called reducible. They describe sets
of decoupled boundary conditions. We thus restrict our attention to irreducible sets of
matrices.
2.3. WZW sl(2) theories
For theories with the affine (current) algebra b
sl(2) as a chiral algebra, the problem of
classifying representations of the fusion algebra was solved long ago [46]. The integrable
highest weight representations
of b
sl(2)k at level k N are labelled by an integer 1 6
q
j 6 k + 1, Sij =
ij
2
k+2 sin k+2 and the
S2j
j
S1j = 2 cos k+2 . The
Cardy equation says that the generator n2 =
only symmetric irreducible matrices with nonn2 has eigenvalues

negative integer entries and eigenvalues less than 2 are the adjacency matrices of A-DE-T graphs [47] of Fig. 2 (see also Table 1). The tadpole graphs are given by Tn :=
A2n /Z2 . Here the level k is related to the Coxeter number by g = k + 2. Only the A-D-E
solutions are retained as their spectrum matches the spectrum of b
sl(2) theories, known by
their modular invariant partition functions [4850]. For a theory classified by a Dynkin
717
Fig. 2. The A-D-E-T graphs.

Table 1
The Coxeter number g and Coxeter exponents m of the A-D-E-T
graphs
G
m Exp(G)
An
n+1
1, 2, 3, . . . , n
Dn
2n 2
1, 3, 5, . . . , 2n 3, n 1
E6
12
1, 4, 5, 7, 8, 11
E7
18
1, 5, 7, 9, 11, 13, 17
E8
30
1, 7, 11, 13, 17, 19, 23, 29
Tn
2n + 1
1, 3, 5, . . . , 2n 1
diagram G of A-D-E type, the set E is the set of Coxeter exponents of G as in Table 1.
The matrices ni are then defined recursively by n1 = I , n2 = G and by Eq. (2.22) which
reduces here to ni+1 = n2 ni ni1 , i = 2, 3, . . . , k. They are the well known fused
adjacency matrices or intertwiners Vi , studied in [46,57] and whose properties are
718
recalled in Appendix B. One verifies that all their entries are non-negative integers. This
set of complete orthonormal solutions of the Cardy equation for b
sl(2) theories is unique up
to a relabelling of the states |ai.
2.4. Minimal sl(2) models
The classification of c < 1 minimal models in the bulk is given in terms of a pair of
Dynkin diagrams (A, G) where G is of A-D-E type [4850]. Let h be the Coxeter number
of Ah1 and g the Coxeter number of G as given in Table 1. Then the complete A-D-E
classification is
M(Ah1 , Ag1),
M(A
h1 , D(g+2)/2), g even,
(2.23)
M(A, G) = M(Ah1 , E6 ),
,
E
),
M(A
h1
7
M(Ah1 , E8 ),
with h, g > 2 and central charges given by
6(h g)2
.
(2.24)
hg
We will use M(A, G) to denote these minimal theories. Since g and h must be coprime and
g is even for all non-A cases, one may always assume, at the price of a possible interchange
in the (A, A) case, that h is odd, h = 2p + 1.
Some members of these series are identified as follows:
c=1
M(A2 , A3 ) = critical Ising,

M(A4 , A3 ) = tricritical Ising,
M(A4 , D4 ) = critical 3-state Potts,
M(A6 , D4 ) = tricritical 3-state Potts,
c = 1/2,
c = 7/10,
c = 4/5,
c = 6/7.
(2.25)
We will use G to denote both the Dynkin diagram and its adjacency matrix. We use r,
r1 , r2 to denote nodes or exponents of Ah1 ; s, s1 , s2 for the nodes (or exponents) of Ag1 ;
a, a1 , a2 , b for the nodes of G. We refer the reader to Appendix B for more data on these
matrices and their eigenvectors.
If Exp(G) denotes the set of exponents of G (see Table 1), the modular invariant partition
function of M(Ah1 , G) reads
1X
2
h1
Z=
|rs (q)|2 + off-diagonal terms.
(2.26)
r=1 sExp(G)
The factor 12 removes the double counting due to the well-known identification of the (r, s)
and (h r, g s) representations of the Virasoro algebra. The diagonal terms in Z, i.e.,
the leftright symmetric (highest weight) states in the spectrum are thus labelled by the set

E = j = (r, s) (h r, g s); 1 6 r 6 h 1; s Exp(G) .
(2.27)
Each of the unitary minimal models M(Ah1 , G) with g h = 1 can be realized as
the continuum scaling limit of an integrable two-dimensional lattice model at criticality,
719
with heights living on the nodes of the graph G. In particular, the critical series with
g h = 1 is associated with the A-D-E lattice models [52,53] and the tricritical series with
g h = 1 is associated with the dilute lattice models [54,55]. In the non-unitary cases
the associated lattice models [56] possess negative Boltzmann weights. In the construction
of the corresponding lattice models as well as in the description of boundary conditions, it
turns out that the two diagrams of the pair (A, G) do not play a symmetric role.
According to the discussion of the previous section, we have to study the fusion algebras
of minimal models and their (integer-valued) representations. The Verlinde formula for the
fusion coefficients makes use of the matrix S of modular transformations of characters
s
8
gh
gh
0
0
(1)(r+s)(r +s ) sin rr 0
sin ss 0
,
(2.28)
Srs,r 0 s 0 =
gh
h
g
with the restriction r, r 0 odd (or any equivalent condition). The fusion coefficients are then
found to be tensor products of those relative to the b
sl(2) algebras of level h 2 and g 2,
up to a symmetrization which accounts for the identification (r, s) (h r, g s)
Nrs,r 0 s 0 r
00 s 00
00
00
00
00
= Nrr 0 r Nss 0 s + Nrr 0 hr Nss 0 gs .
(2.29)
This may be regarded as the regular representation of the fusion matrices Nrs of the
Virasoro algebra of central charge (2.24). Our problem is to find the general non-negative
integer valued representations of this algebra. One observes that Nrs = Nr1 N1s and that
the algebra is thus generated by N21 and N12 . Also, the eigenvalues of N12 and N21 are of
the form
gh
h
S12,r 0 s 0
0
0
0
= (1)r 2 cos s 0 ,
(2.30)
= (1)r +s 2 cos s 0
S11,r 0 s 0
g
g
S21,r 0 s 0
g
0
(2.31)
= (1)s 2 cos r 0 ,
S11,r 0 s 0
h
with 1 6 r 0 6 h 1, 1 6 s 0 6 g 1 and again (r 0 , s 0 ) (h r 0 , g s 0 ).
Turning now to a general (integer valued) representation nrs of the fusion algebra, it is
still true that it is generated by n12 and n21 . In addition, we want the spectrum of the nrs
to be specified by the set of exponents E of (2.23), that is (r 0 , s 0 ) in (2.30), (2.31), with
the eigenvalues labelled by s 0 appearing with some multiplicity in general. To remove the
redundancy in the labelling of eigenvalues, we will usually take r 0 odd, r 0 = 1, 3, . . . , h 2,
and (s 0 , ) Exp(G). In the sequel, we will drop this explicit notation for multiplicities.
We know of course a solution to this problem, namely,
nrs = Nr Vs + Nhr Vgs ,
(2.32)
in terms of the fusion matrices N of b

sl(2) at level h 2 and of the intertwiners V of type G
introduced in the previous subsection (see also Appendix B). More explicitly, this describes
a solution to the Cardy equation between boundary states (r1 , a) and (r2 , b)
nrs;(r1,a) (r2 ,b) = Nrr1 r2 Vsa b + Nhr r1 r2 Vgs a b ,
(2.33)
with 1 6 r, r1 , r2 6 h 1 = 2p, 1 6 s 6 g 1, and a, b running over the nodes of the

Dynkin diagram G.
720
Because of the properties of the N and V matrices recalled in Appendix B, it is readily

seen that
nrs;(r1,a) (r2 ,b) = nrs;(r1,a) (hr2 , (b)) = nrs;(hr1, (a))(r2 ,b)
(2.34)
for an automorphism acting on the nodes of the graph G: this is the identity except for
the A, Dodd and E6 cases, for which it is the natural Z2 symmetry of the diagram. We
conclude that this solution describes boundary states of M(Ah1 , G) labelled by pairs
(r, a) of nodes of the Ah1 and of the G graph, with the identification
(r, a) (h r, (a)).
(2.35)
One checks that the number of independent boundary states |(r, a)i is
1
(2.36)
number of independent boundary states = (h 1)n
2
with n the number of nodes of G, or the number of its exponents. This number (2.36)
coincides with the number of independent leftright symmetric highest weight states
|r, si |r, si in the spectrum of the theory on a cylinder, i.e., with the cardinality of the set
E, as it should.
With such boundary states, the cylinder partition function reads
X
rs (q)Nrr1 r2 Vsa b .
(2.37)
Z(r1 ,a)|(r2,b) = Z(r2,b)|(r1 ,a) = Z(r1 ,a)|(hr2, (b)) =
r,s
Here the sum runs over 1 6 r 6 h 1, 1 6 s 6 g 1.

Let us look more closely at (2.33). There exists a basis in which (2.33) takes a factorized
form. Indeed one may use the identifications (r, s) (h r, g s) and (2.35) to restrict r,
r1 , r2 to odd values (recall that h = 2p + 1 is odd). Then Nhrr1 r2 = 0, the r.h.s. of (2.33)
factorizes and the following expressions
00
(r 00 s 00 )
(2.38)
(r,a) = 2 Srr 00 as , r, r 00 odd, s 00 Exp(G),
written in terms of the modular matrix of b
sl(2) at level h 2 and of eigenvectors of G,
are readily seen to be eigenvectors of nrs . Their eigenvalue is of the form Srs,r 0 s 0 /S11,r 0 s 0
after some reshuffling r 00 , s 00 r 0 , s 0 .
One also shows (see Appendix C) that there exists a basis in which
G
,
(2.39)
n12 = Ip G =
..
n21 = Tp =
0
..
.
..
.
(2.40)
in terms of the tadpole Tp adjacency matrix and of , the matrix that realizes the
automorphism : a b = a (b) .
721
Conversely, suppose we only know that the representation nrs has a spectrum specified
by the set of exponents E. The question is: are these spectral data sufficient to guarantee
that the only nrs are of the form (2.33) in a certain basis? A proof of this fact is relegated
to Appendix C. Notice that our discussion has assumed the classification of modular
invariants to be known. It should be possible to extend it as in the case of WZW b
sl(2)
models and to classify the representations of the Verlinde algebra without this information.
A few spurious cases involving tadpoles, etc., would then have to be discarded.
To recapitulate, we have proved that the only representations of the fusion algebra
of minimal models are given by (2.33). To our knowledge, this is the first proof of the
uniqueness of these (complete orthonormal) boundary states of minimal models.
Some physical intuition about the meaning of these boundary conditions may be helpful.
For this we appeal to the lattice realization of the minimal model as a generalized height
model on the graph G (see [35]). A boundary condition of the type (1, a) describes a fixed
boundary condition, where the height of the model is fixed to value a on the graph. The
interpretation of the other label r is less intuitive. The boundary condition (r, a) is realized
by attaching an r-times fused weight to height a.
The expression (2.37) for the cylinder partition function encompasses and generalizes
cases that were already known:
From the work of Saleur and Bauer [9] who discussed boundary conditions in lattice
height models of A-D-E type on a cylinder in which the heights on the boundaries
are fixed to the values a respectively b. They showed that in the continuum limit, the
partition function reads
X
Vsa b 1s .
Zb|a =
s
From the work of Cardy [10] who showed how to construct new boundary conditions
by fusion.
From the work of Pasquier and Saleur [58], who interpreted the pair of relations
X
(Ah1 ,G)
=
1s Vsa b ,
(2.41)
Z(1b)|(1a)
s
(Ah1 ,Ag1 )
Z(1,s)|(1,1)
= 1s ,
(2.42)
as expressing the decomposition of the representation of the TemperleyLieb algebra

on the space of paths from a to b on graph G onto the irreducible ones on the paths
from 1 to s on graph Ag1 , see point (ii) at the end of Appendix B.
Examples
Let us illustrate these expressions of boundary states by a few simple cases. In the
Ising model (the (A2 , A3 ) minimal model), h = 3, G = A3 , thus n = 3 and there are
1
2 (3 1) 3 = 3 boundary states, generally denoted [10] +, and f . On the lattice,
the first two describe fixed boundary conditions on the spin = 1 or 1 respectively,
while f corresponds to free boundary conditions.
It is then instructive to consider two related examples, see also [22,30]. The first is the
sl(2)4 at level 4, and the other is its cousin, the c = 4/5 minimal
c = 2 D4 solution of b
722
(3-state Potts) model, already mentioned in Section 2.4 and labelled by the pair (A4 , D4 ).
In the former case, we find four boundary states, labelled by 1 to 4, that we attach to the
nodes of the D4 diagram. All these states satisfy the required boundary conditions. The set
sl(2) model is also known to result from the
of exponents is E = {1, 3, 3, 5}. But this D4 b
sl(3)1 . Regarded as an b
sl(3) theory, the model admits
conformal embedding of b
sl(2)4 into b
three boundary states satisfying the more restrictive b
sl(3) conditions (Ln L n )|ai =
(Jn + Jn )|ai = 0, where the choice of corresponds to the diagonal set (j, j ). These
three boundary states may be regarded as the three nodes of a triangular graph A(4) (see
Appendix D and Fig. 11), or as the three extremal nodes of the D4 diagram that have
survived the additional b
sl(3) constraint.
The discussion of the Potts model is quite parallel. From the minimal model standpoint,
it is the (A4 , D4 ) model, h = 5, n = 4 and there are 8 boundary states [22,30]:
A = (1, 1) = (4, 1),
BC = (2, 1) = (3, 1),
ABC = (1, 2) = (4, 2),
B = (1, 3) = (4, 3),

AC = (2, 3) = (3, 3),
N = (2, 2) = (3, 2).
C = (1, 4) = (4, 4),

AB = (2, 4) = (3, 4),
(2.43)
On the lattice, the first three A, B, C describe fixed boundary conditions where the spin
takes at each site of the boundary one of the three possible values. The mixed boundary
conditions AB, BC, AC describe boundary conditions where the spin on the boundary can
take on two values independently. The boundary conditions ABC and N are free boundary
conditions but for N the weights depend on whether adjacent spins are equal or not.
The model may also be regarded as the simplest W3 model. In that picture, one may
impose more stringent boundary conditions. Only the six states denoted above A, B, C,
AB, BC, AC satisfy the additional condition (Wn(3) + W (3)n )|ai = 0. They correspond
to the extremal nodes of the pair (T2 , D4 ) or, alternatively, to the nodes of the pair
(T2 , A(4) ). As will be discussed in more detail in Section 3, the subset of these nodes,
to be denoted T , can be identified in both examples with the representation labels of the
j
corresponding extended chiral algebra. The matrix elements a for a T satisfy [63] the
relation
ext
j
Sa{j
a
}
p
=q
,
ext
S1j
S{1}{j }
where {j } denotes the orbit of the exponent j with respect to the Z2 automorphism and
ext is the modular matrix of the extended theory. This relation implies that
Sa{j
}
ext
X
Sa{j
}
q
|j ii,
ext
S{1}{j
{j }
j
{j
}
}
P
i.e., we can identify j {j } |j ii = |{j }ii with an extended Ishibashi state. The missing
boundary condition corresponds to a twisted boundary condition from the point of view of
the extended algebra.
We conclude that, as expected, the number and nature of the boundary states reflect the
precise conditions that they are supposed to satisfy.
|ai =
723
3. Graph fusion algebras

According to the discussion of Section 2, given a certain chiral algebra A, the sets of
complete orthonormal boundary states of RCFTs consistent with this algebra are classified
by representations of the Verlinde algebra of A, on matrices ni with non-negative integral
entries. Or stated differently: given a certain RCFT with a chiral algebra A, the sets
of complete orthonormal boundary states of this theory consistent with this algebra are
classified by representations of the Verlinde algebra of A, on matrices ni with non-negative
integral entries, with eigenvalues specified by the diagonal part E of the spectrum. These
matrices may thus be regarded as the adjacency matrices of a collection of |I| graphs.
In practice, it is sufficient to look at the smaller number of matrices representing the
generators of the fusion ring. For example, one matrix in the case of sl(2) considered
above, or the N 1 matrices associated with the fundamental representations, in the case
of sl(N).
The simplest case is given by the regular representation of the fusion algebra, when
the matrices ni are the Verlinde matrices themselves, ni = Ni . This is the case of so-called
diagonal theories, when all representations of the set I appear once in the spectrum Spec =
{(i, i)|i I}. This may be regarded as the case of reference from several points of view:
it was the first case analysed in detail [10]; the corresponding graphs are playing a central
role; and finally in that case, Cardy was able to provide a physical argument explaining
why the fusion matrices arise naturally. It is the purpose of this section to extend these
considerations to more general solutions. We shall find that the role of the fusion matrices
in the arguments of Cardy is now played by two sets of matrices. The first is the set of
matrices ni that describe the coefficients of the cylinder partition function; the second is a
ba , forming what is called the graph fusion algebra.
new set of matrices N
On the other hand, since we know that the cylinder partition functions, or equivalently
the matrices ni , contain some information about the bulk theory, through the knowledge
of the diagonal spectrum E, it is expected that this classification of boundary conditions
should have some bearing on the classification of bulk theories, namely on the classification
of torus partition functions and on bulk structure constants. Remarkably, this programme
works even better than expected and the two classification problems seem to be essentially
equivalent, at least for type I theories (see end of Section 1). This will be explained in
Section 3.3 below.
3.1. More on graphs and intertwiners
Suppose we have found a solution to the Cardy equation, namely a set of n n matrices
(2.16), (ni )a b , i = 1, . . . , |I|, a, b = 1, . . . , n. What was said in detail in Section 2 and in
Appendix B in the case of b
sl(2) can be repeated here. As their entries are non-negative
integers, these matrices may be regarded as adjacency matrices of a set of |I| graphs
Gi , with n = |Gi | |G| nodes. We shall refer collectively to these |I| graphs as the
graph G, whereas the basic solution provided by the N s themselves will be called the A
graph (borrowing the notation from the sl(2) case). The eigenvalues of the matrices ni are
724
specified by a set Exp(G) in the sense that they are of the form Sij /S1j , (j, ) Exp(G).
Moreover Exp(G) = E if the RCFT is given and the diagonal spectrum E is known. But
in general, the determination of the set Exp(G) is part of the problem. The fundamental
P
P
relation b (ni )a b (nj )b c = k Nij k (nk )a c may be interpreted in two ways:
Regarded as |G| |G| matrices, the matrices ni form a representation of the fusion
algebra (2.22).
Regarded as a |I| |G| rectangular matrix, each matrix n a for a fixed, (n a )j b := nj a b
intertwines the representatives Ni and ni in the two representations Ni n a = n a ni , or
more explicitly
X
X
Nij k nka c =
nj a b nib c .
(3.1)
k
We shall thus occasionally refer to the matrices ni as intertwiners.

The case of graphs and intertwiners pertaining to b
sl(2) theories has been discussed at
length in Section 2 and in Appendix B. In Appendix D, we outline the discussion of b
sl(3).
In that case, the fusion algebra is generated by two matrices n(2,1) and n(1,2) (labelled by the
two fundamental (shifted) weights of sl(3)), and as these two representations are complex
conjugate to one another, the matrices ni are related by transposition n(2,1) = nT(1,2) . Then
according to (2.22), ni is given by the same polynomial of n(1,2) and n(2,1) with integral
coefficients as that representing Ni in terms of N(1,2) and N(2,1) . It is thus sufficient to list
all possible graphs representing the matrix n(2,1) , provided one checks that all ni have nonnegative integral entries. In contrast with the case of b
sl(2), no complete solution is known
for b
sl(3). The current state of the art is presented in Appendix D with tables, figures and
relevant comments.
3.2. Graph fusion algebras
To see what is playing the role of the fusion algebra in the argument of Cardy, we
have to introduce the graph fusion algebra. The graph fusion algebra, as first discussed
j
by Pasquier [59], is a fusion-like algebra attached to a connected graph G. Let a be the
common orthonormal eigenvectors of the adjacency matrices G labelled by j Exp(G).
In general, these eigenvectors can be complex. In the case of degenerate eigenvalues
the associated eigenvectors need to be suitably chosen. We assume that the graph has a
j
distinguished node labelled 1 = 1 such that 1 > 0, for all j Exp(G).
One then defines the numbers
bab c =
N
a b (c )
j Exp(G)
(3.2)
ba )b c = N
bab c satisfy N
bab c = N
bac b and NaT = Na .
ba with elements (N
and the matrices N
ba has a single non-vanishing
b1 = I . Since each matrix N
Because of orthonormality, N
b1 )a b = ab , the matrices N
ba are linearly
ba )1 b = (N
entry in the row labelled 1, namely (N
c
b
independent. The Nab are the structure constants of the graph fusion algebra satisfied by
b matrices
the N
ba N
bb =
N
bab c N
bc
N
725
(3.3)
which is an associative and commutative algebra. Of course, if the graph G is of type A,

this boils down to the ordinary Verlinde fusion algebra since the matrix of eigenvectors
is nothing but the modular matrix S.
ba and ni matrices have the same eigenvectors, it is easy to derive the matrix
Since the N
relation
X
ba =
bb .
nia b N
(3.4)
ni N
b
b1 = I , and all ni appear as linear combinations with non-negative

In particular for a = 1, N
bs
integer coefficients of the N
X
bca b .
ni1 c N
(3.5)
nia b =
c
Alternatively (3.4) may be used as a starting point to reconstruct the graph algebra, as
explained in Section 3.5.
It should be stressed that the definition of a graph fusion algebra is not unique. In
general, it depends on the choice of the distinguished node 1 and, when there are degenerate
j
eigenvalues, also on the choice of the eigenvectors a . To view the graph fusion algebra as
bab c to be non-negative
a proper fusion algebra we would like the structure constants N
integers. But even the rationality of these numbers is not obvious and it is therefore
surprising that, for appropriate choices of the s and of node 1 and for most cases, they
turn out to be integers of either sign. Among all the examples known to us in b
sl(N) theories,
(12)
2 6 N 6 5, it fails in only two cases: the graph called E5 in Fig. 12, for which there is no
b algebra involves fractions of denominator 4; and a
node 1 satisfying 1 = 1 , and whose N
bab c of either sign occur. Adopting
graph in the b
sl(4)4 theory, [63], in which half-integer N
(2.22), (3.4) in the framework of subfactors theory the latter example has been reinterpreted
by Xu [65] by trading commutativity of the graph fusion algebra for integrality.
b is only possible for certain graphs which we call
Finally the non-negativity of the N
proper fusion graphs. For example, for the sl(2) theories, the A-D-E graphs that admit a
proper graph fusion algebra are
proper A-D-E graphs = An , D2q , E6 , E8 .
(3.6)
The choice of distinguished node for the sl(2) A-D-E graphs is explained in Appendix B.
We note that the set of proper sl(2) fusion graphs matches the modular invariant
partition functions listed as type I at the end of Section 1. The situation is somewhat
different for sl(3) graphs, for which we have to introduce a further distinction. In this
bs are not associated with type I theories. We
case, some graphs with non-negative N
reserve the terminology type I graph for those graphs associated with type I theories
(see Appendix D and Tables 23). Moreover, as is clear from the b
sl(4)4 example above,
some type I modular invariant partition functions are associated with graphs with nonbab c . In the following, we discard these exceptional cases and
integer and/or non-positive N
restrict ourselves to type I graphs that are associated with type I RCFTs. In general, the
726
question of precisely which graphs admit type I fusion algebras should be related to the
classification of type I RCFTs, and thus is a very interesting open question.
3.3. Fusion rules and block characters
Given a solution to the Cardy equation, that is a set of partition functions
X
nia b i (q)
Za|b (q) =
(3.7)
iI
and the corresponding graphs, we assume as in Section 3.2 above that there exists a special
node called 1. We then introduce the combinations of characters (or block characters)
X
n ic i (q),
(3.8)
c (q) =
iI
where
n ic = ni1 c
(3.9)
is referred to as the basic intertwiner. Thanks to (3.5), (3.7) may be rewritten as

X
bca b c (q),
N
Za|b (q) =
(3.10)
where the coefficients are now given by the structure constants of the graph fusion algebra
of G.
Equation (3.10) is a mathematical identity and as such is valid and consistent
independent of the choice of the distinguished node and eigenvectors of G. Physically,
bab c are non-negative integers is the most interesting. In that
however, the case where the N
bab c gives the number of times
case, following Cardys discussion [10], it is suggested that N
that the propagating mode or representation c appears in the strip or cylinder with boundary
bab c are nonconditions a and b. Thus if G is a type I graph, i.e., if the structure constants N
negative integers, we have a possible interpretation: the nodes a of the graph(s) G under
consideration label a class of representations of some extended chiral algebra. The blocks
bab c are their fusion coefficients.
a are their characters, and the integer coefficients N
To probe this interpretation, let us see how it confronts the results in the bulk, in
particular how it is consistent with the form of the torus partition function. There, it has
been observed already long ago that (for type I theories) the torus partition function (cf.
(1.5)) may be recast in the form
X
| a |2 ,
(3.11)
Ztorus =
aT
i.e., as a diagonal sum over a subset T of block characters. The subset T corresponds
b algebra, in the sense that if a, b T , N
bab c 6= 0 only if c
to a subalgebra of the N
c
T . This interpretation of ni1 as a multiplicity of representation i in the block c, that
was first observed empirically [60,61], was subsequently derived in a variety of cases of
type I sl(N) theories based either on conformal embeddings or on orbifolding [63,64].
More recently, it appeared as an important ingredient in the investigation of the algebraic
structure underlying these theories [6568].
727
The following interpretation is thus suggested. The nodes a T label representations

of the maximally extended algebra A0 of the RCFT (of type I) under consideration. The
b algebra is the conventional fusion algebra of this RCFT. The other
subalgebra of the N
b algebra would
nodes a 6 T might label other twisted representations. The entire N
0
describe the fusion of all, twisted and untwisted, representations of A . This interpretation
in terms of twisted representations seems corroborated by the fact that some a are
known also to occur in partition functions on a torus in the presence of twisted boundary
conditions. The fact that general boundary conditions on a cylinder also appeal to these
representations was first observed in the Potts model in [22]. See for some work in this
direction [69], and the more systematic developments [3133] along the lines of [74]. The
concept of twisted representations in other cases, like conformal embeddings, remains to
be understood.
Having discussed the situation for type I theories and graphs, we return to RCFTs of type
II. There the situation is more elusive. On the one hand, as discussed above, the boundary
conditions on a cylinder are labelled by nodes of an improper graph G, and although we
can still write an expression of the form (3.10), its physical interpretation is unclear. On the
other hand, from (1.6), we know [84] that the torus partition function may be expressed in
terms of block characters pertaining to a parent type I theory with graph G0
X

Ztorus =
a (q) (a) (q) ,
(3.12)
aT
b
where T is once again a subset of the nodes of G0 corresponding to a subalgebra of the N
(c)
c
b
b
= Nab .
algebra, and is an automorphism of that subalgebra N(a)(b)
For example, the b
sl(2)16 theory labelled by the Dynkin diagram E7 is known to be
related in that way to the D10 theory. Their respective torus partition functions read
(D )
10
= |1 + 17 |2 + |3 + 15 |2 + |5 + 13 |2 + |7 + 11 |2 + 2|9 |2
Ztorus

X
X (D10 ) 2

n ia i ,
=

a=1,3,5,7,9,10
aD10
(3.13)
(3.14)
(E )
7
= |1 + 17 |2 + |5 + 13 |2 + |7 + 11 |2 + |9 |2
Ztorus

+ (3 + 15 )9 + c.c.
X
X

X
(D10 )
(D10 )
n ia i
n i(a) i ,
=
a=1,3,5,7,9,10
aD10
(3.15)
(3.16)
with exchanging the two nodes 3 and 10 of the D10 diagram.

It seems that the parent graph G0 also plays a role for cylinder partition functions of
type II theories. Indeed, to obtain cylinder partition functions expanded with non-negative
coefficients in terms of block characters, we just have to expand in the block characters of
G0 . Specifically, we find
X
(G)
(GG0 ) b (G0 )
nca
c (q),
Zb|a (q) =
cG0
728
Fig. 3. The torus partition function reconstructed from two cylinder partition functions.
where the G-G0 intertwiners are given by

(GG0 ) b
nca
c(G ) m
(G0 ) m
mExp(G) 1
(G) m
a(G) m b
.
0
(GG ) b
> 0 and satisfy the G0 graph algebra.
These turn out to be non-negative integers nca
Here it is assumed that the distinct exponents of Exp(G) are in Exp(G0 ) and that the
sum is over exponents of G counting multiplicities. Moreover, if there is more than one
eigenvector of G0 corresponding to m Exp(G), then any of these eigenvectors can be
matched with the given m Exp(G). The formula can be derived in the same way as our
previous formulas. In particular, this formula applies for G = E7 and G0 = D10 . In this
case there is an ambiguity as to which D10 eigenvector is taken for m = 9 but in fact one
can take either. The matrices are changed by the Z2 symmetry but the cylinder partition
(12)
(12)
functions agree. The formula also holds for the type II sl(3) theories G = E2 , E4 and
E5(12) and G0 = D(12) .
Putting everything together, we finally observe that in general for a rectangular torus
with two periods 2L and iT , made by pasting together two cylinders (see Fig. 3),
X (G0 ) (G0 )
(G)
Za|1 Z(a)|1,
(3.17)
Ztorus =
aT
i.e., the partition function may be obtained as the sum over a special set of boundary
conditions of cylinder partition functions. This expression is of course deeply rooted in
all the connections between bulk and boundary theories, open and closed strings, etc., but
still we find its simplicity intriguing.
3.4. Examples
More examples can be given to the previous general scheme.
b
sl(N): the classification of the representations of the fusion algebra of b
sl(N)k is a
well posed problem on which we have only partial results. In particular, classes of
graphs pertaining to b
sl(3) as well as some cases for higher N have been expounded
from various standpoints in [46,6064,70] (see Appendix D). In all known cases, the
previous discussion may be repeated: intertwiners, type I graphs, and other concepts
introduced above, still apply. We refer the reader to the above references.
sl(N)1 are
The case of b
sl(N)1 may be described in detail. The representations of b
labelled by an integer 0 6 i 6 N 1 (we depart here from our previous convention,
729
with i = 0 denoting the identity). The fusion rules are isomorphic to the addition
of integers modulo N , and the algebra defined by Ni Nj = Ni+j mod N is thus
generated by N1 , Nj = (N1 )j . The eigenvalues of Nj are exp(2ij l/N). The regular
representation is provided by N N matrices, generated by (N1 )i j = j,i+1 mod N .
All the previous eigenvalues are reached once and we may thus say that the system
has exponents l = 0, . . . , N 1 (mod N).
In general, a representation {ni } of the fusion algebra is associated with each divisor
q of N = p.q, including q = 1 and q = N : q denotes the order of the matrix n1
which is (q q)-dimensional and such that (n1 )a b = ba,1 mod q , for a labelling
of the nodes a, b = 1, . . . , q. (This exhausts all integer-valued representations of the
algebra. Indeed the conditions that nT1 = nN1 and n1 nN1 = I imply that the only
entries of n1 are 0 and 1, and that n1 is a permutation matrix. Being of order q
and indecomposable, n1 is a matrix of a cycle of length q. Q.E.D.) Obviously the
matrices ni = (n1 )i are all integer-valued, n0 = I , and nTi = ni = nNi . The graph
of adjacency matrix n1 is an oriented q-gon. In that case, we may say that the q
exponents are 0, p, . . . , p(q 1).
This census of representations of the b
sl(N)1 fusion algebra matches almost perfectly
that of modular invariant partition functions carried out by Itzykson [71] and
Degiovanni [72]. We recall that according to these authors, a different modular
invariant is associated with each divisor of n, where n = N if N is odd and n = N/2
if it is even. Thus, only the case N even, q = 1 has to be discarded in our list of
representations of the fusion algebra, as it does not correspond to a modular invariant.
3.5. More on graph algebras
b by (3.2) and derived (3.4). Instead of
In Section 3.2, we have introduced the matrices N
b,
looking at the graph as a collection of points we can look at it as a collection of matrices N
providing a basis of a commutative, associative algebra with identity, and an action of the
intertwiners ni given by (3.4), that is, we take (3.4) as a starting point. Given the graph G,
in particular the coefficients ni1 c , it is possible in many cases to invert (3.5) and solve for
ba as linear combinations with integral coefficients of the intertwiners ni , or equivalently,
N
as polynomials of the fundamental adjacency matrices. Similarly the relation (3.4) written
in terms of the eigenvalues j (i) = Sj i /S1i , a (i) = ai /1i
X
j (i) a (i) =
nj a b b (i), i E, j I,
(3.18)
b
is a recursive relation determining (the rows of) the eigenvector matrix ai . In general,
typically in the presence of degenerate eigenvalues, the matrix ai is not determined
uniquely, or alternatively, (3.5) cannot be inverted for all a V. For the type I cases,
however, as explained above, there exists an extended fusion algebra isomorphic to a
can be identified
subalgebra of the graph algebra, so that the extended fusion matrices NBext
i
b
with a subset Na with the nodes a T V [60,61]. In most of these type I cases one
ba in terms of the ni s and N ext , or alternatively express aj in terms
can solve for all N
Bi
730
of the modular matrices Sij , SBexts Bl . A particularly simple subclass of Type I for which
one can go quite far in the programme of reconstructing the graph G and all the related
structures is presented by the orbifold theories, in particular the ones associated with groups
generated by simple currents. In our approach they can be described by graphs obtained by
orbifolding the fundamental graphs of the initial diagonal theory, the simplest example
being provided by the WZW sl(2) D2l series obtained by orbifolding the Dynkin diagram
A4l3 over the Z2 group generated by the automorphism . In these cases as well as in
their sl(N) generalisations defined in [73] involving the group ZN , one can algorithmically
construct the eigenvector matrix, see Appendix B for an illustration in the simplest N = 2
case. In a different approach, using tools similar to the original orbifold treatment of [74],
an elegant general formula for the eigenvector matrix was derived recently in [3133]. It
should be noted that the same graph (orbi)folding procedure leads also to type II graphs,
e.g., the sl(2) Dodd series, or their sl(3) generalisations for k 6= 0 mod 3, see Appendix D.
We have assumed up to now in this discussion that the graphs are already known. On
the other hand the relations (3.4), (2.22) can be taken as the starting point for finding new
graphs, typically exceptional graphs not covered by the previous orbifold constructions.
Since any graph in the vicinity of the identity resembles the original diagonal (A) graph,
one can first try to identify ni s for which the r.h.s. of (3.5) reduces to one term, i.e.,
bai . According to (3.18) this also determines
ni1 a = aai and hence one can identify ni = N
j
j
ai by i (j ) once 1 is known. This is a problem which is reduced to the computation of
some Verlinde fusion multiplicities. Indeed let us take the first matrix element a = b = 1
of the matrix relation (2.22) we have
X
X
X
ni1 c nj 1 c =
Nli j nl1 1 =
Nli j ,
(3.19)
c
where in the last sum = {l I|nl1 1 6= 0} and l is counted nl1 1 times. Let us assume
first that in (3.19) i = j . Whenever the r.h.s. of (3.19) is equal to 1, since by definition
ni1 a are integers, the l.h.s. summation reduces to one term, i.e., we recover ni1 a = aai .
Furthermore plugging this into the l.h.s. of (3.19) taken for j 6= i we recover nj 1 ai as being
given by the sum of Verlinde fusion multiplicities in the r.h.s. of (3.19), i.e., we determine
the multiplicity with which Nai appears in nj , see (3.5). Similarly, a value 2 or 3 for the
r.h.s. of (3.19) with i = j would lead to 2, respectively 3 terms in (3.5), while 4 could be
interpreted either as leading to 4 terms with multiplicity one, or 1 term with multiplicity
two, i.e., ni1 a = 2aai . What we only need in order to check all these possibilities is to
know the content of the set , i.e., ni1 1 . This data is provided in type I theories for all of
which encodes the content of the identity representation of the extended algebra. More
generally, ni1 a = multBa (i), identifying a with a representation Ba of the extended algebra.
The relation (3.19) and its consequences just described are the first steps in a consistent
algorithmic procedure proposed by Xu [65] in the abstract framework of subfactors theory
(see also [6668] for further developments). In particular, the subset of ni which can be
bai are related to irreducible sectors with the sum in the l.h.s. of
identified with some N
(3.19) interpreted as a scalar product (ai , aj ). The algorithm reduces systematically the
ba in type I cases to data provided by the Verlinde fusion matrices
determination of ni , N
731
N and N ext . All graphs previously found in [46,60,61,63] were recovered in [65] by this
b 6 modular invariant was found in
method and a new example corresponding to the sl(4)
[64].
Finally let us point out that to some extent this algorithm for finding new solutions of
the equations (3.4) and (2.22), i.e., new graphs, can be applied to type II cases where we
do not know a priori ni1 1 , i.e., the set . One can start with some trial set and compute
P
P
P
(j,) 2
j
) = S1j l Slj . A first consistency check is that j E (1 )2 = 1. Then one
(1
can proceed as in type I. For example the E7 Dynkin diagram may be reconstructed using
= {1, 9, 17}. 3 Different (consistent) choices of the set might lead to the same graph,
reflecting the possibility of different choices of the identity node.
In some simple cases it is possible to recover a complete set of boundary conditions by
applying formula (3.4) to a known subset in such a way that only one term appears in the
sum in the r.h.s. In terms of the equivalent formula (3.18) for the eigenvalues we obtain
a new solution bi , b = b(a, j ), by fusing a given one ai with the Verlinde eigenvalue
j (i). This seems to be the idea of the so-called Affleck fusion conjecture [17], which
clearly has a restricted application, with the general formulae (3.4), (3.18) being the correct
substitute for it.
Another approach to constructing Type I graphs was discussed in [63] and used to find
new solutions for higher rank cases. It is based on the use of a relation for the structure
constants of the Pasquier algebra, the dual of the graph algebra, with structure constants
labelled by elements in the set Exp(G) and given by a formula analogous to (3.2), however
with the summation running over the nodes of the graph; this algebra will be discussed
further in Section 4.4.1 below.
4. Bulk and boundary operator algebras

In this section we investigate the algebras of fields in the presence of boundaries and the
equations for their structure constants resulting from duality constraints. Our discussion
parallels that of Cardy and Lewellen [11], but generalises it in two respects: to higher rank
and to non-diagonal theories. This results in additional multiplicities associated with the
more general representations of the Verlinde algebra (2.22). Our presentation makes use
of concepts used by Moore and Seiberg for bulk RCFTs and extends them appropriately
for this new setting. This leads to a richer structure in the equations and the appearance of
ba , Mj ). Separately these algebras have appeared before but the
a triplet of algebras (ni , N
inter-relation between these algebras has not been shown in this context.
4.1. Ground state degeneracies
As stressed by Affleck and Ludwig [16], the logarithm of the partition function, in the
limit L/T , contains not only the universal term proportional to L and to the central
3 Note added in proof: this was independently discussed in the recent paper [98], see also [84]. The paper [98]
provides a systematic approach in the framework of the subfactors theory to both types of modular invariants.
732
charge (in unitary theories), but also an L independent term, interpreted as a boundary
condition dependent ground state degeneracy ln ga gb . Indeed in that limit
L
c
4 + ln a1 + ln b1 ln S11 ,
(4.1)
ln Zb|a
24
T
where as before we denote by 1 the representation of conformal weight 0, corresponding
to the identity operator. We therefore identify
1
(4.2)
ga = a .
S11
Thus in unitary minimal models, using (2.38) and (2.28), we have the following
expression for the boundary states
0
|(r, a)i =
r 0 ,s 0
r 0 odd, s 0 Exp(G)
1/4
S (h)0 s
q rr a |r 0 , s 0 ii
(h) (g)
S1r
0 S1s 0
(4.3)
and their g factor

(h)
S 1
hh1, 1|(r, a)i
= 21/4 q r1 a ,
g(r,a) = q
(h) (g)
(h) (g)
S11
S11
S11
S11
(4.4)
sl(2) at levels h 2 and g 2, |g h| = 1.

in terms of the modular matrices S (h) and S (g) of b
For example, for the critical 3-state Potts model, we obtain

5 5 1/4
= 0.550936,
gA =
30
1 + 5 1 + 5
: 3:
3
gA : gAB : gABC : gN = 1 :
2
2
(h)
(h)
in agreement with [22]. As a particular case of (4.4), the ratio g(r,a)/g(1,a) equals Sr1 /S11 ,
in agreement with (A.3) of [22] and the fact that one obtains the boundary state |(r, a)i by
fusion (in the sense of Cardy) of boundary states (1, a) and (r, 1).
In non-unitary cases, these expressions have to be slightly amended. If j0 denotes the
representation of smallest conformal weight hj0 < 0 and assumed to belong to E, then
L
ceff
j
j
4 + ln a0 + ln b0 ln S1j0
24
T
with ceff := c 24hj0 . Also
ln Zb|a
(4.5)
1
ga = p a .
S1j0
(4.6)
For simplicity of notation in most of what follows we shall restrict to unitary theories.
Denoting ga = h1ia from now on,
lim
L/T
Zb|a e
c L
6 T
/gb = h1ia .
(4.7)
One can consider furthermore the partition function with some field insertions at the same
limit [11,26,27]; we shall normalise them similarly so that only a dependence on ga is
733
retained, i.e., h1ia will coincide with the 1-point function of the identity operator in this
limit.
4.2. Bulk and boundary fields, OPE
4.2.1. Boundary fields
According to Cardy [10], boundary conditions can be interpreted as created by the
a
(x) living on the boundary, Im z = 0, x = Re z of the upper
insertion of fields b j,
half-plane z H+ . Here j I, a, b V, and accounts for the multiplicity nj a b of
such fields, to be called of type
= 1, 2, . . . , nj a
b
j a
. Thus can be interpreted as a coupling index
and the boundary fields as

of chiral vertex operators (CVO)
kind
b
, a, b V, j I. This is a formal
j a
associated with a second type of couplings
analogy since the boundary states |ai, |bi labelled by a, b are superpositions of Ishibashi
states. The multiplicity index is traditionally omitted, but it should be stressed that even
in the sl(2) case, in all but the diagonal cases (E = I), there are always some nontrivial
we shall retain this index. Since n1a b = ab
multiplicities nj a b > 1, so most of the time

the index associated with the coupling b1 b takes only one value and will be denoted
= 1b , or just 1, or, altogether omitted. On the other hand in all non-diagonal cases there
is a nontrivial subset {1 j1 , j } of boundary fields, with 1 I , where the set has
been introduced in Section 3.5.
To make contact with Section 2.2, consider the finite strip w = L
log z equipped with the
(H )
c
Hamiltonian Hab = L (L0 24 ). The space of states is generated by all the descendent
a (0)|0i by
states created for fixed a, b from the (properly normalised) vacuum state b j,
the modes of the Virasoro algebra generating the real analytic conformal transformations.
This includes besides the sum over j I a summation over the multiplicity nj a b of these
states for fixed a, b, j , i.e., we can think of the Vir representation spaces Vj, as being
labelled by pairs (j, ). The dual vacuum state is defined by a boundary field placed
P
at infinity 0 limx ca,b,j ;, 0 x 2j h0|a jb , 0 (x), where ca,b,j ;, 0 is a normalisation

constant and 0 = 1, . . . , nj b a is an index of type ja b . Accordingly the trace of the
operator eT Hab computed imposing the periodicity w w + T in time direction can be
written as a sum over (j, ) of characters j, , with the summation over leading to
(2.15), with a and b exchanged.
a (x) appear as ordinary fields (x) decorated by a pair of indices
Boundary fields b j,
j
a, b according to some rules. In the limit L/T their 1-, 2- and 3-point functions are
given by the corresponding invariants of j with respect to sl(2, R) with normalisation
coefficients depending on a, b. As for the ordinary fields the 1-point function is non-zero
only for the identity operator

b a
(4.8)
0 j, (x)0 = j 1 ba 1a h1ia
with the restriction on a, b coming from n1a b = ab . As for the ordinary CVO the product
b c d a
i,1
j,2 of two boundary fields is defined only for coinciding c = d. Similarly the
734
initial and the final indices in a vacuum expectation value of a product of boundary fields,
are restricted to coincide (due to the periodicity in the strip time direction, the boundary
half-line being effectively closed) but in distinction with the ordinary CVO they can be
arbitrary and not just equal to the identity 1. The 2- and 3-point functions read
Ciab

a b
i;
c
1 2

0
=
(x
)
,
0 j,1 (x1 ) b i,
2
j i ac
2
|x12|2j
x1 6= x2 ,
(4.9)
a b
c
d
(x2 ) c k,
(x3 )0 t
0 i,1 (x1 ) b j,
2
3
= ad
Cijabc
k;1 2 3 ;t
k
|x12|ij |x23|jk |x31|ki
x1 6= x2 6= x3 6= x1 ,
(4.10)
where kij = i + j k and xij = xi xj .

The functions (4.9), (4.10) are invariant with respect to SL(2, R), with representations
denoted by a pair (, = ), see [51]; here we choose = 1 corresponding to taking the
modulus of the multiplier of the SL(2, R) transformations and the expressions in (4.9),
b k WZW
(4.10) imply trivial monodromy of the boundary field correlators. In the sl(n)
models the fields carry an additional tensor index, or, in a functional realisation, depend
on an additional (multi)variable X accounting for the representations of the isospin
sl(n, R) algebra and the n-point functions involve also n-point invariants with respect to
b k WZW case the fields j (x, X) can be described
this algebra. For example, in the sl(2)
in terms of a pair of real variables [79], the coefficients in the polynomial expansion
with respect to X representing the horizontal algebra descendants. In this case the isospin
labels are 2j = 0, 1, . . . , k, and the 2- and 3-point invariant correlators contain additional
2
2j
factors Xil along with any xil j . For simplicity we adapt the notation for the minimal
Wn models rather than their WZW counterparts, omitting the explicit indication of the
isospin variables and the corresponding invariants. To keep track of the various possible
three-point invariants, we shall retain the multiplicity index t as in (4.10).
Up to the normalisation constant and up to phases, (4.10) is the 3-point invariant function

h0|i,10i j,t k,1k |0i of the ordinary CVO j,t (x). Here t is a coupling index of type ij k ,
t = 1,2, . .. , Njk i . Two kinds

of permutations
act on these couplings, see [37,38]: 23 :

p
p
p
j i , and 13 : i j ji p . For simplicity in the sequel we denote the
i j
0
one value indices indicating couplings with
one label
of type I set
to 1, like 23 (1i ), 1i ,
i
1
a
or, 1a (corresponding to couplings of type i 1 , i i , or 1 a , resp.), simply by 1.
Motivated by the form of the 3-point function, the operator product expansion (OPE) of
c
(x) is defined according to
(primary) boundary fields b i,
b
c
a
i,
(x1 ) c j,
(x2 )
1
2

(1)
Fcp
p,,t
X
p,,t

(1)
Fcp
i
b
i
b
j
a
j
a
1 2
1 2
a
hp, P |i,t (x12 )|j, 0i b p,;P
(x2 )
1
b a
p, (x2 ) + , (4.11)
|x12|i +j p
735
Fig. 4. Graphical representation of (4.13). To stress the presence of two types of vertices, we
distinguish them explicitly on this particular figure only.
where P is an index for the descendent states of the representation

Vpwith
p I. The

b
c
b
,
indices 1 , 2 , account for the multiplicity of vertices of type i c
j a
p a ,
respectively,
a, b, c V, i.e., 1 = 1, 2, . . . , nic b , etc., while t is that of a standard vertex

p
p
i j , t = 1, 2, . . . , Nij . We will often restrict for simplicity to the sl(2) case, so that the
index t can be omitted. From the 1-point function

(1)
Fcp

(1)
Fcp
i
b
1
a
1
b
j
a
= pi ac 2 1 t 1 1 ,
1 2
t
= p j b c 1 1 t 1 2 .
(4.12)
1 2

With the normalisation of the CVO pi j implied by the second equality in (4.11) the
t

t
i
j
with i, j, p I, a, b, c V, represent the OPE
numerical coefficients (1)Fcp
b a 1 2
coefficients of the boundary fields and their determination is part of the problem. They
are reminiscent of the matrix elements of the fusing (or crossing) matrices F (whence the
notation here for this second fusing matrix), which serve as OPE coefficients of the usual
CVO [37,38]. The definition (4.11) extends to descendent fields in the l.h.s. as for the usual
CVO. Symbolically (4.11) can be written as

b
i c

1 , x1
c
j a

=
2 , x2

(1)
Fcp
p,,t
i
b
j
a
1 2
p
i j
t, x12
b
p a

(4.13)
, x2
and can be depicted similarly as the standard MooreSeiberg

diagrams, see Fig. 4.
b
b the space of boundary fields of type
b =
we have dim Upa
Denoting by Upa
p b
p
npa b while the space of standard CVO, Uij , has dimension given by the Verlinde fusion
p
multiplicity Nij p , dim Uij = Nij p . Thus we can interpret (1)F as a linear operator
M
c
b
Uic
Ujca
b
Uij Upa
,
(4.14)
the dimension of the two sides being identical, according to (2.22). Given the 1- and 2point correlators above the computation of the general boundary field n-point
is

functions
.
reduced to the computation of the conformal blocks of the standard CVO pi j
t, x
736
(a)
(b)
Fig. 5. (a) and (b): boundary field 2- and 3-point functions.
Comparing with the 2- and 3-point functions, see Fig. 5(a) and (b), we have
1 10

i i a i
ab
(1)
h1ia ,
Cii ;1 2 = Fb1
a a 1 2
Cijabc
k;1 2 3 ;t =

(1)
Fbk

(1)
Fci
i
a
j
c
j
b
k
a
23 13 (t )

(1)
Fc1
1 2
t

(1)
Fb1
2 3
i
a
k
a
i
a
1a 10
k
a
(4.15)
h1ia
1a 10
h1ia .
(4.16)
The 2- and 3-point normalisation coefficients are assumed to satisfy the symmetry
conditions

ba
ab
Ciab
i; = Cii ; = Ci i; ( ) ( ) ,
13
13

bca
cba
Cijabc
k; 1 2 3 ;t = Cj ki; 2 3 1 ;13 23 (t ) = Ck j i ;13 ( ) 13 ( ) 13 ( );13 (t ) . (4.17)
3
The first equalities in (4.17) are cyclic symmetry relations, see Fig. 6, while the second
equalities come from an antilinear (CPT) transformation, which in particular sends the
c
(x) to its conjugate c jb ,13 ( ) (x) with multiplicity indices consistent with
field b j,
nj a c = nj c a = nj a c .
The cyclic symmetry relations imply
1 10

j j a j
j
(1)
Fc1
h1ia = (1)Fa1
a a 1 2
c
j
c
1c 10
j
2 1
h1ic ,
(4.18)
737
Fig. 6. The cyclic symmetry of 3-point functions.

(1)
Fak
j
b
s
c

(1)
bs
t

(1)
Fc1
k
c
j
a
k
b
23 13 (t )
1b 10
k
b
h1ib

(1)
Fc1
s
a
s
a
1a 10s
h1ia ,
(4.19)
while the the second equalities in (4.17) lead to

( )23 (t ) !

t

j s
s
j 13
(1)
(1)
Fak
=
Fak
.
c b 13 ( ) 13 ( )
b c
(4.20)
Combining (4.18) and (4.19), one recovers (4.16).

4.2.2. Bulk fields and bulk-boundary coefficients
We turn to the second ingredient of the CardyLewellen boundary CFT, the bulk fields.
The half-plane bulk fields I (z, z ), z = x + iy H+ , z = x iy transform under a
representation of L(H ) [42,43] realised by differential operators
LH
n = Ln (i , z) + Ln (i , z )
(4.21)
of weights. In cases when there is more than one field

and characterised by a pair I = (i, i)
) is needed, but usually

with the same labels (i, i) a more involved notation like I = (i, i;
omitted for simplicity. For type I theories as well as for arbitrary scalar fields both i, i E,
while in general i, i I.
(H )
The invariance with respect to the subalgebra spanned by L1,0 determines the 1-point
a
) as well as the 2-point function ha p,
function of (i,i)
(z, z
i, e.g.,
(i,i)
a
(x1 ) (i,i)
h0|a p,
(z, z )|0i
a
Cp,(i,
i),,t
(z z )i +i p (x1 z)i +p i (x1 z )i +p i
x1 > Re z,
(4.22)
a a
while h(i,i)
p, i is defined for Re z > x1 by the analogous expression with x1 z,
x1 z replaced by z x1 , z x1 . Requiring the symmetry of this function under the
exchange of the two fields, i.e., the independence of the ordering, leads to the constraint
i i Z.
z)|0i,
The r.h.s. of (4.22) is the 3-point block of standard CVO h0|p,10p (x1 ) i,t (z) i,1
(
i

p
p
with t a coupling index of type i i , t = 1, 2, . . . , Ni i . Consistently with this the
(primary) bulk field can be represented for small z z via the decomposition
738
(i,i)
) =
(z, z
X
a,,pI ,t
a,,pI ,t
a,,pI ,t
a, p,t
B(i,i)
0i a a
hp, P |i,t (z z )|i,
z)
p,;P (
P
a, p,t
B(i,i)
a, p,t
B(i,i)
1
a a
p, (z) +
(z z )i +i p
1
(2iy)i +i p
a
p,
(x) + ,
(4.23)
a (z) are unphysical generalised CVO obtained

which extends to descendents. Here a p,
extending to the (full) plane the boundary fields of the previous section. Their OPE is
determined by the same fusing matrix (1)F , i.e., as in (4.11), the latter extended to complex
arguments zi , Re z12 > 0, with |x12 | replaced by z12 ; we shall need only this fusing
property.
p,t
The constants a,B(i,i)
(bulk-boundary reflection coefficients) in this decomposition

p
depend on two couplings of different types, i i and pa a . Note that the coefficients
t
used here differ by a phase from the traditionally normalised coefficients [11,12], which
p,t
will be denoted a,B(i,i)
(CL), i.e.,
a, p,t
B(i,i)
= ei 2 (i +i p ) a,B(i,i)
(CL).
p,t
The decomposition (4.23), symbolically written as

X
p
a
a, p,t
) =
B(i,i)
,
(i,i)
(z, z
i i t, zz p a , z
a,,p,t
(4.24)
(4.25)
see Fig. 7, reduces the computation of the n-point functions of (i,i)

to the computation
b (z), which combined with their OPE (the
of the blocks of the generalised CVO a p,
extension from |x12| to z12 of (4.11)) allows to recover all correlators in terms of standard
conformal blocks. The invariant 1-point function projected onto the boundary state a reads
(i,i)
(z, z )
a,1B 1,1
(i,i)
= i i
a
2
(z z ) i
h1ia .
(4.26)
Omitting the trivial indices and simplifying the label (i, i ) to i, one has in particular
aB 1 = 1 for any a.
1
The OPE of the half-plane bulk fields (k,k)
(z, z ) is defined according to
(k,k)
(z1 , z 1 ) (l,l)
(z2 , z 2 )
X (j,j);t,t X

D(k,k)(l,
hj, J |k,t (z12 )|l, 0i j, Jk,
z12 )l, 0 (j,j);(J,J) (z2 , z 2 )
=
t(
l)
j,j,t,t
J,J
(j,j);t,t
D(k,k)(l,
l)
j,j,t,t
z12kl z 12kl
(j,j);t,t
(j,j) (z2 , z 2 ) + .
(4.27)
The coefficients D(k,k)(l,

l)
are related to the full-plane bulk OPE coefficients, see below.
739
Fig. 7. Graphical representation of the decomposition (4.25) of bulk fields.
4.3. Boundary CFT duality relations

4.3.1. CardyLewellen equations rederived
We collect in this section the set of equations resulting from the sewing constraints on
the various OPE expansions [12]; some of these equations can be interpreted as expressing
locality (symmetry) of the boundary CFT correlators. For simplicity of notation
we shall

i
sometimes omit the explicit indication of the coupling indices of type j k and the
corresponding summations, i.e., the equations will be written essentially for the simplest
sl(2) case. However we shall keep the charge conjugation in the indices of I so that the
general formulae can be easily recovered. In the sl(2) WZW case the braiding phases
Sug
Sug
are given by the shifted scaling dimensions j j , instead of j (since the pair
of coordinate and isospin variables is moved as a whole). Then formulae work equally
well with the same fusing and braiding matrices, without additional signs, as for the
corresponding subfamily (1, 2j + 1) of fields in Virasoro minimal models.
Applying (4.11) in different ways to the 4-point function of boundary fields , that is
demanding associativity, leads to a relation connecting the two types of fusing matrices
(1)F and F , the fusing matrix for the ordinary CVO, which reads symbolically
F
(1)
F = (1)F
(1)
(1)
(4.28)
or, more explicitly,

u u

t

i j 2 3 (1)
i m 1 2 (1)
j
Fbl
Fcm
l k t2 t3
a d 1 2
b
m, 2 ,t3 ,t2

u

u
X
p k 1 2 (1)
i j 1 3
(1)
=
Fcl
Fbp
.
a d 1 3
a c 1 2
X
Fmp
k
d
2 t3
2 3
(4.29)
The identity (4.29), when restricted to the sl(2) case, is a slightly simplified version of the
equation (L 3.29) in [12] and can be also obtained from the latter using the relation (4.16)
and dropping a (non-zero) factor of type (1)Fa1 . The direct derivation of this pentagonlike identity depicted in Fig. 8 is analogous to the derivation of the standard pentagon
equation for the fusing matrices F since the boundary field n-point blocks are analogs
of the ordinary (n + 2)-point conformal blocks with an additional constraint due to the
delta function in the 2-point boundary block. 4 The relation (4.16) is reproduced from the
pentagon identity (4.29) for particular values of the indices.
4 Mixed pentagon identities analogous to (4.28) appear in the framework of weak Hopf algebras as part of
740
Fig. 8. The mixed pentagon identity.
Imposing the symmetry of the 3-point function

b a

b a
a b
a b
j, (x1 ) (i,i)
(z, z ) k, (x2 ) = j, (x1 ) k, (x2 )(i,i)
(z, z )
one derives following Fig. 9,
X
, 0

b, p
B(i,i)
X
s, , 0
Fb1
p
b
p
b

a, s
B(i,i)
X
m
h1ib
(1)
h1ia (1)Fa1
1
(1)
Fap
s
a
s
a
i(2i 2m +k +j p )
j
b
k
b

1
(1)
0
Fsm
Fbs
j
k
0
k
a
j
a
0

i
k
Fmp
i
i

j
.
i
(4.30)
On Fig. 9 the braiding matrices B() appear, see Appendix E. In the r.h.s. of (4.30) we have
also used the cyclic symmetry relations (4.18), (4.19). Using furthermore these relations
a factor of type (1)Fa1 (and the related summation) can be dropped in both sides of (4.30)
which leads to a slighly simplified version as compared with the original equation (L 3.32)
in [12].
a Big Pentagon identity [76]. The counterparts of (1) F are interpreted as kind of 3j-symbols along with the
standard interpretation of the fusing matrices F as 6j-symbols.
741
Fig. 9. Derivation of (4.30).
From the 2-point function h(k,k)

2 )ia using either the OPE formula
(z1 , z 1 )(l,l)
(z2 , z
(4.27), or (4.23) (follow Fig. 10 with i = 1), we obtain
1 10

X
r r a r
a, r,s1 a, r ,s2 (1)
B(k,k)
B(l,l)
Fa1
a a
,
a 1 i(k +lr j )
Bj e
X
t,t
(j,j );t,t
D(k,k)(l,
l)
Fj r
k
l
k
l
23 13 (s2 ) 23 (s1 )

, (4.31)
13 (t) t
or, equivalently,
(j,j );t,t
a 1 i(k +lj )
D(k,k)(l,
l)
Bj e
r,s1 a, r ,s2 (1)

ei r a,B(k,
B(l,l)
Fa1
k)
,,r,s1 ,s2
r
a
r
a
1a 10r

Fr j
k
k
l
l
12 (t) t
. (4.32)
13 (s1 ) s2
a a
Lastly from the 3-point function h(k,k)
(z1 ) (l,l)
(z2 ) i, (x)i, we obtain, see Fig. 10

X
r t
a, r
a, t
(1)
B(k,k)
B(l,l) Fai
a a
,
X (j,j)
a, i
=
D(k,k)(l,
B(j,j) ei(k r j )
l)
j,j
X
s

eis Fjs
j
i

k
l
r
F
j
l
k

s
l
F
st
k
r

l
.
i
(4.33)
742
Fig. 10. Derivation of (4.33).
For i = 1 (4.33) reduces to (4.31). The sum over s in the r.h.s. of (4.33) represents
up to phases one of the sides in (an auxiliary) hexagon identity, resulting in permuting
to {k,
i, l}, i.e., can be written as B23 ()B12 ()B23 () and thus can be replaced
{l, i, k}
by B12 ()B23 ()B12 (). This gives an alternative representation of the r.h.s. of (4.33)
j, r) (l,
l, j, t ). This is the
obtained from the above by replacing everywhere (k, k,
original form of Eq. (L 3.35) in [12], when (4.24) is inserted, with furthermore inverse
operator ordering convention and opposite overall sign of the phase.
The locality of this 3-point function (the symmetry under the exchange of the fields )
implies also
(j,j);t,t
(j,j); (t ),23 (t)
sk +sl sj
23
D(l,l)(k,
D(k,k)(l,
l)
= (1)
k)
sk := k k ,
(4.34)
while from the associativity of the OPE (4.27) one obtains in particular the relation
(j,j);t,t
(1,1)
(k ,k ); 23 (t ),13 23 (t)
13
D(k,k)(l,
l)
D(j,j)(j ,j ) = D(l,l)(j
,j )
(1,1)
D(k ,k )(k,k)
.
(4.35)
743
All the above equations hold true as well with a sign = 1, inserted in the exponents
of all the phases in these equations including (4.24), and replacing the bulk-boundary
coefficients B with B . Thus when rewritten in terms of the CardyLewellen normalised
coefficients B(CL) all equalities are true for both choices of sign.
Remark
The Lewellen equations in the diagonal sl(2) case were recently confirmed in [81].
The seemingly different version of Eq. (4.30) in [14] is in fact equivalent to the original
Lewellen equation, after taking into account one of the duality relations (a hexagon identity,
see (E.6) in Appendix E) for the braiding matrices. On the other hand the derivation of
the versions of (4.31), (4.33) in [14] appears to be affected by a missing phase in the
intermediate (and needless) formula (17) in [14]. This phase is compensated in the final
formulae following from (4.31) (like (4.44) below) by another phase due to the presumably
neglected difference in the normalisation of the bulk-boundary coefficients (like in (4.24))
as compared with that in [11,12].
4.3.2. More pentagon relations
Before we turn to a discussion on the implications of the CardyLewellen equations we
shall introduce one more ingredient to the scheme. It is natural to assume that there exists a
third fusing matrix, a matrix inverting (1)F , (3)F (1)F = I = (1)F (3) F , or more explicitly,
( ) ( )

( ) t
X
c
k j 13 2
k 12 3 12 2 (1)
(3)
Fpb
Fbs
= ps 2 2 t 0 t ,
c a 2 3
a j 23 (2 ) t 0
b,2 ,3

(3)
Fpb

(3)
Fpb
a
c
1
k
a
c
j
1

= a b k p t 1 1 ,
t

= c b j p t 1 1 .
(4.36)

Along with the standard CVOs pj k , this matrix involves new couplings of type

c
which can be thought of as obtained by a permutation 23 from the boundary fields
a p
0

c
, whence the notation 0 = 23 (). The matrix (3)F satisfies a mixed pentagon
p a
identity analogous to (4.28)

F (3)F F = (3)F (3)F.
(3)
(4.37)
12 (3 ) 12 (2 )
j
and
Furthermore multiplying both sides of (4.29) with (3)Fm0 c
d k t
summing over c, 2 , 3 , using (4.36) in the r.h.s., we obtain another equation of similar
form

F (1)F (1)F = (1)F F,
(3)
which implies various useful relations obtained for particular values of the indices. One of
them reproduces the inverse property of (3)F , another one reads
744

(3)
1b
a
a
j
j
23 () 23()
Fa1
11

where dj = Sj 1 /S11 = F11
dj
(3)
F1b

(1)
a
a
j
j
j
j
j
j
j
b
j
b
1 1
=
13 ( ) 13 ()
1
, ,
dj
(4.38)
1
is the quantum dimension. It furthermore implies
23 () 23()

(1)
Fa1
11
j
b
j
b
1 1
= nbja .
(4.39)
13 () 13 ()
Using the inverse matrix (3)F , as introduced here, the half-plane bulk field can be also
written as a product of generalised boundary fields (a bilocal operator)
12 23 () 1223 () !

X
X
i
a
p,t
a,
(3)
H
) =
B(i,i)
Fpb
(i,
(z, z
i)
a i
12 23 ( ) t
p, ,t
a,b,,
b
a
(z) b i,
z),
a i,
(
(4.40)
which reproduces the small z z expansion in (4.23); compare (4.40) with the chiral
decomposition of the (full) plane physical fields
P
(i,
(, ) =
i)
X
l,t,t
0
k,k,l,
0)
l k
(pl) (l,l);(t,t
l k
D(j,j)(k,k)
0 ( ).
i,t ( ) i,t
Finally we shall exploit the inverse (3)F of the matrix (1)F to rewrite (4.30) in another
equivalent form to be used in the next section. Namely we apply the inverse to (1)Fbs in
the r.h.s. and obtain

1
X
s
s
a, s
(1)
B(i,i)
h1ia Fa1 a
a 13 ( 0 )

X
j i
k j
i(2i 2m +k +j p )
e
Fsm Fmp
i
i
k i
m

1
X
p p
b, p
(1)
B(i,i)
h1i
F
=
b
b1
b b 0
0
b,,
(3)
Fs b
b,, 0 ,
k
j
23 () 23( )
b, p
(1)
B(i,i)
h1ib Fbk
(3)
Fs b
a
a
k
j

(1)
Fap
23 ( 0 )
a
a
p
b
j
a
k
b
k
b
k
b
0
13 () 13 ( )
0
23 () 23( )
23 ( 0 )
j
b
13 ()

(1)
Fa1
1
In the last equality we have used the symmetry relation (4.16).
0 13 ( )
(4.41)
745
4.4. Consequences of the bulk-boundary equations

4.4.1. The Pasquier algebra and its dual
In this section we analyse some important consequences of the set of equations derived.
We start with Eq. (4.41), an inverted version of the first Lewellen bulk-boundary equation
(4.30), in which we take s = 1 = p. This implies k = j , i = i , = 1 = 0 = = 0 . The
sum over m in the l.h.s. is proportional to the modular matrix Sj i , see (E.9), while the sums
over the coupling indices , are worked out using (4.39), the final result being
X
Sj i a 1
Bi h1ia =
nj a b bBi1 h1ib .
(4.42)
S1i
b
For simplicity we have done this computation in the sl(2) case but it extends straightforwardly to arbitrary rank leading to the same formula.
Comparing (4.42) with (3.18), we see that it can be identified with the realisation (3.18)
of the relation (3.4) in terms of the eigenvalues a (i) = ai /1i of the graph algebra
ba . Namely we can identify the ratio bB 1 h1ib /aB 1 h1ia with the ratio b (i)/a (i).
matrices N
i
i
Recalling the expression for h1ia in (4.6) we find a relation between the boundary state
coefficients ai and the bulk-boundary coefficients aBi1
s
i
Cii
a 1
Bi = eii aBi1 (CL) = a1 eii
,
(4.43)
di
a
where for the time being Cii is an arbitrary constant, C11 = 1. Conversely, if we assume
the identification (4.43) (as, e.g., derived in the sl(2) case by other means in [11], with
Cii = 1, see also [26,27]) we recover the relation (3.18) or (3.4) directly from one of the
bulk-boundary equations. As discussed in Section 3.5, from this relation we reconstruct the
graph algebra.
j
In the diagonal case E = I, where a = Saj , the relation (3.18) coincides with the
Verlinde formula, i.e., the standard fusion algebra realised by its characters. On the other
hand the Verlinde formula is known [14] to be recovered from the diagonal version of the
other bulk-boundary equation, the CardyLewellen equation (4.31) to which we now turn.
This equation simplifies for r = 1, leading to k = k , l = l . Using (4.43) and denoting
pi (a) = ai /a1 , Eq. (4.31) turns into
X
Mkl j pj (a),
(4.44)
pk (a) pl (a) =
j
where
Mkl j =
(j,j );t,t
d(k,k )(l,l )
t,t
s
:=
dk dl
dj

Cjj X (j,j );t,t
k
D(k,k )(l,l ) Fj 1
l
Ckk Cll
t,t
k
l
1 1
13 (t) t
(4.45)
and Mkl j = 0 if the corresponding Verlinde multiplicity Nkl j vanishes. Alternatively,

inverting (4.44),
746
Mkl j =
X k l aj
a a
,
a1
k, l, j E.
(4.46)
aV
Let us first look at the diagonal case in which according to (4.46) the constants Mkl j
coincide with the Verlinde fusion rule multiplicities Nkl j . This is confirmed also directly
by the alternative expression (4.45) provided by Eq. (4.31) as we shall now show. In the
(j,j );t,t
(j,j );t,t
diagonal case, denoting D(k,k )(l,l ) = C(k,k )(l,l ) , we can use the inverted equation (4.32)
taken for a = 1, a choice which trivialises all summations, since nr1 1 = Nr1 1 = 1r , with
the result (pointed out in the sl(2) case in [81])
1B 1
j
(j,j );t,t
ei(k +l j ) C(k,k )(l,l )
1B 1 1B 1
k
l

= F1j
k
k
l
l
12 (t) t
.
(4.47)
11
Taking in particular j = 1 (4.47) gives

1 11 1
Bk Bk
(1,1)
= e2ik dk C(k,k )(k ,k) .
(4.48)
Comparing with (4.43) taken in the diagonal case we see that we can identify the
undetermined constant Cii with the normalisation constant of the bulk 2-point function.
We shall retain this identification of Cii in the non-diagonal cases (at the same level as
the given diagonal case) which amounts to setting the relative 2-point normalisation to
1. Combined with (4.35) and (4.43) the relation (4.47) leads to a symmetry of the fusing
matrices analogous to the cyclic symmetry (4.18)
t t

t t

l k 1 2
j k 2 1
d
=
F
dj .
(4.49)
F1j
l
1l
j k 11
l k 1 1
Inserting (4.47) back into (4.45) and using (4.49) reduces the sum over t to the standard
pentagon identity specialised for some choice of the indices (cf. the analogous
relation
(4.38)). Finally we are left with the sum over the coupling index t = t jk l , which
reproduces the Verlinde multiplicity Nkl j and completes the argument; alternatively the
same conclusion is achieved using the simple choice of gauge (E.2).
Note that the relation (4.47), with (4.48) accounted for, is a linear version of the
standard (quadratic) relation for the full plane diagonal OPE coefficients which results
from locality of the (full) plane bulk fields 4-point functions, see Appendix E. In the sl(2)
case this identifies the OPE coefficients of the half- and full-plane diagonal bulk fields.
The identification extends to the nondiagonal sl(2) scalar OPE coefficients as can be seen
generalising to 2-point bulk correlators the computation of the limit L/T of the
1-point correlators in [11], leading to (4.43).
In the general (non-diagonal) sl(2) cases characterised by a fixed level (central charge)
we can express the fusing matrix in the r.h.s. of (4.45) in terms of that in the r.h.s. of (4.47).
Using once again the sl(2) versions of the identities just described, we express it in terms
of the diagonal OPE coefficients at the same level, obtaining for Nkl j = 1
(j,j )
(j,j )
(j,j )
Mkl j = d(k,k)(l,l) = D(k,k)(l,l)/C(k,k)(l,l).
(4.50)

(j,j )
747
The relative scalar OPE coefficients d(k,k)(l,l) have been computed for the sl(2) WZW and
the Virasoro (unitary) minimal models, see, e.g., [62] for an exhaustive list of references.
Now using the expression for the eigenvectors (2.38), they can be computed as in (4.46)
for all minimal models.
The matrices (Mk )l j = Mkl j can be seen as a matrix realisation of an associative
commutative algebra with identity, distinguished basis and an involution . In (4.44)
the algebra is realised by its 1-dimensional representations (characters) given by ratios
of elements of the eigenvector matrix defining ni . This algebra, traditionally called the
b-algebra
Pasquier algebra (M-algebra), is dual in the sense of Ref. [82] to the graph N
considered in Section 3 but unlike its dual, its structure constants are not in general integral,
but rather algebraic numbers. In the simplest sl(2) case the squares (Mkl j )2 of these
constants are rational numbers for all A-D-E cases; this rule persists for most of the sl(3)
(12)
(12)
cases but is broken by two of the graphs E1 and E2 corresponding to the exceptional
modular invariant at level k + 3 = 12, see Appendix D. The type II sl(2) cases, Dodd and
E7 are again distinguished by the fact that the sign of some of the multiplicities Mkl j is
negative and this is a basis independent statement in the sense that there is no choice of
j
basis to make all Mkl non-negative, contrary to the Type I cases Deven , E6 , E8 , and this is
a general feature of Type II theories.
The formula (4.50) extends beyond the sl(2) case for (k, l, j ) such that Nkl j = 1, i.e.,
in cases with trivial Verlinde multiplicity the matrix elements Mkl j provide the (relative)
(j,j )
OPE coefficients d(k,k )(l,l ) . For nontrivial Verlinde multiplicities Nkl j > 1 the relation
between the constants Mkl j (4.46) and the OPE coefficients is not so direct. Let us give a
sl(3) WZW example which illustrates the relation (4.45). There are three graphs found
in [46] which correspond to the exceptional block-diagonal modular invariant at level
(12)
k + 3 = 12, see Appendix D, where these graphs are denoted by Ei , i = 1, 2, 3. One
can pick up triplets of weights (i, j, l) such that the Verlinde multiplicity of the diagonal
sl(3) model at level k +3 = 12 is trivial, Nij l = 1 and check the values of the corresponding
Pasquier algebra structure constants Mij l for each of the three graphs. The result is that,
comparing in particular E1(12) and E3(12), there exist such triplets leading to different values
of Mij l for the two graphs. Since for trivial Verlinde multiplicities the formula (4.45) gives
(l,l );1,1
a direct relation between the two types of constants, Mij l = d(i,i
)(j,j ) , this result suggests
that there are two different solutions for the bulk OPE coefficients in this case. Only one
of these two non-diagonal solutions, namely the one which can be associated with the
Type I graph E1(12) was recovered in [63], exploiting a set of equations for the M-algebra
structure constants. This set was derived from the bulk CFT locality equations assuming
an additional (quadratic) constraint on the OPE coefficients in theories with an extended
symmetry; some of its consequences were also reproduced in the abstract framework of
[65], in particular the relation ni1 a = multa (i) discussed in Section 3. Precisely this relation
fails (and hence the assumptions on the OPE coefficients in [63]) for the graph E3(12) , which
otherwise satisfies all the requirements of Type I.
We conclude with a comment on the OPE coefficients. As discussed in [26,27] one can
relate in the limit L/T the correlators of the half- and full-plane bulk fields IH (z, z )
748
and IP0 (, ), looking at the two dual representations of the partition function with field
insertions; in particular (4.43) was recovered in this way. Though this transformation needs
to be elaborated for higher rank cases it seems reasonable to expect (and in agreement with
(4.47)) that using the two choices of the automorphism , discussed in Section 2.1, we can
identify in this way the OPE coefficients of the two bulk fields with either I 0 = I = (j, j)
or I 0 = (j, j ).
A bit of history
The algebra (4.44) defined through the eigenvectors of the A-D-E Cartan matrices first
appeared in the context of the sl(2) A-D-E lattice models proposed by Pasquier [59]
a short time before the Verlinde fusion rule formula (the A algebra in the sl(2) case)
was found. The interpretation in terms of a pair of dual C (Character)-algebras was
proposed in [60,61] in the discussion of the set of graphs found in [46] as a generalisation
of the Dynkin diagrams associated with the modular invariants of sl(3) WZW and minimal
(j,j )
models. The fact that the relative scalar OPE coefficients d(k,k)(l,l) of all A-D-E series of
the sl(2) WZW (or the subfamily of fields (1, s) in unitary minimal models) coincide in
a suitable basis with the Pasquier algebra (4.44) structure constants Mkl j was the main
result of [62]. It was established through a case by case check, supported by a lattice
model derivation in which the same coefficients appear considering representations of the
TemperleyLieb algebra. CFT locality constraints resulting in formulae quite similar in
spirit to (4.45) were furthermore exploited in [63,64] as an ingredient in the construction of
generalised Pasquier algebras and thus of new examples of graphs related to sl(n) modular
invariants, extending the results in [46,60,61]. The authors of [6264] were, however, not
aware of the parallel development of boundary CFT, and in particular of [11,12], where the
equation (4.31) first appeared. The importance of the algebra obtained from this equation
at r = 1 was recognised and stressed in [14], where a representative example of the sl(2)
WZW Dodd series was considered, for which the set V and the characters i (a) of the
algebra were explicitly described. Presumably the authors of [14] were not aware of the
general A-D-E result in [62]. In the same framework of boundary CFT the Pasquier
algebra reappeared recently in a systematic study of orbifold theories, see [3133] and
references therein, under the name (total) classifying algebra.
4.4.2. Relation to the MooreSeiberg set of duality equations
We have seen that the two Lewellen bulk-boundary equations (4.30) and (4.31) when
p
restricted to some particular values of the indices p in bB(i,i)
become in some sense dual
to each other, recovering the two dual C-algebras, the graph and Pasquier algebras. These
algebras are identical in the diagonal case, reproducing the Verlinde fusion algebra, which
suggests that in this case the above two equations might be related.
On the other hand let us recall that the original derivation [3739] of the Verlinde
formula relies on the use of one of the basic MooreSeiberg duality relations, namely the
equation resulting from the modular property of the two-point functions on the torus, see
(E.7). It involves the fusing/braiding matrices F or B and the modular matrix Sij (p) (in
0
(p)
general Sijt t (p)) for the 1-point functions j (, z) on the torus, the index p standing for

the (representation) label of the inserted CVOs
i
p i

t, z0
and
j
p j

t, z
749
, log z0 = (log z)/,
see [37,38].
The alert reader may have already noticed the full analogy between the MooreSeiberg
torus identity (E.7) and the second version (4.41) of Eq. (4.30). It suggests that the quantity
taking over the role of the modular matrix S(p) is the bulk-boundary reflection coefficient
B p . In the diagonal case this correspondence is precise, i.e., the two are identical up to a
constant.
Indeed first note that Eq. (4.41), still considered in the general (non-diagonal) case,
simplifies for s = 1, that is k = j , i = i , 0 = 1. Inserting in the first line the expression
for the modular matrix Sij (p), see (E.8), and using in the third line (4.38), we obtain an
expression for the modular matrix S(p),

X
Sj i (p ) X b, p
1
p j
(1)
a 1

Bi h1ia
=
B
h1i
F
. (4.51)
b
bj
(i,i)
b a
j j
Si1
b,
Fj 1
p p
Let us concentrate now on the diagonal case I = E. The sums in the r.h.s. of (4.51) can
be reduced to one term choosing a = 1 and using (4.12), since nj 1 b = Nj 1 b = j b , = 1.
Alternatively, one can take a = 1 directly in the original equation (4.30) the resulting
(linear) formula for B p in terms of the fusing matrices F (instead of the formula for its
square derived in [11]) was first explicitly written down by Runkel [81]. More explicitly
we have in the sl(2) case,
j p
B(i,i)
1B 1
i
di dj

Fi1
1B 1
Sij (p)
1
1

= i
S11
di dj
i i
j
Fj 1
p p
p
j
p
Sj i (p)
.
S11
(4.52)
Both (4.51) and (4.52) easily extend beyond the sl(2) case. In particular restoring all
coupling indices the latter formula reads in general
j, p ,t
B(i,i )
= 1Bi1
X
u

F1p
j
j
j
j
u 13 23 ()
11
u 23 13 (t )
Sj i
S1i
(p)
(4.53)
With the help of one of the consequences of the pentagon identity, (4.53) can be inverted
and brought into a form analogous to that of (4.52).
The coincidence of the two seemingly very different quantities, the coefficients B p in
the expansion (4.23) of the half-plane bulk field and the modular matrix S(p) of the torus
1-point blocks, is quite surprising and needs a better understanding. We were led to this
observation trying to find a connection between the two duality schemes, the one of Moore
Seiberg involving the torus, the other, of CardyLewellen, involving the cylinder. Thus to
bring Eq. (4.41), derived from the first of the Lewellen bulk-boundary equations (4.30),
into a form identical to the original MooreSeiberg torus duality relation we furthermore
need to identify the three fusing matrices, (1)F , F , (3)F , i.e.,

t

t
k j
k j
(1)
Fbp
= Fbp
,
a c
a c
750

(3)
Fpb
c
a
j
k
0 0
0 t0

= Fpb
c
a
j
k
0 0
0 t0
(4.54)
This identification is consistent since in the diagonal case I = E both mixed pentagon
identities (4.29) and (4.37) then become the ordinary MooreSeiberg pentagon identity;
see also [81], where the identification of the boundary field OPE coefficients with the
fusing matrices in the sl(2) case was first established by a more elaborate argument.
It is now straightforward to show that the first equality in (4.41) reproduce the two sides
of the MooreSeiberg identity (E.7). Taking into account (4.52) and (4.54) the second
Lewellen bulk-boundary equation (4.31) is seen also to be a consequence of the first, i.e.,
of the same MooreSeiberg torus duality relation. To show this, one has to insert in (4.31)
the expression (4.47) for the OPE coefficients and to compare the equation with (4.41) with
p = 1, see also Appendix E.
We thus see that in the diagonal case the two basic bulk-boundary equations (4.30),
(4.31) are not independent and are equivalent to one of the basic MooreSeiberg duality
relations. The third bulk-boundary equation (4.33), a more general version of (4.31), is an
identity which involves again only the MooreSeiberg duality matrices F , B, S, and thus
can be expected, following the completeness argument of [37,38], to be derivable using the
basic MooreSeiberg duality relations. This in particular implies that any solution of the set
of MooreSeiberg (chiral) duality relations provides a solution of the diagonal boundary
CFT equations.
Remark
Rewritten in terms of S(p) the diagonal case Lewellen equation (4.31) can be also
interpreted as a generalised Verlinde fusion formula with (non-integral) multiplicities
l given by some particular F matrix elements. The matrices F are diagonalised
Fqk
q
by S(p) with the usual eigenvalues Sqi (1)/S1i (1). Because of this they realise another
representation of the usual Verlinde algebra. This formula, which derives from the Moore
Seiberg torus duality identity, appears to have been already considered, following a
different motivation, in [83].
We conclude this section with a few comments on the general non-diagonal cases. The
CardyLewellen boundary CFT can be looked at as a purely chiral alternative of the
usual CFT approach in which we combine left and right chiral blocks imposing consistency
conditions. It has its price in that everything effectively splits the set I is replaced
by two dual sets V and E (for Type I, while for Type II we have to retain the whole I
to describe non-scalar fields), there are two representations of the Verlinde fusion algebra
and a related new fusion algebra (at least in Type I cases); there are two types of chiral
vertex operators, new duality matrices, in particular a second fusing matrix (1)F and
its inverse, (3)F , along with the standard F , satisfying new duality relations, the mixed
pentagon relations, generalising one of the basic genus zero polynomial identities; instead
of one relation involving the modular matrix S(p), there are two independent relations
the two bulk-boundary equations in which the role of S(p) is taken over by the reflection
coefficients B p . It remains to find a consistent solution of the equations at least in the
751
sl(2) case. Some of the ingredients are already known and have been recalled above. In
particular the solution for the D-series has just been obtained by Runkel [94].
5. Conclusions and outlook

In this paper we have reexamined various aspects of boundary effects in RCFTs.
We have in particular analyzed the consistency conditions that determine the allowed
boundary states and fields and their characteristic data, OPE coefficients, etc. We have
seen that boundary conditions are naturally associated with a graph, or a collection of
graphs, whose spectral properties (eigenvalues) encode the diagonal spectrum of the bulk
theory. This legitimates empirical observations made previously on the role of graphs
in the classification of RCFTs. We have seen that the torus partition function may
be fully reconstructed from the information contained in these graphs. We have seen
also that in several cases (b
sl(2), b
sl(N)1 theories), this approach provides a substantially
simpler route to the classification of RCFTs than the study of bulk properties (modular
invariants, . . .). We have finally seen that further important information about some
boundary effects (g-factors, boundary structure constants) is also encoded in the spectral
properties (eigenvectors) of these graphs. The bottom line of this analysis is that a triplet of
ba , Mi ) plays a central role in the whole discussion. These algebraic
matrix algebras (nj , N
structures have been also confirmed by the detailed analysis of the basic equations of
the boundary field theory. In the diagonal case the triplet of algebras reduces to one, the
Verlinde fusion algebra. Accordingly, we have seen that in this case the basic boundary
CFT Lewellen equations can be identified with a set of genus 0 (the pentagon) and genus
1 duality identities of MooreSeiberg. This leads to an identification of some of the basic
notions in the two approaches, namely, the boundary fields OPE coefficients (1) F and the
bulk-boundary reflection coefficients B p , with the chiral CFT fusing matrix F and the
modular matrix S(p), respectively (see the text for precise formulae).
The more general representations ni of the Verlinde fusion algebra and the dual pair
b
{N, M} of associative, commutative (semisimple) algebras have been introduced in earlier
work on bulk (and later on boundary) conformal field theories, but it seems to us that
the consistency of the whole scheme now appears in its full generality and that boundary
RCFTs reveal these features in a simpler and more compelling way than in the bulk. In a
loose sense, the boundary effects expose better the underlying chiral structure of the theory
and its algebraic pattern. This should certainly not come as a surprise, as this is in the same
spirit as the old connection between open and closed strings.
The study of a RCFT through its boundary conditions, its algebra triplet, etc., still
requires a lot of work. The derivation of the Cardy equation relies on a technical assumption
that has been only partially justified, namely the proper definition of unspecialized
characters with linear independence and good modular properties for general chiral
algebras. Also it would be good to have a better understanding of the completeness
assumption: given a certain number of boundary conditions satisfying the Cardy equation,
is it obvious that we may always supplement them into a complete set in the sense
752
discussed in Section 2? Then many questions have been only partially treated: Justify in
full generality the validity of expressions (3.11), (3.12) which have been established so far
only for particular cases; understand better the nature and fusion rules of twisted block
representations that appear in this discussion; set up a general scheme for the systematic
classification of integer valued representations of fusion algebras; set up with more rigour
the formalism of generalized chiral vertex operators, their fusing matrices and the ensuing
duality equations as a consistent chiral approach, alternative to the MooreSeiberg scheme,
etc., such are some of the outstanding problems that are awaiting a proper treatment.
Also it remains to see how our discussion of boundary conditions must be generalized
j
b
in theories where there is no choice of a common diagonalising matrix a leading to N
algebra with integer structure constants. In the approach of [6567], in which the numbers
b are integers, one has to drop the axiom of commutativity, replacing this algebra by a
N
non-commutative structure. In that respect, a better understanding of the relation of our
work with other more abstract approaches Ocneanu theory of subfactors, weak Hopf
algebras would be most profitable.
Directions for future work also include the discussion of other cases: rational or
irrational theories at c = 1, c = 2, or N = 2 superconformal theories are particularly
important cases in view of their physical applications to condensed matter or to string
theory. The generalization to other types of twisted boundary conditions along the cycle
of the cylinder, as examined recently in [95], might constitute another useful approach.
Finally the parallel discussion of these boundary conditions and algebraic structures in
lattice models should be extremely instructive and will be the object of a forthcoming
publication.
Appendix A. The Cardy equation

In this appendix, we rederive the Cardy equation (Section 2.2) in the presence of sources,
which have the effect of introducing unspecialized characters in the partition function. We
restrict to a conformal field theory with a current algebra. Let {J } denote the generators
in the Cartan subalgebra, and be charges coupled to them. We consider the theory on
the cylinder L T of Section 2.2, call w = u + iv the local variable, 0 6 v 6 L, u periodic
of period T , and modify the energymomentum tensor T (w) into

2i X
k X 2 2
J (w)
,
(A.1)
T 0 (w) = T (w)
T a
2
T

2i X
k X 2 2
T 0 (w)
= T (w)

J (w)

.
(A.2)
T
2
T
As an elementary calculation shows, the last term is dictated by the requirement that T 0
satisfies the conventional OPE of an energy momentum tensor. The central charge is not
affected by the additional terms.
One then computes the evolution operators in the two channels of Section 2.2, see Fig. 1.
For the cylinder, mapped to the plane by = e2iw/T , the Hamiltonian reads
cyl
1
=
2
ZT

1
du T 0 (w) + T 0 (w)
=
2
dw T 0 (w) + T 0 (w)
753
(A.3)

X
c
2
(P)
(P)
(P)
(P)
+
J0
J0
L0 + L0
.
=
T
12
(A.4)
Note that the additional term in (A.1) and (A.2) has not contributed to the integral over a
(P)
(P)
=
closed cycle. Taking into account the fact that on boundary states L(P)
0 = L0 and J0
(P)
J0 , we find that the first expression of the partition function reads

X j j j (q,
4 L
c P
)

a b
,
(A.5)
Zb|a = b e T (L0 24 + J0 ) a =
Sj 1
j E

j (q, z) := trVj q L0 24 e2i
c
z J0
(A.6)
and as above q = e4L/T , hence = 2iL/T .

In the other channel, the time evolution on the strip is described by the Hamiltonian
ZiL
Hba =
dw 0
T (w) +
2i
ZiL
dw 0
T (w)
2i
(A.7)
and upon mapping on the upper half plane H by z = ew/L , we find

c
L X 2 2
2i X
(H )
k
J0H
,
L0
Hba =
L
24
T
2
T
(A.8)
where now the additional piece in (A.1) contributes the last term. Since the theory
with energy-momentum tensor (A.1) and (A.2) has still the same operator content and
multiplicities ni a b = nib a as before,we may write
LP
2 X
ni a b i (q, )
(A.9)
Zb|a = tr eT Hba = e2k T
i
with q = eT /L . We then use the modular transformation of unspecialized characters

(see [45, p. 264]):
P z2 X
Sij j (q,
)
(A.10)
i (q, ) = e2ik 2
j
together with the linear independence of the i (q, z) to conclude that (2.16) is indeed true
in full generality.
Appendix B. A-D-E diagrams and intertwiners

In this appendix we establish notations on A-D-E Dynkin diagrams and on the
associated intertwiners.
754
Let G be a Dynkin diagram of the A-D-E type with Coxeter number g. It has n nodes
that may be coloured with two colours, i.e., its n n adjacency matrix Gab connects only
nodes of different colours. This matrix is symmetric and it may thus be diagonalized in an
orthonormal basis. We call this orthonormal basis am , it is labelled by the node a and the
exponent m (see Fig. 2 and Table 1). Hence
X
s 0 s 0
0
.
Ga b bs = 2 cos
(B.1)
g a
b
The s satisfy orthonormality conditions, namely

X 0 00
as as = s 0 s 00 , s 0 , s 00 Exp(G),
(B.2)
as bs
= ab .
(B.3)
s 0 Exp
Because of the 2-colourability of G, one may attach a Z2 grading to each node a. One
proves that if s 0 is an exponent, so is (s 0 ) = g s 0 and the s may be chosen to satisfy
0
a (s ) = (1) (a)as .
(B.4)
Moreover, all graphs having even exponents, viz. the A, Dodd and E6 diagrams, have a Z2
automorphism acting on their nodes and preserving their adjacency matrix (i.e., Ga b =
G (a) (b) , this is the natural Z2 symmetry of these graphs) such that
0
s (a) = (1) (s ) as .
(B.5)
Finally, one may find in the graph G a distinguished node labelled a = 1 such that 1m > 0
for all m. This special node is typically an extremal node, i.e., the end of a branch for the
A-D-E graphs; this is generally the end of a long leg, but for the Dodd graphs, for which
we must choose 1 as the end point of one of the two short legs.
We list hereafter the explicit expressions of eigenvectors of the various Dynkin diagrams.
The D g +1 series are the simplest examples of orbifold models. Their fundamental
2
graphs can be obtained by folding the Ag1 Dynkin diagram so that the nodes ai =
agi D g +1 , i = 1, 2, 3, . . . , g2 1 are identified with the orbit {i} of i under the Z2
2
automorphism , (i) = g i, while the fixed point i = g/2 is resolved into two points
a g , on the graph, denoted n, n 1 in Fig. 2. This implies for the adjacency matrix
2
elements Gai aj = Ai j + Ai (j ) = Aij , for i, j 6= g2 , and Ga g , aj = A g j and allows us
2
to determine the eigenvectors a of G in terms of the eigenvectors Sij of the A adjacency

j
j
matrix. To simplify notation we shall use sometimes i = ai .
B.1. Eigenvectors of the D2l adjacency matrix
j
i =
2 Sij ,
Sgj
j
g , = 2 ,
2
2
i, j 6=
j 6=
g
,
2
g
,
2
( g ,)
i 2
= Si g ,
2
i 6=
g
,
2

( g , 0 )
g 2,
2
ext
= S{(
=
g
,)}{( g , 0 )}
2
755

p
1
S g g + 0 i (1)l .
2 22
(B.6)
For i odd the orbits {i} belong to I ext and can be identified with the subset T = {ai , i =
ext
1, 3, . . . , g2 2, a g , }. The matrix S{i}{j
} is the extended theory modular matrix. The
2
expressions (B.6) can be rewritten in the compact form
s
S1j
j
(B.7)
a = Sa{j }
ext ,
S{1}{j
}
ext
where Sa{j } is a rectangular matrix coinciding for a T with Sa{j
} , while for a = ai 6 T ,
P
Sai {( g ,)} = Si g = 0 and Sai {j } = l{i} Slj = 2Sij for j 6= g/2.
2
B.2. Eigenvectors of the D2l+1 adjacency matrix

j
ai = (1)
j1
2
g , = (1)
2 Sij ,
j1
2
i, j 6=
g
,
2
1
1
Sgj = ,
2
g
2
a2i = 0,
j 6=
i 6=
g
,
2
g
1
g2 , = .
2
2
g
,
2
(B.8)
The identity node is chosen to coincide with one of the fork nodes 1 = a g ,+ (denoted
j
by L in Fig. 2) so that the dual PerronFrobenius eigenvector 1 = g ,+ has positive

2
g
2
a1
entries (while = 0). The fundamental node f is identified with

bg 1 . Also a = a for all a, while (a g ,+ ) = a g , .
N
2
g
2
1, i.e., G =
Next we display the eigenvectors of the exceptional Er Dynkin diagrams as a matrix

j
{a }, with the row index a running over the nodes, following the numbering of Fig. 2,
and the column index j over the exponents in the same order as in Table 1. There too, Sij
denote the eigenvectors of the diagonal graph adjacency matrix A with the same Coxeter
number.
B.3. Eigenvectors of the E6 adjacency matrix
1
2
1
2
1
2
12
b
a a
b

c
0
d
d
0
c
j
,
a =
1
1
b 2 a a 2 b
a 12 b b 12 a
d
where
1
a=
2
0 c
3 3
,
6
0 d
c
s
1
b=
2
(B.9)
3+ 3
,
6
s
1
c=
2
3+ 3
,
3
s
1
d=
2
3 3
3
756
are determined from

1i
s
X
= S1i
Sj i ,
6i
= (S4i + S8i ) P
S1i
Sj i
for i E,
= {1, 7}.
j
a =
1
3
b a 1
d e
1
6
1 1
6
6
a c
0 d
1
6
0 f
1
3
f e
b c
d f
,
1
1
6
6
e d
a b
j
where a, b, c, d, e, f are determined from 1 =

j
and 2 =
Explicitly,
S2j
S1j
(B.10)
q
S1j
1 . (The values in the 5th row come from

1/2
,
a = 18 + 12 3 cos 18

11 1/2
,
b = 18 + 12 3 cos 18

13 1/2
,
c = 18 + 12 3 cos 18
i Sij , where
j
5 = 2 S6j for
= {1, 9, 17},
j = 1, 5, 7.)

1/2
d = 12 1 + cos 9
,

5 1/2
e = 12 1 + cos 9
,

7 1/2
f = 12 1 + cos 9
.
(B.11)
a f
c d d
b e h g g
c d a f f
d a f c c
j
a =
e h g b b
f c d a a
g b e h h
h g b
e e
c f
a
h e b
a d c
f a d
,
g h e
d c f
e b g
b
(B.12)
g h
j
where a, b, c, d, e, f , g, h are determined from 1 =

for j E, = {1, 11, 19, 29}. Explicitly,
q
S1j
P
i
Sij and 2 =
S2j
S1j
"
a=
c=
d=
15(3 +
5) +
#1/2
15(130 + 58 5)
2
q
#1/2
"
15(3 + 5) 15(130 + 58 5)
757

q

1/2
b = 15 + 75 30 5
,

q

1/2
e = 15 75 + 30 5
,
2
q
#1/2
"
15(3 5) 15(130 58 5)
,
2
q
#1/2
"
15(3 5) + 15(130 58 5)
,
f=
2
q

1/2
g = 15 + 75 + 30 5
,

q

1/2
h = 15 75 30 5
.
(B.13)
To such a graph G, one then attaches matrices Vi as follows. The case of reference is
the Ag1 diagram of same Coxeter number g as G. For this A graph, both the nodes and
the exponents take all integer values in {1, . . . , g 1}. The s are then nothing other than
the entries of the (symmetric, unitary) matrix S of modular transformations of characters
of the affine algebra b
sl2 at level g 2
s
ii 0
2
(A) i 0
sin
,
(B.14)
= Sii 0 =
i
g
g
in terms of which the fusion coefficients Ni1 i2 i3 may be expressed through Verlinde
formula. Note also that
Ngi i1 gi2 = Nii1 i2
(B.15)
= S.
because of property (B.4) applied to
We now return to the graph G of Coxeter number g. The fused adjacency matrices Vi
with i = 1, . . . , g 1 are n n matrices defined recursively by the sl(2) fusion algebra
(A)
Vi = V2 Vi1 Vi2 ,
2 < i 6 g,
(B.16)
and subject to the initial conditions V1 = I and V2 = G. (One may see that Vg = 0.) The
matrices Vi are symmetric and mutually commuting with entries given by a Verlinde-type
formula
X Sim
m m .
(B.17)
Via b = (Vi )a b =
S1m a b
mExp(G)
Regarded as (g 1) n rectangular matrices, for a fixed, the Via b intertwine the A and G
adjacency matrices
X
X 0
0
A i i Vi 0 a b =
Via b Gb0 b .
(B.18)
i0
b0
Regarded as n n matrices, the Vi satisfy not only their defining relation (B.16) but also
the whole b
sl(2) fusion algebra (2.22)
758
Vi1 Vi2 =
Ni1 i2 i3 Vi3 .
(B.19)
i3
From their recursive definition and initial conditions, it follows that the entries of the V
matrices are integers. What is not obvious is that these entries are non-negative integers.
This follows either from a direct inspection or from an elegant group theoretic argument
due to Dorey [85]. We refer the reader to [46,57] for the explicit expressions of these
intertwiners. As a consequence of the existence of the automorphism defined above
(B.5), a Z2 symmetry on A, Dodd , E6 graphs and the identity for Deven , E7 , E8 , one has
Vgs a (b) = Vsa b .
(B.20)
P
ba = b Via b N
bb ) one
Using (B.16) and (3.4) (i.e., in the notations of this appendix, Vi N
ba matrices for all but the Deven cases as polynomials of V s
can express the graph algebra N
with integer coefficients. This in particular ensures that they have integer matrix elements.
has to be added since V g =
For Deven one of the extended fusion algebra generators N{ext
g
2
2 ,}
g
ext
ext
bi = Vi = Vgi for i = 1, 2, . . . , 1. For the three exceptional
+N g
while N
Ng
{ 2 ,+}
{ 2 ,}
cases Er we have
bi = Vi ,
N
i = 1, 2, . . . , r 3,
br1 = Vr1 Vr3 ,

N
br2 = Vr Vr4 ,
N
br = Vr2 + Vr4 Vr ,
N
j
which translates into relations between the eigenvalues and given 1 allows to express any
j
a in terms of the modular matrix S elements.
Over recent years, these matrices have made repeated appearances in a variety of
problems. Originally introduced in the discussion of local height probabilities in lattice
models [59] and of boundary partition functions [9,46] (see below), they have also appeared
in the following contexts:
(i) The cells or intertwiners of Boltzmann weights of height models [46,57].
(ii) The decomposition of the representation of the TemperleyLieb algebra on the
space of paths from a to b on graph G onto the irreducible ones on the paths from
L
(G)
(A)
1 to s on graph Ag1 [58] according to Ra b = s Vsa b R1 s .
(iii) The counting of essential paths on graphs [84]; see also recent mathematical
work by Xu, Bckenhauer and Evans [6567].
(iv) The expression of the blocks of the partition function (1.2) as (3.11), (3.12), see
Section 3.3.
(v) The sl(2) intertwiners appear in the computation of the multiplicities mbs0 of an
irreducible representation b of the finite group, associated with G in the McKay
correspondence [86,87], in the SU(2) representations of dimension s 0 [96]. Namely
the coefficients of the Kostant polynomials in the generating function Fb of these
P
b
multiplicities are given for a non trivial b by
c G0c Vsc , where Gab is the
adjacency matrix of the affine Dynkin diagram and a = 0 is the affine node deleted
in passing from the affine Dynkin diagram to the ordinary one. The proof of this
fact is reduced to the recursive relation (B.16).
759
(vi) These same entries seem to appear ubiquituously in the description of S-matrices
of affine Toda theories [88] and in the description of the excitation spectrum of
integrable lattice models [8991].
Appendix C. Uniqueness of the boundary conditions of minimal models

C.1. Matrices with spectrum < 2
We first recall general results on symmetric matrices with non-negative integer entries
and with eigenvalues between 2 and 2.
It is a standard result that symmetric matrices with non-negative integer entries and
eigenvalues ]2, 2[ may be classified. A lemma of Kronecker asserts that the
eigenvalues are of the form 2 cos phi i for integers pi and hi and for the largest one(s),
p = 1. One may regard any such matrix as the adjacency matrix of a graph. Irreducible
matrices correspond to connected graphs, and by an abuse of language one may call a
matrix bicolourable if the graph has that property. One proves [47] that any irreducible
bicolourable symmetric matrix with spectrum in ]2, 2[ is the adjacency matrix of one of
the simply laced Dynkin diagrams of type A-D-E.
If one relaxes the assumption of bicolourability, with any symmetric non-bicolourable
irreducible matrix G
one may associate a bicolourable symmetric matrix with a block
0 G
0
. The corresponding graph is irreducible and has a Z2 symmetry that
form G =
G 0
exchanges the two colours. Any eigenvalue of G gives rise to two eigenvalues for G0
and one thus concludes that G0 is of A-D-E type, and its irreducibility forces G0 = A2p .
Its Z2 quotient G is what we call the tadpole graph Tp = A2p /Z2 .
Finally if one relaxes the assumption of irreducibility, one concludes that any matrix
(with non-negative entries and spectrum between 2 and 2) is the direct sum of A-D-E or
tadpole graphs
M
Gi of A-D-E or tadpole type
G=
Gi ,
and this decomposition is unique, up to the permutation of factors. The uniqueness may
be easily proved by induction on the number of terms or on the dimension of the matrix:
Given a matrix G, one first identifies its largest eigenvalue, of the form 1 = 2 cos h1 . By
the previous statement, there is an A-D-E or tadpole graph G1 with Coxeter number h1
m
and exponents mi , such that all its eigenvalues 2 cos h1j appear in the spectrum of G. Thus
G = G1 G00 , and one may apply on G00 the induction hypothesis. If 1 has multiplicity 1,
this suffices to establish the uniqueness of the decomposition (up to permutations), while
the case where 1 has nontrivial multiplicity is also easily dealt with. The uniqueness of this
decomposition implies a property used several times in the text, namely that the spectrum
(between 2 and 2) determines the form of the matrix up to a permutation of its rows and
columns.
760
C.2. Representatives of n12 and n21

We now return to minimal models.
Explicit form of n12 and n21
It is convenient to work in a basis different from that used in (2.38). In the basis r1 =
1, . . . , p, a G, the second term in (2.33) does not contribute to n12 since N2p r1 r2 =
r1 ,2p+1r2 = 0 for 1 6 r1 , r2 6 p. Thus n12 = Ip V2 = Ip G, where Ip is the p
dimensional unit matrix
G
.
(C.1)
n12 = Ip G =
..
.
G
As for n21 , in the same basis, the second term of (2.33) receives a contribution only
from r1 = r2 = p, namely (N2p1 )p p = 1, while Vg1 = , the matrix that realizes the
automorphism :
a b = a (b).
Thus one finds that
0 I
n
In 0
n21 =
..
(C.2)
In
..
.
In
In
After conjugation by a block-diagonal matrix with and In in alternating positions,
which leaves the form (C.1) of n12 unchanged, n21 may be recast in the form
0
= Tp ,
(C.3)
n21 =
.. ..
.
.

in terms of the tadpole Tp adjacency matrix. All the other nrs are obtained as universal
polynomials of the two matrices n12 and n21 .
Uniqueness of the form of n12 and n21
Conversely, suppose we only know that the representation nrs has a spectrum specified
by the set of exponents E. We want to prove that there exists a basis in which n12 and n21
take the forms (C.1) and (C.3).
We first make use of the property that the set Exp(G) is stable modulo the Coxeter
number g of G under multiplication by any integer coprime to g and is also stable under
the reflection s g s. We then find that the spectrum of n12 is made of p copies of
Exp(G). As explained above in C.1, this implies that in some basis
n12 = Ip G.
(C.4)
For the other generator n21 , one observes first that the set of numbers that appear in
gr 0
r 00
}, r 0 = 1, 3, . . . , 2p 1, is simply the set {(1)g+12 cos 2p+1
},
(2.31), namely {2 cos 2p+1
761
r 00 = 1, 3, . . . , 2p 1, which is (1)g+1 times the spectrum of the tadpole Tp . According

0
to (2.31), this has to be multiplied by (1)s , as s 0 runs over the exponents of G. Thus if
g is even (which is the general case except when G = A2l ) the spectrum of n21 is made of
as many copies of that of Tp (respectively, Tp ) as there are odd (resp. even) exponents in
G. For A2l which has as many even as odd exponents, the same conclusion is still correct!
Finally one notices that these signs are just the eigenvalues of the matrix, and one thus
concludes that
n21 Tp ,
(C.5)
where the sign means that it holds in some basis obtained from that of (C.4) by a
simultaneous permutation of rows and columns. From this expression, one can see that
n21 has no row or column with more than two 1s,
(C.6)
n21 has exactly n rows and columns with one 1,
(C.7)
properties invariant under permutations of rows and columns.

We also know that n21 must commute with n12 . In a basis in which n12 takes the form
(C.4), n21 may thus be regarded as made of n n blocks that commute with G. We shall
combine these facts about n21 as follows:
A non-vanishing matrix X with elements in N which commutes with G cannot have
a row or a column of zeros.
Proof: let 1 be the PerronFrobenius eigenvector of G, GX = XG implies that X 1
is an eigenvector of G with the same eigenvalue, hence proportional to 1 , X 1 =
c 1 , with c 6= 0 since the entries of both X and of 1 are non-negative. If X had a
vanishing row, X 1 would have a vanishing component, which is impossible for the
PerronFrobenius eigenvector. If X has a vanishing column, one repeats the argument
with XT .
Any matrix X with elements in N which commutes with G and which appears in the
block decomposition of n21 cannot have more than one 1 per row or column.
Proof: If a matrix X with more than one 1 in a row (respectively, column) was a
block of n21 , because of the property (C.6), all the other blocks on the left or the right
(respectively, above or below) of X would have to have at least one vanishing row
(respectively, column), which is impossible by (i) above, or to vanish altogether. In
the latter case, after a possible reshuffling of rows and columns leaving (C.4) invariant,
one would have
X 0 0
0 X 0
T
0
X
0 0
or n21 =
(C.8)
either n21 =
...
0
0
..
..
0
.
.
which would lead to a pair n12 , n21 reducible in the same basis.
It follows that the matrices that may appear as blocks in the decomposition of n21
must be matrices with one 1 on each row and column, i.e., permutation matrices
that commute with G. These permutation matrices are the symmetries of the Dynkin
diagram, and thus are readily listed:
762
X = I, ,
if G = An , D2q+1 , E6 ,
= I, i , i = 1, . . . , 5,
(C.9)
if G = D4 ,
(C.10)
= I, ,
if G = D2q , q > 2,
(C.11)
= I,
if G = E7 , E8 .
(C.12)
Here i denote the 5 nontrivial permutations of the nodes of the D4 diagrams, and the
matrix 0 exchanges the two end points of D2q , q > 2.
One then demands that the symmetric matrix n21 made of such blocks is irreducible and
satisfies (C.4)(C.7). This implies that at most one non-vanishing block appears on the
diagonal. Consistency with the form n21 Tp leaves as the only possibility
0
X1T
n21 =
0
X1
0
X2T
0
0
X2
..
.
..
.
0
..
.
T
Xp1
Xp1
(C.13)
where X1 , X2 , . . . , Xp1 are chosen among the symmetry matrices of G. A final

permutation of rows and columns by a block diagonal matrix diag(Y1 , Y2 , . . . , Yp1 , I )
brings n21 into the form n21 = Tp while leaving the form (C.4) of n12 unchanged,
T XT
T
provided Yj = Yj +1 XjT , hence Yj = pj Xp1
p2 Xj . Then both n12 and n21
have their canonical forms (C.1), (C.3). Q.E.D.
Remark
Although it is not required for the present analysis, it may be interesting to look at the
commutant of matrices of A-D-E type.
For any G of A-D-E type, with the exception of Deven , all eigenvalues are distinct. It
follows that any matrix X that commutes with G may be diagonalized in the same basis
as G and consequently be written as a polynomial of G, i.e., as a linear combination of
I, G, G2 , . . . , Gn1 . In order to look at cases where entries of X are requested to take
values 0 or 1 only, and with constraints on the number of 1s, it is advantageous to use rather
the basis of fused graph matrices: X is a linear combination of the linearly independent
b2 = G, . . . , N
bn . The Deven case is slightly more involved, since the
b1 = I , N
matrices N
matrices that appear naturally are not independent.
The commutant of an A-D-E matrix is:
ba for G = A, Dodd , E6 , E7 , E8 .
A linear combination of the graph fusion matrices N
ba and of two of the three matrices ab , a 6= b = 1, 3, 4,
A linear combination of the N
that exchange two of the three extremal points of the D4 graph.
b2q 0 ,
b2q and Y = N
ba and of the two matrices X = 0 N
A linear combination of the N
0
where the matrix exchanges the two end points, for G = D2q , q > 2.
763
Appendix D. b
sl(3) modular invariants and graphs
The WZW b
sl(3) theories may be discussed along the same lines as in Sections 2 and 3.
Solutions ni to the Cardy equation are associated with graphs, with specific spectral
properties: their eigenvalues are given by ratios of elements of the modular S matrix
labelled by weights of the diagonal spectrum E of the bulk theory. Conversely, spectral
properties and the fact that the ns form a representation of the fusion algebra are not
restrictive enough to yield the list of possible bulk spectra, as occurred in b
sl(2) (up
to the unwanted tadpole graphs). There are indeed many solutions, i.e., graphs and
representations of the fusion algebra, that must be discarded as not corresponding to a
modular invariant partition function in the list (Table 2) of Gannon [92,93]: see [46] for
such extra solutions. We may summarise the salient features of the analysis as follows, see
also the accompanying Tables 2 and 3 and Figs. 11 and 12.
At least one graph (or rather one set of n matrices) has been identified for each bulk
theory, i.e., each modular invariant. But it is not known if this list of graphs and ns is
exhaustive.
Note that the hypothesis of 3-colourability of the graphs that looked natural on the
basis of the sl(2) case has to be abandoned if we want to cover all cases. This is
manifest on Table 3 where it appears that in some cases (namely A(n) , and D (n) , n
not a multiple of 3, and E (8) , the set E is not invariant under the automorphism of
(D.1), as it should be if the graph was 3-colourable.
There are a few pairs or even triplets of isospectral graphs, i.e., different sets
of ns that give distinct solutions of the Cardy equation for a given bulk theory.
These graphs/representations should not only describe different sets of complete
orthonormal sets of boundary conditions for that bulk theory, but also presumably
be associated with different operator algebras and lattice realisations.
In Table 3, which summarises the state of the art, we have also indicated if the graph
is of Type I or Type II, following the discussion of Section 3. Some hybrid cases are
b structure constants are both non-negative,
also encountered, in which the M and N
b algebra has no subalgebra isomorphic to some extended fusion algebra.
but the N
Notations and footnotes for Tables 2 and 3
(Shifted) weights of SU(3) = (1 , 2 ) := 1 1 + 2 2 , where 1 , 2 are the
fundamental weights of SU(3), = (2 , 1 ), triality () := (1 1) + 2(2 1)
1 2 mod 3.
Q is the set of weights of triality zero.
Weyl alcove of shifted level, or altitude, n := k + 3,
(n)
= { = 1 1 + 2 2 | 1 , 2 > 1, 1 + 2 6 n 1}.
P++
(n)
Automorphism of P++
(1 , 2 ) := (n 1 2 , 1 ).
(D.1)
(a) One of the two connected parts of the fused Dynkin diagram of type An1 (the one
that possesses the exponent (1, 1)). It looks different depending on whether n is
764
Table 2
List of b
sl(3)k modular invariants; n = k + 3

A(n)
D (n)
Z=
Z=
P
(n)
| |2
(n)
P++
P++
1P
2
3
(n) | + + 2 |
QP++
Z= P
P
2
n
(n) | | +
(n)
QP++

D (n)
E (8)
E (8)
E (12)
(12)

EMS

(12)
EMS

E (24)
if 3 does not divide n
P
P

P
2
2
1
`
`
(n)
`=0
`=0
3
Z = P QP++
P
n
(n) +
(n)
QP++
P++ \Q
if 3 divides n
P++ \Q
if 3 divides n
if 3 does not divide n
Z = |(1,1) + (3,3) |2 + |(3,2) + (1,6) |2 + |(2,3) + (6,1) |2 + |(4,1) + (1,4) |2 + |(1,3) + (4,3) |2 + |(3,1) + (3,4) |2
Z = |(1,1) + (3,3) |2 + ((3,2) + (1,6) )((2,3) + (6,1) ) + c.c. + |(4,1) + (1,4) |2 + ((1,3) + (4,3) )((3,1) + (3,4) ) + c.c.
Z = |(1,1) + (10,1) + (1,10) + (5,5) + (5,2) + (2,5) |2 + 2|(3,3) + (3,6) + (6,3) |2
Z = |(1,1) + (10,1) + (1,10) |2 + |(3,3) + (3,6) + (6,3) |2 + |(5,5) + (5,2) + (2,5) |2 + |(4,7) + (7,1) + (1,4) |2
+ |(7,4) + (1,7) + (4,1) |2 + 2|(4,4) |2 + ((2,2) + (8,2) + (2,8) )(4,4)

+ c.c.
Z = |(1,1) + (10,1) + (1,10) |2 + |(3,3) + (3,6) + (6,3) |2 + |(5,5) + (5,2) + (2,5) |2 + 2|(4,4) |2
+ ((4,7) + (7,1) + (1,4) )((7,4) + (1,7) + (4,1) ) + c.c. + ((2,2) + (8,2) + (2,8) )(4,4)
+ c.c.
Z = |(1,1) + (22,1) + (1,22) + (5,5) + (5,14) + (14,5) + (11,11) + (11,2) + (2,11) + (7,7) + (7,10) + (10,7) |2
+ |(7,1) + (16,7) + (1,16) + (1,7) (7,16) + (16,1) + (5,8) + (11,5) + (8,11) + (8,5) + (5,11) + (11,8)|2
A(n)
Table 3
List of modular invariants and associated known graphs for b
sl(3). Footnotes (a)(e) are explained at the end of Appendix D
Graphs
Altitude
A(n)
A(n)
A(n)
A(n) = (An1 ? An1 )c
D (n)
D(n)
(b)
A(n) /Z3
D (n)
(c)
D(n) = 123 (An1 ? An1 )c
E (8)
E (8)
E (8)
E (8)
E (12)
Ei
(12)
(12)
E4
EMS
(12)
E (24)
EMS
(a)
(e)
i = 1, 2, 3
(d)
Exponents
Type
(n)
P++
{(j, j )},
I
1 6 j 6 b n1
2 c
(n)
(n)
b>0
M, N
n = 0 mod 3
P++ Q, with ( n3 , n3 ) triple
n 6= 0 mod 3
P++ Q
b>0
M, N
{(j, j ), (n 2j, j ), (j, n 2j )},

1 6 j 6 b n1
2 c
b>0
M, N
(1, 1), (6, 1), (1, 6), (3, 3), (3, 2), (2, 3),
(4, 1), (3, 4), (1, 3), (1, 4), (4, 3), (3, 1)
(1, 1), (3, 3), (4, 1), (1, 4)
b>0
M, N
12
(1, 1), (10, 1), (1, 10), (5, 5), (5, 2), (2, 5),
Ei
and twice (3, 3), (3, 6), (6, 3)
(12)
E2 II
(12)
12
(1, 1), (10, 1), (1, 10), (5, 5), (5, 2), (2, 5),
(3, 3), (3, 6), (6, 3) and twice (4, 4)
II
E5
(12)
12
(1, 1), (10, 1), (1, 10), (5, 5), (5, 2), (2, 5),
(4, 1), (7, 4), (1, 7), (1, 4), (7, 1), (4, 7),
(3, 3), (3, 6), (6, 3) and twice (4, 4)
II
E (24)
24
(1, 1), (22, 1), (1, 22), (5, 5), (14, 5), (5, 14),
(7, 7), (10, 7), (7, 10), (11, 11), (11, 2), (2, 11),
(7, 1), (16, 7), (1, 16), (1, 7), (16, 1), (7, 16),
(5, 8), (11, 5), (8, 11), (8, 5), (11, 8), (5, 11)
i = 1, 3: I
765
(12)
Modular invariant
766
Fig. 11. The known graphs in the case of b

sl(3). Conventions: (a) For the 3-colourable graphs, the
triality of nodes is indicated by the colour: black = 0, grey = 1, or white = 2; the graph
represents the matrix n21 if edges are oriented from black to grey, or grey to white, etc. (b) For the
non-3-colourable graphs, either the orientation of all edges (of matrix n21 , say) is indicated, (D(n) ,
3 6 | n, series, E (8) ), or all links are unoriented (A(.) series).
b algebras of A(n) are positive, as they follow simply from

even or odd. The M, N
the Verlinde fusion algebra N of b
sl(2). If n is odd, A(n) is the connected component
of the graph of adjacency matrix A2n1 I made of the nodes a odd (integer spin
in sl(2)). For a triplet of exponents = (l, l), = (m, m) and = (r, r) of A(n) ,
bab c is the restriction of the Verlinde An1 algebra
M, = Nlm r + Nlm nr , and N
b algebras of A(n) = I + A n 1 coincide with
to odd a, b, c. For even n, the M and N
2
the Verlinde algebra of A n2 1 .
(b) The orbifold of A(n) , see [73].
(c) The ordinary Z3 fold of A(n) .
767
Fig. 12. The known graphs in the case of b

sl(3).
(d) The unfolded (and 3-colourable) version of A(n) . Their adjacency matrix is a tensor
product by the permutation matrix
0 1 0
123 = 0 0 1
1 0 0
b algebras are simply obtained from those of A(n) , thus also > 0
their M and N
(e) The Z3 fold of E (8) .
768
Appendix E. Formulae for fusing, braiding and modular matrices

We collect in this appendix some standard formulae for the genus 0 and 1 duality
matrices. The fusing matrices satisfy several identities implied by the pentagon identity
they can be recovered from formulae (4.36), (4.38), (4.16) in the text making the
identification (4.54).
Symmetries:

k j
j k
i
l i
l
= Fp q
.
(E.1)
= Fp q = Fpq
Fpq
i
j k
i l
l
k j
Choice of gauge:
1 10 s

dq
i j i
(0) i
=
13 ( ) .
(E.2)
Fq1
j j
di dj
q
q
In the sl(2) case denote by Ckj the normalisation of the CVO in this gauge. Then for the
fusion matrix corresponding to CVO normalised to 1, one has
v
v

u
q i

u
q q 1

u
u Ckj
Ckj Cql (0) k j
Ci l Cqq (0) k j
k j
t
t
F
Fpq
=
=
,
(E.3)
Fpq
p p 1
i C p pq
i l
i l
i l
Ckp
Cki Cj l Cpp
jl
or
p

p
1
Cki Cj l Cpp
Fpq
k
i
j
l

q
1
= Ckj Ci l Cqq
Fqp
k
j

i
.
l
(E.4)
This equation coincides with the quadratic relation resulting from locality of the physical
i = C (i,i)
i
1
4-point function in the diagonal case. Hence the constants Ckp
(k,k)(p,p) , Ckp Cii =
p
1
Cki Cpp
can be identified with the physical OPE structure constants in this case. For the
minimal models these constants were computed in [80]; the matrices F (0) in the gauge
(E.2) coincide up to signs with a product of standard q-6j symbols, see [97] for the latter.
Braiding matrices:

( )

i j 23 2
i l 2
() = ei(k +l p q ) Fpq
.
(E.5)
Bpq
k l 1 23 (2 )
k j 1 2
The q-analogs of the Racah identity (hexagon identities), = 1:

X
i k iq
l i
Fmq
Fqp
e
j l
j k
q

i l
.
= ei(m+p l j i k ) Fmp
j k
(E.6)
Recall the MooreSeiberg torus duality identity resulting from a relation in the modular
group of the torus with two field insertions, namely, S(j1 , j2 ) a = b S(j1 , j2 ) where
S(j1 , j2 ) is the modular matrix of two-point blocks, expressed in terms of F and S(p),
and a, b are the monodromy transformations moving one of the CVO around the a, b
cycles, [37,38],
Sri (s)
e2i(i m ) Fs m
Sqi (p)e
j2
j1
i(p j1 j2 )
i
i
Fsq

Fmp

r
r
j1
i
j2
i
769

j2
j1
Frp
j2
q

j1
.
q
(E.7)
Choose s = 1 = r. This implies that j1 = j2 = j and q = j , hence

X
S1i
j i
j
j
i(2
+2
2
)
m
p
i
j

e
F1m
Fmp
Sj i (p) =
j i
i i
j j
m
F1p
j j

X
S1i
j
j j
i
i(2i +2j 2m )

e
F
F
=
1m
mp
i i
j i
j j
m
F1p
j j

X
S1j
j i
j
j

ei(2i +2j 2m p ) Fpm
F
. (E.8)
=
m1
j i
i i
i
i
m
Fp1
i i
The second equality is obtained reversing the sums in (E.7) and solving for Sri (s) as above,
while the third is obtained from the transposed version of (E.7) taking into account S(p)2 =
Ceip . For p = 1 the formula reproduces the ordinary S = S(1) matrix

X
Si1
j i
j
j
2i(
+
)
m
i
j

e
F1m
Fm1
Sij =
j i
i i
j j
m
F11
j j
X
e2i(i +j m ) dm Nij m .
(E.9)
= S11
m
Acknowledgements
P.A.P. is supported by the Australian Research Council. V.B.P. acknowledges the
hospitality of Arnold Sommerfeld Institute for Mathematical Physics, TU Clausthal,
the support and hospitality of Service de Physique Thorique, CEA-Saclay and the
partial support of the Bulgarian National Research Foundation (contract 8-643). J.-B.Z.
acknowledges partial support of the EU Training and Mobility of Researchers Program
(Contract FMRX-CT96-0012), which made possible an extremely profitable stay in
SISSA, Trieste. We acknowledge useful discussions with M. Bauer, D. Bernard, J. Fuchs,
T. Gannon, A. Honecker, A. Recknagel, P. Ruelle, I. Runkel, C. Schweigert and G. Watts.
References
[1] H. Kawai, D.C. Lewellen, S.-H. Tye, A relation between tree amplitudes of closed and open
strings, Nucl. Phys. B 269 (1986) 123.
770
[2] N. Ishibashi, The boundary and crosscap states in conformal field theories, Mod. Phys. Lett. A 4
(1987) 251264.
[3] N. Ishibashi, T. Onogi, Conformal field theories on surfaces with boundaries and cross-caps,
Mod. Phys. Lett. A 4 (1987) 161168.
[4] C.G. Callan, C. Lovelace, C. Nappi, S.A. Yost, Adding holes and crosscaps to the superstring,
Nucl. Phys. B 293 (1987) 83113.
[5] Z. Bern, D. Dunbar, Coupling of open to closed bosonic strings in four dimensions, Phys. Lett.
B 203 (1988) 109117.
[6] J. Polchinski, Y. Cai, Consistency of open superstring theories, Nucl. Phys. B 296 (1988) 91
128.
[7] G. Pradisi, A. Sagnotti, Open string orbifolds, Phys. Lett. B 216 (1989) 5967.
[8] M. Bianchi, A. Sagnotti, On the systematics of open string theories, Phys. Lett. B 247 (1990)
517524.
[9] H. Saleur, M. Bauer, On some relations between local height probabilities and conformal
invariance, Nucl. Phys. B 320 (1989) 591624.
[10] J.L. Cardy, Boundary conditions, fusion rules and the Verlinde formula, Nucl. Phys. B 324
(1989) 581596.
[11] J.L. Cardy, D.C. Lewellen, Bulk and boundary operators in conformal field theory, Phys. Lett.
B 259 (1991) 274278.
[12] D.C. Lewellen, Sewing constraints for conformal field theories on surfaces with boundaries,
Nucl. Phys. B 372 (1992) 654682.
[13] G. Pradisi, A. Sagnotti, Ya.S. Stanev, The open descendants of non-diagonal SU(2) WZW
models, Phys. Lett. B 356 (1995) 230238.
[14] G. Pradisi, A. Sagnotti, Ya.S. Stanev, Completeness conditions for boundary operators in 2D
conformal field theory, Phys. Lett. B 381 (1996) 97104.
[15] A. Sagnotti, Ya.S. Stanev, Open descendants in conformal field theory, Fortschr. Phys. 44 (1996)
585596.
[16] I. Affleck, A. Ludwig, Universal noninteger ground-state degeneracy in critical quantum
systems, Phys. Rev. Lett. 67 (1991) 161164.
[17] I. Affleck, Conformal field theory approach to the Kondo effect, M.A. Nowak, P. Wegrzyn
(Eds.), Bosonization and Conformal Symmetry in High Energy Physics and Condensed Matter
Physics, Zakopane, 1995, Act. Phys. Pol. 26 (1995) 18691932, and further references therein.
[18] I.V. Cherednik, Factorizing particles on a half-line and root systems, Teor. Mat. Fiz. 61 (1894)
3544.
[19] S. Ghoshal, A. Zamolodchikov, Boundary S matrix and boundary state in two-dimensional
integrable quantum field theory, Int. J. Mod. Phys. A 9 (1994) 38413886; Int. J. Mod. Phys.
A 9 (1994) 4353 (Erratum).
[20] E. Corrigan, P.E. Dorey, R.H. Rietdijk, R. Sasaki, Affine Toda field theory on a half-line, Phys.
Lett. B 333 (1994) 8391; hep-th 9404108.
[21] M. Oshikawa, I. Affleck, Boundary conformal field theory approach to the critical twodimensional Ising model with a defect line, Nucl. Phys. B 495 (1997) 533582.
[22] I. Affleck, M. Oshikawa, H. Saleur, Boundary critical phenomena in the three-state Potts model,
J. Phys. A 31 (1998) 58275842; cond-mat 9804117.
[23] E.K. Sklyanin, Boundary conditions for integrable quantum systems, J. Phys. A 21 (1988)
23752389.
[24] R.E. Behrend, P.A. Pearce, D.L. OBrien, Interaction-round-a-face models with fixed boundary
conditions: the ABF fusion hierarchy, J. Stat. Phys. 84 (1996) 148.
[25] R.E. Behrend, P.A. Pearce, A construction of solutions to reflection equations for interactionround-a-face models, J. Phys. A 29 (1996) 78277835.
[26] A. Recknagel, V. Schomerus, D-branes in Gepner models, Nucl. Phys. B 531 (1998) 185225.
771
[27] A. Recknagel, V. Schomerus, Boundary deformation theory and moduli spaces of D-branes,
Nucl. Phys. B 545 (1999) 233282.
[28] J. Fuchs, C. Schweigert, Branes: from free fields to general backgrounds, Nucl. Phys. B 530
(1998) 99136.
[29] J. Fuchs, C. Schweigert, A classifying algebra for boundary conditions, Phys. Lett. 414 (1997)
251259.
[30] J. Fuchs, C. Schweigert, Completeness of boundary conditions for the critical three-state Potts
model, Phys. Lett. B 441 (1998) 141146.
[31] J. Fuchs, C. Schweigert, Orbifold analysis of broken bulk symmetries, Phys. Lett. B 447 (1999)
266276.
[32] J. Fuchs, C. Schweigert, Symmetry breaking boundaries I. General Theory, hep-th/9902132.
[33] L. Birke, J. Fuchs, C. Schweigert, Symmetry breaking boundary conditions and WZW orbifolds,
hep-th/9905038.
[34] A. Sagnotti, Ya.S. Stanev, in: DESY Meeting on Conformal Field Theory of D-Branes,
Hamburg, September 712, 1998.
[35] R.E. Behrend, P.A. Pearce, J.-B. Zuber, Integrable boundaries, conformal boundary conditions
and A-D-E fusion rules, J. Phys. A 31 (1998) L763L770; hep-th/9807142.
[36] R.E. Behrend, P.A. Pearce, V. Petkova, J.-B. Zuber, On the classification of bulk and boundary
conformal field theories, Phys. Lett. B 444 (1998) 163166.
[37] G. Moore, N. Seiberg, Classical and quantum conformal field theory, Commun. Math. Phys. 123
(1989) 177254.
[38] G. Moore, N. Seiberg, Lectures on RCFT, Physics, Geometry and Topology, Plenum Press, New
York, 1990.
[39] E. Verlinde, Fusion rules and modular transformations in 2-D conformal field theory, Nucl.
Phys. B 300 (1988) 360376.
[40] H. Sonoda, The energymomentum tensor on a Riemann surface, Nucl. Phys. B 281 (1987)
546572.
[41] H. Sonoda, Functional determinants on a punctured Riemann surface and their application to
string theory, Nucl. Phys. B 284 (1987) 157192.
[42] J.L. Cardy, Conformal invariance and surface critical behavior, Nucl. Phys. B 240 (1984) 514
532.
[43] J.L. Cardy, Effect of boundary conditions on the operator content of two-dimensional
conformally invariant theories, Nucl. Phys. B 275 (1986) 200218.
[44] M. Kato, T. Okada, D-branes on group manifolds, Nucl. Phys. B 499 (1997) 583595.
[45] V. Kac, Infinite-Dimensional Lie Algebras, Cambridge Univ. Press, 1995.
[46] P. Di Francesco, J.-B. Zuber, SU(N ) lattice integrable models associated with graphs, Nucl.
Phys. B 338 (1990) 602646.
[47] F.M. Goodman, P. de la Harpe, V.F.R. Jones, Coxeter Dynkin Diagrams and Towers of Algebras,
MSRI Publications, Vol. 14, Springer, Berlin, 1989.
[48] A. Cappelli, C. Itzykson, J.-B. Zuber, Modular invariant partition functions in two dimensions,
Nucl. Phys. B 280 (1987) 445465.
(1)
[49] A. Cappelli, C. Itzykson, J.-B. Zuber, The A-D-E classification of minimal and A1 conformal
invariant theories, Commun. Math. Phys. 113 (1987) 126.
[50] A. Kato, Classification of modular invariant partition functions in two dimensions, Mod. Phys.
Lett. A 2 (1987) 585600.
[51] I.M. Gelfand, M.I. Graev, N.Ya. Vilenkin, Generalised Functions, Vol. 5, Academic Press, New
York, 1966, Ch. VII.
[52] G.E. Andrews, R.J. Baxter, P.J. Forrester, Eight-vertex SOS model and generalized Rogers
Ramanujan-type identities, J. Stat. Phys. 35 (1984) 193266.
[53] V. Pasquier, Two-dimensional critical systems labelled by Dynkin diagrams, Nucl. Phys. B 285
(1987) 162172.
772
[54] P. Roche, On the construction of integrable dilute A-D-E models, Phys. Lett. B 285 (1992)
4953.
[55] S.O. Warnaar, B. Nienhuis, Solvable lattice models labelled by Dynkin diagrams, J. Phys. A 26
(1993) 23012316.
[56] P. Forrester, R. Baxter, Further exact solutions of the eight-vertex SOS model and generalizations of RogersRamanujan identities, J. Stat. Phys. 38 (1885) 435472.
[57] P.A. Pearce, Y.-K. Zhou, Intertwiners and A-D-E lattice models, Int. J. Mod. Phys. B 7 (1993)
36493705.
[58] V. Pasquier, H. Saleur, Common structures between finite size systems and conformal field
theories through quantum groups, Nucl. Phys. B 330 (1990) 523556.
[59] V. Pasquier, Operator content of the ADE lattice models, J. Phys. A 20 (1987) 57075717,
and the preprint version of that article, reprinted in: V. Pasquier, Modles Exacts Invariants
Conformes, Thse dEtat, Orsay, 1988.
[60] P. Di Francesco, J.-B. Zuber, SU(N ) lattice integrable models and modular invariance, in:
S. Randjbar-Daemi, E. Sezgin, J.-B. Zuber (Eds.), Recent Developments in Conformal Field
Theories, Trieste Conference, 1989, World Scientific, 1990.
[61] P. Di Francesco, Integrable lattice models, graphs and modular invariant conformal field
theories, Int. J. Mod. Phys. A 7 (1992) 407500.
[62] V.B. Petkova, J.-B. Zuber, On structure constants of sl(2) theories, Nucl. Phys. B 438 (1995)
347372.
[63] V.B. Petkova, J.-B. Zuber, From CFT to graphs, Nucl. Phys. B 463 (1996) 161193.
[64] V.B. Petkova, J.-B. Zuber, Conformal field theory and graphs, hep-th/9701103.
[65] F. Xu, New braided endomorphisms from conformal inclusions, Commun. Math. Phys. 192
(1998) 349403.
[66] J. Bckenhauer, D.E. Evans, Modular invariants, graphs and -induction for nets of subfactors
II, Commun. Math. Phys. 200 (1999) 57103.
[67] J. Bckenhauer, D.E. Evans, Modular invariants, graphs and -induction for nets of subfactors
III, hep-th/9812110.
[68] J. Bckenhauer, D.E. Evans, Y. Kawahigashi, On -induction, chiral generators and modular
invariants for subfactors, math-OA/9904109.
[69] A. Honecker, Automorphisms of W-algebras and extended rational conformal field theories,
Nucl. Phys. B 400 (1993) 574596.
[70] J.-B. Zuber, Generalized Dynkin diagrams and root systems, in: M. Kashiwara, A. Matsuo,
K. Saito, I. Satake (Eds.), Topological Field Theory, Primitive Forms and Related Topics,
Taniguchi Symposium, Kyoto, 1996, Birkhuser, 1998.
[71] C. Itzykson, Level one KacMoody characters and modular invariance, Nucl. Phys. Proc.
Suppl. 5B (1988) 150165.
[72] P. Degiovanni, Z/NZ conformal field theories, Commun. Math. Phys. 127 (1990) 7199.
[73] I. Kostov, Free field presentation of the An coset models on the torus, Nucl. Phys. B 300 (1988)
559587.
[74] R. Dijkgraaf, C. Vafa, E. Verlinde, H. Verlinde, Operator algebra of orbifold models, Commun.
Math. Phys. 123 (1989) 485526.
[75] P. Di Francesco, P. Mathieu, D. Snchal, Conformal Field Theory, Springer, 1997.
[76] G. Bhm, K. Szlachanyi, A coassociative C -quantum group with non-integral dimensions,
Lett. Math. Phys. 38 (1996) 437456; q-alg/9509008.
[77] P. Roche, Ocneanu cell calculus and integrable lattice models, Commun. Math. Phys. 127 (1990)
395424.
[78] Y.-K. Zhou, P.A. Pearce, Fusion of ADE lattice models, Int. J. Mod. Phys. B 8 (1994) 3531
3577.
[79] A.B. Zamolodchikov, V.A. Fateev, Operator algebra and correlation functions in the twodimensional SU(2) SU(2) WessZumino model, Sov. J. Nucl. Phys. 43 (1986) 657664.
773
[80] Vl.S. Dotsenko, V.A. Fateev, Four-point correlation functions and the operator algebra in 2d
conformal invariant theories with central charge c 6 1, Nucl. Phys. B 251 (1985) 691734.
[81] I. Runkel, Boundary structure constants for the A-series Virasoro minimal models, Nucl. Phys.
B 549 (1999) 563578.
[82] E. Bannai, T. Ito, Algebraic Combinatorics I: Association Schemes, Benjamin-Cummings,
1984.
[83] P. Bntay, P. Vecsernys, Mapping class group representations and generalised Verlinde
formula, Int. J. Mod. Phys. A 14 (1999) 1325; hep-th/9506186.
[84] A. Ocneanu, Paths on Coxeter diagrams: from platonic solids and singularities to minimal
models and subfactors, in: Low Dimensional Topology, Statistical Mechanics and Quantum
Field Theory, 1995.
[85] P. Dorey, Partition functions, intertwiners and the Coxeter element, Int. J. Mod. Phys. A 8 (1993)
193208.
[86] J. McKay, Graphs, singularities, and finite groups, Proc. Symp. Pure Math. 37 (1980) 183186.
[87] B. Kostant, The McKay correspondence, the Coxeter element and representation theory, in:
Astrisque, Socit Mathmatique de France, 1988, pp. 209255.
[88] H. Braden, E. Corrigan, P. Dorey, R. Sasaki, Affine Toda field theory and exact S matrices,
Nucl. Phys. B 338 (1990) 689746.
[89] B.M. McCoy, W. Orrick, Single particle excitations in the lattice E8 Ising model, Phys. Lett.
A 230 (1997) 2432.
[90] M.T. Batchelor, K.A. Seaton, Excitations in the dilute AL lattice model: E6 , E7 and E8 mass
spectra, cond-mat/9803206.
[91] J. Suzuki, Quantum JacobiTrudi formula and E8 structure in the Ising model in a field, condmat/9805241.
[92] T. Gannon, The classification of affine SU(3) modular invariant partition functions, Commun.
Math. Phys. 161 (1994) 233263.
[93] T. Gannon, The classification of SU(3) modular invariant revisited, Ann. Inst. H. Poincar, Phys.
Theor. 65 (1996) 1556; hep-th/9404185.
[94] I. Runkel, Structure constants for the D-series Virasoro minimal models, hep-th/9908046.
[95] P. Ruelle, Symmetric boundary conditions in boundary critical phenomena, hep-th/9904100.
[96] P. Di Francesco, J.-B. Zuber, unpublished, reported in Ref. [62], see also [86].
[97] A.N. Kirillov, N.Yu. Reshetikhin, in: T. Kohno (Ed.), New Developments of the Theory of
Knots, World Scientific, 1990.
[98] J. Bckenhauer, D.E. Evans, Y. Kawahigashi, Chiral structure of modular invariants for
subfactors, math-OA 9907143.

CUMULATIVE AUTHOR INDEX B571B579
Accomando, E.
Aguilar-Saavedra, J.A.
Ahn, C.
Aitchison, I.J.R.
Akchurin, N.
Akemann, G.
Akutsu, Y.
Alexandrou, C.
Alfieri, R.
Alford, M.
Alford, M.
Alimohammadi, M.
Allahverdi, R.
Alls, B.
ALPHA Collaborations
Altarelli, G.
lvarez, E.
Anastasiou, C.
Anastasiou, C.
Andreev, V.A.
Angelantonj, C.
Angelantonj, C.
Antoniadis, I.
Antoniadis, I.
Aoyama, S.
Ardalan, F.
Arefeva, I.Ya.
Arfaei, H.
Armoni, A.
Arnsdorf, M.
Arutyunov, G.
Aschieri, P.
Atamantchouk, A.G.
Aykac, M.
Azaria, P.
B579 (2000) 3
B576 (2000) 56
B572 (2000) 188
B578 (2000) 199
B579 (2000) 277
B576 (2000) 597
B575 (2000) 504
B571 (2000) 257
B578 (2000) 383
B571 (2000) 269
B578 (2000) 367
B577 (2000) 609
B579 (2000) 355
B576 (2000) 658
B571 (2000) 237
B575 (2000) 313
B574 (2000) 153
B572 (2000) 307
B575 (2000) 416
B579 (2000) 277
B572 (2000) 36
B578 (2000) 239
B572 (2000) 36
B579 (2000) 3
B578 (2000) 449
B576 (2000) 578
B579 (2000) 411
B576 (2000) 578
B578 (2000) 239
B577 (2000) 529
B579 (2000) 117
B574 (2000) 551
B579 (2000) 277
B579 (2000) 277
B575 (2000) 439
Bak, D.
Bakas, I.
Balatz, M.Y.
Ball, P.
Ball, R.D.
Bandelloni, G.
Barbieri, R.
B572 (2000) 151

B573 (2000) 768
B579 (2000) 277
B572 (2000) 3
B575 (2000) 313
B577 (2000) 471
B575 (1999) 61
Barbieri, R.
Barnes, K.J.
Bassi, Z.S.
Bastero-Gil, M.
Bastianelli, F.
Bastianelli, F.
Becker, K.
Becker, M.
Beenakker, W.
Behrend, R.E.
Behrndt, K.
Blanger, G.
Belitsky, A.V.
Belitsky, A.V.
Benakli, K.
Berche, B.
Berends, F.A.
Berges, J.
Bialas, P.
Bialas, P.
Bianchi, M.
Bilal, A.
Bill, M.
Binoth, T.
Blas Achic, H.S.
Bobeth, C.
Bocquet, M.
Bode, A.
Bogacz, L.
Bondar, N.F.
Bonora, L.
Bordag, M.
Boudjema, F.
Bouwknegt, P.
Brace, D.
Braden, H.W.
Brahmachari, B.
Bravar, A.
Brignole, A.
Brower, R.C.
Buchbinder, I.L.
Buchmller, W.
B578 (2000) 153

B575 (2000) 397
B578 (2000) 577
B575 (1999) 35
B574 (2000) 107
B578 (2000) 139
B577 (2000) 156
B577 (2000) 156
B573 (2000) 503
B579 (2000) 707
B573 (2000) 127
B576 (2000) 660
B574 (2000) 347
B574 (2000) 407
B579 (2000) 3
B572 (2000) 626
B573 (2000) 503
B571 (2000) 269
B575 (2000) 599
B575 (2000) 645
B573 (2000) 314
B576 (2000) 347
B576 (2000) 241
B572 (2000) 361
B571 (2000) 607
B574 (2000) 291
B578 (2000) 628
B576 (2000) 517
B575 (2000) 599
B579 (2000) 277
B578 (2000) 497
B576 (2000) 430
B576 (2000) 660
B572 (2000) 547
B574 (2000) 551
B573 (2000) 553
B575 (1999) 35
B579 (2000) 277
B579 (2000) 101
B574 (2000) 219
B571 (2000) 358
B576 (2000) 445
776
Bueno, A.
Burda, Z.
B573 (2000) 27
B575 (2000) 599
Cacciari, M.
Campanelli, M.
Campbell, B.A.
Carena, M.
Casas, J.A.
Caselle, M.
Caselle, M.
Castro-Alvaredo, O.A.
Cervera, A.
Chandrasekharan, S.
Chapovsky, A.P.
Chatelain, C.
Chattaraputi, A.
Chen, B.
Chen, B.
Cheng, H.-C.
Chensheng, M.
Chetyrkin, K.G.
Chim, L.
Cho, G.-C.
Chu, C.-S.
Ciuchini, M.
Ciuchini, M.
Cline, J.
Contino, R.
Controzzi, D.
Cooper, P.S.
Cox, J.
Creminelli, P.
Cvetic, M.
Cvetic, M.
Cvetic, M.
Czakon, M.
B571 (2000) 185

B573 (2000) 27
B579 (2000) 355
B577 (2000) 88
B573 (2000) 652
B579 (2000) 635
B579 (2000) 667
B575 (2000) 535
B579 (2000) 17
B576 (2000) 481
B573 (2000) 503
B572 (2000) 626
B573 (2000) 291
B576 (2000) 177
B577 (2000) 23
B573 (2000) 597
B579 (2000) 277
B573 (2000) 617
B572 (2000) 547
B574 (2000) 623
B574 (2000) 275
B573 (2000) 201
B579 (2000) 56
B578 (2000) 259
B578 (2000) 153
B572 (2000) 609
B579 (2000) 277
B576 (2000) 481
B575 (1999) 61
B571 (2000) 358
B573 (2000) 149
B574 (2000) 761
B573 (2000) 57
DAdda, A.
DAppollonio, G.
Daflon Barrozo, M.C.
Damgaard, P.H.
Damgaard, P.H.
Dasgupta, T.
Dauwe, L.J.
Davidenko, G.V.
Dedes, A.
de Forcrand, Ph.
Deger, N.S.
De Holanda, P.C.
del Aguila, F.
Delduc, F.
Delduc, F.
Del Duca, V.
Del Duca, V.
Delepine, D.
De Pietri, R.
B576 (2000) 241

B572 (2000) 36
B574 (2000) 189
B572 (2000) 478
B576 (2000) 597
B572 (2000) 95
B579 (2000) 277
B579 (2000) 277
B576 (2000) 29
B577 (2000) 263
B573 (2000) 275
B573 (2000) 3
B576 (2000) 56
B576 (2000) 196
B577 (2000) 461
B571 (2000) 51
B574 (2000) 851
B576 (2000) 445
B574 (2000) 785
Derendinger, J.-P.
Derkachov, S..
Dersch, U.
Dhar, A.
Diaconescu, D.-E.
Diakonov, D.
Daz, M.A.
Di Renzo, F.
Dirkes, G.
Dixon, L.
Dobrescu, B.A.
Dolgolenko, A.G.
Donini, A.
Donini, A.
Dorey, P.
Dorey, P.
Dorey, P.
Dreiner, H.
Dreossi, D.
Dressler, B.
Dudas, E.
Dudas, E.
Dunning, C.
Dzyubenko, G.B.
B576 (2000) 347

B579 (2000) 56
B579 (2000) 277
B575 (1999) 177
B574 (2000) 245
B571 (2000) 91
B573 (2000) 75
B578 (2000) 383
B579 (2000) 277
B571 (2000) 51
B573 (2000) 597
B579 (2000) 277
B574 (2000) 23
B579 (2000) 17
B571 (2000) 583
B578 (2000) 85
B578 (2000) 699
B574 (2000) 874
B579 (2000) 277
B578 (2000) 293
B572 (2000) 36
B575 (1999) 3
B578 (2000) 699
B579 (2000) 277
Edelstein, J.D.
Edelstein, R.
Eguchi, T.
Ellis, J.
Emediato, L.
Emparan, R.
Endler, A.M.F.
Engelfried, J.
Engels, J.
Engels, J.
Ennes, I.P.
Ermolaev, B.I.
Eschrich, I.
Escobar, C.O.
Espinosa, J.R.
Evans, D.H.
Evans, T.S.
Evdokimov, A.V.
Eyras, E.
B574 (2000) 587

B579 (2000) 277
B577 (2000) 3
B579 (2000) 355
B579 (2000) 277
B573 (2000) 291
B579 (2000) 277
B579 (2000) 277
B572 (2000) 289
B576 (2000) 657
B576 (2000) 313
B571 (2000) 137
B579 (2000) 277
B579 (2000) 277
B573 (2000) 652
B577 (2000) 240
B577 (2000) 325
B579 (2000) 277
B573 (2000) 735
Fabbri, D.
Fendley, P.
Ferrandis, J.
Ferreira, L.A.
Filimonov, I.S.
Fioravanti, D.
Fleischer, J.
Forte, S.
Fosco, C.D.
Frahm, H.
B577 (2000) 547

B574 (2000) 571
B573 (2000) 75
B571 (2000) 607
B579 (2000) 277
B577 (2000) 500
B571 (2000) 511
B575 (2000) 313
B578 (2000) 199
B575 (2000) 485

Franco, E.
Fr, P.
Freidel, L.
Frre, J.-M.
Freund, A.
Freund, M.
Fring, A.
Fring, A.
Frixione, S.
Frizzo, A.
Frolov, S.
Frolov, S.
Fujii, A.
Fujii, Y.
Fujikawa, K.
Fukae, M.
Fukahori, T.
Fukui, T.
B573 (2000) 201

B577 (2000) 547
B574 (2000) 785
B572 (2000) 3
B574 (2000) 347
B578 (2000) 27
B575 (2000) 535
B579 (2000) 617
B571 (2000) 169
B579 (2000) 379
B578 (2000) 139
B579 (2000) 117
B574 (2000) 691
B573 (2000) 377
B577 (2000) 405
B572 (2000) 71
B573 (2000) 377
B575 (2000) 673
Gaberdiel, M.R.
Gaillard, M.K.
Ganchev, A.Ch.
Garcia, D.
Garcia, F.G.
Garca Prez, M.
Garden, J.
Garousi, M.R.
Gaspero, M.
Gavela, M.B.
Gavela, M.B.
Gehrmann-De Ridder, A.
Gerzon, S.
Giller, I.
Girlanda, L.
Giusti, L.
Glover, E.W.N.
Glover, E.W.N.
Gluza, J.
Golovtsov, V.L.
Gmez, C.
Gmez-Reino, M.
Gomez Cdenas, J.J.
Goncharenko, Y.M.
Gonzalez-Garcia, M.C.
Gorsky, A.
Gottschalk, E.
Gouffon, P.
Gracey, J.A.
Grachov, O.A.
Greco, M.
Griehammer, H.W.
Grinstein, B.
Grinza, P.
Grojean, C.
Groves, M.
Gualtieri, L.
B578 (2000) 58
B571 (2000) 3
B571 (2000) 457
B577 (2000) 88
B579 (2000) 277
B577 (2000) 263
B571 (2000) 237
B579 (2000) 209
B579 (2000) 277
B574 (2000) 23
B579 (2000) 17
B578 (2000) 326
B579 (2000) 277
B579 (2000) 277
B575 (2000) 285
B573 (2000) 201
B572 (2000) 307
B575 (2000) 416
B573 (2000) 57
B579 (2000) 277
B574 (2000) 153
B574 (2000) 587
B579 (2000) 17
B579 (2000) 277
B573 (2000) 3
B571 (2000) 120
B579 (2000) 277
B579 (2000) 277
B579 (2000) 56
B579 (2000) 277
B571 (2000) 137
B579 (2000) 313
B577 (2000) 240
B579 (2000) 635
B578 (2000) 259
B573 (2000) 449
B577 (2000) 547
777
Guillet, J.Ph.
Gukov, S.
Glmez, E.
Gnaydin, M.
Gnaydin, M.
Gupta, S.
Gyulassy, M.
B572 (2000) 361

B574 (2000) 169
B579 (2000) 277
B572 (2000) 131
B578 (2000) 405
B577 (2000) 529
B571 (2000) 197
Haack, M.
Hagiwara, K.
Hammou, A.B.
Handoko, L.T.
Hara, Y.
Hart, A.
Hasenbusch, M.
Hashimoto, T.
Hastings, M.B.
Hatayama, G.
Hatsuda, M.
Hebecker, A.
Heinrich, G.
Heinrich, G.
Heitger, J.
Hernndez, P.
Hernndez, P.
Hidaka, H.
Hieida, Y.
Hill, C.T.
Hioki, S.
Ho, P.-M.
Ho, P.-M.
Holland, K.
Hollowood, T.J.
Hoppe, J.
Hou, B.
Houart, L.
Howe, P.S.
Hsu, S.D.H.
Huiszoon, L.R.
Hull, C.M.
B575 (1999) 107

B574 (2000) 623
B573 (2000) 335
B576 (2000) 445
B572 (2000) 574
B572 (2000) 243
B579 (2000) 667
B577 (2000) 263
B572 (2000) 535
B577 (2000) 619
B577 (2000) 183
B571 (2000) 26
B572 (2000) 361
B575 (2000) 359
B571 (2000) 237
B574 (2000) 23
B579 (2000) 17
B573 (2000) 377
B575 (2000) 504
B573 (2000) 597
B577 (2000) 263
B573 (2000) 364
B574 (2000) 275
B576 (2000) 481
B575 (1999) 78
B571 (2000) 479
B575 (2000) 561
B575 (1999) 195
B571 (2000) 71
B572 (2000) 211
B575 (2000) 401
B575 (1999) 231
Ibez, L.E.
Ibarra, A.
Ichie, H.
Ichinose, I.
Ichinose, I.
Ichinose, S.
Iida, S.
Iori, M.
Ishibashi, M.
Ishibashi, N.
Iso, S.
Iso, S.
Itoyama, H.
Itoyama, H.
B576 (2000) 285

B573 (2000) 652
B574 (2000) 70
B575 (2000) 613
B577 (2000) 279
B574 (2000) 719
B573 (2000) 685
B579 (2000) 277
B576 (2000) 501
B573 (2000) 573
B573 (2000) 573
B576 (2000) 375
B576 (2000) 177
B577 (2000) 23
778
Ivanov, E.
Ivanov, E.
Ivin, M.
Iwamoto, A.
B576 (2000) 196

B576 (2000) 543
B577 (2000) 325
B573 (2000) 377
Jaffe, R.L.
Janke, W.
Jegerlehner, F.
Jevicki, A.
Johnston, D.
Johnston, D.A.
Johnstone, G.
Jonsson, T.
Jun, S.Y.
B578 (2000) 367

B578 (2000) 681
B571 (2000) 511
B577 (2000) 47
B575 (2000) 599
B578 (2000) 681
B577 (2000) 646
B575 (2000) 661
B579 (2000) 277
Kabat, D.
Kac, V.G.
Kaczmarek, O.
Kamenskii, A.D.
Kangling, H.
Kar, S.
Karsch, F.
Kataev, A.L.
Kawai, H.
Kawai, H.
Kawamoto, N.
Kaya, A.
Kaya, M.
Kazakov, D.
Kazakov, K.A.
Kazakov, V.
Kemmoku, R.
Kharzeev, D.
Khorrami, M.
Khoze, V.V.
Khuri, R.R.
Kihara, H.
Kikukawa, Y.
Kilgore, W.B.
Kilmer, J.
Kim, V.T.
King, S.F.
Kitao, T.
Kitazawa, Y.
Kitazawa, Y.
Klebanov, I.R.
Klebanov, I.R.
Klinkhamer, F.R.
Kneipp, M.A.C.
Kniehl, B.A.
Kochenda, L.M.
Kokorelis, C.
Komori, Y.
Knigsmann, K.
Konorov, I.
Korff, C.
B571 (2000) 419

B571 (2000) 515
B576 (2000) 657
B579 (2000) 277
B579 (2000) 277
B577 (2000) 171
B576 (2000) 657
B573 (2000) 405
B573 (2000) 573
B576 (2000) 375
B574 (2000) 809
B573 (2000) 275
B579 (2000) 277
B577 (2000) 121
B573 (2000) 536
B571 (2000) 479
B574 (2000) 691
B578 (2000) 351
B577 (2000) 609
B575 (1999) 78
B575 (1999) 231
B577 (2000) 23
B576 (2000) 501
B574 (2000) 851
B579 (2000) 277
B579 (2000) 277
B576 (2000) 85
B578 (2000) 215
B573 (2000) 573
B576 (2000) 375
B574 (2000) 263
B578 (2000) 123
B578 (2000) 277
B577 (2000) 390
B577 (2000) 197
B579 (2000) 277
B579 (2000) 267
B571 (2000) 632
B579 (2000) 277
B579 (2000) 277
B575 (2000) 535
Korff, C.
Koshelev, A.S.
Kostov, I.K.
Kostov, I.K.
Kovchegov, Y.V.
Kozhevnikov, A.P.
Kramer, G.
Krmer, M.
Krasnov, K.
Krivonos, S.
Krivshich, A.G.
Krger, H.
Kubantsev, M.A.
Kubarovsky, V.P.
Kubo, J.
Kulyavtsev, A.I.
Kuniba, A.
Kuropatkin, N.P.
Kurshetsov, V.F.
Kushnirenko, A.
Kwan, S.
B579 (2000) 617

B579 (2000) 411
B571 (2000) 479
B575 (2000) 513
B577 (2000) 221
B579 (2000) 277
B578 (2000) 326
B571 (2000) 169
B574 (2000) 785
B576 (2000) 196
B579 (2000) 277
B579 (2000) 277
B579 (2000) 277
B579 (2000) 277
B574 (2000) 495
B579 (2000) 277
B577 (2000) 619
B579 (2000) 277
B579 (2000) 277
B579 (2000) 277
B579 (2000) 277
Lach, J.
Laenen, E.
Laermann, E.
Lalak, Z.
Lalak, Z.
Lamberto, A.
Landsberg, L.G.
Langfeld, K.
Larin, I.
Larsen, F.
Lavignac, S.
Lazzarini, S.
Lecheminant, P.
LeClair, A.
Leibbrandt, G.
Leikin, E.M.
Lvai, P.
Levin, E.
Levin, E.
Levin, E.
Li, M.
Li, M.
Liang, J.-Q.
Liang, J.-Q.
Lifschytz, G.
Lima, E.
Lindner, M.
Louis, J.
Lozano, C.
Lozano, Y.
Lozano, Y.
L, H.
L, H.
B579 (2000) 277

B571 (2000) 169
B576 (2000) 657
B575 (1999) 151
B576 (2000) 399
B579 (2000) 277
B579 (2000) 277
B572 (2000) 266
B579 (2000) 277
B575 (1999) 211
B576 (2000) 399
B577 (2000) 471
B575 (2000) 439
B578 (2000) 577
B575 (2000) 359
B579 (2000) 277
B571 (2000) 197
B573 (2000) 833
B577 (2000) 221
B578 (2000) 351
B574 (2000) 275
B579 (2000) 525
B578 (2000) 728
B579 (2000) 177
B571 (2000) 419
B572 (2000) 112
B578 (2000) 27
B575 (1999) 107
B576 (2000) 313
B573 (2000) 735
B575 (1999) 195
B572 (2000) 112
B573 (2000) 149

L, H.
Lu, J.X.
Lubicz, V.
Luksys, M.
Lunghi, E.
Lungov, T.
B574 (2000) 761

B579 (2000) 229
B573 (2000) 201
B579 (2000) 277
B574 (2000) 43
B579 (2000) 277
Ma, B.-Q.
Maeshima, N.
Magarrel, D.
Magnea, L.
Magnoli, N.
Maillet, J.M.
Maleev, V.P.
Maltoni, F.
Maltoni, F.
Manashov, A.N.
Mandal, G.
Manvelyan, R.
Mao, D.
Mario, M.
Marshakov, A.
Martn, C.P.
Martinelli, G.
Mas, J.
Masciocchi, S.
Massar, S.
Mateos, D.
Mathew, P.
Mathur, S.D.
Matias, J.
Matsufuru, H.
Matsuo, T.
Mattson, M.
Matveev, V.
Maul, M.
Maul, M.
Maxwell, C.J.
McArthur, I.N.
McCliment, E.
McInnes, B.
McKenna, S.L.
Medvedev, P.B.
Meggiolaro, E.
Mena, O.
Mendes, T.
Mihailescu, M.
Minasian, R.
Minces, P.
Miramontes, J.L.
Mirjalili, A.
Mironov, A.
Misiak, M.
Miyamura, O.
Mizoguchi, S.
Mbius, M.
B574 (2000) 331

B575 (2000) 504
B579 (2000) 277
B579 (2000) 379
B579 (2000) 635
B575 (2000) 627
B579 (2000) 277
B571 (2000) 51
B574 (2000) 851
B579 (2000) 56
B575 (1999) 177
B579 (2000) 177
B579 (2000) 277
B574 (2000) 587
B573 (2000) 553
B572 (2000) 387
B573 (2000) 201
B574 (2000) 587
B579 (2000) 277
B575 (2000) 333
B577 (2000) 139
B579 (2000) 277
B574 (2000) 219
B572 (2000) 3
B577 (2000) 263
B576 (2000) 177
B579 (2000) 277
B579 (2000) 277
B571 (2000) 91
B578 (2000) 293
B577 (2000) 209
B573 (2000) 811
B579 (2000) 277
B577 (2000) 439
B579 (2000) 277
B579 (2000) 411
B571 (2000) 26
B579 (2000) 17
B572 (2000) 289
B577 (2000) 47
B572 (2000) 499
B572 (2000) 651
B575 (2000) 535
B577 (2000) 209
B573 (2000) 553
B574 (2000) 291
B577 (2000) 263
B574 (2000) 691
B577 (2000) 325
779
Moch, S.
Mohapatra, R.N.
Moinester, M.A.
Molchanov, V.V.
Mller, P.
Morales, J.F.
Morales, J.F.
Morariu, B.
Morawitz, P.
Morelos, A.
Moretti, S.
Morozov, A.
Morris, T.R.
Moultaka, G.
Mourad, J.
Mueller, A.H.
Mukhin, V.A.
Mukhopadhyay, S.
Mller, D.
Mller-Kirsten, H.J.W.
Mller-Kirsten, H.J.W.
Multamki, T.
Murakami, K.
Mussardo, G.
B573 (2000) 853

B576 (2000) 466
B579 (2000) 277
B579 (2000) 277
B573 (2000) 377
B573 (2000) 314
B573 (2000) 335
B574 (2000) 551
B574 (2000) 874
B579 (2000) 277
B576 (2000) 29
B573 (2000) 553
B573 (2000) 97
B577 (2000) 121
B575 (1999) 3
B572 (2000) 227
B579 (2000) 277
B576 (2000) 152
B574 (2000) 347
B578 (2000) 728
B579 (2000) 177
B574 (2000) 130
B576 (2000) 177
B578 (2000) 527
Nachtmann, O.
Naculich, S.G.
Nagao, K.
Nakagawa, M.
Nakamura, A.
Nambu, Y.
Nauta, B.J.
Nauta, B.J.
Navarro, I.
Navarro, P.
Navarro-Salas, J.
Nekrasov, N.A.
Nelson, B.D.
Nelson, K.D.
Nemitkin, A.V.
Neoustroev, P.V.
Newsom, C.
Niedermaier, M.
Nierste, U.
Nilles, H.P.
Nilov, A.P.
Nishino, A.
Nishino, T.
Noguchi, T.
Nolte, D.R.
Nurushev, S.B.
B571 (2000) 26
B576 (2000) 313
B577 (2000) 279
B573 (2000) 377
B577 (2000) 263
B579 (2000) 590
B571 (2000) 151
B575 (2000) 383
B573 (2000) 652
B579 (2000) 250
B579 (2000) 250
B574 (2000) 263
B571 (2000) 3
B579 (2000) 277
B579 (2000) 277
B579 (2000) 277
B579 (2000) 277
B579 (2000) 437
B577 (2000) 88
B576 (2000) 399
B579 (2000) 277
B571 (2000) 632
B575 (2000) 504
B576 (2000) 501
B577 (2000) 240
B579 (2000) 277
Ocherashvili, A.
Ohnuki, T.
Ohta, N.
B579 (2000) 277

B573 (2000) 377
B578 (2000) 215
780
Okunishi, K.
Oleari, C.
Oleari, C.
Oleynik, G.
Onel, Y.
Onofri, E.
Ooguri, H.
Orland, P.
Osborn, H.
Oura, Y.
Ovrut, B.A.
Ozel, E.
Ozkorucuklu, S.
B575 (2000) 504

B572 (2000) 307
B575 (2000) 416
B579 (2000) 277
B579 (2000) 277
B578 (2000) 383
B577 (2000) 419
B576 (2000) 627
B571 (2000) 287
B573 (2000) 377
B572 (2000) 112
B579 (2000) 277
B579 (2000) 277
Panagopoulos, H.
Parentani, R.
Parente, G.
Park, D.K.
Passarino, G.
Passarino, G.
Patrichev, S.
Pearce, P.A.
Pelissetto, A.
Pea-Garay, C.
Peng, D.
Penin, A.A.
Penzo, A.
Pepe, M.
Perazzi, E.
Prez-Lorenzana, A.
Perkins, W.B.
Pernici, M.
Petcov, S.T.
Petkova, V.B.
Petkova, V.B.
Petrenko, S.I.
Petrov, A.Yu.
Philipsen, O.
Pillin, M.
Pogodin, P.
Pope, C.N.
Pope, C.N.
Pope, C.N.
Pospelov, M.
Povh, B.
Pradisi, G.
Pradisi, G.
Procario, M.
Pronin, P.I.
Provero, P.
Prutskoi, V.A.
B571 (2000) 257

B575 (2000) 333
B573 (2000) 405
B578 (2000) 728
B574 (2000) 451
B578 (2000) 3
B579 (2000) 277
B579 (2000) 707
B575 (2000) 579
B573 (2000) 3
B575 (2000) 561
B577 (2000) 197
B579 (2000) 277
B576 (2000) 658
B574 (2000) 3
B576 (2000) 466
B573 (2000) 449
B577 (2000) 293
B578 (2000) 27
B571 (2000) 457
B579 (2000) 707
B579 (2000) 277
B571 (2000) 358
B572 (2000) 243
B579 (2000) 535
B579 (2000) 277
B572 (2000) 112
B573 (2000) 149
B574 (2000) 761
B573 (2000) 177
B579 (2000) 277
B573 (2000) 314
B575 (1999) 134
B579 (2000) 277
B573 (2000) 536
B576 (2000) 241
B579 (2000) 277
Rabadn, R.
Raciti, M.
Rajagopal, K.
Ramberg, E.
B576 (2000) 285

B577 (2000) 293
B571 (2000) 269
B579 (2000) 277
Ramgoolam, S.
Ramgoolam, S.
Rapazzo, G.F.
Rasmussen, J.
Rattazzi, R.
Ray, K.
Razmyslovich, B.V.
Reina, C.
Reisz, T.
Rey, S.-J.
Rey, S.-J.
Ridolfi, G.
Ridout, D.
Rigolin, S.
Rigolin, S.
Rinaldi, M.
Ritz, A.
Riva, F.
Rivelles, V.O.
Romanino, A.
Romanino, A.
Rmelsberger, C.
Rothe, H.J.
Rovelli, C.
Roy, S.
Rubbia, A.
Rud, V.I.
Runkel, I.
Runkel, I.
Russ, J.
Russo, R.
B573 (2000) 364

B577 (2000) 47
B579 (2000) 277
B574 (2000) 525
B576 (2000) 3
B576 (2000) 152
B579 (2000) 277
B577 (2000) 547
B575 (2000) 255
B572 (2000) 151
B572 (2000) 188
B574 (2000) 3
B572 (2000) 547
B574 (2000) 23
B579 (2000) 17
B578 (2000) 497
B573 (2000) 177
B577 (2000) 293
B572 (2000) 651
B574 (2000) 675
B578 (2000) 27
B574 (2000) 245
B575 (2000) 255
B574 (2000) 785
B579 (2000) 229
B573 (2000) 27
B579 (2000) 277
B578 (2000) 85
B579 (2000) 561
B579 (2000) 277
B579 (2000) 379
Sagnotti, A.
Sakaguchi, M.
Saleur, H.
Saleur, H.
Samtleben, H.
Snchez-Ruiz, D.
Sato, H.-T.
Sato, N.
Sauser, R.
Savvidy, G.K.
Schfer, T.
Scheglov, Y.
Schellekens, A.N.
Schiavon, P.
Schmidt, I.
Schmidt, M.G.
Schnittger, J.
Schnitzer, H.J.
Schubert, C.
Schwetz, M.
Scimemi, I.
Scorzato, L.
Selivanov, K.
B572 (2000) 36
B577 (2000) 183
B574 (2000) 571
B578 (2000) 552
B579 (2000) 437
B572 (2000) 387
B579 (2000) 492
B574 (2000) 809
B576 (2000) 347
B575 (2000) 661
B575 (2000) 269
B579 (2000) 277
B575 (2000) 401
B579 (2000) 277
B574 (2000) 331
B579 (2000) 492
B574 (2000) 525
B576 (2000) 313
B571 (2000) 71
B572 (2000) 211
B574 (2000) 43
B578 (2000) 383
B571 (2000) 120

Semenoff, G.W.
Semyatchkin, V.K.
Serban, D.
Servant, G.
Sethi, S.
Sezgin, E.
Sfetsos, K.
Shafi, Q.
Sheikh-Jabbari, M.M.
Shin, G.
Shore, G.M.
Shuchen, Z.
Shuster, E.
Sidorov, A.V.
Sierra, G.
Simon, J.
Simon, P.
Sitnikov, A.I.
Skalozub, V.
Skow, D.
Slavnov, N.A.
Smilga, A.V.
Smith, V.J.
Sokatchev, E.
Sommer, R.
Son, D.T.
Sorin, A.
Sousa, N.
Spiesberger, H.
Splittorff, K.
Sridhar, K.
Srivastava, M.
Stamatescu, I.-O.
Stanishkov, M.
Stefanski, Jr., B.
Stefanski, Jr., B.
Steiner, V.
Steinhauser, M.
Stelle, K.S.
Stepanov, V.
Stern, J.
Stern, M.
Strumia, A.
Strumia, A.
Strumia, A.
Stutte, L.
Su, S.
Suganuma, H.
Sugawara, Y.
Sugawara, Y.
Sundborg, B.
Sundell, P.
Suzuki, T.
Svoiski, M.
B576 (2000) 627

B579 (2000) 277
B578 (2000) 628
B578 (2000) 259
B578 (2000) 163
B573 (2000) 275
B573 (2000) 768
B573 (2000) 40
B576 (2000) 578
B572 (2000) 266
B571 (2000) 287
B579 (2000) 277
B573 (2000) 434
B573 (2000) 405
B572 (2000) 517
B579 (2000) 277
B578 (2000) 527
B579 (2000) 277
B576 (2000) 430
B579 (2000) 277
B575 (2000) 485
B571 (2000) 515
B579 (2000) 277
B571 (2000) 71
B571 (2000) 237
B573 (2000) 434
B577 (2000) 461
B575 (2000) 401
B578 (2000) 326
B572 (2000) 478
B576 (2000) 660
B579 (2000) 277
B577 (2000) 263
B577 (2000) 500
B572 (2000) 95
B578 (2000) 58
B579 (2000) 277
B573 (2000) 617
B573 (2000) 149
B579 (2000) 277
B575 (2000) 285
B578 (2000) 163
B575 (1999) 61
B576 (2000) 3
B578 (2000) 153
B579 (2000) 277
B573 (2000) 87
B574 (2000) 70
B576 (2000) 265
B577 (2000) 3
B573 (2000) 349
B573 (2000) 275
B578 (2000) 515
B579 (2000) 277
Takagi, T.
B577 (2000) 619
781
Takahashi, K.
Takaishi, T.
Takemae, S.
Tan, C.-I.
Taormina, A.
Tarasov, O.V.
Tatar, R.
Tateo, R.
Tateo, R.
Tateo, R.
Tavartkiladze, Z.
Taylor IV, W.
Terao, H.
Terashima, H.
Terentyev, N.K.
Termonia, P.
Terras, V.
Teschner, J.
Tetradis, N.
Thomas, G.P.
Thomas, S.
Thomassen, J.B.
Tomasiello, A.
Troyan, S.I.
Tseytlin, A.A.
Tseytlin, A.A.
Tsimpis, D.
Tsvelik, A.M.
Tuchin, K.
B573 (2000) 685

B577 (2000) 263
B578 (2000) 405
B574 (2000) 219
B573 (2000) 291
B571 (2000) 511
B573 (2000) 364
B571 (2000) 583
B578 (2000) 85
B578 (2000) 699
B573 (2000) 40
B573 (2000) 703
B574 (2000) 495
B577 (2000) 405
B579 (2000) 277
B577 (2000) 341
B575 (2000) 627
B571 (2000) 555
B575 (1999) 61
B579 (2000) 277
B575 (1999) 151
B578 (2000) 477
B577 (2000) 547
B571 (2000) 137
B578 (2000) 123
B578 (2000) 139
B572 (2000) 499
B572 (2000) 609
B573 (2000) 833
Uchida, Y.
Ujino, H.
UKQCD Collaborations
Umeda, T.
Uranga, A.M.
Uranga, A.M.
Urban, J.
Uvarov, L.N.
B574 (2000) 809

B571 (2000) 632
B571 (2000) 237
B577 (2000) 263
B576 (2000) 285
B577 (2000) 73
B574 (2000) 291
B579 (2000) 277
Vafa, C.
Valent, G.
Valle, J.W.F.
Valle, J.W.F.
Van Raamsdonk, M.
Van Weert, Ch.G.
Vasiliev, A.N.
Vavilov, D.V.
Verebryusov, V.S.
Veretin, O.L.
Vermaseren, J.A.M.
Vicari, E.
Vicari, E.
Victorov, V.A.
Vilja, I.
Vishnyakov, V.E.
B577 (2000) 419

B576 (2000) 543
B573 (2000) 3
B573 (2000) 75
B573 (2000) 703
B571 (2000) 151
B579 (2000) 277
B579 (2000) 277
B579 (2000) 277
B571 (2000) 511
B573 (2000) 853
B571 (2000) 257
B575 (2000) 579
B579 (2000) 277
B574 (2000) 130
B579 (2000) 277
782
Vitev, I.
Vorobyov, A.A.
Vorwalter, K.
B571 (2000) 197

B579 (2000) 277
B579 (2000) 277
Wadati, M.
Wadia, S.R.
Wagner, C.E.M.
Watts, G.
Watts, G.M.T.
Weiss, C.
Weisz, P.
Wells, J.D.
Wenheng, Z.
West, P.C.
Wiese, U.-J.
Wittig, H.
Wolff, U.
B571 (2000) 632

B575 (1999) 177
B577 (2000) 88
B578 (2000) 85
B571 (2000) 457
B578 (2000) 293
B576 (2000) 517
B576 (2000) 3
B579 (2000) 277
B571 (2000) 71
B576 (2000) 481
B571 (2000) 237
B576 (2000) 517
Yamada, A.
Yamada, Y.
Yang, J.-J.
Yang, S.-K.
Yogendran, K.P.
Yoneya, T.
B576 (2000) 501

B572 (2000) 71
B574 (2000) 331
B572 (2000) 71
B575 (1999) 177
B576 (2000) 219
You, J.
Youm, D.
Youm, D.
Youm, D.
Youm, D.
Youm, D.
Yue, R.
Yunshan, L.
B579 (2000) 277

B573 (2000) 223
B573 (2000) 257
B576 (2000) 106
B576 (2000) 123
B576 (2000) 139
B575 (2000) 561
B579 (2000) 277
Zaffaroni, A.
Zagermann, M.
Zahlten, C.
Zampa, A.
Zhang, Y.
Zhenlin, M.
Zhigang, L.
Zirnbauer, M.R.
Zoupanos, G.
Zraek, M.
Zuber, J.-B.
Zucchini, R.
Zukanovich-Funchal, R.
Zumino, B.
Zwirner, F.
B577 (2000) 547

B572 (2000) 131
B579 (2000) 492
B577 (2000) 547
B579 (2000) 177
B579 (2000) 277
B579 (2000) 277
B578 (2000) 628
B574 (2000) 495
B573 (2000) 57
B579 (2000) 707
B574 (2000) 107
B579 (2000) 277
B574 (2000) 551
B574 (2000) 3

Nucl - Phys.B v.579

Hochgeladen von

Dokumentinformationen

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Nucl - Phys.B v.579

Hochgeladen von

Copyright:

Verfügbare Formate

Nuclear Physics B 579 (2000) 316

Looking for TeV-scale strings

Received 21 January 2000; accepted 22 February 2000

E. Accomando et al. / Nuclear Physics B 579 (2000) 316

As a result, TeV-dimensions can be longitudinal along the world-volume of some branes

E. Accomando et al. / Nuclear Physics B 579 (2000) 316

2. Bulk and boundary states within the brane picture

E. Accomando et al. / Nuclear Physics B 579 (2000) 316

3. The standard (l, l, l) scenario

of massive excitations is enhanced by a factor

2 relative to the lowest (massless) one.

E. Accomando et al. / Nuclear Physics B 579 (2000) 316

E. Accomando et al. / Nuclear Physics B 579 (2000) 316

E. Accomando et al. / Nuclear Physics B 579 (2000) 316

4. The cases (t, l, l), (t, l, t) and (t, t, l)

E. Accomando et al. / Nuclear Physics B 579 (2000) 316

E. Accomando et al. / Nuclear Physics B 579 (2000) 316

5. Limit on the string scale

E. Accomando et al. / Nuclear Physics B 579 (2000) 316

with pi the momentum of state i. Expanding the Gamma-functions at low energies in

E. Accomando et al. / Nuclear Physics B 579 (2000) 316

E. Accomando et al. / Nuclear Physics B 579 (2000) 316

I. Antoniadis, Phys. Lett. B 246 (1990) 377.

E. Accomando et al. / Nuclear Physics B 579 (2000) 316

A. Delgado, A. Pomarol, M. Quirs, hep-ph/9812489.

E. Accomando et al. / Nuclear Physics B 579 (2000) 316

K. Cheung, W.-Y. Keung, hep-ph/9903294.

Nuclear Physics B 579 (2000) 1755

Golden measurements at a neutrino factory

Received 29 February 2000; accepted 5 April 2000

A. Cervera et al. / Nuclear Physics B 579 (2000) 1755

Let U , with (e , , )T = U (1 , 2 , 3 )T , be the leptonic CabibboKobayashi

A. Cervera et al. / Nuclear Physics B 579 (2000) 1755

A. Cervera et al. / Nuclear Physics B 579 (2000) 1755

present paper a higher statistics is considered, allowing to explore smaller values of 13 ,

2. Fluxes and charged currents

A. Cervera et al. / Nuclear Physics B 579 (2000) 1755

A. Cervera et al. / Nuclear Physics B 579 (2000) 1755

A. Cervera et al. / Nuclear Physics B 579 (2000) 1755

J c13 sin 212 sin 223 sin 213

A. Cervera et al. / Nuclear Physics B 579 (2000) 1755

[113 cos 213 A]2 + [113 sin 213 ]2 .

A. Cervera et al. / Nuclear Physics B 579 (2000) 1755

CP transformation 10 , that is, interchanging the probability of neutrinos and antineutrinos.

A. Cervera et al. / Nuclear Physics B 579 (2000) 1755

A. Cervera et al. / Nuclear Physics B 579 (2000) 1755

A. Cervera et al. / Nuclear Physics B 579 (2000) 1755

Fig. 4. The same as in Fig. 3 but for L = 3500 km.

A. Cervera et al. / Nuclear Physics B 579 (2000) 1755

4. Detection of wrong sign muons

Fig. 6. Sketch of the large calorimeter for the neutrino factory.

A. Cervera et al. / Nuclear Physics B 579 (2000) 1755

A. Cervera et al. / Nuclear Physics B 579 (2000) 1755

4.2. Search for wrong-sign muon events

A. Cervera et al. / Nuclear Physics B 579 (2000) 1755

Fig. 8. Definition of the kinematical variables used in this study.

A. Cervera et al. / Nuclear Physics B 579 (2000) 1755

Fig. 9. Signal and backgrounds for CC and NC events.

A. Cervera et al. / Nuclear Physics B 579 (2000) 1755

Fig. 10. Signal and backgrounds for CC and NC events.

A. Cervera et al. / Nuclear Physics B 579 (2000) 1755

5. Analysis in energy bins

A. Cervera et al. / Nuclear Physics B 579 (2000) 1755