Nucl - Phys.B v.572 PDF

Nuclear Physics B 572 Ž2000.
3–35
www.elsevier.nlrlocaternpe
Anatomy of mixing-induced CP asymmetries in left–right

symmetric models with spontaneous CP violation
Patricia Ball a , J.-M. Frere
` b, J. Matias a
a
CERN – TH, CH – 1211 GeneÕa 23, Switzerland
b
´
SerÕice de Physique Theorique, CP 225, UniÕersite´ Libre de Bruxelles, B – 1050 Brussels, Belgium
Received 15 October 1999; accepted 14 December 1999
Abstract
We investigate the pattern of CP violation in K 0 –K 0 , Bd0 –Bd0 and Bs0 –Bs0 mixing in a
symmetrical SUŽ2.L = SUŽ2.R = UŽ1. model with spontaneous CP violation. From a set of fixed
quark masses and mixings, we calculate the phases of the left and right quark mixing matrices
beyond the small phase approximation and perform a careful analysis of all relevant restrictions on
the model’s parameters from D m K , D m B , e , e Xre and the CP asymmetry aCP in Bd0 ™ Jrc K S0.
Of 64 possible sets of phases, most are excluded because they imply opposite signs for Re e and
aCP . For the one remaining set, aCP is predicted to be small Ž( 0.1. for the chosen input quark
masses and mixings. Barring an instability with respect to these input values, the model is thus
soon to be critically tested at B factories. The mass of the right-handed charged gauge boson, M2 ,
is restricted to be in the range 2.75 to 13 TeV, and the mass of the flavour-changing neutral Higgs
bosons, MH , in 10.2 to 14.6 TeV. This means in particular that the decoupling limit M2 , MH ™ `
is already excluded by experiment. q 2000 Elsevier Science B.V. All rights reserved.
1. Introduction
In this paper we investigate in numerical detail an attractive extension of the standard

model ŽSM., the Spontaneously Broken Left–Right model ŽSB–LR.. The model turns
out to be very constrained since, despite the larger number of bosons, symmetries
strongly limit the new Yukawa couplings.
We show that a large fraction of parameter space is already excluded by conservative
bounds arising from the Bd and Bs mass differences, the e parameter for CP violation
in the K system, and the sign of e X . Even if current theoretical uncertainties persist in the
` ., Joaquim.Matias@cern.ch
E-mail addresses: Patricia.Ball@cern.ch ŽP. Ball., frere@ulb.ac.be ŽJ.-M. Frere
ŽJ. Matias..
0550-3213r00r$ - see front matter q 2000 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 Ž 9 9 . 0 0 8 2 4 - X
4 P. Ball et al.r Nuclear Physics B 572 (2000) 3–35
K system, the expected experimental progress in B physics will soon bring conclusive
tests of the model.
It is well known that CP is a natural symmetry of pure gauge theories with massless
fermions.1 As a result, CP violation actually probes the least known sector of unified
theories, namely the scalar and Yukawa couplings. With the current development of
dedicated accelerators to probe CP violation in the B system, it is important to study
possible departures from the ‘‘Standard Model’’ based on the group SUŽ2.L = UŽ1..
Models based on the group SUŽ2.L = SUŽ2.R = UŽ1., and more specifically those
exhibiting spontaneous CP violation, offer the advantage of a well-defined, and actually
quite constraining, context, largely testable experimentally, while presenting a structure
significantly different from the Standard Model. Before going into any detail, let us
stress already that the ‘‘left-handed’’ nature of the charged couplings in the SM,
together with the absence of CP violation in the neutral channels, is extremely
constraining: for instance electric dipole moments, intrinsically a LR transition, are
strongly suppressed. In the same line, the scalar potential for the SM seems incompatible
with a first-order electroweak transition, thereby hampering low-temperature baryogene-
sis.
While very close in many aspects to the SM, and for this reason a natural extension
of it, ‘‘LR’’ models significantly depart from it and provide a rich structure for both
laboratory CP violation and baryogenesis w1x, but also possibly in the leptonic sector, cf.
w2x; the last two aspects will not be discussed in this paper.
By LR model we understand in general a description of electroweak interactions
based on the gauge group SUŽ2.L = SUŽ2.R = UŽ1.. While such a group structure
suggests low-energy parity restoration, a necessary condition for this is the equality of
gauge couplings, g L s g R . This is not necessarily requested, and could actually prove a
difficulty, notably in a cosmological approach: the persistence of an exact discrete
symmetry to low-energy can lead to the formation of domains corresponding to different
orientations of the breaking, and consequently to difficulties with the walls energy.
Some grand unified models, where P, but not SUŽ2.L = SUŽ2.R = UŽ1., is broken at a
very high scale, lead to a low-energy structure where g L / g R . While in this paper we
will for simplicity set g L s g R , the results are easily adapted to the more general case,
as, due to the high mass of WR , the combination in use is generally g R2 rMR2 , or, for the
mixing terms, g L P g R P sin z , Ref. w3x, where z is the mixing angle between L and R
bosons.
Some symmetries are however needed in order to constrain the scalar and Yukawa
sectors of the theory. We thus request P as a symmetry of the Lagrangian Žpossibly, as
stated above, in a weaker form where the interchange of fermions f L | f R is accompa-
nied by g L | g R .. Even the definition of this symmetry is not without ambiguities:
indeed, with in general non-diagonal mass matrices, the L and R partners are not
uniquely defined, namely, a rotation U in flavour space can be allowed for, namely
f L | Uf R . In order to restrict the model further, we implement the attractive feature of
1
This stems from the unitary nature of the groups, or equivalently the fact that gauge couplings are real.
Anomalies can bring in T or CP violation, but such ’’strong CP violation’’ is only defined with respect to the
determinant of fermion masses.
P. Ball et al.r Nuclear Physics B 572 (2000) 3–35 5
spontaneous CP violation. This means that the Lagrangian must be symmetrical under
CP Žor CP generalized to include g L | g R ., which is later broken by ‘‘misaligned’’
phases of the vacuum expectation values. Under these hypotheses, Ecker and Grimus
have shown in Ref. w4x that, except for an exceptional case Žwhich will not be considered
here., the Yukawa couplings can be parametrized in terms of two real symmetrical
matrices. As a result, all phases of the model can be related to a unique phase affecting
vacuum expectation values Žnoted a below. and calculated exactly. This point is
important, as it relates baryogenesis, and in particular the sign of the matter–antimatter
asymmetry, to low-energy CP violation w1x. In practice, for the present analysis, four
parameters are added to the Standard Model, but in counterpart, its single CP phase is
now predicted.
The shorthand SB–LR will refer from now on to this ‘‘Spontaneously Broken
Left–Right model’’. An important result of our analysis is that the SB–LR is in some
sectors more restrictiÕe than the SM itself. Indeed, while the SM is a subset of LR
obtained by sending the R sector masses to infinity, a similar procedure applied to the
SB–LR yields additional constraints since the CKM phase d is no longer independent,
but predicted within the model. For instance, we find that the CP-violating phase in the
resulting SM is too small, < d < - 0.25 or < d y p < - 0.25, whereas the global fit of w5x
yields d s 1.0 " 0.2. Hence the SM limit of the SB–LR is inconsistent by 3.5s with
current experiments. This has the important consequence that the SB–LR is actually
testable, and distinct from the SM: experimental bounds cannot be indefinitely evaded
by simply sending the R sector to infinite masses: scalars and vectors in the range
Ž2–20. TeV are definitely needed.
Experimental constraints on the SB–LR, mainly from the K system, have been
thoroughly investigated in w3x. Since then, many SM parameters, in particular the CKM
angles and the top quark mass, have been measured much more accurately, and the
perspective of finding non-standard CP violation in the B system at the B factories, the
Tevatron or the LHC has prompted a number of new analyses of the SB–LR, for
instance Refs. w6,7x; these, however, all use a certain approximation for calculating the
phases of the left and right CKM matrices. We thus feel that a comprehensive analysis
of the constraints from measured CP conserving and violating observables in both the K
and B systems is timely, which in particular uses exact expressions for the CKM phases.
The main new results of our analysis can be summarized as follows:
– the small phase approximation fails for CKM matrix elements involving the 3rd
generation;
– the role of the Higgs bosons, neglected in most analyses, is crucial;
– the decoupling limit of the model, M2 , MH ™ `, is experimentally excluded,
which implies upper bounds on M2 and MH ;
– with our choice of quark masses and CKM angles, the SB–LR favours opposite
signs of the CP-violating observables Re e and aCP Ž B ™ Jrc K S ., which are
both expected to be positive in the SM; hence, the model cannot accommodate
both the experimentally measured e and the SM expectation aSM CP B ™ Jrc K S
Ž .
f 0.75 and is excluded if aCP will be measured close to its SM expectation.
Although the exact maximum possible value for aCP Ž B ™ Jrc K s . in the
SB–LR may change when taking into account uncertainties in the quark masses
and CKM angles, we interpret this result as indication that a value of aCP as
large as the SM prediction is difficult to reconcile with the SB–LR.
The paper is organized as follows. In Section 2 we define the model underlying our
calculations. In Section 3 we calculate the phases of the left and right quark mixing
matrix. In Section 4 we formulate strategies for measuring andror constraining the
SB–LR parameters. In Section 5 we calculate B mixing in the SB–LR and incorporate
constraints from the measurement of the CP asymmetry in B 0 ™ Jrc K S . In Section 6
we discuss constraints on the model from K physics. In Section 7 we combine the
constraints from both K and B observables. In Section 8, finally, we summarize and
conclude.
2. Definition of the model
We begin with a reminder, namely how the extended gauge group SUŽ2.L = SUŽ2.R
= UŽ1. cascades down to the unbroken electromagnetic subgroup UŽ1.em through the
following simple symmetry-breaking pattern:
Listed underneath each subgroup factor is our nomenclature convention for their
associated generators and coupling constants.
We next specify the quark and scalar content of the model. The quarks transform
under the unbroken gauge group as
Ui Ui
qL i s
ž /
Di L
; Ž 2,1,1r6 . , qR i s
ž /
Di R
; Ž 1,2,1r6 . , Ž 1.
where i is a generation index. The generation of quark masses in the SUŽ2.L = SUŽ2.R
= UŽ1. model requires at least one scalar bidoublet F , i.e. a doublet under both SUŽ2.,
corresponding to two standard doublets:
f 01 fq
1
Fs
ž fy
2 f 02 / ; Ž 2,2,0 . .
As usual, the quarks are given masses by a spontaneous breakdown of the symmetry
such that F acquires the VEV
1 Õ 0
²F : s
'2 ž 0 w
. / Ž 2.
The quark mass matrices read
Ž u.
Õ wU Žd.
w ÕU
M s Gq D, M s Gq D. Ž 3.
'2 '2 '2 '2
In general, both Õ and w are complex, which is the source of spontaneous CP violation.
The phases can of course be redefined by a gauge rotation of the L or R fields, and, in
particular, the phase difference of Õ and w can be rotated away.2 Here we follow the
notations of Ref. w4x and define two phase combinations:
w
sr, arg Ž Õw . s a , arg Ž yÕwU . s l . Ž 4.
Õ
Although the relevant diagrams are the same with an enlarged Higgs sector, only in the
minimal case are the couplings of scalar fields to quarks determined by masses and
mixing angles only. In this case, neglecting for the moment the contributions from the
triplets, which do not couple directly to the quarks, there is a flavour-conserving neutral
scalar field F 1 , the analogue of the SM Higgs, and a single charged scalar field F ", as
well as two neutral scalar fields F 2 , F 3 with flavour-changing couplings to quarks.
Further fields are needed to achieve the complete breakdown3 from SUŽ2.L = SUŽ2.R
= UŽ1. to UŽ1.em ; the simplest choices respecting LR symmetry are either two doublets
Žone L and one R., or two triplets. This latter choice is usually preferred when dealing
with the leptonic sector, since the quantum numbers of these triplets allow for the
generation of heavy neutrino Majorana masses directly related to LR symmetry break-
ing; b decays, however, do not sensitively depend on this precise structure of the scalar
sector. The triplets are
xqq
L xqq
R
q
0
x L s x L ; Ž 3,1,2 . ,
x L0
and acquire the VEVs

q
0
x R s x R ; Ž 1,3,2 .
x R0
1 0
² xL , R : s
'2 ž /
0
ÕL , R
.
In the remainder of this paper we shall assume < ÕL < 2 < < Õ < 2 q < w < 2 < < ÕR < 2 .
2
This rotation was performed in w8x.
3
Spontaneous breaking of CP without undue fine tuning may require the introduction of further fields, for
instance singlets uncoupled to the fermions w9x. We will not discuss here the details of the scalar Lagrangian
which is not critical for this work.
The spontaneous breakdown of SUŽ2.L = SUŽ2.R = UŽ1. to UŽ1.em generates the

charged W boson mass matrix
g L2 Ž 2 ÕL2 q < Õ < 2 q < w < 2 . r4 yg L g R ÕU wr2
MW2 " s
yg L g R ÕwU r2 g R2 Ž 2 ÕR2 q < Õ < 2 q < w < 2 . r4 0
ML2 ML2R eyi l
'
ž ML2R e i l MR2 / .
The eigenvalues
M12 s ML2 cos 2z q MR2 sin2z q ML2R sin2 z ,
M22 s ML2 sin2z q MR2 cos 2z y ML2R sin2 z
and eigenvectors
W1q cos z ye i l sin z WLq
ž /
W2q s
e l sin z
yi
cos z ž /
WRq
of this mass matrix correspond to the physical charged W bosons in the SUŽ2.L =
SUŽ2.R = UŽ1. model. The WL –WR mixing angle is defined as
2 ML2R
tan2 z s y ,
MR2 y ML2
and the charged current reads Žwith g ' g L ' g R and without displaying unphysical
scalars and charged Higgs contributions.
g m yi l m q
g
Lcc s y
'2 Ui cos z Ž VL . i jg PL y e sin z Ž VR . i jg PR Dj W1 m y '2 Ui
= e i l sin z Ž VL . i jg m PL q cos z Ž VR . i jg m PR Dj Wq
2m .
3. Quark mixing
The Yukawa interaction part of the Lagrangian reads

L Y s Gi j q L iF q R j q Di j q L iF˜ q R j q h.c.
yL Ž 5.
˜ U
with F s s 2F s 2 . As discussed in Section 1, the crucial feature of the SB–LR with
spontaneous CP violation is that P invariance coupled to spontaneous CP violation
restricts the coupling matrices G and D: apart from one special case, see Ref. w4x, which
we shall not consider here, both matrices may be taken to be real and symmetric. We
will work in this basis until further notice. After spontaneous symmetry breaking one
obtains the quark mass matrices M Ž u. and M Ž d . of Eq. Ž3.. The diagonalization of M Ž u.
normally requires a bi-unitary transformation, involving two unitary matrices acting
separately on the L and R spinors; two more matrices are then needed for M Ž d .. In this
special case, however, the matrices M Ž u. and M Ž d . are symmetrical in the chosen basis
and can therefore be diagonalized by only two unitary matrices U,V, so that
M Ž u. s UD Ž u.U T , M Ž d . s VD Ž d . V T , Ž 6.
where D Ž u, d . are diagonal mass matrices. Note that in the SB–LR the signs of the quark
masses are observable, so that the entries in D Ž u, d . need in principle not be positive.
This is a difference with the SM, where the sign of the quark masses in the Lagrangian,
ym i Ž q L i q R i q q R i q L i . ,
can always be absorbed into the phase of q R , which is not observable, since right-handed
quarks have no charged weak interactions. This is of course no longer possible in the
SB–LR. For the moment, we thus need to keep track of the possible signs of the masses;
they will later be absorbed into the right quark mixing matrix so that the model has the
standard mass terms in the Lagrangian and the standard quark propagator, 1rŽ pmg m y
m., m 0 0. As the CKM phases can only depend on mass ratios, we thus have a
2 5 s 32-fold multiplicity of solutions corresponding to the different possible choices of
the quark mass signs. Fortunately, as we shall see later, phenomenological constraints
remove most of these solutions.
Diagonalizing the mass matrices introduces two quark mixing matrices, one for the
left, and one for the right sector; the crucial feature in the present case of spontaneous
CP violation is that Eq. Ž6., in the special basis where D and G are symmetrical, implies
that the left and right mixing matrices are complex conjugate to each other:
K ' K L s U † V s K RU .
The remainder of this section will be devoted to the calculation of the phases of that
matrix, but we would like to first make the connection to a more ‘‘standard’’ quark basis
in which the left mixing matrix contains only one phase, d Žthis choice is obviously not
unique, and the numerous parametrizations of the Kobayashi–Maskawa matrix attest to
this; the procedure below is, however, general.. In order to do so, we first rewrite K as
K s z u K˜z dU ,
where z u, d are diagonal matrices with entries z i such that Ž z i . 2 s sign Ž m i .. These
phases are introduced to redefine the masses as positive. The advantage of introducing
K̃ is that for a s 0, i.e. for the case of no CP violation, and for a suitable choice of
gauge, all entries are real. Redefining now the phases of the quark fields by g iu , g jd,
1 ( i, j ( 3, one can bring K˜ into a standard form VL Ži.e. with only one phase left.,
u
˜ ig d .
VL s eyi g Ke
The right mixing matrix then reads
u d u d
VR s h u eyig K˜ U e ig h d s h u ey2 ig VLU e 2 ig h d .
Here h s z 2 are diagonal matrices with entries "1, which specify the signs of the
quark masses. Note that the phase matrices introduce only five independent phases in
VR , as only differences g id y g ju enter. VR can then be written as
Ž VLU . 11 e 2 i a 1 Ž VLU . 12 e i Ž a 1q a 2q e 1 . Ž VLU . 13 e i Ž a 1q a 3q e 1q e 2 .

VR s Ž VLU . 21 e i Ž a 1q a 2y e 1 .
Ž VLU . 31 e i Ž a 1q a 3y e 1y e 2 .
Ž VLU . 22 e 2 i a 2
Ž VLU . 32 e i Ž a 2q a 3y e 2 .
Ž VLU . 23 e i Ž a 2q a 3q e 2 .
Ž VLU . 33 e 2 i a 3 0
Ž 7.
with the five independent phases a i , e i , in which we have also absorbed the signs of the
quark masses; the sixth phase, hidden in Ž VLU . i j , is the usual unique surviving phase of
the SM model. Note that the phases of VL and VR depend on the parametrization chosen
for VL .
We would also like to mention that in addition to different parametrizations of VL ,
also different conventions for the phase in WL –WR mixing are used in the literature.
Most papers, notably Refs. w3,4x Žand ours., keep the phase explicitly in the Lagrangian,
whereas Ref. w8x prefers to shuffle it into VR . This corresponds to a choice of gauge
lw8x s 0, while Ref. w4x uses Õ s ÕU instead, leading to
w EG x
eil s yeyi a .
Denoting matrices in the different conventions by VLw8x for Ref. w8x and VLw EG x for Ref.
w4x, we then have to identify
VRw8x s e i a VRw EG x . Ž 8.
As later on we would like to use formulas given in Ref. w8x, we also have to convert the
phases d 1 , d 2 and g of VRw8x into our language. We find that one has to identify
g y d2 s 2 a1 q a ,
g q d2 s 2 a2 q a ,
g y d1 s a1 q a2 q e1 q a ,
g q d1 s a1 q a2 y e1 q a .
Note that the system is degenerate and contains only three independent relations.
We can now parametrize the matrix K as
Ž VL . 11 eyi a 1 Ž VL . 12 eyi r2 Ž a 1 q a 2 q e 1 . Ž VL . 13 eyi r2 Ž a 1 q a 3 q e 1 q e 2 .

K s Ž VL . 21 e yi r2 Ž a 1 q a 2 y e 1 .
Ž VL . 31 eyi r2 Ž a 1 q a 3 y e 1 y e 2 .
Ž VL . 22 eyi a 2
Ž VL . 32 eyi r2 Ž a 2 q a 3 y e 2 .
Ž VL . 23 eyi r2 Ž a 2 q a 3 q e 2 .
Ž VL . 33 eyi a 3
0 .
Ž 9.
It is clear that the phases will be functions of r and a , and that CP violation is
characterized by y s r sin a , with r ; O Ž m brm t .. This fact led the authors of w4,10x to
calculate K˜ in a linear expansion in y. Later on, in Ref. w8x, the full solutions for the
phases for mixing between the first two generations were calculated; it was found that
the linear approximation works perfectly well for these entries. It is, however, to be
expected that the linear approximation breaks down for the third generation matrix
elements, where the natural ‘‘smallness’’ of the expansion parameter y can be upset by
enhancement factors m trm b , see also Ref. w11x. In this paper we calculate the full
phases beyond the small phase approximation, which, as shown in w8x, amounts to
solving the matrix equation
Ž 1 y r 2 . W˜ D˜ Ž u. W˜ q Ž r 2 e i a y eyi a . D˜ Ž u. s 2 ir sin a K˜ D˜ Ž d . K˜ T , Ž 10 .
which is equivalent to a system of 12 Žreal. coupled equations. The unknowns in these
equations are six real parameters characterizing the unitary symmetrical matrix W˜ and
˜ D˜ s z D z s h D are the diagonal mass matrices
the six phases of the mixing matrix K.
including the signs of the quark masses. In order to solve Ž10., it is convenient to replace
the independent variables r and a by new variables b and b X , defined as w8x
2 r sin a ib
X 1 y r 2 ey2 i a
b s arctan , e s . Ž 11 .
1yr2 <1 y r 2 ey2 i a <
This transformation makes the dependence on one variable, b X , trivial:
X
K˜ s eyi b r2 K X ,
where K X is a solution of
cos b W X D˜ Ž u. W X y D˜ Ž u. s isin b K X D˜ Ž d . K XT , Ž 12 .
X
and W is still unitary and symmetric. For the ‘‘natural’’ choice r ; O Ž m brm t ., to be
motivated in the next section, b X is negligibly small, and we shall neglect it in our final
results.4 The phases are thus to an excellent approximation functions of only one
variable, b . As shown in Ref. w8x, the above equation has solutions only for a restricted
interval in b ,
b mb
tan ( .
2 mt
We have solved Ž12. by a polynomial expansion as suggested in w8x. The input
parameters are the quark masses renormalized at one common scale, which we choose to
be m t ; we use the following values, in the MS scheme:
m t Ž m t . s 170 GeV,
m b Ž m t . s 2.78 GeV m m b Ž m b . s 4.25 GeV,
m c Ž m t . s 0.63 GeV m m c Ž m c . s 1.33 GeV,
Ž 13 .
m s Ž m t . s 0.060 GeV m m s Ž 2 GeV . s 110 MeV,
m d Ž m t . s 0.0030 GeV m m srm d s 20.1,
m u Ž m t . s 0.0017 GeV m m urm d s 0.56.
The value of m s is a compromise between recent lattice calculations as summarized in
w12x and QCD sum rule calculations w13–16x.
One more input are the angles of the CKM matrix. For the entries determined from
tree-level processes, which receive only tiny corrections from LRS contributions, the
particle data group w17x quotes:
< Vu d < s 0.9740 " 0.0010, < Vu s < s 0.2196 " 0.0023,
< Vcb < s 0.0395 " 0.0017, < Vu brVcb < s 0.08 " 0.02. Ž 14 .
For an exactly unitary matrix where all entries lie within the above specified error range,
we fix
< Vu s < s 0.2219, < Vu b < s 0.004, < Vcb < s 0.04. Ž 15 .
In addition, we have to specify the value of the phase d for the case of no CP violation,
i.e. d Ž b s 0. s 0 or p . We will label the corresponding sets of solutions to Ž12. as class
4 X
However, b is, in general, not negligible in the large-mixing region.
Table 1
Identification of solutions of Ž10. with quark mass signatures. We choose m u to be always positive
No. mt mb mc ms md mu
1 q q q q q q
2 q q q q – q
3 q q q – q q
4 q q q – – q
5 q q – q q q
6 q q – q – q
7 q q – – q q
8 q q – – – q
9 q – q q q q
10 q – q q – q
11 q – q – q q
12 q – q – – q
13 q – – q q q
14 q – – q – q
15 q – – – q q
16 q – – – – q
17 – q q q q q
18 – q q q – q
19 – q q – q q
20 – q q – – q
21 – q – q q q
22 – q – q – q
23 – q – – q q
24 – q – – – q
25 – – q q q q
26 – – q q – q
27 – – q – q q
28 – – q – – q
29 – – – q q q
30 – – – q – q
31 – – – – q q
32 – – – – – q
I and class II solutions, respectively. This induces another twofold multiplicity in

addition to the 32-fold one from the different quark mass signs, so that we finally have
to deal with 64 different solutions to Ž12.. In Table 1 we give the explicit identifications
of the solutions with the mass signatures.
Before presenting results, we would also like to stress that K X and K˜ are independent
of the phase convention for VL – they only depend on the modulus < VL <. The phase
convention for VL enters just in the extraction of the six phases d , a i and e i from Eq.
Ž9.. In the following, we will always use the Maiani convention
c12 c13 s12 c13 s13 eyi d

VL s ys12 c 23 y c12 s23 s13 e i d
s12 s23 y c12 c 23 s13 e id
c12 c 23 y s12 s23 s13 e i d
yc12 s23 y s12 c23 s13 e id
s23 c13
c23 c13 0 ,
with, from the experimental results, Eq. Ž14.,

u 12 s 0.2218 " 0.0031, u 23 s 0.0395 " 0.0017, u 13 s 0.0032 " 0.0008.
Our preferred values for CKM matrix elements Ž15. correspond to
u 12 s 0.2238, u 23 s 0.0400, u 13 s 0.004, Ž 16 .
so that
0.9751 0.2219 0.0040
ž
VLŽ d s 0 . s y0.2219
0.0050
0.9743
y0.0399
0.0400 ,
0.9992 /
0.9751 0.2219 y0.0040
VLŽ d s p . s y0.2216
ž 0.0128 y0.0381
0.9743 0.0400 .
0.9992 /
Note that < Vt d < is quite sensitive to the value of d . A full analysis of the impact of CKM
angle and quark mass uncertainties on our results is beyond the scope of this paper.
Fig. 1. Independent phases of the CKM matrices according to Ž7. as functions of b in the Maiani convention,
X
for d Ž b s 0. s 0, i.e. class I, and b s 0, for positive quark masses. The straight lines are the phases
calculated in the small phase approximation, the curves are the full results.
Fig. 2. sin sd , defined in Ž32., as a function of b calculated from the full solution, left, and in the
approximation derived in Ref. w11x, right. Note that with the approximate formula, the value of the sine can
become larger than 1 for certain choices of the mass signs.
In Fig. 1 we plot the independent phases as functions of b in the allowed range

0 ( b ( 2 m brm t s 0.0327, both the full results and the small phase approximation, for
d Ž b s 0. s 0, i.e. class I, and positive quark masses. From the figure it is evident that
the two phases characteristic of the third generation, a 3 and e 2 , deviate significantly
from the small phase approximation for large b , whereas for the other phases the
agreement between the full and the approximate result is very good. This feature is also
observed for the other 63 solutions. It is also evident that the phases characterizing
mixing between the first two generations are smallish, whereas a 3 and e 2 can become
large. This reflects the impact of the enhancement factors ; m trm b , which overcome
the smallness of b and invalidate the small phase approximation. The numerical results
for all the 64 solutions are available from the authors as a MATHEMATICA file.
Finally, we would like to comment on the approximate formulas for the sine of
certain parametrization-independent combinations of CKM entries relevant to B mixing,
sd and ss , to be defined in Section 5, which were originally derived in w11x and used in
w7x and follow-up papers. These approximate formulas rely on the small phase approxi-
mation and retain only leading terms in the ratios of quark masses. In Fig. 2 we plot
sin sd as function of b calculated from the full solutions and from the approximate
formula given in w11x. Apparently, the structure of the full solution is richer than can be
reproduced by a simple approximation formula. Also, for a number of quark mass
signatures and large b , the formula for sin sd predicts values larger than 1 and thus
cannot be used in that region. On the other hand, for sin ss , shown in Fig. 3, the
approximation works quite well and only fails for large b .
Fig. 3. Same as previous figure, but for sin ss . Dashed lines are the small phase approximation.
4. The SB–LR parameter space and strategies for constraints
The new parameters of the SB–LR are the following Žnumerical values will be
discussed in later sections.:
– M2 ; O Ž1 TeV., the mass of the predominantly right-handed weak gauge

boson;
– z , the mixing angle between WR and WL ; z 0 0;
– g R , the coupling of WR . As discussed in the introduction, although low-energy
parity restoration would require g R s g L , this is not necessarily requested if
parity, but not SUŽ2.L = SUŽ2.R = UŽ1., is broken at a higher scale, and
g R / g L may be preferable, so as to avoid domain-wall formation. For definite-
ness we will however set here g R s g L , as the results on boson masses can be
easily adapted;
– 0 ( r ( 1 and 0 ( a ( p , parametrizing the spontaneous breakdown of CP
symmetry, i.e. the VEV of the bidoublet F ;
– extra Higgs masses which, in principle, are quite arbitrary. However, since they
are associated with neutral flavour changing currents, the prejudice is usually to
have heavy extra Higgses with masses MH ; O Ž10 TeV.. In this case, they
cannot mix significantly with the lighter ones and must be nearly degenerate
Žtheir splitting being at most of the order of the weak scale., see also Ref. w4x.
We will in the present study neglect the mass differences, and allow only one
single mass parameter for the heavy Higgs bosons, MH , and also assume
MH ) M2 . A full study, allowing extra Higgses lighter than the right-handed
bosons, could some day be needed; it would however be sensitive to the fine
details of the scalar potential.
Actually not all of these parameters are independent, but they observe the following
relations and constraints:
2
mW 2r
zs ž /
M2 1qr2
, Ž 17 .
mb r sin a
) , Ž 18 .
mt 1yr2
MH
- 13. Ž 19 .
M2
The second of these relations originates from quark mixing, see Section 3, and the third
one comes from requiring convergence of the perturbation expansion w18x.
Let us now discuss for what type of processes we expect measurable effects due to
the SB–LR. First of all, the SB–LR implies strong constraints on the CKM phase d
even in the decoupling limit of the model, i.e. M2 , MH ™ `. An inspection of all sets of
quark mixing phases shows that < d SB - LR < - 0.25 for class I solutions and < d SB - LR y p <
- 0.25 for class II solutions. One can make use of this fact to exclude the SM model
limit of the SB–LR model experimentally by measuring d from D m B d or e . To date, the

extraction of d from these measurements, however, involves considerable theoretical
uncertainties. Nevertheless, a recent global fit of SM CKM parameters w5x, which
includes conservative theory error estimates, finds d SM s 1.0 " 0.2, so that we conclude
that the decoupling limit M2 ,MH ™ ` of the SB–LR is excluded by 3.5 s .
Let us next discuss the case of finite M2 and MH . As for M2 , it enters either directly
via a W2 propagator or indirectly in z via WL –WR mixing. Obviously, with an expected
modification of the amplitude of size Ž mW rM2 . 2 ; O Ž10y3 . it is in practice impossible
to observe either of these effects in tree-level decays. In loop-induced processes,
however, the situation changes, and we expect measurable or even sizeable effects for
the following cases:
– the suppression factor Ž mW rM2 . 2 is partially compensated by large matching

coefficient functions Žtree-level Wilson coefficients. of the effective Hamilto-
nian, radiative corrections or hadronic matrix elements Že.g. chiral enhancement
in K mixing 5 .;
– WL –WR mixing is enhanced by large quark mass terms from spin-flips, e.g.
z ™ z m trm b in b ™ sg w20x; this affects all top-dominated penguin diagrams
and is thus expected to be important for processes with direct CP violation;
– the SM amplitude is forbidden or heavily suppressed Želectric dipole moment of
the neutron..
As for the Higgses, their contribution to SM tree level decays is also heavily
suppressed by factors Ž mW rMH . 2 ; 10y4 or smaller from the propagator. On the other
hand, the neutral FC Higgs contributes to D F s 2 processes at tree level and the charged
Higgs contributions get enhanced by m trm b in b penguins; roughly speaking, their
contribution is similar in size to that of WR for K mixing and becomes dominant in B
mixing.
Based on these relations, we distinguish two regions in parameter space with
qualitatively different phenomenological consequences. These are
– the ‘‘natural’’ region with r ; O Ž m brm t . ; 0.02, which implies z ; O Ž10y4 .;

– the ‘‘large mixing’’ region with z ; 10y3 so that r ; Ž0.1–1..
The small r region is called ‘‘natural’’, because it has been argued to explain the
observed smallness of the CKM mixing angles w21,22x. With the expected size of
M2 ; O Ž1 TeV., one then has a rather small mixing angle z ; 10y4 , which severely
restricts the possible impact of SB–LR contributions on penguin-induced processes and
on CP asymmetries from direct CP violation. On the other hand, condition Ž18. is
fulfilled for any a , which can thus vary freely within 0 and p . In the natural region, z
effectively decouples from mixing-induced CP-violating processes, and we can choose r
and M2 as independent variables.
5
The fact that there is no such chiral enhancement for B mixing led some authors to conclude that the
SB–LR would not have much impact on these processes, see e.g. Ref. w19x; indeed, it is the Higgs contribution
that is dominant in B mixing.
In the large mixing region, on the other hand, we require z to be close to its
maximum experimentally allowed value z f 0.003 w3,8x. Now r can become large, and
consequently, via Ž18., a is restricted to values close to 0 or p . This is the region where
one might expect sizeable SB–LR effects to show up in penguin-induced processes. In
this case it is more appropriate to choose z and M2 as independent variables and
determine r from Eq. Ž17..
In the present paper we restrict ourselves to mixing-induced CP-violating effects and
thus work consistently in the natural region, fixing 6 r s m brm t . The remaining indepen-
dent parameters are then b , M2 and MH . In addition, we have a 64-fold multiplicity of
CKM phases from the different possible choices for quark mass signs and the value of
the phase d in the limit of no CP violation, d s 0 or p . The observables we analyse in
this paper are D m K , D m B d, s, aCP Ž B ™ Jrc K S ., e and e Xre . Another possible observ-
able from the Bs system is DGs , which was analyzed in w23x. Another potentially
powerful constraint can in principle be obtained from the neutron electric dipole moment
ŽEDM.. The LR model contributions to the EDM are discussed in Refs. w8,24x, taking
into account not only the sum of the quark EDMs as done in previous calculations, but
also specific hadronic terms which involve WL –WR exchange between the quark lines of
a neutron. It is indeed an interesting feature of the SB–LR model that it allows CP
violation in this sector already within a 1-generation context. Crucial to such contribu-
tions is the presence of LR mixing, since the EDM is basically a LR transition and hence
suppressed in the SM. In Ref. w24x, the following bound was obtained:
2
2r mW
z sin Ž g y d 2 . s
1qr2 ž /M2
sin 2 a 1 Ž a . q a 4 ( 3 = 10y6 .
Yet, this bound is not without criticism: it is obtained as a sum of large terms of
opposite sign and thus comes with considerable uncertainty. In addition, it is to be
supposed that there are also large contributions from gluonic matrix elements, i.e. strong
CP violation, which might upset the bound. We thus refrain from taking it into account
in our analysis.
Most of the observables we analyse in this paper are related to the matrix element
< D < Fs2 <
² M 0 < Heff M 0 : s 2 m M Ž M12
SM LR
q M12 LD
q M12 ., Ž 20 .
F s S, B and M 0 s K 0 , Bd0 and Bs0 . M12
SM LR
stands for the SM contribution, M12 for the
LD 2
SB–LR contribution and M12 for Ž D F s 1. contributions, which are negligible in B
mixing, but expected to be sizable in K mixing w25x. We also introduce the mixing
angles f MB q and f MK ,
f MB q s arg M12
Bq
, f MK s arg M12
K
. Ž 21 .
In both the B and K systems the mixing between flavour eigenstates can be described in
terms of these mixing angles to an excellent accuracy; the only quantities to be
considered in this paper, for which that approximation is not sufficient, are e and e Xre .
Note also that the mixing angles are convention-dependent quantities.
6
As long as r ( m b r m t , its exact value does not matter.
As for CP conserving quantities, constraints can be derived from D m, the mass

difference between mass eigenstates. For both K’s and B’s, one has
D m s 2 M12 .
The experimental mass differences in the K and B systems provide in principle a
powerful constraint on < M12LR <; for the K system, however, the size of long-distance
contributions to M12LD is not very well known, so in this case one usually makes the
reasonable assumption7 that the LR contribution should at most saturate D m K . We thus
constrain the LR parameters by requiring
2 < M12K ,LR < - D mexp
K . Ž 22 .
For D m B d , as LD contributions are negligible, theory is in a better shape and we can
require
2 < M12B < s D mexp
Bd . Ž 23 .
We also investigate the CP-violating observable aCP Ž B ™ Jrc K S ., to be defined in
Subsection 5.2, which depends on both the K and B mixing angles, and e and e Xre for
the K system, which will be investigated in Section 6. These three observables vanish
for b ™ 0, in contrast to D m K , B . From the analysis of e , in particular, we shall conclude
that the SB–LR is excluded in the decoupling limit M2 , MH ™ `.
5. B 0 –B 0 mixing in the SB–LR
In this section we largely follow the notations and conventions of Ref. w26x.
5.1. Constraints from D m B d and Predictions for D m B s
In the SM, M12 is dominated by box diagrams with WL and top exchange; it is given
by
SM
1 2
M12 s GF2 mW
2
Ž l tL L . S Ž x t . h2B Ž m . ² B 0 < Ž db . Vy A Ž db . VyA Ž m . < B 0 :
32p 2 m B
1 2 y6r23
' GF2 mW
2
Ž l tL L . S Ž x t . hˆ 2B Ž m . a sŽ5. Ž m .
32p 2 m B
=² B 0 < Ž db . Vy A Ž db . VyA Ž m . < B 0 : . Ž 24 .

The hadronic matrix element is parametrized as usual as
1 B
² B 0 < Ž db . Vy A Ž db . VyA Ž m . < B 0 : s y2 1 q
ž Nc / BB Ž m . f B2 m2B eyi f CP , Ž 25 .
where the bag factor BB Ž m . describes the deviation from vacuum saturation Ž BBvac s 1.
7
This assumption appears reasonable unless there are large cancellations between the different contributions
to M12 .
Table 2
Theoretical input parameters for B mixing. Numbers are taken from Refs. w27,28x
mtŽ mt . h 2B,LO Ž m b . hˆ 2B,LO hˆ 2B,NLO BB Ž m b . BˆBNLO (

f B BˆBNLO
Ž170"5. GeV 0.86 0.580 0.551 0.9"0.1 1.4"0.1 Ž207"42. MeV
B
and f CP is an arbitrary phase describing the transformation behaviour of B flavour
eigenstates under CP transformations:
B B
Ž CP. < B 0 : s e i f CP < B 0 : , Ž CP. < B 0 : s eyi f CP < B 0 : .
It goes without saying that physical observables must be independent of that phase.
B
Often, the tacit choice f CP s p is made; f B is the leptonic decay constant of the B
meson, defined as
²0 < Ž db . A < B 0 : s if B pm ,
and the Inami–Lim function S is defined as

4 x t y 11 x t2 q x t3 3 x t3
SŽ xt . s 2
y ln x t
4 Ž x t y 1. 2 Ž1yxt . 3
with x t s m t Ž m t . 2rmW
2
. For the CKM factors, we use the notation
l tA B s Vt qAU Vt Bb
with A, B s L, R. Numerical values of input parameters are given in Table 2. In the
SB–LR, there are several additional contributions, notably the tree-level neutral Higgs
exchange and box diagrams with WR and unphysical scalar exchanges, all of which are
dominated by the top quark. Taking into account only the leading contributions in M2
and MH , one finds: 8
SM W 1W 2 S1W 2 H SM LR
M12 s M12 q M12 q M12 q M12 ' M12 q M12 , Ž 26 .
with w11x
'2 GF 2
M12H s y m t Ž m t . h H Ž m . l tL Rl tR L ² B 0 < OS Ž m . < B 0 : ,
m B MH2
W 1W 2 qS1W 2
GF2 mW
4
m2B 1
M12 ,y 2
BBS Ž m . f B2 m B q l tL Rl tR L
8p M22 mb Ž mb .
2
6
B
= 4h1L R Ž m . F1 Ž x t , b . y h 2L R Ž m . F2 Ž x t , b . 4 eyi f CP
2
with b s mW rM22 .
8
Box diagrams with charged Higgses are suppressed relative to the L–R box diagrams by roughly a factor
Ž M2 r MH . 2 and can thus be neglected as long as M2 < MH . Note also that M12 LR
as given in Ž26. is
gauge-dependent; subsequent formulas are given in the ’t Hooft–Veltman gauge. The diagrams restoring gauge
invariance have been calculated in Refs. w29,30x and were found to be small in that gauge.
Table 3
Matrix elements for Bs , from Ref. w28x
fBs r fB B B s r BB
1.14"0.08 1.00"0.03
Let us first discuss the Higgs contribution. To leading logarithmic accuracy, the
operator OS s Ž dPL b .Ž dPR b . renormalizes multiplicatively with an anomalous dimen-
sion that just compensates that of the factor m2t , such that m2t OS is RG-invariant. The
LO short-distance correction h H can thus be written as
24r23
asŽ m.
h H Ž m. s , h H Ž m b . s 2.0.
as Ž mt .
On the other hand, the matrix element of OS can be written as9
m 2B 1 B
² B 0 < OS Ž m . < B 0 : s y 12 f B2 m 2B BBS Ž m . 2
q eyi f CP , Ž 27 .
mb Ž mb . 2 Nc
where the bag factor BBS Ž m . ; O Ž1., m ; O Ž m b ., contains the full scale dependence of
the matrix element. To our knowledge, BBS Ž m b . has never been estimated by any
non-perturbative method. In Appendix A we calculate the ratio BBSrBB both to leading
order in a 1rNc expansion and from QCD sum rules. The results agree with
BBS Ž m b .
s 1.2 " 0.2. Ž 28 .
BB Ž m b .
Motivated by the results in Table 3, which indicate a small SUŽ3.-breaking for the bag
factor in the SM, we will use Ž28. also for the matrix elements over Bs0 .
W 1W 2qS 1W 2
As for M12 , its operator structure is more complicated than that of the Higgs
LR
contribution. The LO short-distance corrections h1,2 have been calculated in an ap-
proach suggested by Novikov, Shifman, Vainshtein and Zakharov w31,32x, which was
shown in Ref. w33x to be equivalent to the effective theory approach, which is by now
standard. From the results in w4,11x, one finds
h1L R Ž m b . f 1.8, h 2L R Ž m b . f 1.7. Ž 29 .

F1 and F2 are in general complicated functions of x t , x b and b s Ž mW rM2 . 2 .
9 LR
Note that our expressions for M12 and the matrix element of OS are smaller by a factor of 2 than those
w x
quoted in 11 , which is due to the fact that the authors of that paper use a definition of f B that makes it
smaller than ours by a factor of '2 . This is also the source of a factor of 2 discrepancy in the expression for
LR
< M12 SM <
r M12 in Ref. w7x.
However, in the limit x b ™ 0 and for M2 0 1.4 TeV, to within 5% accuracy they are
approximated by
xt x t ln x t
F1 , q 2
y x t b ln b ,
1yxt Ž1yxt .
x t2 2yxt
F2 , q 2
x t2 ln x t y x t ln b .
1yxt Ž1yxt .
Numerically, <4 F1 < < < F2 <. We thus approximate
h 2L R F2 y 4h1L R F1 f h 2L R Ž F2 y 4 F1 . .
We are now in a position to calculate M12 in the SB–LR and to investigate its impact
on phenomenology. Following Ref. w11x, we write Ž26. as
SM
M12 s M12 Ž 1 q k e i sq . , Ž 30 .
with
M12LR
k' SM
, Ž 31 .
M12
LR
M12 Vt Rb Vt Rq U
sq ' arg SM
M12
s arg y
ž Vt Lb Vt LqU / . Ž 32 .
Note that the phase sq is convention-independent and a physical observable. The minus
sign in the definition of sq comes from the fact that, putting all CKM factors equal to 1,
SM LR
M12 and M12 have different relative sign. k is nearly independent of the flavour of
the spectator quark. Numerically, we find10
2 2
BBS Ž m b . 7 TeV 1.6 TeV
ks
BB Ž m b . ž MH / q h 2L R Ž mb . ž M2 /
2
1.6 TeV
½
= 0.051 y 0.013ln
ž M2 /5 , Ž 33 .
which describes the full solution to within 5% accuracy for MH ) 7 TeV and M2 ) 1.4
TeV.
Let us now investigate the predictions of the SB–LR for and the constraints from the
experimental data for D m B . For the remainder of this section we consider < M12 < as a
10
Corrections in Ž1yh 2L R rh1L R . are smaller than neglected terms in 1r M24 .
function of only two variables, k and b , instead of expressing k in terms of M2 and

MH . The mass difference in the Bd0 system has been measured as w34x
D m B d s Ž 0.472 " 0.016 . psy1 , Ž 34 .
whereas there exists, for the mass difference in the Bs0 system, only a lower bound w35x:
D m B s ) 12.4 psy1 ; Ž 35 .
the SM expectation is w5x
D mSM
B s s Ž 14.8 " 2.6 . ps
y1
. Ž 36 .
In the SM and with our values for the CKM angles, Eq. Ž16., Ž34. restricts the phase d
to d SM s 1.17 " 0.44 Žtaking into account that the measured value of e implies
d SM ) 0., where the error comes mainly from f B2 BˆB . Taking into account the theoretical
uncertainties on f B2 BˆB and m t , Ž34. translates into
2
Ž Vt Lb Vt LdU . Ž 1 q k e i s . d s Ž 6.7 " 2.7 . = 10y5 . Ž 37 .
From this result we may derive a constraint for k : in the worst case of negative relative
sign, k evidently cannot be larger than
Ž 6.7 " 2.7 . = 10y5
k max y 1 s 2
,
< Ž Vt Lb Vt LdU . <
which roughly translates into

k - 3,
which we will use in the following. For Bs mixing, on the other hand, we obtain from
Ž35. the lower bound
2
Ž Vt Lb Vt Ls U . Ž 1 q k e i s . s ) 9.6 = 10y4 . Ž 38 .
For the ratio of mass differences, the theory error is much smaller:
2 2
Dm Bs m Bs fBs BˆB s Ž Vt Lb Vt Ls U . Ž 1 q k e i s .
s
Dm Bd
s
m Bd ž /
fBd BˆB d
2
Ž Vt Lb Vt LdU . Ž 1 q k e i s .
d
2
Ž Vt Lb Vt Ls U . Ž 1 q k e i s . s
s Ž 1.31 " 0.19 . 2

, Ž 39 .
Ž Vt Lb Vt LdU . Ž 1 q k e i s . d
which translates into the bound

2
Ž Vt Lb Vt Ls U . Ž 1 q k e i s . s
2
) 17.2. Ž 40 .
Ž Vt Lb Vt LdU . Ž 1 q k e i s . d
In Fig. 4 we plot the left-hand side of Eq. Ž37., normalized by the central value on the
right-hand side, as a function of k for several values of b . In Fig. 5 we plot
Fig. 4. Constraints from D m B d . Predictions of the SB–LR as a function of k for different values of b ,
0,0.01,0.02,0.03. Dashed lines are the experimental result and theory errors. The lower curves are class I
solutions, the upper curves are class II.
<Ž Vt Ls Vt LbU . 2 Ž1 q k e i s s <r0.039 2 , which is expected to be 1 in the SM. From the plots we
conclude the following:
– the decoupling limit k ™ 0 of the SB–LR is excluded; this is due to the fact
that the SM phase as extracted from D m B d is rather large, d SM ; 1, whereas
Fig. 5. As Fig. 4, but for D m B s. Normalization of the vertical axis corresponds to the SM expectation
< Vt s < s 0.039. Dashed line is the lower bound on D m B .
s
the SB–LR predicts values close to 0 or p ; this result depends, however, on

our specific choice of the CKM angles, Eq. Ž16.; yet, we shall derive the
experimental exclusion of the decoupling limit unambiguously from the analy-
sis of e in the next section;
– class II solutions are excluded for b 0 0.021;
– class I solutions require k ) 0.52, class II solutions k ) 0.42. This means that
the Higgs contributions to k are essential, see Eq. Ž33..
We next plot the correlations between predicted values for D m B s and D m B d in Fig. 6.
This plot, too, illustrates the exclusion of the decoupling limit of the SB–LR, which
corresponds to the two crossing-points of the different classes of solutions, best visible
for b s 0. It is also evident that the SB–LR can comfortably accommodate any
non-standard value of D m B s as well as the expected one, Eq. Ž36.. There is in particular
a large fraction of class I and II solutions that predict very large values of D m B s, so that
a measurement of D m B s close to its SM expectation will effectively constrain the
parameter space of the SB–LR.
Fig. 6. Correlation between predictions for D m B d and D m B s for different values of b as a function of k in
w0,3x. Short dashes denote experimental result and theory error for D m B , long dashes denote lower bound on
d
D m B s. The ensemble of lines in the left parts of the plots are class I solutions, the ones in the right halves are
class II.
5.2. Constraints from Bd0 ™ J r c K S0
Other interesting constraints can be obtained from the measurement of the CP

asymmetry in Bd0 ™ Jrc K S0 . Recently, the CDF collaboration has reported the follow-
ing result w36,37x:
G Ž Bd0 Ž t . ™ Jrc K S0 . y G Ž Bd0 Ž t . ™ Jrc K S0 .
aCP s
G Ž Bd0 Ž t . ™ Jrc K S0 . q G Ž Bd0 Ž t . ™ Jrc K S0 .
.41
s Ž 0.79q0
y0 .44 . sin Ž D m B t . , Ž 41 .
and with 90% probability aCP rsin Ž D m B t . ) 0. In the SM, aCP measures just sin2 bCKM
with
VcLd VcbLU
bCK M s arg y
ž Vt Ld Vt LbU
SM
/ ;
q0.05 w x
Ž 42 .
the SM model expectation is sin2 bCK M s 0.73y0.06 5 .

In the SB–LR, however, we have to interpret the measurement differently. For any B
decay into a final CP eigenstate fCP , one can define a convention and parametrization-
invariant quantity l by
q A f CP
l' ž /p B A f CP
with the amplitudes A f CP s AŽ B 0 ™ f . and A f CP s AŽ B 0 ™ f . and the mixing amplitude
Ž qrp .B , yexpŽyi f MB ., defined in Ž21.. In the case of vanishing direct CP violation,
the time-dependent CP asymmetry can be written as
aCP s Im l sin Ž D m B t . ,
so that we have
Im l Ž B 0 ™ Jrc K S0 . s 0.79q0 .41
y0.44 . Ž 43 .
The expression for l itself reads
VcbL VcLsU
l Ž B 0 ™ Jrc K S0 . s exp Ž yi f MB d . hJr c K S0
B K
ž VcbLU VcLs /
=exp i Ž yf CP q f CP . exp Ž i fMK . .
Jrc K S0 is CP odd, hence hJr c K S0 s y1, and the K mixing phase was defined in Eq.
Ž21.. The imaginary part reads
eff 0 0
sin2 bCK M ' Im l Ž B ™ Jrc K S .
M12K ,LR
s sin 2 bCK M q arg Ž 1 q k e i s d . y arg 1 q
ž M12K ,SM / . Ž 44 .
The contribution from K mixing, the third term in square brackets, is numerically small
and can be neglected for the moment.
eff
In Fig. 7 we plot sin2 bCK M as a function of k for several values of b , where we
eff
only show solutions that yield sin2 bCK M ) 0. It is obvious that the SM expectation
eff y4
Fig. 7. sin2 bCK M in the SB–LR as a function of k for several values of b , 10 , 0.01, 0.02, 0.03. Only
eff
solutions yielding positive values of sin2 bCK M are shown.
eff
sin2 bCK M f 0.75 can be accommodated by a number of solutions. It is also visible that
eff
for b - 0.03, there are roughly two branches of solutions, one with small sin2 bCKM -
0.4, the other one spanning all possible values between 0 and 1. A measurement of
eff
sin2 bCK M around its SM expectation value would favour either k f 0.6 or k ) 1.2. Any
more detailed analysis requires the constraints from K mixing to be taken into account.
6. Constraints from the K system
While the K system was the first one to be analyzed in SB–LR, it remains plagued by
theoretical uncertainties.11 The main observables to be considered are obviously D m K , e
and e X . The formulas for D m K are analogous to those for D m B discussed in the last
section. To be specific, we use the LO QCD-corrected formulas for M12LR of Ref. w4x
with the bag factor BKS s 1; radiative corrections to M12SM
are taken from Ref. w27x; we
ˆ w x
also use BK s 0.89 28 . As for e , in the SM, it is usually written as
1 Im M12
es
'2 e ip r4
ž DmK
qj0
/ Ž 45 .
with
Im a0
j0s , aU0 s ²pp Ž I s 0 . < y i Heff
< D S <s1 <
K 0 :weak , Ž 46 .
Re a0
where the matrix element in Ž46. does contain only the weak phase, but no strong
final-state rescattering phases. As the derivation of this formula includes some relations
11
It should not be forgotten that the same is actually true for the SM.
and approximations that need not be valid in the SB–LR, we rederive it, following the
transparent discussion given in Ref. w38x.
The parameter e measures essentially the phase difference between M12 and G 12 ,
where G 12 is the matrix element of the decay matrix G over the K 0 and K 0 states.
Introducing
du M r G s arg M12 y arg G 12 ,
one finds for the hypothetical case of no CP violation
du M r G s p ,
which essentially follows from the fact that D m K ' m L y m s ) 0, but yDGK ' GL y GS
- 0. Note that only the phase difference du M r G is an observable and convention-inde-
pendent quantity, but not arg M12 and arg G 12 separately, which depend on the K
analogue of the arbitrary phase f CP , introduced in Ž25. for B’s, and on the s and d
quark phases, i.e. on the parametrization of the quark mixing matrices. For the analysis
of e , it proves convenient to choose f CP s p , which we shall use in the remainder of
this section. One then has – still for the case of no CP violation and using the Maiani
parametrization of the CKM matrix – arg M12 s 0 and arg G 12 s p . In the real CP-
violating world, du M r G is only slightly different from p : making use of the fact that K
decays are dominated by the 2p channel with isospin 0 Žthe famous D I s 1r2 rule., a
measure of CP violation in the interference of mixing and decay, i.e. of du M r G , is given
by e , which is defined in such a way as to contain no effects from direct CP violation
and which can be written as w38x
x
esy 2 1 q Ž 2 x . i 4 sin du M r G , Ž 47 .
4 x q1
with
DmK
xs s 0.478 " 0.002 f 0.5,
DGK
so that
1
ef e ip r4 Ž ysin du M r G . . Ž 48 .
2'2
In contrast to M12 , G 12 cannot be calculated accurately from theory, and one exploits
the dominance of decays into the 2p Ž I s 0. final state to derive12
arg G 12 f y2arg Ž a 0 e ip r2 . . Ž 49 .
Combining equations Ž48. and Ž49., we finally obtain
1
es e ip r4 sin Ž arg M12 q 2arg a0 . . Ž 50 .
2'2
In contrast to Eq. Ž45., this formula also allows us to demonstrate explicitly that e does
not depend on the parametrization of the quark mixing matrices: M12 contains the
12
The factor e ip r2 arises because arg G 12 is related to the matrix element of Heff , whereas we have defined
aU0 as a matrix element of y i Heff .
generic CKM factor Ž l iA B . 2 , A, B g L, R4 , i g u,c,t 4 , whereas a0 contains the factor

l iA BU , so that phase redefinitions of VL and VR cancel in the sum Ž50..
The experimental result for e , < e < s Ž2.280 " 0.013. = 10y3 w17x, implies
arg M12 q 2arg a0 s Ž 6.449 " 0.037 . = 10y3 . Ž 51 .
In the SM, both terms in the above sum are small so that arg can be replaced by the ratio
of imaginary to real part. In addition, as arg M12 f 0 in our phase convention, one also
has Re M12 f < M12 <, which is just D m K r2, so that one can approximate Ž50. by Ž45. to
excellent accuracy. As 2arg a 0 < arg M12 , this term can safely be neglected in the SM;
in the SB–LR this is no longer true: the contributions of WR to a0LR have been
calculated in Refs. w4,8x, but involve considerable theoretical uncertainties. From Refs.
w4,8x, we find
2
1 TeV
2 <arg aLR <
0 - 0.005 P ž M2 / .
In view of the theoretical uncertainty of aLR0 , which involves a number of only poorly
known hadronic matrix elements, we prefer to include it into the uncertainty of arg M12
in Ž51.. Assuming that the unknown Higgs contributions are not larger than those from
WR , we include twice the value of 2 <arg a0LR < in the uncertainty of arg M12 and thus find
the constraint
2 2
1 TeV 1 TeV
- u˜M - 6.523 = 10
6.375 = 10 y3
y 0.01 P
ž M2 / y3
q 0.01 P
ž M2 /
Ž 52 .
with
2 Re M12
ũ M s arg M12 ,
DmK
which also takes into account two experimental standard deviations and where we have
rescaled Re M12 to its experimental value.
Finally, the expression for e X reads
1 Re a2
eX f e ip r4 Ž j2yj0. ,
'2 Re a 0
where j 2 is defined in analogy to j 0 for the 2p Ž I s 2. final state. In view of the large
theoretical uncertainties associated with the precise value of ReŽ e Xre ., we only require
the SB–LR to predict a positive value.
Let us now discuss the constraints to be obtained from the three observables D m K , e
and e X . First, we consider the decoupling limit M2 , MH ™ `. In this limit we find
u˜M - 2.9 = 10y3 ,
which is less than half of the experimental value and is related to the smallness of the
standard CKM phase d in the SB–LR. From this result, which is insensitive to the exact
value of the uncertain CKM angle u 13 , we firmly conclude that the decoupling limit
M2 ,MH ™ ` is experimentally excluded.
Fig. 8. Allowed values for M2 and MH from the K-physics constraints.
We next investigate the allowed region in the space of mass parameters imposed by
the constraint Ž52. and the one on D m K , Eq. Ž22.. The result is plotted in Fig. 8. We
find in particular the following lower bounds on the extra boson masses:
M2 ) 1.85 TeV, MH ) 5.2 TeV. Ž 53 .
The bound for M2 is in the ballpark of the values usually obtained, cf. Ref. w4x. The one
on MH is smaller; the reason for this is that, in contrast to all previous analyses, we did
not assume charm quark dominance for M12H , but also included the top quark contribu-
tions which can destructively interfere with the charm quark ones, thus lowering the
limit on MH . The experimental limit on u˜M , i.e. e , implies also Žnot very constraining.
upper bounds on the extra boson masses:
M2 - 73.5 TeV, MH - 230 TeV. Ž 54 .
We recall that all these limits and bounds are to be modified by the inclusion of B
physics constraints. This concludes our discussion of constraints from K physics.
7. Combining constraints from K and B system
Combining all the constraints from D m K , D m B d , D m B s, e , e X and sin2 bCKM eff

, our
main finding is that, although the values of the CP conserving observables can be
reproduced by a large range of input parameters, this is not the case for the CP-violating
ones: the crucial point is a strong anticorrelation between the signs of Re e and
eff
sin2 bCK M , which are both known to be positive from experiment. We illustrate this
point in Fig. 9, where we plot the values of e Žto be precise: e P eyi p r4 . as a function of
eff
sin2 bCK M for all sets of input parameters n, b , M 2 , M H with 2 TeV ( M 2 ( 50 TeV
Ž .
and M2 ( MH ( 13 M2 that pass the cuts on the mass differences Žwith a 50% uncer-
tainty on D m B d to account for the uncertainty of the CKM angles. and on the sign of
Re e X . It is obvious that only a few sets of input parameters can reproduce the observed
eff
sign of both e and sin2 bCK M . We find that the class I quark mass signature No. 31 is
the only one accounting for the points in the allowed sector in Fig. 9, and thus the only
one of the initial 64 signatures to survive all cuts. A closer inspection shows that the
maximum possible sin2 bCK eff ˜
M correlated with u M in the range given in 52 is
Ž .
eff ,max
sin2 bCK M s 0.1,
eff
Fig. 9. Allowed values of the CP-violating parameters e and sin2 bCK M whose input parameters pass the D m
X
cuts and yield the correct sign of Re e .
SM
which is incompatible with the SM expectation sin2 bCKM f 0.75. The exclusion of all
but one quark mass signatures also cuts deeply into the allowed range for M2 and MH .
For fixed M2 Žand b ., we have the following constraints on MH :
– a lower bound from D mLR

K - DmK ;
exp
eff eff
– an upper bound from sin2 b SM ) 0 Žbecause sin2 b SM - 0 for MH ™ `.;
– ˜
a lower bound from the upper limit on u M , Eq. 52 ;
Ž .
– an upper bound from the lower limit on u˜M .
The allowed region in Ž M2 , MH . Žalso taking into account the constraints from D m B .
thus gets very much restricted, as shown in Fig. 10. We find the bounds
2.75 TeV - M2 - 13 TeV, 10.2 TeV - MH - 14.6 TeV,
which improve considerably on those from K physics alone, Eqs. Ž53. and Ž54.. The
predictions for D m B s are in the range Ž0.6–1.1. D mSM
Bs
,exp
, i.e. a measurement of D m B s
close to its SM expectation would not pose any additional constraint.
Fig. 10. Allowed region in Ž M2 , MH ., taking into account all constraints.

As for b , we find that

0.009 - b - 0.0327,
i.e. the combination of K and B constraints only results in a lower bound on b , but does
not improve the maximum one, which is still given by Ž18..
8. Summary and discussion
In this paper we have investigated in detail the present status of the left–right
symmetrical model with spontaneous CP violation, based on the gauge group SUŽ2.L =
SUŽ2.R = UŽ1.. The parameter space of this model includes the masses of the predomi-
nantly right-handed charged gauge boson, M2 , of FC neutral and charged Higgs bosons,
which we have assumed to be degenerate with a common mass MH . Also included are
the parameter b , which measures the size of the VEV of the Higgs bidoublet F that
characterizes the spontaneous breakdown of CP symmetry, and n, the 64 different quark
mass signatures, which are observable in the SB–LR. In contrast to previous publica-
tions, e.g. Refs. w3,4,6,7x, in which the constraints on the model from K and B physics
were treated separately, ours is the first one to consider them in a coherent way and to
use the exact results for the CKM phases instead of the small phase approximation. We
have concentrated on experimental constraints imposed by the mass differences D m K , B
and observables describing CP violation, i.e. e , e X and aCP Ž B ™ Jrc K S .. In view of
the large theoretical uncertainties, we only use the sign, but not the absolute value of
Re Ž e Xre . as a constraint, and we do not use the electric dipole moment of the neutron.
Our main finding is that, although the K and B constraints can be met separately by a
large range of input parameters, it is their combination that severely restricts the model.
eff
We find in particular that the CP-violating observables e and sin2 bCK M are crucial: the
sets of input parameters that pass the constraints imposed by the meson mass differences
eff
D m K , B yield to a large majority opposite signs of e and sin2 bCKM . The combination of
all constraints yields the following results:
– all but one quark mass signatures are excluded, only class I solution No. 31
survives;
eff
– aCP Ž B ™ Jrc K S . ' sin2 bCKM - 0.1, which is compatible with the present
experimental result Ž41., but incompatible with the SM expectation of 0.75;
– predictions for D m B s are in the range Ž0.6–1.1. D mSM
Bs
,exp
;
– the masses of the extra bosons are restricted to
2.75 TeV - M2 - 13 TeV, 10.2 TeV - MH - 14.6 TeV;
– the value of b is restricted to 0.009 - b - 0.0327.
We would like to stress that our findings are largely independent of the details of the
scalar potential: the relevant neutral Higgs vertices can be obtained essentially from the
requirement of gauge invariance of S matrix elements, as discussed in Ref. w29x. This
does not apply to the charged Higgs vertices, which in principle do depend on the
characteristics of the scalar potential: we have thus imposed the condition MH ) M2 in
order to suppress all contributions from charged Higgses Žin particular box diagrams..
We would also like to stress that our study does not claim to be exhaustive as we did
not allow the most crucial SM input parameters, the CKM angles and quark masses, to
float within their presently allowed ranges. Taking into account these uncertainties
would certainly affect the phases of the CKM matrices and thus mainly show up in the
CP-violating observables, which, as we have shown, are crucial. It is therefore not to be
excluded that an analysis of the input parameter uncertainties would result in increasing
the viable LR parameter ranges, but we doubt that it will change the anticorrelation
eff
between the signs of e and sin2 bCK M , which implies a small maximum value of
eff
sin2 bCK M attainable in the model.
Another limitation of the present analysis is that we have kept the ratio of Higgs
VEVs, r, constant and equal to m brm t . As stressed before, this quantity governs the
amount of mixing between L and R bosons. For most observables, the relevant
parameter is b ; 2 r sin a , on which our analysis is based. Reducing the value of r
while keeping b fixed Žremember that tanw br2x ( m brm t . would not affect our
conclusions. Increasing r, however, has an impact on the imaginary parts of a0 and a 2 ,
and from there on the values of e and e X . Such an increase is strongly disfavoured if we
take into account the constraint from the neutron’s EDM. Our most important result,
eff
namely the bound on sin2 bCK M is, however, not affected by these considerations.
Acknowledgements
P.B. is supported by DFG through a Heisenberg fellowship. J.M.F. gratefully

acknowledges hospitality and financial support from the Theory Division of CERN,
where this work got started. He is also supported in part by the ‘‘Actions de Recherche
´
Concertees’’ of the ‘‘Direction de la Recherche Scientifique — Communaute´ Française
de Belgique’’, IISN–Belgium, convention No. 4.4505.86. J.M. acknowledges financial
support from a Marie Curie EC Grant ŽTMR–ERBFMBICT 972147..
Appendix A. Non-factorizable contributions to BSB r B B
We estimate the ratio of bag factors BBSrBB , which enters the matrix element M12
describing B 0 –B 0 mixing, by two methods: the 1rNc expansion and QCD sum rules.
Some aspects of the 1rNc expansion for BB have also been discussed in w39x.
To leading order in the 1rNc expansion, it follows from Ž25. that BB s 3r4 and from
Ž27. that BBS s 1rŽ1 q m 2brŽ6 m2B .., so that
BBS Nc™` 4 1
s f 1.2. Ž A.1 .
BB 3 1 m2b
1q
6 m2B
As factorization is exact in the large-Nc limit, BB becomes scale-independent and it is

not clear at what scale ŽA.1. is valid. Another approach, more suited to include the
scale-dependence, is provided by QCD sum rules w40–42x. QCD sum rules for BB have
already been discussed in w43–46x; our results for BBS are new. We consider the
correlation functions
X
P Ž p 2 , pX 2 . s i 2 d 4 x d 4 y e iŽ p xyp y. ²0 < Tj B† Ž x . O SM Ž 0 . j B† Ž y . <0: ,
H
X
P S Ž p 2 , pX 2 . s i 2 d 4 x d 4 y e iŽ p xyp y. ²0 < Tj B† Ž x . O S Ž 0 . j B† Ž y . <0: ,
H
with the currents and operators
j B s Ž m b q m d . dig 5 b,
O SM s Ž db . Vy A Ž db . VyA ,
O S s Ž db . Sy P Ž db . SqP .
In the standard philosophy of QCD sum rules, P and P S are, on the one hand,
calculated in a local operator product expansion in the deep Euclidean region p 2 , pX 2 < 0
and, on the other, analytically continued to the Minkowskian region by dispersion
relations, saturated by the hadronic ground state. Equating these two representations
yields – after some technicalities which are well known to the experts and which we
refrain from describing in this appendix – QCD sum rules for BB and BBS , respectively.
It turns out that the leading non-factorizable contributions come from the dimension-5
mixed condensate ² d s gGd :, which is enhanced by a factor m b ; the sum rules read
m4B f B2 8
2 X2
f B2 m 2B BB s P fact q P ² 5: q . . . ,
Žp y m 2B .Ž p y m2B . 3
m4B f B2
2 f B2 m 2B BBS s P fact
S
q P S ,² 5: q . . .
Ž p 2 y m2B . Ž pX 2 y m2B .
Here the dots denote subleading non-factorizable contributions from O Ž a s . perturbation
theory and the gluon, four-quark and higher condensates, which we neglect in our
estimate.
We can now subtract the factorizable parts, perform Borel transformation and
continuum subtraction Žsome of the technicalities mentioned above., and with B ' 1 q
D B we obtain
D BBS 4 1 BˆPnon
S ,² 5:
- fact
s . Ž A.2 .
D BB 3 m 2B 1 BˆPnon
² 5:
- fact
2
q
mb 6
Unfortunately, both D B in the numerator and denominator are scale-dependent so that

ŽA.2. suffers from a large scale uncertainty. Following Ref. w47x, we thus introduce the
LO RG-invariant quantities Ž M 2 is the Borel parameter.
y g B r Ž2 b 0 .
Pˆ Ž M 2 . s a s Ž M 2rm b . BˆP Ž M 2 . ,
yg B S r Ž2 b 0 .
Pˆ S Ž M 2 . s a s Ž M 2rm b . BˆP S Ž M 2 . ,
Fig. 11. QCD sum rule for D BBS Ž m b .r D BB Ž m b . as a function of the Borel parameter M 2 ; input parameters:
one-loop pole mass m b s 4.8 GeV, s0 s 34 GeV 2 . The arrows denote the fiducial region for the Borel
parameter.
where we have in particular taken into account that the natural scale of QCD sum rules
for heavy hadrons is m s M 2rm b , see the discussion in w47x. This allows us to calculate
D BBSrD BB directly at the scale m s m b with
Ž g B S y g B .r Ž2 b 0 .
D BBS Ž m b . 4 1 as Ž mb .
s
D BB Ž m b . 3 m2B 1 a s Ž M 2rm b .
2
q
mb Ž mb . 6
s0 Ž s q m2b .Ž m2b y s . 2
Hm ds eysr M
= y
1
4
q
1
2
2
b
Hm ds
s0
2
b
s2
Ž s q m2b . m2b
s2
with b 0 s 11 y 2r3 P 4, g B S s y8 and g B s 2; s0 f 34 GeV 2 is the continuum thresh-

e ysr M 2
0 , Ž A.3 .
old.
In Fig. 11 we plot the sum rule as a function of M 2 ; the dependence on the M 2 as
well as on s0 and m b is very mild; we find
D BBS Ž m b .
s y Ž 0.43 " 0.07 . , Ž A.4 .
D BB Ž m b .
which, with D BB Ž m b . s yŽ0.1 " 0.1. from lattice calculations w28x, yields
BBS Ž m b .
s 1.16 " 0.14. Ž A.5 .
BB Ž m b .
This result agrees perfectly well with that from the 1rNc expansion, Eq. ŽA.1.. We thus
quote as our final result
BBS Ž m b .
s 1.2 " 0.2. Ž A.6 .
BB Ž m b .
References
w1x ` et al., Phys. Lett. B 314 Ž1993. 289.

J.-M. Frere
w2x ` J. Liu, Nucl. Phys. B 324 Ž1989. 333.
J.-M. Frere,
w3x P. Langacker, S.U. Sankar, Phys. Rev. D 40 Ž1989. 1569.
w4x G. Ecker, W. Grimus, Nucl. Phys. B 258 Ž1985. 328.
w5x F. Parodi, P. Roudeau, A. Stocchi, hep-exr9903063.
w6x G. Barenboim, J. Bernabeu, M. Raidal, Nucl. Phys. B 478 Ž1996. 527.
w7x G. Barenboim, J. Bernabeu, M. Raidal, Nucl. Phys. B 511 Ž1998. 577.
w8x ` et al., Phys. Rev. D 46 Ž1992. 337.
J.-M. Frere
w9x G.C. Branco, L. Lavoura, Phys. Lett. B 165 Ž1985. 327.
w10x D. Chang, Nucl. Phys. B 214 Ž1983. 435.
w11x G. Ecker, W. Grimus, Z. Phys. C 30 Ž1986. 293.
w12x S. Ryan, Talk given at 1999 Chicago Conference on Kaon Physics ŽK 99., Chicago, IL, June 1999,
hep-phr9908386.
w13x M. Jamin, M. Munz,¨ Z. Phys. C 66 Ž1995. 633.
w14x K.G. Chetyrkin, D. Pirjol, K. Schilcher, Phys. Lett. B 404 Ž1997. 337.
w15x P. Colangelo et al., Phys. Lett. B 408 Ž1997. 340.
w16x M. Jamin, Nucl. Phys. B ŽProc. Suppl.. 64 Ž1998. 250.
w17x C. Caso et al. ŽPDG., Eur. Phys. J. C 3 Ž1998. 1.
w18x F.I. Olness, M.E. Ebel, Phys. Rev. D 32 Ž1985. 1769.
w19x M. Gronau, D. London, Phys. Rev. D 55 Ž1997. 2845.
w20x P. Cho, M. Misiak, Phys. Rev. D 49 Ž1994. 5894.
w21x G. Ecker, W. Grimus, W. Konetschny, Phys. Lett. B 94 Ž1980. 381.
w22x G. Ecker, W. Grimus, W. Konetschny, Nucl. Phys. B 177 Ž1981. 489.
w23x G. Barenboim et al., Phys. Rev. D 60 Ž1999. 016003.
w24x ` et al., Phys. Rev. D 45 Ž1992. 259.
J.-M. Frere
w25x ´
J. Bijnens, J.-M. Gerard, G. Klein, Phys. Lett. B 257 Ž1991. 191.
w26x The BaBar Physics Book, ed. P.F. Harrison, H.R. Quinn, SLAC Report 504 Ž1998..
w27x G. Buchalla, A. Buras, M. Lautenbacher, Rev. Mod. Phys. 68 Ž1996. 1125.
w28x L. Lellouch, Talk given at 34th Rencontres de Moriond, Les Arcs, France, March 1999, preprint
CERN–THr99–140, hep-phr9906497.
w29x W.-S. Hou, A. Soni, Phys. Rev. D 32 Ž1985. 163.
w30x J. Basecq, L.-F. Li, P.B. Pal, Phys. Rev. D 32 Ž1985. 175.
w31x A.I. Vainshtein et al., Sov. J. Nucl. Phys. 23 Ž1976. 540, Yad. Fiz. 23 Ž1976. 1024.
w32x M.I. Vysotsky, Sov. J. Nucl. Phys. 31 Ž1980. 797, Yad. Fiz. 31 Ž1980. 1535.
w33x A. Datta et al., hep-phr9509420 Žunpublished..
w34x The LEP B Oscillation Working Group, preprint LEPBOSC 98r3.
w35x J. Alexander, Plenary talk given at ICHEP 98, Vancouver, July 1998, http:rrichep98.triumf.carprivater
convenorsrtransparenciesrplenary7.pdf.
w36x M.P. Schmidt ŽCDF., Talk given at 34th Rencontres de Moriond, Les Arcs, France, March 1999, preprint
Fermilab–Conf–99r157–E, hep-exr9906029.
w37x T. Affolder et al. ŽCDF., preprint Fermilab–Pub–99r225–E, hep-exr9909003.
w38x T. Nakada, preprint PSI–PR–91–02 Žunpublished..
w39x W.A. Bardeen, Proc. International Symposium on Heavy Flavor and Electroweak Theory, Beijing, P.R.
China, August 1995, p. 88 Žpreprint Fermilab–Conf–95r378–T..
w40x M.A. Shifman, A.I. Vainshtein, V.I. Zakharov, Nucl. Phys. B 147 Ž1979. 385.
w43x A.A. Ovchinnikov, A.A. Pivovarov, Sov. J. Nucl. Phys. 48 Ž1988. 120, Yad. Fiz. 48 Ž1988. 189.
w44x A.A. Ovchinnikov, A.A. Pivovarov, Phys. Lett. B 207 Ž1988. 333.
w45x S. Narison, A.A. Pivovarov, Phys. Lett. B 327 Ž1994. 341.
w46x A.A. Pivovarov, Talk given at the 3rd German Russian Workshop on Progress in Heavy Quark Physics,
Dubna, Russia, May 1996, hep-phr9606482.
w47x E. Bagan et al., Phys. Lett. B 278 Ž1992. 457.
Nuclear Physics B 572 Ž2000. 36–70
Type I vacua with brane supersymmetry breaking

C. Angelantonj a,b, I. Antoniadis b,c , G. D’Appollonio d , E. Dudas e, A. Sagnotti f
a ´ 1
´
Laboratoire de Physique Theorique de l’Ecole ´
Normale Superieure, 24, rue Lhomond,
F-75231 Paris CEDEX 05, France
b 2 ´
´
Centre de Physique Theorique, Ecole Polytechnique, F-91128 Palaiseau, France
c
ITP – Kohn Hall, UniÕersity of California, Santa Barbara, CA 93106-4030, USA
d
Dipartimento di Fisica, UniÕ. Firenze e INFN, Sez. di Firenze, Largo Enrico Fermi 2, I-50125 Firenze, Italy
e
LPT,3 Batiment
ˆ 210, UniÕersity Paris-Sud, F-91405 Orsay, France
f
Dip. di Fisica, UniÕ. Roma ‘‘Tor Vergata’’ and INFN, Sez. di Roma 2, Via della Ricerca Scientifica 1,
I-00133 Rome, Italy
Received 12 November 1999; accepted 21 January 2000
Abstract
We show how chiral type I models whose tadpole conditions have no supersymmetric solution
can be consistently defined introducing antibranes with non-supersymmetric world volumes. At
tree level, the resulting stable non-BPS configurations correspond to tachyon-free spectra, where
supersymmetry is broken at the string scale on some Žanti.branes but is exact in the bulk, and can
be further deformed by the addition of brane–antibrane pairs of the same type. As a result, a scalar
potential is generated, that can stabilize some radii of the compact space. This setting has the
novel virtue of linking supersymmetry breaking to the consistency requirements of an underlying
fundamental theory. q 2000 Elsevier Science B.V. All rights reserved.
1. Introduction
Type I models have become the subject of intense activity during the last few years,
since their perturbative definition w1–4x offers interesting new possibilities for low-en-
ergy phenomenology, and in particular leaves some freedom to lower the string scale
well below the Planck mass, if some extra dimensions are large w5–13x. Their consis-
1
Unite´ mixte du CNRS et de l’ENS, UMR 8549.
2
Unite´ mixte du CNRS et de l’EP, UMR 7644.
3
Unite´ mixte du CNRS, UMR 8627.
PII: S 0 5 5 0 - 3 2 1 3 Ž 0 0 . 0 0 0 5 2 - 3
C. Angelantonj et al.r Nuclear Physics B 572 (2000) 36–70 37
tency and a number of their most amusing features may be traced to the relation to
suitable ‘‘parent’’ models of oriented closed strings, from which their spectra can be
derived. In this procedure, a special role is played by ‘‘tadpole conditions’’ for
Ramond–Ramond ŽRR. states w14x. These may be regarded as global neutrality
conditions for RR charges w15x, constrain the Žinteger. Chan–Paton multiplicities, and
are usually linked to gauge and gravitational anomalies.
The explicit study of type I vacua, however, has revealed an unexpected difficulty: in
some interesting chiral four-dimensional models it is apparently impossible to satisfy
some of the tadpole conditions w16–20x. This peculiar phenomenon can often be traced
to sign flips of some crosscap contributions or, in more suggestive space-time language,
to the reversal of some orientifold charges. Examples are actually known where the
solution would lead to negative Chan–Paton multiplicities, thus violating the positivity
of the annulus amplitude, or where no solution can be found in general, because
crosscaps and boundaries scale differently with the internal volume.
In a recent work, it was shown how this difficulty can be evaded, in a prototype
six-dimensional Z 2 model and in the four-dimensional Z 2 = Z 2 orientifold, if one
relaxes the condition that the brane configuration be supersymmetric, thus allowing for
vacua including both branes and antibranes. As a result, supersymmetry is broken on
some collection of branes at the string scale, while it is preserved Žto lowest order. in
the bulk and possibly on other branes. Aside from its interest for the consistent
definition of type I models, this scenario, termed in w21x ‘‘brane supersymmetry
breaking’’, has clearly some beauty of its own if our non-supersymmetric universe is
modeled as a brane in a bath of higher-dimensional supergravity. Rather than being
introduced as a possible deformation, the breaking of supersymmetry in our low-energy
world would then be seen as a neat consequence of the internal consistency of the
underlying String Theory. Moreover, in this context the present experimental limits on
small-distance deviations from Newtonian gravity leave open the exciting possibility
that supersymmetry, broken on our world brane, be almost exact a millimeter away from
it. We would like to stress that this mechanism links supersymmetry breaking to the
string scale, rather than to geometric scales of the internal space, as Scherk–Schwarz
w22–34x or magnetic w35,36x deformations. The natural distinction between these two
settings, however, is closely linked to a geometric interpretation of the two-dimensional
Conformal Field Theory, and is somewhat blurred in non-geometric settings, such as
those recently explored in w37x.
Actually, all these vacua can be further deformed. In particular, in type I models one
can add arbitrary numbers of D9–D9 and D5–D5 brane pairs without violating any RR
tadpole condition w38,39x, but introducing additional breakings of supersymmetry on the
branes. For instance, starting from the Z 3 orientifold of Ref. w40x, one can thus build
semi-realistic three-generation models w41x. We would like to stress that the introduction
of Žanti.5-branes requires in general that additional Z 2 projections act on the boundaries
consistently with the closed spectrum. In this case the resulting configurations are
generally unstable, due to the presence of tachyonic modes that, however, can often be
lifted by suitable Wilson lines Žor, equivalently, by suitable brane displacements.. This
should be contrasted with the original setting for brane supersymmetry breaking w21x,
where the brane configuration involves branes and antibranes of different types and is by
construction stable. Still, in both settings the models include non-BPS Žanti.brane
38 C. Angelantonj et al.r Nuclear Physics B 572 (2000) 36–70
systems of the type considered in w42–48x, whose interactions with gravity are consis-
tently described by type I string theory.
In this work we discuss in more detail the open descendants of two typical examples
of type IIB orbifolds, where the problem of unsolvable tadpoles was first encountered.
We begin in Section 2 with a detailed discussion of four-dimensional Z 2 = Z 2 models
with discrete torsion, showing how the unavoidable reversal of some of the orientifold
charges is naturally accompanied, in a fully consistent construction, by the simultaneous
presence of branes and antibranes. We also discuss similar modifications of the open
descendants of the corresponding freely acting orbifolds recently studied in w33,34x. In
Section 3 we turn to the T 6rZ 4 orbifold, a canonical case where the tadpole conditions
can not be solved. Here we can display a consistent non-supersymmetric solution,
provided the Klein-bottle projection is also modified, with the net result of lifting the
offending twisted tadpole. We should stress, however, that the difficulties met in these
two classes of orbifolds are rather distinct. In type-I models, the twisted tadpoles are
generally related to non-abelian gauge anomalies, whereas the untwisted ones are related
to gravitational anomalies, absent in four dimensions. In the Z 2 = Z 2 models with
discrete torsion the twisted tadpoles are actually solÕable, whereas the untwisted ones
are not, and as a result even the naive supersymmetric open spectrum is free of
non-abelian gauge anomalies w16,17x. On the other hand, in the Z 4 model with standard
Klein bottle the problem is related to the twisted tadpoles, and as a result the naive
supersymmetric open spectrum has non-abelian gauge anomalies w18–20x. In Section 4
we discuss the possible deformations of type I models to stable vacuum configurations
including both branes and antibranes of the same type. In particular, we discuss
deformations of six-dimensional toroidal compactifications, that may thus include
arbitrary pairs of D9–D9 and D5–D5 branes, and present a similar generalization of the
four-dimensional T 6rZ 4 orientifold studied in Section 3. We show explicitly how the
presence of additional brane–antibrane pairs, via the resulting NS–NS ŽNeveu–Schwarz.
tadpoles, can actually stabilize the radii of the compact internal space. All the models
presented in this paper are meant to provide new instances of the phenomenon discussed
in w21x: to lowest order, the D branes are supersymmetric Žwith suitable diagonal
subgroups of the antibrane gauge groups realized as global symmetries., while the
antibranes are not. Finally, Section 5 contains our Conclusions and the two appendices
collect some relevant properties of the characters used in the text.
2. Z 2 = Z 2 orientifolds and discrete torsion
Let us begin by displaying the torus amplitude for the parent type IIB Z 2 = Z 2
orbifold. Aside from the identity, that we denote by o, the other three elements act on
the three internal two-tori as
g : Ž q,y,y . , f : Ž y,q,y . , h: Ž y,y,q . . Ž 2.1 .

In the Z 2 = Z 2 orbifold, one has the freedom of introducing discrete torsion w49–52x,
that corresponds to associating a sign e s "1 to the independent modular orbit
containing all terms that are twisted and projected by two different orbifold operations.
Omitting for brevity the contributions of the space-time bosons, the torus amplitudes of
the two Z 2 = Z 2 models are
2 2 2
4h 2 4h 2 4h 2
Ts 1
4
½ < To o < 2L1 L 2 L 3 q < To g < 2L1
q 22
q < To f < 2L2
q 22
q < To h < 2L 3
q 22
2 2 2 2
4h 2 4h 2 4h 2 4h 2
q< Tg o < 2L1 q < Tg g < 2L1 q < To f < 2L 2 q < Tf f < 2L 2
q42 q 32 q42 q 32
2 2
4h 2 4h 2
q< Th o < 2L3 q < Th h < 2L 3
q42 q 32
2
8h 3
2 2 2 2
qe < Tg h < q < Tg f < q < Tf g < q < Tf h < q < Th g < q < Th f <
ž 2 2
/ q 2q 3q4 5 , Ž 2.2 .
where the L k are lattice sums for the three internal tori and, throughout the paper, we let
a X s 2. We have expressed the torus amplitude in terms of the 16 quantities Ž k s o, g,h, f .
Tk o s t k o q t k g q t k h q t k f , Tk g s t k o q t k g y t k h y t k f ,
Tk h s t k o y t k g q t k h y t k f , Tk f s t k o y t k g y t k h q t k f , Ž 2.3 .
where the 16 Z 2 = Z 2 characters t k l w16,17x, combinations of products of level-one

SOŽ2. characters, are displayed in Appendix A. The choices e s .1 identify the models
with and without discrete torsion, whose low-energy spectra are quite different: in the
first N s 2 supergravity is coupled to 52 hypermultiplets and 3 vector multiplets, while
in the second it is coupled to 4 hypermultiplets and 51 vector multiplets. These spectra
correspond to orbifold limits of Calabi–Yau manifolds with Hodge numbers Ž51,3. and
Ž3,51., respectively.
The V projections that we are considering are implemented by the Klein-bottle
amplitudes
½
K s 18 Ž P1 P2 P3 q P1W2W3 q W1 P2W3 q W1W2 P3 . To o q 2 = 16
2
h
= e 1 Ž P1 q e W1 . Tg o q e 2 Ž P2 q e W2 . To f q e 3 Ž P3 q e W3 . Th o ž /5q4
,
Ž 2.4 .
where Pk and Wk denote the restrictions of the lattice sums L k to their momentum and
winding sublattices. Discrete torsion has a neat effect w33,34x on Ž Pk q e Wk .: if e s 1,
the massless twisted contributions are diagonal combinations of the t k l , and appear in
the Klein bottle, while if e s y1 they are off-diagonal combinations, and do not
contribute to it. Consistently with the crosscap constraint w53–56x, Ž2.4. can accommo-
date three additional signs e k . Actually, these are not independent, but are linked to the
parameter e by the constraint
e1 e2 e3 s e . Ž 2.5 .
One can write this amplitude as
2
h
Ks 1
8 ½Ž 1
P1 P2 P3 q PkWl Wm . To o q 2 = 16 e k Ž Pk q e Wk . Tk o
2 ž /5 q4
, Ž 2.6 .
where we have resorted to a compact notation, used extensively in the following:

summations over repeated indices and symmetrizations over distinct indices are left
implicit. An S transformation turns this expression into the corresponding vacuum-chan-
nel amplitude
25 Õk
Ks
K̃
8 ½ž Õ1Õ 2 Õ 3W1e W2e W3e q
2 Õ l Õm /
Wke Ple Pme To o
2
Pke 2h
ž
q2 e k Õ kWke q e
Õk / ž /5
To k
q2
, Ž 2.7 .
where the superscript e denotes the restriction of the lattice sums to their even terms and
the Õk denote the volumes of the three internal tori. At the origin of the lattices, the
constraint Ž2.5. leads to an expression whose coefficients are perfect squares,
2
25 Õ1 Õ2 Õ3
K0 s
K̃
8 ½ž (Õ Õ Õ
1 2 3 q e1 ( Õ2 Õ3
q e2 ( Õ 1Õ 3
q e3 ( / Õ 1Õ 2
to o
2
Õ1 Õ2 Õ3
q
ž(
Õ 1Õ 2 Õ 3 q e 1 ( Õ2 Õ3
y e2 ( Õ 1Õ 3
y e3 ( / Õ 1Õ 2
to g
2
Õ1 Õ2 Õ3
q
ž(
Õ 1Õ 2 Õ 3 y e 1 ( Õ2 Õ3
q e2 ( Õ 1Õ 3
y e3 ( / Õ 1Õ 2
to f
2
Õ1 Õ2 Õ3
q
ž(
Õ 1Õ 2 Õ 3 y e 1 ( Õ2 Õ3
y e2 ( Õ 1Õ 3
q e3 ( / 5 Õ 1Õ 2
to h , Ž 2.8 .
which shows rather neatly how the choice e k s y1 reÕerts the charge of the O5k
orientifold plane. While manifestly compatible with the usual positivity requirements,
this reversal clearly affects the tadpole conditions, that require the introduction of
antibranes. In this respect, it should be appreciated that, according to Ž2.5., discrete
torsion implies the reversal of at least one of the O5 charges. Therefore, taking into
account the presence of the e k , one can identify four classes of models, determined by
the independent choices for Ž e 1 , e 2 , e 3 .. If e s 1, the choice Žq,q,q . gives the model
discussed in w16,17,57–59x, with 48 chiral multiplets from the closed twisted sectors,
while the choice Žq,y,y . gives a model with 16 chiral multiplets and 32 vector
multiplets from the twisted sectors. On the other hand, for e s y1 the two choices
Žq,q,y . and Žy,y,y . yield the same massless twisted spectrum, namely 48 chiral
multiplets.
In order to describe the annulus amplitude, it is convenient to introduce a compact
notation. If TkNS
l
ŽTkRl . denote the NS ŽR. parts of the usual combinations of supersym-
metric Z 2 = Z 2 characters, we thus define
T˜kŽl´ . s TkNS R
l y ´ Tk l , Ž 2.9 .
where ´ s "1. Whereas T˜kŽq. l
Žs TkŽq.
l
., which in the following will simply be denoted
by Tk l for the sake of brevity, form a closed set under S modular transformations, the
additional combinations T˜kŽy.
l with reversed RR charges, associated to the interactions
between branes and antibranes, do not. As a result, the corresponding terms in A,
describing open strings stretched between branes and antibranes, contain new combina-
tions TkŽy. Žq.
l , obtained from the Tk l interchanging O 2 with V2 and S2 with C2 in the last
three factors, as explained in Appendix A.
The transverse-channel annulus amplitude is
2y5 D k2; o Õ k
As
Ã
8 ½ž No2 Õ 1Õ 2 Õ 3W1W2W3 q
2 Õ l Õm
Wk Pl Pm To o
/
2
Pk 2h
q4 Ž Nk2 q D k2; k . ÕkWk q D l2/ k ; k
Õk
Tk o
ž /
q4
2 2
2h 2h
q2 No D k ; o ÕkWk TõŽke k . q 2 Nk D k ; k ÕkWk T˜kŽke k .
ž / q2 ž / q3
2
8h 3 Pm 2h
q4Nl D k / l ;l T˜lŽke k . TõŽme k e l .
q 2q 3q4
q Dk ; o Dl ; o
Õm ž /
q2
2
Pm 2h 8h 3
qDk ; m D l ; m
Õm
T˜mŽ emk e l .
ž /
q3
q 4 D k ; k D l ; k T˜kŽme k e l .
q 2q 3q4 5 , Ž 2.10 .
where No , Dg ; o , Df ; o and D h; o are the charges for the D9 branes and for the three sets
of D5 or D5 branes wrapped around the first, second and third torus. In a similar
fashion, Nk , Dg ; k , Df ; k and D h; k Ž k s g, f,h. parametrize the breakings induced by the
three orbifold operations g, f and h. As expected, the RR part of every term describing
the interaction between a brane and an antibrane has a reversed sign. The untwisted
terms at the origin of the lattice sums rearrange themselves into perfect squares:
2
2y5 Õ1 Õ2 Õ3
A0 s
Ã
8 ½ž (
No Õ 1Õ 2 Õ 3 q Dg ; o ( Õ2 Õ3
q Df ; o ( Õ 1Õ 3
q Dh ; o ( / Õ 1Õ 2
toNS
o
2
Õ1 Õ2 Õ3
ž (
y No Õ 1Õ 2 Õ 3 q e 1 Dg ; o ( Õ2 Õ3
q e 2 Df ; o ( Õ 1Õ 3
q e 3 Dh , o ( / Õ 1Õ 2
toRo
2
Õ1 Õ2 Õ3
ž (
q No Õ 1Õ 2 Õ 3 q Dg ; o ( Õ2 Õ3
y Df ; o ( ( /
Õ 1Õ 3
y Dh ; o
Õ 1Õ 2
toNS
g
2
Õ1 Õ2 Õ3
ž (
y No Õ 1Õ 2 Õ 3 q e 1 Dg ; o ( Õ2 Õ3 (
y e 2 Df ; o( / Õ 1Õ 3
y e 3 Dh ; o
Õ 1Õ 2
toRg
2
Õ1 Õ2 Õ3
ž (
q No Õ 1Õ 2 Õ 3 y Dg ; o ( Õ2 Õ3
q Df ; o ( ( /
Õ 1Õ 3
y Dh ; o
Õ 1Õ 2
toNS
f
2
Õ1 Õ2 Õ3
ž (
y No Õ 1Õ 2 Õ 3 y e 1 Dg ; o ( Õ2 Õ3 (
q e 2 Df ; o( / Õ 1Õ 3
y e 3 Dh ; o
Õ 1Õ 2
toRf
2
Õ1 Õ2 Õ3
ž (
q No Õ 1Õ 2 Õ 3 y Dg ; o ( Õ2 Õ3
y Df ; o ( ( /
Õ 1Õ 3
q Dh ; o
Õ 1Õ 2
toNS
h
2
Õ1 Õ2 Õ3
ž (
y No Õ 1Õ 2 Õ 3 y e 1 Dg ; o ( Õ2 Õ3 (
y e 2 Df ; o( / 5 Õ 1Õ 3
q e 3 Dh ; o
Õ 1Õ 2
toRh .
Ž 2.11 .
Moreover, the breaking terms reflect rather neatly the geometry of the Žanti.brane
configuration. Indeed, the coefficient that multiplies a given twisted character is a sum
of squares associated to the fixed tori of the various twisted sectors, and each square
contains the breaking terms for the branes present in the fixed tori, with factors 'Õ if
they are wrapped around them or 1r 'Õ if they are localized on them. The relative
coefficients of these terms are also directly linked to the brane geometry, and are given
by
a of fixed tori
( a of occupied fixed tori
. Ž 2.12 .
Thus, for a given twisted sector, the numerator counts the fixed tori, while the
denominator counts the fixed tori where branes are actually present. Moreover, the R
portions of the characters describing brane–antibrane exchanges have reverted signs also
in these twisted contributions, as expected. For instance, in the g-twisted sector of the
Žqqy . model, that contains D53 branes, the reflection coefficients for the massless
modes in tg h are
2
2y5 1 1
8 ž ( (
Ng Õ1 y 4 Dg ; g Õ 1 y 2 Df ; g
(Õ 1
q 2 Dh ; g
(Õ 1
/
2 2
1 1
ž (
q3 Ng Õ 1 y 2 Df ; g
(Õ 1
/ ž (
q 3 Ng Õ1 q 2 D h ; g
(Õ 1
/ q 9Ng2 Õ1 Ž 2.13 .
for the NS portion, and

2
2y5 1 1
8 ž ( (
Ng Õ1 y 4 Dg ; g Õ 1 y 2 Df ; g
(Õ 1
y 2 Dh ; g
(Õ 1
/
2 2
1 1
ž
q3 Ng Õ 1 y 2 Df ; g( (Õ 1
/ ž (
q 3 Ng Õ1 y 2 D h ; g
(Õ 1
/ q 9Ng2 Õ1 Ž 2.14 .
for the R portion. According to Ž2.12., the coefficient of Ng is Õ 1 , since the D9 are (
wrapped around all fixed tori, the coefficient of Dg ; g is 4 Õ1 , since the D51 are only (
wrapped around one fixed torus, while the coefficients of Df ; g and Dh; g are 2r Õ1 , (
since the D52 and D53 are confined to four of the fixed tori. Finally, out of the 16
g-fixed tori, one sees all the branes, three see only the D9 and the D52 , three see only
the D9 and the D53 and, finally, nine see only the D9.
The direct-channel annulus amplitude is then
D k2; o
As 1
8 ½ž No2 P1 P2
2
P3 q
2
PkWl Wm To o q
/
2
Ž Nk2 q Dk2; k . Pk q Dl2/ k ; kWk
2
2h h h
=To k
ž / q2
q 2 No D k ; o Pk TkŽoe k . ž / q4
y 2 Nk D k ; k Pk TkŽke k . ž /
q3
2
kq l
2h 3 h
q2 i Ž y1 . Nl D k / l ;l TkŽle k . q D k ; o Dl ; oWm TmŽ eok e l . ž /
q 2q 3q4 q4
2
h 2h 3
yDk ; m D l ; mWm TmŽ emk e l . ž /
q3
q 2 i Ž y1 .
mq k
D k ; k D l ; k TmŽ ekk e l .
q 2q 3q4 5 ,
Ž 2.15 .
where in the signs Žy1. kq l and Žy1. mq k k,l,m take the values 1,2,3 for the g, f, and h
generators. The transverse-channel amplitudes K˜ and A˜ determine by standard meth-
¨
ods the transverse Mobius amplitude
2
2hˆ Õk
½
M˜s y 14 No Õ 1Õ 2 Õ 3W1e W2e W3e Tô o q No Õ kWkee k Tô k
ž /q 2̂
q
2 Õ l Õm
2
Pme ˆ Ž e . 2ĥ
=D k ; oWke Ple Pme e k TˆõŽoe k . q D l ; o e k T˜ l q D k ; o ÕkWke TˆõŽke k .
ž Õm o m /ž / 5 q 2ˆ
Ž 2.16 .
¨
and, after a P transformation, the direct-channel Mobius amplitude
2
2ĥ
q 12 D k ; o PkWl Wm e k T˜ôŽoe k .
½
M s y 18 No P1 P2 P3 Tô o y No Pk e k Tô k
ž / q 2̂
2
2ĥ
y Dl ; o e kWm TˆõŽme l . q D k ; o Pk TˆõŽke k .
ž /ž / 5 . Ž 2.17 .
q 2̂
From the transverse amplitudes one can now read the tadpole conditions
No s 32, Ng s Nf s Nh s 0,
D k ; o s 32, D k ; g s D k ; f s Dk ; h s 0. Ž 2.18 .
2.1. Massless spectra
The models where only one e k is negative have discrete torsion and contain one D5.
For the D9 and the two sets of D5 branes, the gauge groups are UŽ8. = UŽ8., with
N s 1 supersymmetry, while for the D5 branes the gauge group is USpŽ8. 4 , with
N s 0. Moreover, the 59 and 5k 5l strings are supersymmetric, while the 95 and 5k 5
strings are not. Let us discuss in some detail the case Ž e 1 , e 2 , e 3 . s Žqqy ., that
contains D5 branes wrapped around the third torus. To this end, let us parametrize the
charges as
No s o q g q o q g , Ng s i Ž o q g y o y g . ,
Nf s i Ž o y g y o q g . , Nh s o y g q o y g ,
Dg ; o s o 1 q g 1 q o 1 q g 1 , Dg ; g s i Ž o 1 q g 1 y o 1 y g 1 . ,
Dg ; f s o 1 y g 1 q o 1 y g 1 , Dg ; h s yi Ž o 1 y g 1 y o 1 q g 1 . ,
Df ; o s o 2 q g 2 q o 2 q g 2 , Df ; g s o 2 y g 2 q o 2 y g 2 ,
Df ; f s i Ž o 2 q g 2 y o 2 y g 2 . , Df ; h s i Ž o 2 y g 2 y o 2 q g 2 . ,
D h ; o s a q b q c q d, D h ; g s a q b y c y d,
D h ; f s a y b q c y d, Dh ; h s a y b y c q d, Ž 2.19 .
and extract the massless spectrum from the amplitudes at the origin of the lattices. The
99, 51 51 and 52 52 sectors have N s 1 supersymmetry, and all give gauge groups
UŽ8. = UŽ8., with chiral multiplets in the representations Ž8,8., Ž8,8., Ž28,1., Ž1,28. and
their conjugates. Moreover, as expected, the 951 , 952 and 51 52 strings are also
supersymmetric, and contain chiral multiplets in the representations
951 : Ž 8,1;1,8 . , Ž 1,8;8,1 . , Ž 8,1;8,1 . , Ž 1,8;1,8 . ,

952 : Ž 8,1;1,8 . , Ž 1,8;8,1 . , Ž 8,1;8,1 . , Ž 1,8;1,8 . ,
51 52 : Ž 8,1;8,1 . , Ž 1,8;8,1 . , Ž 8,1;1,8 . , Ž 1,8;1,8 . .
On the other hand, the strings whose ends live on the antibrane give rise to
supersymmetric spectra, even if the annulus contains supersymmetric characters, since
bosons and fermions are treated differently by M . Thus, the 53 53 sector contributes a
gauge group USpŽ8. 4 , with Weyl spinors in the Ž28,1,1,1. and in three additional
permutations, and chiral multiplets in the Ž8,8,1,1. and in five additional permutations.
Finally, the strings stretched between a brane and an antibrane have non-supersymmetric
spectra, with Weyl spinors and complex scalars in the representations
953 spinors: Ž 8,1;8,1,1,1 . , Ž 1,8;1,8,1,1 . , Ž 1,8;1,1,8,1 . , Ž 8,1;1,1,1,8 . ,

scalars: Ž 8,1;1,8,1,1 . , Ž 1,8;8,1,1,1 . , Ž 1,8;1,1,1,8 . , Ž 8,1;1,1,8,1 . ,
51 53 spinors: Ž 8,1;1,1,8,1 . , Ž 1,8;1,1,1,8 . , Ž 1,8;1,8,1,1 . , Ž 8,1;8,1,1,1 . ,
scalars: Ž 8,1;1,1,1,8 . , Ž 1,8;1,1,8,1 . , Ž 1,8;8,1,1,1 . , Ž 8,1;1,8,1,1 . ,
52 53 spinors: Ž 8,1;8,1,1,1 . , Ž 1,8;1,1,1,8 . , Ž 1,8;1,1,8,1 . , Ž 8,1;1,8,1,1 . ,
scalars: Ž 8,1;1,1,1,8 . , Ž 1,8;1,8,1,1 . , Ž 1,8;8,1,1,1 . , Ž 8,1;1,1,8,1 . .
The choice Ž e 1 , e 2 , e 3 . s Žyyy . also corresponds to a model with discrete torsion. In
this case, however, there are D9 branes and three sets of D5 branes, while the charges
are to be parametrized as
No s a q b q c q d, Ng s a q b y c y d,
Nf s a y b q c y d, Nh s a y b y c q d,
Dg ; o s o 1 q g 1 q o 1 q g 1 , Dg ; g s o 1 y g 1 q o 1 y g 1 ,
Dg ; f s i Ž o 1 q g 1 y o 1 y g 1 . , Dg ; h s i Ž o 1 y g 1 y o 1 q g 1 . ,
Df ; o s o 2 q g 2 q o 2 q g 2 , Df ; g s i Ž o 2 q g 2 y o 2 y g 2 . ,
Df ; f s o 2 y g 2 q o 2 y g 2 , Df ; h s yi Ž o 2 y g 2 y o 2 q g 2 . ,
Dh ; o s o 3 q g 3 q o 3 q g 3 , Dh ; g s i Ž o 3 q g 3 y o 3 y g 3 . ,
Dh ; f s i Ž o 3 y g 3 y o 3 q g 3 . , Dh ; h s o 3 y g 3 q o 3 y g 3 . Ž 2.20 .
The D9 branes have N s 1 supersymmetry, with gauge group SOŽ8. 4 and chiral
multiplets in the Ž8,8,1,1. and five permutations. Moreover, each antibrane gives a
non-supersymmetric spectrum, with gauge group UŽ8. = UŽ8., chiral multiplets in the
Ž8,8., Ž8,8. and in their conjugates, spinors in the Ž28,1., Ž 28,1., Ž1,28., Ž1,28 . and
complex scalars in the Ž36,1., Ž 36,1., Ž1,36., Ž1,36 .. We would like to stress that in this
¨
case the gauginos are massless, since the Mobius amplitude does not affect the adjoint
representations of unitary groups. Finally, 5k 5l sectors give chiral multiplets in the
representations
51 52 Ž 8,1;8,1 . , Ž 8,1;1,8 . , Ž 1,8;8,1 . , Ž 1,8;1,8 . ,

51 53 Ž 8,1;8,1 . , Ž 8,1;1,8 . , Ž 1,8;1,8 . , Ž 1,8;8,1 . ,
52 53 Ž 8,1;8,1 . , Ž 8,1;1,8 . , Ž 1,8;1,8 . , Ž 1,8;8,1 . ,
and the non-supersymmetric 95k sectors contain Weyl spinors and complex scalars in
the representations
951 spinors: Ž 8,1,1,1;8,1 . , Ž 1,8,1,1;8,1 . , Ž 1,1,8,1;1,8 . , Ž 1,1,1,8;1,8 . ,

scalars: Ž 8,1,1,1;1,8 . , Ž 1,8,1,1;1,8 . , Ž 1,1,8,1;8,1 . , Ž 1,1,1,8;8,1 . ,
952 spinors: Ž 8,1,1,1;8,1 . , Ž 1,8,1,1;1,8 . , Ž 1,1,8,1;8,1 . , Ž 1,1,1,8;1,8 . ,
scalars: Ž 8,1,1,1;1,8 . , Ž 1,8,1,1;8,1 . , Ž 1,1,8,1;1,8 . , Ž 1,1,1,8;8,1 . ,
953 spinors: Ž 8,1,1,1;8,1 . , Ž 1,8,1,1;1,8 . , Ž 1,1,8,1;1,8 . , Ž 1,1,1,8;8,1 . ,
scalars: Ž 8,1,1,1;1,8 . , Ž 1,8,1,1;8,1 . , Ž 1,1,8,1;8,1 . , Ž 1,1,1,8;1,8 . .
All chiral spectra thus obtained are free of non-abelian anomalies.
On the other hand, the models without discrete torsion are not chiral. The choice
Ž e 1 , e 2 , e 3 . s Žqqq ., discussed in w16,17x and worked out in detail in w57–59x, leads to
a gauge group USpŽ16. 4 . Another model, without discrete torsion but with two D5
branes, can be obtained letting two of the e k be negative. The D9 and D5 branes give
orthogonal gauge groups with N s 1 supersymmetry, while the two D5 branes give
symplectic gauge groups with no supersymmetry. For instance, with the choice Žqyy .
A0 s 18 ½Ž N 2 2 2 2
o q Dg ; o q Df ; o q D h ; o . To o q 2 No Dg ; oTg o q 2 No Df ; oT oŽyf .
q2 No Dh ; o T hŽ y . Žy . Žy .
5
o q 2 Dg ; o Df ; o T h o q 2 Dg ; o D h ; o T o f q 2 Df ; o D h ; o Tg o ,
M0 s y 14 ½Ž N qD
o g ;o . Ž t o o y t o g q t o f q t o h . y Ž Df ; o q D h ; o .
= Ž toNS NS NS NS R R R R
o y to g q to f q to h . q Ž to o y to g q to f q to h . 5, Ž 2.21 .
and there are no breaking terms in the annulus. After a suitable rescaling of the charge
multiplicities, the D52 and the D53 branes give non-supersymmetric spectra, with
USpŽ16. gauge groups, spinors in the 136 and in three copies of the 120 and scalars in
the 120 and in two copies of the 136. The 99 and 51 51 sectors have N s 1 supersymme-
try, gauge group SOŽ16. and chiral multiplets in the 136 and in two copies of the 120.
Finally, there are two chiral multiplets in the representation Ž16,16. arising from the 951
and the 52 53 sectors and complex scalars and Weyl spinors in bifundamental representa-
tions arising from the 952 , 953 , 51 52 and 51 53 sectors.
The fact that the Z 2 = Z 2 orientifold with discrete torsion naturally leads to non-su-
persymmetric spectra can also be argued considering F-theory on the T 8rZ 2 = Z 2 = Z 2
with discrete torsion w19x, since the blow-up of this eight-dimensional orbifold is only
birational to a Calabi–Yau fourfold 4 .
4
We would like to thank Zurab Kakushadze and Koushik Ray for calling this correspondence to our
attention.
2.2. Comments on freely acting Z 2 = Z 2 orientifolds with brane supersymmetry break-

ing
In all the models discussed in w33,34x, one can introduce D5 branes simply reverting
some signs in the twisted sectors of the Klein bottle, as in the previous case. This
procedure generates models with branes, antibranes and various amounts of supersym-
metry. For instance, the p 23 model of w33,34x, with 12 -momentum shifts along the last
two tori, can be turned into a different model, described by
½
K s 18 To o P1 P2 P3 q P1W2W3 q W1 Ž y1 .
m2
P2W3 q W1W2 Ž y1 .
m3
P3
2
h
y2 = 16Tg o P1 ž /5
q4
,
Dg2 ; o
½
A s 18 To o No2 P1 P2 P3 q
2
P1 Ž W2W3 q W2nq1r2 W3nq1r2 .
2 2
2h m2 2h
qŽ Ng2 q 2 Dg2 ; g . To g P1ž / q2
2
q Nf To f Ž y1 . P2
ž / q2
2 2
m3 2h h
qNh2 To h Ž y1 . P3
ž /
q2
q 2 No Dg ; o TgŽy.
o P1 ž /
q4
2
h
y4Ng Dg ; g TgŽy.
g P1 ž /5q3
,
2
2ĥ
Tô o No P1 P2 P3 y TˆõŽoy . Dg ; o P1W2W3 q Tô g No y TˆõŽgy . Dg ; o P1
Msy 1
8 ½ž / ž / ž /
q 2̂
2 2
2hˆ 2hˆ
qNo Tô f Ž y1 .
m2
P2
ž /
q 2ˆ
q No Tô h Ž y1 .
m3
P3
ž /5
q 2ˆ
. Ž 2.22 .
The 99 sector gives a gauge group SOŽ16 y a. = SOŽ a. = SOŽ16 y c . = SOŽ c ., with
N s 1 supersymmetry and chiral multiplets in the representation Ž16 y a,a,1,1. and in
five additional ones differing by permutations of the entries. The 51 51 sector gives a
gauge group USpŽ8. = USpŽ8., with complex scalars in the Ž36,1. and Ž1,36., Weyl
spinors in two copies of the Ž28,1. and Ž1,28., and one N s 2 hypermultiplet in the
Ž8,8.. Finally, the 951 sector gives Weyl spinors in the representations Ž16 y a,1,1,1;8,1.,
Ž1,a,1,1;8,1., Ž1,1,16 y c,1;1,8., Ž1,1,1,c;1,8. and complex scalars in the representations
Ž16 y a,1,1,1;1,8., Ž1,a,1,1;1,8., Ž1,1,16 y c,1;8,1., Ž1,1,1,c;8,1.. In the decompactifica-
tion limit R 2 , R 3 ™ `, local tadpole cancellation requires a further breaking of the D5
gauge group to USpŽ4. 4 , and the resulting configuration may be linked to the Z 2
orientifold discussed in w21x.
3. Open descendants of the T 6 r Z 4 orbifold
The Z 4 orbifold is obtained identifying the complex world-sheet coordinates X a on

the three internal tori and the world-sheet fermions l a according to X a ; v a X a,
l a ; v a l a, where v a s e 2 p i t a Ž a s 1,2,3. and the twist vector has components Ž t 1 ,t 2 ,t 3 .
s Ž 14 , 14 ,y 12 .. In order to display the contributions of the l a, it is convenient to
decompose the original level-one SOŽ8. characters with respect to SOŽ2. = SOŽ2. =
SUŽ2. = UŽ1.. To this end, we introduce the two level-one SUŽ2. characters Ž x 0 , x 1r2 .,
of conformal weights Ž0, 14 ., and the eight UŽ1. characters j m Ž m s 0," 1," 2," 3,4.
for a boson on a circle of radius R s '8 , of conformal weights h m s m2r16. The
contribution of the l a to the sector twisted by v k and projected by v l , Ž k,l s 0,1,2,3.,
can then be expressed as
l l
c k l s r k 0 q i lr k1 q Ž y1 . r k 2 q Ž yi . r k 3 , Ž 3.1 .
where the characters r k l are collected in Appendix B, while the contributions of the
internal Žlattice. bosons are
d k ,0
pl h
f k l s y2sin ž / 4
u
1r2 q kr4
. Ž 3.2 .
1r2 q lr4
The torus amplitude is then

2 4
2h 2h
Ts 1
4
½ < c 00 < 2L1 L2 L3 q c 01 c 03 f 01
2 2
f 03
q2
q < c 02 < 2
q2
L3
2 4 4
2h h h
2
qc 03 c 01 f 03 2
f 01
q2 ž
q 16 < c 20 < 2
q4
L 3 q < c 22 < 2
q3
L3
/
2 2
2h 2h
q4
ž 2
c 21 c 23 f 21 2
f 23
q2
2
q c 23 c 21 f 23 2
f 21
q2 /
2 2 2
h h h
ž 2
q16 c 10 c 30 f 10 2
f 30
q4
2
q c 11 c 33 f 11 2
f 33
q3
2
q c 12 c 32 f 12 2
f 32
q4
2 2 2
2 2
h 2 2
h 2 2
h
qc 13 c 31 f 13 f 31 q c 30 c 10 f 30 f 10 q c 31 c 13 f 31 f 13
q3 q4 q3
2 2
h h
2
qc 32 c 12 f 32 2
f 12
q4
2
q c 33 c 11 f 33 2
f 11
q3 /5 . Ž 3.3 .
Before the V projection, the massless sector describes N s 2 supergravity coupled to 7

vector multiplets and 32 hypermultiplets, and can thus be associated with the singular
limit of a Calabi–Yau manifold with Hodge numbers Ž h11 ,h 21 . s Ž31,7..
Let us now turn to the construction of the open descendants, starting from the
Klein-bottle amplitude
2
2h
Ks 1
8
½ c 00 Ž P1 P2 P3 q W1W2 P3 . q 2 c 02
ž /
q2
W3
2 2
h h
qe 2 = 16 c 20 ž / q4
P3 q 2 = 4 c 22 ž /
q3
W3
5 , Ž 3.4 .
where, as in the previous cases, we have inserted a sign e s "1 in front of the twisted
contributions. The choice e s 1 gives the usual V projection, while the choice e s y1
inverts the charge of the O5 plane and, as in the previous examples, requires the
introduction of antibranes. In the transverse channel, this amplitude turns into
2
25 Õ3 h P3e
Ks
K̃
8 ½ ž
c 00 Õ1Õ 2 Õ 3W1e W2e W3e q
Õ 1Õ 2 /
P1e P2e W3e q 2 c 20 ž /
q4 Õ3
2 2
2h h P3e
q2 e c 02
ž /
q2
W3e Õ 3 y 2 e c 22 ž / q3 Õ3 5 , Ž 3.5 .
and a closer look at the contributions at the origin of the internal lattices
2 2
25 Õ3 Õ3
K0 s
K̃
8 ½ ž(Õ 1Õ 2 Õ 3 q e ( / Õ 1Õ 2 ž
Ž r 00 q r 02 . q (Õ1Õ 2 Õ 3 y e ( / Õ 1Õ 2
4
= Ž r 01 q r 03 . q
Õ3
1
2 Ž 1 y e . Ž r 20 q r 22 . q Ž 1 q e . Ž r 21 q r 23 . 5
Ž 3.6 .
reveals the presence of a term proportional to 1rÕ 3 . For e s 1, this gives rise to a
massless tadpole that can not be canceled by the annulus and Mobius¨ contributions. For
e s y1, however, this term becomes massive, and one can complete the construction of
the open descendants without any further difficulties. In this case, the projected massless
closed spectrum is N s 1 supergravity coupled to 6 vector multiplets and 33 chiral
multiplets. For the open sector, it is convenient to introduce a compact notation, as in the
previous section, defining
c˜ kŽ l´ . s c kNS R
l y ´ ck l , Ž 3.7 .
and denoting by
l l l
c kŽy.
l s s k 0 q i s k1 q Ž y1 . s k 2 q Ž yi . s k 3 Ž 3.8 .
the combinations of characters obtained from c˜ lŽy.
k after an S modular transformation.
The explicit definition of the s k l may be found in Appendix B.
The transverse-channel annulus amplitude is then

2y5 Õ3
As
Ã
8 ½ž N 2 Õ 1Õ 2 Õ 3W1W2W3 q D 2
Õ 1Õ 2
P1 P2W3 c 00 /
2
h h
q32 Ž R 2
q R 2D . 2
c 30 f 30 ž / q4
q 16 Ž S 2
q SD2 . c 20 ž / q4
Õ 3W3
2
h 2h
q 2 NDc˜ 02 Õ 3W3 q 32 RR D c˜ 32
q32 Ž T 2 q TD2 . c 10 f 10
2
ž /
q4
Žy.
ž / q2
Žy. 2
f 32
2
h h h
= ž / q4
y 8SSD c˜ 22
Žy.
ž /
q3
Õ 3W3 q 32TTD c˜ 12
Žy. 2
f 12 ž /5
q4
Ž 3.9 .
and, in particular, the untwisted contributions at the origin of the lattices are
2
2y5 Õ3
A0 s
Ã
8 ½ž ( N Õ 1Õ 2 Õ 3 q D ( / Õ 1Õ 2
Ž r 00 q r 02 .
NS
2
Õ3
ž(
q N Õ 1Õ 2 Õ 3 y D ( / Õ 1Õ 2
Ž r 01 q r 03 .
NS
2
Õ3
ž(
y N Õ 1Õ 2 Õ 3 y D ( / Õ 1Õ 2
Ž r 00 q r 02 .
R
2
Õ3
ž(
y N Õ 1Õ 2 Õ 3 q D ( / Õ 1Õ 2
Ž r 01 q r 03 .
R
5 . Ž 3.10 .
It should be appreciated that the structure of the breaking terms is consistent with Eq.
Ž2.12.. Thus, in the sector twisted by v 2 they are
2
Ž S q 4SD . q 15S 2 s 16S 2 q 16SD2 q 8SSD , Ž 3.11 .
since the branes are present only in one fixed torus, while in the sector twisted by v Žor
v 3 . they are
2
4 Ž R q 2 R D . q 12 R 2 s 16 R 2 q 16 R 2D q 16 RR D , Ž 3.12 .
since the branes fill the third torus, and are thus present in four of the v-fixed points.
As usual, the transverse-channel amplitudes K˜ and A˜ determine
2
2hˆ Õ3
P1e P2e W3ecˆ˜ 00
M˜s y 1
4 ½ ˆ
NÕ1Õ 2 Õ 3W1e W2e W3ec 00 y NÕ 3W3ec 02
2
ˆ
ž /
q 2̂
yD
Õ 1Õ 2
Žy.
2hˆ 2hˆ
qDÕ 3W3ecˆ˜ 02 q 2 Scˆ 21 y SD cˆ˜ 21Ž y . fˆ 21
Žy.
ž / q 2ˆ ž 2
/ ž / q 2ˆ
2ĥ
q2 Scˆ 23 y SD c˜ˆ 23Ž y . fˆ 23
ž 2
/ ž /5q 2̂
, Ž 3.13 .
and the tadpole conditions, now solvable, are
N s D s 32, R s R D s S s SD s T s TD s 0. Ž 3.14 .
After S and P transformations, one can then recover from Ž3.9. and Ž3.13. the
direct-channel amplitudes
2h
A s 18
½ Ž N 2 P1 P2 P3 q D 2 W1W2 P3 . c 00 q Ž R 2 q R 2D . c 01 f 012 ž /
q2
2
2h 2h
qŽ S 2
q SD2 . c 02 ž / q2
P3 q Ž T 2 q TD2 . c 03 f 03
2
ž /q2
2 2
h 2h h
q2 NDc 20
Žy.
ž /q4
Žy. 2
P3 q 2 RR D c 21 f 21
ž /
q2
q 2 SSD c 22
Žy.
ž /q3
P3
2h
Žy. 2
q2TTD c 23 f 23
ž /5 q2
Ž 3.15 .
and
2 2
2hˆ 2hˆ
y DW1W2 P3 c˜ˆ 00 y DP3 c˜ˆ 02
½
M s y 18 NP1 P2 P3 cˆ 00 q NP3 cˆ 02
ž /
q 2ˆ
Žy. Žy.
ž / q 2ˆ
2hˆ 2hˆ 2hˆ

qScˆ 01 fˆ 01 q Scˆ 03 fˆ 03 y SD c˜ˆ 01
Žy. ˆ 2
2
ž / q 2ˆ
2
ž /
q 2ˆ
f 01
ž /
q 2ˆ
2ĥ
ySD cˆ˜ 03
Žy. ˆ 2
f 03
ž /5 q 2̂
. Ž 3.16 .
As usual, the coefficients in M˜ are fixed by factorization in the tube channel. Some sign
ambiguities apparently present for massive characters, whose coefficients are not
constrained by tadpoles, are fixed demanding that the direct-channel amplitudes have a
correct particle interpretation. The latter calls for the parametrization
N s n q m q p q m, Dsdqrqqqr,
R s n q im y p y im, R D s d q ir y q y ir ,
S s n y m q p y m, SD s d y r q q y r ,
T s n y im y p q im, TD s d y ir y q q ir , Ž 3.17 .
that implements the Z 4 orbifold projections on the Chan–Paton charges.
One can now read the massless open spectrum from the amplitudes Ž3.15., Ž3.16.
restricted to the origin of the internal lattice
A0 q M0 s 12 r 00 Ž n 2 q p 2 q 2 mm q d 2 q q 2 q 2 rr .
½
qr 01 Ž 2 nm q 2 pm q 2 dr q 2 qr .
qr 02 Ž 2 np q m 2 q m2 q 2 dq q r 2 q r 2 .
qr 03 Ž 2 nm q 2 pm q 2 dr q 2 qr . q s 20 Ž nd q pq q mr q mr .
qs 21 Ž nr q mq q pr q md . q s 22 Ž nq q mr q pd q mr .
qs 23 Ž nr q md q pr q mq . y 12 rˆ 00 Ž n q p .
5
yrˆ˜ 00
Žy .
Ž d q q . q rˆ 02 Ž m q m . y rˆ˜ 02Ž y . Ž r q r . 4 , Ž 3.18 .
while the tadpole conditions select the gauge group SOŽ8. 9 = SOŽ8. 9 = UŽ8. 9 = USpŽ8.5
= USpŽ8.5 = UŽ8.5 . The D9 spectrum has N s 1 supersymmetry, with two chiral
multiplets in the Ž8,1,8. and in the Ž1,8,8. representations, and one chiral multiplet in
each of the Ž1,1,28., Ž1,1,28 . and Ž8,8,1. representations. On the other hand, the D5
spectrum is not supersymmetric, and contains, aside from the corresponding gauge
bosons, spinors in the Ž28,1,1., Ž1,28,1., Ž1,1,64., Ž1,1,28., Ž1,1,28 ., complex scalars in
the Ž1,1,36., Ž1,1,36 ., and chiral multiplets in the Ž8,8,1. and in two copies of the Ž8,1,8.
and Ž1,8,8.. Finally, the ND strings contain complex scalars in the Ž8,1,1;8,1,1.,
Ž1,8,1;1,8,1., Ž1,1,8;1,1,8. and Ž1,1,8;1,1,8., and Weyl spinors in the Ž8,1,1;1,1,8.,
Ž1,1,8;1,8,1., Ž1,8,1;1,1,8. and Ž1,1,8;8,1,1.. This spectrum is chiral, but is free of
irreducible gauge anomalies. As in previous examples, if suitable diagonal subgroups of
the D5 factors are regarded as global symmetries, the D9 spectrum has N s 1
supersymmetry.
4. Type I vacua with branes and antibranes of the same type
In the previous examples we have seen how the structure of the closed sector
determines both the types and the total numbers of D branes present in the open
descendants. In these cases, the breaking of supersymmetry on the branes is directly
enforced by the consistency of the model.
Somewhat different scenarios have been recently proposed in w38,39,41x. In the
resulting models, a supersymmetric open sector is deformed allowing for the simultane-
ous presence of branes and antibranes of the same type. Whereas tadpole conditions only
fix the total RR charge, the option of saturating it by a single type of D brane, whenever
available, stands out as the only one compatible with space-time supersymmetry.
However, if one relaxes this last condition, there are no evident obstructions to
considering vacuum configurations where branes and antibranes with a fixed total RR
charge are simultaneously present. The rules for constructing this wider class of models
can be simply presented referring to a ten-dimensional example.
The starting point is the familiar supersymmetric type IIB torus amplitude
T s < V8 y S 8 < 2 , Ž 4.1 .

and the corresponding Klein-bottle projection
K s 12 Ž V8 y S8 . . Ž 4.2 .
In the transverse channel, the latter becomes
25
Ks
K̃ Ž V8 y S8 . , Ž 4.3 .
2
and requires an open sector with a net number of 32 branes in order to cancel the
resulting RR tadpole. Actually, both V8 and S8 develop tadpoles in this case, that in the
usual type I model are related by supersymmetry, but are conceptually quite different.
While NS–NS tadpoles result in redefinitions of the vacuum configuration, RR tadpoles
signal in general genuine inconsistencies, and their presence is a symptom of the
emergence of serious pathologies w14x.
Actually, one can conceive a more general construction in this case w38x, allowing in
the transverse-channel annulus different reflection coefficients for the V8 and S8
characters, so that
2y5 2 2
As
Ã Ž nqq ny . V8 y Ž nqy ny . S8 , Ž 4.4 .
2
where nq and ny actually count the total numbers of D9 and D9 branes. It should be
appreciated how their relative minus sign in the coefficient of S8 accounts neatly for
their opposite RR charges, while they have clearly identical couplings to the graviton,
consistently with the coefficient of V8 . The direct-channel annulus amplitude
2 2
nq q ny
As Ž V8 y S8 . q nq ny Ž O 8 y C8 . Ž 4.5 .
2
reflects the opposite GSO projections for open strings stretched between two D branes
of the same type Ž99 or 99. and of different types Ž99 or 99. w42–48x. While the former
yields the supersymmetric type I spectrum, the latter eliminates the vector and its
spinorial superpartners, while retaining the tachyon and the spinor of opposite chirality.
As a result, supersymmetry is broken and an instability, signaled by the presence of the
tachyonic ground state, emerges.
¨
The Mobius amplitude
M s " 12 Ž nqq ny . Vˆ8 q 12 Ž nqy ny . Sˆ8 Ž 4.6 .

now involves naturally an undetermined sign for V8 , whose tadpole is generally
incompatible with the one of S8 , and is to be relaxed. Together with A, the two signs
lead to symplectic or orthogonal gauge groups with S fermions in Žanti.symmetric
representations and tachyons and C fermions in bi-fundamentals.
In these ten-dimensional models, the only way to eliminate the tachyon consists in
¨
introducing only D9 branes. Depending on the signs in the Mobius amplitude, one thus
recovers either the SOŽ32. superstring or the USpŽ32. model of w38x. On the other hand,
more can be done if one compactifies the theory on some internal manifold. In this case,
one can introduce Wilson lines Žor, equivalently, separate the branes. in such a way that
in the open strings stretched between separate 9 and 9 branes the tachyon actually
becomes massive. It is instructive to analyze in some detail the simple case of circle
compactification. As in the previous example, the Klein-bottle amplitude is not affected,
and is given by
K s 12 Ž V8 y S8 . P , Ž 4.7 .
where P denotes the sum over momentum states. However, the Wilson line affects the
annulus amplitude, that in the transverse-channel now reads
2y5 n 2 n 2
As
Ã
2
Ž nqq Ž y1. ny . V8 y Ž nqy Ž y1 . ny . S8 W , Ž 4.8 .
where Žy1. n W denotes an oscillating winding sum. As a result, in the direct channel
amplitude
2 2
nq q ny
As Ž V8 y S8 . P q nq ny Ž O 8 y C8 . P Ž1r2. , Ž 4.9 .
2
where P Ž1r2. denotes a sum over 12 -shifted momentum states, both the tachyon and the
¨
C spinor are lifted. The open sector is completed by the Mobius amplitude
M s 12 " Ž nqq ny . Vˆ8 q Ž nqy ny . Sˆ8 P , Ž 4.10 .
and at the massless level comprises gauge bosons in the adjoint of SOŽ nq . = SOŽ ny .
Žor USpŽ nq . = USpŽ ny ., depending on the sign of Vˆ8 in M . and S spinors in
Žanti.symmetric representations.
4.1. Toroidal compactifications with nine and fiÕe (anti)branes
Let us now turn to six-dimensional toroidal compactifications. In this case, one has
the interesting option to introduce in the standard type I model of w60x pairs of D5–D5
and D9–D9 branes. The result includes a chiral spectrum confined to the Žnon-super-
symmetric. branes, and calls for the introduction of six-dimensional Green–Schwarz
couplings to the single tensor present in the projected closed spectrum to cancel the
residual gauge and mixed anomalies.
The starting point in this construction is the standard type IIB string compactified on
a four-torus, for which
T s < V8 y S8 < 2 L4 s < V4 O4 q O4 V4 y C4 C4 y S4 S4 < 2 L4 . Ž 4.11 .

One can now add the standard Klein-bottle projection
K s 12 Ž V8 y S8 . P 4 s 12 Ž V4 O4 q O4 V4 y C4 C4 y S4 S4 . P 4 , Ž 4.12 .
where, for later convenience, we have explicitly decomposed the SOŽ8. characters into
products of SOŽ4. ones. The transverse-channel annulus amplitude
2y5 1
As
Ã
2 ½ Ž V4 O4 q O4 V4 y C4 C4 y S4 S4 . Ž Nq2 q Ny2 . ÕW 4 q Ž Dq2 q Dy2 . Õ P 4
n
q2 Ž V4 O4 q O4 V4 q C4 C4 q S4 S4 . Nq Ny Õ W 3 Ž y . W
1 m
qDq Dy P 3 Ž y . P q 2 Ž V4 O4 y O4 V4 y C4 C4 q S4 S4 .
Õ
2
2h
=
ž /
q2
Ž Nq Dqq Ny Dy . q 2 Ž V4 O4 y O4 V4 q C4 C4 y S4 S4 .
2
2h
=
ž /
q2
Ž Nq Dyq Ny Dq . 5 Ž 4.13 .
is somewhat unconventional, and thus deserves some comments. Actually, the 99, 99,
99, 55, 55 and 55 contributions have already been discussed previously and do not need
further explanations, but one should notice the presence of Wilson lines Žbrane displace-
ments. in the 99 Ž55. sectors. These are to affect different directions, and are needed to
lift the tachyons that would otherwise be present in the open spectrum. The transverse-
channel amplitude A˜, however, includes additional mixed terms, absent in ordinary
toroidal constructions, that are to be interpreted as orbifold-like projections. In fact, the
simultaneous presence of 9 and 5 branes, a familiar feature of orbifold models, halves
the number of supersymmetries and results in the corresponding ŽZ 2 . breaking of the
characters V8 and S8 . Of course, due to the simultaneous presence of branes and
antibranes, the complete theory is not supersymmetric, as implied by the different signs
in the RR sectors. The orbifold-like structure of the annulus amplitude can be better
appreciated after an S transformation to the direct-channel amplitude
A s 12 Ž V4 O4 q O4 V4 y C4 C4 y S4 S4 . Ž Nq2 q Ny2 . P 4 q Ž Dq2 q Dy2 . W 4
q Ž O4 O4 q V4 V4 y S4 C4 y C4 S4 . Nq Ny P 3 P Ž 1r2 . q Dq Dy W 3 W Ž 1r2 .
2
h
q Ž O4 C4 y S4 O4 q V4 S4 y C4 V4 . ž /Ž
q4
Nq Dqq Ny Dy .
2
h
q Ž O4 S4 y C4 O4 q V4 C4 y S4 V4 .ž / Ž Nq Dyq Ny Dq . . Ž 4.14 .
q4
While the mixed ND terms have the structure familiar from Z 2 orbifolds, no breaking
terms are actually present in this case both for N and D charges. This is precisely as
demanded by the consistency conditions for open-string constructions, since only the
toroidal bulk states are allowed to propagate in the transverse channel.
¨
One can now derive the transverse-channel Mobius amplitude combining the terms in
K˜ and A˜ at the origin of the lattices, so that
1
½
M˜0 s 22 'Õ e Vˆ4 Oˆ4 Ž Nqq Ny . 'Õ q Ž Dqq Dy .
'Õ
1
qe Oˆ4 Vˆ4 Ž Nqq Ny . 'Õ y Ž Dqq Dy .
'Õ
1
qCˆ4 Cˆ4 Ž Nqy Ny . 'Õ q Ž Dqy Dy .
'Õ
1
qSˆ4 Sˆ4 Ž Nqy Ny . 'Õ y Ž Dqy Dy .
'Õ 5 , Ž 4.15 .
and extract two NS–NS and two RR tadpole conditions for this class of models:
Ž Dqq Dy .
V4 O4 : Ž Nqq Nyq 32 e . 'Õ q s 0,
'Õ
Ž Dqq Dy .
O4 V4 : Ž Nqq Nyq 32 e . 'Õ y s 0,
'Õ
Ž Dqy Dy .
C4 C4 : Ž Nqy Nyy 32 . 'Õ q s 0,
'Õ
Ž Dqy Dy .
S4 S4 : Ž Nqy Nyy 32 . 'Õ y s 0. Ž 4.16 .
'Õ
The RR tadpole conditions demand that the net charge of the D9rD9 branes, Nqy Ny,
be equal to 32, and that the numbers of D5 and D5 branes be the same. It is then
impossible to satisfy the NS–NS tadpoles, and as a result a potential is generated for the
six-dimensional dilaton, f6 , and for the volume Õ of the internal torus:
Ž Dqq Dy .
Veff ; ey f 6 Ž Nqq Nyq 32 e . 'Õ q . Ž 4.17 .
'Õ
This potential has the peculiar property of stabilizing the internal volume to the value
Dqq Dy
Õ0 s , Ž 4.18 .
Nqq Nyq 32 e
while giving a mass to the corresponding Žbreathing-mode. field. This feature is

common to all models with 9, 9, 5 and 5 branes that we shall encounter in the following
sections.
¨
The direct-channel Mobius amplitude
2
2ĥ
Ms 1
2
½ e Ž Nqq Ny . Vˆ4 Oˆ4 q Oˆ4 Vˆ4 P 4 y e Ž Dqq Dy . Vˆ4 Oˆ4 y Oˆ4 Vˆ4
ž / ž /ž / q̂ 2
2
2ĥ
q Ž Nqy Ny . Cˆ4 Cˆ4 q Sˆ4 Sˆ4 P 4 y Ž Dqy Dy . Cˆ4 Cˆ4 y Sˆ4 Sˆ4
ž / ž /ž / 5
q̂ 2
Ž 4.19 .
follows from Ž4.15. after one includes the massive contributions and performs a P
transformation. The massless open spectrum depends on the sign e , that is not fixed by
tadpole conditions. For e s y1, it comprises gauge bosons in the adjoint of SOŽ Nq . 9 =
SOŽ Ny . 9 = USpŽ Dq .5 = USpŽ Dy .5 , with the matter
4 scalars: Ž A,1;1,1 . , Ž 1, A;1,1 . , Ž 1,1; A,1 . , Ž 1,1;1, A . ,

2 scalars: Ž f ,1; f ,1 . , Ž 1, f ;1, f . , Ž 1, f ; f ,1 . , Ž f ,1;1, f . ,
L-spinor: Ž A,1;1,1 . , Ž 1,S;1,1 . , Ž 1,1;S,1 . , Ž 1,1;1, A . ,
R-spinor: Ž A,1;1,1 . , Ž 1,S;1,1 . , Ž 1,1; A,1 . , Ž 1,1;1,S . ,
half L-spinor: Ž 1, f ; f ,1 . , Ž f ,1;1, f . ,
half R-spinor: Ž f ,1; f ,1 . , Ž 1, f ;1, f . ,
where S Ž A. and f denote the Žanti.symmetric and fundamental representations. On the
other hand, for e s q1 symmetric and antisymmetric representations are interchanged
in the bosonic sector Žthus also interchanging orthogonal and symplectic gauge groups.,
while the massless closed states, still supersymmetric, fill the N s Ž1,1. supergravity
multiplet. It should be appreciated that, to lowest order, after a toroidal compactification
to four dimensions, for e s q1 Ž e s y1. the spectrum on the D9 and D5 ŽD9 and D5.
branes has N s 2 supersymmetry, if suitable gauge subgroups for the D9 and D5 ŽD9
and D5. branes are regarded as global symmetries. Thus, to lowest order the resulting
massless spectrum has N s 4 supersymmetry in the bulk, N s 2 supersymmetry on the
branes Žantibranes., while it is not supersymmetric on the antibranes Žbranes.. The open
sector is chiral and leads to residual mixed and gauge anomalies, summarized by the
polynomial
I8 s 18 Ž tr R 2 y tr FN2q q tr FN2y .Ž tr FD2qy tr FD2y . , Ž 4.20 .
and removed by Green–Schwarz couplings w61x involving the single antisymmetric

tensor present in the massless closed spectrum, while the RR tadpole conditions
guarantee the cancellation of all irreducible tr R 4 and tr F 4 anomalies.
One can further deform these spectra, allowing for a non-trivial quantized NS–NS
Bi j of rank r. Due to the peculiar structure of the models, the closed sector behaves as in
standard toroidal compactifications w60x, whereas the annulus amplitude, modified as in

w62,63x, in the transverse channel reads
2y5
As
Ã
2 ½ Ž V4 O4 q O4 V4 y C4 C4 y S4 S4 .
1
= 2 ry 4 Ž Nq2 q Ny2 . Õ Ý Ý Wn e ip n B
i
i i jw j 2
q Ž Dq 2
q Dy . Õ
Ý Pm i
w j 4 s"1 n i 4 mi4
q2 Ž V4 O4 q O4 V4 q C4 C4 q S4 S4 .
n4
= 2 ry 4 Nq Ny Õ Ý Ý Ž y1. Wn i e ip n i B i j w j
w j 4 s"1 n i 4
1 m3
qDq Dy Ý Ž y1. Pm i
Õ m i4
2
2h
q2 = 2 r r2
Ž V4 O4 y O4 V4 y C4 C4 q S4 S4 . ž /
q2
Ž Nq Dqq Ny Dy .
2
2h
q2 = 2 r r2 Ž V4 O4 y O4 V4 q C4 C4 y S4 S4 .
ž /
q2
Ž Nq Dyq Ny Dq . . 5
Ž 4.21 .
¨
The corresponding transverse-channel Mobius amplitude
½
M˜s 22 2 Ž ry4 . r2 e Ž Nqq Ny . Ž V4 O4 q O4 V4 . Õ
2
Ý Ý Wn e ip n B
w j 4s"1 n i 4
i
i i jw j gw
2h
qe Ž Dqq Dy . Ž V4 O4 y O4 V4 .
ž /
q2
q 2 Ž ry4.r2 Ž Nqy Ny .
= Ž C4 C4 q S4 S4 . Õ Ý Ý Wn e ip n B i
i i jwj gw q e Ž Dqy Dy .
w j 4s"1 n i 4
2
2h
= Ž C4 C4 y S4 S4 .
ž /5
q2
Ž 4.22 .
is thus fixed by factorization and, as usual, involves the signs gw that enforce a proper
normalization. As a result, the net RR charge of the D9rD9 branes is reduced by a
factor 2 r r2 and, after a modular transformation to the direct channel, 2 r r2 copies of the
ND sectors are present, while the gw allow for continuous interpolations between
orthogonal and symplectic gauge groups only on the D9 and D9 branes.
4.2. The T 4 r Z 2 orbifold reÕisited
Let us now reconsider the open descendants of the T 4rZ 2 orbifold, allowing for the
simultaneous presence of branes and antibranes of the same type. For the sake of clarity,
let us begin from the torus amplitude
4 4
2h h
Ts 1
4 ½ < Q o q Q Õ < 2L4 q < Q o y Q Õ < 2
q2
q 16 < Q s q Q c < 2
q4
4
h
q16 < Q s y Q c < 2
q3 5 , Ž 4.23 .
and its Klein-bottle projection

2
h
½
K s 14 Ž Q o q Q Õ . Ž P 4 q W 4 . q 2 = 16 e Ž Q s q Q c . ž /5
q4
, Ž 4.24 .
where we have used the standard Z 2 decompositions

Q o s V4 O4 y C4 C4 , Q Õ s O4 V4 y S4 S4 ,
Q s s O4 C4 y S4 O4 , Q c s V4 S4 y C4 V4 . Ž 4.25 .
We have written K in this form, since one can actually modify the Klein-bottle
projection for twisted closed states, as in w21x. Multiple Klein-bottle projections, first
discussed in w53–56x, have also interesting physical applications to tachyon-free non-su-
persymmetric open-string models w64–69x. The choice e s q1 corresponds to the usual
T 4rZ 2 orientifold, and gives a closed unoriented spectrum containing N s Ž1,0.
supergravity coupled to 20 hypermultiplets and 1 tensor multiplet, while the choice
e s y1 gives N s Ž1,0. supergravity coupled to 4 hypermultiplets and 17 tensor
multiplets.
From the Klein-bottle contribution to the massless tadpoles
2 2
25 e e
K˜0 s
4 ½ž 'Õ q
'Õ / Qo q ž 'Õ y
'Õ / 5 QÕ , Ž 4.26 .
one can anticipate the need for a net number of 32 D9 and 32 Žanti.D5 branes Žfor
e s "1, respectively. in order to cancel the RR tadpoles.
Proceeding as in the previous sections, the annulus amplitude in the presence of
branes and antibranes is
½
A s 14 Ž V4 O4 q O4 V4 y C4 C4 y S4 S4 . Ž Nq2 q Ny2 . P 4 q Ž Dq2 q Dy2 . W 4
q2 Ž O4 O4 q V4 V4 y S4 C4 y C4 S4 . Nq Ny P 3 P1r2 q Dq Dy W 3 W1r2
2
2h
q Ž V4 O4 y O4 V4 y C4 C4 q S4 S4 .
ž /Žq2
R 2N q q R 2Ny q R 2D q q R 2D y .
2
h
q2 Ž O4 C4 q V4 S4 y S4 O4 y C4 V4 . ž /Ž q4
Nq Dqq Ny Dy .
2
h
q2 Ž O4 S4 q V4 C4 y C4 O4 y S4 V4 .ž / Ž Nq Dyq Ny Dq .
q4
2
h
q2 Ž O4 C4 y V4 S4 y S4 O4 q C4 V4 .ž / Ž R N q R D qq R N y R D y .
q3
2
h
q2 Ž yO4 S4 q V4 C4 y C4 O4 q S4 V4 .ž / Ž
q3 5
R N q R D yq R N y R D q . , Ž 4.27 .
where N" Ž D " . refer to the D9 and D9 ŽD5 and D5. branes, respectively, and the R’s
are orbifold-induced breaking terms. As in the previous cases, in order to lift the
tachyons, following w39x we have inserted the D5 and D5 branes at different fixed points
of the orbifolds, and we have added Wilson lines for the D9 branes along a different
coordinate.
¨
Finally, the Mobius amplitude
M s y 14
½ž Vˆ4 Oˆ4 q Oˆ4 Vˆ4 y Cˆ4 Cˆ4 y Sˆ4 Sˆ4 / Nq P 4 q e Dq W 4
q Vˆ4 Oˆ4 q Oˆ4 Vˆ4 q Cˆ4 Cˆ4 q Sˆ4 Sˆ4

ž / Ny P 4 q e Dy W 4
2
2ĥ
y Vˆ4 Oˆ4 y Oˆ4 Vˆ4 y Cˆ4 Cˆ4 q Sˆ4 Sˆ4 / ž q̂
ž / 2
Ž e Nqq Dq .
2
2ĥ
y Vˆ4 Oˆ4 y Oˆ4 Vˆ4 q Cˆ4 Cˆ4 y Sˆ4 Sˆ4
ž /ž / q̂ 2
Ž e Nyq Dy . 5 Ž 4.28 .
gives a proper symmetrization of the annulus amplitude, while the tadpole conditions
Nqy Nys 32, Dqy Dys 32 e ,
R N q s R N y s R D q s R D ys 0 Ž 4.29 .
require net numbers of 32 D9 branes and 32 D5-branes Žantibranes. for e s "1.

The choice e s q1 corresponds to a deformation of the supersymmetric UŽ16. =
UŽ16. model w3,4,70x, and requires the introduction of complex charges, so that
N"s n "q n " , R N "s i Ž n "y n " . ,
D "s d "q d " , R D "s i Ž d "y d " . . Ž 4.30 .

The resulting massless open spectrum can be extracted from

A0 q M0 s Ž nq nqq ny nyq dq dqq dy dy . Ž V4 O4 y C4 C4 .
q 12 nq
2
ž 2
q nq 2
q ny 2
q ny 2
q dq 2
q dq 2
q dy 2
q dy Ž O4 V4 y S4 S4 . /
q nq dqq nq dqq ny dyq ny dy Ž O4 C4 y S4 O4 .
ž /
yž n q dyq ny dqq nq dyq ny dq /C O q4 4
q nq dyq nq dyq ny dqq ny dq O4 S4

ž /
y 12 ½ Ž n q n q d q d . ž Oˆ Vˆ y Sˆ Sˆ /
q q q q 4 4 4 4
q Ž n q n q d q d . ž Oˆ Vˆ q Sˆ Sˆ / 5 .
y y y y 4 4 4 4 Ž 4.31 .
The 99 and 55 sectors are supersymmetric, and comprise vector multiplets in the adjoint
of UŽ nq . 9 = UŽ ny . 9 = UŽ dq .5 = UŽ dy .5 , together with hypermultiplets in the antisym-
metric and conjugate antisymmetric representations. On the other hand, the 99 and 55
are not supersymmetric and contain left-handed Weyl fermions in the antisymmetric and
conjugate antisymmetric representations, and a quartet of real scalars in the symmetric
and conjugate symmetric representations. Finally, the 95 and 95 sectors give hypermulti-
plets in the Ž nq, dq . and Ž ny, dy ., respectively, while the 95 and 95 sectors, evidently
not supersymmetric, give a quartet of scalars in the Ž nq, dy ., Ž ny, dq . and right-handed
Weyl fermions in the Ž nq, dy ., Ž ny, dq .. The irreducible gravitational and gauge
anomalies vanish as a result of tadpole conditions, whereas the residual anomaly
polynomial
I8 s y 161 tr R 2 y 2 Ž tr FN2q y tr FN2y . tr R 2 y 2 Ž tr FD2q y tr FD2y . Ž 4.32 .
requires a conventional Green–Schwarz mechanism w61x involving the single antisym-
metric tensor present in the massless closed spectrum.
The choice e s y1 corresponds to a deformation of the original model in w21x, and
calls for the parametrization
N"s n "q m " , R N "s n "y m " ,
D "s p "q q " , R D "s p "y q " , Ž 4.33 .
in terms of real charges. The massless open spectrum can now be read from
A0 q M0 s 12 Ž nq
2 2
q ny 2
q mq 2
q my 2
q pq 2
q py 2
q qq 2
q qy . Ž V4 O4 y C4 C4 .
q Ž mq nqq my nyq pq qqq py qy . Ž O4 V4 y S4 S4 .
q Ž mq qqq nq pqq my qyq ny py . Ž O4 C4 y S4 O4 .
q Ž mq pyq nq qyq my pqq ny qq . O4 S4
y Ž mq qyq nq pyq my qqq ny pq . C4 O4
y 12 Ž nqq mqy pqy qq . Vˆ4 Oˆ4 y Cˆ4 Cˆ4

½ ž /
q Ž nyq myy pyy qy . Vˆ4 Oˆ4 q Cˆ4 Cˆ4ž /5, Ž 4.34 .
and the massless excitations thus comprise gauge bosons in the adjoint of SOŽ mq . 9 =
SOŽ nq . 9 = SOŽ my . 9 = SOŽ ny . 9 = USpŽ pq .5 = USpŽ qq .5 = USpŽ py .5 = USpŽ qy .5 ,
left-handed Weyl fermions in the adjoint of SOŽ mq ., SOŽ nq ., USpŽ pq . and USpŽ qq .,
in the symmetric representations of SOŽ my . and SOŽ ny . and in the antisymmetric
representations of USpŽ py . and USpŽ qy .. They also comprise a full hypermultiplet in
the Ž mq, nq ., Ž my, ny ., Ž pq, qq ., Ž py, qy ., as well as half hypermultiplets in the
Ž mq, qq ., Ž nq, pq ., Ž my, qy . and Ž ny, py .. Finally, the 95 and 59 sectors comprise
complex scalars in the Ž mq, py ., Ž nq, qy ., Ž my, pq . and Ž ny, qq ., and left-handed
symplectic Majorana–Weyl spinors in the Ž mq, qy ., Ž nq, py ., Ž my, qq . and Ž ny, pq ..
The tadpole conditions eliminate all irreducible gauge and gravitational anomalies,
whereas the residual reducible anomaly
I8 s 641 2tr R 2 y Ž tr Fm2q q tr Fn2q y tr Fm2y y tr Fn2y

2
ytr Fp2q y tr Fq2q q tr Fp2y q tr Fq2y .
2
y 641 Ž tr Fm2q q tr Fn2q y tr Fm2y y tr Fn2y q tr Fp2q q tr Fq2q y tr Fp2y y tr Fq2y .
2
q 641 Ž tr Fm2q y tr Fn2q y tr Fm2y q tr Fn2y y tr Fp2q q tr Fq2q q tr Fp2y y tr Fq2y .
2
q 644 Ž tr Fm2q y tr Fn2q q tr Fm2y y tr Fn2y .
2
q 644 Ž tr Fp2q y tr Fq2q q tr Fp2y y tr Fq2y .
2
q 643 Ž tr Fm2q y tr Fn2q y tr Fm2y q tr Fn2y q tr Fp2q y tr Fq2q y tr Fp2y q tr Fq2y .
Ž 4.35 .
can be removed by a generalized Green–Schwarz mechanism w71x.
In both models a potential is generated for the six-dimensional dilaton and the
volume of the internal manifold
Ž Dqq Dyy 32 e .
Veff ; ey f 6 Ž Nqq Nyy 32 . 'Õ q , Ž 4.36 .
'Õ
as a result of uncancelled NS–NS tadpoles. As in the previous model, this stabilizes the
internal volume at a local minimum, that in this case is
Dqq Dyy 32 e
Õ0 s , Ž 4.37 .
Nqq Nyy 32
and gives mass to the corresponding Žbreathing-mode. field.

In the present models discrete deformations for the NS–NS B-field may be intro-
duced as in w60,62,63x, and result in a reduced total RR charge for nine and five
Žanti.branes and in multiplicities for the ND sectors. Moreover, the corresponding signs
¨
in the Mobius amplitude allow continuous interpolations between orthogonal Žsym-
plectic. and unitary gauge groups.
4.3. The non-supersymmetric T 6 r Z 4 orbifold reÕisited
As a last example, we would like to generalize the non-supersymmetric T 6rZ 4

orbifold of Section 3, allowing for the simultaneous presence of branes and antibranes.
As in the previous examples, the torus and Klein-bottle amplitudes are not affected,
¨
whereas the annulus and Mobius amplitudes are now given by
A s 18
½ Ž Nq2 q Ny2 . P1 P2 P3 q Ž Dq2 q Dy2 . W1W2 P3 c 00
2h
q Ž R 2N q q R 2N yq R 2D q q R 2D y . c 01 f 01
2
ž / q2
2
2h
q Ž SN2 q q SN2 yq SD2 q q SD2 y . c 02
ž / q2
P3
2h
q Ž TN2q q TN2yq TD2q q TD2y . c 03 f 03
2
ž / q2
q2 Ž Nq Ny P1 P21r2 P3 q Dq Dy W1W21r2 P3 . c 00
Žy.
2
h
q2 Ž Nq Dqq Ny Dy . c 20 ž /q4
P3
2h
2
q2 Ž R N q R D qq R N y R D y . c 21 f 21
ž / q2
2
h
q2 Ž SN q SD qq R N y R D y . c 22 ž /q3
P3
2h
2
q2 Ž TN q TD qq R N y R D y . c 23 f 23
ž / q2
2
h
q2 Ž Nq Dyq Ny Dq c 20
Žy.
. ž / q4
P3
2h
Žy. 2
q2 Ž R N q R D yq R N y R D q . c 21 f 21
ž / q2
2
h
q2 Ž SN q SD yq R N y R D q . c 22
Žy.
ž /
q3
P3
2h
Žy. 2
q2 Ž TN q TD yq R N y R D q . c 23 f 23
ž /5 q2
Ž 4.38 .
and
2
2ĥ
½
M s y 18 Ž Nq P1 P2 P3 y Dq W1W2 P3 . cˆ 00 q Ž Nqy Dq . P3 cˆ 02
ž / q̂ 2
2
2ĥ
q Ž Ny P1 P2 P3 y Dy W1W2 P3 . cˆ˜ 00 q Ž Nyy Dy . P3 cˆ˜ 02
Žy. Žy.
ž / q̂ 2
2hˆ 2hˆ
q2 Ž SN qy SD q . cˆ 01 fˆ 01 q 2 Ž SN qy SD q . cˆ 03 fˆ 03
2
ž /
qˆ2
2
ž /
qˆ2
2hˆ 2hˆ
q2 Ž SN yy SD y . c˜ˆ 01
Žy. ˆ 2
f 01
ž /
qˆ2
q 2 Ž SN yy SD y . c˜ˆ 03
Žy. ˆ 2
f 03
ž /5
qˆ2
, Ž 4.39 .
where the c ’s and f ’s have been introduced in Section 3, P 1r2 and W 1r2 denote
shifted momentum and winding sums, and we are parametrizing again the 9, 9, 5 and 5
brane charges by N", D " and by their orbifold-induced breakings R, S and T. The Z 4
action suggests the parametrization
N"s n "q m "q p "q m " , D "s q "q r "q s "q r " ,
R N "s n "q im "y p "y im " , R D "s q "q ir "y s "y ir " ,
SN "s n "y m "q p "y m " , SD "s q "y r "q s "y r " ,
TN "s n "y im "y p "q im " , TD "s q "y ir "y s "q ir " Ž 4.40 .
and, as expected from the simpler case discussed in Section 3, the RR tadpole conditions
fix to 32 the net numbers of D9 branes and D5 branes, and require the vanishing of all
the R, S and T breaking coefficients. Even in this case the NS–NS tadpole conditions
can not be satisfied, and as a result the potential
Ž Dqq Dyq 32 .
Veff ; ey f 4 Õ 3 Ž Nqq Nyy 32 . Õ 1Õ 2 q
( ( Ž 4.41 .
(Õ Õ
1 2
is generated for the four-dimensional dilaton and the volume of the internal manifold. As
in the previous cases, it constrains the internal volumes, since it has a minimum for
Dqq Dyq 32
Ž Õ 1Õ 2 . 0 s . Ž 4.42 .
Nqq Nyy 32
From the tadpole conditions and the massless contributions to A and M , one finds
the Chan–Paton gauge group SOŽ nq . 9 = SOŽ pq . 9 = UŽ mq . 9 = SOŽ ny . 9 = SOŽ py . 9
= UŽ my . 9 = USpŽ qq .5 = USpŽ sq .5 = UŽ rq .5 = USpŽ qy .5 = USpŽ sy .5 = UŽ ry .5 .
The 99 and 55 sectors are supersymmetric and, aside from N s 1 vector multiplets in
the adjoint of the gauge group, comprise chiral multiplets in the representations 5
99: 2 Ž nq ,mq . , 2 Ž pq ,mq . , Ž nq , pq . , A m q, A m q,
55: 2 Ž qq ,rq . , 2 Ž sq ,rq . , Ž qq , sq . , Srq, Srq.
The 99 and 55 sectors are non-supersymmetric and, aside from the gauge bosons in the
adjoint of the corresponding gauge groups, comprise Dirac spinors in the representations
99: Sny, S py, Adj m y, Sm y, Sm y,
55: A qy, A sy, Adj ry, A ry, A ry,
scalars in the representations Sm y and Sm y Ž A ry and A ry ., pairs of chiral multiplets in

the representations Ž ny,my . and Ž py,my . ŽŽ qy,ry . and Ž sy,ry .., and a single chiral
multiplet in the representation Ž ny, py . ŽŽ qy, sy .. for the 99 Ž55. sectors. The 95
spectrum contains scalars in the representations Ž nq,qy ., Ž pq, sy ., Ž mq,ry . and
Ž mq,ry ., and spinors in the representations Ž nq,ry ., Ž pq,ry ., Ž mq, sy . and Ž mq,qy .,
while the 59 sector contains chiral multiplets in the representations Ž nq,rq ., Ž pq,rq .,
Ž mq, sq . and Ž mq,qq .. Finally, the 59 and 59 sectors give similar contributions, with
suitable relabelings of the Chan–Paton charges. This chiral spectrum is free of non-
abelian gauge anomalies.
5. Conclusions
In this paper we have discussed type I models where the basic two-dimensional
consistency conditions ŽRR tadpoles., although apparently unsolvable, can actually be
solved provided supersymmetry is broken at the string scale on some Žanti.branes,
while some amount of supersymmetry is left, to lowest order, on other branes, and most
notably in the bulk Žgravitational. sector.
This new feature of our mechanism Žcalled in w21x brane supersymmetry breaking.,
where the breaking is tied to the string scale, is the main difference compared to other
previously known mechanisms in heterotic strings w22–28x, in type I strings w29–36,72–
74x or in M-theory w75–79x. Brane supersymmetry breaking is typically accompanied by
uncanceled NS–NS tadpoles for the dilaton at some moduli fields of the compact
manifold.
5
For the sake of brevity, we denote by Sa Ž Aa . the Žanti.symmetric representations for the a th factor in
the gauge group, by Sa Ž Aa . their conjugates, and by Adj the adjoint of unitary factors.
Since our models contain non-BPS brane–antibrane systems, one might wonder
whether they correspond to stable configurations. This is indeed the case, and actually
the ‘‘minimal’’ D9–D5 pairs of w21x, aside from having no open-string tachyons,
experience no net mutual forces, as can be seen from the vanishing of the corresponding
annulus amplitude. Following Refs. w38,39x, these models can be further deformed by
the inclusion of additional brane–antibrane pairs of the same type ŽD9–D9, D5i –D5i ..
We would like to stress that this additional deformation is an interesting option, not
required by any RR tadpole conditions. Rather, it destroys local tadpole cancellations,
while the new brane–antibrane pairs do experience mutual forces, although the tachyons
resulting from strings stretched between them may be lifted if the endpoints are suitably
separated in the internal space. Interestingly, the NS–NS tadpoles resulting from the
various mutual interactions between Žanti.branes and orientifold planes determine a
scalar potential that can actually stabilize some of the radii of the compact space, giving
masses to the corresponding fields. Large Žsmall. volumes, however, require unnaturally
large numbers of D5–D5 ŽD9–D9. pairs, whose presence, therefore, asks for a dynami-
cal reason. Since the whole approach is perturbative, the dilaton potential has, not
surprisingly, a runaway behavior towards vanishing string coupling. We would like to
stress that all these constructions, despite their attractive features, do not solve the
problem of the Žfour-dimensional. cosmological constant.
Breaking supersymmetry at the string scale is a viable phenomenological alternative
if our world is a non-supersymmetric Žanti.brane and the string scale MI is in the TeV
range, or if our world is on a brane that, to lowest order, is supersymmetric. In the latter
case, supersymmetry breaking could be mediated by gauge interactions, and a realistic
spectrum would ask for a string scale of the order of 100 TeV if the non-supersymmetric
Žanti.brane gauge coupling were of order one, or at an intermediate value if the
Žanti.brane gauge coupling were suppressed by the volume of the internal space.
Alternatively, if this coupling were very tiny, gravitation would mediate supersymmetry
breaking with an intermediate string scale of the order of 10 11 GeV w12,13x. The
Žtree-level. supersymmetry present in the bulk sector can have far-reaching physical
consequences. Indeed, quantum corrections to branerantibrane couplings can also be
interpreted in terms of effective brane couplings to bulk fields. If these respect the bulk
symmetry, one could contemplate the fascinating possibility of living in a non-super-
symmetric world where quantum corrections are governed by a supersymmetric bulk
sector. In this case, the gauge hierarchy would be protected and the quantum corrections
would be very similar to those of a supersymmetric brane theory. We will return to these
interesting issues in the near future.
Acknowledgements
C.A. and A.S. would like to thank the Physics Department of the Humboldt
University, C.A. and E.D. would like thank the Physics Department of the University of
´
Rome ‘‘Tor Vergata’’, and A.S. would like to thank the Centre de Physique Theorique
´
of the Ecole Polytechnique for the warm hospitality extended to them while this work
was in progress. This research was supported in part by the EEC under TMR contract
ERBFMRX-CT96-0090, and in part by the National Science Foundation, under Grant
No. PHY94-07194.
Appendix A. Characters for the T 6 r Z 2 = Z 2 orbifolds
In this appendix we list the Z 2 = Z 2 characters needed for the models in Section 2.
They may be expressed as ordered products of the four SOŽ2. level-one characters, O 2 ,
V2 , S2 and C2 :
t o o s V 2 O 2 O 2 O 2 q O 2 V 2 V2 V 2 y S 2 S 2 S 2 S 2 y C 2 C 2 C 2 C 2 ,
t o g s O 2 V2 O 2 O 2 q V 2 O 2 V2 V2 y C 2 C 2 S 2 S 2 y S 2 S 2 C 2 C 2 ,
t o h s O 2 O 2 O 2 V2 q V2 V 2 V2 O 2 y C 2 S 2 S 2 C 2 y S 2 C 2 C 2 S 2 ,
t o f s O 2 O 2 V 2 O 2 q V 2 V2 O 2 V 2 y C 2 S 2 C 2 S 2 y S 2 C 2 S 2 C 2 ,
t g o s V2 O 2 S 2 C 2 q O 2 V2 C 2 S 2 y S 2 S 2 V2 O 2 y C 2 C 2 O 2 V2 ,
t g g s O 2 V2 S 2 C 2 q V 2 O 2 C 2 S 2 y S 2 S 2 O 2 V 2 y C 2 C 2 V 2 O 2 ,
t g h s O 2 O 2 S 2 S 2 q V2 V2 C 2 C 2 y C 2 S 2 V2 V2 y S 2 C 2 O 2 O 2 ,
t g f s O 2 O 2 C 2 C 2 q V2 V 2 S 2 S 2 y S 2 C 2 V 2 V2 y C 2 S 2 O 2 O 2 ,
t h o s V2 S 2 C 2 O 2 q O 2 C 2 S 2 V2 y C 2 O 2 V2 C 2 y S 2 V2 O 2 S 2 ,
t h g s O 2 C 2 C 2 O 2 q V 2 S 2 S 2 V2 y C 2 O 2 O 2 S 2 y S 2 V2 V2 C 2 ,
t h h s O 2 S 2 C 2 V2 q V 2 C 2 S 2 O 2 y S 2 O 2 V2 S 2 y C 2 V2 O 2 C 2 ,
t h f s O 2 S 2 S 2 O 2 q V2 C 2 C 2 V2 y C 2 V2 V2 S 2 y S 2 O 2 O 2 C 2 ,
t o f s V2 S 2 O 2 C 2 q O 2 C 2 V2 S 2 y S 2 V2 S 2 O 2 y C 2 O 2 C 2 V2 ,
t f g s O 2 C 2 O 2 C 2 q V 2 S 2 V 2 S 2 y C 2 O 2 S 2 O 2 y S 2 V 2 C 2 V2 ,
t f h s O 2 S 2 O 2 S 2 q V2 C 2 V2 C 2 y C 2 V2 S 2 V2 y S 2 O 2 C 2 O 2 ,
t f f s O 2 S 2 V 2 C 2 q V2 C 2 O 2 S 2 y C 2 V2 C 2 O 2 y S 2 O 2 S 2 V2 . Ž A.1 .
While these are sufficient to describe all supersymmetric Z 2 = Z 2 amplitudes, when
brane supersymmetry breaking is present additional characters are needed to describe the
open strings stretched between branes and antibranes. These new characters, that we
denote by t kŽy.
l , are obtained from the others in Eq. A.1 interchanging O 2 with V2 and
Ž .
S2 with C2 in the last three factors, that correspond to the three internal tori:
toŽy.
o s O 2 O 2 O 2 O 2 q V2 V2 V 2 V 2 y C 2 S 2 S 2 S 2 y S 2 C 2 C 2 C 2 ,
toŽy.
g s V2 V2 O 2 O 2 q O 2 O 2 V2 V2 y S 2 C 2 S 2 S 2 y C 2 S 2 C 2 C 2 ,
toŽy.
h s V 2 O 2 O 2 V2 q O 2 V2 V2 O 2 y S 2 S 2 S 2 C 2 y C 2 C 2 C 2 S 2 ,
toŽy.
f s V 2 O 2 V2 O 2 q O 2 V 2 O 2 V 2 y S 2 S 2 C 2 S 2 y C 2 C 2 S 2 C 2 ,
tgŽy.
o s O 2 O 2 S 2 C 2 q V2 V2 C 2 S 2 y C 2 S 2 V2 O 2 y S 2 C 2 O 2 V2 ,
tgŽy.
g s V2 V2 S 2 C 2 q O 2 O 2 C 2 S 2 y C 2 S 2 O 2 V2 y S 2 C 2 V2 O 2 ,
tgŽy.
h s V 2 O 2 S 2 S 2 q O 2 V2 C 2 C 2 y S 2 S 2 V2 V2 y C 2 C 2 O 2 O 2 ,
tgŽy.
f s V2 O 2 C 2 C 2 q O 2 V2 S 2 S 2 y C 2 C 2 V2 V2 y S 2 S 2 O 2 O 2 ,
t hŽy.
o s O 2 S 2 C 2 O 2 q V 2 C 2 S 2 V2 y S 2 O 2 V2 C 2 y C 2 V2 O 2 S 2 ,
t hŽy.
g s V2 C 2 C 2 O 2 q O 2 S 2 S 2 V2 y S 2 O 2 O 2 S 2 y C 2 V2 V2 C 2 ,
t hŽy.
h s V 2 S 2 C 2 V2 q O 2 C 2 S 2 O 2 y C 2 O 2 V2 S 2 y S 2 V2 O 2 C 2 ,
t hŽy.
f s V2 S 2 S 2 O 2 q O 2 C 2 C 2 V2 y S 2 V2 V2 S 2 y C 2 O 2 O 2 C 2 ,
toŽy.
f s O 2 S 2 O 2 C 2 q V2 C 2 V2 S 2 y C 2 V2 S 2 O 2 y S 2 O 2 C 2 V2 ,
t fŽy.
g s V2 C 2 O 2 C 2 q O 2 S 2 V2 S 2 y S 2 O 2 S 2 O 2 y C 2 V2 C 2 V2 ,
t fŽy.
h s V2 S 2 O 2 S 2 q O 2 C 2 V2 C 2 y S 2 V2 S 2 V2 y C 2 O 2 C 2 O 2 ,
t fŽy.
f s V 2 S 2 V 2 C 2 q O 2 C 2 O 2 S 2 y S 2 V2 C 2 O 2 y C 2 O 2 S 2 V2 . Ž A.2 .
Appendix B. Characters for the T 6 r Z 4 orbifold
In this appendix we list the characters needed for the Z 4 orientifold of Section 3:
r 00 s V2 O 2 x 0 j 0 q O 2 V2 x 0 j 4 y S2 C2 x 0 j 2 y C2 S2 x 0 jy2 ,
r 01 s V2 V2 x 1r2 jy2 q O 2 O 2 x 1r2 j 2 y S2 S2 x 1r2 j 0 y C2 C2 x 1r2 j 4 ,
r 02 s V2 O 2 x 0 j 4 q O 2 V2 x 0 j 0 y S2 C2 x 0 jy2 y C2 S2 x 0 j 2 ,
r 03 s V2 V2 x 1r2 j 2 q O 2 O 2 x 1r2 jy2 y S2 S2 x 1r2 j 4 y C2 C2 x 1r2 j 0 ,
r 10 s V2 S2 x 0 jy3 q O 2 C2 x 0 j 1 y S2 O 2 x 0 jy1 y C2 V2 x 0 j 3 ,
r 11 s V2 C2 x 1r2 j 3 q O 2 S2 x 1r2 jy1 y S2 V2 x 1r2 jy3 y C2 O 2 x 1r2 j 1 ,
r 12 s V2 S2 x 0 j 1 q O 2 C2 x 0 jy3 y S2 O 2 x 0 j 3 y C2 V2 x 0 jy1 ,
r 13 s V2 C2 x 1r2 jy1 q O 2 S2 x 1r2 j 3 y S2 V2 x 1r2 j 1 y C2 O 2 x 1r2 jy3 ,
r 20 s V2 O 2 x 1r2 j 0 q O 2 V2 x 1r2 j 4 y S2 C2 x 1r2 j 2 y C2 S2 x 1r2 jy2 ,
r 21 s V2 V2 x 0 jy2 q O 2 O 2 x 0 j 2 y S2 S2 x 0 j 0 y C2 C2 x 0 j 4 ,
r 22 s V2 O 2 x 1r2 j4 q O 2 V2 x 1r2 j 0 y S2 C2 x 1r2 jy2 y C2 S2 x 1r2 j 2 ,
r 23 s V2 V2 x 0 j 2 q O 2 O 2 x 0 jy2 y S2 S2 x 0 j 4 y C2 C2 x 0 j 0 ,
r 30 s V2 C2 x 0 j 3 q O 2 S2 x 0 jy1 y S2 V2 x 0 jy3 y C2 O 2 x 0 j 1 ,
r 31 s V2 S2 x 1r2 j 1 q O 2 C2 x 1r2 jy3 y S2 O 2 x 1r2 j 3 y C2 V2 x 1r2 jy1 ,
r 32 s V2 C2 x 0 jy1 q O 2 S2 x 0 j 3 y S2 V2 x 0 j 1 y C2 O 2 x 0 jy3 ,
r 33 s V2 S2 x 1r2 jy3 q O 2 C2 x 1r2 j 1 y S2 O 2 x 1r2 jy1 y C2 V2 x 1r2 j 3 . Ž B.1 .
The s characters are obtained interchanging in the internal part O 2 with V2 , j 0 with
j 4 , S2 with C2 , and j 2 with jy2 :
s 00 s V2 V2 x 0 j 4 q O 2 O 2 x 0 j 0 y S2 S2 x 0 jy2 y C2 C2 x 0 j 2 ,
s 01 s V2 O 2 x 1r2 j 2 q O 2 V2 x 1r2 jy2 y S2 C2 x 1r2 j 4 y C2 S2 x 1r2 j 0 ,
s 02 s V2 V2 x 0 j 0 q O 2 O 2 x 0 j 4 y S2 S2 x 0 j 2 y C2 C2 x 0 jy2 ,
s 03 s V2 O 2 x 1r2 jy2 q O 2 V2 x 1r2 j 2 y S2 C2 x 1r2 j 0 y C2 S2 x 1r2 j 4 ,
s 20 s O 2 O 2 x 1r2 j 0 q V2 V2 x 1r2 j 4 y S2 S2 x 1r2 jy2 y C2 C2 x 1r2 j 2 ,
s 21 s V2 O 2 x 0 jy2 q O 2 V2 x 0 j 2 y S2 C2 x 0 j 0 y C2 S2 x 0 j 4 ,
s 22 s V2 V2 x 1r2 j 0 q O 2 O 2 x 1r2 j 4 y S2 S2 x 1r2 j 2 y C2 C2 x 1r2 jy2 ,
s 23 s V2 O 2 x 0 j 2 q O 2 V2 x 0 jy2 y S2 C2 x 0 j 4 y C2 S2 x 0 j 0 . Ž B.2 .
References
w1x A. Sagnotti, in: Cargese ’87, Non-Perturbative Quantum Field Theory, ed. G. Mack et al. ŽPergamon,
Oxford, 1988. p. 521.
w2x G. Pradisi, A. Sagnotti, Phys. Lett. B 216 Ž1989. 59.
w3x M. Bianchi, A. Sagnotti, Phys. Lett. B 247 Ž1990. 517.
w4x M. Bianchi, A. Sagnotti, Nucl. Phys. B 361 Ž1991. 519.
w5x I. Antoniadis, Phys. Lett. B 246 Ž1990. 377.
w6x E. Witten, Nucl. Phys. B 471 Ž1996. 135.
w7x J.D. Lykken, Phys. Rev. D 54 Ž1996. 3693.
w8x N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 429 Ž1998. 263.
w9x K.R. Dienes, E. Dudas, T. Gherghetta, Phys. Lett. B 436 Ž1998. 55.
w10x I. Antoniadis, N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 436 Ž1998. 263.
w11x For a recent review see I. Antoniadis, hep-thr9909212, and references therein.
w12x K. Benakli, Phys. Rev. D 60 Ž1999. 104002.
w13x C. Burgess, L.E. Ibanez,
´˜ F. Quevedo, Phys. Lett. B 447 Ž1999. 257.
w14x J. Polchinski, Y.C. Cai, Nucl. Phys. B 296 Ž1988. 91.
w15x J. Polchinski, Phys. Rev. Lett. 75 Ž1995. 4724.
w16x M. Bianchi, Ph.D. Thesis, preprint ROM2F-92r13.
w17x A. Sagnotti, hep-thr9302099.
w18x G. Zwart, Nucl. Phys. B 526 Ž1998. 378.
w19x Z. Kakushadze, G. Shiu, S.H.H. Tye, Nucl. Phys. B 533 Ž1998. 25.
w20x G. Aldazabal, A. Font, L.E. Ibanez,
´˜ G. Violero, Nucl. Phys. B 536 Ž1998. 29.
w21x I. Antoniadis, E. Dudas, A. Sagnotti, Phys. Lett. B 464 Ž1999. 38.
w22x R. Rohm, Nucl. Phys. B 237 Ž1984. 553.
w23x C. Kounnas, M. Porrati, Nucl. Phys. B 310 Ž1988. 355.
w24x I. Antoniadis, C. Bachas, D.C. Lewellen, T.N. Tomaras, Phys. Lett. B 207 Ž1988. 441.
w25x S. Ferrara, C. Kounnas, M. Porrati, F. Zwirner, Nucl. Phys. B 318 Ž1989. 75.
w26x C. Kounnas, B. Rostand, Nucl. Phys. B 341 Ž1990. 641.
w27x I. Antoniadis, Phys. Lett. B 246 Ž1990. 377.
w28x I. Antoniadis, C. Kounnas, Phys. Lett. B 261 Ž1991. 369.
w29x I. Antoniadis, E. Dudas, A. Sagnotti, Nucl. Phys. B 544 Ž1999. 469.
w30x I. Antoniadis, G. D’Appollonio, E. Dudas, A. Sagnotti, Nucl. Phys. B 553 Ž1999. 133.
w31x R. Blumenhagen, L. Gorlich,
¨ Nucl. Phys. B 551 Ž1999. 601.
w32x ¨
C. Angelantonj, I. Antoniadis, K. Forger, Nucl. Phys. B 555 Ž1999. 116.
w33x I. Antoniadis, G. D’Appollonio, E. Dudas, A. Sagnotti, hep-thr9907184, to appear in Nucl. Phys. B.
w34x A.L. Cotrone, hep-thr9909116.
w35x C. Bachas, hep-thr9503030.
w36x M. Bianchi, Ya.S. Stanev, Nucl. Phys. B 253 Ž1998. 193.
w37x M. Bianchi, J.F. Morales, G. Pradisi, hep-thr9910228.
w38x S. Sugimoto, hep-thr9905159.
w39x G. Aldazabal, A.M. Uranga, hep-thr9908072.
w40x C. Angelantonj, M. Bianchi, G. Pradisi, A. Sagnotti, Ya.S. Stanev, Phys. Lett. B 385 Ž1996. 96.
w41x ´˜ F. Quevedo, hep-thr9909172.
G. Aldazabal, L.E. Ibanez,
w42x A. Sen, J. High Energy Phys. 9806 Ž1998. 007.
w47x For recent reviews, see A. Sen, hep-thr9904207.
w48x A. Lerda, R. Russo, hep-thr9905006.
w49x C. Vafa, Nucl. Phys. B 273 Ž1986. 592.
w50x C. Vafa, E. Witten, J. Geom. Phys. 15 Ž1995. 189.
w51x J. Blum, Nucl. Phys. B 486 Ž1997. 34.
w52x M.R. Douglas, hep-thr9807235.
w53x D. Fioravanti, G. Pradisi, A. Sagnotti, Phys. Lett. B 321 Ž1994. 349.
w54x G. Pradisi, A. Sagnotti, Ya.S. Stanev, Phys. Lett. B 354 Ž1995. 279.
w57x M. Berkooz, R.G. Leigh, Nucl. Phys. B 483 Ž1997. 187.
w58x K. Mohri, Nucl. Phys. B 521 Ž1998. 161.
w59x S. Mukhopadhyay, K. Ray, hep-thr9909107.
w60x M. Bianchi, G. Pradisi, A. Sagnotti, Nucl. Phys. B 376 Ž1992. 365.
w61x M.B. Green, J.H. Schwarz, Phys. Lett. B 149 Ž1984. 117.
w62x Z. Kakushadze, G. Shiu, S.H.H. Tye, Phys. Rev. D 58 Ž1998. 086001.
w63x C. Angelantonj, hep-thr9908064, to appear in Nucl. Phys. B.
w64x A. Sagnotti, hep-thr9509080,
w65x A. Sagnotti, hep-thr9702093.
w66x C. Angelantonj, Phys. Lett. B 444 Ž1998. 309.
w67x ¨ hep-thr9904069.
R. Blumenhagen, A. Font, D. Lust,
w68x R. Blumenhagen, A. Kumar, hep-thr9906234.
w69x ¨
K. Forger, hep-thr9909010.
w70x E. Gimon, J. Polchinski, hep-thr9601038.
w71x A. Sagnotti, Phys. Lett. B 294 Ž1992. 196.
w72x J. Blum, K.R. Dienes, Phys. Lett. B 414 Ž1997. 260.
w73x J. Blum, K.R. Dienes, Nucl. Phys. B 516 Ž1998. 83.
w74x R. Blumenhagen, C. Kounnas, D. Lust,¨ hep-thr9910094.
w75x I. Antoniadis, M. Quiros, Phys. Lett. B 392 Ž1997. 61.
w76x E. Dudas, C. Grojean, Nucl. Phys. B 507 Ž1997. 553. hep-thr9704177.
w77x I. Antoniadis, M. Quiros, Nucl. Phys. B 505 Ž1997. 109. hep-thr9705037.
w78x I. Antoniadis, M. Quiros, Phys. Lett. B 416 Ž1998. 327.
w79x E. Dudas, Phys. Lett. B 416 Ž1998. 309.
Mordell–Weil lattice via string junctions

Mitsuaki Fukae a , Yasuhiko Yamada a , Sung-Kil Yang b
a
Department of Mathematics, Kobe UniÕersity, Rokko, Kobe 657-8501, Japan
b
Institute of Physics, UniÕersity of Tsukuba, Ibaraki 305-8571, Japan
Received 27 September 1999; received in revised form 23 December 1999; accepted 5 January 2000
Abstract
We analyze the structure of singularities, Mordell–Weil lattices and torsions of a rational

elliptic surface using string junctions in the background of 12 7-branes. The classification of the
Mordell–Weil lattices due to Oguiso–Shioda is reproduced in terms of the junction lattice. In this
analysis an important role played by the global structure of the surface is observed. It is then
found that the torsions in the Mordell–Weil group are generated by the fraction of loop junctions
which represent the imaginary roots of the loop algebra Eˆ9 . From the structure of the Mordell–Weil
lattice we find 7-brane configurations which support non-BPS junctions carrying conserved
Abelian charges. q 2000 Elsevier Science B.V. All rights reserved.
PACS: 11.25.yw; 02.10.Rn

Keywords: IIB 7-branes; String junctions; Rational elliptic surface; Mordell–Weil lattice
1. Introduction
In our previous paper w1x, and in a related work w2x, elliptic curves have been
constructed for the 7-brane configurations on which the affine Lie algebras Eˆn Ž1 F n F 8.
and Eñ Ž n s 0,1. are realized. Upon deriving these curves from a rational elliptic surface
S, we recognize that the brane picture is very efficient to deal with the geometry of S. In
this construction, however, we have only probed the local geometry of S with the aid of
the 7-brane technology. Our purpose in this paper is to show that 7-branes and string
junctions stretched among them precisely capture the global structure of a rational
elliptic surface.
To explain what kind of global structures we will discuss, let us start with briefly
reviewing the heterotic stringrF-theory duality. Duality between F-theory on an elliptic
K 3 surface SF and the heterotic string theory on a two-torus T 2 has been the source of
inspiring various duality relations in lower dimensions w3–6x. In the type IIB picture,
singularities of S F are described in terms of coalescing 7-branes which are in general
mutually non-local. The sub-lattice G 18,2 of the 22-dimensional homology lattice G 19,3
PII: S 0 5 5 0 - 3 2 1 3 Ž 0 0 . 0 0 0 1 3 - 4
72 M. Fukae et al.r Nuclear Physics B 572 (2000) 71–94
of K 3 is spanned by junctions stretched among 7-branes w7x. When the size of T 2 is

very large, S F can be viewed as consisting of two rational elliptic surfaces S1 , S2 , each
of which may be associated to one of the heterotic E8 = E8 gauge groups w8x. More
precisely, in this regime, S1 and S2 intersect along an elliptic curve E) whose complex
structure is identified as that of T 2 on the heterotic side. While keeping E) fixed, then,
deformations of the complex structure of Si are dual to deformations of the correspond-
ing E8 bundle on T 2 w8,9x.
The gauge symmetry in the heterotic string is now understood in terms of shrinking
two-cycles in Si , or equivalently coalescing 7-branes, on the F-theory side. As found in
Refs. w10,11x, while the resulting singularity fixes the root lattice, and hence the gauge
symmetry algebra, the torsion part of the Mordell–Weil group plays a crucial role to
determine the gauge group. The Mordell–Weil group, as will be described in more
detail in the text, is an Abelian group generated by rational sections of the elliptic
fibration, and decomposed into its free part and its torsion part. When the torsion part is
non-trivial the corresponding gauge group acquires non-trivial p 1 , and thus there appear
non-simply-connected gauge groups w11x.
Thus it is seen that the Mordell–Weil group reflects a global structure of a rational
elliptic surface. Furthermore it is known that the Mordell–Weil group equipped with a
natural bilinear pairing possesses the lattice structure which is referred to as the
Mordell–Weil lattice w12x. The classification of the Mordell–Weil lattice of a rational
elliptic surface has already been completed, thanks to Oguiso–Shioda w13x.
In this paper, describing the singularity structures of a rational elliptic surface S in
terms of 12 7-branes with trivial total monodromy, we will show that the junction
lattices on the 7-brane backgrounds precisely produce all the Mordell–Weil lattices of S
as listed in Ref. w13x. Especially the torsion elements are found to be identified as the
‘‘fraction’’ of global loop junctions which, on the other hand, are related to the
imaginary roots of E-type affine Lie algebras.
This paper is organized as follows: In Section 2, we introduce some elementary
arithmetic of elliptic curves and the notion of the Mordell–Weil group of S using
several explicit examples of S presented as elliptic curves. In Section 3, after a short
review of 7-branes and junctions, we determine the brane configurations which describe
the structure of the Mordell–Weil lattice given in Ref. w13x. In this calculation we
recognize a non-trivial role played by the global structure of the surface. In Section 4,
the charge integrality condition is re-considered to obtain the weight lattice and torsions
from string junctions. In particular, torsions are expressed as fractional loop junctions.
Finally we discuss possibly stable non-BPS states in F-theory on K 3 on the basis of the
structure of the Mordell–Weil lattice.
2. Preliminaries
2.1. Mordell–Weil group
Consider an elliptic curve E defined over Q,

y 2 s 4 x 3 y g2 x y g3 , g 2 , g 3 g Q. Ž 2.1 .
M. Fukae et al.r Nuclear Physics B 572 (2000) 71–94 73
Any cubic in P 2 can be transformed into this canonical form. A point P s Ž x, y . on E

is called a rational point if x, y g Q, the totality of which is denoted as EŽ Q ..
Determination of the structure of EŽ Q . is a deep arithmetic problem. The fundamental
fact is that EŽ Q . has the structure of an Abelian group, known as the Mordell–Weil
group w15x.
In order to see the group law explicitly, let us recall that a point P s Ž x, y . on E is
parametrized by the Weierstrass `-function as x s `Ž t ., y s `X Ž t .. The addition formu-
lae for ` read
2
`X Ž s . y `X Ž t .
` Ž s q t . s y` Ž s . y ` Ž t . q 1
4
ž `Ž s . y `Ž t . / ,
3
X X X
`X Ž s . y `X Ž t .
` Ž s q t . s y` Ž s . y ` Ž t . q 1
4
ž `Ž s . y `Ž t . /
` Ž s . `X Ž t . y ` Ž t . `X Ž s .
q3 . Ž 2.2 .
`Ž s . y `Ž t .
It is obvious that, given Pi s Ž x i , yi . s Ž`Ž t i .,` X Ž t i .. g EŽ Q . with i s 1,2, the ‘‘sum’’
P3 s P1 q P2 defined by P3 s Ž x 3 , y 3 . s Ž`Ž t 1 q t 2 .,` X Ž t 1 q t 2 .. is also in EŽ Q .. The
special point at infinity O s Ž`,`. is regarded as the unit of this addition. The inverse of
P s Ž x, y . is given by yP s Ž x,y y .. When P1 s P2 , the r.h.s. of Ž2.2. is evaluated by
taking a limit s ™ t, see Ž3.2. for explicit formulae.
When the curve E is expressed as a complex torus Et s CrŽ Z q t Z ., the addition is
nothing but the usual sum of complex numbers t i g Et . Note that if t 1 q t 2 q t 3 s 0 in
Et then
` Ž t1 . `X Ž t1 . 1
X
det ` Ž t 2 . ` Ž t2 . 1 s 0. Ž 2.3 .
X
`Ž t3 . ` Ž t3 . 1
This means that three distinct points Pi g E satisfy P1 q P2 q P3 s 0 if and only if they
are on the same line. This geometric rule of the addition, classically known to Fermat
and Euler, is applicable for any cubic which is not necessarily in the canonical form
Ž2.1..
Mordell’s theorem says that the group EŽ Q . is finitely generated,
E Ž Q . ' Z [r [ E Ž Q . tor , Ž 2.4 .
where the torsion part EŽ Q . tor is generated by Žat most two. generators P g EŽ Q . such
that mP s O for some m g Z ) 0 Žsee Ref. w11x and references therein..
2.2. Elliptic surfaces
What we have described in the previous subsection can be readily generalized for
other field K than Q. The relevant case for our analysis in this paper is that K is a field
of rational functions of one variable, namely K s C Ž z . s aŽ z .rb Ž z . N
aŽ z .,bŽ z . polynomials in z 4 . Then the curve
y2 sx3 qf Ž z . xqg Ž z . , Ž 2.5 .
where f Ž z ., g Ž z . g K, represents an elliptic surface S over P 1 which is a family of

elliptic curves E s Ž x, y .4 parametrized by z g P 1. Every elliptic surface over P 1 with
section can be recast in this form. At the point z on P 1 where the discriminant
D s 4 f Ž z . 3 q 27g Ž z . 2 vanishes, the fiber becomes singular. Possible singular fibers
were classified by Kodaira, according to which the blowing-up diagram of each
singularity is represented by the Dynkin diagram of A, D and E type w16,17x. In the IIB
brane picture, the singularities correspond to the coinciding 7-branes w3x.
Now, a point P in the Mordell–Weil group EŽ K . is a solution P s Ž x Ž z ., y Ž z .. of
Ž2.5. such that x Ž z ., y Ž z . g K. In view of the elliptic surface S, a point P g EŽ K . is a
rational section of the elliptic fibration. Applying the addition formulae Ž2.2. fiberwise,
we see that EŽ K . has the structure of an Abelian group.
In the following let us take several examples of elliptic surfaces and compute rational
sections to show the Abelian group property explicitly. We note in passing that
determining sections of elliptic surfaces is essential when one constructs the Seiberg–
Witten differential in N s 2 supersymmetric gauge theories with matter w18–21x. In
each example below, No. a refers to the entry a of Table 2 which will appear in
Section 3. We obtain sections of these examples by using the method in Ref. w21x.
Example (No. 72): N s 2 SU(2) theory with massless Nf s 3 flaÕors [18]

L23 2
y2 sx2 Ž xyz . y Ž xyz. . Ž 2.6 .
64
The discriminant is D s L 23 Ž256 z y L23 . z 4r4096 and the singularities are given by
A3 : Ž z s 0. , D5 : Ž z s `. , A0 : Ž z s L32r256 . . Ž 2.7 .
This curve has four sections generated by a single element P,
i L3 i L3
P s 0,yž 8 /
z , 2 P s Ž z ,0 . , 3 P s 0, ž 8 /
z , 4 P s O s Ž `,` . .
Ž 2.8 .
Hence the Mordell–Weil group consists of the torsion part only, i.e. EŽ K . s Zr4Z.
Example (No. 66): Massless Eˆ3 curÕe [1,2,22]

y 2 s x 3 q Ž z 2 q 10 z y 23 . x 2 q 128 Ž 1 y z . x. Ž 2.9 .
The discriminant is D s y16384Ž z q 1 .3Ž zy1 .2 Ž z q 17. and the singularities are
A5 : Ž z s `. , A2 : Ž z s y1. , A1 : Ž z s 1. , A0 : Ž z s y17. .
Ž 2.10 .
This curve has six sections generated by a single element P,
P s Ž 8 Ž 1 y z . ,y 8 Ž 1 y z 2 . . , 2 P s Ž 16,y 16 Ž 1 q z . . , 3 P s Ž 0,0 . ,
4 P s Ž 16,16 Ž 1 q z . . , 5P s Ž 8 Ž 1 y z . ,8 Ž 1 y z 2 . . , 6 P s O s Ž `,` . .
Ž 2.11 .
Hence we have EŽ K . s Zr6Z.
Example (No. 44 and its degeneration to No. 70): MassiÕe and massless Eˆ1 curÕe
[1,2,22]
The massive Eˆ1 curve reads
16 65536
ž ž
y2 sx3 q z2 y2 pq
p / /
z q p 2 y 224 x 2 q
p2
x. Ž 2.12 .
The discriminant is
D s y4294967296 Ž z y p y 16 . Ž z y p q 16 . Ž pz y p 2 q 16 p y 32 .
= Ž pz y p 2 y 16 p y 32 . rp 6 . Ž 2.13 .
The singularities are
A7 : Ž z s `. ,
32
A0 q A0 q A0 q A0 :
ž z s p " 16, p q
p
" 16 .
/ Ž 2.14 .
The Mordell–Weil group EŽ K . is generated by

16 q p 2
P s Ž 0,0 . ,
ž
Q s 256,256 z y 256
p / ,
256 256 16 q p 2
Rs
ž ,
p2 p2
z y 256
p3
.
/ Ž 2.15 .
Since 2 P s O, Q q R s P we have EŽ K . s Z [ Zr2 Z where the free part Z and the

torsion part Zr2 Z are generated by Q and P, respectively.
Under the degeneration at p s 1 Žmassless Eˆ1 curve. the discriminant becomes
2
D s y4294967296 Ž z q 15 . Ž z y 49 . Ž z y 17 . Ž 2.16 .
and the singularities are
A7 : Ž z s `. , A1 : Ž z s 17 . , A0 q A0 : Ž z s y15,49 . . Ž 2.17 .
In this case, we have Q s R and 4Q s 2 P s O, thereby EŽ K . reduces to EŽ K . s Zr4Z.
Example (No. 27): MassiÕe A 2 curÕe [21]

y 2 s x 3 q ux q Õ q z 2 . Ž 2.18 .
4 2 3 2
The discriminant is D s 27z q 54Õz q 4 u q 27Õ and the singularities are
E6 : Ž z s `. ,
A0 q A0 q A0 q A0 : Ž D Ž z . s 0. , Ž 2.19 .
This curve has six fundamental sections,
"Pi s Ž a i ," z . , i s 1,2,3, Ž 2.20 .
where a i are determined through Ž x y a1 .Ž x y a 2 .Ž x y a 3 . s x 3 q ux q Õ. These six
sections "Pi correspond to the six fundamental weights of the A 2 algebra "L1 ,
"Ž L2 y L1 . and "ŽyL2 .. By addition formulae Ž2.2., one can generate a section Pl, m
corresponding to a weight l s l L1 q m L 2 for each Ž l,m. g Z [2 . Thus we see that

EŽ K . ' Z [2 as an abstract group, but it has a more detailed structure as a lattice
EŽ K . s A)2 . Here a ‘‘lattice’’ is introduced as a free Abelian group L s Z [r with a
symmetric bilinear form ŽP,P .: L m L ™ Z. The equality EŽ K . s A)2 means the isomor-
phism as a lattice. The bilinear form on EŽ K . is defined by the ‘‘height pairing’’ which
can be explicitly evaluated in terms of an intersection pairing on an elliptic surface S,
see the next section.
Finally it is mentioned that, extending the computation in Ref. w21x, we have shown
that the Seiberg–Witten differential for the curve Ž2.18. can be constructed so that it has
the poles with residues located on sections Ž3.6.. The details may appear elsewhere.
3. Mordell–Weil lattice versus junction lattice
3.1. Mordell–Weil lattice
Let us start with recapitulating what we have discussed in the previous section, then
we introduce the notion of the Mordell–Weil lattice, following Ref. w12x. We take an
elliptic curve E defined over a field K
y 2 s x 3 q fx q g , f , g g K . Ž 3.1 .
2
A point P s x, y g K on the curve E is called K-rational point. Let E K denote a
Ž . Ž .
set of all the K-rational points plus the point at infinity O s Ž`,`.. For given
Pi s Ž x i , yi . g EŽ K . with i s 1,2, the third point P3 s P1 q P2 s Ž x 3 , y 3 . g EŽ K . is
defined by
x 3 s yx 1 y x 2 q m2 , y 3 s ym Ž x 3 y x 1 . y y 1 , Ž 3.2 .
where
Ž y1 y y 2 . r Ž x 1 y x 2 . , if P1 / P2 ,
ms
½ Ž 3 x 12 q f . r Ž 2 y1 . , if P1 s P2 .
Ž 3.3 .
With respect to this addition law, EŽ K . has the structure of an Abelian group, called the
Mordell–Weil group. The point at infinity O g EŽ K . is the unit of this addition.
In this paper we are mainly concerned with the case of K s C Ž z . Žthe field of
rational functions on z .. Then Ž3.1. is naturally considered as an elliptic surface S over
P 1, for which the points P s Ž x Ž z ., y Ž z .. g EŽ K . represent the rational sections of this
fibration. The Kodaira classification of singular fibers is presented in Table 1. We
denote by T the lattice corresponding to the singular fibers.
Table 1
Kodaira classification, ADE singularities and 7-branes
Fiber type Singularity lattice 7-branes
I n Ž nG1. A ny1 A n s A ny1
II,III,IV Ž ns 0,1,2. An A nq1 C s Hn
I n) Ž nG 0. Dnq4 A nq4 B C s Dnq4
II ) ,III ) ,IV ) Ž ns8,7,6. En A ny1 B C 2 s En
There exists a deep relation between the singularities T and the sections EŽ K .,
reflecting the global structure of the elliptic surface S. The essential idea in studying
such a relation is to equip EŽ K . with the lattice structure which can be described in
terms of intersections on S.
The lattice structure, or the height pairing ŽP,P . of the Mordell–Weil group EŽ K .
was introduced by Shioda using the intersection pairing w12x. The Mordell–Weil group
equipped with the height pairing is called the Mordell–Weil lattice. In Theorem 8.1 of
Ref. w12x, the height pairing Ž P,Q . for sections P,Q g EŽ K . is explicitly calculated as
follows:
Ž P ,Q . s P P O q Q P O y P P Q q x Ž OS . y Ý contr Õ Ž P ,Q . ,
Õ
Ž P , P . s 2 x Ž OS . q 2 P P O y Ý contr Õ Ž P . , Ž 3.4 .
Õ
where P P Q denotes the intersection pairing, x Ž OS . is the arithmetic genus of S Žs 1

for rational elliptic surfaces. and contr Õ is the local contribution from each singular
point Õ,
0, if i s 0 or j s 0,
contr Õ Ž P ,Q . s
½ Ž Cy1
Õ ij,
. otherwise,
Ž 3.5 .
and contr Õ Ž P . s contr Õ Ž P, P .. Here C Õ is the Cartan matrix Žof finite type. correspond-
ing to the singularity at Õ and the indices i, j label the components of the singular fibers
with which P or Q intersects. The component intersecting with the zero section O is
specified by i s 0.
Example (No. 27): The massiÕe A 2-curÕe (2.18) continued

The rational section Pl, m corresponding to a weight vector l s l L1 q m L2 takes the
form
fŽ z. c Ž z.
xŽ z. s 2
, yŽ z. s 3
, Ž 3.6 .
x Ž z. x Ž z.
where deg c Ž z . s d l, m s l 2 q lm q m2 , deg x Ž z . s 13 d l, m y 1 wresp. 13 d l, m y 13 x and
deg f Ž z . s 23 d l, m wresp. 23 d l, m y 23 x for l ' m Žmod 3. wresp. otherwisex. For this section
Pl, m , the second formula in Ž3.4. can be checked since we evaluate Ž Pl, m , Pl, m . s Ž l L1 q
m L2 . 2 s 23 d l, m , contr zs`Ž Pl, m . s 0 wresp. 43 x and O P Pl, m s deg x Ž z .1.
Noting that the structure of the Mordell–Weil lattice EŽ K . is essentially determined
by the singularity lattice T, Oguiso–Shioda classified the Mordell–Weil lattice EŽ K . of
a rational elliptic surface w13x. This will be described in detail in Section 3.3. Our task is
now to figure out how the Mordell–Weil lattice EŽ K . and the singularity lattice T are
related to each other in terms of the junctions on rational elliptic surfaces 2 .
1
In the homogeneous coordinates Ž X:Y:Z ., the curve and sections are rewritten as ZY 2 s X 3 q f Ž z . XZ 2
q g Ž z . Z 3 , P s Ž X:Y:Z . s Ž fx , c , x 3 .. Hence, the intersection of P with zero-section O s Ž0:1:0. is given
by deg x Ž z ..
2
In case of more general elliptic surfaces, elliptic K 3 for instance, the junction lattice will contain
transcendental cycles also.
3.2. Brane configurations and junction lattice
We concentrate on a rational elliptic surface S defined by Ž3.1. where f s f Ž z . and

g s g Ž z . are some polynomials in z, and deg f F 4, deg g F 6 with D / constant.
Generically there exist 12 singular points z i where DŽ z i . s 0. Each singular fiber at
z s z i has a local monodromy K w p i , q i x labeled by pi ,qi g Z. Here the monodromy matrix
takes the form
1 q pq yp 2
Kw p,qx s
ž q2 1 y pq / , Ž 3.7 .
corresponding to a vanishing cycle pa q q b with a , b being homology cycles of a

fiber torus. Physically a singular point is interpreted as the position of a 7-brane X w p, q x
on which a Ž p,q .-string with a boundary homologous to pa q q b can end w3x. Among
various Ž p,q . 7-branes X w p, q x it is convenient to express the representative ones as
X w1,0x s A, X w1,y 1x s B, X w1,1x s C. The Kodaira singularities can then be described as a
coalescence of collapsible sub-configurations of 7-branes, see Table 1 w23,24x.
To specify a brane configuration, we place the 12 branes, say, on the real axis of the
z-plane and draw downwards the branch cuts emanating from the branes. Thus, for a
brane configuration
X w p 1 , q1 x X w p 2 , q 2 x PPP X w p 12 , q12 x , Ž 3.8 .
we have the total monodromy
K s K w p 12 , q12 x PPP K w p 2 , q 2 x K w p 1 , q1 x , Ž 3.9 .
which should be trivial, i.e. K s 1, to describe a rational elliptic surface S. A standard
realization of such a brane configuration is w7x
Eˆ9 s A8 B C B C. Ž 3.10 .
A topological configuration of strings Žor string junctions. associated to the branes
can be parameterized as
12
Js Ý Qi si or J s Ž Q1 ,Q2 , . . . ,Q12 . , Ž 3.11 .
is1
where s i stands for the outgoing Ž pi ,qi .-string starting at X w p i , q i x and Q i is the net
number of outgoing Ž pi ,qi .-strings. By definition the charges Q i must be integral and
are called the invariant charges w23x. The total Ž p,q . charges of a string junction are
12
Ž p,q . s Ý Q i Ž pi ,qi . . Ž 3.12 .
is1
A junction represents a closed two-cycle in S if and only if its total charges vanish.
A Ž p,q .-junction J is in general decomposed into two pieces J1 , J2 as depicted in
Fig. 1. If J is a singlet with respect to the symmetry realized on the branes, then J1 s 0.
The charge z 0 s Ž r, s . of the tadpole loop J2 is determined by charge conservation at
the trivalent vertex. When the tadpole loop goes around the A n , Dn and En branes of
Fig. 1. A string junction.
Table 1, the loop charges Ž r, s ., invariant charges Q i and outgoing charges Ž p,q . are
related as follows:
An : d r , s s ys Ž s 1 q . . . qs nq1 . , Ž p,q . s Ž y Ž n q 1 . s,0 . , Ž 3.13 .
Dn : d r , s s ys Ž s 1 q . . . qs n . y Ž r y Ž n y 1 . s . s nq1 y Ž r y Ž n y 3 . s . s nq2 ,
Ž p,q . s Ž y2 r q Ž n y 4 . s,y 2 s . , Ž 3.14 .
En : d r , s s ys Ž s 1 q . . . qs ny1 . y Ž r y Ž n y 2 . s . s nq1
y Ž r y Ž n y 4 . s . Ž s nq 1 q s nq2 . ,
Ž p,q . s Ž y3r q Ž 2 n y 9 . s,y r q Ž n y 6 . s . . Ž 3.15 .
It should be remembered that if one employs a different 7-brane configuration from the
ones given in Table 1 to describe the ADE singularities, the assignment of the Ž r, s .
charges will also change.
The junctions form a lattice which is endowed with a symmetric intersection pairing
defined by w23x
Ž s i , s i . s y1, Ž s i , s j . s Ž s j , s i . s 12 Ž pi q j y pj qi . , for i - j. Ž 3.16 .
For the sub-configurations in Table 1, the junction lattice with Ž p,q . s 0 is known to be
isomorphic to the corresponding root lattice w23x. In these cases, the root junctions
represent the BPS strings responsible for the symmetry enhancement.
For Eˆ9 the junction lattice forms an indefinite lattice with signature Ž2 q ,10 y ..
Under the condition that the total Ž p,q . charges Ž3.12. vanish, this lattice reduces to the
semi-definite one isomorphic to
Z d 1 [ Z d 2 [ Ž yE8 . . Ž 3.17 .
Here d 1 and d 2 are the null junctions. A choice of basis is
d 1 s Ž 0,0,0,0,0,0,0,0,y 1,y 1,1,1 . ,
d 2 s Ž y1,y 1,y 1,y 1,y 1,y 1,y 1,y 1,7,5,y 3,y 1 . . Ž 3.18 .
The d 1 , d 2 junctions are expressed as loops of Ž1,0., Ž0,1. strings, respectively,
surrounding the Eˆ9 branes Ž3.10. counterclockwise, see Fig. 2. These null junctions
Fig. 2. A loop junction.

represent two imaginary roots of the loop algebra Eˆ9 w7x and will play a crucial role in
our study of torsions in Section 4.
3.3. Mordell–Weil lattice from junctions
A rational elliptic surface S is a special kind of 9 points blown-up of P 2 and

b 2 s dim H2 Ž S . s 10. H2 Ž S . is a unimodular Lorentzian lattice with signature Ž1 q ,9 y ..
Let O and F be a class of zero section and generic fiber. Their intersections are
O P O s y1, O P F s 1 and F P F s 0. The orthogonal complement of ² O, F : in H2 Ž S .
is isomorphic to yE8 .
Some elements of H2 Ž S . appear as components of singular fibers. They generate the
root lattice corresponding to the singularity type. Let T ; E8 be the lattice generated by
the components of singular fibers. Theorem 10.3 in Ref. w12x states that the structure of
the Mordell–Weil group of the rational elliptic surface S is described as
E Ž K . , L) [ Ž T XrT . , Ž 3.19 .
) H X
where L is the dual of L s T Žin E8 . and T s T m Q l E8 . From Ž3.19. the torsion
subgroup EŽ K . tor of EŽ K . is read off as EŽ K . tor , T XrT, and hence EŽ K .rEŽ K . tor ,
L) . According to this theorem, the computation of EŽ K . is reduced to the embedding of
lattice T in E8 .
In Ref. w13x all the possible structures of T and EŽ K . are classified. They are listed
in Table 2 where r s rank EŽ K . 3. In the last column for EŽ K ., ² k : denotes a rank one
lattice Zx with Ž x, x . s k which does not correspond to a weight lattice except for the
case k s 1r2 Ž A1) s ²1r2:.. For higher rank cases, it is interesting that we have also
several lattices of non-Cartan type. For such cases occurring in No. ), the lattice EŽ K .
can be represented by the following matrices LŽ ) . :
2 1 0 y1
3 1 y1
LŽ12. s 1
6
7
1
0
y1
1
5
3
1
2
3
6
3
1
3
5
,
0 LŽ17. s 101
ž 1
y1
7
3
3 ,
7 /
2 1
LŽ19. s 121 1
2 ž 7
2 4/
2 , LŽ23. s 61 ž 1 2/,
2 1 2 1 2 1
LŽ25. s 17 ž /, LŽ31. s 151 ž , / LŽ33. s 101 ž . / Ž 3.20 .
1 4 1 8 1 3
To understand this result of Ref. w13x in terms of junctions, we have to solve the
following two problems:
1. Find the collapsible sub-configurations of 7-branes which support the singularity

lattice T as the junction lattice.
2. Compute the lattice L using string junctions on the 7-brane backgrounds.
3
EŽ K . s A1) [²1r6: for No. 32 and Ž Z r2 Z . 2 for No. 70 in Ref. w13x should read EŽ K . s LŽ23. and
Z r4Z, respectively. See Ref. w14x.
The problem Ž1. is solved by finding a suitable rearrangement of the Eˆ9 configura-
tions Ž3.10. with the use of the brane move w24,25x
X z i X z iq 1 s X zXi X zXiq 1 , Ž 3.21 .
where
Ž a. zXi s z iq1 , zXiq1 s z i q Ž z i = z iq1 . z iq1 ,
Ž b. zXi s z iq1 q Ž z i = z iq1 . z i , zXiq1 s z i , Ž 3.22 .
with z i = z j s pi q j y pj qi for z i s Ž pi ,qi .. Under the brane move Ž3.21., the charges Q i
also change as
Ž a . QXi s Q iq1 y Ž z i = z iq1 . Q i , QXiq1 s Q i ,
Ž b . QXi s Q iq1 , QXiq1 s Q i y Ž z i = z iq1 . Q iq1 , Ž 3.23 .
in such a way that
Q i z i q Q iq1 z iq1 s QXi zXi q QXiq1 zXiq1 , Ž 3.24 .
thereby the total charges are kept invariant. Moreover, what is important is that the
brane moves preserve the junction lattice. That is, under the unimodular transformation
Ž3.23. we can prove the relation
yJ 2 s Ý Q i Ž s i , s j . Q j s Ý QXi Ž sXi , sXj . QXj s yJ X 2 . Ž 3.25 .
i, j i, j
To this end, let us rewrite J 2 as

yJ 2 s Ý Q i Q i , Qi s Qi y Ý Ž z k = z i . Qk . Ž 3.26 .
i k-i
Then, invariance of J 2 follows from the transformation formulae for Q j given by
Ž a . QXi s Q iq1 , QXiq1 s Q i q Ž z i = z iq1 . Q iq1 ,
Ž b . QXi s Q iq1 q Ž z i = z iq1 . Q i , QXiq1 s Q i . Ž 3.27 .

Now we find brane configurations for every T as shown in Table 2 where the branes
put in the parentheses are the mutually coinciding ones corresponding to the singular
fibers. We note that the A n branes are used to represent the A n singularity with
n s 0,1,2, though one may use Hn as well without changing the structure of EŽ K .. All
the configurations in Table 2 can be obtained from the Eˆ9 configuration Ž3.10. by
suitable brane moves. This means that the lattice T in Table 2 can be realized as a
sub-lattice of the junction lattice on Eˆ9 , hence T ; E8 .
Example (No. 63)

The move from A8 B C B C to A9 X w2,y 1x X w1,y 2x C is given as follows:
A8 B C B C s A8 B C 2 X w3 ,1xs A7 B X w0,1x C 2 X w3,1xs A7 B A 2 X w0,1x X w3,1x
s A9 X w3 ,y 1x X w0,1x X w3,1xs A8 X w2,y 1x C X w4,1x A
s A8 X w2 ,y 1x X w1,y 2x C As A9 X w2,y 1x X w1,y 2x C. Ž 3.28 .
It is also possible to find corresponding curves. Some of them have already been
given in Section 2, from which it is clear how to identify a curve with an entry of
Table 2
Brane configurations
No. r T Branes EŽ K .
8
1 8 0 A BCBC E8
2 7 A1 Ž A 2 . A6 B C B C E7)
3 6 A2 Ž A3 . A5 B C B C E6)
4 A1 [2 Ž A2 . 2 A4 B C B C D6)
5 5 A3 Ž A4 . A4 B C B C D5)
6 A 2[A1 Ž A3 .Ž A 2 . A 3 B C B C A)5
7 A1 [3 Ž A2 . 3 A2 B C B C D4) [A1)
8 4 A4 Ž A5 . A3 B C B C A)4
9 D4 Ž A4 B C . A4 B C D4)
10 A 3[A1 Ž A4 .Ž A2 . A2 B C B C A)3 [A1)
11 A 2 [2 Ž A3 . 2 A2 B C B C A)[2
2
12 A 2[A1 [2 Ž A3 .Ž A 2 . 2 A B C B C LŽ12.
13 A1 [4 Ž A2 . 4 B C B C D4) [ Z r2 Z
14 A1 [4 Ž A2 . 4 A X w2,y 1x X w1,y 2x C A1)[4
15 3 A5 Ž A6 . A 2 B C B C A)2 [A1)
16 D5 Ž A5 B C . A3 B C A)3
17 A 4[A1 Ž A5 .Ž A 2 . A B C B C LŽ17.
18 D4[A1 Ž A4 B C .Ž A2 . A2 B C A1)[3
19 A 3[A 2 Ž A4 .Ž A3 . A B C B C LŽ19.
20 A 2 [2[A1 Ž A3 . 2 Ž A2 . B C B C A)2 [²1r6:
21 A 3[A1 [2 Ž A4 .Ž A2 . 2 B C B C A)3 [ Z r2 Z
22 A 3[A1 [2 Ž A4 .Ž A2 . 2 A X w2,y 1x X w1,y 2x C A1)[2[²1r4:
23 A 2[A1 [3 Ž A3 .Ž A 2 . 3 X w2,y 1x X w1,y 2x C A1) [LŽ23.
24 A1 [5 Ž A2 . 2 Ž B 2 . 2 Ž X w0,1x
2 . 2
X w2,1x A1)[3[Z r2 Z
7
25 2 A6 Ž A . ABCBC LŽ25.
26 D6 Ž A6 B C . A 2 B C A1)[2
27 E6 Ž A5 B C 2 . X w3,1x A3 A)2
28 A 5[A1 Ž A6 .Ž A2 . B C B C A)2 [ Z r2 Z
29 A 5[A1 Ž A6 .Ž A2 . A X w2,y 1x X w1,y 2x C A1) [²1r6:
30 D5[A1 Ž A5 B C .Ž A2 . A B C A1) [²1r4:
31 A 4[A 2 Ž A5 .Ž A 3 . B C B C LŽ31.
32 D4[A 2 Ž A4 B C .Ž A3 . A B C LŽ23.
33 A 4[A1 [2 Ž A5 .Ž A 2 . 2 X w2,y 1x X w1,y 2x C LŽ33.
34 D4[A1 [2 Ž A4 B C .Ž A2 . 2 B C A1)[2[ Z r2 Z
35 A 3[A 3 Ž A4 . 2 B C B C A1)[2[ Z r2 Z
36 A 3[A 3 Ž A4 . 2 A X w2,y 1x X w1,y 2x C ²1r4:[2
37 A 3[A 2[A1 Ž A4 .Ž A3 .Ž B 2 . X w1,y 3x C A A1) [²1r12:
38 A 3[A1 [3 Ž A4 .Ž A2 .Ž B 2 .Ž X w0,1x2 .
BC A1) [²1r4:[ Z r2 Z
39 A 2 [3 Ž A3 . 3 X w2,y 1x X w1,y 2x C A)2 [Z r3Z
40 A 2 [2[A1 [2 Ž A3 . 2 Ž B 2 .Ž X w0,1x
2 .
BC ²1r6:[2
41 A 2[A1 [4 Ž A3 .Ž B 2 .Ž X w0,1
2 .Ž 2 .Ž
x B
2 .
X w0,1x A LŽ23.[Z r2 Z
42 A1 [6 Ž A2 . 2 Ž B 2 . 2 Ž X w0,1
2 .Ž 2
x X w2,1x . A1)[2[Ž Z r2 Z . 2
43 1 E7 Ž A6 B C 2 . X w3,1x A2 A1)
44 A7 Ž A8 . B C B C A1) [ Z r2 Z
45 A7 Ž A8 . A X w2,y 1x X w1,y 2x C ²1r8:
46 D7 Ž A7 B C . A B C ²1r4:
47 A 6[A1 Ž A7 .Ž B 2 . X w1,y 3x C A ²1r14:
48 D6[A1 Ž A6 B C .Ž A2 . B C A1) [ Z r2 Z
49 E6[A1 Ž A2 .Ž A5 B C 2 . X w3,1x A ²1r6:
50 D5[A 2 Ž A5 B C .Ž A3 . B C ²1r12:
Table 2 Ž continued .
No. r T Branes EŽ K .
51 A 5[A 2 Ž A6 .Ž A3 . X w2,y 1x X w1,y 2x C A1) [Z r3Z
52 D5[A1 [2 Ž A5 B C .Ž B 2 .Ž X w0,1x
2 .
A ²1r4:[Z r2 Z
53 A 5[A1 [2 Ž A6 .Ž B 2 .Ž X w0,1x
2 .
BC ²1r6:[ Z r2 Z
54 D4[A 3 Ž A4 B C .Ž A4 . B C ²1r4:[ Z r2 Z
55 A 4[A 3 Ž A5 .Ž A 4 . X w2,y 1x X w1,y 2x C ²1r20:
56 A 4[A 2[A1 Ž A5 .Ž A 3 .Ž B 2 . X w1,y 3x C ²1r30:
57 D4[A1 [3 Ž A4 B C .Ž A2 .Ž B 2 .Ž X w0,1x2 .
A1) [Ž Z r2 Z . 2
58 A 3 [2[A1 Ž A4 . 2 Ž B 2 . X w1,y 3x C A1) [ Z r4Z
59 A 3[A 2[A1 [2 Ž A4 .Ž B 3 .Ž X w0,1x
2 .Ž 2 .
X w2,1x X w3,1x ²1r12:[Z r2 Z
60 A 3[A1 [4 Ž A4 .Ž B 2 .Ž X w0,1x
2 .Ž 2 .Ž 2 .
B X w0,1x ²1r4:[Ž Z r2 Z . 2
61 A 2 [3[A1 Ž A3 . 2 Ž B 3 .Ž X w1,
2
y 2x C
. ²1r6:[Z r3Z
62 0 E8 Ž A7 B C 2 . X w3,1x A 0
63 A8 Ž A9 . X w2,y 1x X w1,y 2x C Z r3Z
64 D8 Ž A8 B C . B C Z r2 Z
65 E7[A1 Ž A2 .Ž A6 B C 2 . X w3,1x Z r2 Z
66 A 5[A 2[A1 Ž A6 .Ž B 3 .Ž X w1,
2
y 2x C
. Z r6Z
67 A 4 [2 Ž A5 .Ž B 5 . X w2,y 3x C Z r5Z
68 A 2 [4 Ž A3 .Ž B 3 .Ž X w0,1x
3 .Ž 3 .
C Ž Z r3Z . 2
69 E6[A 2 Ž A3 .Ž A 5 B C 2 . X w3,1x Z r3Z
70 A 7[A1 Ž A8 .Ž B 2 . X w1,y 3x C Z r4Z
71 D6[A1 [2 Ž A6 B C .Ž B 2 .Ž X w0,1x
2 . Ž Z r2 Z . 2
72 D5[A 3 Ž A5 B C .Ž B 4 . X w1,y 2x Z r4Z
73 D4 [2 Ž A4 B C .Ž A4 B C . Ž Z r2 Z . 2
74 Ž A 3[A1 .[2 Ž A4 .Ž B 4 .Ž X w0,1x
2 .Ž 2 .
X w2,1x Z r4Z[ Z r2 Z
Table 2. Let us enumerate more examples based on Refs. w1,2,22x. The massive Eˆn
Ž n s 8, . . . ,1. and E˜1 curves correspond to Nos. 1, 2, 3, 5, 8, 15, 25, 44 and 45,
respectively. The massless Eˆn Ž n s 8, . . . ,3,1. and E˜0 curves correspond to Nos. 62, 65,
69, 72, 67, 66, 70 and 63, respectively. The Seiberg–Witten curves for N s 2 SUŽ2.
theory with various flavor symmetries w18x can be identified with No. 64 Ž Nf s 0., No.
46 Ž Nf s 1., Nos. 26, 48, 71 Ž Nf s 2., Nos. 16, 30, 50, 52, 72 Ž Nf s 3. and Nos. 9, 18,
32, 34, 54, 57, 73 Ž Nf s 4.. In each case, the number of independent mass parameters is
given by r s rank EŽ K .. For the extremal cases r s 0, the curves and sections were
explicitly obtained earlier in Ref. w26x.
When we consider curves in view of Table 2 there is a point to notice. As remarked
above, one can use either A n or Hn branes to describe the A n singular fibers with
n s 0,1,2. This does not change the structure of EŽ K ., whereas it does change the
explicit form of corresponding elliptic curves, and hence we are led to different physical
interpretation. To clarify the point, let us examine the case of EŽ K . s E8 , E7) , E6) . The
7-branes are rewritten as
Ž A9y N . A Ny 1 B C B C s Ž A9y N . X w Ny6 ,1x A Ny 1 B C 2 , Ž 3.29 .

for N s 8,7,6. Since the configuration Ž A9y N . X w Ny6,1x is equivalent to Ž A9y N . C up to
SLŽ2,Z . conjugation, further coalescence can occur,
Ž A9y N . X w Ny6,1x ™ Ž A9y N X w Ny6,1x . s H8yN , Ž 3.30 .

Fig. 3. Junctions orthogonal to T.
which also yields T s A 8y N . Although the singularity structure of T is identical, it is

shown in Ref. w1x that the limit Ž3.30. gives rise to the compactification of 5D N s 1 EN
theories down to 4D N s 2 EN theories for N s 8,7,6. Thus the structure of EŽ K .
seems not sensitive enough to distinguish these two theories.
In solving the problem Ž2., we recognize that the global structure of the surface S
plays a relevant role in determining the lattice L. This point is now illuminated by
working out several examples.
3.3.1. Case of Cartan type

Let us first consider the case No. 7 using the brane configuration Ž A2 . 3 A2 B C B C.
Accordingly, we put s i s a i Ž1 F i F 8., s 9 s b 1 , s 10 s c 1 , s 11 s b 2 and s 12 s c 2 , where
a, b and c denote outgoing Ž1,0.-, Ž1,y 1.- and Ž1,1.-strings attached to A, B and C
branes, respectively. ŽNotice that the assignment of s j may be different from this for
other cases depending on the brane configuration.. Choose the root junctions generating
the lattice T s A1 [3 as
a 1 s s1 y s 2 , a 2 s s 3 y s4 , a 3 s s5 y s6 . Ž 3.31 .
These junctions are supported by the collapsible branes A 1 A 2 , A 3 A 4 and A 5 A 6 ,
respectively, and there remain 6 branes A2 B C B C of Eˆ9 . From Table 2 we see that the
corresponding Mordell–Weil lattice is EŽ K . s L) s D4) [ A1) . These lattices are canon-
ically realized on the 7-branes A4 B C and A2 . Thus one needs apparently 8 branes,
which is more than the remaining ones. This puzzle, however, can be resolved if we
consider junctions containing strings which encircle the branes supporting T. To do so,
we note that the general form of junctions which are orthogonal to T with Ž p,q . s Ž0,0.
is parameterized as
j s Q1 Ž s 1 q s 2 . q Q 3 Ž s 3 q s 4 . q Q 5 Ž s 5 q s 6 . y Ž 2 Q1 q 2 Q 3 q 2 Q 5 q Q 8
q2Q10 q 2 Q12 . s 7 q Q8 s 8 q Ž Q10 y Q11 q Q12 . s 9 q Q10 s10 q Q11 s11
q Q12 s 12 . Ž 3.32 .
Thus these junctions span the 7-dimensional lattice. This junction lattice has elements
along the null junctions Ž3.18. and the remaining ones form a lattice of rank 5, which is
expected to be isomorphic to the lattice D4 [ A1. In fact one can find the generators of
L s D4 [ A1 as
j 1 s s 7 q s 8 y s 9 y s 10 , j 2 s ys 9 q s 11 , j 3 s s5 q s6 y s 9 y s 10 ,
j 4 s s 3 q s 4 y s 9 y s 10 , j5 s s 7 y s 8 , Ž 3.33 .
Table 3
The basis junctions for r G 5
No. Basis junctions of L yŽ j, j .
s 2 y s 3 , s 3 y s 4 , s 4 y s5 , s5 y s6 , s6 y s 7 ,
1 ½ s 7 q s 8 y s11 y s 12 , s 10 y s 12 , s 7 y s 8 C Ž E8 .
s 3 y s 4 , s 4 y s5 , s5 y s6 , s6 y s 7 ,
s 3 y s 4 , s 4 y s5 , s5 y s6 , s6 y s 7 ,
s10 y s12 , y s 3 y s 4 y s10 q s11 q s 12 ,

4
½ s 3 q s4 q s 7 q s 8 q s10 y2 s11 y3s12 ,
s6 y s 7 , s5 y s6 , s 7 y s 8
s10 y s12 , s 7 q s 8 y s 9 y s 10 ,
C Ž D6 .
5 ½ s6 y s 7 , s5 y s6 , s 7 y s 8 C Ž D5 .
~°y s q s ,
s4 q s5 q s6 q s 7 q s10 y2 s 11 y3s 12 ,
y s6 q s 7 ,
6
¢ 7 8
y s 7 y s 8 y s10 q s11 q2 s12 , s 10 y s 12
CŽ A5 .
° s 7 q s 8 y s 9 y s10 , 2
y1
y1
2
0
y1
0
y1
0
0
~
7
¢
y s 9 q s11 , s5 q s6 y s 9 y s10 ,
s 3 q s 4 y s 9 y s 10 ,
s 7 y s8 0
0
0
y1
y1
0 0
2
0
0
2
0
0
0
2
0
where j i with 1 F i F 4 are for D4 and j5 is for A1. Some of the strings in these
junctions are ending on the 7-branes which support the lattice T generated by Ž3.31.. It
is observed, however, that they take the form of tadpole loops of Ž3.13. locally around
each of the collapsible branes A 1 A 2 , A 3 A 4 and A 5 A 6 Žsee Fig. 3., and hence the
generators Ž3.33. are in fact orthogonal to T. It should be noted that any junction
orthogonal to the lattice T in Table 2 is of the form of Ž3.13. – Ž3.15. locally around the
coinciding A n , Dn and En branes which describe the singular fibers.
In a similar manner we construct the basis junctions of L for the other cases of
Cartan type intersections. The results are presented in Tables 3–7 where C Ž G . denotes
the Cartan matrix of the Lie algebra G and the self-intersection Žtimes Žy1.. of the ith
junction is given by the Ž i,i . element of the intersection matrix in the third column.
3.3.2. Case of non-Cartan type

In Table 2 there exists a class of EŽ K . which are characterized by the intersection
matrices of non-Cartan type Ž3.20.. It is intriguing that we can indeed derive these
intersection matrices from the junctions. Some examples are given as follows:
Example (No. 12)

We have the 7-branes Ž A3 .Ž A2 . 2A B C B C. The junctions orthogonal to T s A 2 [ A[2
1
are
j 1 s s 4 q s5 q 2 s 8 q s 10 y 2 s 11 y 3s 12 , j 2 s s6 q s 7 y s 11 y s 12 ,
j 3 s s 10 y s 12 , j 4 s s4 q s5 y s 11 y s 12 . Ž 3.34 .
Table 4
The basis junctions for r s 4
° s10 y s12 ,
2 y1 0 0
~s q s y s 11 y s 12 ,
8
¢
7 8
s6 y s 7 ,
s 7 y s8
y1
0
0
2
y1
0
y1
2
y1
0
y1
2
0
° y s7 q s8 ,
2 y1 0 0
~ y s8 q s9 ,
9
¢ y s 9 q s10 ,
y s 9 y s10 q s11 q s12
y1
0
0
2
y1
y1
y1
2
0
y1
0
2
0
° y s7 q s8 ,
2 0 0 0
~ y s 7 y s 8 q s11 q s12 ,
10
¢ y s10 q s12 ,
y s5 y s6 q s11 q s12
0
0
0
2
y1
0
y1
2
y1
0
y1
2
0
° s 4 q s5 q s6 q s 7 q s10 y2 s11 y3s12 ,
2 y1 0 0
~ y s7 q s8 ,
11
¢ y s10 q s12 ,
y s 7 y s 8 q s11 q s12
y1
0
0
2
0
0
0
2
y1
0
y1
2
0
° s 4 q s5 q2 s 8 q s10 y2 s11 y3s12 ,
4 y1 0 1
~ s6 q s 7 y s11 y s12 ,
12
¢ s10 y s12 ,
s 4 q s5 y s11 y s12
y1
0
1
2
y1
0
y1
2
y1
0
y1
2
0
° y s 7 y s 8 q s11 q s12 ,
2 y1 0 0
~ y s10 q s12 ,
13
¢ y s 3 y s 4 q s11 q s12 ,
y s5 y s6 q s11 q s12
y1
0
0
2
y1
y1
y1
2
0
y1
0
2
0
° s 3 q s4 q s 9 y s11 y2 s12 ,
2 0 0 0
~ s5 q s6 q s 9 y s11 y2 s12 ,
14
¢ s 7 q s 8 q s 9 y s11 y2 s12 ,
s 3 q s4 q s5 q s6 q s 7 q s 8 y2 s11 y4 s12
0
0
0
2
0
0
0
2
0
0
0
2
0
These consist of the basis junctions of L. The intersection matrix turns out to be
4 y1 0 1
y Ž ji , j j . s
y1
0
1
2
y1
0
y1
2
y1
0
y1
2
.
The inverse of this yields the matrix LŽ12. of Ž3.20..

0 Ž 3.35 .
Example (No. 17)

We have the 7-branes Ž A5 .Ž A2 . A B C B C. The junctions orthogonal to T s A 4 [ A1
are
j 1 s 2 s 8 q s 10 y s 11 y 2 s 12 , j 2 s s6 q s 7 y s 11 y s 12 , j 3 s s 10 y s 12 ,
Ž 3.36 .
Table 5
s 7 q s 8 q s10 y s 11 y2 s12 , 2 y1 0
15
½ y s10 q s12 ,
s 7 y s8
s8 y s9 ,
ž y1
0 0
2 0
2 /
2 y1 0
16
½ y s 8 q s10 ,
s 8 q s 9 y s11 y s12
2 s 8 q s10 y s11 y2 s12 ,
ž y1
0
2
y1
y1
2 /
4 y1 1
17
½ s6 q s 7 y s11 y s12 ,
s10 y s12
s 9 y s10 ,
ž y1
1
2
y1
y1
2 /
2 0 0
18
½ s 7 q s 8 y s11 y s 12 ,
s 9 q s10 y s11 y s12 ž 0
0
2
0
0
2 /
~°s
s5 q s6 q s 7 q s 8 q s 10 y2 s11 y3s12 , 2 0 y1
19
¢ 10 y s 12 ,
y2 s 8 y s10 q s11 q2 s12
ž 0
y1
2
y1
y1
4 /
~°
s10 y s12 , 2 y1 0
20
¢ s 7 q s 8 y s11 y s12 ,
2 s4 q2 s5 q2 s6 q s 7 q s 8 q2 s10 y4 s11 y6 s12
ž y1
0 0
2 0
6 /
~°
s10 y s12 , 2 y1 0
21
¢ s5 q s6 y s11 y s12 ,
y s5 y s6 y s 7 y s 8 y s10 q2 s11 q3s12
ž y1
0
2
y1
y1
2 /
~°
s5 q s6 q s 9 y s11 y2 s12 , 2 0 0
22
¢ s 7 q s 8 q s 9 y s11 y2 s12 ,
s5 q s6 q s 7 q s 8 y s 9 y s11 y2 s 12
ž 0
0
2
0
0
4 /
s4 q s5 q s6 q s 7 q s 8 q s 9 y2 s 11 y4 s 12 , 2 0 0
23
½ y s 4 y s5 q s6 q s 7 ,
s 4 q s5 y s 8 y s 9
s 3 q s 4 y s 7 y s 8 y s 9 y s10 ,
ž 0
0
4
y2
y2
4 /
2 0 0
24
½ s 3 q s 4 q s 7 q s 8 q2 s 9 q2 s10 y s11 y s12 ,
s11 y s12 ž 0
0
2
0
0
2 /
whose intersection matrix is
4 y1 1
1ž
y Ž j i , j j . s y1 2
y1
y1 .
2 / Ž 3.37 .
Example (No. 25)

We have the 7-branes Ž A7 . A B C B C. The junctions orthogonal to T s A 6 are
j 1 s 2 s 8 y s 11 y s 12 , j 2 s s 10 y s 12 , Ž 3.38 .
Table 6
y2 s 8 q s11 q s12 , 4 y1
25 ½ y s10 q s12 ž y1 2 /
y s 9 y s10 q s11 q s12 , 2 0
26 ½ y s 9 q s10 ž 0 2 /
y s10 q s11 , 2 y1
27 ½ y s11 q s12 ž y1 2 /
y s10 q s12 , 2 y1
28 ½ y s 7 y s 8 q s11 q s 12 ž y1 2 /
y s 7 y s 8 y s 9 q s 11 q2 s12 , 2 0
29 ½ s 7 q s 8 y2 s 9 ž 0 6 /
y s 8 y s 9 q s11 q s12 , 2 0
30 ½ y2 s10 q s11 q s12 ž 0 4 /
2 s6 q2 s 7 q2 s 8 q2 s10 y3s11 y5 s12 , 8 y1
31 ½ y s10 q s12 ž y1 2 /
y s 7 y s 8 y s 9 q s 10 q s11 q s12 , 4 y2
32 ½ y2 s10 q s11 q s12 ž y2 4 /
y s6 y s 7 y2 s 8 y2 s 9 q2 s11 q4 s12 , 6 y2
33 ½ y s6 y s 7 q s 8 q s 9 ž y2 4 /
s 9 q s10 y s11 y s12 , 2 0
34 ½ s 7 q s 8 y s11 y s 12 ž 0 2 /
y s5 y s6 y s 7 y s 8 y s10 q2 s11 q3s12 , 2 0
35 ½ y s10 q s12 ž 0 2 /
y s5 y s6 y s 7 y s 8 q s 9 q s 11 q2 s12 , 4 0
36 ½ s5 q s6 q s 7 q s 8 q2 s 9 y2 s11 y4 s12 ž 0 4 /
y s5 y s6 y s 7 q s 10 q3 s11 y s12 , 2 0
37 ½ s5 q s6 q s 7 y3 s12 ž 0 12 /
y s5 y s6 q s11 q s12 , 2 0
38 ½ y s5 y s6 y s 9 y s10 q2 s12 ž 0 4 /
s 4 q s5 q s6 y s11 y2 s12 , 2 y1
39 ½ s 7 q s 8 q s 9 y s11 y2 s12 ž y1 2 /
y2 s4 y2 s5 y2 s6 y s 9 y s10 q2 s11 q4 s12 , 6 0
40 ½ y s 9 y s10 y s11 q s12 ž 0 6 /
s 8 q s 9 q s10 q s11 y2 s 12 , 4 y2
41 ½ y s6 y s 7 q s10 q s11 ž y2 4 /
s 3 q s 4 q s 7 q s 8 q2 s 9 q2 s10 y s11 y s12 , 2 0
42 ½ s 3 q s 4 y s 7 y s 8 y s 9 y s 10 ž 0 2 /
whose intersection matrix reads
4 y1
y Ž ji , j j . s
y1 ž 2
. / Ž 3.39 .
Table 7
The basis junctions for r s1
43 s11 y s 12 2
44 s10 y s12 2
45 3s 9 y s11 y2 s12 8
46 2 s10 y s11 y s12 4
47 s10 q3s11 y4 s12 14
48 s 9 q s10 y s11 y s 12 2
49 s 3 q . . . q s 7 q2 s 9 q2 s10 y4 s11 q3s12 6
50 2 s 8 q2 s 9 q2 s10 y3s11 y3s12 12
51 s 7 q s 8 q s 9 y s11 y2 s12 2
52 s 8 q s 9 q s10 q s11 y2 s12 4
53 s 9 q s10 q s11 y s 12 6
54 s 7 q s 8 q s 9 q s10 y2 s11 y2 s12 4
55 3s6 q3s 7 q3s 8 q3s 9 y4 s11 y8 s12 20
56 4 s6 q4 s 7 q4 s 8 y3s11 y9s12 30
57 s 7 q s 8 y s 9 y s10 y s 11 y s 12 2
58 s5 q s6 q s 7 q s 8 y s 11 y3s 12 2
59 s 8 q s 9 y3s10 y3s11 q4 s12 12
60 s 7 q s 8 y s11 y s 12 4
61 2 s4 q2 s5 q2 s6 y s10 y s11 y4 s 12 6
Further computations enable us to write down the basis junctions and their intersec-
tions for all the cases of non-Cartan type Ž3.20.. The results are shown in Tables 4–7.
Finally, for the case of rank EŽ K . s 1 ŽNos. 43–61., the lattice L is one-dimensional
and we have determined its generator. The result is presented in Table 7. Thus, we have
completed the construction of the lattice L listed in Table 2 in terms of string junctions,
producing the classification Tables 3–7.
4. Weight lattice and torsions
4.1. Weak integrality of inÕariant charges
The Mordell–Weil group EŽ K . consists of the free part EŽ K .rEŽ K . tor and the
torsion part EŽ K . tor . According to Theorem 9.2 in Ref. w12x the former should be
isomorphic to the weight lattice L) of L s T H in E8 .
We first would like to discuss the property of L) in view of junctions. Naively, the
weight junctions v i can be obtained as the dual to the basis junctions j i , i.e.
v i s Ž Cy1 . i j j j , where C is the intersection matrix. Though the invariant charges Q i of
these dual basis junctions v i are in general not integer, we can remedy this partially by
adding null junctions.
Example (No. 25)

The basis junctions of L read
j 1 s Ž 0,0,0,0,0,0,0,y 2,0,0,1,1 . , j 2 s Ž 0,0,0,0,0,0,0,0,0,y 1,0,1 . , Ž 4.1 .
for which the dual basis junctions v 1 , v 2 are obtained as
v 1 s 17 Ž 2 j 1 q j 2 q 3 d 2 . s Ž y 37 ,y 37 ,y 37 ,y 37 ,y 37 ,y 37 ,y 37 ,y 1,3,2,y 1,0 . ,
v 2 s 17 Ž j 1 q 4 j 2 q 5d 2 . s Ž y 57 ,y 57 ,y 57 ,y 57 ,y 57 ,y 57 ,y 57 ,y 1,5,3,y 2,0 . ,
Ž 4.2 .
where d 2 is given by Ž3.18..
Example (No. 7)
We have L s D4 [ A1. Since L is generated by Ž3.33., the basis of the weight lattice
is found to be
v 1 s Ž 0,0, 12 , 12 , 12 , 12 ,1,1,y 3,y 2,1,0 . , v 2 s Ž 0,0,1,1,1,1,1,1,y 5,y 3,2,0 . ,
v 3 s Ž y 12 ,y 12 ,0,0, 12 , 12 ,0,0,0,0,0,0 . , v 4 s Ž y 12 ,y 12 , 12 , 12 ,0,0,0,0,0,0,0,0 . ,
v 5 s Ž 21 , 21 , 21 , 21 , 21 , 21 ,1,0,y 3,y 2,1,0 . . Ž 4.3 .

These examples show that some of the charges Q i still remain fractional. However,
the fractionality is restricted by a certain condition which we call weak integrality. In
general, a junction orthogonal to the lattice T can be represented by using tadpole
junctions Ž J2 in Fig. 1. that go around the collapsed branes and do not touch them
directly. Accordingly, for each of the collapsed branes, say Ž X w p i , q i x PPP X w p j, q j x ., we
require the integrality not for indiÕidual charges Q k Ž i F k F j ., but for their total Ž p,q .
charges. Namely
Ž p,q . s Q i Ž pi ,qi . q . . . qQ j Ž pj ,q j . g Z 2 . Ž 4.4 .

We call this condition weak integrality.
We now wish to point out that the junctions orthogonal to T subject to this weak
integrality condition form the full Mordell–Weil lattice EŽ K . rather than the lattice L
Žcalled the narrow Mordell–Weil lattice in Refs. w12,13x.. To see this, we have first
checked that the torsion free part of EŽ K . is the dual lattice L) of L by explicitly
constructing the dual basis in terms of string junctions. The dual basis junctions have
been worked out for all the cases Nos. 1–61. They indeed satisfy the weak integrality
condition.
4.2. Torsions as fractional null junctions
Let us next consider the torsion part EŽ K . tor . By definition, a section P g EŽ K . is a

torsion if and only if mP s O for some m g Z ) 0 . Then its height pairing vanishes
Ž P,Q . s 0 for any Q g EŽ K . since mŽ P,Q . s Ž O,Q . s 0. Hence P corresponds to a
null junction. We know there are two independent null junctions d 1 , d 2 . By virtue of
the Hanany–Witten effect, these null junctions can be transformed to the canonical form
with integer charges Q i , see Ž3.18. for instance. We call such null junctions strongly
integral ones. As we remarked above, a fractional junction which is integral only in
weak sense is also allowed for EŽ K .. It is then shown that the weakly integral null
junctions, which we refer to as fractional loop junctions, are identified as the torsions
Žmodulo strongly integral null junctions.. Our idea is explained by presenting some
examples explicitly.
Example (No. 73)

The brane configuration consists of two collapsed branes Ž A4 B C .Ž A 4 B C .. Each of
them supports one of the D4 components in T s D4 [ D4 . The junctions orthogonal to T
become null junctions of the form
j s Q1 Ž s 1 q s 2 q s 3 q s 4 . y Ž 2 Q1 q Q12 . s5 y Q12 s6 y Q1 Ž s 7 q s 8 q s 9 q s 10 .
q Ž 2 Q1 q Q12 . s 11 q Q12 s 12 . Ž 4.5 .

Thus the Ž p,q . charges exchanged between the two D4 components are p s 2Ž Q1 y Q12 .
and q s 2Ž Q1 q Q12 .. The integrality condition on Ž p,q . requires 2 Q1 ' 2 Q12 ' 0.
Hence we obtain the Ž Zr2 Z . 2 torsion. The generators of the torsion junctions are
j s Ž 0,0,0,0,y 12 ,y 12 ,0,0,0,0, 12 , 12 . ,
j X s Ž y 12 ,y 12 ,y 12 ,y 12 , 32 , 12 , 12 , 12 , 12 , 12 ,y 32 ,y 12 . , Ž 4.6 .
which in fact obey j 2 s j X 2 s 0 and 2 j s d 1 , 2 j X s d 2 . They are represented as the
fractional loop junctions with the charges Ž r, s . s Ž 12 ,0. for j and Ž r, s . s Ž0, 12 . for j X 4.
Example (No. 66)

We have the 7-branes Ž A6 .Ž B 3 .Ž X w1,
2
y 2x C and the torsion group reads Zr6Z. This
.
is generated by the torsion junction
j s Ž y 16 ,y 16 ,y 16 ,y 16 ,y 16 ,y 16 , 23 , 23 , 23 ,y 12 ,y 12 ,0 . , Ž 4.7 .
which obeys j 2 s 0 and 6 j s d 1 q d 2 . As a fractional loop junction this carries the
charges Ž r, s . s Ž 16 , 16 ..
Example (No. 74)

We have the 7-branes Ž A4 .Ž B 4 .Ž X w0,1x
2 .Ž 2 .
X w2,1x and the torsion group reads Zr4Z [
Zr2 Z. This is generated by the torsion junctions
j s Ž y 14 ,y 14 ,y 14 ,y 14 , 34 , 34 , 34 , 34 ,2,2,y 12 ,y 12 . ,
j X s Ž 0,0,0,0,y 12 ,y 12 ,y 12 ,y 12 ,y 32 ,y 32 , 12 , 12 . Ž 4.8 .
which obey j 2 s j X 2 s 0 and 4 j s d 2 , 2 j X s d 1. As fractional loop junctions they carry
the charges Ž r, s . s Ž0, 14 . for j and Ž r, s . s Ž 12 ,0. for j X .
4
We note again that the charges Q i for d 1 , d 2 depend on the brane configurations.
Following this procedure we can express the generators of all the torsion groups in
Table 2 as the fractional loop junctions with the Ž r, s . charges. Our result is
Zr2 Z : Ž r , s . s Ž 12 , 12 . , Nos. 13, 21, 28, 34, 35, 38, 44, 48, 53, 54, 59, 64,
65;
Zr2 Z : Ž r , s . s Ž 0, 12 . , No. 24;
Zr2 Z : Ž r , s . s Ž 21 ,0 . , Nos. 41, 52;
1 1
Zr3 Z : Ž r , s. s Ž , . ,
3 3 Nos. 39, 51, 61, 63;
Zr3 Z : Ž r , s . s Ž 0, 13 . , No. 69;
1 1
Zr4 Z : Ž r , s. s Ž , . ,
4 4 Nos. 58, 70;
Zr4 Z : Ž r , s . s Ž 14 , 12 . , No. 72;
1 1
Zr5 Z : Ž r , s. s Ž , . ,
5 5 No. 67;
Zr6 Z : Ž r , s . s Ž 16 , 16 . , No. 66;
2
Ž Zr2 Z . : Ž r , s . s Ž 12 ,0 . , Ž 0, 12 . , Nos. 42, 57, 60, 71, 73;
2
Ž Zr3 Z . : Ž r , s . s Ž 13 ,0 . , Ž 0, 13 . , No. 68;
1 1
Zr4Z [ Zr2 Z : Ž r , s . s Ž 0, . , Ž ,0 . ,
4 2 No. 74.
The result for the torsion group we have obtained from the junction consideration is in
agreement with that in Ref. w13x.
5. Discussion
In this paper, we have systematically studied the structure of singularities, Mordell–

Weil lattices and torsions of a rational elliptic surface by making use of the 7-brane-
junction technology. Our results are in nice agreement with Oguiso–Shioda’s Main
Theorem in Ref. w13x. Consequently we found explicit correspondence between sections
and junctions, which is summarized in Table 8.
Though we have restricted ourselves to the case of rational elliptic surfaces, general-
ization to other elliptic surfaces is clear and stated as follows 5 : For a general elliptic
surface p:S ™ C over a curve C, the lattice of strongly integral tadpole junctions can be
identified with the cohomology group H 1 Ž C, R 1 p) Z . whose H 1,1 part is isomorphic to
the narrow Mordell–Weil group of S. ŽFor a rational elliptic surface, all the elements in
H 1 Ž C, R 1 p) Z . belong to its H 1,1 part, however, it is not so in general.. Since the sheaf
R 1 p) Z is a local system whose fiber at x g C is H 1 Ž py1 Ž x .,Z . s Z a [ Z b , the
cohomology H 1 Ž C, R 1 p) Z . can be evaluated by using group cohomology associated
with the monodromy representation r :p 1Ž C y singularity4. ™ SLŽ2,Z .. The 7-brane-
junction technology may offer an efficient way to calculate this cohomology
H 1 Ž C, R 1 p) Z . and intersections on it.
5
We thank M. Saito for pointing out this interpretation. See Ref. w27x.
Table 8
Sections and junctions
Section Junction
O zero section strongly integral null junctions
EŽ K . tor torsions weakly integral null junctions
L narrow Mordell–Weil lattice strongly integral tadpole junctions
EŽ K . Mordell–Weil lattice weakly integral tadpole junctions
One may apply our results to gain a physical understanding of torsion groups which
play an important role to determine the gauge group rather than the gauge algebra w11x.
Since the structure of Mordell–Weil lattices is related to the Wilson lines on the
heterotic side under F-theoryrheterotic duality, it will be interesting to think of the issue
from the heterotic string point of view.
Another application is found when we consider stable non-BPS states in F-theory. It
has recently been recognized that, in string theory, there exist solitonic states which are
stable but not BPS w28,29x. These states are the lightest ones which carry certain
conserved charges, i.e. there exist no other BPS or non-BPS states of lower mass having
the same charges. Thus their stability is ensured by charge conservation. In Ref. w2x the
analysis of stable non-BPS states in F-theory on K 3 was initiated. As we will see now,
the structure of the Mordell–Weil lattice determines the 7-brane configurations support-
ing non-BPS junctions which could be candidates for stable non-BPS states in a region
of the moduli space of F-theory on K 3.
Let us consider the region of the moduli space where 24 7-branes for an elliptic K 3
are split into two Eˆ9 w7x. This corresponds to the so-called stable degeneration of K 3
w8,9x. Two Eˆ9 brane configurations are properly isolated from each other in the sense of
Ref. w2x. Thus we may focus on string junctions stretched on a single Eˆ9 to analyze
non-BPS states. The BPS junctions J BPS have to obey the holomorphy condition which
2
is stated as J BPS G y2 w30,31x. Thus junctions J with J 2 - y2 are non-BPS.
Inspecting Table 7 we first observe that the basis junctions of one-dimensional L, except
for the case L s A1 , are all non-BPS. As for the higher rank EŽ K ., we also have
non-BPS basis junctions in Nos. 12, 17, 19, 20, 22, 23, 25, 29–33, 36–38, &
40 and 41.
We note that the cases No. 45 Ž E˜1 ., No. 46 Ž D 1 . and No. 25 Ž E2 . have already
appeared in Ref. w2x. Each non-BPS basis junction, which we denote as j uŽ1. , generates
the UŽ1. symmetry associated to the one-dimensional lattice. It is clear that any junction
can be written as
J s j H qnj uŽ1. , Ž 5.1 .
H
where n g Z and j stands for the orthogonal components of J with respect to the
UŽ1. direction. Then the self-intersection is obtained as J 2 s j H 2 q n 2 j uŽ1.
2
. Since
H 2 2 2
j F 0 and j uŽ1. - y2 we have J - y2 for any n / 0. Therefore the junctions with
non-vanishing component along j uŽ1. represent non-BPS states. Among these non-BPS
states there could be stable states against decay. For instance, non-BPS junctions in Nos.
45, 46 and 25 are shown to & be stable by the charge conservation argument since the
configurations E˜1 , D 1 and E2 are isolated w2x. Beyond the isolated configurations,
however, identifying such stable states requires a detailed dynamical analysis, as
discussed in Section 5 of Ref. w2x, which is beyond the scope of this paper.
Acknowledgements
We would like to thank M. Noumi and M. Saito for valuable discussions. We also
wish to thank T. Shioda and K. Oguiso for ascertaining some corrections in the table in
Ref. w13x and Y. Ohtake for preparing the figures in this paper. The work of SKY was
supported in part by Grant-in-Aid for Scientific Research on Priority Area 707 ‘‘Super-
symmetry and Unified Theory of Elementary Particles’’, Japan Ministry of Education,
Science and Culture.
References
w1x Y. Yamada, S.-K. Yang, Affine 7-brane backgrounds and five-dimensional EN theories on S 1 , Nucl.
Phys. B, to be published; hep-thr9907134.
w2x A. Sen, B. Zwiebach, Stable non-BPS states in F-theory, hep-thr9907164.
w3x C. Vafa, Nucl. Phys. B 469 Ž1996. 403. hep-thr9602022.
w4x A. Sen, Nucl. Phys. B 475 Ž1996. 562. hep-thr9605150.
w5x D.R. Morrison, C. Vafa, Nucl. Phys. B 473 Ž1996. 74. hep-thr9602114.
w6x D.R. Morrison, C. Vafa, Nucl. Phys. B 476 Ž1996. 437. hep-thr9603161.
w7x O. DeWolfe, T. Hauer, A. Iqbal, B. Zwiebach, Uncovering infinite symmetries on w p,q x 7-branes:
Kac–Moody algebras and beyond, hep-thr9812209.
w8x R. Friedman, J. Morgan, E. Witten, Comm. Math. Phys. 187 Ž1997. 679. hep-thr9701162.
w9x P.S. Aspinwall, D.R. Morrison, Nucl. Phys. B 503 Ž1997. 533. hep-thr9705104.
w10x P.S. Aspinwall, J. High Energy Phys. 9804 Ž1998. 019. hep-thr9802194.
w11x P.S. Aspinwall, D.R. Morrison, J. High Energy Phys. 9807 Ž1998. 012. hep-thr9805206.
w12x T. Shioda, Comment. Math. Univ. St. Pauli. 39 Ž1990. 211.
w13x K. Oguiso, T. Shioda, Comment. Math. Univ. St. Pauli. 40 Ž1991. 83.
w14x T. Shioda, Proc. Jap. Acad. A 68 Ž1992. 251.
w15x J.H. Silverman, J. Tate, Rational points on elliptic curves, Undergraduate Texts in Mathematics ŽSpringer,
Berlin, 1992..
w16x K. Kodaira, Ann. Math. 77 Ž1963. 563.
w17x K. Kodaira, Ann. Math. 78 Ž1963. 1.
w18x N. Seiberg, E. Witten, Nucl. Phys. B 431 Ž1994. 484. hep-thr9408099.
w19x J.A. Minahan, D. Nemeschansky, Nucl. Phys. B 482 Ž1996. 142. hep-thr9608047.
w20x J.A. Minahan, D. Nemeschansky, Nucl. Phys. B 489 Ž1997. 24. hep-thr9610076.
w21x M. Noguchi, S. Terashima, S.-K. Yang, Nucl. Phys. B 556 Ž1999. 115. hep-thr9903215.
w22x J.A. Minahan, D. Nemeschansky, N.P. Warner, Nucl. Phys. B 508 Ž1997. 64. hep-thr9705237.
w23x O. DeWolfe, B. Zwiebach, Nucl. Phys. B 541 Ž1999. 509. hep-thr9804210.
w24x O. DeWolfe, T. Hauer, A. Iqbal, B. Zwiebach, Uncovering the symmetries on w p,q x 7-branes: Beyond
the Kodaira classification, hep-thr9812028.
w25x M.R. Gaberdiel, T. Hauer, B. Zwiebach, Nucl. Phys. B 525 Ž1998. 117. hep-thr9801205.
w26x R. Miranda, Math. Ann. 255 Ž1981. 379.
w27x D. Cox, S. Zucker, Invent. Math. 53 Ž1979. 1.
w28x A. Sen, J. High Energy Phys. 9806 Ž1998. 007. hep-thr9803194.
w30x A. Mikhailov, N. Nekrasov, S. Sethi, Nucl. Phys. B 531 Ž1998. 345. hep-thr9803142.
w31x O. DeWolfe, T. Hauer, A. Iqbal, B. Zwiebach, Nucl. Phys. B 534 Ž1998. 261. hep-thr9805220.
Non-BPS states and heterotic – type I X duality

´
Tathagata Dasgupta, Bogdan Stefanski, Jr.
Department of Applied Mathematics and Theoretical Physics, UniÕersity of Cambridge, SilÕer Street,
Cambridge, CB3 9EW, UK
Received 9 November 1999; received in revised form 31 January 2000
Abstract
There are two families of non-BPS bi-spinors in the perturbative spectrum of the nine-dimen-
sional heterotic string charged under the gauge group SO Ž16. = SO Ž16.. The relation between
these perturbative non-BPS states and certain non-perturbative non-BPS D-brane states of the dual
type IX theory is exhibited. The relevant branes include a Z 2 charged non-BPS D-string, and a
bound state of such a D-string with a fundamental string. The domains of stability of these states
as well as their decay products in both theories are determined and shown to agree with the duality
map. q 2000 Elsevier Science B.V. All rights reserved.
1. Introduction
Over the past couple of years stable non-BPS states and D-branes have opened a new
direction to our understanding of string theory. Reviews of these developments can be
found in w1–3x. Two approaches have been used to construct and analyse non-BPS
D-branes. In the first approach w4–8x non-BPS D-branes are constructed by tachyon
condensation as bound states of brane-anti-brane pairs. This construction permits for a
classification of D-brane charges in terms of K-theory w9x. The second approach uses the
boundary state formalism w10–14x, to describe D-branes as coherent states in the closed
string theory satisfying a number of consistency conditions w15–18x. Since this latter
approach provides an explicit boundary conformal field theory description of non-BPS
D-branes, we use this second approach.
SO Ž32. heterotic string theory is conjectured to be non-perturbatively dual to type I
string theory in ten dimensions w19x. It should therefore be possible to identify suitable
perturbative non-BPS states of the heterotic string with non-BPS D-brane states in the
type I theory. The most familiar example of stable non-BPS states are the states
E-mail addresses: t.dasgupta@damtp.cam.ac.uk ŽT. Dasgupta., b.stefanski@damtp.cam.ac.uk

´
ŽB. Stefanski, Jr...
PII: S 0 5 5 0 - 3 2 1 3 Ž 0 0 . 0 0 0 3 9 - 0
96 ´ Jr.r Nuclear Physics B 572 (2000) 95–111
T. Dasgupta, B. Stefanski,
transforming in the spinor representation of the gauge group of the SO Ž32. heterotic
string theory. They arise in the first excited level and are absolutely stable due to charge
conservation but are not BPS as N s 1 supersymmetry algebra has no central charges.
The dual state in the type I theory is a stable Z 2-valued non-BPS D-particle w7,9x, which
can be described as a tachyonic kink solution on the D1- D1 pair w5x.
In this paper we test the S-duality between the heterotic string theory on S 1 with
gauge group SO Ž16. = SO Ž16. and the type I X theory on S 1 with the same gauge group.
In such a configuration the two heterotic theories are T-dual to each other w20,21x. The
type I X theory is an orientifold of type IIA by I9 V , where I9 reverses the sign of x 9
and V is the world-sheet parity operator. This orientifold can be thought of as two
O8-planes w22–24x at x 9 s 0 and x 9 s p R 9 , which in the SO Ž16. = SO Ž16. point in
moduli space has eight D8-branes, and their images placed on each of the O8-planes to
cancel the tadpole locally.1 The positions of the D8-branes on the interval correspond in
the T-dual type I theory to a Wilson line.
The conserved charges of the type I string theory are Kaluza–Klein ŽKK. momentum
and D-string winding number, and those of type I X are winding and D-particle numbers.
The duality map relating various parameters between type I X and heterotic theory is
given by w26x
1 R IX X R IX
Rh s , lh s , GMh N s GMI N , Ž 1.1 .
(R I
X lIX lIX lIX
where GM N is the nine-dimensional metric Ž M, N s 0, . . . ,8., l h and l I X are the

heterotic and type I X coupling constants, and R h and R I X are the radii of the circle when
measured in the heterotic and type I X metric, respectively. From this relations one can
see that the heterotic KK momentum is mapped to type I X winding, and heterotic
winding is mapped to the type I X D-particle number.
Under the unbroken SO Ž16. = SO Ž16. gauge group the states in the spinor conjugacy
class of SO Ž32. representations decompose into two sets of bi-spinors, denoted A and
B. In the type I X theory, A are shown to correspond to a Z 2-valued non-BPS D-string
stretching along the interval which is T-dual to the Z 2-valued D-particle on S 1 in type I
theory. We show that these bi-spinors are unstable against a decay into two single spinor
states for radii less than a critical radius R h . In the type I X theory, this corresponds to the
appearance of a tachyon in the open string spectrum with endpoints on the Z 2-valued
D-string. For radii greater than a certain critical radius R I X , the D-string decays into a
D0-brane at one O8-plane and an anti-D0-brane on the other O8-plane. The Z 2 charge
of the decaying D-string is encoded in the Z 2 choice of locations for the decay products.
We determine the masses of the various states and show that at the critical radius they
are equal to one another, indicating that the deformation is marginal. Further we show
that the critical radii R h and R I X agree with Ž1.1. qualitatively. This provides a test of
the type I X-heterotic duality beyond the constraints of BPS states.
The second class of bi-spinors B can be thought of as bound states of the A
bi-spinors and certain bi-vectors. In type I X the B bi-spinors correspond to a bound state
of a Z 2-valued non-BPS D-string and F-string both stretching along the interval. We
1
For a non-technical review of Z 2 orientifolds of type II theories see for example, Ref. w25x.
´ Jr.r Nuclear Physics B 572 (2000) 95–111
T. Dasgupta, B. Stefanski, 97
show that such a bound state does indeed exist in type I X . Although the B bi-spinor Žthe
ŽF, D. bound state. is stable against decay into an A bi-spinor Žthe non-BPS D-string.
and the bi-vector Žthe fundamental string. as we will see, it is not always stable against
decay into the two different single spinor states Ža D0– D0 pair. and a bi-vector Ža
fundamental string stretching along x 9 .. The mass of the bound state and domains of
stability of the heterotic and type I X states are computed. The regimes of stability in the
two theories are shown to be qualitatively the same. In the T-dual picture, the ŽF, D.
bound state corresponds to a Z 2-valued D-particle with a constant velocity along S 1 as
the effect of adding a fundamental string to the D-string of type I X is to give the type I
D-particle a KK momentum. We show that, unlike in the previous case, presently the
non-BPS mass of the ŽF, D. bound state does not match with its decay product at the
critical radius. This indicates that the transition from the non-BPS bound state to the
D-particle pair and an F-string is not a marginal deformation.
The paper is organised as follows. In Section 2 we briefly review the heterotic string
on S 1 and point out that there are two kind of non-BPS bi-spinors arising from the
unbroken gauge group SO Ž16. = SO Ž16.. The duality map Ž1.1. is tested by comparing
the masses of certain BPS states, namely bulk D-particle and fractional D-particle and
anti-D-particle, in Section 3. In Section 4 the non-BPS bi-spinor states are analysed. In
Section 5 we give the type I X analysis following boundary state approach. We conclude
and raise some open problems in Section 6.
2. Review of heterotic string on S 1
The left- and right-moving momenta of an SO Ž32. heterotic string compactified on

S 1 are given by w21x 2
ph wh R h ph wh R h
Ž pL < pR . s ž V q Awh ,
Rh
q
2 Rh
y
2 / , Ž 2.1 .
where V is an element of the internal G 16 lattice, R h and p h are the compactification

radius and physical momentum, respectively, while wh g Z is the winding number. In
terms of the background gauge field, A Žor Wilson line. p h is given by
p h s n h y V P A y w h A2r2 , Ž 2.2 .
where n h g Z denotes the Kaluza–Klein ŽKK. momentum. Physical states satisfy the
level-matching condition
pL2 q 2 Ž NL y 1 . s pR2 q 2 Ž NR y c R . , Ž 2.3 .
where NL and NR are left- and right-moving excitation numbers, and c R s 0 and 1r2 for
the right-moving fermions in the periodic ŽR. and anti-periodic ŽNS. sectors, respec-
tively. For BPS states all the right-moving oscillators should be in the ground state:
NR s c R w28x. The heterotic mass formula is given by
m2h s 12 pL2 q Ž NL y 1 . q 12 pR2 q Ž NR y c R . , Ž 2.4 .
2 X X
Throughout the paper we follow the convention a h s 2, a I X s1.
which using the level-matching condition Ž2.3. becomes

m2h s pR2 q 2 Ž NR y c R . . Ž 2.5 .
2
The states with NL s 1 and V s 0 are KK excitations of either the gravity multiplet or
one of the vector multiplets associated with the Cartan subalgebra of the gauge group.
But there are additional massless states having NL s 0 and pL2 s 2. In the zero winding
sector Ž wh s 0. we have V 2 s 2 and p h s 0, hence states ŽV. are roots of SO Ž32.. For a
non-trivial Wilson line A, these states correspond to the roots of the unbroken subgroup
of SO Ž32.. Without a Wilson line, the heterotic string has a R ™ 1rR symmetry giving
rise to an enhanced SUŽ2. gauge symmetry at the self-dual point. More generally, for a
non-trivial Wilson line, A, the winding sectors give additional massless states at some
critical radius. In this paper we consider only the Wilson line, A s Ž0 8 ,Ž 12 . 8 ., which
corresponds to an SO Ž16. = SO Ž16. gauge group and critical radius zero. In type I X
theory this is equivalent to placing eight D8-branes with mirrors at each O8-plane giving
rise to a constant dilaton and metric background in the bulk w26,29x.
The 2 ny 1-dimensional spinor representation is the smallest representation of the
ten-dimensional spinor conjugacy class of SO Ž32.. In nine dimensions, under the Wilson
line A s Ž0 8 ;Ž 12 . 8 . with unbroken gauge group SO Ž16. = SO Ž16., this decomposes as
X X
2 15 ™ Ž 2 7 ,2 7 . [ Ž Ž 2 7 . , Ž 2 7 . . , Ž 2.6 .
giving two kinds of bi-spinors with
1 1 1
Ž A. : ž^" ,`. . . ," ‘q ’; " 2 , . . . ," 12 7
/ g Ž 2 ,2 . ,7
2
_ 2
even no . of ^ ` _even no . of ‘q ’
1 1 1 7 X 7 X
Ž B. : ž^" ,`
. . . ," ‘q ’; " 2 , . . . ," 12 / g Ž Ž 2 . ,Ž 2 . . .
2
_ 2
odd no . of ^ ` _odd no . of ‘q ’
Ž 2.7 .
In later sections we will show that in type I X theory, the first set of bi-spinors
corresponds to a Z 2-charged non-BPS D-string stretching along the interval with
end-points at the two O8-planes. The second set corresponds to a bound state of the
Z 2-charged non-BPS D-string with an F-string, both stretching along the interval.
3. Single spinor states
In this section we describe heterotic single spinor states that are charged under one of
the SO Ž16.’s. In type I X , they turn out to be fractional D-particles Žor D0 f . stuck on an
O8-plane. Their masses and bulk RR charges are half of those of the bulk D-particles.
A bulk D-particle is dual to a heterotic string with non-trivial winding, w h s 2 and
the physical momentum, p h s 0. As V q Awh s 0 with V s Ž0 8 ;Žy1. 8 ., these states are
not charged under either of the SO Ž16.’s. The level-matching condition implies that
NR s c R and NL s 1. Hence these are BPS states and are KK excitations of either the
gravity multiplet or the vector multiplets in the Cartan subalgebra. The heterotic mass
formula gives
m h Ž D0 bulk . s R h . Ž 3.1 .
Using Ž1.1. the corresponding type I X bulk D-particle mass turns out to be
1
m I X Ž D0 bulk . s l1r2
h m h Ž D0 bulk . s . Ž 3.2 .
lIX
Consider next the single spinor states
8 8
V1 s " 12
ž^`_
Ž . ‘q ’;0
8
/ , V2 s 0 8 ; Ž " 12 .
ž / . Ž 3.3 .
even no . of ^`_even no . of ‘q ’
where V1 is of the form V q Awh , with V s ŽŽ" 12 . 8 ;Žy 12 . 8 . g G 16 , with an even
number of ‘q’ signs, w h s 1 and vanishing physical momentum: p h s 0 Žfor KK
momentum, n h s y1.. The type I X D-particle number, is given by w29x n I X s whr2 s
1r2. The level-matching condition gives NL s 0 and NR s c R , thus V1 is a BPS state
whose mass is
Rh
m h Ž V1 . s . Ž 3.4 .
2
Using the duality relations Ž1.1., the type I X mass is
1
m I X Ž V1 . s , Ž 3.5 .
2 lIX
which is half of the D-particle’s mass. This shows that V1 corresponds to a fractional
D-particle stuck on the x 9 s 0 O8-plane.
Similarly, V2 is of the form V q Awh , where V s Ž0 8 ;12 n,0 8y2 n . g G 16 with wh s
y1, and n s 0, . . . ,4. The level-matching condition implies that the state is BPS
Ž NR s c R ., with vanishing physical momentum, p h s 0 Ži.e. KK momentum, n h s n y 1..
The mass formula gives
Rh
m h Ž V2 . s , Ž 3.6 .
2
which in type I X units is
1
m I X Ž V2 . s . Ž 3.7 .
2 lIX
This shows that V2 corresponds to a fractional anti-D-particle stuck at the x 9 s p R 9

O8-plane.
4. Bi-spinor states
In this section we discuss in some detail the bi-spinors A and B. In particular we

demonstrate that they are non-BPS states, which are stable in certain regions of the
moduli space.
4.1. Bi-spinors in A
The lightest states in A come from the zero winding sector. Since A P V g Z we
choose p h s 0. For NL s 0 the level-matching condition implies that the states are
non-BPS: NR s 1 q c R . The mass of the states with trivial winding, vanishing momen-
tum, NL s 0 and NR s 1 q c R is
m h Ž V g Ž 128 ,128 . . s '2 . Ž 4.1 .

Note that there could be BPS states satisfying NR s c R with wh s "1, p h s 0. But for
such winding numbers the modified lattice vector Ž V q Awh ., does not belong to G 16 .
The bi-spinors A are charged under both the SO Ž16.’s. In fact these non-BPS states
have the same charges as V1 and V2 , the single spinor states discussed in Section 3. This
suggests that a decay process into the BPS single spinor states is possible. Since the
mass of the single spinor states is radius dependent the decay too will depend on the
radius. As the mass of the single spinor states is R hr2, the bi-spinor states A decay into
the single spinor states V1 and V2 for
R h - '2 . Ž 4.2 .
In the following section we construct a D-brane state which corresponds to A. This
turns out to be a Z 2 D-string stretching along the interval. This D-string is the T-dual of
the Z 2 D-particle of type I.
4.2. Bi-spinors in B
The states, V X g Ž128X ,128X . Ži.e. in sector B . have an odd number of q 12 ’s in both
the first and second eight entries. Hence they can be expressed as sum of a state,
V g Ž128,128. Ži.e. in sector A . and a bi-vector state, V bv g Ž16 ,16.:
1
ž
^ `
"
2
, . . . ," 12
_odd no . of
‘q ’; " 2
^ ` _
2
1
, . . . ," 1
odd no . of
‘q ’
/
1
s "
ž
^ ` 2
, . . . ," 12
_even no . of
‘q ’; " 2
^ ` _
2
1
, . . . ," 1
even no . of
‘q ’
/ " Ž 1,0 7 ;" 1,0 7 . .
Ž 4.3 .
In type I X the state " Ž 1,0 7 ;" 1,0 7 . corresponds to a fundamental string stretching
along the interval. The overall sign corresponds to the orientation of the string.
The lightest states in B are those with wh s 0. As A P V g Z q 1r2, these have
p h s "1r2. For NL s 0 the level matching condition implies that the states are
non-BPS: NR s 1 q c R . Here 2 n q 1 Ž n s 0, . . . ,3. is the number of q 12 ’s in the last
eight entries of V g Ž128,128.. The mass formula for such states is
1r2
X X X
1
m h Ž V g Ž 128 ,128 . . s
ž 4 R 2h
q2
/ . Ž 4.4 .
On the other hand, the bi-vector states, V bv with w h s 0, NL s 0 are BPS Ž NR s c R .. The
mass of these BPS states with physical momentum, p h s "1r2 Žas A P V s "1r2. is
given by
1
m h Ž V bv g Ž 16 ,16 . . s . Ž 4.5 .
2 Rh
Since m2h Ž V g Ž 128 ,128 . . s 2, we have
m h Ž V X g Ž 128X ,128X . . - m h Ž V g Ž 128 ,128 . . q m h Ž V bv g Ž 16 ,16 . . , Ž 4.6 .
for a finite radius. Hence the states in B are bound states of states in A and the
bi-vector states Ž4.3.. Although the bi-spinor, B is stable against a decay into the
bi-spinor A and a bi-vector state, it is not always stable against decay into the single
spinor states, V1 and V2 , and a bi-vector state. In particular this bi-vector is unstable for
Rh -1 . Ž 4.7 .
In Section 5 we show that in type I X theory the bi-spinor B corresponds to a bound state
of a non-BPS Z 2 D-string with an F-string, both stretching along the interval. Equiva-
lently, the states in B describe a non-BPS D-string with a constant electric field on its
world-sheet w27x. This Z 2 D-string with constant electric field becomes under T-duality,
a Z 2 D-particle with constant velocity along S 1 w37,38x.
5. Type I X analysis
The states in B describe a non-BPS D-string with a constant electric field on its
world-sheet. Such a bound state will exist as long as the constant electric field on a Z 2
D-string is invariant under V I9 . If the gauge field is invariant under V I9 the ŽF, D.
bound state can be stabilised from an unstable non-BPS D-string with constant electric
flux in type IIA by orientifolding. The NSNS B field equation of motion has F s F q B
as a source term w12–14,27,32x, where F s dA is the gauge field strength. Recall that in
type I, the gauge field vertex operator has tangential derivatives of the form dX irdt
which are odd under V , while the transverse scalars have normal derivatives. As a result
the gauge field is projected out while the transverse scalars survive.3 In type I X theory,
on the other hand, the projection is V I9 . As a result the gauge field component along
S 1 survives the projection. Similarily, while the NSNS two-form B is odd under V and
hence projected out in type I, Bm9 is also odd under I9 and thus survives the projection.
This guarantees the gauge invariance of the bound state.
As with most non-BPS configurations w1,18,30x, the stability of the Z 2 D-string, and
its bound state with a fundamental string, depends crucially on the size of the compact
directions. We investigate this dependence by constructing boundary states which
describe the configurations studied presently. In particular we obtain the critical radius at
which the configurations become unstable and compute the tensions of these states.
3
In fact a Z 2 subgroup of the original UŽ1. survives. This describes the GSO projection on the current
algebra fermions in the heterotic string in the type I-heterotic duality context w26x.
The boundary state representing a Z 2 D-string is given by

N
B1 Z 2 :s Ž B1,q :NSNS y B1,y :NSNS . , Ž 5.1 .
2
where N is the normalisation obtained by factorising on an open string partition
function Žsee for example Ref. w18x for more details on obtaining such normalisations.
and
` 1
< B1,k ,h : NSNS s exp
Ý
ns0 n
m
ayn n
Smn a˜ yn q ih Ý
rgNq
1
2
m
cyr Smn c˜yr
n
0
= Ý e i u w 9 < B1,k ,w 9 ,h :Ž0. m < B1,h :ghost . Ž 5.2 .
w9
Here w 9 is the winding number, h s "1, u is a Wilson line and for compactness of
notation we do not write u on the left-hand side of the above equation. The matrix S
encodes the boundary conditions of the D-string and is a 10 = 10 diagonal matrix given
by
S s diag Ž 1,1, . . . ,1,y 1 . . Ž 5.3 .
The D-string is taken to lie along directions x 0 and x 9 with x 9 compactified along a
circle of radius R I X , and we work in the Minkowski metric. As the Z 2 D-string stretches
in directions 0 and 9 while the O8-planes extend in directions 0, . . . ,8 it is not possible
to work in the light-cone gauge and ghost and superghost contributions will have to be
taken into account. These are taken as in Refs. w10,12,31x. The ground state
< B1,k,w 9 ,h :Ž0. carries momentum k in the directions transverse to the D-string and is
unique in the NSNS sector. To obtain a localised D-brane, we have to Fourier transform
the above state,
8
m
< Bp, y,h : s H mŁž
s1
dk m
eik ym
/Ý
w9
e i u w 9 < Bp,k ,w 9 ,h : . Ž 5.4 .
Here y denotes the location of the boundary state along the transverse directions, and is
suppressed throughout as we take it to be y s 0. The normalisation N is equal to that of
a type IIB D-string from which it follows that the tension of the Z 2 D-string is the same
as that of a conventional type II D-string. This is a '2 greater than the tension of a type
I BPS D-string. This follows since the open string partition function for the Z 2 D-string
has no GSO projection but does have the orientifold projection Ž1 q V I9 .r2.
Next consider the strings that end on a D8-brane. The boundary state of a D8-brane is
N8
< B8: NSNS s Ž < B8,q : NSNS y < B8,y : NSNS . ,
2
4 i N8
< B8: RR s Ž < B8,q : RR q < B8,y : RR . ,
2
< B8: s < B8: NSNS q < B8: RR . Ž 5.5 .
Explicitly the NSNS boundary state is as in Eq. Ž5.2. but now the matrix S has eight
entries equal to y1, with only the first and last one equal to 1. The RR sector is not
discussed here as it does not enter into our analysis. Factorisation of the cylinder
diagram on an annulus fixes the normalisation N8 . Strings that stretch between a
D8-brane and the Z 2 D-string have nine ND and one NN boundary conditions. As a
result, the ground state energy in the NS sector is positive, while as always in the R
sector it is zero, thus these are tachyon free.
Next we consider the Mobius ¨ strip diagrams corresponding, in the closed string
channel to the exchange between the D-string and the O8-planes. A crosscap state
representing an O8-plane is given by a coherent state very similar to that of a D8-brane,
the only difference being an extra factor of Žy1. n in the coherent state exponentials
w10,11x. To check that there are no open string tachyons in the theory we compute the
open string partition functions. Since strings stretching between the D-string and the
D8-brane have no tachyons we disregard these in the following. The relevant amplitude
in the closed string channel is given by
2
A1 s dl ² BZ 2 1 < eyl H c < BZ 2 1: q
H Ý ² BZ 1 < eyl H < Ci 8: q ² Ci 8 < eyl H < BZ 1:
2
c c
2
is1
V f 38 Ž q . y f 48 Ž q . `
2
s 2y6 R I X dl ly4
H Ý eylp Ž w 9 R IX .
2p f 18 Ž q. w 9sy`
V f 49 iq Ž . f 1 Ž iq . f 39 Ž iq . f 1 Ž iq .
q
2p
4 Ž y2 y5r2
. Hdl
ž 2y9 r2 f 29 Ž iq . f 3 Ž iq .
y y9r2 9
2 f 2 Ž iq . f 4 Ž iq . /
V 1 dt y1r2
f 38 Ž q˜ . y f 28 Ž q˜ . `
2
s H Ž2 t . Ý ey2 tp Ž m r R IX .
2p 2 2 t f 18 Ž q˜ . msy`
V dt y1 r2
q 2y3r2 H 2 t Ž2 t .
2p
f 39 Ž iq˜ . f 1 Ž iq˜ . f 49 Ž iq˜ . f 1 Ž iq˜ .
= e
ž yip r4
2y9 r2 f 29 Ž iq˜ . f 4 Ž iq˜ .
ye ip r4
2y9r2 f 29 Ž iq˜ . f 3 Ž iq˜ . / , Ž 5.6 .
where q s ey2 p l , t s 1r2 l, q˜ s eyp t and Hc is the closed string Hamiltonian
9 `
Hc s p p 2 q 2 p m
Ý Ý Ž ayn a nm q a˜ yn
m
a˜ nm . q Ý m
r cyr
ž crm q c˜yr
m ˜m
cr /
ms0 ns1 1
rgNq
2
q 2p Cc q ghosts . Ž 5.7 .
The constant Cc is y1.
In order to see if there are open string tachyons we expand in q˜ to suitable order
V 1 dt y1 r2
1 2 1
A1 s H Ž2 t .
2p 2 2 t q˜
Ž 1 q 2 q˜ 2r R . y q˜ q . . . . Ž 5.8 .
It then follows that for R I X ( '2 , the Z 2 D-string is tachyon free and hence stable. In
Eq. Ž5.6. we have fixed the normalisation constants by requiring that the cylinder and
¨
Mobius strip diagrams factorise on suitable open string partition functions. The decay of
the non-BPS D-string into a D0– D0 pair restricted to the orientifold planes is possible
in the region
R I X ) '2 . Ž 5.9 .
This is also confirmed by the fact that, in this region, the classical mass of the D-string,
given below, is bigger than that of two fractional D-particles, given by Ž3.5. and Ž3.7., in
the above region
R IX
m I X Ž D1 non - BPS . s . Ž 5.10 .
'2 l I X
The numerical factor above follows by noting that the tension of the Z 2 D-string is '2
bigger than the tension of a type I BPS D-string. As expected, the corresponding masses
in the two theories are not related by the duality map, since for non-BPS states the
masses are not protected from quantum corrections. In terms of heterotic string theory
the decay corresponds to Ž3.3.. The regimes of stability of the non-BPS state in the two
dual theories, Ž4.2. and Ž5.9., are qualitatively the same, given the duality relation Ž1.1..
Consider next a bound state of the Z 2 D-string with a fundamental string. We show
that the tension of this state is lower than the individual tensions of the D- and F-strings,
thus forming a bound state. Further we will analyse the stability of this state and find a
dependence on the radius. Boundary states with gauge fields were first considered in
Ref. w12x and in the context of D-branes in Refs. w13,14,31,32x. In the NSNS sector the
following changes occur in the boundary state description. The matrix S now becomes
hyF
hqF
Ss
y1
..
.
y1
0 , Ž 5.11 .
where h is the Minkowski metric, there are still eight entries equal to y1 corresponding
to the transverse directions and the field strength F is
0 yf
Fs
ž f 0
.
/ Ž 5.12 .
We have placed the x 0 and x 9 coordinates in positions 1,2 in the matrix. In the above f
is given by w31x
m
fsy , Ž 5.13 .
'n2 q m2
where m s 1, is the number of fundamental strings, and n s 1 is the number of
D-strings. The background gauge field has no effect on ghost or superghost contribu-
tions. In the presence of a gauge field the open string momentum eigenvalues on a circle
become w33x
n 1
pn s . Ž 5.14 .
(
R 1yf 2
The non-compact open string momentum integrals get modified in an analogous fashion.
This can be viewed from the closed string channel as an extra normalisation factor
(y det Ž h q F . s (1 y f 2
, Ž 5.15 .
of the boundary state corresponding to the non-BPS bound state relative to the Z 2
D-string w12x. We are now in a position to compute the amplitude corresponding to A1
for the ŽF, D. bound state. Since S is an orthogonal matrix, the non-zero-mode
contributions to the cylinder amplitude are as before. Further, the non-zero-modes of the
¨
Mobius strip amplitude do not change either as shown in Appendix A. The amplitude is
2
A2 s dl ² B1 Z 2 , F < eyl H c < B1 Z 2 , F : q
H Ý ² B1Z 2
, F < eyl H c < C8 i :
is1
V f 38 Ž q . y f 48 Ž q .
q ² C8 i < eyl H c < B1 Z 2 , F :s 2 y6
R Ž1 y f 2
. Hdl l y4
2p f 18 Ž q .
` V
2
Ž1yf 2 .
= Ý eyl p Ž w 9 R. q
2p
4 Ž y2y5r2 . 1 y f 2 (
w 9sy`
f49 Ž iq . f 1 Ž iq . f 39 Ž iq . f 1 Ž iq .
= dlH ž 2y9 r2 f 29 Ž iq . f 3 Ž iq .
y
2y9r2 f 29 Ž iq . f 4 Ž iq . /
V dt y1r2
f 38 Ž q˜ . y f 28 Ž q˜ . ` 2
s (1yf H Ž2 t . 2
Ý ey2 tp Ž m '
R 1yf 2 .
2p 2t f 18 Ž q˜ . msy`
V dt y1 r2
q
2p
(1 y f 2
2y3r2 H 2t 2 Ž2 t .
f 39 Ž iq˜ . f 1 Ž iq˜ . f 49 Ž iq˜ . f 1 Ž iq˜ .

ž
= eyip r4
2y9 r2 f 29 Ž iq˜ . f 4 Ž iq˜ .
y e ip r4
2y9r2 f 29 Ž iq˜ . f 3 Ž iq˜ . / , Ž 5.16 .
which we again expand to suitable order
V 1 dt y1 r2
1 2 2 1
A2 s
2p 2
(1yf H Ž2 t .
2t
2
q˜
Ž 1 q 2 q˜2r Ž R Ž1yf . . . y 2y1r2 q˜ q . . . .
Ž 5.17 .
(
It then follows that for R I X ( 2 Ž 1 y f 2 . s 2 the ŽF, D. bound state is stable and
tachyon-free. Hence the decay of the non-BPS ŽF, D. bound state into a D0– D0 pair and
a fundamental string is possible in the region
R IX ) 2 . Ž 5.18 .
Again the regimes of stability of the non-BPS state in the two dual theories, Ž4.7. and
Ž5.18., are qualitatively the same, given the duality relation Ž1.1.. The tension of a
general configuration of m F-string and n D-string bound state is Ž m2 q n2 .1r2 times
the D-string tension w31x. Non-threshold bound states are realised if m and n are
relatively prime integers. In our case the non-BPS D-string mass is given by Ž5.10..
Hence the mass of the non-BPS D-string with one unit of electric field Žwith m s n s 1.
is given by
R IX
m I X Ž D1 non - BPS q F . s . Ž 5.19 .
lIX
Unlike the case of the non-BPS D-string, the mass of the non-BPS ŽF, D. bound state is
not same as its decay products Žnamely the D0– D0 pair with Ž8,8. F-string. at the
critical radius Ž5.18.. Here the F-string mass is R I Xr2 obtained from Ž4.5. using duality
map Ž1.1.. There is a loss of energy due to the interaction between the decay products in
the presence of an electric field. The transition from the non-BPS bound state to the
D-particle pair and a fundamental string at the critical radius does not turn out to be a
marginal deformation.
6. Type I analysis
In this section we briefly outline what happens under T-duality to the analysis of the
previous section. As expected T-duality remains valid when analysing non-BPS states.
Under T-duality the Z 2 D-string becomes a Z 2 D-particle analysed in w7,36x. The
combined cylinder and Mobius ¨ strip diagram amplitude Ž5.6. undergoes only one
2
change; the open string momentum sum Ý m ey2 tŽ m r R IX . now becomes a winding sum
2
Ý w ey2 tŽ w R I . , thus stabilising the D-particle for
1
RI ) , Ž 6.1 .
'2
in agreement with w7,36x. Under T-duality an electric field becomes a velocity, and so
the Z 2 D-string with electric flux is dual to a D-particle with constant velocity in the
compactified direction. The open string momentum sum in Ž5.16. now becomes a
winding sum
2
Ý ey2 t Ž w R '1yf .
2
I . Ž 6.2 .
w
The f-dependent square-root is easily identified as a Lorentz contraction of the compact-

ification radius as viewed by the moving Z 2 D-particle. This moving D-particle is stable
for
1
RI ) , Ž 6.3 .
2
a radius smaller than that of the static D-particle, which can be regarded as a
consequence of special relativity.
7. Final remarks and conclusion
Let Ž p,q . denotes a bound state of p F-strings and q D-strings with p,q any
relatively prime pair in type IIB string theory. Existence of this bound state is a
consequence of conjectured SLŽ2,Z. duality symmetry of the type IIB string theory
w34,35x. The Ž p,1. bound states exists as a consequence of the D-string structure w27x.
But the situation is different in our case with Z 2-charged non-BPS D-string due to the
following observation. From Eq. Ž4.3., we see that if we add eÕen number of bi-vector
states to any bi-spinor in A we get a bi-spinor in A itself. In type I X theory, this
observation corresponds to the fact that Ž p,1. bound states exist only for p s 1. For p
odd, the bound state decays into a p s 1 bound state and a number of winding states. On
the other hand, if we add two Ž8,8. F-strings to the non-BPS D-string the system is
unstable against decay into the D-string itself and Žprobably. a full winding state or a
closed string state. As far as charge conservation is concerned this is equivalent to the
fact that two such D-strings together are unstable and decay into massless states. The
fact that two D-strings in our example are not stable is not surprising as two spinor
states can annihilate to give various massless states. For example, if we take two
bi-spinor states of the form ŽŽ 12 . 8 ;Ž 12 . 8 . they together carry the same charge and same
mass Žat all radii. as the state Ž 116 . which can decay into various massless states in the
adjoint representation describing Ž8,8. F-strings with both the D8-branes on the same
orientifold planes.
Since the mass of a state with unit winding is R I X , in the case of the D-string with
two units of electric field, using Ž5.19., the inequality
m I X Ž D-stringq 2 F . ) m I X Ž D-string. q m I X Ž winding state . Ž 7.1 .
holds for sufficiently small l I X at all radius. Hence, for a sufficiently small type I X
coupling, the D-string with two units of electric field is unstable against a decay into a
D-string and a closed string state. Since the configuration studied is non-BPS, it is not
surprising that the above inequality only holds for small type I X coupling.
In this paper we have tested the duality between the heterotic and type I X string with
SO Ž16. = SO Ž16. gauge group beyond the BPS limit. We have found agreement
between the regions of stability and decay products of non-BPS states in both theories.
In particular we have found that the domains of stability are qualitatively related by the
duality map between the two theories. This fact was not guaranteed a priori as the
masses of non-BPS states are not protected by supersymmetry. Unlike the case of
non-BPS D-string the mass of the non-BPS ŽF, D. bound state is not the same as its
decay product at the critical radius. This indicates that the transition from the non-BPS
bound state to the D-particle pair and a fundamental string is not a marginal deforma-
tion.
It would be interesting to perform a similar analysis for other Wilson lines, where the
gauge group is SO Ž16 y 2 N . = SO Ž16 q 2 N .. This is achieved by moving N D8-branes
from one O8-plane to the other and to see whether there is any possible modification in
string creation and gauge enhancement phenomena in the presence of these non-BPS
D-branes.
Acknowledgements
We are grateful to M.B. Green, A. Sen, M.C. Daflon Barrozo for many useful
discussion and in particular to M.R. Gaberdiel for support throughout the project. T.D. is
supported by a Trinity College Scholarship. B.S. is supported by the Cambridge
Overseas Trust.
¨
Appendix A. Mobius strip diagram
¨
In this appendix we explicitly derive the Mobius strip part of Eq. Ž5.16.. The term we
are then interested in is
M s ² C8 < eyl H c < B1 Z 2 , F : . Ž A.1 .
The contributions from the ghosts, superghosts and directions transverse to the Z 2
D-string are as in the case of F s 0. We focus here on the matter contributions in the
directions x 0 and x 9. We write
hyF cosh Ž 2 n . sinh Ž 2 n .
hqF
s
ž sinh Ž 2 n . cosh Ž 2 n .
,
/ Ž A.2 .
since the matrix is orthogonal relative to the inner product defined by the Minkowski
metric h. The bosonic contribution to M is given by
² C8 < eyl H c < BZ 1, F : 0,9

2
m
` Ž y1.
s ²0 <exp Ý m
Ž a m0 a˜ m0 q a m9 a˜ m9 .
ms1
` 1
=eyl H c exp Ý 0
Ž ayn 0
Ž a˜yn 9
cosh Ž 2 n . q a˜ yn sinh Ž 2 n . .
ns1 n
9 9 0
yayn Ž a˜yn cosh Ž 2 n . q a˜ yn sinh Ž 2 n . . . <0:
n k n l
Ž y1. Ž y1.
s Ł
`
Ý
`
²0 <
ž n
a n0a n0 ˜ /ž n
a n9 a n9
˜ /
ns1 k ,l , j, p , s,ts0 k! l!
j p
1 0
1 9 2n 9
ž ayn q 2 na˜ yn
0
cosh Ž 2 n . /ž y ayn q a˜ yn cosh Ž 2 n . /
n n
=
j! p!
s t
1 0
1 9 2n 0
ž ayn q 2 na˜ yn
9
sinh Ž 2 n . /ž y ayn q a˜ yn sinh Ž 2 n . /
n n
= <0:
s! t!
n kq j
k
Ž y1. 1
sŁ Ý ²0 <
ž n
a n0ã n0
/ ž n
0
ayn q 2 nayn
0
˜
cosh Ž 2n . /
n k ,l , j Ž kqj. ! k!
n lqj
l
Ž y1. 1 9 2n 9
=
ž n
a n9 ã n9
/ ž y ayn
n
q a˜ yn cosh Ž 2 n . /
Ž lqj. ! l!
j
1
ž y 2
q 4 nayn
0 0
a˜ yn 9
ayn 9
a˜ yn sinh2 Ž 2 n . /
n
= <0:
j! j!
Ž k q j . ! Ž l q j . ! Ž iq . 2 n kqlq2 j cosh kql Ž 2 n . sinh2 j Ž 2 n .

Ž .
sŁ Ý 6
n k ,l , j k!l!n4 j Ž j! .
j j
=²0 < Ž a n0a˜ n0a n9 a˜ n9 . Ž yayn
0 0
a˜ yn 9
ayn 9
a˜ yn . <0:
kqj lqj 2j
Ž iq . 2 n kqlq2 j cosh kql Ž 2 n . Ž ysinh Ž 2 n . .
Ž .
sŁ
n
Ý
k ,l , j
ž /ž /
j j
2n yjy1 yjy1
s Ý Ž 1 q Ž iq . cosh Ž 2 n . . Ž 1 y Ž iq . 2 n cosh Ž 2 n . .
j
4n j 2 n y1 y1
= Ž Ž iq . sinh2 Ž 2 n . . s Ł Ž 1 y Ž iq .
n
. Ž 1 q Ž iq . 2 n . . Ž A.3 .
In the above we have used the following:

`
kqj k yjy1
Ý
ks0
ž /
j
a s Ž 1 y a. . Ž A.4 .
This demonstrates that the bosonic non-zero-modes contribute the same factor to the
¨
Mobius strip amplitude with or without a gauge field. The fermions are substantially
easier as the coherent state exponentials terminate, due to the anti-commutative nature of
the crm. Explicitly we have
² C8 < eyl H c < BZ 1, F : 0,9

2
r
s ²0 <exp Ý Ž y1. i ž cr0c˜r0 q cr9c˜r9 /
1
rgNq
2
=eyl H c exp Ý 0
i cys
ž Ž c˜ys0 cosh Ž 2 n . q c˜ys
9
sinh Ž 2 n . .
1
sgNq
2
9
ycys Ž c˜ys
9
cosh Ž 2 n . q c˜ys
0
sinh Ž 2 n . . <0: /
r r
s Ł ²0 < Ž 1 q i Ž y1 . cr0c˜r0 .Ž 1 q i Ž y1 . cr9c˜r9 .
1
rgNq
2
= Ž 1 q iq 2 r cosh Ž 2 n . cr0c˜r0 .Ž 1 y iq 2 r cosh Ž 2 n . cr9c˜r9 .
= Ž 1 q iq 2 r sinh Ž 2 n . cr0c˜r9 .Ž 1 y iq 2 r sinh Ž 2 n . cr9c˜r0 . <0:

r r
s Ł ²0 < 1 q i Ž y1 . cr0c˜r0 q i Ž y1 . cr9c˜r9
ž
1
rgNq
2
2r
y Ž y1 . cr0c˜r0cr9c˜r9 1 q iq 2 r cosh Ž 2 n . cr0c˜r0 y iq 2 r cosh Ž 2 n . cr9c˜r9
/ž
qq 4 r cosh2 Ž 2 n . cr0c˜r0cr9c˜r9 1 q q 4 r sinh2 Ž 2 n . cr0c˜r9cr9c˜r0 <0:
/ž /
s Ł Ž 1 y Ž iq . 4 r Ž cosh2 Ž 2 n . y sinh2 Ž 2 n . . .
1
rgNq
2
`
s Ł Ž 1 y Ž iq . 2 ny1 .Ž 1 q Ž iq . 2 ny1 . , Ž A.5 .
ns1
which is indeed the same as for the case with no gauge field.
References
w1x A. Sen, Non-BPS states and branes in string theory, hep-thr9904207.

w2x L. Lerda, R. Russo, Stable non-BPS states in string theory: a pedagogical review, hep-thr9905006.
w3x J.H. Schwarz, TASI lectures on non-BPS D-brane systems, hep-thr9908144.

w9x E. Witten, J. High Energy Phys. 12 Ž1998. 019. hep-thr9810188.
w10x J. Polchinski, Y. Cai, Nucl. Phys. B 296 Ž1988. 91.
w11x C.G. Callan Jr., C. Lovelace, C.R. Nappi, S.A. Yost, Nucl. Phys. B 293 Ž1987. 83.
w12x C.G. Callan, C. Lovelace, C.R. Nappi, S.A. Yost, Nucl. Phys. B 308 Ž1988. 221.
w13x M. Li, Nucl. Phys. B 460 Ž1996. 351. hep-thr9510161.
w14x M.B. Green, M. Gutperle, Nucl. Phys. B 476 Ž1996. 484. hep-thr9604091.
w15x O. Bergman, M.R. Gaberdiel, Nucl. Phys. B 499 Ž1997. 183. hep-thr9701137.
w16x O. Bergman, M.R. Gaberdiel, Phys. Lett. B 441 Ž1998. 133. hep-thr9806155.
w17x M.R. Gaberdiel, A. Sen, Non-supersymmetric D-brane configurations with Bose–Fermi degenerate open
string spectrum, hep-thr9908060.
w18x ´ Jr., Dirichlet branes on orbifolds, hep-thr9910109.
M.R. Gaberdiel, B. Stefanski
w19x E. Witten, Nucl. Phys. B 443 Ž1995. 85. hep-thr9503124.
w20x K.S. Narain, M.H. Sarmadi, E. Witten, Nucl. Phys. B 279 Ž1987. 369.
w21x P. Ginsparg, Phys. Rev. D 35 Ž1987. 648.
w22x A. Sagnotti, Open strings and their Symmetry Groups, Cargese ` 1987, Non-perturbative Quantum Field
Theory, ed. G. Mack et al. ŽPergamon, New York, 1988..
w23x ˇ
P. Horava, Phys. Lett. B 231 Ž1989. 251.
w24x ˇ
P. Horava, Nucl. Phys. B 327 Ž1989. 461.
w25x S. Mukhi, Orientifolds: the unique personality of each space-time dimension, hep-thr9710004.
w26x J. Polchinski, E. Witten, Nucl. Phys. B 460 Ž1996. 525. hep-thr9510169.
w28x A. Dabholkar, J. Harvey, Phys. Rev. Lett. 63 Ž1989. 478.
w29x O. Bergman, M.R. Gaberdiel, G. Lifschytz, Nucl. Phys. B 524 Ž1998. 524. hep-thr9711098.
w30x O. Bergman, M.R. Gaberdiel, J. High Energy Phys. 03 Ž1999. 013. hep-thr9901014.
w31x P. Di Vecchia, M. Frau, A. Lerda, A. Liccardo, ŽF, D p . bound states from the boundary state,
hep-thr9906214.
w32x C. Schmidhuber, Nucl. Phys. B 467 Ž1996. 146. hep-thr9601003.
w33x A. Abouelsaood, C.G. Callan, C.R. Nappi, S.A. Yost, Nucl. Phys. B 280 Ž1987. 599.
w34x J.H. Schwarz, Phys. Lett. B 360 Ž1995. 13.
w35x J.H. Schwarz, Phys. Lett. B 364 Ž1995. 252. hep-thr9508143.
w36x M. Frau, L. Gallot, A. Lerda, P. Strigazzi, Stable non-BPS D-branes in Type I string theory, hep-
thr9903123.
w37x C. Bachas, Phys. Lett. B 374 Ž1996. 37. hep-thr9511043.
w38x J. Polchinski, TASI lectures on D-Branes, hep-thr9611050.
Supercharges, Killing spinors and intersecting gauge

five-branes
E. Lima a , H. Lu¨ a,1, B.A. Ovrut a,2 , C.N. Pope b,3
a
Department of Physics and Astronomy, UniÕersity of PennsylÕania, Philadelphia, PA 19104, USA
b
Centre for Theoretical Physics, Texas A&M UniÕersity, College Station, TX 77843, USA
Received 19 November 1999; accepted 24 December 1999
Abstract
We obtain new solutions where a string and a pp-wave lie in the common world-volume
directions of the non-standard intersection of two gauge 5-branes in the heterotic string. The two
5-branes are supported by independent SUŽ2. Yang–Mills instantons in their respective Žnon-over-
lapping. transverse spaces. We present a detailed study of the unbroken supersymmetry, focusing
especially on a comparison between a direct construction of Killing spinors and a counting of zero
eigenvalues in the anticommutator of supercharges. The results are in agreement with some
previous arguments, to the effect that additional zero eigenvalues resulting from a ‘‘fine-tuning’’
between positive-energy and negative-energy contributions from different components in an
intersection are spurious, and should not be taken as an indication of supersymmetry enhance-
ments. These observations have a general applicability that goes beyond the specific example we
study in this paper. q 2000 Elsevier Science B.V. All rights reserved.
1. Introduction
Over the years, a large menagerie of BPS p-brane solutions of the various supergravi-
ties has been discovered. These include the basic half-supersymmetric solutions such as
the string w1x and 5-brane w2,3x in D s 10, and the membrane w4x and 5-brane w5x in
D s 11. In addition, there are BPS solutions that preserve smaller fractions of supersym-
metry, which admit an interpretation as intersections w6x of the basic half-supersymmet-
ric building blocks.
The intersections themselves can be divided into two broad categories. Firstly, there
are the ‘‘standard’’ intersections, which can all be interpreted, by means of a toroidal
1
Research supported in part by DOE grant DE-FG02-95ER40893.
2
Research supported in part by DOE grant DE-AC02-76ER03071.
3
Research supported in part by DOE Grant DE-FG03-95ER40917.
PII: S 0 5 5 0 - 3 2 1 3 Ž 9 9 . 0 0 8 2 9 - 9
E. Lima et al.r Nuclear Physics B 572 (2000) 112–130 113
reduction to some sufficiently low dimension, as multi-charge p-branes for some single
specific p, where the charges are carried by different field strengths of the lower-dimen-
sional theory. The p-brane for each charge species by itself oxidises back to one specific
component of the intersection in the higher dimension, with the full set of intersecting
objects arising when all the different charge species are turned on. This draws attention
to the fact that one should also include BPS configurations that go somewhat beyond
what one would normally call a ‘‘ p-brane.’’ If the charge-carrying field in the lower
dimension is a Kaluza–Klein vector, then back in the original higher dimension it will
have become part of the off-diagonal structure of the internal part of the metric. If the
Kaluza–Klein vector carried an electric charge in the lower dimension, this will give rise
to a higher-dimensional metric with a pp-wave propagating along one of the internal
directions. If, on the other hand, the Kaluza–Klein field carried a magnetic charge in the
lower dimension, then, from the higher-dimensional standpoint, the metric will have a
Taub-NUT-like monopole structure, sometimes simply called a ‘‘NUT’’ for short.
Various p-brane examples can be found in w7x and references therein. Four-dimensional
black hole solutions were classified in w8x. The intersection rules for p-branes in D s 11
and D s 10 were classified in w9x. The classification of p-branes and standard intersec-
tions in maximal supergravities for 2 ( D ( 11 was given in w11,12x.
The second category of intersecting solutions consists of what may be called
‘‘non-standard’’ intersections. These are examples where there is no simple lower-di-
mensional interpretation as a multi-charge p-brane. This is because in these solutions the
harmonic functions for the intersecting ingredients are all independent of the overall
transverse space. The first such example was constructed in w13x, and further examples
were studied in w9,10,14,15x. A four-dimensional solution with three perpendicular
intersecting membranes Ždomain walls in D s 4. was constructed in w16x, admitting an
interpretation as a cosmological lattice universe model. In fact three is the maximal
number of intersections Žwith all pair-wise intersections non-standard. that can occur in
supergravity theories. The solutions that we shall be constructing in this paper involve a
combination of standard and non-standard intersections.
Intersections in D s 10 can arise in both the type II theories and also in the heterotic
theory. Many of the solutions in the heterotic theory can also be viewed as solutions in
the type II theories, since the subset of the latter that make use only of the NS–NS fields
can be transferred across directly as solutions in the heterotic string. However, there are
also further possibilities in the heterotic string, owing to the presence of the Yang–Mills
fields. One possibility is to find solutions in which the Yang–Mills fields themselves
carry charges that play a role ˆ in supporting the p-brane or intersection. However, in
such circumstances one typically finds that the solution will not be a BPS one. Another
possibility is to use the Yang–Mills fields to construct an instanton configuration in a
four-dimensional transverse space, which can act as a non-singular source in place of the
more customary point-charge singular sources in the harmonic functions describing the
solution. One can think of the Yang–Mills fields here as playing the role ˆ of a
‘‘regulator,’’ which smears out the point-charge singularities. The first example of such
an instanton-supported soliton was the ‘‘gauge 5-brane’’ constructed in w2,3x. A dyonic
string in D s 6, where both the electric and magnetic charges are supported by
Yang–Mills instantons, was constructed in w17x. Such a configuration can also support a
pp-wave propagating in the string world-sheet w18x.
114 E. Lima et al.r Nuclear Physics B 572 (2000) 112–130
The new solutions that we shall construct in this paper involve many of the
ingredients mentioned above. Specifically, what we obtain is a ten-dimensional solution
of the heterotic theory, describing the intersection of two 5-branes, a string, and a
pp-wave. If the 5-brane charges are turned off, the stringrwave intersection is of the
‘‘standard’’ type, which can be viewed as a 2-charge black hole in D s 9, supported by
the winding vector and the Kaluza–Klein vector respectively. We shall first obtain the
more general intersections, with the two 5-branes, a string and a pp-wave, as solutions
using only the fields of the N s 1 truncation of type II supergravities; in other words the
metric, the 2-form potential and the dilaton. As usual in such solutions, there will be
singular sources, corresponding to the locations of point charges or distributions of
charges. Then, we shall show that we can generalise the solutions by using the
Yang–Mills fields of the heterotic string to ‘‘smear out’’ the two sets of 5-brane
charges. A novel feature, associated with the fact that the intersection between the two
5-branes is non-standard, is that we can introduce separate self-dual Yang–Mills
instanton configurations in the two distinct transverse 4-spaces of the two 5-branes.
Having obtained the intersecting solutions, we then turn to a detailed discussion of
their supersymmetry. There has been a rather confusing literature on the subject of the
supersymmetry of multi-charge p-branes, and intersecting p-branes, and in the present
paper we attempt to clarify some of these issues. Although our discussion will be
focused on the particular case of interest here, it actually provides insights of a more
widespread applicability. We shall be particularly concerned with addressing the issue of
how one should interpret the occurrence of zero-eigenvalues of the matrix Q,Q4
obtained by anticommuting the supercharges, and to what extent such zero eigenvalues
can be taken as an indication of the corresponding existence of Killing spinors. In
particular, when one calculates Q,Q4 for BPS configurations involving more than one
kind of charge species, one commonly finds that the number of zero-eigenvalues can
become enhanced for particular fine tunings of the charges, by cancelling one contribu-
tion against another.4 This would appear to imply that the supersymmetry can be
enhanced at these special charge values, leading even to preserved supersymmetry
fractions such as 34 or 78 in some cases. Examples of this apparent phenomenon were
found w19x and subsequently laid to rest w20x in the past. In this paper we examine the
issue in the context of the new intersections of two 5-branes, a string and a pp-wave that
we obtain here. We compare the results from the Q,Q4 anticommutator with the results
of direct computation of the Killing spinors, and we conclude again that the apparent
‘‘supersymmetry enhancements’’ suggested by the enlarged numbers of Q,Q4 zero-ei-
genvalues at special charge values are spurious. The point is that the derivation of the
connection between zero-eigenvalues of Q,Q4 and the existence of Killing spinors
involves certain assumptions about the global structure of the solutions, including the
absence of naked singularities in the metric, and these assumptions are violated in all the
cases where ‘‘fine-tuning’’ of charge parameters enlarges the number of zero-eigenval-
ues. ŽSome detailed discussion of this point was given in w22x.. Thus our results here
support the previous contention that no enhancements of supersymmetry occur at
fine-tuned non-vanishing charge values.
4
As opposed to setting charges to zero, which obviously can enlarge the supersymmetry.
The paper is organised as follows. In Section 2 we construct the new solutions,

comprising a non-standard intersection of two 5-branes together with a string and a
pp-wave, within the framework of the N s 1 truncation of type II supergravity. All the
ingredients, including the 5-branes, have singular sources. Then, in Section 3, we
generalise the solutions within the framework of the heterotic theory, by introducing
self-dual Yang–Mills instantons to replace the singular 5-brane sources. In Section 4 we
examine the supersymmetry of the solutions, both from the type IIA or M-theoretic
viewpoint and from the heterotic viewpoint. We show how the explicit results from
solving the Killing-spinor equations compare with a counting of zero-eigenvalues in the
anticommutator of supercharges, which clarifies the issue of when the zero-eigenvalue
counting procedure gives trustworthy results for the determination of unbroken super-
symmetry. In Section 5, we discuss the near-horizon structure of the intersecting
solutions. The paper ends with conclusions in Section 6.
2. Intersections with singular sources
In this section, we construct the new solution, which is a non-standard intersection of

two 5-branes, a string and a pp-wave, within the framework of the N s 1 truncation of
the type II strings. Specifically, it is a solution of the theory described by the
ten-dimensional Lagrangian
L10 s R)1 y 12 ) d f n d f y 12 eyf ) FŽ3. n FŽ3. , Ž 2.1 .
and is given by
2
2
ds10 s Ky3r4 Hy1r4 H˜y1r4 yWy1 dt 2 q W Ž dx q hw Ž Wy1 y 1 . dt .
ž /
qK 1r4
H 3r4
H˜
y1 r4
Ž dy12 q . . . qdy42
.
qK 1r4
Hy1r4
H˜ 3r4 2 2
Ž dz1 q . . . qdz 4 . ,
f s y log Kr Ž HH˜ . ,
1
2
FŽ3. s h e f ) Ž H˜ dt n dx n d 4 z n dHy1 . q h˜ e f ) Ž H dt n dx n d 4 y n dH˜y1 .

q he dt n dx n dKy1 , Ž 2.2 .
where H s H Ž y ., H˜ s H˜ Ž z ., K s K Ž y, z . and W s W Ž y, z ., and the ten coordinates
have been split as Ž t, x, y, z . with y s Ž y 1 , y 2 , y 3 , y4 . and z s Ž z 1 , z 2 , z 3 , z 4 .. The quanti-
ties h h˜ , he and hw can each independently be chosen to be "1, giving a total of 16
equivalent solutions.5
We find that Eqs. Ž2.2. give a solution provided that the functions H and H˜ are
harmonic with respect to the flat spaces corresponding to their indicated coordinate
dependences,
I y Hs0 , I z H˜ s 0 , Ž 2.3 .
5
These 16 solutions are all equivalent purely within the framework of the bosonic sector of the
supergravity, but they are not all equivalent when supersymmetry is taken into account. This is because field
strengths enter quadratically in the bosonic equations, but linearly in the supersymmetry transformation rules.
while He and W satisfy the equations
Hy1 I y K q H˜y1 I z K s 0 , Hy1 I y W q H˜y1 I z W s 0 . Ž 2.4 .

The harmonic functions H and H˜ are associated with the two 5-branes in the
intersection, K is associated with the string, and W with the pp-wave. Since there is no
overlap between the coordinate dependences of the H and H˜ harmonic functions, it is
evident that the intersection is of a non-standard type, since there is no lower dimension
where the configuration could become a multi-charge p-brane. Furthermore, the func-
tions K and W do not simply satisfy Laplace equations, but instead satisfy coupled
equations that involve the 5-brane harmonic functions H and H. ˜ This also is a
characteristic feature of non-standard intersections. Of course, we can find simple
solutions for K and W by taking
K Ž y, z . s c 1 Ž y . c 2 Ž z . q c 3 Ž y . q c4 Ž z . ,
W Ž y, z . s x 1 Ž y . x 2 Ž z . q x 3 Ž y . q x4 Ž z . , Ž 2.5 .
where the functions c i and x i are all harmonic in their respective subspaces. Note that
solutions of this kind, but without the wave component Ži.e. with W s 1., were obtained
previously in w23,24x. The solution can also be obtained from the dimension reduction of
the intersections of the wave, M2- and M5-branes in D s 11, which were constructed in
w25x.
It is worth remarking that since the Lagrangian Ž2.1. can also be viewed as a
consistent truncation of the type IIA theory, we can also regard the above solutions as
originating from D s 11. In fact we shall exploit this later, when we calculate the
supersymmetry of the solutions.
3. Instanton-supported intersections
We shall now show that we can generalise the above intersection, by using the
Yang–Mills fields of the heterotic string to provide instanton configurations that will
allow the harmonic functions H and H˜ associated with the 5-branes to be replaced by
non-singular solutions in the heterotic string. A novel feature that arises here is that,
owing to the non-overlapping nature of the four-dimensional subspaces coordinatised by
y and z, where the functions H Ž y . and H˜ Ž z . find their support, we can introduce
independent Yang–Mills instantons for the two 5-branes, using separate SUŽ2. factors in
the E8 = E8 or SO Ž32. gauge group.
The low-energy effective action of the heterotic string is N s 1 supergravity in
D s 10, coupled to E8 = E8 or SO Ž32. Yang–Mills matter fields. We shall focus on two
orthogonal SUŽ2. subgroups of E8 = E8 . The Lagrangian for the bosonic sector is given
by
L10 s R)1 y 12 ) d f n d f y 12 eyf ) FŽ3. n FŽ3.

1
y 12 ey 2 f Ž )GŽ2.
a a
n GŽ2. a
q ) GŽ2. a
n GŽ2. ., Ž 3.1 .
a a
where the fields GŽ2. and GŽ2. are the Yang–Mills field strengths given by
a
GŽ2. a
s dBŽ1. q 12 e a b c BŽ1.
b c
n BŽ1. ,
a a
GŽ2. s d BŽ1. q 21 e a bg BŽ1.
b g
n BŽ1. , Ž 3.2 .
and FŽ3. is the three-form field strength, given by

a a
FŽ3. s dAŽ2. q 12 BŽ1.
a a
n dBŽ1. q 16 e a b c BŽ1.
a b
n BŽ1. c
n BŽ1. q 12 BŽ1. n d BŽ1.
q 16 e abg BŽ1.
a b
n BŽ1. g
n BŽ1. . Ž 3.3 .
It satisfies the Bianchi identity

a a
dFŽ3. s 12 GŽ2.
a a
n GŽ2. q 12 GŽ2. n GŽ2. . Ž 3.4 .
We find that the Lagrangian Ž3.1. admits solutions of precisely the same form Ž2.2. as
we obtained in the Introduction, describing the intersection of two 5-branes, a string and
a pp-wave, except that now the functions H, H, ˜ K and W satisfy the more general
equations of motion
I y H s y 14 Giaj Giaj , I z H˜ s y 14 Gman Gman ,
žH y1
I y q H˜y1 I z K s 0,
/ žH y1
I y q H˜y1 I z W s 0 , / Ž 3.5 .
where the index contractions in Giaj Giaj and Gman Gman are performed simply using the
metrics d i j and dm n of the flat four-dimensional transverse spaces dy i dy i and dz m dz m
a a
respectively. The SUŽ2. Yang–Mills fields GŽ2. and GŽ2. satisfy the self-duality equa-
a a a a
tions )Gi j s Gi j and ) Gm n s Gm n in the four-dimensional flat transverse spaces
respectively, where ) denotes Hodge duality in these flat spaces. To be precise, we
should remark that the bosonic equations can be satisfied by taking the Yang–Mills
fields to be either self-dual or anti-self-dual Žwith independent such choices for the two
SUŽ2. factors.. As with the signs in the expression for the 3-form field strength in Ž2.2.,
the different choices that one makes can impinge upon the supersymmetry of the
solutions, as we shall see later.
The equations for K and W depend on H and H. ˜ In this paper, we shall focus on
solutions where K and W are independent of H and H. ˜ Thus, as in Ž2.5., we may
simply take solutions for K and W built from harmonic functions in the two subspaces:
X XX XXX
2 Q ea 2 Q ea 2 Q ea 2 Q ea
ž
Ks 1q Ý
a < y y ya < 2 /ž 1q Ý
aX < z y za X < 2 / qÝ
a XX < y y ya XX < 2
qÝ
a XXX < y y ya XXX < 2
,
X XX XXX
2 Pwb 2 Pwb 2 Pwb 2 Pwb
ž
Ws 1q Ý
b < y y yb < 2 /ž 1q Ý
bX < z y zb X < 2 / qÝ
b XX < z y zb XX < 2
qÝ
b XXX < z y zb XXX < 2
.
Ž 3.6 .
Our notation here is that ya , ya X , etc., denote independent sets ofX instanton locations for
each type of index a , a X , etc. Likewise, the quantities Q ea , Q ea , etc., denote indepen-
dent sets of charges at the different sets of locations.
The equations for the 5-brane functions H and H˜ have Yang–Mills source terms.
We shall consider the situation where the source in each equation is an SUŽ2.
Yang–Mills instanton Žusing a different SUŽ2. subgroup for each equation.. The use of
single-charge and certain classes of multi-charge SUŽ2. instanton solutions as sources
for the 5-brane were discussed in w17,18x. We have
N la N 1
ž
H s 1 q 14 I y log 1 q Ý
as1 < y y ya < 2 / q Ý
a s1 < y y ya < 2
,
X X
N l̃b N 1
1
4
ž
H̃ s 1 q I z log 1 q Ý
bs1 < z y zb < 2
/ q Ý
b s1 < z y zb < 2
. Ž 3.7 .
As discussed in w18x, the final terms in these expressions for H and H˜ serve the purpose
of subtracting out singular-source contributions to the functions H and H. ˜ Clearly, one
can always add in any harmonic solution of the homogeneous equations I y H s 0 and
I z H˜ s 0 to the solutions of the inhomogeneous equations given in Ž3.5.. However, we
are interested in the case where the sources are entirely non-singular, coming only from
the Yang–Mills instantons. It turns out that the terms involving the logarithms in Ž3.7.
actually include singular-source contributions, and the final terms in the expressions for
H and H˜ are put in precisely to subtract these out.
Thus, as discussed in w18x, the functions H and H˜ given in Ž3.7. satisfy the equations
of motion throughout the space with no singularities. There are two types of phase
transition that can occur in certain limits of the instanton moduli, i.e. the instanton sizes
and their relative locations. If the size of an instanton shrinks to zero, the associated H
function becomes harmonic, with a delta-function singularity at the location of the
instanton, implying that a point-like fundamental 5-brane is created w18x. The second
type of phase transition occurs if two instantons coalesce, leading to the creation of a
point-like fundamental 5-brane w18x. As we shall see later, the vanishing of the instanton
degrees of freedom, and hence the creation of the fundamental 5-brane, turns the null
area of the horizon to a non-vanishing one.
4. Supersymmetry
To begin, we shall consider the situation where we take the scale sizes of the
Yang–Mills instantons to zero, so that the function H and H˜ become harmonic,
satisfying Ž2.3.. Having studied the supersymmetry in the framework of type IIA
supergravity and M-theory, and then in the heterotic framework, we shall then consider
the situation when the Yang–Mills instantons replace the singular sources for the
5-branes. We shall see that the preserved supersymmetry will be the same whether or
not the vanishing-instanton limit is taken.
4.1. M-theory perspectiÕe
When the Yang–Mills instanton scales are both set to zero, the solution in the
heterotic theory can be embedded into type IIA supergravity, and hence into M-theory.
We shall make use of this in order to calculate the explicit conditions for the existence
of Killing spinors, by viewing the configuration as a solution in D s 11. Then, we shall
compare these explicit results with what one learns by studying the matrix of anticom-
mutators of supercharges. This latter method can be a useful tool for determining the
fraction of unbroken supersymmetry, although as we shall see, the results that come
from it must be interpreted with care. Specifically, it can sometimes give a false
indication of ‘‘enhanced’’ supersymmetry for special values of the charges, but these
always turn out to be spurious, being associated with configurations with negative-mass
contributions in the metric and naked singularities. ŽSimilar issues were discussed
previously in w20x..
4.1.1. Killing spinor construction

We may view the configuration comprising the intersection of the two 5-branes,
string and a pp-wave as a solution of D s 11 supergravity. To do this, we oxidise from
D s 10 type IIA to D s 11 using the standard Kaluza–Klein rules:
4
1 f
3 2
ˆ2
ds11 s ey 6 f 2
ds10 qe Ž d j q AŽ1. . ,
F̂Ž4. s FŽ4. q FŽ3. n Ž d j q AŽ1. . . Ž 4.1 .

Thus we find that the solution in D s 11 is given by
y1 r3 2
2
dsˆ11 s Ky2r3 Ž HH˜ . ž yW y1
dt 2 q W Ž dx q Ž Wy1 y 1 . dt . /
2r3
q K 1r3 H 2r3 H˜y1r3 dy 2 q K 1r3 Hy1r3 H˜ 2r3 dz 2 q Ky2r3 Ž HH˜ . dj 2 ,
FˆŽ4. s ) y dH q ) z dH˜ q dKy1 n dt n dx n d j ,

ž / Ž 4.2 .
where ) y and ) z denote the Hodge duals in the four-dimensional y and z subspaces,
in the metrics dy i dy i and dz m dz m , respectively. Note that here the solution describes a
non-standard intersection of two 5-branes, a membrane, and a pp-wave. We have made
the specific choice of the solution where all four h sign parameters in Ž2.2. are taken to
be q1. We shall discuss the effects of including the h parameters later.
To solve for the Killing spinors, it is useful first to calculate the spin connection for
the class of metrics
2 2
dsˆ11 s ye 2 A dt 2 q e 2 A W 2 Ž dx q Ž Wy1 y 1 . dt . q e 2 B dy 2 q e 2 C dz 2
q e2 f dj 2 , Ž 4.3 .
0 A 9 A y1
taking the natural orthonormal basis e s e dt, e s W e Ž dx q ŽW y 1. dt . Žwhere
we choose to take x 0 s t and x 9 s x .; e i s e B dy i , e m s e C dz m ; and e a s e f d j .
ŽThere should be no confusion between vielbeins and exponentials! Note that a,
pronounced ‘‘ten,’’ denotes the vielbein component in the extra dimension.. We find
that the spin connection is
v 09 s y 12 eyB Wy1 E iW e i y 12 eyC Wy1 EmW e m ,
v 0 i s eyB Ž yE i A e 0 y 12 Wy1 E iW e 9 . ,
v 0 m s eyC Ž yEm A e 0 y 12 Wy1 EmW e 9 . ,
v 9 i s eyB Ž E i A q Wy1 E iW . e 9 y 12 eyB Wy1 E iW e 0 ,
v 9 m s eyC Ž Em A q Wy1 EmW . e 9 y 21 eyC Wy1 EmW e 0 ,
v i j s eyB Ž E j B e i y E i B e j . , v i m s eyC Em B e i y eyB E i C e m ,
v m n s eyC Ž En C e m y Em C e n . ,
v ia s yeyB E i f e a , v ma s yeyC Em f e a . Ž 4.4 .
The supersymmetry transformations in D s 11 are given by
1
dcA s DA e y 288 GA BCDE
e FB C D E q 361 G B C D e FA B C D . Ž 4.5 .
Consider first the A s 0 vielbein components of this equation. Substituting the eleven-
dimensional solution Ž4.2. into this, and using Ž4.4., we obtain
1r6
dc 0 s K 1r3 Ž HH˜ . E 0 e y Ky1r6 Hy1r3 H˜ 1r6 1
6 Ky1E i K Ž Gm i q em nGn ia .
q 121 Hy1E i H Ž Gm i q 61 e i jk l Gm jk l a . q 41 Wy1 E iW Ž G 0 i q G 9 i . e
y Ky1 r6 H 1r6 H˜y1r3 1

6 Ky1Em K Ž Gm m q em nGn ma .
q 121 H˜y1Em H˜ Ž Gm m q 61 e m n p q Gm n p qa . q 41 Wy1 EmW Ž G 0 m q G 9 m . e . Ž 4.6 .
From this, we see that we shall have solutions of dc 0 s 0 if E 0 e s 0, and the following
conditions hold:
Ž Gi q 16 ei jk l Gjk l a . e s 0 , Ž Gm q 16 em n p q Gn p qa . e s 0 ,
Ž G 0 q G 9a . e s 0 , Ž G0 q G9 . e s 0 . Ž 4.7 .
Note that the first two conditions come respectively from the coefficients of E i H and
Em H˜ in Ž4.6., while the last two conditions come respectively from the coefficients of
˜ K or W is trivial Ži.e.
Ž E i K, Em K . and Ž E iW, EmW .. Thus if any of the functions H, H,
equal to 1 , then the associated condition in 4.7 will be absent. Note also that in
. Ž .
deriving separate conditions associated with each function, we have implicitly assumed
that the functions are independent Ži.e. not proportional to one another..
Proceeding with the A s 9, i, m and j components of dcA in a similar fashion, we
find that the conditions for the existence of Killing spinors are just precisely those
already found in Ž4.7., together with
y1 r12
e s Ky1r6 Ž HH˜ . Wy1r4 e 0 , Ž 4.8 .
where e 0 is a constant spinor.
If we take the indices i and m in the y and z spaces to range over the values
Ž1,2,3,4. and Ž5,6,7,8. respectively, it follows from Ž4.7. that the conditions for the
existence of Killing spinors are that the constant spinor e 0 must satisfy
H: e 0 s yG 1234a e 0 ,
H̃ : e 0 s yG5678a e 0 ,
K: e 0 s G 09a e 0 ,
W: e 0 s G 09 e 0 , Ž 4.9 .
where we have made explicit which condition is associated with which metric function.
Any one of the four equations in Ž4.9. by itself has 16 independent solutions for e 0 .
Thus with just one of the four charges turned on, the solution preserves 12 of the
eleven-dimensional supersymmetry. The number of solutions that one gets when more
charges are turned on depends on various factors, including one’s choice of gamma-ma-
trix conventions. In particular, one should bear in mind that the product G 0123456789a
must be either q1 or y1. If we make the convention choice
G 0123456789a s q1 , Ž 4.10 .
then we find, for example, that the conditions from H and from H˜ are equivalent, and
so introducing the charge for H˜ as well as for H would yield no further constraints. On
the other hand, if the opposite gamma-matrix convention to Ž4.10. were chosen, then
introducing H˜ as well as H would cause all Killing spinors to be lost. We shall not
enumerate here all the supersymmetry fractions for the various possible non-vanishing
sets of charges, since the results can be summarised more succinctly later. Let us just
remark for now that with optimally chosen conventions, one finds that the solution with
all four charges active preserves 18 of the supersymmetry.
We observed in Section 2 that there are actually 16 independent solutions to the
bosonic equations of motion that follow from Ž2.1., where we allow the independent
choice of q1 or y1 for each of the parameters h , h˜ , he and hw in Ž2.2.. Thus our
specific choice in Ž4.2. corresponds to Žh ,h˜ ,he ,hw . s Žq1,q 1,q 1,q 1.. It is clear that
when we reinstate the parameters, the conditions Ž4.9. will be replaced by
H: e 0 s yh G 1234a e 0 ,
H̃ : e 0 s yh˜ G5678a e 0 ,
K: e 0 s he G 09a e 0 ,
W: e 0 s hw G 09 e 0 . Ž 4.11 .
We shall return to a discussion of the possible sign choices later.
Note that using Ž4.10., we can replace the gamma-matrix combinations G 1234a and
G5678a in Ž4.11. by G 056789 and G 012349 respectively. This then means that the indices
on the first three gamma-matrix combinations in Ž4.11. can be viewed as lying in the
world-volumes of the two 5-branes and the membrane respectively. The last combina-
tion, for W, lies in the plane in which the pp-wave propagates.
4.1.2. Superalgebra analysis

It is now instructive to compare the explicit results that we have obtained for the
Killing spinors with what one learns from the eleven-dimensional supersymmetry
algebra. We shall follow some of the notation and conventions of w21x. One finds that
the anticommutator of the supercharges Q gives the expression
1 1
Q,Q 4 s C G M PM q
ž G M 1 M 2 ZM 1 M 2 q G M 1 . . . M 5 ZM 1 . . . M 5 ,
/ Ž 4.12 .
2! 5!
where C is the charge-conjugation matrix, which can be taken to be C s G 0 , and ZM 1 M 2
and ZM 1 . . . M 5 are 2-form and 5-form charges. In the present case, where we have two
5-branes and a membrane supported by FŽ4. in D s 11, the charges will be given by the
asymptotic integrals of the three terms in FŽ4. given in Ž4.2.. Let us call the 5-brane and
membrane charges q5 , q˜5 and q2 respectively. Thus we will have non-zero Z’s given by
Z12349 s q5 , Z56789 s q˜5 , Z9a s q2 . Ž 4.13 .
In addition, the pp-wave will contribute to the momentum PM in its direction of
propagation, and so
P9 s q w . Ž 4.14 .
We may choose a basis for the eleven-dimensional gamma matrices where
G 012349 s diag Ž 1,1,1,1,y 1,y 1,y 1,y 1 . m 1 4 ,
G 056789 s diag Ž 1,1,y 1,y 1,1,1,y 1,y 1 . m 1 4 ,
G 09a s diag Ž 1,1,y 1,y 1,y 1,y 1,1,1 . m 1 4 ,
G 09 s diag Ž 1,y 1,y 1,1,y 1,1,1,y 1 . m 1 4 , Ž 4.15 .
since G 012349 , G 056789 , G 09a and G 09 all commute with one another. We therefore find
that
Q,Q 4
s diag Ž E y q5 y q˜5 y q2 y qw , E y q5 y q˜5 y q2 q qw , E y q5 q q˜5 q q2 q qw ,
E y q5 q q˜5 q q2 y qw , E q q5 y q˜5 q q2 q qw , E q q5 y q˜5 q q2 y qw ,
E q q5 q q˜5 y q2 y qw , E q q5 q q˜5 y q2 y qw . m 1 4 . Ž 4.16 .
where the total energy E is given by
E s p5 q p˜5 q p 2 q pw . Ž 4.17 .
The quantities p5 , p˜5 , p 2 and pw are the individual contributions to the ADM mass
coming from the two 5-branes, the membrane and the pp-wave. They correspond
directly to the overall asymptotic coefficients of leading-order inverse power-law
coordinate dependences of the four metric functions H, H, ˜ K and W. The correspond-
ing charges q5 , q˜5 , q2 and qw are related to them by
Ž q5 ,q˜5 ,q2 ,qw . s Ž h p5 ,h˜ p˜5 ,he p 2 ,hw pw . , Ž 4.18 .
where Žh ,h˜ ,he ,hw . are the 16 " sign choices in the bosonic solutions that we discussed
previously.
Consider first the case where Žh ,h˜ ,he ,hw . s Žq1,q 1,q 1,q 1.. We then find that
the anticommutator Ž4.16. is given by
Q,Q 4 s 2 diag Ž 0, pw , p˜5 q p 2 q pw , p˜5 q p 2 , p5 q p 2 q pw , p5 q p 2 , p5

qp˜5 , p5 q p˜5 q pw . m 1 4 . Ž 4.19 .
Clearly in general, namely with all four parameters non-zero, this will have just 4 zero
eigenvalues, giving a counting of 18 unbroken supersymmetry that is in precise agree-
ment with our findings from the explicit solutions for the Killing spinors. For the various
possible combinations of non-vanishing subsets of parameters Ž p5 , p˜5 , p 2 , pw ., the
corresponding numbers of zero eigenvalues in Ž4.19. can be read off. In each case it is
easy to verify that the counting agrees precisely with our previous derivation of the
constraint Ž4.9. for the Killing spinors, provided that, as usual, the spurious zero
eigenvalues that could apparently be achieved by allowing negative p parameters are
discarded.
For a total of 8 of the 16 possible sign choices in Ž4.18., the story is similar. It is best
summarised by discussing the eigenvalues of Q,Q4 , rather than keeping track of the
ordering of diagonal entries, which are permuted around in the various cases. Thus we
may say that for 8 of the sign choices, we obtain the eigenvalues
l s 2 Ž 0, p5 q p 2 , p˜5 q p 2 , p5 q p˜5 , pw , p5 q p 2 q pw , p˜5 q p 2 q pw , p5 q p˜5 q pw . ,

Ž 4.20 .
each occurring with degeneracy 4. ŽThis is the same as the set of eigenvalues in the
specific example Ž4.19... For these supersymmetric solutions, the massrcharge relations
are given by
E ' p5 q p˜5 q p 2 q pw s q2 " Ž q5 q q˜5 . " qw , or yq2 " Ž q5 y q˜5 . " qw .

Ž 4.21 .
For the remaining 8 possibilities, the eigenvalues are given by
2 Ž p 2 , p5 , p˜5 , p5 q p˜5 q p 2 , p 2 q pw , p5 q pw , p˜5 q pw , p5 q p˜5 q p 2 q pw . ,

Ž 4.22 .
and hence these solutions break all the supersymmetry if all the charges are non-vanish-
ing. For these non-supersymmetric solutions the massrcharge relations are given by
E ' p5 q p˜5 q p 2 q pw s q2 " Ž q5 y q˜5 . " qw , or yq2 " Ž q5 q q˜5 . " qw .

Ž 4.23 .
Naively, it might seem that we could achieve different, sometimes unusual, fractions
of unbroken supersymmetry by making certain special non-vanishing choices for the
Ž p5 , p˜5 , p 2 , pw . parameters. For example, choosing p5 s p˜5 s yp 2 , with pw s 0, in
Ž4.20., we could apparently get 24 zero eigenvalues and hence 34 preserved supersymme-
try. On the other hand, we saw no indication at all from the explicit solutions for the
Killing spinors that such ‘‘supersymmetry enhancements’’ could occur at special values
of the charges.6
The resolution to this puzzle, as discussed in w20,22x, is that the occurrence of zero
eigenvalues in the anticommutator of supercharges can sometimes give a false impres-
sion of the existence of additional Killing spinors, and here we have encountered
precisely such an example. The reason why the supercharge argument is unreliable here
is that in order to achieve the extra zeroes in Ž4.19. it was necessary to have at least one
of the contributions Ž p5 , p˜5 , p 2 , pw . that appear in the energy Ž4.17. be negative. This
means that the associated harmonic function will have a negative asymptotic coefficient,
and hence it means that there will be a naked singularity in the metric in some region.
Under such circumstances, the assumed conditions under which one derives a relation
between zeroes of the anticommutator Ž4.12. and unbroken supersymmetry generators
are violated, and so one cannot trust the result. The fact that no additional Killing
spinors actually arise, as seen from our earlier explicit calculation, shows that this is
indeed what has happened in this case.
4.2. Heterotic theory perspectiÕe
To study the fractions of preserved supersymmetry in the heterotic theory, we can use
the same Killing-spinor calculations as we did before, but now we impose the additional
ten-dimensional chirality condition on the Killing spinor,
e 0 s Ga e 0 . Ž 4.24 .
ŽOf course we might instead impose this condition with a minus sign, depending on our
conventions.. Thus the number of independent components of unbroken supersymmetry
is determined by solving Eqs. Ž4.9., together with Ž4.24.. As usual, if any of the charges
in Ž4.9. is zero, then the associated condition is omitted. Rather than stating the results
for the various supersymmetry fractions here, it is more convenient first to give the
discussion of the anticommutator of supercharges in this heterotic case. As in the
previous M-theory discussion, we again find that the two approaches agree, provided
that we discard any apparent supersymmetry enhancements that would naively appear to
occur in the Q,Q4 eigenvalue calculation when there are negative energy contributions.
Since we are not for now concerned with the contributions of the Yang–Mills fields
in the anticommutator algebra Q,Q4 , we can make use of the same formalism Ž4.12. as
in D s 11, but with the additional requirement that we should project all matrices onto
the positive eigenspace of the chirality operator Ga . From Ž4.15., we see that in the
gamma matrix conventions we are using here Ga will be given by
Ga s diag Ž 1,y 1,1,y 1,1,y 1,1,y 1 . m 1 4 . Ž 4.25 .
The projection onto the positive eigenspace of Ga therefore amounts to keeping only
6
We did remark at the stage when we obtained the conditions Ž4.7. for the existence of Killing spinors that
we were assuming that the functions H, H, ˜ K and W were not proportional to one another. It is crucial to
appreciate that the quantities p5 , p˜5 , p 2 and pw are precisely the coefficients appearing in these functions, and
so there can be no possibility of two of the functions being proportional in the kinds of cases we are
considering here, where one of their p coefficients is the negatiÕe of the other.
the 1st, 3rd, 5th and 7th entries in Ž4.16.. It turns out that for 4 out of the possible 16
sign choices in Ž q5 ,q˜5 ,q2 ,qw . s Ž"p5 ," p˜5 ," p 2 ," pw . we obtain eigenvalues
l s 2 Ž 0, p˜5 q p 2 , p5 q p 2 q pw , p5 q p˜5 . , Ž 4.26 .
for Q,Q4 , each with degeneracy 4. This expression summarises all the information
about the possible supersymmetry fractions that can be achieved; as usual, all the
quantities Ž p5 , p˜5 , p 2 , pw . should be considered to be non-negative. The various super-
symmetry fractions implied by taking all possible non-zero subsets of the Ž p5 , p˜5 , p 2 , pw .
agree completely with those obtained by explicitly solving the Killing-spinor equations.
The massrcharge relation for these supersymmetric solutions is given by
E ' p5 q p˜5 q p 2 q pw s Ž q2 q qw . " Ž q5 q q˜5 . ,
or y Ž q2 q qw . " Ž q5 y q˜5 . . Ž 4.27 .
Note that the other 12 possible sign choices in Ž q5 ,q˜5 ,q2 ,qw . s Ž"p5 ," p˜5 ," q2 ,"
qw . divide into three further sets of 4, with each set leading to eigenvalues as follows:
l s Ž p5 q p 2 , p˜5 q p 2 , pw , p5 q p˜5 q pw . ,
l s Ž p 2 , p5 q pw , p˜5 q pw , p5 q p˜5 q p 2 . ,
l s Ž p5 , p˜5 , p 2 q pw , p5 q p˜5 q p 2 q pw . . Ž 4.28 .
Thus with all four charges turned on, none of these other sets gives rise to any preserved
supersymmetry. Note also that if we make the opposite sign choice for the chirality
projection, then the same sets of eigenvalues Ž4.26. and Ž4.28. occur, but now with
different combinations of sign choice in Ž q5 ,q˜5 ,q2 ,qw . s Ž"p5 ," p˜5 ," q2 ," qw . being
associated with each set of eigenvalues.
The above example provides another manifestation of a supersymmetry rule obtained
in w12x. Namely, when a new intersecting ingredient is introduced in a set of intersec-
tions, if it breaks a further half of the supersymmetry then the structure of the
eigenvalues of Q,Q4 is independent of the sign of the new charge. If, on other hand, the
introduction of the new charge does not break supersymmetry further, then it would
break the supersymmetry completely if it were instead introduced with the opposite sign.
Thus for N intersecting objects that preserve 1r2 n of the supersymmetry with n ( N, 2 n
out of the 2 N possible choices of solutions are supersymmetric while the rest are
non-supersymmetric.
Finally, in our discussion of supersymmetry in the heterotic framework, we examine
the situation where we include the Yang–Mills fields, and consider the solutions where
H and H˜ are non-singular functions with Yang–Mills instanton sources, as given in
Ž3.7.. The supersymmetry transformation rule for the gravitino is unchanged from the
one in the N s 1 truncation of the type II theory, and so our previous calculation of the
supersymmetry in the gravitino sector goes through unchanged. ŽIn the derivation of the
Killing spinors, for example in Ž4.6., it was not important that the functions H and H˜
be harmonic.. However, we now have to consider also the supersymmetry transforma-
tions of the gaugini x a and x a , which are the superpartners of the SUŽ2. Yang–Mills
a a
fields BŽ1. and BŽ1. respectively. These take the form
dx a s Giaj Gi j e , dx a s Gman Gm n e . Ž 4.29 .
Now, if Giaj is self-dual or anti-self-dual, we shall therefore find that the Killing spinors
e should satisfy
Ž Gi j " 12 ei jk l Gk l . e s 0, Ž 4.30 .
a
respectively, with similar conclusions in the z space for dx . Equivalently, we can
express these conditions as
e s .G 1234 e , e s .G5678 e . Ž 4.31 .
Bearing in mind that we also have the chirality condition Ž4.24. Žor its opposite., we see
that the conditions Ž4.31. are of the same form as the ones already encountered in Ž4.11..
Thus the question of whether the solutions are supersymmetric or not comes down to the
issue of achieving a proper correlation of signs, with the self-duality or anti-self-duality
choice for the Yang–Mills instantons being correlated with the chirality sign convention
for the spinors of the heterotic theory. Provided the signs are properly chosen, the
preserved supersymmetry fractions for the Yang–Mills instanton-supported solutions in
the heterotic theory will be the same as for their corresponding singular-source limits.
4.3. Further comments
In the previous discussion we demonstrated that the correspondence between the

counting of zero-eigenvalues in the anticommutator Q,Q4 and the counting of Killing
spinors holds if the energy contribution from each intersecting ingredient is non-nega-
tive. One cannot trust any additional zero-eigenvalues that arise by virtue of having any
negative-energy contributions from any of the intersecting ingredients. It should be
emphasised, however, that one does not always get the wrong conclusion from the
Q,Q4 calculation when there are negative-energy contributions. For example, for a
simple extremal p-brane solution preserving 12 supersymmetry the analogous result is
Q,Q4 s 2Ž0, p . = 1 16 , and this correctly implies that there will continue to be 16
unbroken components of supersymmetry even if the mass is taken to be negative. ŽThe
Killing spinor equation continues to admit 16 solutions, even if one sets the mass
negative.. But the Q,Q4 calculation is giving the ‘‘correct conclusion for the wrong
reason’’ if the mass is negative. On the other hand, it seems that in all cases where there
appear to be enhanced supersymmetry fractions for particular tuned sets of charges that
involve negative-energy contributions, the conclusion is always wrong.
Although Killing spinor solutions still exist even if an intersecting ingredient
contributes a negative energy, the behaviour of the Killing spinor will become singular.
To see this, recall that the Killing spinor solution Ž4.8. holds regardless of the detailed
structure of the harmonic functions K, H, H˜ or W. When each intersecting ingredient
has positive energy, these harmonic functions take values between 1 and `, and hence
the Killing spinors are finite. On the other hand, if any intersecting object contributes a
negative energy, then the associated harmonic function takes values between 1 and 0,
with a naked singularity occurring at the latter value. As can be seen from Ž4.8., the
Killing spinor blows up at the naked singularity.
In most examples, requiring the regularity of the Killing spinor leads to a positive
energy contribution, which in turn implies cosmic censorship w26x. However, as observed
in w22x, when a 5-brane with negative mass is supported by a Yang–Mills instanton of
sufficient scale size, the previous singularity in metric can be smeared out by the
instanton, and hence the Killing spinor is also well-behaved. In this case, the continued
violation of the relation between the zero-eigenvalue counting in Q,Q4 and the counting
of Killing spinors is caused by the fact that the energy–momentum tensor of the
Yang–Mills instanton fields now violates the positive-energy condition w22x.
For N intersecting objects that preserve 1r2 N of the supersymmetry, the eigenvalues
of the anticommutator Q,4 have the form
l s Ž E " p 1 " . . . " pN .
s 2 Ž 0, p 1 , p 2 , . . . , pN , p 1 q p 2 , p 1 q p 3 , . . . , p 1 q . . . pN . , Ž 4.32 .
where the total energy is E s Ý i pi . Thus the anticommutator Q,Q4 would have
negative eigenvalues if any of the objects contributes a negative energy. On the other
hand, if the N intersecting objects preserve 1r2 n of the supersymmetry with n - N,
then Q,Q4 can still remain positive even if some of the individual energy contributions
are negative, since in these cases, not all the individual energy contributions appear in
isolation as eigenvalues. For example, the eigenvalues in Ž4.19. will remain non-nega-
tive even if we set q2 q q5 s 0. The phenomenon was observed in w27x in the context of
4-charge black holes in four-dimensional heterotic string theory, where the eigenvalues
of Q,Q4 are 2 0, p 1 q p 2 , p 3 q p4 , p 1 q p 2 q p 3 q p4 4 and the energy is E s p 1 q p 2 q
p 3 q p4 . Thus one can obtain a massless black hole by setting p 1 q p 2 s 0 and
p 3 q p4 s 0, without there being any negative eigenvalues. Another example in the
literature is the dyonic string in N s 1 supergravity, for which Q,Q4 has eigenvalues
2 0, p 1 q p 2 4 , with energy E s p 1 q p 2 w17,22x. It follows that the dyonic string becomes
tensionless when p 1 q p 2 s 0. ŽOf course as usual, one should not take the occurrence
of extra zero-eigenvalues in these limits as indicating enhanced supersymmetry.. It
seems that solutions such as the above massless ones, where no negative eigenvalues
occur in Q,Q4 , may have a more solid relation to states in the quantum theory than ones
where masslessness is achieved at the price of negative eigenvalues.
In our present case in this paper, from an M-theory perspective, there is no choice of
parameters such that the energy E s p5 q p˜5 q p 2 q pw vanishes while the eigenvalues
in the associated anticommutator Ž4.19. all remain positive. On the other hand, from the
heterotic perspective the eigenvalues in the superalgebra are truncated to Ž4.26., and
hence it is possible to obtain a massless solution with purely non-negative eigenvalues in
the associated superalgebra. Of course achieving the massless solution requires
negative-energy contributions from some of the intersecting ingredients, and hence
either the Killing spinor blows up, or the Yang–Mills energy density becomes negative.
Such a pathology may imply a phase transition of the type discussed in w17,30x.
5. Near-horizon structure
For a single-centre configuration, the solutions to Eqs. Ž3.5. can be taken to be

2 Qe 2 QXe 2 Qw 2 QXw
Ks 1q ž r2 /ž 1q
r2 / , Ws 1q ž r2 /ž 1q
r2 / ,
2 Ž r 2 q 2 a2 . 2Ž r 2 q 2 b2 .
Hs1q 2
, H˜ s 1 q 2
, Ž 5.1 .
Ž r 2 q a2 . Ž r 2 q b2 .
where r 2 s y i y i and r 2 s z m z m. Here a and b are the sizes of the instantons supporting
each of the 5-branes.
Let us now consider what happens if the two instanton sizes a and b vanish. In this
situation, the functions H and H˜ become harmonic functions. Let us consider the
horizon region with r r ™ 0 and rrr non-vanishingly finite. In this region, the additive
constants ‘‘1’’ in these two functions H and H˜ can be dropped. Also K ;
4Q e QXerŽ r 2 r 2 .. In this region the dilaton scalar f becomes constant, and the metric
becomes Žfor simplicity, we consider Q e QXe s 1.
2
ds10 s ds42 q 2 d V 32 q 2 d V˜ 32 , Ž 5.2 .
where d V 32 and d V˜ 32 are the metrics for unit 3-spheres, and
2 2 dr 2 2 dr2
ds42 s 14 r 2 r 2 yWy1 dt 2 q W Ž dx q Ž Wy1 y 1 . dt . q
ž / q . Ž 5.3 .
r2 r2
The four-dimensional configuration Ž5.3. is the solution to the Lagrangian

2 2
ey1 L s R y 12 Ž Ef . y 12 e 2 f Ž Ex . q m2 e f , Ž 5.4 .
which is the scalar Lagrangian of four-dimensional SUŽ2. = SUŽ2. gauged supergravity,
constructed in w31x.
Making the coordinate transformation y s logŽ r r . and z s logŽ rrr ., the metric
Ž5.3. becomes
2
ds42 s 14 e 2 x yWy1 dt 2 q W Ž dx q Ž Wy1 y 1 . dt . q dy 2 q dz 2 ,
ž /
W s Ž 1 q Q w eyy yz . Ž 1 q QXw eyy qz . . Ž 5.5 .
When Q w s QXw s 0, the four-dimensional metric becomes AdS 3 = S 1, as discussed in
w23,24x. ŽVarious intersections whose near horizon structures that give rise to AdS 3
space-time were given in w10x.. When the 1’s are dropped from W, the above space-time
becomes K 3 = S 1, where K D denotes the generalised Kaigorodov metric in D dimen-
sions w32x. Note that K 3 is locally equivalent to the BTZ black hole constructed in w33x.
The near horizon structure of this intersection were also discussed in w25x from D s 11
point of view.
The area of the horizon for the metric Ž5.2., and hence the entropy, is proportional to
(Q w QXw . On the other hand, when the instanton sizes a and b are non-vanishing, the
area of the horizon, and hence the corresponding entropy, would be zero. An analogous
phase transition occurs also when two instantons coalesce, which increases the area of
the horizon. The entropy associated with the non-vanishing area of the horizon can be
understood from the two-dimensional boundary conformal field theory of the AdS 3
space-time.
Note that the Lagrangian Ž3.1. also admits a different type of four-object intersection,
namely a string, 5-brane, pp-wave and NUT. This intersection is of the standard type,
and gives rise to a 4-charge black hole w28x in D s 4. The near-horizon structure is
BTZ = Ž S 3rZn . = E4 w29x. By contrast, the near-horizon structure of the four-object
intersection discussed in this paper is K 3 = S 3 = S 3 = S 1.
6. Conclusions
In this paper, we have constructed extremal solutions in D s 10, comprising the

non-standard intersection of two 5-branes together with a string and a pp-wave. This can
arise as a solution in the N s 1 truncation of the type II theory, with singular sources for
all the ingredients in the intersection. This configuration can be oxidised to a solution in
D s 11, where the string now becomes a membrane living in the common world-volume
directions of the 5-branes. It can instead be viewed as a solution in the heterotic theory,
in which case it is possible to replace the singular sources for the 5-branes by self-dual
Yang–Mills instantons. An unusual feature here is that, owing to the non-standard
nature of the 5-brane intersection, in which they have non-overlapping four-dimensional
transverse spaces, there can be a separate SUŽ2. instanton for each 5-brane. This
solution therefore makes use of an SUŽ2. = SUŽ2. subgroup of the E8 = E8 or SO Ž32.
gauge group of the heterotic string.
We presented a detailed discussion of the supersymmetry of the intersecting solution.
In particular, we compared the results from an explicit construction of the Killing
spinors with a counting of the zero eigenvalues of the anticommutator of supercharges,
Q,Q4 . We showed that the two are in agreement, provided one discounts as ‘‘spurious’’
the additional zero eigenvalues of Q,Q4 that can arise for special ‘‘tuned’’ non-vanish-
ing values for certain of the charges. We argued that, as discussed in previous examples
in the literature w20,22x, the naive counting of zero eigenvalues of Q,Q4 can give
misleading results, if any of the components in the intersection is giving a negative
contribution to the total energy. Having exhibited this phenomenon in specific examples,
the implication is that one should always treat apparent supersymmetry enhancements
seen from supercharge anticommutators with suspicion, unless there is some compelling
argument for why they are not spurious.
Acknowledgements
We are grateful to Mirjam Cveticˇ for valuable discussions on Killing spinors and the
Bogomoln’yi bound in four-dimensional black-hole solutions, and to Joachim Rahmfeld
for raising again a question about supersymmetry enhancements.
References
w1x A. Dabholkar, G.W. Gibbons, J.A. Harvey, F. Ruiz Ruiz, Nucl. Phys. B 340 Ž1990. 33.
w2x A. Strominger, Nucl. Phys. B 343 Ž1990. 167.
w3x A. Strominger, Nucl. Phys. B 353 Ž1991. 565. ŽE..
w4x M.J. Duff, K.S. Stelle, Phys. Lett. B 253 Ž1991. 113.
w5x ¨
R. Guven, Phys. Lett. B 276 Ž1992. 49.
w6x A.A. Tseytlin, Nucl. Phys. B 475 Ž1996. 149. hep-thr9604035.
w7x M.J. Duff, R.R. Khuri, J.X. Lu, Phys. Rep. 259 Ž1995. 213. hep-thr9412184.
w8x ˇ D. Youm, Nucl. Phys. B 472 Ž1996. 249. hep-thr9512127.
M. Cvetic,
w9x E. Bergshoeff, M. de Roo, E. Eyras, B. Janssen, J.P. van der Schaar, Nucl. Phys. B 494 Ž1997. 119.
hep-thr9612095.
w10x H.J. Boonstra, B. Peeters, K. Skenderis, Nucl. Phys. B 533 Ž1998. 127. hep-thr9803231.
w11x ¨ C.N. Pope, Nucl. Phys. B 465 Ž1996. 127. hep-thr9512012.

H. Lu,
w12x ¨ C.N. Pope, T. A Tran, K.W. Xu, Nucl. Phys. B 511 Ž1998. 98. hep-thr9708055.
H. Lu,
w13x R. R Khuri, Phys. Rev. D 48 Ž1993. 2947. hep-thr9305143.
w14x K. Behrndt, E. Bergshoeff, B. Janssen, Phys. Rev. D 55 Ž1997. 3785. hep-thr9604168.
w15x J.P. Gauntlett, D.A. Kastor, J. Traschen, Nucl. Phys. B 478 Ž1996. 544. hep-thr9604179.
w16x M.J. Duff, P. Hoxha, H. Lu, ¨ R.R. Martinez-Acosta, C.N. Pope, Phys. Lett. B 451 Ž1999. 38.
astro-phr9712301.
w17x M.J. Duff, H. Lu,¨ C.N. Pope, Phys. Lett. B 378 Ž1996. 101. hep-thr9603037.
w18x E. Lima, H. Lu,¨ B.A. Ovrut, C.N. Pope, Instanton moduli and brane creation, to appear in Nucl. Phys. B,
hep-thr9903001.
w19x ¨ C.N. Pope, Int. J. Mod. Phys. A 12 Ž1997. 437. hep-thr9512153.
H. Lu,
w20x H. Lu,¨ C.N. Pope, p-brane taxonomy, in Trieste 1996, High energy physics and cosmology, hep-
thr9702086.
w21x J.P. Gauntlett, C.M. Hull, BPS states with extra supersymmetry, hep-thr9909098.
w22x ¨ C.N. Pope, Nucl. Phys. B 529 Ž1998. 137. hep-thr9711089.
M.J. Duff, J.T. Liu, H. Lu,
w23x P.M. Cowdall, P.K. Townsend, Phys. Lett. B 429 Ž1998. 281.
w24x P.M. Cowdall, P.K. Townsend, Phys. Lett. B 434 Ž1998. 458. hep-thr9801165, ŽE..
w25x J.P. Gauntlett, R.C. Myers, P.K. Townsend, Phys. Rev. D 59 Ž1999. 025001. hep-thr9809065.
w26x R. Kallosh, A. Linde, T. Ortin, A. Peet, A. Van Proeyen, Phys. Rev. D 46 Ž1992. 5278. hep-thr9205027.
w27x ˇ D. Youm, Phys. Lett. B 359 Ž1995. 87. hep-thr9507160.
M. Cvetic,
w28x ˇ D. Youm, Phys. Rev. D 53 Ž1996. 584. hep-thr9507090.
M. Cvetic,
w29x ˇ H. Lu,
M. Cvetic, ¨ C.N. Pope, Nucl. Phys. B 549 Ž1999. 194. hep-thr9811107.
w30x N. Seiberg, E. Witten, Nucl. Phys. B 471 Ž1996. 121. hep-thr9603003.
w31x D.Z. Freedman, J.H. Schwarz, Nucl. Phys. B 137 Ž1978. 333.
w32x ˇ H. Lu,
M. Cvetic, ¨ C.N. Pope, Nucl. Phys. B 545 Ž1999. 309. hep-thr9810123.
w33x M. Banados, C. Teitelboim, J. Zanelli, Phys. Rev. Lett. 69 Ž1992. 1849. hep-thr9204099.
The gauging of five-dimensional, N s 2 Maxwell–Einstein

supergravity theories coupled to tensor multiplets q
a,b
¨
M. Gunaydin , M. Zagermann b
a
CERN, Theory DiÕision, 1211 GeneÕa 23, Switzerland
b
Physics Department, Penn State UniÕersity, UniÕersity Park, PA 16802, USA
Received 9 December 1999; accepted 22 December 1999
Abstract
We study the general gaugings of N s 2 Maxwell–Einstein supergravity theories ŽMESGT. in

five dimensions, extending and generalizing previous work. The global symmetries of these
theories are of the form SUŽ2.R = G, where SUŽ2.R is the R-symmetry group of the N s 2
Poincare´ superalgebra and G is the group of isometries of the scalar manifold that extend to
symmetries of the full action. We first gauge a subgroup K of G by turning some of the vector
fields into gauge fields of K while dualizing the remaining vector fields into tensor fields
transforming in a non-trivial representation of K. Surprisingly, we find that the presence of tensor
fields transforming non-trivially under the Yang–Mills gauge group leads to the introduction of a
potential which does not admit an AdS ground state. Next we give the simultaneous gauging of
the UŽ1.R subgroup of SUŽ2.R and a subgroup K of G in the presence of K-charged tensor
multiplets. The potential introduced by the simultaneous gauging is the sum of the potentials
introduced by gauging K and UŽ1.R separately. We present a list of possible gauge groups K and
the corresponding representations of tensor fields. For the exceptional supergravity we find that
one can gauge the SO ) Ž6. subgroup of the isometry group E6Žy 26. of the scalar manifold if one
dualizes 12 of the vector fields to tensor fields just as in the gauged N s 8 supergravity. q 2000
Elsevier Science B.V. All rights reserved.
q
Work supported in part by the National Science Foundation under Grant Number PHY-9802510.
¨
E-mail addresses: murat@phys.psu.edu ŽM. Gunaydin ., zagerman@phys.psu.edu ŽM. Zagermann..
PII: S 0 5 5 0 - 3 2 1 3 Ž 9 9 . 0 0 8 0 1 - 9
132 ¨
M. Gunaydin, M. Zagermannr Nuclear Physics B 572 (2000) 131–150
1. Introduction
Gauged supergravity 2 theories in various dimensions have been studied extensively

in the early and mid-eighties Žsee, e.g., Refs. w1–6x..
In the last few years there has been a renewed intense interest in gauged supergravity
theories. This interest is driven mainly by the work on AdSrCFT Žanti-de Sitterrconfor-
mal field theory. dualities w7–13x. For example, the IIB superstring theory on the
background manifold AdS 5 = S 5 with N units of five-form flux through the five-sphere,
is conjectured to be equivalent Žat least in a certain limit. to 4 d N s 4 super
Yang–Mills theory with gauge group SUŽ N ., which is a conformally invariant quantum
field theory. In the limit of small string coupling and large N, the classical Ži.e. tree
level. IIB supergravity approximation becomes valid and can be used to discuss the
large N limit of the corresponding dual Yang–Mills theory. The importance of gauged
supergravity lies in the fact that 5d gauged N s 8 supergravity w3–5x is believed to be a
consistent non-linear truncation of the lowest lying Kaluza–Klein modes of IIB super-
gravity on AdS 5 = S 5 w14,15x.3 Many aspects of the AdSrCFT correspondence, such as
the renormalization group flows of certain non-conformal deformations of the Yang–
Mills theory with a smaller number of supersymmetries, can therefore be studied entirely
within the framework of 5d gauged supergravity due to the lack of interference with the
higher Kaluza–Klein modes w18,19x. Thus, gauged supergravity theories lie at the core
of AdSrCFT dualities.
On the other hand, five-dimensional N s 2 gauged supergravity is the natural
framework for so-called brane world scenarios in which our 4 d world is realized as a
domain wall in an effectively five-dimensional theory w20–22x. In fact, certain M-theory
compactifications w23–28x seem to suggest theories which appear five dimensional at a
certain intermediate length scale, at which the effective field theory is given by a certain
5d N s 2 gauged supergravity plus 4 d Standard Model-type matter fields on the 4 d
boundaries of this 5d space-time.
Motivated by the above-mentioned applications, as well as others, we study the most
general gaugings of 5d, N s 2 supergravity theories coupled to vector as well as tensor
multiplets. The work presented here represents a generalization and an extension of
earlier work on the gaugings of N s 2 supergravity coupled to vector multiplets
w29–34x.
The organization of the paper is as follows. For the convenience of the reader,
Section 2 briefly summarizes the basic features of ungauged Maxwell–Einstein super-
gravity theories. Focusing on the global symmetries of these ungauged theories, we list
the possible types of their gaugings. The subsequent four sections describe each of these
2
The term ‘‘gauged supergravity’’ commonly refers to Žusually N-extended. supergravity theories in which
a subgroup of the automorphism group Žalias ‘‘R-symmetry group’’. of the underlying supersymmetry algebra
is realized as a local ŽYang–Mills-type. gauge symmetry. Sometimes, this term is also used for gaugings of
other global symmetry groups that are not subgroups of the R-symmetry group. In this paper, we will refer to
the latter type of theories as ‘‘Yang–MillsrEinstein supergravity theories’’. In contrast, ‘‘ungauged’’
supergravity theories are those for which the R-symmetry group is just a global symmetry group of the
Lagrangian.
3
The consistency of the non-linear truncation of the S 7 and S 4 compactifications of 11-dimensional
supergravity was shown in w16x and w17x, respectively.
¨
M. Gunaydin, M. Zagermannr Nuclear Physics B 572 (2000) 131–150 133
gauge types in detail: Section 3 summarizes the gauging of a UŽ1.R subgroup of the
N s 2 R-symmetry group SUŽ2.R . Sections 4 and 5 are devoted to the gauging of a
subgroup K of the isometry group G of the scalar manifold: Section 4 summarizes the
well-known case without tensor fields, whereas Section 5 covers the case when tensor
fields have to be introduced. The simultaneous gauging of UŽ1.R and K is treated in
Section 6. We conclude with a classification of possible gauge groups and the corre-
sponding representations of tensor fields in Section 7 and a short discussion of our
results in Section 8.
2. Ungauged N s 2 Maxwell–Einstein supergravity theories and their global

symmetries
In this section, we briefly recall the most relevant features of the Žungauged. N s 2
Maxwell–Einstein supergravity theories ŽMESGT. constructed in w29,30x. Unless other-
wise stated, our conventions will coincide with those of Refs. w29,30x, where further
details can be found. In particular, we will use the metric signature Žyqqqq . and
impose the ‘symplectic’ Majorana condition on all fermionic quantities.
¨
The fields of the N s 2 supergravity multiplet are the funfbein emm , two gravitini Cmi
Ž i s 1,2. and a vector field Am . An N s 2 vector multiplet contains a vector field Am ,
two spin-1r2 fermions li and one real scalar field w . The fermions of each of these
multiplets transform as doublets under the USpŽ2.R ( SUŽ2.R R-symmetry group of the
N s 2 Poincare´ superalgebra; all other fields are SUŽ2.R inert.
The N s 2 MESGT’s constructed in w29,30x describe the coupling of n˜ vector
multiplets to supergravity. Hence, the total field content is
˜
½em i
m ,Cm , AmI , li a˜ , w x˜ 5 Ž 2.1 .
with
I˜s 0,1, . . . ,n,
˜
a˜ s 1, . . . ,n,
˜
x˜ s 1, . . . ,n,
˜
where we have combined the ‘graviphoton’ with the n˜ vector fields of the n˜ vector
˜
multiplets into a single Ž n˜ q 1.-plet of vector fields AmI labelled by the index I.˜ The
indices a,b, ˜
˜ . . . and x, ˜ y,
˜ . . . should be interpreted as flat and curved indices, respec-
˜
tively, of the n-dimensional target space manifold M of the scalar fields. ŽOur indices
˜ ˜ x˜ . correspond to the indices Ž I,a, x . in Refs. w29–32x..
Ž I,a,
The generic Maxwell–Einstein supergravity Lagrangian was found to be Žup to
4-fermion terms. w29,30x
( ˜ ˜
ey1 L s y 12 R Ž v . y 12 CmiG mnr =nCr i y 14 a I˜J˜FmnI F Jmn
˜ ˜ ˜
y 12 li a˜ G m=m d a˜ b q V xã˜ bG mEm w x˜ l bi y 12 g x˜ y˜ Ž Em w x˜ . Ž E mw y˜ .
ž /
i ˜ i
y li a˜G mG nCm i f xã˜En w x˜ q 14 h aI˜˜li a˜G mG l r Cm i FlIr q
2 2'6
= ž 1
4 d a˜ b˜ h I˜q Ta˜ b˜ c˜ h cI˜˜ li a˜G mnl bi˜ FmnI˜
/
134 ¨
3i ˜ ˜
y h I˜ CmiG mnrsCn i FrsI q 2C m iCin FmnI
8'6
ey1 mnrsl I ˜ ˜ ˜
q ˜ ˜´
CI˜JK Fmn FrsJ AlK Ž 2.2 .
6'6
with the supersymmetry transformation laws Žto leading order in fermion fields.
d emm s 12 ´ iG m Cm i ,
i nr ˜
dCm i s =m Ž v . ´ i q h I˜ Ž Gm y 4dmn G r . FnrI ´ i ,
4'6
d AmI˜ s qmI˜,
i ˜
dl ai˜ s y f xã˜ G m Ž Em w x˜ . ´ i q 14 h aI˜˜G mn´ i FmnI ,
2
i
dw x˜ s f a˜x˜´ il ai˜ , Ž 2.3 .
2
where
i'6
qmI˜' y 12 h aI˜˜ ´ iGm l ai˜ q ˜
h I Cmi´ i . Ž 2.4 .
4
Here, e denotes the funfbein ¨ determinant, whereas RŽ v . and =m s =mŽ v . are the scalar
curvature and the space-time covariant derivative with respect to the ordinary spin
˜ ˜
connection vmm n Ž e .. FmnI are the field strengths of the Abelian vector fields AmI . The
various scalar field dependent quantities that contract the different types of indices are as
˜
follows: f xã˜, g x˜ y˜ and V xã˜ b denote the n-bein,
˜ the metric and the spin connection,
˜ ˜
respectively, of the target manifold M . The quantities h I˜, h I, h aI˜˜, h aI˜ , Ta˜ b˜ c˜ and a (I˜J˜ are
x̃
w -dependent functions that are subject to various algebraic and differential constraints
Žsee Refs. w29,30x for details. as required by supersymmetry. These constraints also
˜ ˜ and imply that all scalar field dependent quantities are completely
involve f xã˜, g x˜ y˜ , V xãb
determined by the constant symmetric tensor CI˜JK ˜ ˜ that appears in the F n F n A term
in Ž2.2.. The CI˜JK ˜ ˜ thus uniquely determine the whole theory. In particular, the scalar
field target manifold M can be viewed as an n-dimensional ˜ hypersurface
˜ ˜
I J K ˜
CI˜JK
˜ ˜ h h h s1 Ž 2.5 .
˜
of an Ž n˜ q 1.-dimensional ambient space parametrized by Ž n˜ q 1. coordinates h I. The
resulting geometry of these theories was later referred to as ‘‘ very special geometry’’. In
w32x it was suggested that the compactification of 11-dimensional supergravity over a
Calabi–Yau threefold would lead to d s 5, N s 2 MESGT’s coupled to hypermulti-
plets. The Calabi–Yau compactifications of 11d supergravity were later studied in
w35–37x where it was explicitly shown that they lead to N s 2 MESGT’s with
Ž hŽ1,1. y 1. vector multiplets coupled to Ž hŽ2,1. q 1. hypermultiplets. Ž hŽ1,1. and hŽ2,1. are
the Hodge numbers of the corresponding Calabi–Yau manifold..
¨
˜ ˜ themselves are not completely arbitrary. Going to a particular basis w29,30x,

The CI˜JK
they can be brought to the following form:
C000 s 1, C0 i j s y 12 d i j , C00 i s 0 Ž 2.6 .
and the remaining coefficients Ci jk Ž i, j,k s 1,2, . . . ,n˜ . may be chosen at will. We shall
refer to this basis as the canonical basis.
The arbitrariness of the Ci jk shows that, even for a fixed number n˜ of vector
multiplets, various target manifolds M are possible. A classification of these ‘‘ very
special real’’ manifolds has been given in w38x for the case that M is a homogeneous
space. This class contains the subclass of symmetric spaces, which were classified
already long time ago w29,30,32x. Although our further discussion is not at all restricted
to symmetric Žor even homogeneous. M , we will look at the symmetric spaces in a
somewhat greater detail in Section 7. Let us therefore list the possible symmetric spaces
for later reference. The symmetric spaces M fall into two different categories, depend-
ing on whether they are associated with Jordan algebras or not:
Ži. M s Str0 ŽJ.rAutŽJ., where Str0 ŽJ. and AutŽJ. are the reduced structure group
and the automorphism group, respectively, of a formally real, unital, Jordan
algebra, J, of degree three w29,30,39x. This ‘‘Jordan class’’ can be further
divided into two subclasses:
v ‘‘Generic’’ or ‘‘reducible’’ Jordan class:

SO Ž n˜ y 1,1 . = SO Ž 1,1 .
J s R [ Sñ : Ms , n˜ 0 1.
SO Ž n˜ y 1 .
Here, Sñ is a Jordan algebra of degree two associated with a quadratic
form that has a Minkowskian signature.
v ‘‘Irreducible’’ or ‘‘magical’’ Jordan class. The corresponding Jordan
algebras are simple and are isomorphic to the Hermitian Ž3 = 3.
matrices over the four division algebras R,C,H,O with the product
being the anticommutator. They lead to the following target spaces:
J3R : M s SL Ž 3,R . rSO Ž 3 . Ž n˜ s 5 . ,
J3C : M s SL Ž 3,C . rSU Ž 3 . Ž n˜ s 8 . ,
J3H : M s SU ) Ž 6 . rUsp Ž 6 . Ž n˜ s 14 . ,
J3O : M s E 6Žy26.rF4 Ž n˜ s 26 . .
Žii. M s SO Ž1,n˜ .rSO Ž n˜ ., n˜ ) 1. This class is not associated with Jordan algebras
and will therefore be referred to as the ‘‘symmetric non-Jordan-family’’
w32x. We will now turn to the global symmetries of a generic MESGT
Žwith possibly non-symmetric or non-homogeneous M . described by
Ž2.2.. Two different global symmetries have to be distinguished:
v Any N s 2 MESGT is globally invariant under the R-symmetry group

SUŽ2.R . This symmetry is inherited from the underlying supersymme-
try algebra and acts exclusively on the fermions Cmi and li ã Ži.e. on
their index i ..
136 ¨
v Any group G of linear transformations

˜ ˜ ˜ ˜ ˜ ˜
h I ™ B JI˜ h J , AmI ™ B JI˜ AmJ
that leaves the tensor CI˜JK
˜ ˜ invariant
˜X ˜X ˜X
BI˜I B JJ˜ BKK˜ CI˜X J˜X K˜ X s CI˜JK
˜˜
is automatically a symmetry of the whole Lagrangian Ž2.2., since the
latter is uniquely determined by the CI˜JK ˜ ˜ . In particular, these symme-
tries give rise to isometries of the scalar manifolds M , which becomes
manifest if one rewrites the kinetic energy term for the scalar fields as
w29,30,38x
˜ ˜ ˜
y 12 g x˜ y˜ Ž Em w x˜ . Ž E mw y˜ . s 32 CI˜JK I J m K
˜ ˜ h Em h E h
˜
with the h I constrained according to Ž2.5..
Important Žbut not the only. examples with such a non-trivial symmetry group G are
given by the aforementioned symmetric space cases. In the Jordan class, G coincides
with the full isometry group of M Ži.e. with the full ‘‘numerator group’’ Str0 ŽJ... For the
symmetric non-Jordan family, G s w SO Ž n˜ y 1. = SO Ž1,1.x(TŽ ny1.
˜ where ( denotes the
semi-direct product and TŽ ny1.
˜ is the group of translations in an Ž n˜ y 1.-dimensional
Euclidean space. Note that for this family G is only a subgroup of the target space
isometry group SO Ž1,n˜ . w40x.
The fact that the total global symmetry group of Ž2.2. factorizes into SUŽ2.R = G is a
consequence of the SUŽ2.R invariance of the scalar fields belonging to the vector
multiplets and allows to study the gaugings of the two factors separately. In general
matter coupled extended supergravity theories the R-symmetry group is non-trivially
embedded into some larger global symmetry group if the scalar fields are not singlets
under it.
Let us now turn to the possible gaugings of subgroups of SUŽ2.R = G. Since the
vector fields are all SUŽ2.R inert, they cannot serve as non-Abelian gauge fields for the
full SUŽ2.R4 . We will therefore only consider gaugings of subgroups of UŽ1.R = G,
where UŽ1.R denotes the UŽ1. subgroup of SUŽ2.R . This obviously leaves the following
possibilities:
Ži. One can simply gauge the UŽ1.R subgroup of SUŽ2.R by coupling a linear
combination of the vector fields to the fermions w31x, which are the only fields
that transform non-trivially under SUŽ2.R . In general, this kind of gauging
Žwhich we will refer to as ‘‘gauged MESGT’’. introduces a scalar potential
Žsee Section 3..
Žii. Another possibility is to gauge a subgroup K of G. In this case, which we will
refer to as ‘‘Yang–MillsrEinstein supergravity’’, at least a subset of the
vector fields has to transform in the adjoint representation of K so that these
vector fields can serve as the corresponding Yang–Mills gauge fields. If there
are additional vector fields Ž‘spectator vector fields’. beyond these gauge
4
One could, however, try to identify SUŽ2.R with an SUŽ2. subgroup of G and then gauge this diagonal
subgroup, yet we have not considered such a possibility in the present paper.
¨
fields, there are two possibilities. They are either K-singlets or some of them
transform non-trivially under K. In the former case, there are no technical
difficulties and the gauging can be performed as described in w31x and leads to
a theory without scalar potential Žsee Section 4.
Žiii. If there are vector fields that are charged under K, one faces the same
problem that was first encountered in the context of maximally extended
gauged supergravity in seven w2x and subsequently in five dimensions w3–5x.
The problem is that a naive gauging of K would introduce masses for these
vector fields, thereby leading to a mismatch between bosonic and fermionic
degrees of freedom. The only known solution to this problem is to convert the
charged vector fields into two-form fields with ‘‘self-dual’’ field equations
w42x. In the maximally extended theories, this idea is also supported by the
analysis of the spectra of the underlying Kaluza–Klein compactifications
w14,15x.
Živ. Finally, one can combine Ži. and Žii., or alternatively Ži. and Žiii., and
simultaneously gauge both UŽ1.R and K ; G. We will refer to this type of
gauging as ‘‘gauged Yang–MillsrEinstein supergravity’’.
The first two possibilities were studied in w31,33,34x with special emphasis on the
cases were M is a symmetric space of the Jordan family. It is the purpose of this paper
to extend some of the aspects that were discussed in w31,33,34x for the gaugings of type
Ži. and Žii. to more general M and, moreover, study the so far uncovered gaugings of
type Žiii. and Živ., thereby closing a gap in the existing literature.
3. Gauged Maxwellr
r Einstein supergravity
In order to gauge the UŽ1.R subgroup of the SUŽ2.R R-symmetry group, one
˜
promotes a linear combination of the Ž n˜ q 1. vector fields AmI to the UŽ1.R-gauge field
˜
Am U Ž 1 . R s VI˜ AmI , Ž 3.1 .
where VI˜ are Ž n˜ q 1. constants, and replaces the derivatives of the fermionic fields by
UŽ1.R-covariant derivatives
i
=m l ai˜ ™ Ž Dm l a˜ . ' =m l ai˜ q g R VI˜ AmI˜ d i jl aj˜ ,
i
=mCn i ™ Ž DmCn . ' =mCn i q g R VI˜ AmI˜ d i j Cn j , Ž 3.2 .
where g R denotes the UŽ1.R-coupling constant. The appearance of the d i j is due to the
convention that the SUŽ2.R indices i, j, . . . are raised and lowered with the antisymmet-
ric metric ´ i j s y´ ji , ´ 12 s ´ 12 s 1 w29–31x. This UŽ1.R covariantization in the La-
grangian Ž2.2. and the transformation laws Ž2.3. breaks the original supersymmetry. In
order to restore it, some g R-dependent gauge invariant terms have to be added. The
138 ¨
additional terms in the Lagrangian are w31,34x Žthe numerical factors are chosen for
convenience.
i'6 1
ey1 L X s y g RCmiG mnCn jd i j P0 Ž w . y g R li a˜G m iCmj d i j Pã Ž w .
8 '2
i ˜
q g R li a˜l jbd i j Pa˜ b˜ Ž w . y g R2 P Ž R. Ž w . , Ž 3.3 .
2'6
whereas the transformation laws have to be modified by
i
d XCmi s g R P0 Ž w . Gm d i j´ j ,
2'6
1
d Xli a˜ s g R P a˜ Ž w . d i j´ j . Ž 3.4 .
'2
The new scalar field dependent quantities P0 , P ã, Pa˜ b˜ , and the scalar potential P Ž R. are
fixed by supersymmetry
˜
P a˜ s '2 h a˜ I VI˜, Ž 3.5 .
˜
P0 s 2 h I VI˜, Ž 3.6 .
1
Pa˜ b˜ s d a˜ b˜ P0 q 2'2 Ta˜ b˜ c˜ P c̃ ,
2 Ž 3.7 .
2
P Ž R. s y Ž P0 . q P a˜P a˜ . Ž 3.8 .
Ž R.
The scalar potential P can be written in the form w31,34x
˜˜ ˜
P Ž R. s y4C I JK VI˜VJ˜ h K˜ , Ž 3.9 .
where the I,˜ J,
˜ K˜ are raised with the inverse of a (I˜J˜. In our metric signature, a critical
point wc of the scalar potential with P Ž R. Ž wc . - 0 corresponds to an anti-de Sitter
ground state. The critical points of the potential Ž3.8. have been analyzed in w31,34x for
the Jordan cases. If a critical point exists, it was found that, depending on the linear
combination Ž3.1. of the vector fields, one either gets an N s 2 supersymmetric anti-de
Sitter ground state, or the scalar potential vanishes identically, and thus admits a
Minkowski vacuum with spontaneously broken supersymmetry.
4. Yang–Mills r Einstein supergravity without tensor fields
We now consider the gauging of a subgroup K of G. As mentioned earlier, this type

requires that a subset AmI ; I, J, . . . s 1, . . . dim K 4 of the vector fields transforms in the
adjoint representation of K. In this section, we assume that if there are additional
spectator vector fields AmM ; M, N, P s 1, . . . ,Ž n˜ q 1. y dim K 4 , they are all K-singlets
Ži.e. we are dealing with the gauging of type Žii...
The only fields that transform under K are the scalar fields w x̃ , the spinor fields li ã
and the vector fields AmI , Ž I s 1, . . . dim K .. The K-covariantization is thus achieved by
¨
replacing the corresponding derivativesrfield strengths by their K-gauge covariant

counterparts:
Em w x̃ ™ Dm w x̃ ' Em w x̃ q gAmI K Ix̃ ,
=m li a˜ ™ Dm li a˜ ' =m li a˜ q gAmI LaI˜ b˜li b˜ ,

FmnI ™ FmnI ' FmnI q g f JK
I
AmJ AnK . Ž 4.1 .
Here, g denotes the K-coupling constant, K Ix̃ are the Killing vectors of M that
˜
correspond to the subgroup K of its isometry group G Žcf. w31x., LaI˜ b are the Žscalar field
i ã I
dependent. K-transformation matrices of the fermions l Žcf. w31,33x. and f JK are the
structure constants of K. These replacements in the Lagrangian Ž2.2. and the transforma-
tion laws Ž2.3. are subject to one exception: The proper gauge-covariantization of the
y1 mnrsl I˜ ˜ ˜
F n F n A term in Ž2.2. leads to a Chern Simons term, i.e. e CI˜JK ˜ ˜´ Fmn FrsJ AlK
'
6 6
gets replaced by
ey1 mnrsl ˜ ˜ ˜ ˜ ˜ ˜ ˜ ˜
˜ ˜´
CI˜JK FmnI FrsJ AlK q 32 gFmnI ArJ Ž f L˜KM˜ AsL AlM .
½
6'6
˜ ˜ ˜
q 35 g 2 f NJ˜P˜ AnN ArP
ž / Ž f ˜˜ ˜ A ˜ A ˜ . A ˜ 5 ,
K
LM
L
s
M
l
I
m Ž 4.2 .
˜
where it is understood that f I˜KJ˜ is zero whenever one of the indices I,˜ J,
˜ K˜ corresponds
M
to one of the spectator vector fields Am .
Again, supersymmetry is broken by these replacements. This time, however, its
restoration requires little modification; the Žcovariantized. transformation laws remain
unchanged, and only a Yukawa-like term has to be added to the Žcovariantized.
Lagrangian w31,33x
i ˜
L X s y g li a˜l bi K I w a˜ h bI˜ x . Ž 4.3 .
2
In particular, no scalar potential is introduced so that only Minkowski ground states are
possible.
5. Yang–Mills r Einstein supergravity with tensor fields
We now turn to case Žiii. of our gauge type classification and consider the gauging of
K ; G, when not all the spectator vector fields are K-singlets. As mentioned earlier,
consistency with supersymmetry requires that these K-charged spectator vector fields
have to be dualized to ‘‘self-dual’’ two-form fields w3–5,14,15,42x. We will therefore
˜
split the vector fields AmI of the ungauged theory Ž2.2. – Ž2.3. of Section 2 into two sets.
The first set contains the vector fields in the adjoint representation of the gauge group K
plus possible K-singlets. The second set contains the remaining K-charged vector
fields. We will use indices I, J, K, . . . s 1, . . . ,n for the first and M, N, P, . . . s 1, . . . 2 m
for the second set, where n q 2 m s n˜ q 1. The reason for the even number 2 m is that
the ‘‘self-duality’’ condition of Ref. w42x requires complex tensor fields for d s 5, which
we will always consider as being decomposed into their real and imaginary parts. The
140 ¨
gauging now proceeds as follows. First, one has to replace all Abelian field strengths
FmnI by the corresponding non-Abelian generalizations FmnI ' FmnI q gf JK
I
AmJ AnK , with f JK
I
5 M
being the structure constants of K , and the Fmn by the above-mentioned ‘‘self-dual’’
M
two-form fields Bmn :
˜ ˜
FmnI ™ HmnI :s Ž FmnI , Bmn
M
.. Ž 5.1 .
Again, the only exception to this replacement is the F n F n A term of the ungauged
theory. Since no ‘naked’ AmM can appear anymore, we first require
CM N P s 0 Ž 5.2 .
M I J
and since terms of the form B n F n A appear to be impossible to supersymmetrize
in a gauge invariant way Žexcept possibly in very special cases. we shall also assume
that
CM I J s 0. Ž 5.3 .
Hence, the only non-vanishing CI˜JK ˜ ˜ have the index structure CI JK and C IM N . The
covariantization of the CI JK F I n F J n A K term again leads to a Chern–Simons term
Žsee below.. The term of the form CIM N A I n B M n B N has its natural place in the
M Ž
gauge-invariant kinetic energy term for the tensor fields Bmn cf. Eqs. Ž5.4. and Ž5.6...
The gauge covariant derivative of these tensor fields reads
Dm BnrM ' =m BnrM q gAmI L IN
M
BnrN , Ž 5.4 .
M
where the constant matrices L IN
are the corresponding representation matrices of K.
The remaining gauge covariantizations involve the scalar and spinor fields, for which
we again make the replacements
Em w x̃ ™ Dm w x̃ ' Em w x̃ q gAmI K Ix̃ ,
=m li a˜ ™ Dm li a˜ ' =m li a˜ q gAmI LaI˜ b˜li b˜ . Ž 5.5 .

After all these modifications, the original supersymmetry of the ungauged theory
Ž2.2. – Ž2.3. is again badly broken. This time, however, the supersymmetry breaking is
not only due to the gauge covariantization alone. An additional source for the break-
down of supersymmetry is provided by the loss of the Bianchi identity for the tensor
M Ž ˜
fields Bmn i.e. dB M / 0 in general.. The corresponding Bianchi identity dF I s 0 for the
I˜
Fmn in the ungauged theory is needed at several places to cancel certain supersymmetry
variations in Ž2.2..
Remarkably enough, supersymmetry can again be restored by adding further g-de-
pendent gauge invariant terms to the Lagrangian and the transformation laws. This
procedure is very similar to what had to be done in the N s 8 theory w3–5x. We omit
the details here and quote the final result.
The Lagrangian is given by Žup to 4-fermion terms.
( ˜ ˜
ey1 L s y 12 R Ž v . y 12 CmiG mnr =nCr i y 14 a I˜J˜HmnI H Jmn
˜ ˜ ˜
y 12 li a˜ G m Dm d a˜ b q V xã˜ bG m Dm w x˜ l bi y 12 g x˜ y˜ Ž Dm w x˜ . Ž D mw y˜ .
ž /
5
In the presence of K-singlets, the corresponding f JIK are again assumed to be zero Žcf. Section 4.
¨
i ˜
y li a˜G mG nCm i f xã˜ Dn w x˜ q 14 h aI˜˜li a˜G mG l rCm i HlIr
2
i i a˜ mn b˜
q 1 c˜
4 d a˜ b˜ h I˜q Ta˜ b˜ c˜ h I˜ l G
ž l i HmnI˜
/
'
2 6
3i ˜ ˜
y h I˜ CmiG mnrsCn i HrsI q 2C m iCin HmnI
8'6
ey1
q CI JK ´ mnrsl FmnI FrsJ AlK q 32 gFmnI ArJ Ž f LKF AsL AlF .
½
6'6
ey1
q 35 g 2 Ž fGJ H AnG ArH . Ž f LKF AsL AlF . AmI q 5 ´ mnrslV M N Bmn
M N
Dr Bsl
4g
˜ ˜
q g li a˜G m Cm iW a˜ q g li a˜l bi W a˜ b y g 2 P . Ž 5.6 .
The transformation laws are Žto leading order in fermion fields.
d emm s 12 ´ iG m Cm i ,
i nr ˜
dCm i s =m Ž v . ´ i q h I˜ Ž Gm y 4dmn G r . HnrI ´ i ,
4'6
d AmI s qmI ,
'6 g ig
M
d Bmn s 2 Dw mqn Mx q V M N h NC w im Gn x ´ i q V M N h aN˜ li a˜Gmn ´ i ,
4 4
i ˜
dl ai˜ s y f xã˜ G m Ž Dm w x˜ . ´ i q 14 h aI˜˜G mn´ i HmnI q gW a˜´ i ,
2
i
dw x˜ s f a˜x˜´ il ai˜ Ž 5.7 .
2
with
i'6
qmI˜' y 12 h aI˜˜ ´ iGm l ai˜ q ˜
h I Cmi´ i . Ž 5.8 .
4
The quantities which are not already present in the ungauged theory are a Žconstant.
real symplectic metric V M N
V M N s yV N M , V M N V N P s d MP , Ž 5.9 .
˜
two tensors W a˜Ž w . and W a˜ b Ž w .
'6
W a˜ s y h aM˜ V M N h N ,
8
˜ ˜ ˜ i'6 ˜
W a˜ b s yW b a˜ s ih J w a˜K Jb x q h J K Jw a˜ ; b x , Ž 5.10 .
4
142 ¨
where the semicolon denotes covariant differentiation on the target space M , and a
scalar potential P Ž w .
P s 2W ã W ã . Ž 5.11 .
Furthermore, one finds the relation
N
2 2
L IM s V N P CM P I m V N P L IM
P
s CM N I , Ž 5.12 .
'6 '6
N
which, because of CM N I s CN M I , means that the L IM have to form a symplectic
representation of the gauge group K. Supersymmetry also requires the following two
relations:
'6
W a˜ s h J K Ja˜ , Ž 5.13 .
4
˜ ˜
P , a˜ s 4 iW a˜ b W b , Ž 5.14 .
where the comma denotes partial differentiation with respect to the scalar fields. These
last two conditions, however, can be shown to follow automatically from the various
other constraints.
The above scalar potential P Ž w . deserves some comments.
First of all, it is a bit surprising that there is a scalar potential at all, since no minimal
couplings to the gravitini have been introduced at this point, and, as we have seen in
Section 4, the pure Yang–MillsrEinstein supergravity theories without antisymmetric
tensor fields do not involve a scalar potential. In fact, the necessity for the scalar
potential in the above Lagrangian can eventually be traced back to the loss of the
M
Bianchi identity for the Bmn , which are not present in the theories considered in Section
4.
The second important point about the potential is its sign. As mentioned at the end of
Section 3, in our metric signature, a critical point with P Ž wc . - 0 would correspond to
an anti-de Sitter solution. The explicit form Ž5.11. of our potential, however, is
manifestly non-negative. Therefore, the N s 2 Yang–MillsrEinstein supergravity theo-
ries with tensor multiplets do not admit an anti-de Sitter solution. This might at first
seem surprising, since it was, among other things, the representation theory of the AdS 5
superalgebra SUŽ2,2 <4. that hinted towards the dualization of twelve vector fields to
antisymmetric tensor fields in the gauging of the N s 8 supergravity theory in d s 5
w3–5x. For N s 2 however, this argument does not apply anymore, since the N s 2
anti-de Sitter graviton supermultiplet also contains only one vector field and no tensor
fields, giving rise to the same field content as its N s 2 super Poincare´ counterpart.
Thus, for N s 2, the antisymmetric tensor fields do not necessarily have to be
associated with anti-de Sitter space-times anymore.
6. Gauged Yang–Mills r Einstein supergravity with tensor fields
We will now come to case Živ. of our list of possible gaugings and simultaneously
gauge the UŽ1.R R-symmetry subgroup and a subgroup K of G. We will do this for the
most general case with tensor fields, since the case without tensor fields can easily be
¨
recovered as a special case. Our starting point will be the Yang–MillsrEinstein

supergravity with tensor fields presented in Section 5, i.e. Eqs. Ž5.6. – Ž5.7..
As in Section 3, we will take a linear combination of the vector fields AmI as the
UŽ1.R-gauge field
Am U Ž 1 . R s VI AmI Ž 6.1 .
with some constants VI , which at this point are completely arbitrary. ŽNote, however,
that we do not sum over I˜ like in Section 3, i.e. ‘‘VM s 0’’.. The gauging of UŽ1.R then
obviously requires the UŽ1.R-covariantization of all fermionic derivatives,
i
Dm l ai˜ ™ Ž D m l a˜ . ' Dm l ai˜ q g R VI AmI d i jl aj˜ ,
i
=mCn i ™ Ž D mCn . ' =mCn i q g R VI AmI d i j Cn j , Ž 6.2 .
where g R again denotes the UŽ1.R-coupling constant and Dm is the K-covariant
derivative introduced in Ž5.5.. Again, this gauge covariantization breaks supersymmetry,
and to restore it, new g R-dependent terms have to be added to the Lagrangian and the
transformation laws.
The additional terms in the Lagrangian are
i'6 1
ey1 L X s y g RCmiG mnCn jd i j P0 Ž w . y g R li a˜G m Cmjd i j Pã Ž w .
8 '2
i ˜
q g R li a˜l jbd i j Pa˜ b˜ Ž w . y g R2 P Ž R. Ž w . , Ž 6.3 .
2'6
whereas the transformation laws have to be modified by
i
d XCmi s g R P0 Ž w . Gm d i j´ j ,
2'6
1
d Xli a˜ s a˜ ij
'2 g R P Ž w . d ´ j . Ž 6.4 .
The new scalar field dependent quantities P0 , P ã, Pa˜ b˜ , and the scalar potential P Ž R. are
fixed by supersymmetry
P a˜ s '2 h a˜ I VI , Ž 6.5 .
I
P0 s 2 h VI , Ž 6.6 .
Pa˜ b˜ s 12 d a˜ b˜ P0 q 2'2 Ta˜ b˜ c˜ P c̃ , Ž 6.7 .
2
P Ž R. s y Ž P0 . q P a˜P a˜ . Ž 6.8 .
Furthermore, the VI are constrained by
I
VI f JK s 0. Ž 6.9 .
Supersymmetry also requires the relations
P ã K Iã s 0,
˜ ˜
Pa˜ b˜W b s yi2'3 Wa˜ b˜ P b q 52 Wa˜ P0 ,
y'3 1 ˜
P ;a˜x˜ s P0 f xã˜ y Pa˜ b˜ f x˜b ,
4 2'3
144 ¨
2
P0, x˜ s y P a˜ f xã˜ ,
'3
5 1 ˜
P ,Žx˜R. s P0 P a˜ f xã˜ y f xã˜ Pa˜ b˜ P b , Ž 6.10 .
2'3 '3
however, these can be shown to be consequences of the other constraints and therefore
do not give rise to additional restrictions.
It should be noted that the constraints Ž6.5. – Ž6.8. are almost the same as in the case
of the pure UŽ1.R-gauging described in Section 3. Yet there are two important
differences. The first is that the Žcompletely arbitrary. VI˜ of Section 3 are now subject to
two constraints, namely Eq. Ž6.9. and ‘‘VM s 0’’, which is merely a trivial consequence
of Ž6.1..
The second difference is that Ž6.8. is not the full scalar potential. The latter is now a
sum of the UŽ1.R-related potential P Ž R. and the potential P, which was due to the
introduction of the 2-form fields Žcf. Eqs. Ž5.6. and Ž5.11..,
ey1 Lpot s yg 2 P y g R2 P Ž R. Ž 6.11 .
These differences have some interesting implications:
Eq. Ž6.9. gives a new constraint on the possible gauge groups K, since for it to be
I
true, the f JK have to admit a non-trivial eigenvector VI with eigenvalue 0. This means
that either there has to be at least one spectator vector field AmI or K has to have at least
one Abelian factor Žboth of them together could also be true..
As for the potential, one sees that the UŽ1.R-gauging introduces a negative contribu-
tion to the total scalar potential so that anti-de Sitter solutions might now be possible. In
fact, the experience with the gauged MESGT’s in w31,34x and certain truncations of the
N s 8 theory w4x make this possibility quite plausible. A more detailed analysis of the
potential and its critical points, however, is now complicated by the additional scalar
potential term yg 2 P induced by the tensor field dualization and the additional
constraints on the VI˜ and will therefore be given elsewhere w41x.
7. Allowed gauge groups and the corresponding representations of the tensor

multiplets
In this section we will give a partial classification of the possible gauge groups and
the representations under which the tensor fields transform. An attempt at a complete
classification will be made elsewhere w41x.
We will start our discussion of possible gauge groups with the ‘‘Magical’’ supergrav-
ity theories defined by simple Jordan algebras of degree 3.
Ži. The largest of the magical N s 2 supergravity theories is defined by the

exceptional Jordan algebra with the scalar manifold E6Žy 26.rF4 , which we shall
refer to as the exceptional supergravity theory. The exceptional supergravity
theory and its counterparts in four and three dimensions share many of the
remarkable properties of the maximally extended supergravity theories in the
¨
respective dimensions. In the exceptional theory one can gauge the SO ) Ž6. s
SUŽ3,1. subgroup of the isometry group E6Žy 26. of the scalar manifold while
dualizing twelve of the vector fields into tensor fields that form a symplectic
representation Ž6 q 6. of SO ) Ž6.. The pure maximal Yang–Mills Einstein
subsector of this theory is the unique unified Yang–Mills Einstein supergravity
in five dimensions that was studied in w33x. To gauge a UŽ1.R subgroup of the
R-symmetry group SUŽ2.R one needs to break the non-Abelian gauge group
SUŽ3,1. down to a subgroup. One possibility is to gauge the UŽ1.R such that the
SUŽ3. subgroup of SUŽ3,1. is unbroken. In this case we obtain a gauged
Yang–Mills Einstein supergravity theory with the gauge group UŽ1.R = SUŽ3.
and 18 tensor multiplets in the symplectic representation Ž3 q 3 q 3 q 3 q 3 q
3. of SUŽ3.. The subsector of this theory involving only 6 tensor multiplets
corresponds to the N s 2 truncation of the gauged N s 8 theory with the
gauge group SUŽ3. = UŽ1.R w4x which admits an AdS ground state. One can
also gauge the subgroup UŽ1.R such that one has a vanishing potential P Ž R.. In
this case the unbroken non-Abelian symmetry is the SUŽ2,1. subgroup of
SUŽ3,1. with 18 tensor multiplets.
Žii. The magical N s 2 MESGT defined by the Jordan algebra of 3 = 3 Hermitian
matrices over the quaternions has the scalar manifold SU ) Ž6.rUSpŽ6.. One
can gauge the SO ) Ž6. subgroup of the isometry group resulting in the unique
unified Yang–Mills Einstein supergravity w33x with no tensor multiplets. To
obtain a Yang–Mills Einstein supergravity with tensor multiplets one has to
gauge a subgroup of SO ) Ž6.. One can gauge the maximal compact subgroup
SUŽ3. = UŽ1. of SO ) Ž6. by dualizing 6 of the vector fields to tensor fields
transforming in the symplectic representation Ž3 q 3. of SUŽ3. = UŽ1.. One can
similarly gauge the non-compact subgroup SUŽ2,1. = UŽ1. of SO ) Ž6.. In both
cases one can use the gauge field associated with the Abelian factor to gauge
the UŽ1.R symmetry thereby obtaining gauged Yang–MillsrEinstein supergrav-
ity theories with the non-Abelian gauge groups SUŽ3. and SUŽ2,1. and six
tensor multiplets, respectively. We expect the generic SUŽ2,1. = UŽ1.R gauging
to lead to a vanishing potential P Ž R..
Žiii. The magical MESGT defined by the Jordan algebra of 3 = 3 Hermitian
matrices over the complex numbers has the scalar manifold SLŽ3,C.rSUŽ3.. In
this theory one can gauge the full compact symmetry group to obtain a
Yang–MillsrEinstein supergravity theory with the gauge group SUŽ3. w31x.
The remaining vector field Žgraviphoton. can be used to gauge the UŽ1.R
symmetry with a non-vanishing potential and an AdS ground state. To obtain a
Yang–Mills Einstein supergravity with tensor multiplets one needs to gauge a
subgroup of SUŽ3.. One can, for example, gauge the SUŽ2. = UŽ1. subgroup
while dualizing four of the vector fields to tensor fields in the symplectic
representation Ž2 q 2.. One can then use the graviphoton to obtain a UŽ1.R
gauged version of this theory. Non-compact analogs of these theories also exist
with SUŽ3. and SUŽ2. replaced by SLŽ3,R. and SLŽ2,R., respectively.
Živ. The smallest of the magical MESGT’s has the scalar manifold SLŽ3,R.rSO Ž3..
In this case one can gauge the SLŽ2,R. subgroup of the isometry group while
dualizing two of the vector fields into tensor fields. The remaining vector field
can be used to gauge the UŽ1.R symmetry.
146 ¨
Žv. For the generic Jordan family the scalar manifold of the N s 2 MESGT is
SO Ž n˜ y 1,1 . = SO Ž 1,1 . rSO Ž n˜ y 1 . . Ž 7.1 .

On the other hand, the scalar manifold of the generic symmetric non-Jordan
family is of the form
˜ . rSO Ž n˜ . .
SO Ž n,1 Ž 7.2 .
For the latter family, not all the isometries of the scalar manifold can be
extended to symmetries of the Lagrangian w40x. Only the subgroup w SO Ž n˜ y 1.
= SO Ž1,1.x(TŽ ny1.
˜
Ži.e the Euclidean group in Ž n˜ y 1. dimensions times
˜ . extends to a full symmetry of the action. This can
dilatations. of SO Ž n,1
simply be understood by the fact that there is no irreducible symmetric
˜ ..
invariant tensor of rank three of SO Ž n,1
One can treat the generic Jordan and non-Jordan families in a unified manner as
was shown in w32x. Consider a vector m in an n-dimensional
˜ Euclidean space
with components m i . Then the non-vanishing components of the tensor CI˜JK ˜˜
can be written in the form Žcf. Eq. Ž2.6..
C000 s 1,
C0 i j s y 12 d i j ,
Ci jk s 23 mŽ i d jk . y m i m j m k Ž 7.3 .
˜ For the generic Jordan family the length squared of

where i, j, . . . s 1,2, . . . ,n.
the vector m is two
mPms2 Ž 7.4 .
hile for the non-Jordan family one has
m P m s 12 . Ž 7.5 .
It is easy to verify that the above Ci jk can provide a symplectic representation
of only an Abelian subgroup of the compact symmetry group of N s 2
MESGT. Therefore, if we are to have tensor fields transforming non-trivially
under the compact gauge group then only products of UŽ1.’s are allowed in the
Yang–Mills Einstein supergravity with tensor multiplets. Of course one still has
the option to gauge a non-Abelian subgroup of the compact symmetry group so
long as the tensor fields are inert under it.
Žvi. The scalar manifolds listed above exhaust the list of N s 2 MESGT’s whose
scalar manifolds are symmetric spaces. In addition there is a large set of other
theories whose scalar manifolds admit isometries that extend to symmetries of
the full action. These include theories whose scalar manifolds are homogeneous
spaces as well as those that are not homogeneous. A complete list of possible
homogeneous spaces was given in w38x. This classification was achieved by
showing that the requirement of a transitive isometry group allows one to bring
¨
the most general solution for the symmetric tensor given in the ‘‘canonical
basis’’ above to the form
C011 s 1,
C0 mn s yd mn ,
C1 i j s yd i j ,
C m i j s gm i j Ž 7.6 .
where the indices I˜ are now split such that I˜s 0,1, m ,i with m s 1,2, . . . ,q q 1
and i s 1,2, . . . ,r. The coefficients gm i j are Ž q q 1. real r = r matrices that
generate a real Clifford algebra of positive signature C Ž q q 1,0.. The allowed
homogeneous Žbut not symmetric. spaces are, in general, quotients of ‘‘para-
bolic groups’’ G modded out by their maximal compact subgroups H. The Lie
algebra g of the group G is a semi-direct sum,
g s g 0 [ gq1 , Ž 7.7 .
g 0 s so Ž 1,1 . [ so Ž q q 1,1 . [ Sq Ž P ,Q . ,
gq1 s Ž spinor, vector . , Ž 7.8 .
where spinor denotes a spinor representation of soŽ q q 1,1. Žof dimension
Dqq 1 . and Õector denotes the vector representation of Sq Ž P,Q . which is of
dimension Ž P q Q ..6 The isotropy group H is
H s SO Ž q q 1 . m Sq Ž P ,Q . . Ž 7.9 .
The possible groups Sq Ž P,Q . and the associated real Clifford algebras were
given in w38x which we list in Table 1.
Now the gamma matrices g m i j provide a symplectic representation of a group
only for q s 1 or q s 2 i.e for UŽ1. or SUŽ2.. Hence one can gauge SUŽ2.
symmetry of the N s 2 MESGT for q s 2 while dualizing the 2 P vector
fields to tensor fields. One can then use the remaining two SUŽ2. singlet
vector fields to gauge the UŽ1.R symmetry andror the Abelian UŽ1. factor in
UŽ P . s UŽ1. = SUŽ P .. For q s 1 one can gauge the SO Ž2,1. symmetry while
dualizing the 2 P vector fields into tensor fields. The remaining SO Ž2,1.
singlet vector field can then be used to gauge the UŽ1.R symmetry of these
theories.
Žvii. As is clear from above, the coupling to the tensor fields restricts the possible
non-Abelian symmetry groups greatly for those theories whose scalar mani-
folds are symmetric spaces or homogeneous spaces. We would like to point
out that there does exist a novel class of Žgauged. N s 2 Yang–Mills Einstein
supergravity theories coupled to tensor multiplets with a rich set of possible
non-Abelian groups that admit symplectic representations. To construct these
6
We should note that in case the scalar manifold is a symmetric space the above Lie algebra gets extended
by additional symmetry generators belonging to grade y1 space transforming in the conjugate representation
of gq1 with respect to g 0 .
148 ¨
Table 1
Real Clifford algebras C Ž q q1,0.. R, C and H are the division algebras of real, complex numbers and
quaternions, respectively, while Dqq 1 denotes the real dimension of an irreducible representation of the
Clifford algebra. The Sq Ž P,Q . is the metric preserving group in the centralizer of the Clifford algebra in the
Ž P q Q . Dqq 1-dimensional representation
q C Ž q q1,0. Dqq 1 Sq Ž P,Q .
y1 R 1 SO Ž P .
0 R[R 1 SO Ž P .mSO Ž Q .
1 RŽ2. 2 SO Ž P .
2 CŽ2. 4 UŽ P .
3 HŽ2. 8 USpŽ2 P .
4 HŽ2.[HŽ2. 8 USpŽ2 P .mUSpŽ2 Q .
5 HŽ4. 16 USpŽ2 P .
6 CŽ8. 16 UŽ P .
7 RŽ16. 16 SO Ž P .
nq8 RŽ16.mC
C Ž nq1,0. 16 Dn as for q s n
theories one simply chooses the arbitrary tensor Ci jk in the canonical basis
Ž2.6. as follows. Split the indices i, j,k . . . as i s Ž i, M ., j s Ž j, N ., . . . where
i, j s 1,2, . . . ,n y 1 and M, N, . . . s 1, . . . ,2 m and identify
Ci jk s d i jk Ž 7.10 .
3
Ci M N s CM i N s CM N i s ( 2 Ž L i . M N s ( 32 Ž L i . N M , Ž 7.11 .
where d i jk are the completely symmetric Gell-Mann d-symbols of a Lie
group K and the Ž L i . are the matrices of a 2 m-dimensional symplectic
representation of K. Now the d-symbols vanish for all simple groups except
for the groups SUŽ N ., N ) 2 and SpinŽ6. which is isomorphic to SUŽ4.. For
vanishing d-symbols and m s 0 the cubic form reduces to that of the generic
Jordan family. Thus the theories defined by non-vanishing Ci jk s d i jk can be
considered as the non-trivial generalizations of the generic Jordan family. The
first non-trivial example i.e the case of K s SUŽ3. d-symbols Žwith m s 0.
lead to the magical N s 2 MESGT with the scalar manifold SLŽ3,C.rSUŽ3..
The scalar manifold obtained by taking Ci jk to be the d-symbols of SUŽ N .
for N ) 3 cannot be a symmetric or homogeneous space. This follows from
the fact that for such theories SUŽ N . act as isometries of the scalar manifold
that extend to symmetries of the full Lagrangian. However,it is clear from the
list of possible homogeneous spaces w38x that it does not include manifolds
with such properties. Hence the isometries of the scalar manifolds correspond-
ing to Ci jk s d i jk for N ) 3 in the canonical basis cannot act transitively. This
is perhaps expected from the fact that the number of independent invariants of
a group in its adjoint representation is equal to its rank i.e. the number of
Casimir operators. The term involving d i j in the cubic form corresponds to
the quadratic invariant and the term involving d i jk corresponds to the third
order Casimir. Only for SUŽ3. do they form a complete set of invariants and
the resulting scalar manifold is a symmetric space. For higher SUŽ N . Ž N ) 3.
one has invariants of order up to N.
¨
As for the symplectic representations L i of SUŽ N ., one can, for example,

choose the reducible Ž N q N . representations corresponding to the standard
embedding of UŽ N . in USpŽ2 N . by taking m s N and n s N 2 .
8. Conclusions
Our results imply several interesting conclusions.

Whereas the R-symmetry group and the isometry group G of the scalar manifold M
are entangled with each other for N ) 2 MESGT’s, and for simple supergravities for
N ) 4 the case N s 2 allows a separate discussion of the gaugings of subgroups of
these two groups. In particular, the issues of the tensor field dualization and the
gravitino coupling to gauge fields can be completely separated. It turns out that both
mechanisms require their own scalar potential. The potential due to the introduction of
the tensor fields is manifestly non-negative and does therefore not admit an anti-de Sitter
solution. This is in contrast to the pure UŽ1.R gauging, which involves minimal coupling
to the gravitini and leads to an indefinite potential which can sustain anti-de Sitter vacua.
Combining both types of gauging, one observes surprisingly little interference. In
particular, the scalar potential is just a sum of the two potentials of the individual
gaugings. Nevertheless, the analysis of the critical points seems to be more complicated,
but is, in general, expected to allow anti-de Sitter solutions w41x. A particular example of
such a theory obtained by a truncation of the gauged N s 8 supergravity does admit an
AdS vacuum w4x.
The introduction of the tensor fields leads to strong constraints on the possible gauge
groups K ; G and the representations under which the tensor fields transform. The
simultaneous UŽ1.R gauging further restricts these gauge groups K, which is one of the
few places where these two types of gaugings interfere with each other. We gave a list
of possible gauge groups and the corresponding representations of the tensor fields using
the known classification of N s 2 MESGT’s whose scalar manifolds are symmetric or
homogeneous spaces. We also pointed out the existence of a novel family of N s 2
MESGT’s whose scalar manifolds are, in general, not homogeneous, but admit SUŽ N .
isometries. The latter class of theories lead to a richer class of gaugings with some of the
vector fields dualized to tensor fields.
Acknowledgements
We would like to thank Sergio Ferrara, Renata Kallosh, Raymond Stora and Antoine
van Proeyen for useful discussions.
References
w1x B. de Wit, H. Nicolai, Nucl. Phys. B 208 Ž1982. 323.

w2x M. Pernici, K. Pilch, P. van Nieuwenhuizen, Phys. Lett. B 143 Ž1984. 103.
w3x M. Gunaydin,
¨ L.J. Romans, N.P. Warner, Phys. Lett. B 154 Ž1985. 268.
150 ¨
w4x M. Gunaydin,
¨ L.J. Romans, N.P. Warner, Nucl. Phys. B 272 Ž1986. 598.
w5x M. Pernici, K. Pilch, P. van Nieuwenhuizen, Nucl. Phys. B 259 Ž1985. 460.
w6x For further references on the vast subject of gauged supergravity theories see the reprint volumes,
Supergravity in Different Dimensions, ed. A. Salam, E. Sezgin ŽWorld Scientific, Singapore, 1989..
w7x J. Maldacena, Adv. Theor. Math. Phys. 2 Ž1998. 231. hep-thr9711200.
w8x S. Gubser, I.R. Klebanov, A.M. Polyakov, Phys. Lett. B 428 Ž1998. 105. hep-thr9802109.
w9x E. Witten, Adv. Theor. Math. Phys. 2 Ž1998. 253. hep-thr9802150.
w11x For the relationship between the earlier work on Kaluza–Klein supergravity theories and the Maldacena
¨
conjecture see M. Gunaydin, D. Minic, Nucl. Phys. B 253 Ž1998. 145, hep-thr9802047.
w12x M.J. Duff, H. Lu,¨ C. Pope, hep-thr9803061.
w13x For an extensive list of references on AdSrCFT dualities see the recent review paper O. Aharony, S.S.
Gubser, J. Maldacena, H. Ooguri, Y. Oz, hep-thr9905111.
w14x M. Gunaydin,
¨ N. Marcus, Class. Quantum Gravity 2 Ž1985. L11.
w15x H.J. Kim, L.J. Romans, P. van Nieuwenhuizen, Phys. Rev D 32 Ž1985. 389.
w17x H. Nastase, D. Vaman, P. van Nieuwenhuizen, hep-thr9905075.
w18x D.Z. Freedman, S.S. Gubser, K. Pilch, N.P. Warner, hep-thr9904017.
w19x For a recent overview see N.P. Warner, hep-thr9911240.
w20x L. Randall, R. Sundrum, hep-thr9905221.
w21x L. Randall, R. Sundrum, hep-thr9906064.
w22x K. Behrndt, M. Cvetic, ˇ hep-thr9909058.
w23x P. Horava,
ˇ E. Witten, Nucl. Phys. B 460 Ž1996. 506. hep-thr9510209.
w24x P. Horava,
ˇ E. Witten, Nucl. Phys. B 475 Ž1996. 94. hep-thr9603142.
w26x A. Lukas, B. A Ovrut, K.S. Stelle, D Waldram, hep-thr9803235.
w27x A. Lukas, B. A Ovrut, K.S. Stelle, D Waldram, hep-thr9806051.
w28x J. Ellis, Z. Lalak, S. Pokorski, W. Pokorski, Nucl. Phys. B 540 Ž1999. 149. hep-phr9805377.
w29x M. Gunaydin,
¨ G. Sierra, P.K. Townsend, Nucl. Phys. B 242 Ž1984. 244.
w30x M. Gunaydin,
¨ G. Sierra, P.K. Townsend, Phys. Lett. B 133 Ž1983. 72.
w31x M. Gunaydin,
¨ G. Sierra, P.K. Townsend, Nucl. Phys. B 253 Ž1985. 573.
w32x M. Gunaydin,
¨ G. Sierra, P.K. Townsend, Class. Quantum Grav. 3 Ž1986. 763.
w33x M. Gunaydin,
¨ G. Sierra, P.K. Townsend, Phys. Rev. Lett. 53 Ž1984. 332.
w34x M. Gunaydin,
¨ G. Sierra, P.K. Townsend, Phys. Lett. B 144 Ž1984. 41.
w35x A.C. Cadavid, A. Ceresole, R. D’Auria, S. Ferrara, Phys. Lett. B 357 Ž1995. 76. hep-thr9506144.
w36x I. Antoniadis, S. Ferrara, T.R. Taylor, Nucl. Phys. B 460 Ž1996. 489. hep-thr9511108.
w37x G. Papadopoulos, P.K. Townsend, Phys. Lett. B 357 Ž1995. 300. hep-thr9506150.
w38x B. de Wit, A. van Proeyen, Commun. Math. Phys. 149 Ž1992. 307. hep-thr9112027.
w39x K. Mc Crimmon, Pacific J. Math. 15 Ž1965. 925.
w40x B. de Wit, A. van Proeyen, Phys. Lett. B 293 Ž1992. 94. hep-thr9207091.
w41x M. Gunaydin,
¨ M. Zagermann, work in progress.
w42x K. Pilch, P.K. Townsend, P. van Nieuwenhuizen, Phys. Lett. B 136 Ž1984. 38.
Holographic view of causality and locality via branes in

AdSrCFT correspondence q
Dongsu Bak a , Soo-Jong Rey b
a
Physics Department, UniÕersity of Seoul, Seoul 130-743, South Korea
b
Physics Department, Seoul National UniÕersity, Seoul 151-742 South Korea
Received 16 September 1999; accepted 6 December 1999
Abstract
We study dynamical aspects of the holographic correspondence between d s 5 anti-de Sitter

supergravity and d s 4 super Yang–Mills theory. We probe causality and locality of ambient
space-time from super Yang–Mills theory by studying transmission of low-energy brane waves
via an open string stretched between two D3-branes in the Coulomb branch. By analyzing two
relevant physical threshold scales, we find that causality and locality is encoded in the super
Yang–Mills theory in terms of an infinite tower of long supermultiplet operators. The massive
W-boson and dual magnetic monopole behave more properly as extended, bilocal objects. We also
study the causal time-delay of a low-energy excitation on a heavy quark or meson and find
excellent agreement between the anti-de Sitter supergravity and super Yang–Mills theory descrip-
tions. We observe that the strong ’t Hooft coupling dynamics and the holographic scale-size
relation thereof play a crucial role to the agreement of dynamical processes. q 2000 Elsevier
Science B.V. All rights reserved.
1. Introduction
It is now widely believed that an anti-de Sitter ŽAdS. supergravity is dual to the large
N limit of a conformal field theory ŽCFT. w1x. A particularly interesting case regarding
this so-called AdSrCFT correspondence is between the semiclassical limit of Type IIB
superstring compactified to a five-dimensional anti-de Sitter space-time and the large-N
limit of super Yang–Mills theories in four dimensions. As exemplified by this case, one
of the most important features is that the ‘holographic principle’ w2–4x underlies all
q
Work supported in part by BK-21 Initiative Program ŽPhysics. KOSEF Interdisciplinary Research Grant
98-07-02-07-01-5, KRF International Collaboration Grant, UOS Academic Research Program, and The Korea
Foundation for Advanced Studies Faculty Fellowship.
E-mail addresses: dsbak@mach.uos.ac.kr ŽD. Bak., sjrey@gravity.snu.ac.kr ŽS.J. Rey..
PII: S 0 5 5 0 - 3 2 1 3 Ž 9 9 . 0 0 7 5 5 - 5
152 D. Bak, S.-J. Rey r Nuclear Physics B 572 (2000) 151–187
AdSrCFT correspondences: semi-classical gravity in Ž d q 1. dimensions is holographi-

cally described by degrees of freedom of a large-N quantum field theory in d
dimensions. The AdSrCFT correspondence has been tested extensively and evidence for
it has been drawn mainly from the agreement between the two sides of various static
quantities, such as symmetries, particle spectrum, Euclidean correlation functions, and
black hole entropy w5–7x.
The holographic principle, however, refers to the correspondence not just for static
only but also for all dynamical properties 2 . This raises an immediate puzzle: how can
possibly local and causal dynamics on an anti-de Sitter space-time be recovered from a
lower-dimensional quantum field theory? Large-N super Yang–Mills theory is certainly
a well-defined quantum field theory w16,17x with locality and causality of four-dimen-
sional Minkowski space-time built in. So is the supergravity in anti-de Sitter space-time,
in the semi-classical limit at the least. Given the apparent locality and causality of
physics, it seems virtually impossible to describe what happens in the bulk of space-time
by dynamical variables located at the boundary. In this paper, taking the correspondence
between Type IIB superstring on d s 5 anti-de Sitter space-time and d s 4 super
Yang–Mills theory as a setup, we will try to provide an answer to this puzzle.
More specifically, we will pose and study the following question: in order to encode
causality and locality of semi-classical supergravity in d s 5 anti-de sitter space-time,
what precise prescription does one need in formulating the d s 4 super Yang–Mills
theory? Our main conclusion will be that, in terms of elementary fields of super
Yang–Mills fields Ž N s 4 vector multiplet, which consistitute the fundamental degrees
of freedom via holography. neither conventional, dimension-four operators only nor
operators descending from expansion of Dirac–Born–Infeld action is sufficient. Only
after an infinite tower of higher-dimensional operators encompassing long supermulti-
plets are taken into account, causality and locality can be decoded out of the super
Yang–Mills theory.
We will address the posed question in the Coulomb branch of the super Yang–Mills
theory, viz. consider SUŽ N . ™ SwUŽ N y 2. = ŽUŽ1. = UŽ1..x by separating two D3-
branes out of the Ž N y 2. remainder and place them at r 1,2 , respectively, in the Coulomb
branch moduli space. We will then place a macroscopic open string Žeither fundamental
or Dirichlet. stretched between the two D3-branes and study transmission of a pulse of
brane waves from the first D3-brane to the second through the string. The stretched open
string is identified with a massive W-boson with mass M w s Ž r 1 y r 2 .rl s2 ' D rrl s2 . As
the string is stretched over a distance D r, under suitable circumstance, one would expect
to detect causal time-delay out of the transmitted brane-waves.
This process, which is a sort of Thomson scattering w18x Žalias Compton scattering at
low energy. of brane waves off a massive W-boson, is closely related to much-studied,
F 4 brane-wave interaction w19–24x. Cutting the one-loop diagram of the latter across
two internal W-boson lines, one obtains a pair of tree-level, Thomson scattering
diagram. In so far as probing causality and locality, the only difference between the two
would be that the F 4 brane-wave interaction involves inter-brane interaction via
exchange of closed string states, while the Thomson scattering interaction describes
2
A Minkowski description of the AdSrCFT correspondence has been studied extensively in Refs. w8,9x.
Various dynamical issues on anti-de Sitter space-time have been addressed in Refs. w10–15x.
D. Bak, S.-J. Rey r Nuclear Physics B 572 (2000) 151–187 153
interaction via excitation of the massive W-boson. In effect, the stretched open string
serves as a sort of one-dimensional ‘waveguide’ of the radiation, which would otherwise
have propagated out all over ambient space-time.
A comparison of Thomson scattering with known results regarding the F 4 brane-wave
interaction brings up another important lesson drawn throughout this work: static and
dynamic information are closely related in flat space-time, but not in anti-de Sitter
space-time. In flat space-time Žcorresponding to N s 2., it is well known that the F 4
brane-wave interaction is exact at one-loop by the non-renormalization theorem of w20x.
Douglas and Taylor w25x have argued that, by taking a coincident Ž N y 2.-brane limit
from a generic point in the Coulomb branch, the non-renormalization theorem should
also hold in anti-de Sitter space-time. As an evidence for this assertion, they have noted
that the supergravity interaction for zero-momentum brane waves is the same for both
flat and anti-de Sitter space-time. Through careful analysis, Das w26x has confirmed the
result further 3. The interaction at zero-momentum, however, is always instantaneous and
retarded interaction will show up only if non-zero momentum configuration is consid-
ered, a well-known fact of non-relativistic expansion in covariant perturbation theory.
For brane waves with non-zero momentum, corrections to the leading order result in
Yang–Mills theory is controlled by the string scale, l s w25x, hinting that stringy
corrections somehow ought to be taken into account into the supergravity description.
Douglas and Taylor have interpreted these corrections as a manifestation of the
space-time uncertainty principle w27x. Das has sketched a calculation of the retarded
interaction between non-zero momentum brane waves in flat space-time and have
argued that the non-renormalization theorem assures the same result in anti-de Sitter
space-time. Underlying to their arguments seems that the massive W-boson, which is a
stretched open string of length D r, is a good ruler for measuring distance and hence
causality of the bulk space-time, both flat and anti-de Sitter.
From the Thomson scattering process, we will find contrasting results. Underlying to
the process, the W-boson, being an open string stretched between the two D3-branes,
sets two competing threshold scales: the winding and oscillation energy of the string.
We will find that causality and locality of the ambient space-time is visible in super
Yang–Mills theory only if the oscillation threshold is lighter than the winding threshold.
The winding threshold remains the same for both flat and anti-de Sitter space-time but,
quite importantly, the oscillation threshold turns out to change significantly 4 . In flat
space-time, crossover between the two threshold takes place when D r ; O Ž l s ., confirm-
ing from dynamical side the well-known result on D-brane world-volume dynamics w28x.
In anti-de Sitter space-time, however, crossover takes place when Žusing Maldacena’s
scaling variable U ' rrl s2 .
2
Ž DU . 1
U1U2
;O
ž /
g st N
. Ž 1.1 .
It shows that, everywhere in the Coulomb branch Žpossibly except at the origin., the
strong ’t Hooft coupling limit ensures that the oscillation threshold is lower than the
3
Intuitively, the result may be interpreted as a consequence of the fact that the operators involved are short
supermultiplets and that the massive W-boson running through the loop is a BPS state.
4
Recall that a winding string is a BPS state, but not when oscillator excitations are added.
winding threshold and hence enables the large-N super Yang–Mills theory to encode
causality and locality of semi-classical supergravity in the anti-de Sitter space-time.
With light oscillation threshold, the super Yang–Mills theory is given in a rather
different form from conventional Žretaining dimension-four operators only. or Dirac–
Born–Infeld theories. The massive W-bosons and dual magnetic monopoles are more
suitably described as a sort of extended, bilocal object 5. Remarkably, particle La-
grangian of the W-boson takes Žan extended form of. Pais–Uhlenbeck w31x model with
transcendental, infinite-order kernels, which is one of the few known examples satisfy-
ing convergence, positive-definiteness and strict causality.
This paper is organized as follows. In Section 2, we will study brane-wave transmis-
sion via Thomson scattering, first in flat space-time. In Section 3, we will study the
holographic relation between energy and size in AdSrCFT correspondence, but now
from dynamical point of view. We will be studying causal time delay of low-energy
excitations added to a macroscopic string representing quark or meson w33,34x. We will
find an excellent agreement between anti-de Sitter supergravity and super Yang–Mills
theory results. In Section 4, we will study brane-wave transmission via Thomson
scattering, but now in anti-de Sitter space-time. In Appendix A and Appendix B, we will
explain technical details of the covariant Green–Schwarz action for an open string
stretched between the two D3-branes, in particular, the consistent gauge-fixing of
k-symmetry. We will also derive a low-energy effective Lagrangian for a bilocal
W-boson, for both flat and anti-de Sitter space-times.
2. Causality and locality in flat space-time
Before dwelling upon anti-de Sitter space-time, we will first study causality and
locality in flat space-time. To make a direct comparison with anti-de Sitter space-time
later, as a probe, we will use a system consisting of two D3-branes and an open F- or
D-string stretched between them.
Consider, in Type IIB string theory, two parallel D3-branes in flat space-time
X s R9,1. The D3-branes are oriented along 0,1,2,3 directions and are located along
4, . . . ,9 directions at Ž r 1 ,0, . . . ,0. and Ž r 2 ,0, . . . ,0., respectively. The low-energy dynam-
ics on the D3-brane world-volume is governed by d s 4, N s 4 super Yang–Mills
theory with gauge group G s UŽ2.. We will denote generators of the UŽ2. gauge group
as Ta Ž a s 1,2,3,4., where T1,2,3 s s 1,2,3 belong to the SUŽ2. and T4 s id to the diagonal
UŽ1. subgroups. As the two D3-branes are separated by a distance D r s < r 1 y r 2 <, the
SUŽ2. gauge group is spontaneously broken to UŽ1. generated by T3 Cartan subalgebra.
Together with the diagonal UŽ1., the gauge group on the two D3-branes, which we label
by 1 and 2, is then given by H s U Ž1. Ž1. = U Ž2. Ž1., generated by diagonal linear
combinations, ŽT3 " T4 .r2. A massive W-boson Žor dual magnetic monopole. associated
with the spontaneous symmetry breaking is represented by an open F-string Žor D-string.
5
rather reminiscent of Yukawa’s old idea w29,30x.
stretched between the two D3-branes. The static mass of the W-boson and the dual
magnetic monopole is given by
T
M w s TD r , Mm s Dr . Ž 2.1 .
g st
They are part of an isospin triplet under G Žor its dual gauge group. and hence, under H,
carry electric Žor magnetic. charges Ž Q,y Q . where Q s "16 .
To investigate causality and locality of the flat Minkowski space-time over the
distance scale D r, one will need to excite slightly, say, the first D3-brane and follow the
subsequent propagation of the excitation, which will eventually arrive at and excite the
second D3-brane. In the limit g st ™ 0, the semi-classical dynamics of the Type IIB
supergravity ought to obey both locality and causality. For low-energy excitation, the
leading order process in this limit will be such that the excitation is transmitted from the
first D3-brane to the second through an open F- or D-string stretched between them7.
When exciting brane-wave on the first D3-brane, there are two channels available:
massless Higgs or U Ž1. Ž1. gauge field waves. In what follows, we will consider
exclusively adding a weak amplitude, plane-wave of the U Ž1. Ž1. gauge field on the first
D3-brane8. The entire dynamical process is then viewed as the classic Thomson
scattering of the U Ž1. Ž1. radiation field off the open string.
The open string stretched between the two D3-branes is casually identified with a
massive W-boson. This sounds paradoxical since the open string is an extended object
whose characteristic scale is the string scale, l s , while the massive W-boson is a
point-like object. An answer to this is well known w28x ever since the advent of
D-branes. In this section, from the dynamical point of view, we will revisit this issue as
it is intimately tied with understanding causality and locality of the ambient space-time
in which the D-branes are embedded. Specifically, we will study the Thomson scattering
process in two opposite limits of the massive W-boson, first in a point-particle limit and
second in a stretched open string limit. We will find shortly that, over the distance scale
D r, super Yang–Mills theory will perceive non-locality and acausality in the limit the
W-boson may be treated as a point particle. In order to recover locality and causality,
the W-boson ought to behave more like an extended, bilocal object. Crossover between
the two limits takes place, as anticipated, at D r f O Ž l s ., the minimum distance scale of
the ambient space-time probed by an F-string. Somewhat surprisingly, we will find that
the same conclusion holds for a massive magnetic monopole, viz. an open D-string
stretched between the two D3-branes.
2.1. Scattering of flat brane-waÕe by point-like W-boson
As is set up, a plane-wave of monochromatic radiation of U Ž1. Ž1. gauge field is

incident on a free, massive W-boson of charge Q and mass M W . The W-boson will be
6
Charge conjugation on the D3-brane world-volume is generated by world-sheet parity reversal of the
attached open strings.
7
Processes involving more than one open strings or closed strings are at least O Ž g st .. At weak coupling,
g st ™0, their contribution is negligible.
8
Interaction mediated by adjoint scalar fields can be included in an analogous manner from that of gauge
fields, as they are related by underlying N s 4 supersymmetry.
Fig. 1. Thomson scattering of brane wave off a W-boson. The incident U Ž1. Ž1. brane wave, whose wave front
is denoted by dashed lines, is scattered off a point-like massive W-boson at rest. The scattered spherical brane
waves of both U Ž1. Ž1. and U Ž2. Ž1. gauge groups, labelled as 1, 2, are emitted instantly and simultaneously.
accelerated and emit radiation and, if the energy of the incident radiation is low enough
compared to the mass M, the emitted radiation will have the same frequency as the
incident radiation. The whole process then can be described by conventional Thomson
scattering process, for which the W-boson is treated as a point particle. As the W-boson
is charged under H, the emitted radiation will consist of both U Ž1. Ž1. and U Ž2. Ž1. gauge
fields. In terms of the D3-branes, this means that both D3-branes will be excited by
dipole oscillation of the stretched open string Žinterpreted as the massive W-boson. after
the incident U Ž1. Ž1. radiation on the first D3-brane scatters off it, as depicted in Fig. 1.
2.1.1. Scattering equation of motion

We will be studying dynamics of a massive W-boson, located at XŽ t ., interacting
with gauge fields of the gauge group H. From the D3-brane point of view, the W-boson
is realized by an open F-string stretched between the two D3-branes. Denote the spatial
position Žon the two D3-branes. of its endpoints by X 1,2 Ž t .. In the limit the W-boson is
treated as a point particle, the position of the W-boson XŽ t . ought to coincide with the
two endpoint positions, X 1,2 Ž t . all the time. Consider an incident plane-wave radiation
of the U Ž1. Ž1. gauge field, whose wave vector and frequency are denoted by k and v ,
respectively. As the W-boson carries electric charge Ž Q,y Q . under H, the electric field
of the incident radiation,
yi v t
E Ž x,t . s E Ž1. Ž t . e i k P x , E Ž1. Ž t . s E Ž1.
0 e , Ž 2.2 .
will exert an instantaneous acceleration to the massive W-boson according to the
Lorentz force equation
Q
¨ 1 Ž t . s X¨ 2 Ž t . s
X E Ž1. Ž t . e i k P x . Ž 2.3 .
Mw
Throughout this paper, we will consider the weak field-strength limit. In this case, the
W-boson is in non-relativistic motion and, at leading order in the weak field expansion,
the spatial dependence of the electric field on the right-hand side of Eq. Ž2.3. may be
ignored.
Eq. Ž2.3. shows that the two ends, even though separated by a distance D r, undergo
instantaneous and, especially, simultaneous acceleration, viz. the stretched open F-string
behaves like a rigid rod. It indicates that, should the open F-string be treated as a point
particle, super Yang–Mills theory on the D3-brane world-volume would entail in-
evitably non-locality over the distance scale D r in the direction perpendicular to the
D3-brane in the embedding space. This is of course as it should be, in order for the
perpendicular direction to be interpreted as color isospin and SO Ž6. R-symmetry
direction.
2.1.2. Transmission rate

To appreciate the significance of Eq. Ž2.3., we will now calculate the rate of energy
transmission T from the first D3-brane to the second through the stretched open string.
From Eq. Ž2.3., one can determine the instantaneous power P Ž t . radiated into a
polarization state ´ by the W-boson as9
dP Q2
s ¨ Ž t. <2.
<e) PX Ž 2.4 .
dV2 4 c2
Averaging over a period 2prv , during which the charged particle moves a negligible
fraction of one wavelength,
2
dP E Ž1.2 Q2
¦ ;
dV2
sc
0
8 ž 4p M w c 2 / ˆ Ž1.
<e ) PE 0
<2. Ž 2.5 .
The first factor ŽE Ž1.2

0 r2 c is nothing but the incident energy flux, viz. the time-averaged
.
Poynting vector for the plane wave. Hence, the differential transmission rate for an
unpolarized incident radiation is given by
2
dT Q2
ž /
dV2 classical
s
ž 4p M w c 2 / P 18 Ž 1 q cos 2u . , Ž 2.6 .
in which u denotes the scattering angle in the laboratory frame. The total transmission
rate T obtained thereof is
2
2p Q2
Tclassical s
3 ž 4p M w c 2 / . Ž 2.7 .
A characteristic feature of classical Thomson scattering is that the cross section is

independent of the frequency of the scattered radiation. The result Eq. Ž2.6. is valid only
in the low frequency limit, v < Mc 2 , for which the gauge field can be treated as a
classical wave. As the energy of the incident radiation becomes larger and, especially,
comparable to the W-boson mass, the scattering process should be treated quantum
mechanically, viz. treating the radiation as photons. Treated in the Coulomb gauge for
9
In the weak field limit of the incident radiation, the W-boson may be treated as a heavy particle. In this
case, spin and charge degrees of freedom decouple from each other. We will henceforth treat the W-boson as a
spinless particle.
which the transition matrix element is identical to the classical amplitude, the modifica-
tion is from the phase space factor and hence is purely kinematical. The result is
2
dT Q2 vX 2
ž /
dV2 quantum
s
ž 4p M w c 2 /ž v / P 18 Ž 1 q cos 2u . , Ž 2.8 .
where the ratio of the outgoing to the incident frequency is given by the well-known
Compton formula
vX
y1
v u
v
s 1q2
ž Mw c 2
sin 2
2 / . Ž 2.9 .
For the low frequency limit, v < M w c 2 , one easily obtains the transmission rate as
2
2p Q2 v
Tquantum s
3 ž 4p M w c 2 /ž 1y2
Mw c 2
q... . / Ž 2.10 .
Had we considered a point-like limit of the magnetic monopole dual to the W-boson,
represented by an open D-string stretched between the two D3-branes, the transmission
rate would be essentially the same as above except that M w is to be replaced by the
mass of the magnetic monopole Mm and that the charge Q is interpreted as the dual
magnetic charge w35,36x.
To recapitulate, in the limit the W-boson is point-like, viz. the stretched F-string
moves rigidly, the transmission of radiation energy from the first D3-brane to the second
through W-boson is instantaneous and is a consequence of the standard field theory
result, Eq. Ž2.3.. Based on this fact, we conclude that the D3-brane world-volume
dynamics, if treated in terms of conventional super Yang–Mills theory and point-like
W-bosons, would perceive non-locality and acausality over a distance scale D r in the
six-dimensional transverse space in X. From the super Yang–Mills theory point of
view, this conclusion should be hardly surprising as the color isospin and R-symmetry
space is only an internal space and is not part of the four-dimensional space-time
Žworld-volume of the D3-branes., on which the non-trivial dynamics under consideration
takes place.
2.2. Scattering of a flat brane-waÕe by a charged open string
From the underlying string theory point of view, non-locality and acausality over the
distance scale D r are extremely bizarre. Transverse to the D3-branes is a six-dimen-
sional subspace in X and string theory ought to exhibit locality and causality, at least in
the long distance limit. For example, energy transfer from the first D3-brane located at
r s r 1 to the second at r s r 2 through the open string between them ought to take a
‘causal time delay’
1 Dr
Dt ' Ž r 2 y r1 . s , Ž 2.11 .
c c
as depicted in Fig. 2.
The resolution of the puzzle is quite simple. As non-locality and acausality has arisen
from the point-particle limit of the W-boson, one expects that locality and causality will
Fig. 2. A stretched F- or D-string connecting two probe D3-branes in a flat space-time background. Vibration
of the stretched string causes retarded signal propagation between the two D3-branes only when the separation
is larger than the string scale.
be restored transparently in the opposite limit, viz. the W-boson Žrealized as an open
string stretched between the two D3-branes. is treated as a full-fledged string. This turns
out to be correct and prompts one to study the Thomson scattering of the incident
U Ž1. Ž1. radiation off a string-like W-boson. Indeed, in this section, we will find that the
W-boson dynamics is drastically modified from that in the point-particle limit and that
the modified dynamics is precisely what allows the restoration of locality and causality
of X.
2.2.1. Scattering equations of motion

Consider again the Thomson scattering of the U Ž1. Ž1. field radiation off the W-boson,
but now in the limit the W-boson is treated as a full-fledged open string. Dynamics of
the string is governed by Type IIB Green–Schwarz action, supplemented by an
appropriate boundary action for the open string attached to the D3-branes. In the
Nambu–Goto formulation, which is sufficient for the Thomson scattering process, the
action is constructed in Appendix A. After fixing the k-symmetry in a gauge compatible
with the boundary conditions and world-sheet reparametrization symmetries in static
gauge, whose steps are explained in detail in Appendix A, one finds that the action for
the transverse string coordinates XŽ r,t . is reduced to the form
T 1 2 2
Istring s dt yM w q
H 2 ž c2 /
Ž E t X. y Ž Er X . q . . .
1
q Ý Hdt Q I A 0Ž I . Ž r,t . q ˙ I P AŽ I . Ž r,t .
X
Is1,2 c rsX I Ž t .
Ž Q1 s yQ2 s "1. . Ž 2.12 .

The first term in the expansion represents the static mass of the string, TD r, which
equals the W-boson mass, M w Žsee Eq. Ž2.1... Consider, as posed above, Thomson
scattering of a low-energy, monochromatic plane wave of the U Ž1. Ž1. gauge field off the
string endpoints. The boundary condition of the string coordinates XŽ t,r . is given by
TEr X Ž t ,r . rs r 1 s QE
Ž1.
Ž t. ,
TEr X Ž t ,r . rs r 2 s 0. Ž 2.13 .
According to the first boundary condition, the string endpoint X 1Ž t . Žon the first
D3-brane. will undergo a dipole oscillation and generate a pulse that will subsequently
propagate along the string. The pulse may be decomposed into spectral components
d v aŽ v . a˜ Ž v .
X Ž t ,r . s H 2p eyi v Ž tyrrc . q eyi v Ž tqrrc . . Ž 2.14 .
2v 2v
The spectral amplitudes aŽ v ., a˜ Ž v . are determined uniquely by the boundary condition
Eq. Ž2.13.
QE Ž1.
0
a Ž v . s a˜ ) Ž v . s Ž cot vD t y i . . Ž 2.15 .
T
From Eqs. Ž2.14., Ž2.15., one immediately obtains an effective equation of motion for
the string endpoint X 1Ž t . on the first D3-brane:
tanh D tE t
¨ 1 Ž t . s QE Ž1. Ž t . .
Mw
ž D tE t / X Ž 2.16 .
From Eq. Ž2.16., one first finds that the inertia mass of the open string equals the static
mass, M w , as is dictated by the underlying Lorentz invariance. Compared to the
point-particle limit of the W-boson, Eq. Ž2.3., the equation of motion of the endpoint
X 1Ž t . is changed to the one involving a kernel of infinite-order time derivatives. As will
be shown in Subsection 2.3, the extra kernel, which has originated from taking into
account the string oscillations, is what enables the low-energy dynamics of the D3-brane
world-volume to reproduce locality and causality in the ambient space-time X. For now,
note that, due to the kernel, one may interpret the endpoint X 1Ž t . to behave effectively
heaÕier than the point-like W-boson mass M w .
What about the dynamics of the second endpoint X 2 Ž t . attached to the second
D3-brane? For point-particle limit of the W-boson, we have seen that the two endpoints
behave identically, see Eq. Ž2.3.. What one now finds, however, is that the dynamics of
the second endpoint is governed by a different equation of motion:
sinh D tE t
¨ 2 Ž t . s QE Ž1. Ž t . .
Mw
ž D tE t / X Ž 2.17 .
Comparison with Eq. Ž2.16. shows that the infinite-order kernel associated with the
second endpoint is different from that of the first one. In particular, the kernel lets us
interpret the second endpoint X 2 Ž t . to behave lighter than the point-like W-boson mass
Mw .
Eqs. Ž2.16., Ž2.17. constitute the defining equations for the Thomson scattering off
the stretched open F-string. Compared to the Thomson scattering off the massive
W-boson, new effects have entered through D3-brane-separation-dependent Žsee Eq.
Ž2.11.., infinite-order kernels. As the two endpoints obey different equations of motion,
at any moment, one would expect that X 1Ž t . / X 2 Ž t .. Therefore, from the super
Yang–Mills theory perspective, it suggests to view the W-boson more naturally as a sort
of extended, bi-local object.
2.2.2. Transmission rate across open string

Following the same steps as in Subsection 2.1.2, it is straightforward to calculate the
transmission rate T Ždefined earlier in Subsection 2.1.2. from the new equation of
motion, Eq. Ž2.17., of the string endpoint on the second D3-brane.
Actually, one will need to take into account another important corrections to the
transmission rate: virtual W-boson pair and radiation reactive force effects. They are
entirely of field-theoretic origin and have nothing to do with string oscillations. Because
of these effects, in a field theory treatment, the W-boson might be viewed roughly as a
particle with an effective size set by the W-boson Compton wavelength. Heuristically,
the effects may be understood as follows w37x. The low-energy dynamics of a single
W-boson may be described by taking the non-relativistic limit of the W-boson gauge
field Wm :
1 i 2
W0 s eyi M w c t D P F Ž r,t . ,
(2 M w Mw c
1 2
Ws eyi M w c t F Ž r,t . , Ž 2.18 .
(2 M w
in which the charge-conjugate, anti-particle part is suppressed. Expanding the super

Yang–Mills Lagrangian in the non-relativistic limit, one finds
1 Q
L SYM s y 14 F mn Fmn q iF † P Dt F q F† P D 2 F y i B P F† = F q . . . .
2 Mw Mw c
Ž 2.19 .
This can be reduced further to a single particle Lagrangian of the W-boson,
1
˙ P X˙ q Q A 0 q
L w s 12 M w X ˙ P A q S P B q . . . y dr Ž 14 F mn Fmn . .
ž c
X / H Ž 2.20 .
Here, parts containing the classical spin dynamics and interaction with the Higgs field
are suppressed for brevity. Following an Abraham–Lorentz-type approach w18x, it is
straightforward to derive an effective equation of motion that takes the radiative reaction
force into account. One obtains
Q2 Et
¨ Ž t . s QE 0 Ž t . .
ž
Mw 1 y
4p M w
q... X
/ Ž 2.21 .
If one takes virtual W-boson pair corrections into account, one also obtains a term
similar to the second on the left-hand side. At low frequency, the second term is
suppressed compared to the first by O Ž vrMw . and hence, according to Eq. Ž2.4., gives
rise to O Ž v 2rM w2 . correction to the transmission rate.
Taking into account the above effects, one obtains the transmission rate for F-string
as
2 2
2p Q2 v v
TF - string s
3 ž 4p M w c 2 /ž 1y2
Mw c 2 /ž ž
q . . . P 1y
Mw c 2 / q...
/
2
vD t
P . Ž 2.22 .
sin vD t
The first two brackets represent the result derived from the limit the W-boson is treated
as a point-particle. The third bracket is the correction due to the virtual W-boson and
radiation reactive force effect estimated above. The last bracket comes from the string
oscillation effects, viz. from the fact that the massive W-boson is actually a stretched
open string.
It is instructive to consider the transmission rate for, instead of an F-string, a D-string
is stretched between the two D3-branes. From super Yang–Mills point of view, the open
D-string corresponds to a magnetic monopole dual to W-boson. In this case, scattering
of the brane wave takes place from interaction of the open D-string with magnetic
component of the radiation field. The corresponding transmission rate is obtained
straightforwardly:
2
2p Q2 v
TD - string s
3 ž 4p M m c 2 /ž
2
1y2
Mm c 2
q... /
2
v vD t
P 1y
ž ž / / Mw c 2
q... P
sin vD t
, Ž 2.23 .
where Q should now be interpreted as the magnetic charge of the dual W-boson. One
again finds that, apart from the kinematical, Compton scattering correction, corrections
of O Ž v 2rM w2 . arise from the classical finite-size of the magnetic monopole, set by the
Compton wavelength of the W-boson, and radiation reactive force effects w37x.
We emphasize that, for both the W-boson and its dual magnetic monopole, the
correction due to finite-size and radiation reactive force effects is set by the W-boson
mass, M w . Given that it is the only low-energy threshold scale in the system at weak
coupling limit, the fact that corrections for both W-boson and magnetic monopole is
governed by the same low-energy scale should not be surprising 10 .
2.3. Locality and causality in flat space-time
Having analyzed transmission of brane waves across open string at two different
limits, we will now study causality and locality of ambient space-time as seen by super
Yang–Mills theory.
2.3.1. Retarded dynamics of string endpoints
We have already seen from Eq. Ž2.3. that, in the limit the massive W-boson behaves
like a point particle, the string endpoint on the second D3-brane, despite being separated
10
Incidentally, the correction also entails an interesting deviation from S-duality of the ds 4, N s 4 super
Yang–Mills theory w37x.
by a distance D r, responds to the incident brane wave localized on the first D3-brane
instantaneously. As such, the scattered brane wave will come out of both isospin
components of T3 Žlabelling the two D3-branes. simultaneously, with no apparent causal
time-delay between the two isospin components. See Fig. 1. Indeed, the transmission
rate, Eq. Ž2.10., is exactly the same as the classic Thomson scattering cross section. As a
result, one will conclude that any super Yang–Mills theory with pointlike W-bosons will
not be able to encode causality and locality of ambient space-time.
In the opposite limit where the W-boson is a full-fledged oscillating F-string, Eq.
Ž2.14. shows that a pulse propagating along the string is described by characteristic
functions of the form f Ž t " rrc .. Clearly, it will take a causal time delay D t, Eq. Ž2.11.,
for the pulse, induced by the incident brane wave on the first D3-brane, to reach the
second D3-brane. To appreciate the causal time-delay more transparently, let us rewrite
the equation of motion of second string endpoint, Eq. Ž2.17., into an integral equation
form
D tE t
¨ 2 Ž t . s dtX Q Ž t y tX . ² t <
Mw X H < tX : P QE Ž1. Ž tX . , Ž 2.24 .
sinh D tE t
where the Heaviside step function follows from the convergence requirement, as is usual
in any non-relativistic dynamics. The integral kernel in Eq. Ž2.24. shows manifestly that
the causal time-delay between the two ends of the F-string is governed by D t s D rrc.
Note that the integral kernel exhibits infinitely many poles but no zeros.
The equation of motion, Eq. Ž2.17., can be put into an equivalent to Eq. Ž2.24., but
even more suggestive form by observing the fact that the infinite-order kernel is in fact a
difference operator:
t X
X 2 Ž t q Dt . y X 2 Ž t y Dt . s 2 Dt H dt QE Ž1.
Ž tX . , Ž 2.25 .
displaying manifestly that it takes causal time-delay D t for the brane wave on the first
D3-brane, E Ž1., to arrive at the second. The fact that the time-delay is exactly D t as
estimated from the geometric distance between the two D3-branes, Eq. Ž2.11. also
implies that the boundary interactions between F-string and D3-branes are local in the
r-direction.
While encoding causality and locality quite successfully, the massive W-boson in this
limit is far from the point-like object one casually take in Yang–Mills theory. Rather,
the two endpoints X 1,2 Ž t . of the open string evolves independently. Hence, projected on
the D3-brane world-volume, the W-boson is now replaced by a sort of extended, bilocal
object whose dynamics is captured by the equations of motion, Eqs. Ž2.16., Ž2.17..
2.3.2. Physical thresholds: ‘ winding’ Õersus ‘momentum’

As is evident from the transmission rates, Eqs. Ž2.22., Ž2.23., frequency-dependent
corrections are characterized by two distinct physical threshold scales inherent to the
problem11 . The first threshold is set by the W-boson mass, M w s TD r, defining a
physical threshold Žof field theoretic origin. above which the description in terms of
11
As the open string is stretched over a distance D r and only low-energy string oscillation is considered, we
find it more convenient to study the open string dynamics in non-relativisitc limit.
abelian gauge theory with gauge group H breaks down. The second is set by the energy
gap of the string oscillation, 1rD t s crD r, above which a point-particle description of
the W-boson breaks down. Being proportional to D r and inversely to D r, respectively,
we shall be referring the first threshold as ‘winding threshold’ L w and the second as
‘momentum threshold’, Lm :
1 "c
L w s M w s TD r , Lm s s . Ž 2.26 .
Dt Dr
As there are two distinct physical thresholds, depending on which one sets the lower
scale, the low-energy dynamics of the super Yang–Mills theory will become quite
different. As the winding and momentum thresholds depend on the spatial separation D r
inversely, as D r is varied, the two will compete with each other. The competition may
be summarized into a form of string uncertainty relation ŽT ' 1rl s2 .
"c
L w Lm f O
ž /
l s2
. Ž 2.27 .
Note that the right-hand side, a combination of fundamental constants only, depends
crucially on the string scale, l s . It implies that any effects related to causal retardation,
which arises from oscillation threshold corrections, will inevitably depend on it, as
ErLm f l s2 M w E . This explains the observation of Douglas and Taylor w25x on the F 4
interaction between brane waves with non-zero momentum. As M w A D r A D t, one
again finds that successive power corrections in the oscillation threshold are nothing but
the well-known expansion in powers of retardation time, as summarized in Eq. Ž2.24..
Hence, if L w < Lm , threshold effects to low-energy processes are dominated by the
‘winding threshold’. As string oscillation excitations are completely irrelevant, the
stretched open string behaves literally like a rigid-rod, to which the D3-branes serve as a
pair of guard rail. Behaving as a rigid body, the open string may be treated as a
point-particle, massive W-boson. From the condition
L w < Lm ™ D r < l s , Ž 2.28 .

one finds that the limit in which the D3-brane world-volume dynamics is best approxi-
mated by super Yang–Mills theory, either in conventional Žrenormalizable dimension-
four operators. or Dirac–Born–Infeld form, is when the D3-brane separation is over a
sub-stringy scale, reproducing the well-known result w28x.
Exactly the same condition applies if the stretched F-string is replaced by a D-string.
An open D-string stretched between two D3-branes is casually identified with a
magnetic monopole dual to the massive W-boson. Its classical size12 is set by the
Compton wavelength of the massive W-boson M w s TD r and defines the field theoretic,
winding threshold scale L w . The requirement that this scale is much smaller than the
oscillation threshold scale Lm , which is the same for both F- and D-string, again sets the
condition Eq. Ž2.28..
12
At weak coupling, classical size of the stretched D-string TD r is much larger than the Compton
wavelength of the monopole TD rr g st and hence a semi-classical treatment of the monopole is justifiable.
3. Dynamical aspect of the energy–size relation
An important feature of the AdSrCFT correspondence is the holographic relation

between the radial position, ‘energy scale’, in anti-de Sitter space-time and ‘size’ in
super Yang–Mills theory. As this relation will be playing a crucial role for foregoing
discussions, we will be studying first the dynamical aspect of the relation. More
specifically, for heavy quarks and mesons studied in w33,34x, we will estimate the causal
time-delay for a pulse generated by a quark either along the string in anti-de Sitter
space-time or in Minkowski space-time, using the harmonic fluctuation Lagrangian of
the string studied previously w32,33,38x. We will find that, after utilizing the holographic
energy–size relation, the two estimates agree perfectly with each other.
A charged probe necessarily accompanies the long-range Coulomb field, so it will be
necessary to control the field configuration in the infrared. A convenient way is to move
into the Coulomb branch, viz. separating some of the D3-branes and then study a state
charged under the gauge group of the separated D3-branes. The N coincident D3-branes
sitting at the conformal point then generates the anti-de Sitter space-time as a near-hori-
zon geometry w1x
U2 dU 2
ds 2 s l s2 Ž yc 2 dt 2 q dx 2I . q geff q g eff d V 52 Ž U ' Tr , T s 1rl s2 . .
g eff U2
Ž 3.1 .
In this coordinate, the energy-scale holographic relation asserts that the scale position U
in the anti-de Sitter space-time is related directly to the scale size R in the super
Yang–Mills theory:
g eff
Rs . Ž 3.2 .
U
This relation has first been derived from the evaluation of the Wilson loop w33,34x and
later extended to more general contexts w7x.
3.1. Holography of heaÕy quark dynamics
Consider a heavy quark of large-N super Yang–Mills theory at zero temperature. We

will separate a single D3-brane to the Coulomb branch and consider a quark coupled to
it. Denote the location of the displaced probe D3-brane as U) . Then, a heavy quark
Žmonopole. of the super Yang–Mills theory is realized by a macroscopic F- ŽD-. string
ending on the displaced D3-brane at U s U) and stretched outward along the U-direc-
tion. The dynamics of the macroscopic string may be studied using the Nambu–Goto
action in anti-de Sitter space-time, Eq. Ž3.1.. Taking a static gauge Žt , s . s Ž t,U . and
expanding string transverse coordinates ŽX I , V 5 . up to harmonic fluctuations, one finds
w33x
` `
Ž2.
L NG s y HU dU qHU dU L q L boundary Ž 3.3 .
) )
Fig. 3. Coulomb branch view of gauge field generated by a test quark Žmonopole..
where
1 1 U4 2
g eff 1
˙ 2I y XXI2 q U 2 V˙ 52 y V 5X 2
L Ž2. s
2 ž c 2
X 2
g eff / ž U 4 c2 / q... Ž 3.4 .
and L boundary specifies the boundary conditions at U s U) .

Let us begin with the static properties of the macroscopic string. The first term in
Eq. Ž3.3. represent the static mass of the string
L UV
Mstring Ž U) . s HU dU, LU V ™ `. Ž 3.5 .
)
The static mass diverges as the cutoff L UV is removed. As shown first in Refs. w33,34x
and discussed further in Refs. w7,11x, this is nothing but an anti-de Sitter space
manifestation of the divergent self-energy of a heavy quark. A physically meaningful,
finite quantity would be a residual static mass between strings ending on different
locations of U) , which we may define as
U) dMstring
D Mstring Ž U) . ' dU H s U) . Ž 3.6 .
0 dU
Using the scale–size relation Eq. Ž3.2., one then finds immediately that D Mstring ŽU) .
s g effrR ) corresponds to the Coulomb field energy measured at radius R ) . To
understand this scale better, let us interpret the macroscopic string from the super
Yang–Mills theory side. At the origin of the Coulomb branch Žviz. U) s 0., the
macroscopic string creates not only a heavy quark Žtransforming as the defining
representation of SUŽ N .. itself but also a long-range color Coulomb field around it, as
depicted in the left of Fig. 3. Away from the origin of the Coulomb branch, one might
expect that the long-range Coulomb field is cut off at the Compton wavelength U)y1 of
the massive gauge bosons13 . However, as pointed out in Refs. w33,34x, the strong ’t
Hooft coupling dynamics cuts off the long-range field at a larger scale, g eff U)y1 . Thus,
we find that D Mstring is nothing but the ‘field energy’, the energy associated with
Coulomb field outside radius R ) s g effrU) in the Yang–Mills theory. Summarizing,
the static property Žsuch as the static mass. of the macroscopic string is determined by
the quantity HdU.
What about dynamical properties? Consider exciting a low-energy pulse at the outer
end of the string, as depicted in Fig. 4. The pulse will then propagate at the speed of
13
Recall that the massive gauge bosons are BPS particles.
Fig. 4. String attached to a probe D3-brane in anti-de Sitter space.
light Žwhich has been explicitly checked in w33x. along the string toward U) . According
to the energy–size relation, in super Yang–Mills theory, the pulse traveling along the
string is mapped, in super Yang–Mills theory, to a spherical brane wave generated at the
position of the heavy quark and then propagating outward. For the spherical wave to
propagate over a radial size R ) , it will take Yang–Mills time:
1
Ž D t . YM s R ) . Ž 3.7 .
c
If causality is implied by the holographic energy–size relation, then the time-delay
for the pulse propagation along the string should be the same as that for the spherical
brane wave propagating outward. Let us check this explicitly. The string pulse propa-
gates along the string, viz. a null geodesic in ŽU,t . subspace,
U2 g eff
Dt 2 s DU 2 . Ž 3.8 .
g eff U2
Thus, measured from anti-de Sitter space-time in Yang–Mills time unit, the causal
time-delay for the pulse to travel from U s ` to U s U) along the string is given by
1 ` dU 1 g eff
Ž D t . AdS s g effH 2
s . Ž 3.9 .
c U) U c U)
Remarkably, the holographic energy–size relation, Eq. Ž3.2., permits the causal time
delay in the anti-de Sitter space-time to agree exactly with that along the boundary!
3.2. Holography of heaÕy quark dynamics at finite temperature
Consider next the super Yang–Mills theory at non-zero temperature. The near-hori-
zon geometry of near extremal D3-branes is described by the anti-de Sitter–
Schwarzschild space-time, whose Ž t,U . subspace is given by
y1
2
U04 U2 g eff U04
2
ds s l s yc 1 y2
ž U4 / g eff
2
dt q
U2 ž 1y
U4 / dU 2 . Ž 3.10 .
Fig. 5. Pulse propagation along semi-infinite string in anti-de Sitter Schwarzschild space-time.
From the Nambu–Goto Lagrangian, one again finds that the static mass of the string is
given by
`
Ms HU dU, Ž 3.11 .
0
the same as the value at zero temperature.

Let us introduce a heavy quark to the Yang–Mills theory at finite temperature. The
quark is described by a macroscopic F-string stretched along U-direction w38,39x, as
depicted in Fig. 5. In the present case, the string terminates at the Schwarzschild
horizon, U s U0 , where it will dissipate the F-string flux into the inner horizon region.
This phenomenon has a direct counterpart in super Yang–Mills theory. At zero
temperature, the Coulomb field of the static quark extends everywhere, as depicted at
the left in Fig. 4. At finite temperature, however, the Coulomb field becomes Debye-
screened and is expected to extend to the scale of the Debye wavelength U0y1 s
Ž g eff T .y1 . Using the holographic energy–size relation, one finds that the Coulomb field
is cut off at a larger scale g eff U0y1 , again due to the strong ’t Hooft coupling dynamics.
See Fig. 6.
In fact, by demanding causality, one can understand the fact that the string must end
at U s U0 . Consider a weak pulse injected at U s ` and propagating inward. The
time-delay for the pulse to reach the position U ŽU0 ( U ( `. is easily estimated:
g eff U0rU dx
Ž D t . AdS s H . Ž 3.12 .
U0 0 1yx4
The time-delay diverges logarithmically as U ™ U0 . So, evolved in terms of Yang–Mills
theory time, which is an asymptotic observer’s time in the anti-de Sitter space-time, the
Fig. 6. Aerial view of an isolated quark in super Yang–Mills theory at finite temperature.
Fig. 7. A pulse propagating along the Wilson loop in anti-de Sitter space-time.
super Yang–Mills theory covers the space-time region only outside the horizon. If a
different global time is chosen as an evolution operator, then the string might be able to
enter inside the horizon. In this case, however, the string would not be in a static
configuration w10x.
At finite temperature, the scale R 0 s g eff U0y1 corresponds to a physical threshold
scale in Yang–Mills theory. At the weak ’t Hooft coupling limit, the Matsubara mass
scale associated with dimensionally reduced fermions Žsatisfying an anti-periodic bound-
ary condition around Euclidean time. is O ŽT ., which equals U0rg eff . If this observation
somehow holds also for super Yang–Mills theory at zero temperature, then we are
prompted to expect that the scale R ) ' g eff U)y1 actually corresponds to a new physical
threshold scale. Recall that this scale is not a threshold of elementary excitations as the
Compton wavelength of massive gauge bosons U)y1 . Nevertheless, the fact that the scale
R ) is g eff times the Compton wavelength of massive gauge bosons is tantalizingly
similar to the observation in a different context w40x.
3.3. Holography of heaÕy meson dynamics
A more intricate bulk probe is the macroscopic string configuration corresponding to

the heavy meson w33,34x. In the static gauge, the string configuration is specified by a
first-integral of motion:
2
g eff
G2UX2 q G s . Ž 3.13 .
U)4
Again, consider a pulse propagating along the string, U s UŽ s . Žsee Fig. 7.. The time
delay for the pulse to travel between the quark and anti-quark Žboth ‘located’ at U s `.
along the string is again calculated straightforwardly by solving string equation of
motion along the null geodesic:
1
'G dU 2 s 2 2 2
'G Ž yc dt q d s . . Ž 3.14 .
This yields
1 qdr2
Ž D t . AdS s H d s(GU X2
q1 . Ž 3.15 .
c ydr2
Fig. 8. Aerial view of the quark anti-quark pair and dipole field configuration produced thereof.
Using Eq. Ž3.13., one finds immediately that

1 g eff ` 1 1 2'p G Ž 5r4 . g eff
Ž D t . AdS s 2 HU dU (U rU s . Ž 3.16 .
c U)2 )
4 4
) y1
c G Ž 3r4 . U)
From the super Yang–Mills theory side, natural scale of the meson is the inter-quark
distance d. Hence, one expects that
1
Ž D t . YM s A d, Ž 3.17 .
c
where A is a O Ž1. numerical coefficient, which should reflect the details of the dipole
field configuration produced by the quark–antiquark pair at the strong ’t Hooft coupling
limit. For mesons, the precise form of the holographic energy–size relation has been
derived w33,34x:
d 'p G Ž 3r4. g eff
s , Ž 3.18 .
2 G Ž 1r4 . U)
Hence, equating D t of anti-de Sitter space-time with that of Yang–Mills theory, one
finds that
G Ž 5r4 . G Ž 1r4 .
As s 2.188 . . . Ž 3.19 .
G 2 Ž 3r4 .
The numerical factor, A s 2.188 . . . , associated with the anticipated dipole electric
field configuration of the quark–antiquark pair as depicted in Fig. 8, may be viewed as a
prediction for the strong ’t Hooft coupling dynamics from anti-de Sitter supergravity.
4. Causality and locality in anti-de Sitter space-time
Having understood the dynamical aspect of the energy–size relation better, we will
now revisit the Thomson scattering of brane waves by an open string and the consequent
implications to the issue of causality and locality, but now all in anti-de Sitter
space-time.
We will be showing that the ‘momentum’ threshold of the open string stays always
below the ‘winding’ threshold everywhere in the anti-de Sitter space-time. Moreover,
quite importantly, both thresholds do not depend on the string scale, l s . Thus, in sharp
contrast to the flat space-time situation, we will be able to encode causality and locality
of semi-classical supergravity in anti-de Sitter space-time in terms of large-N super
Yang–Mills theory, at least in the strong ’t Hooft coupling limit. The resulting super
Yang–Mills theory is not the standard one, however. At a generic point in the Coulomb
branch moduli space14 , we will be finding that an infinite tower of long supermultiplet
operators should be taken into account in order to be able to encode causality and
locality of the five-dimensional bulk space-time. Once again, this conclusion is drawn
from the fact that W-bosons are more suitably described by a sort of extended, bilocal
objects.
4.1. D3-branes in the Coulomb branch
The anti-de Sitter space-time is generated as the near-horizon geometry of infinitely

many Ž N ™ `. D3-branes when they become coincident with one another. Therefore, to
explore causality and locality of the anti-de Sitter space-time, from the viewpoint of
super Yang–Mills theory, one will need to examine the low-energy dynamics near the
origin of the Coulomb branch moduli space, where the Higgs expectation values are
interpreted as positions of individual D3-branes in anti-de Sitter space-time w1,25x.
As in flat space-time, we will be considering two D3-branes out of N coincident
branes to be located at positions U1 ,U2 , respectively, and a macroscopic open string
stretched between them. We will again excite the displaced probe D3-branes with
low-energy brane waves and study Thomson scattering off the open string. The
configuration corresponds to the following sequential symmetry breaking:
SU Ž N . ™ G s S U Ž N y 2 . = U Ž 2 . ™ H s S U Ž N y 2 . = Ž U Ž 1 . = U Ž 1 . . .
Ž 4.1 .
From super Yang–Mills theory perspectives, this appears quite a complicated prob-
lem, as the excitation of, say, the first D3-brane will reach the second D3-brane not only
through the prescribed open string stretched between them but also indirectly through
the coincident Ž N y 2. D3-branes, with which the two D3-branes interact. At the least,
because the number of possible transmission channels through the ‘background’ Ž N y 2.
D3-branes is humongous, O Ž N 2 ., one might be tempted to conclude that, in the super
Yang–Mills theory, brane wave transmission is not mediated through the massive
W-bosons but by some sort of complicated large-N, strong ’t Hooft coupling dynamics.
This, however, should not be a problem. The point is that the interaction between the
probe Yang–Mills theory with the gauge group UŽ2. or UŽ1. = UŽ1. and that of the
conformal sector with gauge group SUŽ N y 2. are controlled by the mass scales, U1,2 .
Taking DU s < U1 y U2 < much lighter and g st sufficiently small, the dynamics of the two
probe D3-brane sector can be studied in a controllable manner. Thus, starting from the
14
Except possibly at the origin, which we will be discussing later.
super Yang–Mills theory with gauge group G, we will be integrating out over the
Coulomb branch, viz. open strings transforming in ŽN I 2,2. and its complex-conjugate
representations. In the g st ™ 0, N ™ ` and large ’t Hooft coupling limit, the integration
amounts to a resummation of large-N Feynman diagrams. As have argued in w1x Žsee
also discussions in w25x., the result is to produce anti-de Sitter space-time, inside which
the two displaced probe D3-branes are localized.
Intuitively, the above assertion may be understood by observing that the variance of
the positions of the Ž N y 2. coincident D3-branes and two displaced D3-branes are of
order Ž g st N .1r4 and Ž g st .1r4 , respectively. That is, in the large-N limit, the position of
the two probe D3-branes is sharply localized inside the anti-de Sitter space-time. It
should also become clear that, even after adding a stretched open string between the two
displaced D3-branes Žassociated with the symmetry breaking G ™ H ., integration over
the Coulomb branch yields essentially the same result, viz. the dynamics of both the two
probe D3-branes and the stretched open string between them will perceive the ambient
background as anti-de Sitter space-time.
Henceforth, at g st ™ 0, N ™ ` and in the strong ’t Hooft coupling regime, we will be
able to study causality and locality of the anti-de Sitter space-time in a controllable
manner by focusing only on low-energy excitations of massive W-bosons in ‘quantum
corrected’ super Yang–Mills theory with symmetry breaking pattern UŽ2. ™ UŽ1. =
UŽ1..
4.2. Scattering of brane waÕe by charged open string
The dynamics of the open F-string stretched between the two D3-branes is governed
by a Type IIB Green–Schwarz action in anti-de Sitter space-time. In Appendix B, for
the Nambu–Goto form of the action, details of gauge fixing of local k- and
reparametrization symmetries are explained. It turns out the gauge-fixing as well as the
analysis of string dynamics thereof become enormously simplified once one adopts a
new coordinate, R ' g effrU. According to the energy–size holography relation, Eq.
Ž3.2., the new coordinate R is seen to be nothing other than the ‘size’ variable of super
Yang–Mills theory. It is interesting that, in analyzing a ‘bulk’ process in anti-de Sitter
space-time, the description in terms of a ‘boundary’ variable turns out more suited and
efficient. Indeed, in terms of the ‘size’ coordinate R, the metric of the anti-de Sitter
space-time is given in Poincare´ coordinates as
2
g eff
dsAdS s 2 yc 2 dt 2 q dx 2I qdR 2 q R 2 d V 52 .
R
As elaborated on in Appendix B, this coordinate choice leads the Green–Schwarz action
to an almost identical form, after gauge fixing, to the one in flat space-time, for both
bosonic and fermionic parts. Moreover, as has been observed in w33x, the harmonic
dynamics of the open string turns out to be considerably simplified in the Poincaré
coordinate system. After the gauge-fixing, the open string action is given by
g eff R 2 dR 1 2 2
Istring s dt yM w c 2 q
H H ž
Ž E X. y Ž E R X. q . . . /
2 R1 R 2 c 2 t
1
q Ý Hdt Q I A 0Ž I . Ž r,t . q ˙ I Ž t . P AŽ I . Ž r,t .
X . Ž 4.2 .
Fig. 9. A stretched F- or D-string connecting two probe D3-branes in the Coulomb branch. Vibration of the
stretched string causes retarded radiation between the two D3-branes.
First, one notes that the static mass of the string as measured in anti-de Sitter
space-time, the first term in Eq. Ž4.2., g eff HdRrR 2 equals ŽU2 y U1 . in terms of the
original ‘energy’ coordinate. ŽSee Fig. 9.. A similar result holds also for an open
D-string stretched between the two D3-branes. Hence,
1
M w s DU, Mm s DU. Ž 4.3 .
g st
Recalling that DU s TD r Žcf. Eq. Ž3.1.., one discovers that the static mass of the string
in anti-de Sitter space-time is the same as that in flat space-time. This is, from the super
Yang–Mills theory perspective, as it should be. Both the W-boson and the dual
magnetic monopole are BPS-saturated states and hence their masses are not renormal-
ized as one interpolates from the weak to the strong ’t Hooft coupling regime.
From the Thomson scattering process in flat space-time we have observed that the
static mass of the W-boson and the causal time-delay are governed by one and the same
length scale, D r, the separation between the two D3-branes. As the W-boson is a BPS
state, one might have guessed that it will be the same in anti-de Sitter space-time. After
all, the distance between the two D3-branes is the only relevant length scale inherent in
the problem. This expectation, however, turns out not quite right. We will be showing
below that the causal time-delay is governed not by the scale DU, but by
1 1 1 1
Dt ' DR s g eff < y <. Ž 4.4 .
c c U1 U2
It is clear that Dt has no simple functional relation to DU, the scale that has set the
W-boson mass for both flat and anti-de Sitter space-time and the energy variance that
the space-time uncertainty relation w27x utilizes. In fact, the reason why DŽ g effrU ., and
not DU, should be the relevant variable for the time-delay may be gleaned from the fact
that SO Ž4,2. invariant line element between two points Ž t,U,x.1,2 is given by
2
U1U2 2
1 1 2
2
D sAdS s
g eff
2
yc Ž t 1 y t 2 . 2
q g eff
ž U1
y
U2 / q Žx1 y x 2 . . Ž 4.5 .
4.2.1. Scattering equation of motion

Consider the excitation of a low-energy, monochromatic plane wave of the U Ž1. Ž1.
gauge field on the first D3-brane. As in flat space-time, we will be studying Thomson
scattering of the brane-wave off the open string. We have argued above that, in so far as
2
g st s g YM ™ 0, low-energy processes between the two probe D3-branes can be trusted.
Actually, the Thomson scattering process depends on the combination of g YM E 0rM w .
Thus, even in the situation where g YM itself may not be sufficiently small, the foregoing
calculations will be completely controllable by taking the brane waves sufficiently weak.
In static gauge, the boundary condition of the transverse string coordinates XŽ t, R . is
given by
g eff
E XŽ t , R . s QE Ž1. Ž t . ,
R2 R RsR 1
g eff
E XŽ t , R . s 0. Ž 4.6 .
R2 R RsR 2
From Eq. Ž4.2., one finds that a weak pulse propagating along the string is given by
XŽ t , R .
d v aŽ v . R a˜ Ž v . R
s H 2p ž 1yiv / eyi v Ž tyRrc . q ž 1qiv / eyi v Ž tqR r c. .
2v c 2v c
Ž 4.7 .
Again, the boundary condition, Eq. Ž4.6., fixes the spectral amplitudes aŽ v ., a˜ Ž v .
uniquely:
R 1rg eff eyi v R 2 r c
a Ž v . s a˜ Ž v . s
)
v ž isin vDt /. QE Ž1.
0 Ž 4.8 .
Substituting Eq. Ž4.8. into Eq. Ž4.7., one obtains the equation of motion for the string
endpoint X 1Ž t . on the first D3-brane:
y1
R1 DtE t
¨ 1 Ž t . s QE Ž1. Ž t . .
Mw 1 q
ž R2 ž tanh DtE t
y1
// X Ž 4.9 .
Comparison with the corresponding equation Ž2.16. in the flat space-time indicates that
the infinite-order kernel has been changed completely both in functional form and in the
dependence on the D3-brane positions, U1,2 . Moreover, the kernel does not respect
translational invariance in the U-coordinate as it depends on the individual positions U1,2
separately. If probed the anti-de Sitter space-time using a static ruler such as the
W-boson, one would have concluded incorrectly the translational invariance from the
W-boson mass, Eq. Ž4.3.. Dynamic information such as causal time-delay Ž4.4. does not
exhibit translational invariance, either in ‘energy’ or ‘scale’ coordinates. Despite these
differences, the fact that the kernel makes the endpoint effectively heavier holds the
same in anti-de Sitter space-time as well.
Proceeding similarly, one also finds an equation of motion for the string endpoint
X 2 Ž t . on the second D3-brane:
sinh DtE t
¨ 2 Ž t . s QE Ž1. Ž t . .
Mw
ž DtE t
X
/ Ž 4.10 .
Interestingly, in this case, the functional form of the infinite-order kernel is exactly the
same as in the flat space-time, Eq. Ž2.17.. However, the argument of the kernel, which
will set the causal time-delay, is completely changed: in flat space-time, it was
D r s l s2DU proportional to the W-boson mass, while in anti-de Sitter space-time, it is
D R s DŽ g effrU . and has no apparent functional relation with the W-boson mass. As R
is the natural ‘size’ variable in super Yang–Mills theory Žsee Eq. Ž3.2.., one again finds
a hint from Eq. Ž4.10. that causality and locality of the anti-de Sitter space-time ought to
be encoded in super Yang–Mills theory as a difference in the size of the brane waves.
4.2.2. Transmission rate across open string

It is straightforward to estimate the transmission rate of the brane-wave in anti-de
Sitter space-time. As argued above, the transmission has taken place effectively via the
stretched open string in the anti-de Sitter space-time background. Compared to the flat
space-time, the equation of motion Eq. Ž4.10. of the second endpoint depends on
Dt s D Rrc Ždefined in Eq. Ž2.11.., separation between the two D3-branes measured in
terms of the ‘size’ variable in the super Yang–Mills theory. Apart from this, as the
equation of motion has exactly the same form as in flat space, Eq. Ž2.17., from Eq. Ž2.4.,
one finds straightforwardly the transmission rate across F-string as
2 2
2p Q2 v v
TF - string s
3 ž 4p M w c 2 /ž 1y2
Mw c 2 /ž ž
q . . . P 1y
Mw c 2 / q...
/
2
vDt
P . Ž 4.11 .
sin vDt
Again, the first two terms represent the classical and quantum results, and the third
summarizes the radiation reactive force and virtual W-boson pair effects. As the
W-boson is a BPS state, its mass is protected in the strong ’t Hooft coupling limit. This
explains the fact that no change occurs in the first three terms from the result in flat
space-time. The last term, the string oscillation correction, is not protected by supersym-
metry, even though its functional form has remained the same, at least in the weak brane
wave limit.
Similarly, the transmission rate TD - string for a D-string in anti-de Sitter space-time is
obtained as
2 2
2p Q2 v v
TD - string s
3 ž 4p M m c 2 /ž 1y2
Mm c 2 /ž ž
q . . . P 1y
Mw c 2 / q...
/
2
vDt
P . Ž 4.12 .
sin vDt
One again finds that the first three terms are protected against strong ’t Hooft coupling
corrections, as both both the magnetic monopole and W-boson masses are BPS
saturated.
It should be emphasized again that the derivation of the above results is sufficiently
controllable, despite the fact that the large-N super Yang–Mills theory is in the strong ’t
Hooft coupling regime. The expansion parameter g YM E 0rM w is controllably small as
long as the brane wave is kept weak enough.
4.3. Locality and causality in anti-de Sitter space-time
4.3.1. Retarded dynamics of string endpoints

As in flat space-time, one can observe the causal time-delay, Eq. Ž2.11., in various
ways. First, a pulse propagating along the string in anti-de Sitter space-time is described
by characteristic functions of the form g Ž t " Rrc ., see Eq. Ž4.7.. As such, it will take a
causal time delay Dt , as claimed in Eq. Ž2.11., for the pulse to propagate across the two
D3-branes. Alternatively, the equation of the second string endpoint, Eq. Ž4.10., written
in an integral equation form
DtE t
¨ 2 Ž t . s dtXQ Ž t y tX . ² t <
Mw X H < tX : P QE Ž1. Ž tX . Ž 4.13 .
sinh DtE t
or in a difference equation form,
t X
X 2 Ž t q Dt . y X 2 Ž t y Dt . s 2 Dt H dt QE Ž1.
Ž tX . Ž 4.14 .
displays manifestly the causal time-delay Dt for brane-wave transmission. Agreement of

the time-delay with the geometric distance, Eq. Ž2.11., implies that the boundary
interaction between the F-string and D3-branes are local, even in the strong ’t Hooft
coupling limit.
Compared to that in flat space-time, an important distinction of causal time-delay in
anti-de Sitter space-time is the fact that Dt does not depend on the sting scale at all. In
flat space-time, as D t s l s2 M w , the oscillation threshold and hence the causal time-delay
has disappeared completely in the supergravity limit, l s ™ 0. This has now changed
fundamentally in anti-de Sitter space-time. The fact that the oscillation threshold is
finite, independent of the string scale, everywhere on the Coulomb branch moduli space
suggests that they will play an important role in defining the precise nature of the super
Yang–Mills theory, to which we now turn.
4.3.2. Physical thresholds in anti-de Sitter space-time

From the discussion of the last subsection, especially the transmission rate formulae
Ž4.11., Ž4.12., one finds that two physical threshold scales in the super Yang–Mills
theory are the ‘winding’ threshold scale set by the W-boson mass and the ‘momentum’
threshold scale set by the Yang–Mills ‘scale’ size:
1 U1U2 1
L w s M w s DU; Lm s s . Ž 4.15 .
Dt g eff DU
As discussed already, the ‘winding’ threshold scale L w keeps the same value as in the
flat space-time, as it is protected by supersymmetry. The ‘momentum’ threshold scale
Lm , however, has changed significantly. In particular, it has survived miraculously the
scaling limit, l s ™ 0, in obtaining the supergravity regime in anti-de Sitter space-time.
In this sense, the stretched open string is a sort of non-critical string, whose tension is
governed by a low-energy scale in the super Yang–Mills theory.
A new uncertainty relation for anti-de Sitter space-time, expressed in terms of

physical scales in the Yang–Mills theory, is then given by
U1U2
L w Lm f O
ž /
g eff
. Ž 4.16 .
Note that, in contrast to the flat space-time situation, the uncertainty relation does
depend on the location of the two D3-branes, U1,2 , viz. the location in the moduli space
of the Coulomb branch. Thus, in any effects related to causal retardation, the size of the
threshold correction will depend on the location, as ErLm f Ž g effrU1U2 . M w E . For a
fixed M w , the correction is parametrically pronounced as the probe D3-branes zoom into
the origin.
Stated differently, one finds that, as the ‘winding’ threshold scale L w is held fixed at
a generic point in the Coulomb branch moduli space Žaway from the origin., the
‘momentum’ threshold L m becomes smaller than L w in the strong ’t Hooft coupling
limit, g eff ™ `. Causality and locality of the semi-classical supergravity in anti-de Sitter
space-time will then be encoded to the super Yang–Mills theory provided
U1U2
L w 4 Lm ™ DU 4 O
ž( /g eff
. Ž 4.17 .
As noted above, on top of the suppression in the strong ’t Hooft coupling limit, the
condition is better satisfied near the origin of the Coulomb branch moduli space.
What about at the origin of the Coulomb branch? As the ‘momentum’ threshold
depends both on g eff and on U1,2 , the answer clearly depends on the order in which one
takes the limits g eff ™ ` and U1,2 ™ 0. If one takes g eff ™ 0 first and then U1,2 ™ 0, it is
clear that the ‘momentum’ threshold stays light all the time. If one takes them
oppositely, for example, DU ™ 0 first and U1,2 ™ 0 afterward, the ‘winding’ threshold
always dominates and the non-critical string states will disappear at the conformal point.
4.3.3. Holographic encoding in super-Yang–Mills theory

We have now seen that, in large-N super Yang–Mills theory, the strong ’t Hooft
coupling limit is what enables us to encode causality and locality of the anti-de Sitter
space-time. Let us now ask how the Thomson scattering process would look like on the
super Yang–Mills theory side. As the ‘momentum’ threshold stays always lighter than
the ‘winding’ one, the W-boson is described more properly as an extended, bilocal
object. Super Yang–Mills theory is not the one defined either in conventional Ždimen-
sion-four operators. or in Dirac–Born–Infeld form, as they certainly do not contain such
an object.
As the two endpoints can move independently, the two isospin components of the
scattered brane waves can carry the causal time-delay information. Let us see how this
looks like in the super Yang–Mills theory. The open string is now holographically
projected to a shell of spherically symmetric Yang–Mills field configuration, as depicted
in Fig. 10. The inner and outer radii are R 1 s g effrU1 and R 2 s g effrU2 , respectively.
Once the W-boson is hit by the incident brane wave on the first D3-brane, concentric
spherical waves of the two isospin components will emanate. An observer sitting at
spatial infinity on D3-branes will then perceive the radiation as originating from the
Fig. 10. Yang–Mills theory view of the Thompson scattering.
same point in space Žthe center in Fig. 10., but with a time delay Dt . As is dictated by
Eq. Ž2.11., the difference in ‘size’ coordinate of the wave-fronts in the super Yang–Mills
theory exactly equals Dt .
Having concluded that super Yang–Mills theory defined either in renormalizable or
in DBI Lagrangians is not capable of encoding causality and locality of anti-de Sitter
space-time, we now ask what modifications, if possible, can do so. The answer to this
question can be drawn again from Eq. Ž4.13.. Expanding the infinite-order kernel for
low-energy brane-wave, one finds
q` 1
n
Mw X 2 Ž t . s Ý Ž y. H dtX ² t < 2
< tX : QE Ž1. Ž tX . .
nsy` t0t
X
E t2 q Ž nprDt .
q` ` 2 mq2
mq n Dt
sQ Ý Ý Ž y.
nsy` ms0
ž / np
O2 m Ž t . , Ž 4.18 .
where
O2 m s E t2 m E Ž t . ; =m 1 . . . =m 2 m Fa b . Ž 4.19 .
Thus, the interaction of massive W-boson with the brane waves is governed by an
infinite tower of higher-dimensional operators, O2 m ’s. Brane waves can be gauge or
Higgs fields, so there will be similar higher-derivative operators including Higgs scalars.
In d s 4, N s 4 super Yang–Mills theory, operators of these structures are classified as
long supermultiplets.
Perhaps it should not be surprising that long supermultiplets are responsible for
encoding causality and locality. After all, we have argued that string oscillator excita-
tions are what is needed for probing causality and locality. It has been already known
w5,6x that the long supermultiplets arise precisely from the massive string excitations, but
the novelty of the above argument is that they can be seen explicitly even at the level of
a single W-boson.
From Eq. Ž4.18., one also deduces that the Žcovariantized. interaction between
massive W-bosons and abelian gauge fields is schematically given by
`
Linteraction ; d 4 x
H Ý c2 n J m Ž x . E 2 nAm Ž x . . Ž 4.20 .
ns0
Generically, the coefficients c 2 n’s are functions of moduli parameters of the Coulomb
branch and, in particular, can be expanded in positiÕe powers of the moduli parameters.
The low-energy effective Lagrangian of abelian gauge fields is then derived by
integrating out massive W-bosons. Taking into account the interactions Ž4.20., one finds
straightforwardly that the low-energy effective Lagrangian contains not only dimension-
four operators but also infinite towers of higher-derivative operators, viz. long supermul-
tiplets. Note, however, that this is not simply an expansion in inÕerse powers of the
moduli parameters. In various terms, positive powers coming from the coefficients C2 n’s
are cancelled by negative powers from propagators of virtual massive W-bosons.
Appendix A. Effective field theory of string endpoints
In this appendix, using the covariant Green–Schwarz formalism, we will explain the
derivation of the gauge-fixed, open string world-sheet action, Eqs. Ž2.12., Ž4.2.. The
derivation turns out somewhat non-trivial, as two Majorana–Weyl spinors of the
covariant Green–Schwarz action are constrained both by k-symmetry gauge-fixing and
by open string boundary conditions. We will show that both requirements lead to the
so-called ‘D-brane gauge’ as a unique choice of the spinor projection. The ‘D-brane
gauge’ has been proposed previously as a natural gauge-fixing condition of the k
symmetry. Interestingly, we find that exactly the same condition also arises from the
analysis of open string boundary conditions.
We will also derive, after integrating out string excitations, an effective field theory
describing the low-energy dynamics of two endpoints of the string. As pointed out in
Sections 2 and 4, the dynamics is described by a Lagrangian with infinite-order kernels,
which turns out exactly the same as Ža two-particle generalization of. the Pais–Uhlen-
beck’s model. Remarkably, according to the analysis of Pais and Uhlenbeck, the
particular infinite-order kernel is the one compatible with convergence, positive-definite-
ness and causality.
A.1. Green–Schwarz action in flat space-time
Dynamics of an open F- or D-string stretched between the two parallel D3-branes, as

depicted in Fig. 2, is governed by the covariant Green–Schwarz action w41x. Denote
bosonic and fermionic string coordinates as Ž X M ;QaA . Ž M s 0, . . . ,9; A s 1,2; a s
1, . . . ,32., which map the world-sheet S Žspanned by s i Ž i s 0,1.. to the coset
superspace PoincareŽ9,1 <2.rSO Ž9,1.. In Type IIB string theory, Q 1,2 are Majorana–Weyl
spinors of same chirality. In the Nambu–Goto formulation, the world-sheet action is
given by
Istring s IGS q I D3 - Q ,
IGS s yT HSdt d s w(y det L ˆ L ˆ q iHM L ˆ n L t

a a
i j
a I I J aˆ J
3 G nL . Ž A.1 .
3
Here, I D3yQ is a boundary action of the string endpoints, M denotes a three-dimen-

sional manifold whose boundary equals the string world-sheet, and
Laˆ s d Maˆ dX M y iQ AG aˆdQ A Ž aˆ s 0,1, . . . ,9 . ,
I I
L s dQ . Ž A.2 .
are invariant Cartan one-forms of the coset superspace. The Wess–Zumino term in Eq.
ŽA.1. is given by
L WZ s 2 idX aˆ n Ž Q 1G aˆdQ 1 y Q 2G aˆdQ 2 . y 2 Ž Q 1G aˆdQ 1 . n Ž Q 2G aˆdQ 2 . .

Ž A.3 .
Apart from the boundary Lagrangian, the covariant Green–Schwarz action ŽA.1. is
invariant under N s 2 global supersymmetry Ž A s 1,2.:
de Q A s e A ; de X aˆ s i e AG aˆQ A , Ž A.4 .
and local k symmetry, when expressed in terms of world-sheet scalar spinors k˜ 15 ,
A
dk˜ Q A s Ž 1 q Ž y . P . k˜ A , dk X aˆ s iQ AG aˆdk˜ Q A , Ž A.5 .
where
ˆ
1 Laˆ n Lb
Ps G aˆ bˆ , P 2 s 1. Ž A.6 .
2 'y h
A.1.1. Open string boundary conditions

The last term in Eq. ŽA.1. is the Lagrangian describing interaction of the open string
endpoints with gauge and Higgs fields on the world-volume of the two D3-branes.
Denoting the d s 4, N s 4 vector supermultiplet as Ž A, l., the boundary Lagrangian in
the lowest order in derivative expansion reads
L D3yQ s Q EESdt X˙ Ž A ˆ Ž X . y lŽ X . Gû . .

â
a a Ž A.7 .
The boundary action represents turning on gauge and Higgs fields on the D3-brane
world-volume. The boundary conditions, derived from variation of the Green–Schwarz
action, are given by 16
D: d X H s 0, Q 1G H
dQ 1 q Q 2G H
dQ 2 s 0 Ž A.8 .
N: PsI s 0, Q 1G I dQ 1 y Q 2G I dQ 2 s 0, Ž A.9 .
for Dirichlet and Neumann directions, respectively. To proceed further with gauge
fixing, one will need to find linear boundary conditions for fermions that are consistent
15
The world-sheet vector k and world-sheet scalar k˜ are related to each other by k i1,2 sy i G â P i âk˜ 2,1 .
16
Once, on the D3-brane world-volume, Higgs and gauge fields are turned on, the boundary conditions are
expected to be modified. It is easy to show, however, that the Higgs field background does not modify the
Dirichlet boundary condition at all, while the gauge field background affects the Neumann boundary condition.
As we will be considering background fields that are weak enough, these modifications will be ignored
throughout.
with Eqs. ŽA.8., ŽA.9.. Because of a sign difference between the A s 1,2 terms in Eqs.
ŽA.8., ŽA.9., a unique choice of the linear boundary condition is of the form
Q A s G5 MBAQ B , G5 ' G 0G 1 . . . G 3 , Ž A.10 .
where Ž2 = 2. matrix M is determined to be MA B s e A B by the requirement that the
boundary condition is compatible with Majorana and Weyl conditions and with invert-
ibility of Eq. ŽA.10.. Hence, for an open Green–Schwarz superstring attached between
the two D3-branes, the boundary conditions are
d X Hs 0, Es X Is 0,
Ž dA B y G5 e A B . Q B s 0, Ž dA B q G5 e A B . Es Q B s 0. Ž A.11 .
A.1.2. Gauge fixing

Due to the local k-symmetry, half of the degrees of freedom of Q 1,2 ’s are redundant.
Since the spinors Q 1,2 have the same chirality, the k-symmetry may be gauge fixed
conveniently by setting Q 1 s G5Q 2 ' 12 C . Using the equation of motion for Lâ ’s, a
straightforward calculation yields the k-gauge-fixed action
i
IGS s T dt d s y det E i X aÊ j X aˆ q dX aˆ n Ž CG aˆdC . .
HH ( Ž A.12 .
2
In the gauge-fixed action, half of N s 2 supersymmetry is realized linearly
ˆ
1 dX aˆ n dX b
de X aˆ s i eG aˆC , deC s y G aˆ bê , Ž A.13 .
2 < dX n dX <
and the other half as a non-linear fermionic symmetry
dj X â s 0, djC s j . Ž A.14 .
For the reparametrization invariance, we will choose the static gauge s s t, s 1 s r. 0
After expanding the action to quadratic order and reinstating the speed of light c, one
obtains the gauge-fixed action of the open string as
IGS s Istring q I D3 - Q Ž A.15 .
where
T r2 1 2 2
Istring s dt yM w q
H H
2 r 1
dr ž c2
Ž E t X . y Ž Er X . q . . . , / Ž A.16 .
1
I D3yQ s Ý Hdt Q I A 0Ž I . Ž r,t . q ˙ I Ž t . P AŽ I . Ž r,t .
X ,
Ž Q1 s yQ2 s "1. . Ž A.17 .
A.1.3. Low-energy effectiÕe Lagrangian of string endpoints

We will begin with solving the equations of motion for XŽ t,r ., subject to a fixed but
arbitrary location of the string endpoints, X 1,2 Ž t .. Consider a Fourier transformed,
harmonic solution
˜ Ž v ,r . s dteqi v t X Ž t ,r . s a Ž v . e i v r r c q a˜ Ž v . eyi v r r c .
X H
The boundary condition specified,
˜ Ž v ,r . < rs r 1 s X˜ 1 Ž v . ,
X ˜ Ž v ,r . < rsr 2 s X˜ 2 Ž v . ,
X Ž A.18 .
relates the spectral amplitudes aŽ v .,a˜ Ž v . uniquely to the ŽFourier transformed. string
endpoints X˜ 1,2 Ž v .. Using the relation, it is possible to express normal derivatives at the
string endpoints
˜ 1 Ž v . ' Er X˜ Ž v ,r . < rs r 1 ,
Er X ˜ 2 Ž v . ' Er X˜ Ž v ,r . < rsr 2
X Ž A.19 .
˜ 1,2 Ž v .. The result is Ž D t ' D rrc .
in terms of X
vD t ˜ 1 Ž v . cos vD t y X˜ 2 Ž v .
X
˜ 1Ž v . s y
Er X
sin vD t ž c Dt / ,
vD t ˜ 1 Ž v . y X˜ 2 Ž v . cos vD t
X
˜ 2Ž v . sy
Er X
sin vD t ž c Dt / . Ž A.20 .
In order to obtain an effective action for the string endpoints, we will need to
integrate out massive string excitations. As the string action, Istring , is quadratic in XŽ r,t .
at leading order, it amounts to imposing the equation of motion of XŽ r,t . back to the
action, Eq. ŽA.17.. One then obtains, after a Fourier transform back to X 1,2 Ž t .,
T
Istring s dt yM w y
H Ž X 2 Ž t . P Er X 2 Ž t . y X 1Ž t . P Er X 1Ž t . . q . . . . Ž A.21 .
2
The string action is now expressed entirely in terms of boundary data of the open string
positions on the two D3-branes.
Inserting Eq. ŽA.20. into Eq. ŽA.21., one finally obtains the low-energy effective
action of the string endpoints
IGS s Istring q I D3 - Q , Ž A.22 .
T
Istring s dt yM w y
H Ý X I Ž t . KIflat
J Ž D tE t . X J Ž t . ,
2 I, Js1,2
1
X
Ž Q1 s yQ2 s "1. .
In the action, Istring , the infinite-order kernels are defined by
1 D tE t
K11flat s K22flat s q
ž
D r tanh D tE t / ,
1 D tE t
K12flat s K21flat s y
ž
D r sinh D tE t / . Ž A.23 .
Let us now focus on the Thomson scattering by D3-brane waves, studied in Section
2. Initially, the radiation field is present only for the first D3-brane, viz. AŽ2. Ž2.
0 s A s 0.
In this case, the equation of motion for X 2 Ž t . is reduced simply to
Ž cosh D tE t . X 2 Ž t . s X 1 Ž t . , Ž A.24 .
viz. a retardation relation
1
X 2 Ž t . s dtX² t <
H < tX :X 1 Ž tX . . Ž A.25 .
cosh D tE t
After X 2 Ž t . is integrated out of Istring using Eq. ŽA.24., the low-energy effective
Lagrangian for the Thomson scattering process is obtained:
Mw tanh D tE t
˙ 1Ž t . ˙ 1Ž t . q . . .
L Thomson s yM w q
2
X
ž D tE t / X
1
q Q A 0Ž 1 . Ž r,t . q ˙ 1 P AŽ 1 . Ž r,t .
X . Ž A.26 .
c rsX 1Ž t .
Eqs. Ž2.16., Ž2.17. follow immediately from the above effective Lagrangian and the
retardation relation Eq. ŽA.24..
Appendix B. Green–Schwarz action in anti-de Sitter space-time
The Green–Schwarz action in AdS 5 = S 5 can be obtained in a closed form via

super-coset space construction w42x17 . In Nambu–Goto form, the action reads
1
IAdS s T HHdt d s (y det L ˆ L ˆ q 4 iH0 dsL ˆ Ž s . Qt
a a
i j
a
i 3G
aˆ
Lj Ž s. . Ž B.1 .
Here, aˆ s Ž a,aX . s Ž0,1, . . . ,4,5, . . . ,9., I, J s 1,2 and t 1,2,3 are Pauli matrices. The
invariant world-sheet one-forms LI Ž s . and Lâ Ž s . are given by
I
sinh s M
LI Ž s . s ž DQ / ,
M
I
sinh2 12 s M
aˆ
L Ž x. s e Maˆ M
dX y 4 iQ G I aˆ
ž M2
DQ
/ , Ž B.2 .
where Ž X M ,Q I . are the bosonic and fermionic F-string coordinates and

IJ X I I X X X J X X
Ž M 2 . s g a Ž it 2 Q . Q Jg a y g a Ž it 2 Q . Q Jg a q 12 g a bQ I Ž Q it 2 . g a b
J
yg a bQ I Ž Q it 2 . g a b ,
Ž DQ . s ž E q 14 v aˆ bˆGaˆ bˆ / Q I q 12 E aˆGaˆ Ž t 2 Q . .
I I
Ž B.3 .
Also, LI ' LI Ž s s 1., Lâ ' Lâ Ž s s 1..
17
For a D-string, the action can be constructed similarly w43x.
The action is invariant under the N s 2 global supersymmetry and local k-symme-
try. Expressed again in terms of world-sheet scalar spinors, the symmetry transforma-
tions rules are essentially the same as Eqs. ŽA.5., ŽA.6., except that now the indices are
repeated over AdS 5 and S 5.
B.1. Gauge fixing
In order to fix the local k-symmetry, it turns out most convenient to make a change
of variable R s g effrU and use the Poincare´ coordinates. Up to a conformal rescaling,
the ten-dimensional space-time is flat. To proceed in an analogous manner to the flat
space-time situation, we will follow the observation of Kallosh w44–46x and take the
gauge-fixing 18
I IJ
QyI s 0 where Q "
I
' Ž P "Q . , IJ
P" s 12 Ž t 0 " i G5t 2 . . Ž B.4 .
The k-gauge-fixed action is quartic in fermions. However, if one makes a change of
variable to the ‘size’ variable, R s g effrU, as mentioned in Section 4, the quartic
fermion terms are turned into quadratic ones only. Moreover, in the R-coordinates, as
we will see below, the reparametrization gauge-fixed action also simplifies consider-
ably 19 .
Again, choosing the static gauge s 0 s t, s 1 s R, one finds that the gauge-fixed
Green–Schwarz action of the open string in anti-de Sitter space-time is given by
IAdS s Ž Lstring q L D3 - Q . Ž B.5 .
where
g eff R2 dR 1 2 2
Istring s dt yM w q
H 2
HR
1
R 2 ž c2 /
Ž E t X. y Ž E R X. q . . . , Ž B.6 .
1
L D3 - Q s Ý Hdt Q I A 0Ž I . Ž r,t . q ˙ I Ž t . P AŽ I . Ž r,t .
X
= Ž Q1 s yQ2 s "1 . Ž B.7 .

and the boundary interaction action is exactly the same as in the flat space-time.
B.2. Low-energy effectiÕe Lagrangian of string endpoints
One can obtain the low-energy effective Lagrangian of the string endpoints repeating
steps of flat space-time case, Section A.1.2. After Fourier transformation, a harmonic
˜ Ž v , R . now takes the form
solution of the string coordinate X
˜ Ž v , R . s dteqi v t X Ž t , R .
X H
R R
ž
s aŽ v . 1 y i v / e i v R r c q a˜ Ž v . 1 q i v ž / eyi v R r c ,
c c
18
Different gauge-fixing choice has been advocated recently w47x.
19
In w46x, a closely related observation has been made but was interpreted as the effect of T-duality along all
directions of D3-brane world-volume. According to the present argument, this interpretation seems unneces-
sary.
Ž " .Ž
where the harmonic modes are expanded in terms of Hankel functions, H3r2 v R ..
˜ 1,2 Ž v . and Ez X˜ 1,2 Ž v ., one finds the relations Ž Dt '
Eliminating aŽ v .,a˜ Ž v . between X
D zrc .
1 v2 R2 R1
˜ 1Ž v . s ˜ 1Ž v . y v ˜ 2Ž v. ,
R1
ER X
N ž sin vDt q v
c
cos vDt X / c
X
1 v2 R2 R1
˜ 2Ž v. s ˜ 1 Ž v . q sin vDt y v ˜ 2Ž v. ,
R2
ER X
N
v
c
X ž c
cos vDt X / Ž B.8 .
where
R1 R 2
Ns 1qv2 ž / sin vDt y vDt cos vDt . Ž B.9 .
c2
Imposing the equation of motion for XŽ t, z . to the action ŽB.6., one obtains
g eff 1 1
Istring s dt yM w y
H 2 ž R 22
X 2 Ž t . P ER X 2 Ž t . y
R 12 /
X 1Ž t . P E R X 1Ž t . q . . . .
Ž B.10 .
Inserting Eq. ŽB.8. to the action ŽB.10., after Fourier-transforming back, one finally
arrives at the low-energy effective action of the string endpoints in anti-de Sitter
space-time:
IAdS s Ž Istring q I D3 - Q . Ž B.11 .
g eff
Istring s dt yM w y
H Ý X I Ž t . P KIAdS
J Ž DtE t . P X J Ž t . q . . . ,
2 I, Js1,2
1
X
= Ž Q1 s yQ2 s "1 . .
The infinite-order kernels in the action Istring are now given by
y1
1 R 2rc DtE t R1 R 2 DtE t
AdS
K11 sy
R1 ž 1q
Dt tanh DtE t /ž 1y
c2
E t2 y
tanh DtE t /
y1
1 R 1rc DtE t R1 R 2 DtE t
AdS
K22 sq
R2 ž 1y
Dt tanh DtE t /ž 1y
c2
E t2 y
tanh DtE t / ,
y1
1 DtE t R1 R 2 DtE t
AdS
K12 AdS
s K21 sq
ž
D R sinh DtE t /ž 1y
c2
E t2 y
tanh DtE t / . Ž B.12 .
Let us again consider Thomson scattering of a D3-brane wave, studied in Section 4,

by setting AŽ2. Ž2.
0 s A s 0. From Eq. B.11 , the equation of motion for X 2 t yields a
Ž . Ž .
retardation relation
R1 DtE t sinh DtE t
ž 1q
R2 ž tanh DtE t
y1
//ž DtE t / X 2 Ž t . s X 1Ž t . . Ž B.13 .
Integrating out X 2 Ž t . from the action Istring , using Eq. ŽB.13., one finally obtains the
low-energy effective Lagrangian for the Thomson scattering of D3-brane waves:
y1
Mw R1 DtE t
˙ 1Ž t . 1 q ˙ 1Ž t . q . . .
L Thomson s yM w q
2
X
R2 ž tanh DtE t
y1
/ X
1
q Q A0Ž 1 . Ž r,t . q ˙ 1 Ž t . P AŽ1. Ž r,t .
X . Ž B.14 .
c rsX 1Ž t .
From the effective Lagrangian ŽB.14. and the retardation relation ŽB.13., one easily gets
the equations of motion of the string endpoints, Eqs. Ž4.44., Ž4.45..
References
w1x J. Maldacena, Adv. Theor. Math. Phys. 2 Ž1998. 231.

w2x G. ’t Hooft, Dimensional Reduction in Quantum Gravity, in: Salamfest, ed. A. Ali, J. Ellis ŽWorld
Scientific, Singapore, 1993. pp. 284–296.
w3x L. Susskind, J. Math. Phys. 36 Ž1995. 6377.
w4x C.B. Thorn, Reformulating String Theory With The 1r N Expansion, in: Sakharov Conference, ed. L.V.
Keldysh, V.Ya. Fainberg ŽNova Science, 1992. pp. 447–454.
w5x S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Phys. Lett. B 428 Ž1998. 105.
w6x E. Witten, Adv. Theor. Math. Phys. 2 Ž1998. 253.
w7x L. Susskind, E. Witten, The holographic bound in anti-de Sitter space, hep-thr9805114.
w8x V. Balasubramanian, P. Kraus, A. Lawrence, Phys. Rev. D 59 Ž1999. 46003.
w9x V. Balasubramanian, P. Kraus, A. Lawrence, S.P. Trivedi, Phys. Rev. D 59 Ž1999. 104021.
w10x T. Banks, M.R. Douglas, G.T. Horowitz, E. Martinec, AdS dynamics from conformal field theory,
hep-thr9808016.
w11x J. Polchinski, A. Peet, Phys. Rev. D 59 Ž1999. 065011. hep-thr9809022.
w12x G.T. Horowitz, N. Itzhaki, Black holes, shock waves, and causality in the AdSrCFT correspondence,
hep-thr9901012.
w13x J. Polchinski, S-matrices from AdS space-time, hep-thr9901076.
w14x L. Susskind, Holography in the flat space limit, hep-thr9901079.
w15x D. Kabat, G. Lifschytz, J. High Energy Phys. 9905 Ž1999. 005. hep-thr9902073.
w16x G. ’t Hooft, Comm. Math. Phys. 86 Ž1982. 449.
w17x G. ’t Hooft, Comm. Math. Phys. 88 Ž1983. 1.
w18x See, for example, J.D. Jackson Classical Electrodynamics, 3rd edition, ŽAddison-Wesley, Reading, MA,
1998..
w19x S.J. Gates, M.T. Grisaru, M. Rocek, W. Siegel, Superspace ŽBenjaminrCummings, Menlo Park, CA,
1987..
w20x M. Dine, N. Seiberg, Phys. Lett. B 409 Ž1997. 239.
w21x J. Maldacena, Phys. Rev. D 57 Ž1998. 3736.
w22x V. Periwal, R. von Unge, Phys. Lett. B 430 Ž1998. 71. hep-thr9801121.
w23x M. Berkooz, Nucl. Phys. B 553 Ž1999. 205. hep-thr9807230.
w24x D. Lowe, R. von Unge, J. High Energy Phys. 9811 Ž1998. 14.
w25x M.R. Douglas, W. Taylor IV, Branes in the bulk of anti-de Sitter space, hep-thr9807225.
w26x S.R. Das, J. High-Energy Phys. 9902 Ž1999. 012. hep-thr9901004.
w27x T. Yoneya, Mod. Phys. Lett. A 4 Ž1989. 1587.
w29x H. Yukawa, Phys. Rev. 77 Ž1950. 219.
w30x H. Yukawa, Phys. Rev. 80 Ž1950. 1047.
w31x A. Pais, G.E. Uhlenbeck, Phys. Rev. 79 Ž1950. 145.
w32x S.-J. Rey, J.-T. Yee, Nucl. Phys. B 526 Ž1998. 229.
w33x S.-J. Rey, J.-T. Yee, Macroscopic strings as heavy quarks in large-N gauge theories and anti-de Sitter
supergravity, hep-thr9803001.
w34x J. Maldacena, Phys. Rev. Lett. 80 Ž1998. 4859.
w35x D. Bak, C. Lee, Nucl. Phys. B 403 Ž1993. 315.
w36x D. Bak, C. Lee, Nucl. Phys. B 424 Ž1994. 124.
w37x D. Bak, H. Min, Phys. Rev. D 56 Ž1997. 6665.
w38x S.-J. Rey, S. Theisen, J.-T. Yee, Nucl. Phys. B 527 Ž1998. 171.
w39x A. Brandhuber, N. Itzhaki, J. Sonnenschein, S. Yankielowicz, Phys. Lett. B 434 Ž1998. 36.
w40x S.H. Shenker, Another length scale in string theory?, hep-thr9509132.
w41x M.B. Green, J.H. Schwarz, Phys. Lett. B 136 Ž1984. 367.
w42x R.R. Metsaev, A.A. Tseytlin, Nucl. Phys. B 533 Ž1998. 109.
w43x J. Park, S.-J. Rey, J. High Energy Phys. 9811 Ž1998. 008.
w44x R. Kallosh, Superconformal actions in Killing gauge, hep-thr9807206.
w45x R. Kallosh, J. Rahmfeld, Phys. Lett. B 443 Ž1998. 143.
w46x R. Kallosh, A.A. Tseytlin, J. High Energy Phys. 9810 Ž1998. 16.
w47x A. Rajaraman, M. Rozali, On the quantization of GS string on AdS5 = S5 , hep-thr9902046.
More CFTs and RG flows from deforming M2rM5-brane

horizon q
Changhyun Ahn a , Soo-Jong Rey b
a
Department of Physics, Kyungpook National UniÕersity, Taegu 702-701, South Korea
b
School of Physics, Seoul National UniÕersity, Seoul 151-742, South Korea
Abstract
Near-horizon geometry of coincident M2-branes at a conical singularity is related to M-theory

on AdS 4 times an appropriate seven-dimensional manifold X 7 . For X 7 s N 0,1,0 , squashing
deformation is known to lead to spontaneous Žsuper.symmetry breaking from N s Ž3,0. to
N s Ž0,1. in gauged AdS 4 supergravity. Via AdSrCFT correspondence, it is interpreted as
renormalization group flow of strongly coupled three-dimensional field theory with SUŽ3. = SUŽ2.
global symmetry. The flow interpolates between N s Ž0,1. fixed point in the UV to N s Ž3,0.
fixed point in the IR. Evidences for the interpretation are found both from critical points of the
supergravity scalar potential and from conformal dimension of relevant chiral primary operators at
each fixed point. We also analyze cases with X 7 s SO Ž5.rSO Ž3. max ,V5,2 Ž R ., M 1,1,1,Q 1,1,1 and find
that there is no non-trivial renormalization group flows. We extend the analysis to Englert type
vacua of M-theory. By analyzing de Wit–Nicolai potential, we find that deformation of S 7 gives
rise to renormalization group flow from N s 8, SO Ž8. invariant UV fixed point to N s 1, G 2
invariant IR fixed point. For AdS 7 supergravity relevant for near-horizon geometry of coincident
M5-branes, we also point out a non-trivial renormalization group flow from N s 1 superconfor-
mal UV fixed point to non-supersymmetric IR fixed point. q 2000 Elsevier Science B.V. All
rights reserved.
PACS: 11.25.-w; 11.25.Sq; 11.15.Pg; 04.65.qe
1. Introduction
Few examples are known for three-dimensional interacting conformal field theories
or for renormalization group flows among themselves, mainly due to strong coupling
q
Work supported in part by the BK-21 Initiative Physics Program, KRF International Collaboration Grant
and the KOSEF Interdisciplinary Research Grant 98-07-02-07-01-5.
E-mail addresses: ahn@kyungpook.ac.kr ŽC. Ahn., sjrey@gravity.snu.ac.kr ŽS.-J. Rey..
PII: S 0 5 5 0 - 3 2 1 3 Ž 0 0 . 0 0 0 0 8 - 0
C. Ahn, S.-J. Rey r Nuclear Physics B 572 (2000) 188–207 189
dynamics in the infrared limit. For example, N s 1 superconformal field theories,

which arise in various situations involving D-brane dynamics, do not have any continu-
ous R-symmetries, holomorphy constraints or any known non-renormalization theorems.
In the previous paper w1x, as an alternative route, we have proposed to classify
three-dimensional Žsuper.conformal field theories by utilizing the AdSrCFT correspon-
dence w2–4x and earlier, exhaustive study of the Kaluza–Klein supergravity w5x.
The simplest spontaneous compactification of the eleven-dimensional supergravity
w6,7x is the Freund–Rubin w8x compactification2 to a product of AdS 4 space-time and an
arbitrary compact seven-dimensional Einstein manifold X 7 of positive scalar curvature.
Continuous deformations among X 7 ’s is of interest 3, as they would be interpreted, in the
strongly interacting d s 3 quantum field theory dual to the AdS 4 supergravity, to
renormalization group flows among interacting Žsuper.conformal fixed points. The best
known example is provided by round and squashed S 7. The standard Einstein metric of
the round S 7 yields a vacuum with SO Ž8. gauge symmetry and N s 8 supersymmetry.
The S 7 also admits the second, squashed Einstein metric w9,10x, yielding a vacuum with
SO Ž5. = SO Ž3. gauge symmetry and N s 1 or N s 0 supersymmetry, depending on the
orientation of the S 7 w11–13x.
In Ref. w1x we have shown that the well-known spontaneous Žsuper.symmetry
breaking deformation from round- to squashed-S 7 is mapped to a renormalization group
flow from N s 0 or 1, SO Ž5. = SO Ž3. invariant fixed point in the UV to N s 8, SO Ž8.
invariant fixed point in the IR. In particular, in w1x, we have shown that Ž1. the squashing
deformation corresponds to an irrelevant operator at the N s 8 superconformal fixed
point and a relevant operator at the N s 1 or 0 Žsuper.conformal fixed point, respec-
tively, and Ž2. the renormalization group flow is described geometrically, as in AdS 5
supergravity w14–16x, by a ‘static’ domain wall of the sort studied earlier in w17,18x,
which interpolates the two asymptotically AdS 4 space-times with X 7 round and
squashed S 7 ’s.
In this paper, we will be studying further known examples of Kaluza–Klein super-
gravity vacua and reinterpret them in terms of three-dimensional Žsuper.conformal field
theories and associated renormalization group flows. First, we will be exploring various
Freund–Rubin type spontaneous compactifications on AdS 4 = X 7 . For M2-branes on an
eight-dimensional manifold, the near-horizon geometry X 7 is expected to change as the
branes are placed at or away a conical singularity of the manifold w19,20x. More
specifically, we will consider X 7 being Ž1. 3-Sasaki, Ž2. Sasaki–Einstein or Ž3. weak
G 2 holonomy manifolds, describing near-horizon geometry of M2-branes at relevant
conical singularities. Each of them preserves Ž3,0.,Ž2,0. and Ž1,0. supersymmetries,
respectively, where Ž NL , NR . denotes chirality of supercharges on AdS 4 .
Explicit examples of the above X 7 ’s are
Ž1. NI0,1,0 space with N s Ž3,0.;

Ž2. M 1,1,1,Q 1,1,1,V5,2 ŽR. spaces with N s Ž2,0.;
2
corresponding to the near-horizon geometry of coincident, infinitely planar M2-branes
3
We restrict our foregoing discussions only to deformations among X 7 ’s of same topology. While
examples of deformation between X 7 ’s of distinct topology would be extremely interesting, we are not aware
of any explicit example.
190 C. Ahn, S.-J. Rey r Nuclear Physics B 572 (2000) 188–207
Ž3-L. Squashed S 7,SO Ž5.rSO Ž3. max and other NIp, q, r space with N s Ž1,0.;
Ž3-R. all NIIp, q, r space with N s Ž0,1..
In Section 2, by analyzing relevant scalar potentials developed by Yasuda w21,22x for

each case, we will be finding further examples of supergravity dual to a three-dimen-
sional quantum field theory exhibiting a non-trivial renormalization group flow.
0,1,0
The first example is provided by X 7 s NI,II . The manifold NIp, q, r has been studied
originally by Castellani and Romans w23x, identified as a coset manifold of the form
w SUŽ3. = UŽ1.xrwUŽ1. = UŽ1.x. Embedding of wUŽ1. = UŽ1.x in w SUŽ3. = UŽ1.x is speci-
fied by a choice of the integers Ž p,q,r .. Each choice of p,q,r leads to an Einstein
manifold, yielding N s Ž3,0. supersymmetry and SUŽ3. = SUŽ2. gauge symmetry for
N 0,1,0 , or N s Ž1,0. supersymmetry and SUŽ3. = UŽ1. gauge symmetry otherwise.
Later, Page and Pope w24x have completed the coset manifold construction by showing
existence of another series of Einstein manifold, NIIp, q, r , except for those corresponding
to q s 0. The NIIp, q, r ’s, which can be obtained from geometric squashing of the NIp, q, r ’s,
retain the same gauge group as the latter but instead preserve N s Ž0,1. supersymmetry.
As in the case of X 7 s S 7, the existence of two Einstein metrics on N p, q, r ’s offers an
interpretation in terms of spontaneous Žsuper.symmetry breaking: scalar field corre-
sponding to the squashing deformation acquires a non-zero vacuum expectation value,
leading to Žsuper.-Higgs mechanism. The symmetry breaking pattern depends on the
choice of the X 7 orientation. With one orientation, for N 0,1,0 ’s, the squashing interpo-
lates between a N s 3 supersymmetric vacuum and another with N s 0 supersymme-
try. With opposite orientation, it interpolates between a non-supersymmetric vacuum and
a supersymmetry restored one with N s 1 supersymmetry. In both cases, however, the
gauge group is locally SUŽ3. = SUŽ2. and remains unbroken.
Via AdSrCFT correspondence, this implies that the three-dimensional quantum field
theory dual to the AdS 4 = N 0,1,0 supergravity exhibits two types of non-trivial renormal-
ization group flow: one between N s 3 superconformal UV fixed point and non-super-
symmetric IR fixed point, and another between non-supersymmetric UV fixed point and
N s 1 superconformal IR fixed point. For both, along the renormalization group flow
trajectory, the global symmetry SUŽ3. = SUŽ2. is always maintained. Analogously, for
all other N p, q, r ’s except q s 0, the dual quantum field theory exhibits two renormaliza-
tion group flows between N s 1 superconformal fixed point and non-supersymmetric
one. We present these analysis in Subsection 2.1.
In Subsection 2.2, we will be studying other examples of X 7 ’s. We will find that
p, q, r
NI,II ’s give rise to a non-trivial, SUŽ3. = UŽ1. invariant renormalization group flow
between N s 1 superconformal fixed point and non-supersymmetric conformal fixed
point. On the other hand, for X 7 s M 1,1,1 ,Q 1,1,1 ,V5,2 ŽR.,SO Ž5.rSO Ž3. max , it turns out the
squashing deformation does not lead to any non-trivial renormalization group flow 4 .
We next consider M-theory compactification vacua of Englert type. By generalizing
compactification vacuum ansatz to the non-linear level, solutions of the eleven-dimen-
sional supergravity were obtained directly from the scalar and pseudo-scalar expectation
values at various critical points of the N s 8 supergravity potential w26x. This way, it
4
Other aspects of three-dimensional superconformal field theories from Sasaki seven-manifold have been
studied in w25x.
was possible to reproduce all known Kaluza–Klein solutions of the eleven-dimensional

supergravity: round S 7 w27x, SO Ž7.y invariant, parallelized S 7 w28x, SO Ž7.q-invariant
vacuum w29x, SUŽ4.y-invariant vacuum w30x, and a new one with G 2 invariance. Among
them, round S 7- and G 2-invariant vacua are stable, while SO Ž7. "-invariant ones are
known to be unstable w31x. In all these vacua, generically, either the metric is endowed
with a non-trivial warp factor or the four-form magnetic flux Ga b c d is non-vanishing,
corresponding to turning on vacuum expectation values for scalar or pseudoscalar field,
respectively. The novelty of vacua with a non-trivial warp factor is that they corresponds
to inhomogeneous deformations of S 7. In Section 3, we will analyze the above vacua by
investigating de Wit–Nicolai potential. We will be identifying a deformation which
gives rise to a renormalization group flow associated with the symmetry breaking
SO Ž8. ™ G 2 Žboth of which are stable vacua. and find that the deformation operator is
relevant at the SO Ž8. fixed point but becomes irrelevant at the G 2 fixed point.
In Section 4, we study near-horizon geometry of coincident M5-branes, which are
described by AdS 7 gauged supergravity. The geometrical interpretation of seven-dimen-
sional solutions from the eleven-dimensional viewpoint has been obtained recently
w32–34x. By taking a non-linear Kaluza–Klein ansatz to the eleven-dimensional super-
gravity, they were able to obtain a consistent S 4 reduction to seven-dimensional, N s 1
gauged supergravity. Field content of the latter includes a scalar field parametrizing
inhomogeneous deformations of the S 4 , Yang–Mills gauge fields, and a topologically
massive three-form potential. Turning on the scalar field induces again spontaneous
Žsuper.symmetry breaking, and we interpret it as being dual to a renormalization group
flow from N s 1 supersymmetric UV fixed point to a non-supersymmetric IR fixed
point of a putative d s 6, N s Ž1,0. quantum theory. The existence of stable, non-super-
symmetric d s 6 conformally invariant quantum theory should be of considerable
interest w35x, as the latter includes non-critical non-supersymmetric strings as part of the
spectrum.
2. 3d CFTs from Freund–Rubin compactifications
Spontaneous compactification of M-theory to AdS 4 = X 7 is obtained from near-hori-

zon geometry of N coincident M2-branes. The configuration is equivalent to Freund–
Rubin compactification of the eleven-dimensional supergravity: through X 7 , the M2-
branes thread non-vanishing flux of four-form field strength
1
Gabgd s Qey7sea bgd , Q ' 96p 2 N l Pl6 . Ž 1.
(y g4
The parameter Q refers to so-called ‘Page’ charge w36x, Q ' py4HX 7 Ž ) G q C n G ..

Here, the d s 11 coordinates with indices A, B, . . . are decomposed into AdS 4 coordi-
nates x with indices a , b , . . . and X 7 coordinates y with indices a,b, . . . . We will also
adopt the eleven-dimensional metric convention as Žy,q, . . . ,q .. A Weyl rescaling of
AdS 4 and X 7 , appropriate for M-theory description, is given by g a b ™ e7s g a b and
g a b ™ ey2 s g a b .
In this section, we will be studying critical points of the scalar potential for various
choices of X 7 and, in the corresponding three-dimensional conformal field theories,
identifying an operator that gives rise to a renormalization group flow among the critical
points.
2.1. N 0 ,1,0 space
We will begin with the case X 7 s N p, q, r space w23x, which is a homogeneous space
w SUŽ3. = UŽ1.xrwUŽ1. = UŽ1.x. Consider a subset of all continuous deformation of
N p, q, r , in which the vielbein B a is given by
1 1 X 1 1 X
Bas
ž a
V a,
b
Ž pV 8 q q V 3 q r V 3 . ,
X X
g
V A, V A .
d / X
Here, V a s 1,2., V , V A s 4,5., V Ž A s 6,7., V and V X3 are the left invariant
aŽ 3 AŽ A 8
one-forms corresponding to the coset generators l a, l3 , l A , l A , l8 of SUŽ3. and the

generator of UŽ1. factor respectively. The integer p,q,r characterize the embedding of
UŽ1. = UŽ1. in SUŽ3. = UŽ1.. It is convenient to define
a s e sq u r2yÕ , b s e sy 3 u , g s e sq Ž uqÕqw .r2 , d s e sq Ž uqÕyw .r2 .
Ž 2.
The deformation considered above can be summarized compactly in terms of four-di-
mensional effective Lagrangian w21x
2 2 2 2
L4 s 12 y g 4 yR 4 y 632 Ž Em s . y 212 Ž Em u . y 3 Ž Em Õ . y Ž Em w . y V Ž s,u,Õ,w . ,
( ž /
where
V Ž s,u,Õ,w .
ž
s e 9 s y 32 e u Ž ey2 Õ q e Õqw q e Õyw . q 14 e u Ž ey2 Õq2 w q ey2 Õy2 w q e 4 Õ .
2 2
Ž1qx . Ž1yx .
q 14 e 8 uy4 Õ q
16
e 8 uq2 Õq2 w q
16 /
e 8 uq2 Õy2 w q Q 2 e 21 s
and Q is the aforementioned Page charge. Critical points of the deformation are
determined by the stationarity condition E V E us E V E Õs E V E ws 0 5. It turns out the
solution w23,37x is specified by one parameter c 6 :
64 5 2
16 Ž 54 c 2 q 3c q 2 .
a2s 2 Ž4
c q 3c q 2 . e 2 , bs" e,
Ž c q 2. Ž c q 2 . Ž 3c q 2 .
64 d 64 d
g 2s q 32 e 2 , d 2s q 32 e 2 ,
. ž / . ž /
2
cq 2
cy Ž 3.
Ž cq2 2 Ž cq2 2
5
Actually this is equivalent to the requirement w23x that the internal 7 manifold has to be an Einstein space.
The rescalings are fixed by this condition w23x.
6
The normalization convention we adopt here is as follows: Rba sy24 e 2dba , R ba s12 e 2d ba and Ga bgd s
e eabgd .
with y1 ( c ( 1 and d s " '1 y c 2 and c is related to
3p Ž 5c q 6 . d
x' sy . Ž 4.
q 3c q 2
It is known w23x that for a particular choice of p,q,r the isometry of N p, q, r is

SUŽ3. = SUŽ2. in which N s 3 supersymmetry survives. We call it type I solution. For
all the other values of p,q,r, supersymmetry is broken to N s 1 and the isometry group
is SUŽ3. = UŽ1.. Moreover, as we mentioned in Section 1, Page and Pope worked out a
different construction of metrics on the N 0,1,0 space and showed that there exists an
another Einstein metric, squashed from the first. The isometry is SUŽ3. = SUŽ2. and the
supersymmetry is N s Ž0,1.. We call this type II solution. Strictly speaking, the
analysis of Killing spinors in w24x shows that a possible different root of squashing
parameter gives different supersymmetry. In other words, Type I solutions are known to
preserve Ž3,0.. Similarly, the case of Type II solutions have Ž0,1. supersymmetry. Thus,
for the squashing with left-handed orientation, the renormalization group flow interpo-
lates between the boundary conformal field theories with N s 3 and N s 0 supersym-
metry while for the squashing with right-handed orientation, the flow interpolates
between conformal field theories with N s 0 and N s 1. From now on we will restrict
the foregoing discussions to the Ž p,q,r . s Ž0,1,0. case.
2.1.1. Type I solution for the space NI0 ,1,0

Type I solution with N s Ž3,0., as studied by Castellani and Roman w23x, is defined
by
2
y1 ( c ( y , d s '1 y c 2 .
'5
In this case, Eqs. Ž3. become
a 2 s 16 e 2 , b s "4 e, g 2 s 32 e 2 , d 2 s 32 e 2 , Ž 5.
where we put c s y1 and d s 0 by plugging p s 0 into Eq. Ž4.. Differentiating V with
respect to u,Õ,w with x s 0, one obtains
Vu u < s 21 = 2 3r7e 9 s I , VÕ Õ < s 6 = 2 3r7e 9 s I , Vw w < s y2 3r7e 9 s I ,
Vu Õ < s 0, Vu w < s 0, VÕ w < s 0,
evaluated at s I ,u I ,Õ I and w I which are the extremum values corresponding to field

s,u,Õ, and w, respectively,
9 1
e 12 s I s 18r7
, u I s 212 ln2, Õ I s 31 ln2, wI s 0
2 Q2
that can be obtained explicitly using Eq. Ž2. and Eq. Ž5. Žsee Fig. 1..
The conformal dimension of the perturbation operator representing squashing is
determined by fluctuation spectrum of the scalar fields. After rescaling '63 s s s,'21 u
Fig. 1. Ž u,Õ .-subspace slice of the N 0,1,0 scalar potential over us wy0.45,q0.15x, Õ s wy0.5,q0.5x. The
Type I vacuum is the saddle point in the middle-left corner, and the Type II vacuum is the local minimum in
the middle-right corner.
s u,'6 Õ s Õ,'2 w s w, one finds that the non-zero fluctuation spectrums for u,Õ and w
fields around the NI0,1,0 take
E 2V
Mu2u Ž NI0,1,0 . s ,u I ,Õ I ,w I s q2
3r7 9 s I
e ,
E u2 sI
E 2V
M Õ2Õ Ž NI0,1 ,0 . s ,u I ,Õ I ,w I s q2
3r7 9 s I
e ,
E Õ2 sI
E 2V 1
Mw2 w Ž NI0,1,0 . s ,u I ,Õ I ,w I s y 2 = 2 3r7e 9 s I .
E w2 sI
The cosmological constant L I is

3r4
VI 9'6 1 3
LI s
2
sy
16 ž /Q 2
'y 2
r IR l Pl2
.
One finds that the fluctuation spectrum for u,Õ,w fields around N s 3 fixed point takes
two positive values and one negative value:
1
Mu2u Ž NI0,1,0 . s q4 2
,
r IR
1
M Õ2Õ Ž NI0,1 ,0 . s q4 2
,
r IR
1
Mw2 w Ž NI0,1,0 . s y2 2
.
r IR
Via AdSrCFT correspondence, one finds that in d s 3 conformal field theory with
N s 3 supersymmetry, the squashing ought to be an irrelevant perturbation of conformal
dimension D s 4 for u,Õ fields and an relevant perturbation of conformal dimension
D s 2 or D s 1 for w field. According to the result of w38x, mass spectrum of the
Lichnerowitz scalar field f can be obtained from that of transverse spinor l T via
mf2 s ml TŽ ml T y 4. in the normalization convention, ŽI y 32 q mf2 . f s 0. Moreover,
from Ref. w39x, mlT s D q 3, where D is the Dirac operator of each spinor harmonics.
Hence, canonically normalized mass of the Lichnerowitz scalar is given by mf̃2 s yŽ D
q 3.Ž D y 1. q 32. Unfortunately, at present, complete eigenvalues of the Dirac opera-
tor D is not known for the transverse harmonics, even though there have been attempts
recently to classify the full mass spectra w40,41x. We anticipate that the linear combina-
tion of masses of u and Õ fields should correspond to mf̃2 .
2.1.2. Type II solution for the space NII0 ,1,0

The Type II solution with N s Ž0,1. supersymmetry exists, as pointed out in w24x, in
the range of c:
2
( c ( 1, d s y'1 y c 2 .
'5
For NII0,1,0 , it leads to
a 2 s 400 2
9 e , b s " 203 e, g 2 s 160 2
9 e , d 2 s 160 2
9 e ,
where we have put c s 1 and d s 0. One again obtains

2 3r7 9 s II
2 3r7 9 s II
2 3r7
Vu u < s 189 e , VÕ Õ < s y6 e , Vw w < s 11 e 9 s II ,
510r7 510r7 510r7
2 3r7
Vu Õ < s y84 e 9 s II , Vu w < s 0, VÕ w < s 0,
510r7
all evaluated at the extremum
81 1
e 12 s II s 18r7 10r7
, u II s 212 ln 52 , Õ II s 31 ln 52 , w II s 0.
2 =5 Q2
From these, one calculates mass spectrum of the scalar fields straightforwardly:
E 2V 2 3r7
Mu2u Ž NII0,1,0 . s ,u II ,Õ II ,w II s q9 = e 9 s II ,
E u2 s II
510r7
2
E V 2 3r7
Mu2Õ Ž NII0,1,0 . s '
,u II ,Õ II ,w II s y2 14 = e 9 s II ,
E uE Õ s II
510r7
E 2V 2 3r7
M Õ2Õ Ž NII0,1 ,0 . s ,u II ,Õ II ,w II s y e 9 s II ,
E Õ2 s II
510r7
E 2V 11
2 3r7
Mw2 w Ž NII0,1,0 . s ,u II ,Õ II ,w II s 2 = e 9 s II .
E w2 s II
510r7
Diagonalizing the mass matrix, one obtains the mass eigenvalues

2 3r7 2 3r7 2 3r7
M 2 s y5 = 10r7
e 9 s II , q13 = 10r7
e 9 s II , q 112 = e 9 s II .
5 5 510r7
The eigenvector '2 u q '7 Õ s '42 Ž u q Õ . Žcorresponding to the ‘tachyonic’ eigen-

value. in fact represents squashing of N 0,1,0 manifold, whose magnitude is parametrized
by l2 :
7
Ž uqÕ .
2 1 3
l s e 4 .
For NI0,1,0 , l2 s 1r2, while, for NII0,1,0 , l2 s 1r10.
The cosmological constant of Type II solution L II is
3r4
VII 729'10 1 3
L II s
2
sy
2 = 10 3 ž / Q 2
'y 2
r UV l Pl2
.
One finds that the fluctuation spectrum for u,Õ,w fields around N s 1 fixed point takes
one negative value and two positive values:
1
Mu˜2u˜ Ž NII0,1,0 . s y 209 2
,
r UV
1
M Õ˜2Õ˜ Ž NII0,1 ,0 . s 529 2
,
r UV
1
Mw2 w Ž NII0,1,0 . s 229 2
,
r UV
14
where u˜ s ( 3
Ž u q Õ . and Õ˜ s
1 Ž7u y 2 Õ ..
'3
From the above mass spectrum, one finds that, in N s 1 superconformal field theory,
the squashing deformation ought to be a relevant perturbation of conformal dimension
D s 5r3 or 4r3. The scaling dimensions of other deformations are D s 13r3 for Õ̃
field and D s 11r3 for w field, respectively.
2.2. SO(5) r SO(3)max ,V5,2 ( R ),M 1,1,1,Q 1,1,1,N p ,q ,r
We will now study other homogeneous Einstein manifolds, for which four-dimen-
sional scalar potential is known on an appropriate subspace.
2.2.1. N p ,q / I0,II
,r
As recalled above, Page and Pope have shown that, for q / 0, by geometric
squashing, NIp, q, r manifold leads, much as in N 0,1,0 ’s, to NIIp, q, r. It is straightforward to
show that the effective potential Eq. Ž4., which is valid for all x hence Ž p,q / 0,r .,
have two critical points. The supersymmetry preserved by NIp, q, r and NIIp, q, r is N s Ž1,0.
and N s Ž0,1., respectively. On the other hand, depending on the value of x, relative
magnitude of the cosmological constants, V I,II , changes. Hence, in three-dimensional
p, q, r
conformal field theory with SUŽ3. = UŽ1. global symmetry dual to AdS 4 = NI,II ’s,
there will be two classes of renormalization group flows: one flowing from a N s 1 UV
fixed point to a non-supersymmetric IR fixed point, and another flowing from a
non-supersymmetric UV fixed point to a N s 1 IR fixed point.
2.2.2. SO(5) r SO(3)max

The SO Ž5.rSO Ž3. max seven-manifold is constructed from maximal embedding of
SO Ž3. in SO Ž5.. Let us consider squashing each coset direction as specified by the
vielbein:
1 1 1
Bas
ž a
V a,
b
V aˆ ,
g /
V0 ,
where V a Ž a s 1,2,3., V â Ž aˆ s 1,2,3. and V 0 are the left invariant one-forms corre-
sponding to the generators X a, X â, X 0 of SO Ž5., respectively. Parametrizing as
a s e sq Ž uqÕ.r2 , b s e sq Ž uyÕ.r2 , g s e sy 3 u ,
the squashing deformation can be summarized compactly in terms of the following
four-dimensional effective Lagrangian w22x:
2 2 2
L4 s y g 4 R 4 y 632 Ž Em s . y 212 Ž Em u . y 32 Ž Em Õ . y V Ž s,u,Õ . .
( ž /
Here, the scalar potential is given by
V Ž s,u,Õ . s e 9 s Ž y12 e uq Õ y 7e uy Õ y 2 ey6 u q e 8 u q ey6 uq2 Õ
qey6 uy2 Õ q e uy3 Õ y 30 e u y 16 ey Ž 5uqÕ . r2 . q Q 2 e 21 s .
Again, Q is the conserved ‘Page’ charge. It is straightforward to check that there exists
only one critical point, u s Õ s 0. In the subspace of the perturbations considered above,
one finds that the strongly interacting d s 3 quantum field theory dual to AdS 4 =
SO Ž5.rSO Ž3. max has only one isolated Žsuper.conformal fixed point, hence, no non-triv-
ial renormalization group flow.
2.2.3. V5,2 ( R )
The V5,2 ŽR. is a real Stiefel manifold, constructed by embedding SO Ž3. in SO Ž5. in
such a way that 5 of SO Ž5. branches into 3 of SO Ž3. plus two singlets. The isometry of
V5,2 Ž R . is SO Ž5. = SO Ž2.. Consider deformations rescaling each coset direction as
1 1 1
Bas
ž a
V m,
b
V mˆ ,
g
V0 ,
/
where V m Ž m s 1,2,3., V m̂ Ž m
ˆ s 1,2,3. and V 0 are the left invariant one-forms corre-
sponding to the generators X m s T 4 m , X m̂ s T 5m , X 0 s T 45 where T i j Ž i, j s 1,2,3,4,5.
are generators of SO Ž5.. In terms of parametrization
a s e sq Ž uqÕ.r2 , b s e sq Ž uyÕ.r2 , g s e sy 3 u ,
the four-dimensional effective Lagrangian w22x can be written as
2 2 2
L s y g 4 R 4 y 632 Ž Em s . y 212 Ž Em u . y 32 Ž Em Õ . y V Ž s,u,Õ . ,
( ž /
where
V Ž s,u,Õ . s e 9 s Ž y6 e uq Õ y 6 e uy Õ y 2 ey6 u q ey6 uq2 Õ
qey6 uy2 Õ q e 8 u . q Q 2 e 21 s
and Q is the Page charge. The potential exhibits only one critical point at u s 17 ln 32 ,Õ s 0.
As in SO Ž5.rSO Ž3. max case, there is no renormalization group flow in the correspond-
ing three-dimensional conformal field theory.
2.2.4. M 1,1,1
The manifold M 1,1,1 is a homogeneous space w SUŽ3. = SUŽ2. = UŽ1.xrw SUŽ2. =
UŽ1. = UŽ1.x. For the vielbein B a , we rescale each coset direction as
1 1 X 1
Bas ž V m, Ž '3 V 8 q V 3 q 2 V 3 . , a V A / ,
b c
X
where V m Ž m s 1,2., V 3 , V 8 , V A Ž A s 4,5,6,7., and V 3 are the left invariant one-forms
corresponding to the coset generators s m and s 3 of SUŽ2., l8 and l A of SUŽ3., and
the generator of UŽ1. factor respectively. In terms of parametrization
a s e sq Ž uyÕ.r2 , b s e sq u r2qÕ , c s e sy 3 u ,
2 2 2
L s 12 y g 4 yR 4 y 632 Ž Em s . y 212 Ž Em u . y 3 Ž Em Õ . y V Ž s,u,Õ . .
( ž /
The scalar potential is given by
V Ž s,u,Õ . s e 9 s Ž y3e uyÕ y e uq2 Õ q 98 e 8 uy2 Õ q 14 e 8 uq4 Õ . q Q 2 e 21 s ,
where again Q is the conserved Page charge. The stationary conditions E V E ss E V E us
E V E Õs 0 lead to a set of cubic equations for t ' 3ey3 Õ :
t 3 y 3t 2 q 4 t y 4 s Ž t y 2 . Ž t 2 y t q 2 . s 0.
One finds that there is no extra critical point except t s 2. Hence, one concludes that
there ought to be no non-trivial renormalization group flow in the dual, d s 3
Žsuper.conformal field theory.
2.2.5. Q 1,1,1
The manifold Q 1,1,1 is a homogeneous space w SUŽ2. = SUŽ2. = SUŽ2.xrwUŽ1. = UŽ1.
= UŽ1.x. For the vielbein B a , we rescale each coset direction as
1 1 X XX 1 X 1 XX
Bas ž V i, Ž V 0qV 0 qV 0 . , V i , V i , /
a d b c
X X X XX XX XX
where V i s 1,2., V , V Ž i s 4,5., V , V Ž i s 6,7. and V 0 are the left invariant
iŽ 0 i 0 i
one-forms corresponding to the coset generators s i and s 3 of SUŽ2.,SUŽ2. andSUŽ2.

respectively. By parametrizing
a s e sq Ž uqÕqw .r2 , b s e sq Ž uqÕyw .r2 , c s e sq u r2yÕ , d s e sy 3 u ,
2 2 2 2
L s 12 y g 4 yR 4 y 632 Ž Em s . y 212 Ž Em u . y 3 Ž Em Õ . y Ž Em w . y V Ž s,u,Õ,w . ,
( ž /
where
V Ž s,u,Õ,w . s e 9 s Ž ye uqÕqw y e uqÕyw y e uy2 Õ q 14 e 8 uq2 Õq2 w
q 14 e 8 uq2 Õy2 w q 14 e 8 uy4 Õ . q Q 2 e 21 s .
Table 1
>
Classification of Einstein spaces X 7 and renormalization group flows. The flow found in w1x is marked ,
while the ones found in Subsections 2.1 and 2.2 are marked w and †
X7 Isometry Holonomy Supersymmetry RG flow
Round S 7
SO Ž8. 1 Ž8,8. Yes>
Squashed S 7 SO Ž5.= SUŽ2. G2 1 Yes>
SO Ž5.rSO Ž3. ma x SO Ž5. G2 Ž1,0. No
V5,2 ŽR. SO Ž5.=UŽ1. SUŽ3. Ž2,0. No
M 1,1,1 SUŽ3.= SUŽ2.=UŽ1. SUŽ3. Ž2,0. No
M p, q, r SUŽ3.= SUŽ2.=UŽ1. SO Ž7. 0
Q1,1,1 SUŽ2. 3 =UŽ1. SUŽ3. Ž2,0. No
Q p, q, r SUŽ2. 3 =UŽ1. SO Ž7. 0
NI0,1,0 SUŽ3.= SUŽ2. SUŽ2. Ž3,0. Yes w
NII0,1,0 SUŽ3.= SUŽ2. G2 Ž0,1. Yes w
NIp, q / 0, r SUŽ3.=UŽ1. G2 Ž1,0. Yes†
NIIp, q / 0, r SUŽ3.=UŽ1. G2 Ž0,1. Yes†
The stationary conditions E V E us E V E Õs E V E ws 0 yield a set of simple algebraic

equations
a Ž1ya . sb Ž1yb . sg Ž1yg . sa 2 qb 2 qg 2 ,
where a ' a2 2 d 2, b ' b 2 2 d 2,g ' c 2 2 d 2. The only solution is
2 a s 1 y '1 y t s 2 b s 2g , 3 y 3'1 y t y 2 t s 0.
It is easy to see that there exists only one extremum value when t s 3r4. One again
concludes that there is no non-trivial renormalization group flow in the dual, d s 3
Žsuper.conformal field theory.
We have summarized our result of this section in Table 1. In the subspace of
squashing deformations considered, among various known X 7 ’s, we have found that
p, q / 0, r
only NI,II ’s turn out to be dual to non-trivial renormalization group flows of
strongly coupled three-dimensional field theories, connecting two Žsuper.conformal
fixed points. One novelty of these conformal field theories is that, even though the
superconformal symmetry is changed, the global symmetry is not changed at all, in
contrast to the deformation interpolating between round- and squashed-S 7 w1x.
Although we have not considered non-supersymmetric seven-manifolds M p, q, r or
p, q, r
Q in this paper, it is known that M p, q, r remains stable for the specific region of
98r243( p 2rq 2 ( 6358r4563 and Q p, q, r solutions are stable in a certain region
containing the point p s q s r s 1. We therefore anticipate that, if there exists more
than one critical points of the corresponding scalar potential, they will provide gravity
dual to strongly interacting, stable, non-supersymmetric field theories in three dimen-
sions, whose renormalization group flows interpolate interacting conformal fixed points
w35x.
3. 3d CFTs from Englert compactifications
So far, we have deduced existence of three-dimensional conformal field theories out

of Freund–Rubin compactification of M-theory. Let us now consider more general
compactifications by relaxing the restriction that the eleven-dimensional metric is a

product space and that there is no magnetic four-form field strength flux threaded on
X 7 . The first solution of this sort has been found by Englert w28x. In contrast to the
Freund–Rubin compactifications, the symmetry of the vacuum is no longer given by the
isometry group of X 7 but rather by the group which leaves invariant both the metric g ab
and four-form magnetic field strength Ga b c d . In the Englert compactification w28x,
non-vanishing Ga b c d on the round S 7 breaks SO Ž8. down to SO Ž7.. The precise
interpretation of Englert compactification in terms of microscopic configuration of
coincident M2-branes is still lacking. Nevertheless, AdSrCFT correspondence implies
that there ought to be a three-dimensional Žsuper.conformal field theory for each
Englert-type compactification as well.
In Kaluza–Klein supergravity, it is well known that the four-dimensional N s 8
gauged supergravity can be embedded consistently into the eleven-dimensional super-
gravity. As shown in w42x, the 70 scalars of N s 8 supergravity live on the coset space
E7rSUŽ8. and are described by an element V Ž x . of the fundamental 56-dimensional
representation Ž56-bein. of E7 :
u iI jJ Ž x . Õi j K L Ž x .
V Ž x. s ,
Õ klIJ Ž x . u Kk lLŽ x .
where SUŽ8. index pairs w ij x, . . . and SO Ž8. index pairs w IJ x, . . . are antisymmetrized
and therefore u and Õ are 28 = 28 matrices. Complex conjugation is done by raising or
lowering indices, for example, Ž u iI jJ . w s u iI Jj . Under local SUŽ8. and local SO Ž8., the
matrix V Ž x . transforms as V Ž x . ™ UŽ x . V Ž x . Oy1 Ž x ., where UŽ x . ; SUŽ8. and
O Ž x . ; SO Ž8. and matrices UŽ x ., O Ž x . are elements of the 56-dimensional representa-
tion. By appropriate gauge fixing of the local SUŽ8. symmetry, the 56-bein V Ž x . can
be brought into the following form:
'2
0 y f i jk l
4
V Ž x . s exp ,
'2 mn pq
y f 0
4
where f i jk l is a complex self-dual tensor describing the 35 scalars 35 v Žthe real part of
f i jk l . and 35 pseudoscalar fields 35c Žthe imaginary part of f i jk l . of the N s 8
supergravity. Note that, after gauge fixing, there is no distinction between SO Ž8. and
SUŽ8. indices.
The scalar potential of the gauged N s 8 supergravity is known to possess four
critical points with at least G 2 invariance w43,44x. The maximally supersymmetric
vacuum with SO Ž8. symmetry, S 7, is where expectation value of both scalar and
pseudoscalar fields vanish. Let us denote self-dual and anti-self-dual tensors of SO Ž8.
I JK L I JK L
tensor as Cq and Cy , respectively, satisfying
I JM N M N K L
C" C" s 12 d KI JL " 4C "
I JK L
.
I JK L
Turning on the scalar fields proportional to Cq yields an SO Ž7.q invariant vacuum.
I JK L
Likewise, turning on pseudoscalar fields proportional to Cy yields SO Ž7.y invariant
Fig. 2. Scalar potential V Ž a , l.. The left axis corresponds to Õ s cos a and right one does l. The extremum
value V sy7.19 for G 2 occurs around Õ s 0.56 and l s 0.73 while the extremum value V sy6 for SO Ž8.
appears around l s 0. We take g 2 as 1 for simplicity.
vacuum. Both SO Ž7. " vacua are non-supersymmetric. However, simultaneously turning
I JK L I JK L
on both scalars and pseudoscalar fields proportional to Cq and Cy , respectively,
7
one obtains G 2-invariant vacuum with N s 1 supersymmetry . The most general
vacuum expectation value of 56-bein retaining G 2-invariance can be parametrized as
l
² f I JK L : s
2'2
Ž cos a CqI JK L q isin a CyI JK L . .
In this case, the elements of 56-bein V Ž x . can be written as
u KI JLŽ l . s 2 p 3 d KI JL q 12 Ž 1 q cos2 a . pq 2 Cq
I JK L
q 12 Ž 1 y cos2 a . pq 2 Cy
I JK L
y ipq 2 sin2 a Dy
I JK L
,
Õ I JK L Ž l . s 12 Ž 3e i a q ey3 i a . q 3 d KI JL q p 2 qcos a Cq
I JK L
y ip 2 qsin a Cy
I JK L
q 12 Ž e i a y ey3 i a . q 3 Dq
I JK L
, Ž 6.
where DI"JK L ' 12 Ž I JM N M N K L
Cq Cy I JM N M N K L
" Cy Cq ., p ' coshŽ lr2'2 . and q ' sinhŽ lr2
'2 ..
In the above parametrization, the scalar potential is given by w26x
V Ž a , l . s 2 g 2 Ž Ž 7Õ 4 y 7Õ 2 q 3 . c 3 s 4 q Ž 4Õ 2 y 7 . Õ 5 s7 q c 5 s 2 q 7Õ 3 c 2 s 5 y 3c 3 . ,
Ž 7.
where c ' coshŽ lr '2 ., s ' sinhŽ lr '2 . and Õ ' cos a . The potential is plotted in Fig.
2 and the four critical points are summarized in Table 2.
In this section, we will be identifying a renormalization group flow associated with
global symmetry breaking SO Ž8. ™ G 2 in a three-dimensional strongly coupled field
theory 8. We will show that the perturbation operator is relevant at the SO Ž8. invariant
7
The G 2 is the common subgroup of SO Ž7.q and SO Ž7.y.
8
As the SO Ž7. " vacua are unstable w31x, we will not consider in the foregoing discussions.
Table 2
Summary of four critical points: symmetry group, supersymmetry, vacuum expectation values of scalar and
pseudoscalar fields, and cosmological constants
Gauge symmetry Supersymmetry cos a c2 V
SO Ž8. N s8 1 y6 g 2
SO Ž7.y N s0 0 5
y
'
25 5
g 2 sy6.99 g 2
4
8
SO Ž7.q N s0 1 1
2
3
ž' / q1 y2P53r4 g 2 sy6.69 g 2
5
G2 N s1 1
(Ž 3y'3 . s 0.56 1
Ž3q2'3 . y
'
216 2 1r 4
3 g 2 sy7.19 g 2
2 5
'
25 5
UV fixed point but becomes irrelevant at the G 2 invariant IR fixed point. To identify
conformal field theory operator corresponding to the perturbation while preserving G 2
symmetry, we will consider harmonic fluctuations of space-time metric and l scalar
field around AdS 4 = S 7. From the scalar potential Ž7., one finds that the cosmological
constant L is given by
3
LSOŽ8. s y6 g 2 ' y ,
2
r UV l Pl2
where r UV is the radius of AdS 4 and l Pl is the eleven-dimensional Planck scale.
Conformal dimension of the perturbation operator representing this deformation is
calculated by fluctuation spectrum of the scalar fields. The kinetic term w45x for l field
can be obtained from y< Aai jk l < 2r96, where
Aai jk l s y2'2 Ž u Ii Jj Ea Õ k l I J y Õ i j I JEa u IkJl . .
Then, from the explicit forms of u KI JL and Õ I JK L in Eq. Ž6., one obtains < Aai jk l < 2 s
84Ž Ea l. 2 , where we have kept to quadratic order in the fluctuation of l. The resulting
kinetic term is y 78 Ž Ea l. 2 . After rescaling the l field as l s 74 l, one finds that the (
mass spectrum of the l field around SO Ž8. fixed point is given by
1
E l2 V Ž SO Ž 8 . . < ls0 s y2 g 2 l Pl2 s y2 2
. Ž 8.
r UV
Via AdSrCFT correspondence, one finds that in the corresponding N s 8 superconfor-

mal field theory, the G 2 symmetric deformation ought to be a relevant perturbation of
conformal dimension D s 1 or D s 2. Recall that, on S 7, mass spectrum of the
representation corresponding to SO Ž8. Dynkin label ( n ,0,2,0 ) is given by
&
2
M 2 s Ž Ž n q 1. y 9 . m2 ,
where m2 is the mass-squared

&
parameter of a given AdS 4 space-time and a scalar field S
2
is defined by Ž DAdS q M . S s 0. This follows from the known mass formula w46x
M 2 s ŽŽ n q 1. 2 y 1. m 2 for OyŽ1. and the fact that M 2 is traditionally defined accord-
&
ing to Ž DAdS y 8 m2 q M 2 . s 0. For 35c corresponding to n s 0, M 235c s y8m 2 and
2
this ought to equal to Eq. Ž8.. Recalling that r UV s r S27r4 s 1r4m2 ,
1 &
E l2 V Ž SO Ž 8 . . < ls0 s y2 2
s M 235c
r UV
On the other hand, the mass spectrum & of the representation corresponding to SO Ž8.
2
Dynkin label ( n H 2,0,0,0 ) is given by M 2s Ž Ž n y 1 . y 9 . m 2 . This follows from the
2 2 2
known &mass formula w46x M s ŽŽ n y 1. y 1. m for OqŽ1.. For 35 v corresponding to
n s 0, M 235 v s y8m 2 and this ought to equal to Eq. Ž8..
Let us next consider the conformal fixed point corresponding to the G 2 symmetry.
Again, from the scalar potential Ž7., one finds that cosmological constant L is given by
216'2 3
LG 2 s y 31r4 g 2 ' y .
25'5 2
r IR l Pl2
The mass spectrum for the l field around G 2 fixed point takes a positive value:
1
E l2 V Ž G 2 . < c 2s 1 s 15.446 g 2 l Pl2 s 6.443 2
.
5
'
Ž3q 2 3 . r IR
One finds that in the corresponding three-dimensional conformal field theory with
N s 1 supersymmetry, the G 2 symmetric deformation ought to be an irrelevant pertur-
bation of conformal dimension D s 4.448.... We thus conclude that the perturbation
operator dual to the l field induces non-trivial renormalization group flow from N s 8
superconformal UV fixed point with SO Ž8. symmetry to N s 1 superconformal IR
fixed point with G 2 symmetry.
It is known that, in N s 8 supergravity, there also exists a N s 2 supersymmetric,
SUŽ3. = UŽ1. invariant vacuum w47x. To reach this critical point, one has to turn on
expectation values of both scalar and pseudoscalar fields as
1
² f I JK L : s Ž l XqI JK L q i lX XyI JK L . ,
2'2
where
1234 5678 1256 3478 1278 3456
Xq
i jk l s q Ž d i jk l q d i jk l . q Ž d i jk l q d i jk l . q Ž d i jk l q d i jk l . ,
1357 2468 1268 2457

Xy
i jk l s y Ž d i jk l y d i jk l . q Ž d i jk l y d i jk l .
q Ž d i1458 2367 1467 2358

jk l y d i jk l . y Ž d i jk l y d i jk l . .
The scalar potential is
V s 12 g 2 Ž sX 4 Ž Ž x 2 q 3 . c 3 q 4 x 2 Õ 3 s 3 y 3Õ Ž x 2 y 1 . s 3 q 12 xÕ 2 cs 2
y6 Ž x y 1 . cs 2 q 6 Ž x q 1 . c 2 sÕ . q 2 sX 2 Ž 2 Ž c 3 q Õ 3 s 3 . q 3 Ž x q 1 . Õs 3
q6 xÕ 2 cs 2 y 3 Ž x y 1 . cs 2 y 6 c . y 12 c . ,
where
c ' cosh Ž lr'2 . ,
s ' sinh Ž lr'2 . ,
cX ' cosh Ž lXr'2 . ,
sX ' sinh Ž lXr'2 .

Õ ' cos a , x s cos2 f .
At the critical point, s s 1r '3 , sX s 1r '2 and a s 0, one finds that the cosmological
constant is given by V s y 9'3 g 2 . Hence, one expects that there ought to exist a renormalization group
2
flow between N s8 SO Ž8. fixed point and N s 2 SUŽ3.=UŽ1. fixed point.
4. 6d CFTs from inhomogeneous compactification
We finally study the case of near-horizon geometry of coincident M5-branes, as

described by AdS 7 gauged supergravity. Recently, in w32–34x, geometrical interpretation
for vacua in seven-dimensional N s 1 gauged supergravity has been given in terms of
Englert type compactification of eleven-dimensional supergravity, having non-vanishing
electric or tilted magnetic four-form fluxes and inhomogeneous metric deformation, by a
set of consistent non-linear ansatz. The inhomogeneous deformations of the S 4 is
parametrized by a scalar field f , while the SUŽ2. gauge fields, which are the surviving
subgroup of the SO Ž5. Yang–Mills fields of the maximal gauged supergravity after the
deformation are associated with the right translations under the SUŽ2.. The full
two-parameter potential of the resulting seven-dimensional gauged supergravity turns
out to be equal to w48x Žsee also Ref. w49x.
2
f
8 3 '5
2 y f y f 2
V Ž f . s 16 h e '5 y 8'2 hg e '5 y g e , Ž 9.
1
where h and g are arbitrary real constants and ey '5 f represents the SUŽ2. gauge
coupling constant. As shown in Fig. 3, provided hrg ) 0, the scalar potential has two
extrema.
Fig. 3. Schematic shape of the scalar potential of ds 7, N s1 gauged supergravity.

A local maximum of the potential Eq. Ž9. is located at

1 8'2 h
fIs ln ,
'5 g
whose curvature equals
2r5
E 2V '2 h 1
Ef 2
fs f I
s y2 = 2 1r5
g 2
ž / g
' y8 2
r UV
. Ž 10 .
One then infers the radius of AdS97 as
2
4
r UV s 2r5
s y15rVI ,
Ž 2 hrg . g2
where
2r5
15 '2 h
VI s y
2 = 2 4r5
g 2
ž /g
.
According to the analysis of w50,51x, if scalar field fluctuation about the maximum of the
2 .
potential V satisfy ŽI q a r UV f s 0, then perturbative stability is guaranteed provided
2
a ( Ž d y 1. r4. In the present case, the stability bound yields a s s 9 when d s 7. Eq.
Ž10. is less than the bound, hence, is stable. In fact, the vacuum is N s 1 supersymmet-
ric. Conformal dimension of the perturbation operator representing the squashing is
calculated by fluctuation spectrum of the scalar fields. From Eq. Ž10., via AdSrCFT
correspondence, one finds that, in six-dimensional N s Ž1,0. superconformal field
theory, the perturbation operator representing the scalar field deformation ought to be a
relevant perturbation of conformal dimension D s 2 or D s 4.
A local minimum of the scalar potential Ž9. is located at
1 4'2 h
f II s ln ,
'5 g
around which
2r5
E 2V '2 h 1
Ef 2
f s f II
s2=2 4r5
g 2
ž / g
' q12 2
r IR
.
One finds that the radius of AdS 7 is changed to

2
r IR s y15rVII ,
where
2r5
5 '2 h
VII s y
2 1r5
g 2
ž / g
.
9 2
The radius of AdS 7 space is defined by Ra bgd syŽ g ag g bd y g a d g bg .r r UV .
Dual to the critical point is a six-dimensional conformal field theory with no supersym-
metry, for which the perturbation associated with the squashing deformation is an
irrelevant perturbation of conformal dimension D s 3 q '21 .
One thus finds that, in the subspace representing inhomogeneous deformation of S 4 ,
there ought to be a six-dimensional quantum theory which interpolates between N s
Ž1,0. superconformal fixed point in the UV and non-supersymmetric fixed point in the
IR. Note that the conformal field theory around the IR fixed point is a stable one, despite
the absence of any supersymmetry, which follows from the general argument of Ref.
w35x. The existence of a stable, non-supersymmetric d s 6 conformal field theory would
be of considerable interest, as the theory would contain stable, non-critical non-super-
symmetric strings as part of the spectra.
Acknowledgements
We are grateful to H. Nicolai and O. Yasuda for helpful correspondences on their

works.
References
w1x C. Ahn, S.-J. Rey, Nucl. Phys. B 565 Ž2000. 210.

w4x S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Phys. Lett. B 428 Ž1998. 105. hep-thr9802109.
w5x M.J. Duff, B.E.W. Nilsson, C.N. Pope, Phys. Rep. 130 Ž1986. 1.
w6x E. Cremmer, B. Julia, J. Scherk, Phys. Lett. B 76 Ž1978. 409.
w7x E. Cremmer, B. Julia, Nucl. Phys. B 159 Ž1979. 141.
w8x P.G.O. Freund, M.A. Rubin, Phys. Lett. B 97 Ž1980. 233.
w9x G. Jensen, Duke Math. J. 42 Ž1975. 397.
w10x J.-P. Bourguignon, H. Karcher, Ann. Sci. Normale Sup. 11 Ž1978. 71.
w11x M.A. Awada, M.J. Duff, C.N. Pope, Phys. Rev. Lett. 50 Ž1983. 294.
w12x M.J. Duff, B.E.W. Nilsson, C.N. Pope, Phys. Rev. Lett. 50 Ž1983. 2043.
w13x M.J. Duff, B.E.W. Nilsson, C.N. Pope, Phys. Rev. Lett. 51 Ž1983. 846, errata.
w14x L. Girardello, M. Petrini, M. Porrati, A. Zaffaroni, J. High Energy Phys. 9812 Ž1998. 022. hep-
thr9810126.
w15x M. Porrati, A. Starinets, Phys. Lett. B 454 Ž199. 77. hep-thr9903085.
w16x D.Z. Freedman, S.S. Gubser, K. Pilch, N.P. Warner, hep-thr9904017.
w17x M. Cvetic, S. Griffies, S.-J. Rey, Nucl. Phys. B 381 Ž1992. 301. hep-thr9201007.
w18x M. Cvetic, S. Griffies, S.-J. Rey, Nucl. Phys. B 389 Ž1993. 3. hep-thr9206004.
w19x B.S. Acharya, J.M. Figueroa-O’Farrill, C.M. Hull, B. Spence, Adv. Theor. Math. Phys. 2 Ž1998. 1249.
hep-thr9808014.
w20x D.R. Morrison, R. Plesser, Adv. Theor. Math. Phys. 3 Ž1999. 1. hep-thr9810201.
w21x O. Yasuda, Phys. Rev. Lett. 53 Ž1984. 1207.
w22x O. Yasuda, Phys. Rev. D 31 Ž1985. 1899.
w23x L. Castellani, L.J. Romans, Nucl. Phys. B 238 Ž1984. 683.
w24x D.N. Page, C.N. Pope, Phys. Lett. B 147 Ž1984. 55.
w25x D. Fabbri et al., 3D superconformal theories from Sasakian seven-manifolds: New non-trivial evidence
for AdS 4 r CFT3 , hep-thr9907219.
w26x B. de Wit, H. Nicolai, N.P. Warner, Nucl. Phys. B 255 Ž1985. 29.
w27x M.J. Duff, C.N. Pope, Kaluza–Klein supergravity and the seven sphere, in: Supersymmetry and
Supergravity ’82, ed. S. Ferrara, J.G. Taylor, P. van Nieuwenhuizen ŽWorld Scientific, Singapore, 1983..
w28x F. Englert, Phys. Lett. B 119 Ž1982. 339.
w29x B. de Wit, H. Nicolai, Phys. Lett. B 148 Ž1984. 60.
w30x C.N. Pope, N.P. Warner, Phys. Lett. B 150 Ž1985. 352.
w32x H. Nastase, D. Vaman, P. van Nieuwenhuizen, Consistent non-linear K-K reduction of 11-d supergravity
on AdS 7 = S 4 and self-duality in odd dimensions, hep-thr9905075.
w33x H. Lu, C.N. Pope, Exact embedding of N s1,ds 7 gauged supergravity in ds11, hep-thr9906168.
w34x H. Lu, C.N. Pope, Mod. Phys. Lett. A 12 Ž1997. 1087.
w35x M. Berkooz, S.-J. Rey, J. High-Energy Phys. 9901 Ž1999. 014. hep-thr9807200.
w36x D.N. Page, Phys. Rev. D 28 Ž1983. 2976.
w37x L. Castellani, Nucl. Phys. B 254 Ž1985. 266.
w38x R. D’Auria, P. Fre,
´ Ann. Phys. 162 Ž1985. 372.
w39x R. D’Auria, P. Fre,
´ Ann. Phys. 157 Ž1984. 1.
w40x P. Termonia, The complete N s 3 Kaluza–Klein spectrum of 11-d supergravity on AdS 4 = N 010 ,
hep-thr9909137.
w41x P. Fre,
´ L. Gualtieri, P. Termonia, The structure of N s 3 multiplets in AdS 4 and the complete
OSpŽ3 <4.= SUŽ3. spectrum of M-theory on AdS 4 = N 010 , hep-thr9909188.
w42x E. Cremmer, B. Julia, Phys. Lett. B 80 Ž1978. 48.
w43x N.P. Warner, Phys. Lett. B 128 Ž1983. 169.
w44x N.P. Warner, Nucl. Phys. B 231 Ž1984. 250.
w46x B. Biran, A. Casher, F. Englert, M. Rooman, P. Spindel, Phys. Lett. B 134 Ž1984. 179.
w47x H. Nicolai, N.P. Warner, Nucl. Phys. B 259 Ž1985. 412.
w48x L. Mezincescu, P.K. Townsend, P. van Nieuwenhuizen, Phys. Lett. B 143 Ž1984. 384.
w49x K. Skenderis, P.K. Townsend, Gravitational Stability and Renormalization-Group Flow, hep-thr9909070.
w50x P. Breitenlohner, D.Z. Freedman, Phys. Lett. B 115 Ž1982. 197.
w51x P. Breitenlohner, D.Z. Freedman, Ann. Phys. 144 Ž1982. 249.
Magnetic interactions, the renormalization group and color

superconductivity in high density QCD
Stephen D.H. Hsu a , Myckola Schwetz b
a
Department of Physics, UniÕersity of Oregon, Eugene, OR 97403-5203, USA
b
Department of Physics and Astronomy, Rutgers UniÕersity, Piscataway, NJ 08855-0849, USA
Received 24 August 1999; received in revised form 6 October 1999; accepted 8 October 1999
Abstract
We investigate the effect of long range magnetic interactions on the renormalization group
ŽRG. evolution of local Cooper pairing interactions near the Fermi surface in high density QCD.
We use an explicit cut-off on momentum modes, with special emphasis on screening effects such
as Landau damping, to derive the RG equations in a gauge invariant, weak coupling expansion.
We obtain the Landau pole D ; m gy5 expŽy3p 2r '2 g ., although the structure of our equations
differs from previous results. We also investigate the gap equation, including condensates of
higher angular momentum. We show that rotational invariance is unbroken at asymptotically high
density, and verify that D is the correct value of the gap when higher modes are included in the
analysis. q 2000 Elsevier Science B.V. All rights reserved.
PACS: 11.15.Ex; 12.38.yt; 12.38.Mh

Keywords: QCD; Barion density; Color superconductivity
1. Introduction
In this paper we study the behavior of quark matter at high density and low
temperature w1–25x. Under these conditions QCD exhibits color superconductivity,
caused by the condensation of diquarks. This condensation is analogous to the Cooper
pairing observed in ordinary superfluids, and can be shown to occur in the presence of
even very weakly attractive interactions. The reason for this is the special nature of
physics very close to the Fermi surface ŽFS..
At high density, characterized by a chemical potential m which is much larger than
the current quark masses and the QCD scale LQC D , the typical momentum transfer in
quark–quark interactions is of order m , and therefore it is plausible that the dynamics
E-mail addresses: hsu@duende.uoregon.edu ŽS.D.H. Hsu., myckola@baobab.rutgers.edu ŽM. Schwetz..
PII: S 0 5 5 0 - 3 2 1 3 Ž 9 9 . 0 0 6 5 5 - 0
212 S.D.H. Hsu, M. Schwetzr Nuclear Physics B 572 (2000) 211–226
can be understood via perturbative gluon exchange. Of course, small angle scatterings,
which involve small momentum transfers, are still problematic and require special
attention.
In recent work, Son w11x showed that long range magnetic interactions lead to a
modification of the RG equations originally derived in Refs. w8,9x for the case of local
interactions. The magnetic effects are strong enough to modify the parametric depen-
dence of the position of the Landau pole Žand hence the superfluid gap D . on the gauge
coupling constant. Son finds
3p 2
D ; m gy5exp y
ž '2 g / . Ž 1.
That D should scale like expŽycrg . is easy to see w12x by considering the gap equation
with a massless, or weakly damped, gauge propagator. In the usual case of a local four
fermion interaction, the gap integral exhibits a logarithmic divergence which, roughly
speaking, is cut off near the FS by the gap itself. Solving for D yields a result of the
form D ; expŽycrG ., where G is the four fermion coupling, and is of order g 2 if it
arises from the exchange of a gauge boson. However, if the four fermion interaction is
replaced by a weakly damped gauge propagator, an additional logarithmic divergence
appears due to small angle scattering of the fermions. This divergence is again regulated
by a scale related to the gap itself, and the resulting exponent of the solution is roughly
the square root of the what appeared in the local case: D ; expŽycrg ..
Our intention here is to understand this behavior in terms of the evolution of Cooper
pairing interactions near the Fermi surface. Despite the long range of the magnetic
interactions, we find that the problem can still be formulated in terms of local operators
which are, essentially, terms in the expansion of the magnetic gluon form factor. The
reason that this is possible is because we retain at all times an explicit cutoff on long
wavelength modes, which keeps all quantities finite. This approach is somewhat
different from that of Son w11x, who studied the RG evolution of scattering amplitudes
themselves. As discussed below, our results differ from his. Primarily, we believe that
this is due to conceptual problems in applying the RG directly to scattering amplitudes.
The paper is organized as follows. In Section 2 we give a description of our cutoff
scheme and the resulting effective Lagrangian. In Section 3 we discuss the problem of
small angle scattering and gluon screening effects. In Section 4 we compute our RG
equations and compare them with Son’s. In Section 5 we investigate the possibility of
breaking rotational symmetry. The final section contains additional discussion of our
results.
2. RG scheme
We adopt a Wilsonian RG procedure, with a hard IR cutoff on spatial momentum, L

w8,9,26,27x. As L ™ 0, only quark excitations very near the FS, as well as soft gluons,
are left in the effective theory. Our prescription differs from what is often used in QFT,
where a hard cut-off is imposed on energy as well as momentum, but it has some
advantages. In particular, in our scheme integration over modes corresponds to shrinking
the Hilbert space of the model in the basis of energy eigenstates.
S.D.H. Hsu, M. Schwetzr Nuclear Physics B 572 (2000) 211–226 213
The effective Lagrangian has the form

LL s LQCD q Ý On , Ž 2.
n
where the On are local operators involving quark and gluon fields, which are the result
of the integration over higher frequency quarks and gluons. The most important of these
operators are the marginal Cooper pairing interactions which involve four quarks. All
other quark interactions can be shown to be irrelevant in the limit of small L w8,9x.
The Cooper pairing interaction is of the form
G Ž k y q . cq Ž k 0 ,k . gm PL , R cq Ž q0 ,q . cq Ž k 0 ,y k . g m PL , R cq Ž q0 ,y q . , Ž 3.
where cq denotes the projection of the quark field
cqs 12 Ž 1 q a P pˆ . c Ž p . , Ž 4.
and consists of quark, rather than antiquark degrees of freedom. We will be interested in
the case where all of the external quarks in this operator are essentially on-shell. Note
that the incoming and outgoing quarks have almost equal and opposite momenta. Near
the FS, the form factor GŽ k y q . becomes a function of angle u s Ž k P q .rkq, since
< k < , < q < , m , and k 0 ,q0 , 0. ŽStrictly speaking, in the case of Landau damping, it is a
function of energy as well as angle. We will always assume that the energy transfer
k 0 y q0 in the gluon line is much less than but of order L.. In this paper we restrict
ourselves to the 3 color channel, which is attractive, and to the LL Žor equivalently RR.
chirality channel, which has been shown to dominate the LR channel w8,9x. It is
straightforward to derive the related RG equations for other channels using our
techniques.
In previous work we took the form factor GŽ u . to be a constant w8,9x. This is
appropriate at sufficiently low energy, if screening masses exist for both the magnetic
and electric components of the gluon, which is likely to be the case at intermediate
densities where the coupling is not small. However, as argued by Son w11x, a magnetic
mass for the gluon is unlikely to arise within perturbation theory. The magnetic mass is
due solely to nonperturbative effects, and is presumably of order expŽy1rg 2 .. At high
density it is therefore likely to be smaller than the eventual superfluid gap, and hence
plays no role in the analysis. Instead, w11x focused on the role of Landau damping on the
magnetic interactions.
In order to consider long range magnetic interactions, it is necessary to expand the
form factor GŽ u . in components with definite angular momentum. We can then study
the RG evolution of each of these components. Let
G Ž u . s Ý Ž 2 l q 1 . Pl Ž cos u . G l , Ž 5.
l
so that each component

1
G l s 12 Hy1d Ž cos u . P Ž cos u . G Ž u . .
l Ž 6.
This integral exhibits a logarithmic divergence in the case of a massless, or weakly

damped, gluon ŽFig. 1.. However, in our regularization scheme GŽ u . contains only the
Fig. 1. One gluon exchange.
effects of gluons which have been integrated out above the cutoff L. For nonzero cutoff
the components Gl will be finite, but exhibit a logarithmic dependence on L1. It is this
logarithmic dependence that leads to the constant O Ž g 2 . term in the RG equations noted
by Son w11x.
3. Screening effects
In this section we discuss the incorporation of screening effects in our procedure. As

mentioned previously, small angle scattering of quarks must be considered carefully, as
the simple perturbative expansion in powers of g 2 may break down. Indeed, it is easy to
see that vacuum polarization corrections to any graph can become important if the gluon
momentum is sufficiently small. Resummation of these effects leads to a screened
propagator which is well behaved at small momentum, as long as the energy is nonzero.
The gluon propagator, including vacuum polarization effects from virtual quarks w28x,
has the following form in covariant gauge:
1 T
1 Qm Qn
Dmn s 2
Pmn q 2
PmnL y j , Ž 7.
GqQ FqQ Q4
T
where Q s Ž q4 ,q . s Žyv ,q . is the gluon Euclidean 4-momentum, and Pmn and PmnL are
transverse and longitudinal projectors correspondingly. In our leading order calculations
the propagator will always appear contracted with gamma matrices next to on-shell
external quark lines. Thus the gauge dependent part of Ž7. will vanish due to the
equations of motion, leading to a gauge-invariant result. Henceforth we will simply set
j s 0 in Ž7..
1
The precise form of this logarithm is dependent on the IR behavior of the propagator, as we will see
below.
The functions G and F are related to the gluon polarization tensor Pmn :
Q2 qi q j
Fs 2
P44 , G s 12 Pmn
T
Pmn s 12 d i j y ž / Pi j ,
q q2
which reflects the fact that the Lorentz symmetry is broken to 3D rotational symmetry.
The explicit expression of Pmn to one loop is
d4K
Pmn Ž Q . s g 2 H Ž 2p . 4
Tr gm Kugn Ž Ku y Qu . D Ž K . D Ž K y Q . , Ž 8.
where DŽ K . s 1rK 2 .
If the energy and momentum transfer in Fig. 2 are small then one may neglect Q in
the numerator of Ž8., as the dominant momenta in the loop will be k ; m Žthis is the
equivalent of the hard thermal loop approximation.. Then F and G take the familiar
form w28x
2 m2D Q 2 iv iv
Fs
q 2 ž ž //
1y
q
L0
q
,
2
iv iv iv iv
G s m2D
q ž ž // ž /
1y
q
L0
q
q
q
, Ž 9.
where
xq1
L0 Ž x . s 12 ln , Ž 10 .
xy1
and m2D s Nf Ž g 2m2r2p 2 . is the Debye screening mass. The small-x expansion of G
leads to the Landau damped magnetic gluon propagator
T
T
Pmn
Dmn Ž q0 ,q . s , Ž 11 .
2
1
q qi p m2D Ž < q0 <rq .
2
while the expansion of F leads to the usual longitudinal propagator, with Debye
screening. The effect of Landau damping is to cut off the small-q divergence in Ž11. at
q ; q01r3 m2r3
D .
One must be careful to compute the screening effects using a Wilsonian cutoff. When
we integrate over a shell in momentum space only the contribution of quarks which have
already been integrated out is to be included in the screening effects. This means that we
must re-examine the calculation which leads to the Debye mass and Landau damping,
and use cutoff-dependent versions of the vacuum polarization ' PmnL in our RG.
The terms in F and G can be shown to result from integration of quark modes within
roughly q of the FS. For example, the Landau damping term originates from an integral
of the form Žarising from Ž8. w28x.:
k4 n Ž E1 . y n Ž E2 .
I Ž v ,q . s dk d V H , Ž 12 .
E1 E2 i v y E1 q E2
Fig. 2. Vacuum polarization correction to gluon propagator.
where E1 s k, E2 s < k y q < , k y qcos u and

1
nŽ E . s b Ž Ey m . Ž 13 .
e q1
is the Fermi–Dirac distribution for quarks. nŽ E . is a theta function of Ž m y E . in the
low temperature limit in which we work. For a given value of u s Ž k P q .rkq, the modes
which contribute to Landau damping must be within , qcos u of the FS due to the theta
functions.
If the momentum transfer q - L < m , then none of these modes are included in the
calculation of PmnL . The corresponding F L and G L are changed drastically: in particular
the leading m2 behavior of F vanishes, and G becomes proportional to m2 . Fortunately,
our interest is only in the logarithmic divergences of diagrams, which are dominated by
gluon momenta satisfying the limit L < q < m , due to the form of Landau damping.
Ži.e., L ; D, while the dominant momentum transfer is q ; D1r3 m 2r3 D . In this limit the
.
results for F and G given in 9 are accurate.
Ž .
Finally, we mention the issue of screening due to the diquark condensate itself, which
is necessary for a self-consistent description of the region near the FS w23,24x. In a
conventional superconductor the relative size of the magnetic penetration depth l and
the correlation length j ; Dy1 determine whether one is in the type I Ž j 4 l. or type II
Ž l < j . regime. In a type II superconductor the magnetic screening length can be
computed using the London formula, and is proportional to the total density of
superconducting particles. In our case this would lead to a rather large effective
screening mass ly1 ; g 2m2 relative to the gap size. However, in high density quark
matter we are actually in the type I, or Pippard, regime. In this regime the effective
screening mass is much smaller, and scales with the gap D. A direct calculation of the
gluon vacuum polarization using the quark propagator in the presence of a gap Žsee Ž31.
below. shows that the London-type screening applies only to long wavelength gluons
with momentum less than ; D. Harder gluons experience a much smaller screening of
the Pippard type, which is similar in form to that of Landau damping Žsee, e.g., Refs.
w23,24x.. Thus, roughly speaking, the Meissner effect alters the coefficient of the Landau
damping term. This does not affect our RG calculation at leading order: the dependence
of the Landau pole on g is unchanged. However, the overall numerical prefactor Žwhich
is not fixed by the RG analysis. is affected.
4. RG equations
To obtain the RG equations we need to evaluate the scale-dependent quantum

corrections to the form-factor GŽ k y q .. Let us consider the effect of integrating out
quark states in the momentum shell LX - < q < y m - L, and gluon states in the momen-
tum shell LX - < q < - L. The tree level contribution comes directly from one-gluon
exchange, Fig. 1, while the one loop contribution comes from the box diagram, Fig. 3. It
is important to note that only one topology of the box diagram contributes. The diagram
with ‘‘crossed’’ gluon lines does not have the same kinematic structure as an iterated
Cooper pairing interaction, and is subleading. Actually, in our effective theory the box
diagram contains several different contributions. The most important contains two local
four fermion interactions, and is actually a ‘‘bubble’’ diagram, with running form factor
coefficients. These coefficients contain the effects of previously integrated shells of
quarks and gluons. The other contributions involve at least one exchange of a soft gluon
within the momentum shell, and are suppressed in the thin-shell limit. Thus, as we
discuss below, the result of the one loop part of our calculation is essentially the same as
iterating bubbles with form factor vertices.
Solving the RG equations is equivalent to summing up an infinite series of ladder
graphs corresponding to nearly colinear scattering mediated by gluon exchange. This
corresponds to the ‘‘rainbow’’ approximation in which the gap equation is solved in
Section 5.
Fig. 3. Box diagram.

Let us elaborate on why the tree and box graphs are sufficient to compute the leading
order solution to the RG equations. Consider integrating out all modes between an upper
cutoff ; m and a lower cutoff given by D. For simplicity, we can consider doing this in
a single step, rather than shell by shell. A systematic expansion for the result is possible
in powers of the gauge coupling g and powers of t ; ln L. Since the RG evolution
terminates at a Landau pole of order ; expŽycrg ., t ; 1rg in our counting scheme.
As we shall see below, Fig. 1 is of order g 2 t, and so contributes a term of order g 2 to
the RG equation. The box diagram ŽFig. 3. is of order g 4 t 3 and so contributes terms of
order g 4 t 2 ; g 2 . Higher order loop corrections to either of these graphs Že.g. from
vertex or wavefunction renormalization. are suppressed by at least g 2 t ; g. Of course,
as discussed in the previous section, the screened gluon propagators must be used in our
computations, since some of the logarithms which arise are due to small angle
scattering. Additional radiative corrections beyond vacuum polarization effects on the
gluon propagators are suppressed by at least a power of g 2 t.
Because the rotational SOŽ4. symmetry is broken to SOŽ3., the coupling in Ž3. will
split into its ‘‘temporal’’ G4 and ‘‘spatial’’ Gi parts, which we refer to as A and B
components, according to the notation previously used in Refs. w8,9x
A Ž k y q . cq Ž k 0 ,k . g4 PL cq Ž q0 ,q . cq Ž k 0 ,y k . g 4 PL cq Ž q0 ,y q .
q B Ž k y q . cq Ž k 0 ,k . g i PL cq Ž q0 ,q . cq Ž k 0 ,y k . g i PL cq Ž q0 ,y q . . Ž 14 .
Due to the form of the screening effects described in the previous section it is easy to
see that the RG equations for the A-type couplings will not contain terms of order g 2 .
The Debye mass m D in the longitudinal part of the gluon propagator Ž7. removes any IR
logarithm from the one gluon exchange diagram.
In contrast, the B coupling may receive such corrections. The leading contribution
will come from the region of shell integration where one may neglect the Landau
damping term GŽ Q . in gluon propagator Ž7.. In this region the scattering angle satisfies
the following condition: 1 y x 2 r 3 - z s cos u - 1 y x X 2 r 3 , with x s
Žpr2 5r2 .Ž m2D Lrm3 .. The angular momentum l-component Bl receives the contribution
g2 2r3 dz
Bl Ž L . s y 13 23
4m 2 Hy11yx 1yz
Pl Ž z . Ž 15 .
Žthe factor of 13 corresponds to the attractive 3 channel, while the factor of 23 arises from
the transversality of the magnetic gluon propagator.. The correction is of the form
d Bl ; Ž g 2rm2 . d t, proportional to the RG scaling parameter d t s lnŽ LrLX .. Note that
the leading term in d Bl does not depend on the details of the angular cut-offs – that
ambiguity is cancelled in the definition of t, thus confirming our expectations that
Meissner effect will not affect the dependence of the Landau pole on g.
2
For notational simplicity, we suppress the chiral projectors PL in our expressions, although they appear in
our calculations.
The box diagram of Fig. 3 has the following integral representation2 :
d4 p
I Ž t . s yg 4 H Ž 2p . 4
Kmr , nk g m Ž i m q p4 . g 4 q p P g g n
m g r Ž i m y p4 . g 4 y p P g g k , Ž 16 .
where Kmr , nk incorporates both fermion and propagators Ž7.:
Dmr Ž p4 , k y p . Dnk Ž p4 , p y q .
Kmr , nk s 2 2
. Ž 17 .
Ž i m q p4 . q p Ž i m y p4 . q p 2 2
In order to evaluate the contribution to I Ž t . one should apply the familiar decomposition
of momenta near FS: p s m p q pn , k s m k , q s m q with normal component being
integrated in the limits LX 2 - pn2 - L 2 w8,9x. Note that due to screening effects the
dominant regions of pn and p4 are around the origin, close to the FS.
Consider, for example, the part of I Ž t . with both gluon lines being transverse. All
other cases are done analogously. Capturing only the leading O Žln3L. behavior one gets
g4 dp4 dpn
IŽ t. T , 2 HH Ž 2p . Hd V 4 p PiTj PkTl
16 m
g ig 4g k m g jg 4g l q g i Ž prm P g . g k m g j Ž prm P g . g l
= , Ž 18 .
Ž p42 q pn2 . Ž 1 y z1 . Ž 1 y z 2 .
where z 1 is the cosine of the angle between m k and m p , and z 2 is the cosine of the
angle between m q and m p . Then, use the definition Ž15. of Bl in the expansion of
transverse gluon propagators
1 18 m2
1yz
sy
g2
Ý Ž 2 l q 1. Bl Pl Ž z . Ž 19 .
l
on the interval 3 1 - z - 1 y x 2r3. Applying orthogonality conditions 4 :

4p
Hd V P Ž z . P Ž z
l 1 l
X
2 . s d l lX
2lq1
Pl Ž z . ,
di j 4p
Hd V pˆ pˆ P Ž z . P Ž z
i j l 1 l
X
2 . s d l lX
3 2lq1
Pl Ž z . , Ž 20 .
one obtains the final answer for the I Ž t .T in the leading divergence approximation
m2
IŽ t. T ,y
4p 2
Ž 5g4 m g 4 q 133 g i m g i . Bl2 Ž L . t . Ž 21 .
3
The longitudinal gluon propagators may be expanded in Legendre polynomials Pl Ž z . on the interval
y1- z -1.
4
The averaging over coordinate systems is assumed in the second orthogonality condition of Ž20..
The final answer for the ŽMinkowskian. RGE equations is

dA l N
dt
sy
2
Ž A2l y 2 A l Bl q 5Bl2 . ,
dBl N g2
dt
s
6
Ž A2l y 10 A l Bl q 13 Bl2 . q 27m2 , Ž 22 .
where N s m2r2 p 2 . These RG equations are the same as obtained previously for the
case of local interactions w8,9x, except for the constant term in the Bl sector. A simple
way to understand this is as follows: by expanding the four quark form factor in terms of
orthogonal local operators G l , we reduce the problem to one in which the gluon
exchanges are replaced by local interactions. Hence the analysis of w8,9x should apply
and the same RG equations obtained for each value of l. This is true as well for LR
helicity scattering, which we do not consider explicitly here. Note that though the A l
couplings originate in the Debye screened sector, it would be a mistake to discard them
in the RGE equations due to the fact that there is mixing with the Bl , which diverge near
the FS. Diagonalization of Ž22. gives
d 2
g2
Ž A l y 3 Bl . s yN Ž A l y 3 Bl . y ,
dt 9m2
d N 2
g2
Ž A l q Bl . s y Ž A l q Bl . q . Ž 23 .
dt 3 27m2
It is straightforward to solve these equations. For convenience, we define the spin 0

combination Sl ' A l y 3 Bl and the spin 1 combination Tl ' A l q Bl . We find that Sl
reaches a Landau pole at the scale
3p 2
D ; m gy5exp y
ž '2 g / , Ž 24 .
which agrees with the result Ž1., despite differences between the RG equations of w11x
and ours Žsee below.. Tl , due to the opposite sign of the g 2 term, does not diverge and
reaches the asymptotic value of Tl Ž t ™ `. s '2 p gr3 m 2 .
In order to compare our results with those of w11x, it is necessary to convert four
quark operators into scattering amplitudes. There are additional angular dependences
introduced by the spin angular momentum of the quarks. Let us classify amplitudes by
their total angular momentum, which is the sum of the spin and orbital components.
Thus, for example, the j s 0 channel receives contributions from both the l s 0, s s 0
and l s 1, s s 1 operators. First we note that the spinor part of operators of type A
introduce an additional factor of 2cos 2 Ž 12 u . to the scattering amplitude, while the type B
operators introduce a factor of 2cos 2 Ž 21 u . q 4sin2 Ž 21 u .. This leads to the following
expression for the amplitude:
f Ž u . s Ý Ž 2 l q 1 . w Sl q Tl cos u x Pl Ž cos u . . Ž 25 .
l
The components of total angular momentum j are given by

j jq1
f j s Sj q Tjy1 q Tjq1 , Ž 26 .
2 jq1 2 jq1
where we have used the identity

Ž 2 l q 1 . zPl Ž z . s Ž l q 1 . Plq1 Ž z . q lPly1 Ž z . . Ž 27.
The RG equations for f j can be deduced easily from Ž26. and Ž23.. For the lowest
component, we have
d 2 g2 N
f0 s y 2
y NS02 y T12 . Ž 28 .
dt 27m 3
The equations for higher components are of the form
d 2 g2 N j jq1
dt
fj s y
27m 2
y NS j2 y
3 ž 2 jq1
2
Tjy1 q
2 jq1
2
Tjq1
/. Ž 29 .
Several remarks are in order.
Ø These results are similar to, but different from, those of w11x. In particular, we do not
find that the RG equations take the simple form w11x
d g2
fj s y y Nf j2 . Ž 30 .
dt 9m2
The rhs of our Eq. Ž29. cannot be organized in terms of any simple amplitude f j2 . We
believe that the treatment of quantum corrections in Section 2 of that paper is too
crude. In particular, iterating the amplitude f Ž p,k . in order to obtain the f j2 terms in
the RG equation neglects some important spinor structure of the vertex which our
calculation takes into account.
Ø Fermi statistics constrain the operators as follows. We restrict ourselves to the color 3
and isospin singlet channels, which are both antisymmetric. The remaining part of the
wavefunction must be antisymmetric. In the antisymmetric s s 0 channel, we must
therefore have l s 0,2,4, . . . , while in the symmetric s s 1 channel we must have
l s 1,3,5, . . . . Thus the operators Sl vanish for odd l and Tl vanish for even l. From
Ž26. we see that f j vanishes for all odd values of j. Note that this analysis is
modified in the LR channels, which we do not consider here.
Ø It is not necessary to solve the f j RG equations, as their behavior can be deduced
from that of Sl and Tl . Since Tl never diverges, near the FS f j f S j and diverges at
the scale Ž24.. Note that the 13 NTj2 term cannot be neglected in Ž29.; its asymptotic
value is g 2r27m 2 .
Ø Because the Landau poles in all the j channels are the same, we naively expect to
find condensates with non-zero angular momentum, leading to the breaking of
rotational symmetry w11x. However, we will see in the next section by studying the
gap equation that this is not the case.
5. Rotational symmetry breaking and gap equation
In this section we explore the issue of condensates with non-zero angular momentum
using the gap equation. Our RG results suggest that the gap function could violate
rotational invariance. Some recent papers have studied the s-wave condensate using the
non-local gap equation, including the effects of magnetic gluons w23–25x. We will
generalize their approach to consider gap functions DŽ q0 ,q . which are functions of
orientation. Remarkably, it is possible to show that in the leading order approximation
only an s-wave condensate is formed.
Let us introduce a two component field C s Ž c , c T .. The inverse quark propagator
takes the form
qu q mu y m D
Sy1 Ž q . s
ž D Ž qu q mu q m .
T
/ , Ž 31 .
where D s g 0 D †g 0 . The gap is a matrix in color, isospin, and Dirac space, and in our
analysis we allow it to depend on orientation. As discussed, the RG analysis shows that
the condensate will form in the LL ŽRR. channels. Given this, the form of the gap
matrix is w1–4,12x
ab
Diajb Ž q . s Ž l2 . Ž t 2 . i j Cg 5 Ž Dq Ž q0 ,q . 12 Ž 1 q a P qˆ . q Dy Ž q0 ,q . 12 Ž 1 y a P qˆ . . .
Ž 32 .
In our present weak coupling discussion, we are only interested in Dq. Dy corresponds
to a condensate of anti-quarks, and does not influence Dq. Henceforth we shall only
refer to D ' Dq. Making the usual FS approximations, the gap equation has the form
d4 k D Ž q0 y k 0 ,q y k . D Ž k 0 ,k ,kˆ .
D Ž q0 ,q,qˆ . s yig 2 H Ž 2p . 4 2
, Ž 33 .
e Ž k . q k 02 q < D Ž k 0 ,k ,kˆ . < 2
where the interaction kernel DŽ q0 y k 0 ,q y k . is essentially the gluon propagator up to
some additional factors arising from the particle projector and gamma matrices. Here
k s < k <, q s < q < and e Ž k . s k y m ; we assume q02 ; q02 < m2 and k ; q , m. For the
next step in our analysis it is useful to separate the interaction kernel into angular
momentum channels,
D Ž q0 y k 0 ,q y k . s Ý Ž 2 l q 1 . Pl Ž cos u k q . D l Ž q0 y k 0 , < q y k < . . Ž 34 .
l
The D l coefficients are obtained by integrating the kernel against Legendre polynomials
Pl Žcos u .. Due to the divergence at small angle, all of the D l have the same value at
leading logarithmic order.
The gap equation can now be rewritten as
D Ž q0 ,q,qˆ . s y4p ig 2 Ý Yml Ž uq , f q .

lm
4
d k D l Ž q0 y k 0 , < q y k < . D Ž k 0 ,k ,kˆ .
= H Ž 2p . Yml ) Ž u k , f k . . Ž 35 .
4
e 2 Ž k . q k 02 q < D Ž k 0 ,k ,kˆ . < 2
For D l s DŽ q0 y k 0 , < q y k <. independent of l, the sum over spherical harmonics reduces
to a delta function: d Ž f q y f k . d Žcos uq y cos u k .. The angular integral on the rhs of the
equation can be trivially evaluated, yielding
dk 4 dk k 2 D Ž q0 y k 0 , < q y k < . D Ž k 0 ,k ,qˆ .

D Ž q0 ,q,qˆ . s yig 2 H 3
. Ž 36 .
Ž 2p . e 2 Ž k . q k 02 q < D Ž k 0 ,k ,qˆ . < 2
ˆ
This equation can be regarded as a set of identical equations, one for each orientation q.
Since each equation is identical, the solution must be independent of orientation.
This result could have been guessed directly from the fact that colinear scattering
dominates the magnetic effects. In that approximation the kernel D is proportional to a
delta function d Ž kˆ y qˆ ., and it is clear that Ž33. only has an s-wave solution.
Our analysis thus far has been within the leading logarithmic approximation. We can
relax this condition by considering a gap function
D Ž q . s D 0 q D1 Ž q0 ,q . , Ž 37 .
where D 0 is the constant Žin orientation. solution obtained from Ž33., and D1 is a small
perturbation which can depend on orientation. We will show that D1 is at most of order
expŽycrg 2 . and hence negligible relative to the D 0 . Substituting Ž37. into Ž33., we
obtain two gap equations. The leading order equation is just the usual one for an s-wave
condensate, and determines D 0 . Note that we retain the complete interaction kernel
DŽ q0 y k 0 ,q y k . here, without making the leading log approximation. The second
equation contains terms of O Ž D1 .:
dk 0 d V k q D Ž q0 y k 0 , u k q . < D0 < 2
D1 Ž q0 ,q . , g 2m2 H 3
1y 2
D1 Ž k 0 , k . .
Ž 2p . (
2 k 02 q < D 0 <
2 k 02 q < D 0 <
Ž 38 .
This equation was obtained after first performing the integral over k, in the approxima-
tion that D and D are slowly varying for k , m. The term in brackets in Ž38. suppresses
the logarithmic divergence in the integral over k 0 near the FS, although there is still a
potential divergence from the small angle behavior of DŽ q0 y k 0 , u k q .. Hence the
solution D1 is at most of order expŽy1rg 2 ., and is negligible relative to D 0 in the
weak coupling limit.
The discussion in terms of D 0 and D1 is quite general. We can also apply it to lower
densities, where the magnetic interaction is presumably screened by non-perturbative
effects, and quark interactions are effectively local. In this case the largest condensate
can be shown to be rotationally invariant w1–4,8,9x. Because D in Ž38. is non-singular, a
non-zero solution D1 only exists above some critical coupling Žif at all.. This is unlike
the usual Cooper pairing instability in which an arbitrarily weak interaction can lead to a
condensate. It seems that a large s-wave component tends to inhibit condensates of
higher angular momentum.
Having made some general observations about rotational invariance, we now concen-
trate on solving the gap equation explicitly, in order to check our result for the Landau
pole Ž24.. The authors of w23–25x do not consider higher orbital angular momentum
components Dl ) 0 in their analyses, but it is straightforward to do so.
The explicit gap equation for D1 ' D, neglecting the suppressed antiparticle contribu-
tion, is w24,25x
2 ig 2 d4 k DŽ k0 , k . 3
2 y 12 kˆ P qˆ
D Ž q0 ,q . s y H Ž 2p . 4
. Ž 39 .
3 e 2 Ž k . q k 02 q < D Ž k 0 , k . < 2 Ž k y q . 2 q G
Here we have written explicitly the factors resulting from the particle projector and
gamma matrices. G is the Landau damping term which appears in the magnetic gluon
propagator; we neglect the effect of the electric gluon.
In order to consider higher angular modes of D we make the expansion:
D Ž q0 ,q . s Ý Ž 2 l q 1 . Pl Ž qˆ . Dl , Ž 40 .
l
as well as a similar expansion of the gluon propagator with coefficients

1 1 Pl Ž z .
Dl s 2 Hy1dz 1 y z q Gr Ž2 m . , 2 Ž 41 .
2m
which are all of the same size at leading order. Using the identity
2p msl
)
Pl Ž cos u k q . s Ý Yl m Ž u k , f k . Yl m Ž uq , f q . , Ž 42 .
2lq1 msyl
we obtain the following coupled gap equations:
2 ig 2 dk 0 dk m2
Dl s y H Ž 2p . 2
3p e 2 Ž k . q k 02 q < D Ž k 0 , k . < 2
3
1
= 2 D l Dl y Ž Ž l q 1. Dlq1 Dl q lDly1 Dl . . Ž 43 .
2 Ž 2 l q 1.
In obtaining this equation we have neglected the angular dependence of < DŽ k 0 , k .< 2
which occurs in the denominator. This is justified if the gap turns out to be rotationally
invariant, as expected from our previous arguments. It is easy to see that a self-con-
sistent solution exists with all l ) 0 gaps zero, and the solution for D s D0 given by
Ž24..
6. Discussion
In this paper we investigated the renormalization group behavior of QCD at high

density, concentrating on the effects of long range magnetic interactions. Our approach
was somewhat different from that of w11x in that we focused on individual local
operators rather than scattering amplitudes. The resulting RG equations are different,
although the location of the Landau pole is still given by Ž1.. The disagreement results
from two causes: ŽI. the transverse form of the propagator does not appear to have been
used in Refs. w11x, leading to a different coefficient in the constant g 2 term of the RGE
and ŽII. the treatment of spinor properties of the quantum corrections is different in the
two calculations. We believe that the renormalization group applied directly to ampli-
tudes does not properly compute the loop corrections. Some additional issues we
attempted to clarify include the validity of the use of Landau damping in resummed
gluon propagators, the gauge invariance of the computation and the size of subleading
corrections. Our RG equations Ž23. are gauge invariant, and represent the leading order
result in a self-consistent expansion. Corrections to the coefficients in Ž23. are of order
O Ž g . in the weak coupling limit.
We also used the gap equation to investigate whether rotational symmetry is broken
at asymptotically high densities. The gap equation analysis shows that scattering which
is predominantly colinear leads to a rotationally invariant condensate. We found, in
disagreement with a naive interpretation of the RG results, that any condensates of
higher angular momentum are exponentially smaller than the s-wave condensate. We
also checked that the solution of the gap equation agrees with the value of our Landau
pole.
Acknowledgements
The authors would like to thank Nick Evans, Deog-Ki Hong, James Hormuzdiar, Rob
Pisarski and Dirk Rischke for useful discussions and comments. This work was
supported in part under DOE contracts DE-FG02-91ER40676 and DE-FG06-85ER40224.
References
w1x D. Bailin, A. Love, Nucl. Phys. B 190 Ž1981. 175.

w4x D. Bailin, A. Love, Phys. Rep. 107 Ž1984. 325.
w5x M. Alford, K. Rajagopal, F. Wilczek, Phys. Lett. B 422 Ž1998. 247.
w6x R. Rapp, T. Schafer, E.V. Shuryak, M. Velkovsky, Phys. Rev. Lett. 81 Ž1998. 53.
w7x R. Rapp, T. Schafer, E.V. Shuryak, M. Velkovsky, hep-phrA9904353.
w8x N. Evans, S.D.H. Hsu, M. Schwetz, Nucl. Phys. B 551 Ž1999. 275.
w9x N. Evans, S.D.H. Hsu, M. Schwetz, Phys. Lett. B 449 Ž1999. 281.
w10x T. Schafer, F. Wilczek, Phys. Lett. B 450 Ž1999. 325, hep-phr9810509.
w11x D.T. Son, Phys. Rev. D 54 Ž1999. 094019.
w12x R.D. Pisarski, D.H. Rischke, nucl-thr9811104.
w13x D.K. Hong, hep-phr9812510.
w14x D.K. Hong, hep-phr9905523.
w15x D.K. Hong, M. Rho, I. Zahed, hep-phr9906551.
w16x M. Alford, K. Rajagopal, F. Wilczek, Nucl. Phys. B 537 Ž1999. 443.
w17x J. Berges, K. Rajagopal, Nucl. Phys. B 538 Ž1999. 215.
w18x S. Hands, J.B. Kogut, M.-P. Lombardo, S.E. Morrison, hep-latr9902034.
w19x S. Hands, S.E. Morrison, hep-latr9902012.
w20x S. Hands, S.E. Morrison, hep-latr990521.
w21x J.B. Kogut, M.A. Stephanov, D. Toublan, hep-phr9906346.
w22x T. Schafer, F. Wilczek, Phys. Rev. Lett. 82 Ž1999. 3956, hep-phr9903503.
w23x D.K. Hong, V.A. Miransky, I.A. Shovkovy, L.C.R. Wijewardhana, hep-phr9906478, version 3.
w24x T. Schafer, F. Wilczek, hep-phr9906512.
w25x R.D. Pisarski, D.H. Rischke, nucl-thr9907041.
w26x R. Shankar, Physica A 177 Ž1991. 530.
w27x R. Shankar, Rev. Mod. Phys. 66 Ž1993. 129.
w28x M. Le Bellac, Thermal Field Theory ŽCambridge Univ. Press, Cambridge, 1996..
Toward equilibration in the early stages after a high energy

heavy ion collision
A.H. Mueller 1
Physics Department, Columbia UniÕersity, New York, NY 10027, USA
Received 15 June 1999; received in revised form 2 August 1999; accepted 17 August 1999
Abstract
The early stages in the evolution of the gluons produced in the central region of a head-on
high-energy heavy ion collision are studied. An equation is given for the rate of change of
transverse momentum into longitudinal momentum where the longitudinal direction is along the
collision axis. We are able to follow the system up to the time where equilibration seems to be
setting in, but we are unable to actually follow the system as it reaches equilibrium. q 2000
1. Introduction
At very small values of x the gluon density in a proton’s light-cone wavefunction

reaches saturation w1–4x, that is for gluons having transverse momentum below the
saturation momentum, Q s , there is a density on the order of Ž Nc2 y 1 . a Nc gluons per
unit of transverse phase space Žtransverse coordinate times transverse momentum space..
Parton saturation in the proton may already have been observed at HERA w5,6x. This
regime of high density gluons is reached at more moderate values of x for a large
nucleus. In a head-on collision of high energy heavy ions saturated gluons, those having
transverse momentum at or below Q s , are freed at very early times after the valence
quarks of the two ions have passed each other, say in the center of mass frame. It is
widely assumed that this gluon system then quickly reaches kinetic equilibrium and
somewhat more slowly reaches equilibrium between gluons and quarks.
There are Monte Carlo calculations which have followed the early stages of the
evolution of a heavy ion collision and which give good evidence for equilibration w7–9x.
There is also an interesting Monte Carlo calculation w10x which followed the time
evolution of the gluon fields which start from the saturated distribution occurring in the
1
This work is supported in part by the US Department of Energy, grant DE-FG02-94ER-40819.
PII: S 0 5 5 0 - 3 2 1 3 Ž 9 9 . 0 0 5 0 2 - 7
228 A.H. Muellerr Nuclear Physics B 572 (2000) 227–240
McLerran–Venugopalan model. Our purpose here is somewhat different. The object is

to give a rough analytic calculation of how gluons evolve in momentum and in space
after being freed in a high energy heavy ion collision. We are able to follow a typical
gluon through its many small angle scatterings up to the point where the interactions
which should actually give equilibration begin to occur. At the time when equilibration
actually begins to set in our technique breaks down, although we are able to say at what
time and at what momentum, relative to the saturation momentum, equilibration starts to
set in.
The starting point of our discussion is the McLerran–Venugopalan w11–13x model for
the light-cone wavefunction of a heavy ion w14x. This model is characterized by one
parameter the saturation momentum, Q s , below which gluon densities reach their
maximum value. We suppose that a finite fraction Žour parameter c in Ž16.. of all gluons
having transverse momentum less than or equal to Q s are freed in a head-on heavy ion
collision w15x, and that those gluons having small rapidity are freed in a time about equal
to 1rQ s . Gluons having small rapidity can then scatter only with other gluons having
small rapidity, because the rate of scattering is slow and higher rapidity gluons quickly
separate from lower rapidity gluons. It is then straightforward to write an equation for
the rate of change of the transverse momentum as zero rapidity gluons scatter. The
equation is given in Ž49.. In arriving at Ž49. we have also had to determine where the
infrared cutoff lies for small angle scattering. We have done this by requiring that the
soft gluon field, coming from the hard gluons we are dealing with, also be limited in
magnitude to be no larger than 1rg. This determines a minimum momentum transfer,
l m , which is not quite the same as has been suggested earlier w16x. The difference
between l m given in Ž44. and the screening length of Ref. w16x, given in Ž45., changes
the constants in Ž49., but not the form of the equation.
What has been said for zero rapidity gluons also applies to gluons having rapidity y,
by simply considering their interactions in a frame boosted by yy. Thus gluons of
rapidity y only interaction with other gluons having rapidity y and Ž49. remains valid
for the rate of change of transverse momentum if j ™ j q ln cosh y.
At early times the rate of change of transverse momentum is much less than that
corresponding to the one-dimensional expansion of a gas of gluons in equilibrium.
However, because of the time dependence of the infrared cutoff the rate of loss of the
square of the transverse momentum of gluons increases so that at a time, for zero
rapidity gluons, on the order of Ž1 Q s. e
'2p c a N
c
, and at a transverse momentum given
2
(
by Ž l H rQ s . ; c a Nc 2p , the expansion of our gluon gas more closely resembles that
of a system in equilibrium. At this same time scatterings which change longitudinal
momentum into transverse momentum, not included in Ž49., begin to become important.
This is also the time at which gluon interactions no longer take place simply among
gluons of the same rapidity so that our whole formalism is breaking down. Finally, this
is the time at which the z-extent of the medium is becoming greater than the mean free
path for gluon interactions which can change a gluon’s momentum by an amount
comparable to its momentum. It seems very likely that this is the time at which
equilibration is setting in, although we are unable to follow the system further because
of the limitations of our present formalism. It is important to note that the time when
equilibration seems to be setting in is parametrically short compared with the radius of
A.H. Muellerr Nuclear Physics B 572 (2000) 227–240 229
the initial colliding ions. We consider this evidence that equilibrium indeed occurs in
very high energy relativistic heavy ion collisions while the system is still undergoing a
one-dimensional expansion.
2. Small-x wavefunctions and parton saturation
In this section we shall review the parton description of the light-cone wavefunction
of a heavy ion. There is a very simple model, the McLerran–Venugopalan w11–13x
model, which very nicely illustrates parton Žgluon. saturation. Indeed, it has recently
been claimed w4x that the main results of the McLerran–Venugopalan model should be
general results in QCD. We shall briefly review the physical basis of this model and
then comment on the general expectation of the reliability of the model.
We begin, following McLerran and Venugopalan, by looking at the distribution of
valence quarks in a high energy ion. The valence quarks are found in a Lorentz-con-
tracted longitudinal disc of size D z s 2 Rm p, where R is the nucleon size and p the
momentum per nucleon of the ion. For purposes of our qualitative discussion we
consider D z s 0 so that one can imagine the valence quarks having a two-dimensional
number density of
n q Ž b . s 6 r'R 2 y b 2 Ž 1.
valence quarks per unit area where r is the normal nuclear density while b is the impact
parameter measured from the center of the nucleus in a direction perpendicular to the
direction of motion of the nucleus.
The valence quark density given in Ž1. is the source of soft gluons corresponding to
¨
the Weizsacker–Williams field of n q w3,17,18x. The color charge at a given impact
parameter comes from a random addition of the color charges of each of the valence
quarks at that impact parameter. A single quark gives a C F lnQ 2rm2 gluons at scale Q 2
p
and per unit rapidity so that one expects the number density of gluons per unit area and
per unit rapidity to be
dxGA Ž x ,Q 2 . a CF
2
s nq Ž b. lnQ 2rm2 Ž 2.
d b p
leading to a gluon distribution in the nucleus of
a CF
xGA Ž x ,Q 2 . s 3 A lnQ 2rm2 s AxG Ž x ,Q 2 . . Ž 3.
p
ŽWe note that for a proton Ž3. is not too bad when 10y1 - x - 10y2 and Q 2 s 5–10
GeV 2 .. The unintegrated gluon distribution
dx GA Ž x , l 2 . a CF
2 2
s nq Ž b. 2 Ž 4.
d bd l H p2 lH
has the interpretation, in light-cone gauge, of the number of gluons per unit rapidity per
unit of transverse phase space Žimpact parameter space times transverse momentum
space.. Of course Ž2. – Ž4. are correct only in the low-density limit. In the high density
regime the distribution of gluons in the wavefunction of the nucleus in the McLerran–
Venugopalan model is w2,14x
dx G Nc2 y 1 d 2 x1 2 Q s 2r4
2 2
s 4 H 2 Ž 1 y ey x . e i lP x , Ž 5.
d bd l H 4p a Nc x
where the saturation momentum, Q s , is given by
8p 2a Nc
Q s2 s 'R 2 y b 2 r xG ž x ,4 x2 .
/ Ž 6.
Nc2 y 1
The large l 2 limit given in Ž4. comes from the logarithmic singularity of xG when
2 2
evaluating Ž5.. In the region l H rQ s2 < 1 one can neglect the x dependence of Q s2 and
write
dx G Nc2 y 1 ` dt 2 2
2 2
s 3 H1 eyt l H r Q s Ž 7.
d bd l H 4p a Nc t
with the scale of xG in Ž6. taken as 4 x2s Q s2 . From Ž7. one finds, when l H
2
rQ s2 < 1,
dx G Nc2 y 1 2
s lnQ s2rl H , Ž 8.
d 2 bd 2 l H 4p 3a Nc
which exhibits saturation, the logarithm naturally present in the momentum space
expression. In Ref. w4x a result essentially identical to Ž8. has been derived using BFKL
dynamics. We expect the essential features of Ž7., and Ž8., to be general though it is
difficult to reliably set the overall constant in Ž8..
Despite our notation it should be emphasized that Ž8. refers to the distribution of
gluons in a light-cone wavefunction. It has no relation to the gluon distribution, or its
derivatives, as determined by an operator product expansion analysis of deep inelastic
lepton–nucleus scattering. The result in Ž8. does, however, give the transverse momen-
tum distribution of gluons produced in a hard scattering off a large nucleus.
Of course, the whole approach discussed here only makes sense if the parameter Q s2
is in the perturbative regime. For RHIC energies Ž6. gives Q s f 1 GeV while for LHC
energies Q s should be 2–3 GeV. Thus, our perturbative approach is certainly marginal
for central RHIC collisions while it should be more applicable for heavy ion collisions at
the LHC. Of course the same phenomenon of saturation also occurs at very small x
values for a proton and, indeed, gluon saturation may have been observed at HERA.
3. Toward equilibration in the central region of a head-on heavy ion collision
In this section we shall discuss, in a semiquantitative manner, the evolution from very
early times up to the time where equilibrating interactions begin to occur for the central
region of a zero impact parameter heavy ion collision. There is good evidence from
Monte Carlo simulations that equilibration does indeed occur in collisions having
kinematics similar to those which will take place at RHIC and at the LHC w7–9x. Our
purpose here is of a more theoretical nature, to try and understand in analytic detail the
early stages of evolution of a dense gluon system and to see that equilibrating
interactions do begin well before the system falls apart.
3.1. Which gluons are freed and when?
At the time of collision of the two ions the gluons in the wavefunctions of the ions
are distributed according to Ž5.. For l 2rQ s2 < 1 the distribution is given by Ž8., a
distribution which is, except for a logarithm, given by phase space. Since there are few
2
gluons in the wavefunctions having l H rQ s2 < 1 we shall simply ignore them com-
2
pletely in determining the initial distribution of gluons. When l H rQ s2 4 1 the distribu-
tion of gluons in the wavefunctions is given by Ž2. and Ž4.. These gluons are additive in
the various nucleons in the nucleus so that one may write the two gluon Žjet. production
cross section as
ds
dy 1 dy 2 d 2 l H d 2 b
dx 1GA Ž x 1 l 2 . dx 2 GA Ž x 2 , l 2 . d ŝ
s d 2 b1 d 2 b 2
H d Ž b1 y b2 y b . , Ž 9.
d 2 b1 d 2 b2 dtˆ
where xGA is given by Ž2. and where the gluon–gluon elastic scattering cross section is
2
d sˆ a Nc 4p 3 ˆˆ
ut ˆˆ
us ˆˆ
st
dtˆ
s ž /Ž p Nc2 y 1 . sˆ 2 ž 3y
sˆ 2
y 2
tˆ
y
uˆ 2 / Ž 10 .
2 2
with tˆs yl H . The main interest of Ž9. and Ž10. for us here is that for l H rQ s2 4 1 the
y4
cross section in Ž9. scales like l H . Thus one gets more particles, and more transverse
energy, by lowering l H . Thus the dominant region for gluon production, and for the
production of transverse energy, in a very high energy heavy ion collision is the region
where the produced gluons have transverse momentum on the order of Q s w15x and
where the gluons come from the interaction of gluons having transverse momenta on the
order of Q s in the wavefunctions of the colliding ions.
Now that it is clear that the main contribution to the initial transverse energy density
in a high energy heavy ion collision comes from gluons having transverse momentum on
the order of Q s , the saturation momentum, we need to ask at what time are these gluons
produced. Suppose we focus on a unit of rapidity centered about Y0 ,y 12 q Y0 - y - 12
q Y0 with quanta in this region being right-movers in the frame of choice. Then define
P̂ by
P̂ s Q s sinhY0 . Ž 11 .
P̂ is the typical z-component of the momentum of the gluons we are considering. The
time over which the gluons in our unit of rapidity are freed is
2 Pˆ
tˆ s . Ž 12 .
Q s2
tˆ is the time over which gluons in differing units of rapidity physically separate with
those we are considering. It is also the time over which our gluons physically separate
from the valence quarks. We may estimate the time of this latter separation by
neglecting the z-extent of the right-moving valence quarks. The z-extent of the gluons in
y 12 q Y0 -- y - 12 q Y0 is
1
Dzs Ž 13 .
P̂
while their longitudinal velocity is
Pˆ Q s2
Õs f1y . Ž 14 .
(Pˆ q Q
2 2
s
2 Pˆ 2
The time which it takes the right-moving valence quarks to separate from the gluons in
our unit of rapidity is
Dz
Dt s . Ž 15 .
1yÕ
Using Ž13. and Ž14. we see that Dt s tˆ .
Thus the gluons on which we are focusing are freed during a time given by Ž12.. We
have described these gluons as being virtual gluons in the light-cone wavefunction
which are ‘‘freed’’ during the collision, and this is indeed the picture we have in mind.
However, we do not preclude that a significant number of these gluons may be created,
in a light-cone gauge description, after the valence quarks of the colliding nuclei passed
each other. In this case it is more proper to describe these gluons as produced by the
collision.
It would be nice to have a reliable calculation of the production cross section for
freeing gluons in a central ion-ion collision. Unfortunately, this has not been achieved,
up to now, even in the McLerran–Venugopalan model. There is an interesting numerical
calculation of some of the features of early time production w10x, but so far it is not clear
how to convert these calculations into an evaluation of the number of gluons produced at
time tˆ . In order to proceed further we simply suppose that all the gluons in the
light-cone wavefunction having l 2 f Q s2 are freed. Thus we take as our initial
distribution of freed gluons
dN dx G Nc2 y 1
s cX p Q s2 s c Q s2 , Ž 16 .
d 2 b dy d2 b d2 l l 2
sQ s2 4p 2a Nc
where c is a constant expected to be of order one. c cannot be calculated without
knowing exactly which gluons are freed during the collision. At early times we can use
free streaming kinematics
z s t tanh y Ž 17 .
to get
dN Ž Nc2 y 1 . Q s2 t
2
sc 2
. Ž 18 .
d b dz 4p a Nc t yz2
2
Later on we shall argue that Ž18. should be a reasonable approximation for all times well
before equilibration takes place.
3.2. The dominant two-body scatterings at early times
Immediately after the gluons in a given rapidity region are freed they begin to move
freely along straight line trajectories until scattering with other freed gluons makes them
deviate from their original direction. We suppose that the main scattering which occurs
is gluon–gluon elastic scattering. Such an assumption should be justified for sufficiently
large Q s . Consider those gluons having zero rapidity in a given frame of reference,
where we suppose the z-direction is the collinear direction of the colliding ions. If
X X
gluons l 1 and l 2 scatter into gluons l 1 and l 2 the kinematics of the scattering can
be represented as
l 1 m s Ž l 10 , l 1 x , l 1 y , l 1 z . s l 1 Ž 1,1,0,0 . , Ž 19 .
l 2 m s l 2 Ž 1,y cos x ,y sin x ,0 . , Ž 20 .
l 1X m s l 1X Ž 1,cos u ,sin u cos w ,sin u sin w . , Ž 21 .
l 2X m s Ž l 1 q l 2 y l 1X . m Ž 22 .
X X
with l 1 determined by Ž l 2 . 2 s 0. The initial gluons lie in the x, y plane and approach
each other at an angle x with the initial gluon l 1 being scattered by an angle u from
X
its original direction into the final gluon l 1. w is an azimuthal angle, but here about the
x-axis. Useful invariants are
2
sˆ s Ž l 1 q l 2 . s 2 l 1 l 2 Ž 1 q cos x . , Ž 23 .
X 2 X
ytˆs y Ž l 1 y l 1 . s 2 l 1 l 1 Ž 1 y cos u . , Ž 24 .
X
while Ž l 1 q l 2 y l 1 . 2 s 0 gives
l 1 l 2 Ž 1 q cos x .
l 1X s . Ž 25 .
l 1 Ž 1 y cos u . q l 2 Ž 1 q cos x cos u q sin x sin u cos w .
We shall be dealing with small angle scattering Ž u small. so that
l 1 l 2 Ž 1 q cos x .
l 1X s . Ž 26 .
u2
l 2 Ž 1 q cos x . q l 2 u sin x cos w q Ž l 1 y l 2 cos x .
2
In what follows we shall be following the change in transverse momentum of a gluon
due to scattering with other gluons in the medium. For the gluon initially labeled by l 1
(
the transverse momentum is l 1 H s l 12x q l 12y s l 1. After scattering
u2
( 2 2
l 1X H s Ž l 1X x . q Ž l 1X y . s l 1X 1 y ž 2
sin2f
/ Ž 27 .
in the small u approximation. Thus

u2
l 1 H yl 1X H s l 1 y l 1X q l 1 sin2f Ž 28 .
2
in the small u approximation. Using Ž26. it is straightforward to arrive at

X
l 1 H yl 1H 1 2 l 2 u 2 sin2x cos 2f
Ž l 1X .
2
s
2 l1 l2 ½ 1 q cos x
q u 2 l 1 q l 2 Ž sin2f y cos x cos 2f . 5
1
P . Ž 29 .
1 q cos x
After averaging over the azimuthal angle f according to
X
D l 1H 2p df D l 1 H 2p d f l 1 H yl 1 H
ž /
Ž l 1X .
2
' H0 2p Ž l 1X .
2
' H0 2p Ž l 1X .
2 Ž 30 .
one finds
D l 1H u2
ž /
Ž l 1X .
2
s
4 l 1 l 2 Ž 1 q cos x .
2 l 1 q 3 l 2 Ž 1 y cos x . . Ž 31 .
3.3. The rate of change of gluon transÕerse momentum at early times
Now that we know how much transverse momentum a gluon loses during an early
time collision, given by Ž29. and Ž31., it is not difficult to give an expression for the rate
of change of the transverse momentum of a zero rapidity gluon due to its initial small
angle scatterings. The result is
d l 1H dN ds
sy Hd 3 2
< z12 < P D l 1 H d V 1X d 2 l 2 , Ž 32 .
dt xd l2 d V 1X
where
d V 1X s sin u du d w , Ž 33 .
(
< z12 < s Ž z1 y z2 . 2 y Ž z1 xz2 . 2 . Ž 34 .
Eq. Ž32. expresses the fact that the rate of change of transverse momentum is given by
the change of transverse momentum in a single scattering event times the rate of
scatterings. The rate of scatterings is given by the cross section times the relative flux,
< z12 <, times the density of scatterers. The density of scatterers is given by Ž18. as
dN Q Ž Q s2 y l 22 . dN Ž Nc2 y 1 .
s sc Q Ž Q s2 y l 22 . , Ž 35 .
d3 x d2 l 2 p Q s2 d 2 b dz 4p 3a Nc t
where we approximate the l 2 spectrum by a step function. < z12 < can be evaluated from
Ž19., Ž20. and Ž34. as
< z12 < s 1 q cos x Ž 36 .
while, for small angle scattering, using Ž10. and Ž24.,
2
X ds a Nc 4p 3 du 2 d w
l1 l 1
dV1 X
X
dV1 s ž / p Nc2 y 1 u 4 2p
. Ž 37 .
The d w in Ž37. and the lack of any w dependence in any of the other factors, beside
X X X
D l 1 H rŽ l 1 . 2 , in Ž32. allow one to replace Ž D l 1 H rŽ l 1 . 2 .d w 2p by D l 1 H rŽ l 1 . 2 as
given by Ž31.. Thus
d l 12H a Nc du 2 Qs
s yc Hu H 2
d l 2 Ž2 l 1 q 3 l 2 . Ž 38 .
dt pt
or
d l 12H 5c a Nc 1 du 2 5 c a Nc 1 du 3
sy Q s2 Hu sy Q s2 Hu , Ž 39 .
dt 2p t 2
m
u2 3 pt 3
m
u3
where the u-integration is cut off at um , which quantity will be the topic of our next
section. Finally, suppressing the subscript 1 on l 1 H , one finds
2
d lH 3c a Nc
sy Q s2 ln1rum2 , Ž 40 .
dt 2p t
a formula accurate in the logarithmic approximation, and where , for convenience of
notation, we have redefined c by letting c ™ 35 c.
3.4. The minimum scattering angle
The cutoff on the u-integration in Ž39. corresponds to a cutoff on the momentum

ˆ given by
transfer squared, t,
ytˆm s l 12H um2 s l m2 Ž 41 .

with l m the minimum momentum transfer allowed. One can view Ž39. as arising from
the scattering of a particular hard gluon with the Žrelatively. soft gluon field due to all
the other hard gluons which have been produced at a similar rapidity. However, this soft
field should also be limited to size 1rg because of saturation. We can impose the
saturation condition by requiring that our hard gluon have only one scattering with
momentum transfer greater than or equal to l m in a distance equal to 1rl m . Thus, l m ,
and hence um is determined by
ytˆ m
d s dN 1
Hy` s 1. Ž 42 .
dtˆ d 3 x l m
Using Ž10., Ž18. and Ž41. one finds
3 a Nc Q s3
um3 s c ž /l 3 Ž 43 .
5 p 1H Ž Qs t .
or
3
3 a Nc Q s3
l ms
5
c ž /
p Qs t
, Ž 44 .
with c as redefined just below Ž40.. This is not quite the same screening mass suggested
in Ref. w16x. If one uses Eq. Ž3. of Ref. w16x with dNA A given from Ž16. above and with
d 2 l dy
all freed gluons taken in the integral of Eq. Ž3. one finds
6 Q s2
m2D s c . Ž 45 .
5 Qs t
When Ž a Nc . 2 Q s t is much bigger than one l m , as given by Ž44., is larger than m D , as
given in Ž45.. For our purposes the exact form of the cutoff in the infrared is not so
important so long as the cutoff decreases no faster than some fractional power of Q s t.
3.5. The time dependence of l H
Now go back to Ž40.. It is convenient to write this equation as

2
1 d lH Ž t. c a Nc Q s l H3 Ž t .
sy ln . Ž 46 .
Q s2 dt p Qs t l m3
Using Ž44.,
2
dŽ l H Ž t . rQ s2 . a Nc 3 l H2 p
dlnQ s t
s yc
p 2
ln
Q s2
q lnQ s t q l n ž /
c a Nc
. Ž 47 .
Defining
3 c a Nc
j s lnQ s t , j 0 s ln ž 5 p / Ž 48 .
one finds
2
d Ž l H rQ s . a Nc
s yc j y j 0 q 3ln Ž l H rQ s . Ž 49 .
dj p
which is our final equation.
So far we have derived Ž49. when t, and hence j , is not too large, and thus in the
regime where l H rQ s is very close to 1. In the next section we shall argue that Ž49. is
valid in a much wider range of times. Thus, we here give a brief discussion of the
properties of the solution to Ž49. even when j corresponds to very large times. We are
interested in finding the solution to Ž49. in the region j ) 0 with q H rQ s s 1 at j s 0.
The dependence of l H rQ s on j is qualitatively clear. For j not too large one may
neglect the ln l H rQ s term on the right-hand side of Ž49. to get
2 c a Nc 2
Ž l H rQ s . f1y Ž j y j 0 . y j 02 . Ž 50 .
2p
l H rQ s continues to decrease with increasing j . When j is large, and hence l H rQ s is
small, the j y dependence of l H rQ s is obtained by setting the right-hand side of Ž49.
equal to zero. This gives
2
2
Ž l H rQ s . ™ ey 3 Ž jy j 0 . . Ž 51 .
jy j o™`
The transition from Ž50. to Ž51. takes place when j is extremely close to j 1 with j 1
given by
2p
j1s ( c a Nc
q j 02 q j 0 Ž 52 .
or, equivalently, when t is near t 1 given by

a Nc 2p
t1 s c
p Qs
exp ( c a Nc
q j 02 . Ž 53 .
Eq. Ž49. represents scatterings which transform transverse momenta into longitudinal
momenta. At early times there is a negligible transfer of longitudinal momenta into
transverse momenta. As we shall see in the next section it is exactly at the time when
longitudinally moving particles have a significant probability of scattering and creating
additional transverse momenta that equilibration is beginning to set in.
There is another way to estimate when equilibration begins to set in. An equilibrated
expanding gas in QCD should obey the ideal gas expansion law
y1 r3
T t
ž / ž /
T0
s
t0
Ž 54 .
with T the Žlarge. temperature of the gas. For an equilibrated system transverse
momentum and temperature are proportional so one should have
2
yd Ž l H rQ s . 2
s 23 Ž l H rQ s . . Ž 55 .
dj
At early times, j y j 0 small, it is clear from Ž49. that the transverse momentum in our
expanding gluon system is not decreasing anywhere near fast enough to be close to
equilibration. However, because of the lowering of l m as t increases the rate of decease
of l H rQ s grows with time and we may estimate the time when equilibration is likely
to begin to happen by equating Ž49. and Ž55.. ŽIn the next section we shall see that this
estimate is the same as that coming from requiring that the mean free path of
longitudinally moving gluons be less than the longitudinal length of the medium.. One
finds
2 a Nc
y 23 Ž l H rQ s . s yc j y j 0 q 3ln l H rQ s . Ž 56 .
p
Taking Ž l H rQ s . 2 on the left-hand side of Ž56. from Ž50. and Žtemporarily. neglecting
ln l H rQ s one gets
a Nc 2
3c a Nc
1yc Ž jyj0. s Ž jyj0. Ž 57 .
2p 2p
giving j s j 2 with j 2 determined by
2p
j2s ( c a Nc
q j 0 y 32 s j 1 y 32 Ž 58 .
and
c a Nc
Ž l H rQ s . j
2
2
s3 ( 2p
. Ž 59 .
One can now go back to Ž56. and check that neglect of the ln l H rQ s term on the
right-hand side of that equation amounts only to a very small correction to j 2 .
Thus, our guess is that equilibration starts to set in at a time j 2 given by Ž48. and
Ž58. and at that time Ž l H rQ s . 2 is given by Ž59.. Also at j 2 , from Ž44.,
ln Ž Q s2rn 2 . y ln Ž l m2rn 2 . s 23 Ž j y j 0 . . Ž 60 .
Using Ž58. one sees that at j s j 2
1 1 1
aŽ Q s2 .
s
aŽ l m
2
.
sO
ž' /a
Ž 61 .
so that running coupling effects should not be important in the analysis we have given.
Finally, we note that
1 c a Nc 2p 3
t2 s
Qs
ej2s
p Qs
exp ( c a Nc
y
2
. Ž 62 .
Using the fact that parametrically Q s2 A Ža Nc Nc2 y 1. R with R the nuclear radius,
equilibration can take place before the one-dimensional expansion changes to a three-di-
mensional expansion.
3.6. From early times to equilibration
Suppose we take as our initial reference frame the center of mass of the central
ion–ion collision with the right-moving valence quarks crossing the left-moving valence
quarks at t s 0. At j s 0 gluons at zero rapidity are freed and begin to interact. At early
times a gluon at zero rapidity can only interact with other gluons having small rapidity
because all other gluons separate quickly in the z-direction from zero rapidity gluons. A
gluon having rapidity y has a longitudinal velocity Õz s tanhy and can interact only
with other gluons having rapidity close to y because it, too, separates longitudinally
from all other gluons. Of course, if two gluons having rapidity y interact we can change
to a coordinate system, boosted by rapidity yy from our original system, where these
gluons have zero rapidity and our calculation given above applies. Thus, our calculation
for changes in transverse momentum, which is boost invariant, is also true for gluons
having rapidity y. In particular, Ž49. remains valid for the rate of change of transverse
momentum of a gluon having y if one simply identifies j as lnQ s t y ln cosh y and j 0
as lnŽ c a Nc . y ln cosh y with Ž49. being valid for j ) 0.
p
To see when the approximation of scattering only between gluons having identical
rapidities breaks down it is convenient to introduce z by
2p
jyj0s ( c a Nc
q j 02 y z . Ž 63 .
Then Ž49. and Ž50. become

2
d Ž l H rQ s . c a Nc 2p
dz
s
p ( c a Nc
y z q 3ln Ž l H rQ s . Ž 64 .
and
c a Nc c a Nc
Ž l H rQ s .
2
s2 ( 2p
zy
2p
z2 Ž 65 .
with
c a Nc 2p
ts
p Qs
exp
ž( c a Nc
yz
/ s t 1 ey z . Ž 66 .
As long as ddz Ž l H rQ s . Ž l H rQ s . is small the approximation of having only gluons of

equal rapidity interact is good. This is perhaps most easily seen by imagining that at
time t a particular gluon begins to cross, making a finite angle with the z-axis, the gluon
medium of thickness D z f 2 t. If our particular gluon is right-moving, then during a
time t it will cross all the left-moving gluons in the medium. If during that time,
corresponding to Dz f 1, the gluon does not suffer enough interactions to change its
momentum by a significant amount, then its important interactions must occur much
later and with gluons having a similar rapidity since only such gluons have the same
longitudinal velocity and so do not separate from our particular gluon which is
essentially moving along a straight line trajectory. Now, using Ž64. and Ž65., one finds
d 1 d 2
Ž l rQ s .
dz H 2
Ž l rQ s .
dz H 1
s 2
s . Ž 67 .
Ž l H rQ s . Ž l H rQ s . 2z
Thus as long as z is large our procedure of calculation should be reasonable. However,

somewhere when z becomes on the order of 1 our whole procedure begins to break
down as scattering between unequal rapidity gluons becomes important. In addition, at
this time, scatterings which transform longitudinal momentum into transverse momen-
tum are also becoming important, and they are not included in our formalism.
Thus when z is of order 1 we expect the interactions important for equilibration to
begin and our approximation to break down. As can be seen by comparing Ž58. and Ž63.
this is essentially the same criterion we found in the last section. Thus, in our simple
approach we are not able to follow gluons when the interactions essential for equilibra-
tion begin. We are, however, able to follow the system, on average, up to the time at
which equilibrating interactions begin to occur. We have a good estimate of that time
which is close to t 1 , given in Ž53., and of the transverse momentum of the gluons, given
by Ž59., at the time when equilibration starts. However, because the density of gluons,
given by Ž18., is of size cŽNc2 y 1 4p 2a Nc. Q s3 ey j 1 when equilibration starts we expect
2
3
that when equilibration is complete and dN dx; l H that Ž l H rQ s . 2 ; ey j 1 . It is
3
encouraging that equilibration starts well before the expansion changes from one-dimen-
sional to three-dimensional.
Acknowledgements
I am grateful to Professor Larry McLerran for emphasizing to me the importance of

the small angle gluon–gluon scatterings in this problem. This observation was crucial
for the discussion given above. I would like to thank Dr. Dominique Schiff and the
Laboratoire de Physique Theorique, where this work was completed, for their hospital-
ity.
References
w1x L.V. Gribov, E.M. Levin, M.G. Ryskin, Phys. Rep. 100 Ž1983. 1.
w2x A.H. Mueller, Nucl. Phys. B 335 Ž1990. 115.
w3x J. Jalilian-Maarian, A. Kovner, L. McLerran, H. Weigert, Phys. Rev. D 55 Ž1997. 5414.
w4x A.H. Mueller, hep-phr9904404.
w5x A. Caldwell at DESY Workshop Ž1997..
w6x H. Abramowicz, A. Caldwell, DESY report DESY 98-192 Ž1998..
w7x K. Kinder-Geiger, Comput. Phys. Commun. 104 Ž1997. 70.
w8x B. Zhang, Comput. Phys. Commun. 109 Ž1998. 193.
w9x B. Zhang, thesis, Columbia University Ž1998..
w10x A. Krasnitz, R. Venugopalan, hep-phr9808332.
w11x L. McLerran, R. Venugopalan, Phys. Rev. D 49 Ž1994. 2233.
w14x Yu. Kovchegov, A.H. Mueller, Nucl. Phys. B 529 Ž1998. 451.
w15x J.-P. Blaizot, A.H. Mueller, Nucl. Phys. B 289 Ž1987. 747.
w16x T. Biro, B. Mueller, X.-N. Wang, Phys. Lett. B 283 Ž1992. 171.
w17x Yu.V. Kovchegov, Phys. Rev. D 541 Ž1996. 5463.
w18x Yu.V. Kovchegov, Phys. Rev. D 44 Ž1997. 5445.
The spectrum of the three-dimensional adjoint Higgs model

and hot SU Ž2. gauge theory
A. Hart a , O. Philipsen b
a
Department of Physics and Astronomy, UniÕersity of Edinburgh, Edinburgh EH9 3JZ, Scotland, UK
b
Theory DiÕision, CERN, CH-1211 GeneÕa 23, Switzerland
Received 26 August 1999; accepted 23 November 1999
Abstract
We compute the mass spectrum of the SUŽ2. adjoint Higgs model in 2 q 1 dimensions at
several points located in the Žmetastable. confinement region of its phase diagram. We find a
dense spectrum consisting of an almost unaltered repetition of the glueball spectrum of the pure
gauge theory, and additional bound states of adjoint scalars. For the parameters chosen, the model
represents the effective finite temperature theory for pure SUŽ2. gauge theory in four dimensions,
obtained after perturbative dimensional reduction. Comparing with the spectrum of screening
masses obtained in recent simulations of four-dimensional pure gauge theory at finite temperature,
for the low-lying states we find quantitative agreement between the full and the effective theory
for temperatures as low as T s 2Tc . This establishes the model under study as the correct effective
theory, and dimensional reduction as a viable tool for the description of thermodynamic proper-
ties. We furthermore compare the perturbative contribution ; gT with the non-perturbative
contributions ; g 2 T and ; g 3T to the Debye mass. The latter turns out to be dominated by the
scale g 2 T, whereas higher order contributions are small corrections. q 2000 Elsevier Science
B.V. All rights reserved.
PACS: 12.38.Gc; 11.10.Wx; 12.38.Mh; 11.10.Kk; 14.80.Cp

Keywords: Dimensional reduction; Quark–gluon plasma; Debye mass
1. Introduction
The thermodynamic properties of non-Abelian field theories at finite temperature are

known to be non-perturbative due to the confining nature of the Matsubara zero-mode
sector. Lattice simulations of the full problem are numerically expensive and especially
difficult if fermions are present. A way to circumvent both problems is to perturbatively
integrate out the non-zero Matsubara and momentum modes of order ; T and higher to
arrive at a purely bosonic, three-dimensional effective theory describing the static
quantities of the original theory w1,2x. Three-dimensional gauge theories are non-per-
PII: S 0 5 5 0 - 3 2 1 3 Ž 9 9 . 0 0 7 4 2 - 7
244 A. Hart, O. Philipsenr Nuclear Physics B 572 (2000) 243–265
turbative, but readily accessible by lattice simulations. This programme has been
successfully carried out in studies of the electroweak phase transition, where the phase
structure and nature of the transition were computed from the dimensionally reduced
theory, both by numerical w3–5x and analytical w6–8x methods. The results have been
found to agree to remarkable accuracy with the answers obtained recently from
four-dimensional simulations w9,10x.
In view of the experimental search for the quark gluon plasma in heavy ion collisions
at RHIC and LHC it is of particular interest to see to what extent this programme can be
carried over to QCD. The reduction step for a SUŽ N . gauge theory with Nf flavours of
fermions has been carried out to two loops in w11x Žsee also Refs. w12,13x.. The effective
theory obtained in this case is a three-dimensional SUŽ N . adjoint Higgs theory, with the
scalars corresponding to the electric gauge potential, A 0 , in the unreduced theory. In the
framework of dimensional reduction Nf only modifies the parameters of the effective,
bosonic theory. To be specific, and to compare with existing four-dimensional work, we
discuss in this study the simplest case of Nf s 0, N s 2.
Simulations of the three-dimensional SUŽ2. adjoint Higgs model revealed that this
theory has a phase diagram with continuously connected Higgs and confinement regions
w11,14x, which are partially separated by a line of first order phase transitions. This is in
contrast to the four-dimensional SUŽ2. Yang–Mills theory at finite temperature, which
displays a second-order phase transition w15x. Hence the deconfinement phase transition
cannot be described by the effective theory. Based on the apparent convergence
properties of the reduction step, however, it has been argued w11x that the effective
theory still yields the correct description of static correlation functions at temperatures
sufficiently above the critical temperature, Tc , and hence valuable information about the
plasma state.
In recent publications w16,17x it was established that dimensional reduction indeed
works in the sense that the spectrum measured in four-dimensional simulations at finite
temperature can be classified by the symmetry group of a purely three-dimensional
theory. This does not yet determine the precise form of the effective theory. Earlier
simulations of the SUŽ2. adjoint Higgs model have concluded it to be a good effective
theory by comparing the static potential w18x and gauge-fixed propagators w19,20x
between the full and the effective theory. The question of the dynamical role ˆ of the A 0
and whether there is a further hierarchy gT < g 2 T in the reduced theory, however, is
still not precisely answered. The purpose of the present paper is to quantitatively test the
agreement of a wide range of gauge-invariant correlation functions between the full and
the effective theory at temperatures not much larger than the critical temperature. We
compute correlation functions of the three-dimensional SUŽ2. adjoint Higgs model on
the lattice and extract the corresponding screening masses for various quantum number
channels. These are then compared with the four-dimensional calculation of finite
temperature SUŽ2. Yang–Mills theory in w17x. We find indeed that dimensional reduc-
tion works remarkably well, even at the quantitative level. Furthermore, we are able to
identify the lightest gauge-invariant state to consist of A 0 , which hence may not be
integrated out.
The continuum action of the SUŽ2. adjoint Higgs model is given by
1 2
S s d3 x
H ½ 2 Tr Ž Fi j Fi j . q Tr Ž Di w Di w . q m23 Tr Ž ww . q l3 Ž Tr Ž ww . . 5, Ž 1.
A. Hart, O. Philipsenr Nuclear Physics B 572 (2000) 243–265 245
where Fi j s E i A j y E j A i q ig w A i , A j x, Di w s E i w q ig 3w A i , w x and Fi j , A i , and w are all

traceless 2 = 2 Hermitian matrices Ž w s w asar2, etc... The physical properties of the
theory are fixed by the two dimensionless ratios
l3 m 23
xs , ys . Ž 2.
g 32 g 34
In the framework of dimensional reduction these parameters are completely determined

by the four-dimensional gauge coupling, g 2 , and the temperature, T. Since g 2 is a
running coupling its value is in turn fixed by a renormalisation scale L MS . The details of
the derivation of x, y to two loops can be found in w11x. Here we merely quote the
results to furnish the connection between the three-dimensional parameters and the
four-dimensional, finite T situation. Choosing the renormalisation scale as in w11x and
expressing L MS through the critical temperature as measured on the lattice w15x,
Tc s 1.23Ž11. L MS, one has
10.7668 0.3636
g 32 s T, xs ,
ln Ž 8.3TrTc . ln Ž 6.6TrTc .
2 1
yŽ x. s 2
q qO Ž x. . Ž 3.
9p x 4p 2
With these equations, specifying TrTc completely fixes the parameters x, y of the
reduced model.
In this paper we study specifically the temperatures T s 2Tc , T s 4Tc and T s 5Tc
Žwith the corresponding values of x and y given in Table 1.. Note, however, that these
parameters lie in the Higgs phase of the model, whereas the connection to four-dimen-
sional physics is only valid in the confinement phase w11,21x. The dimensional reduction
programme would be futile were it not that the confinement phase is metastable in this
region of the phase diagram. Starting with a configuration in the confinement phase, we
may re-thermalise and make many measurements at the appropriate parameters before a
tunnelling transition occurs to the Higgs phase. In practice, the phases are so strongly
separated that no tunnelling occurred in our simulations.
In order to discuss correlation functions of the effective theory, it is most convenient
to ignore the finite T context and to discuss the properties of the 2 q 1-dimensional
Higgs model by itself, as we shall do in the following two sections. This will also enable
us to compare the dynamics of the SUŽ2. adjoint Higgs model in its confining phase
Table 1
The lattice parameters and sizes simulated
T x y b s9 b s12 b s18
2Tc 0.141 0.185 L2PT s 28 3 , 38 3 L2PT s 38 3 –
4Tc 0.111 0.228 – L2PT s 38 3 –
5Tc 0.104 0.242 L2PT s 24 3 , 34 3 L2PT s 34 3 L2PT s 48 3
with that of the 2 q 1-dimensional pure gauge theory and fundamental Higgs model,
whose spectra are known in some detail w22–24x, and thus to learn about some general
features of three-dimensional confinement. In Section 2 we introduce the corresponding
lattice action, the operators considered and some technical details about our numerical
simulation. In Section 3 we present our detailed numerical results for the spectrum of the
model, including checks for finite size effects and a numerical continuum extrapolation.
Then we return in Section 4 to an interpretation and discussion of our results in the
context of finite temperature pure gauge theory, before giving our conclusions.
2. Lattice methods
2.1. Action and parameters
The discretised form of the action Ž1., with leading order lattice spacing corrections
of O Ž a., is
Ssb Ý Ž 1 y 12 Tr Ui j Ž x . . q 2 Ý Tr Ž w Ž x . w Ž x . .
x ,i)j x
ˆ a. U † Ž x . .
y 2 k Ý Tr Ž w Ž x . Um Ž x . w Ž x q m
x,m
2
q l Ý Ž 2Tr Ž w Ž x . w Ž x . . y 1 . . Ž 4.
x
ŽThe lattice scalar field has been rescaled relative to the continuum.. The parameters of
the continuum and lattice theory are up to two loops related by a set of equations
specifying the lines of constant physics in the lattice parameter space b , k , l4 w25x,
4 xk 2
bs , ls ,
ag 32 b
b2 1 2 xk Sb
ys
8 ž k
y3y
b / q
4p
Ž 1 q 54 x .
1
q Ž 20 x y 10 x 2 . Ž ln Ž 32 b . q 0.09 . q 8.7 q 11.6 x , Ž 5.
16p 2
where S s 3.17591. For a given pair of continuum parameters x, y these equations

determine the lattice parameters k , l as a function of lattice spacing, and hence govern
the approach to the continuum limit, b ™ `. It is along such trajectories that we should
observe a scaling of dimensionful quantities such as the masses. Due to superrenormalis-
ability of the theory, the above perturbative relation is exact in the continuum limit.
Based on previous experience w14x we expect it to be accurate for all b ) 6. As a
consequence of the presence of scalar fields in the action, the leading order lattice
spacing corrections to mass ratios should be linear in a.
2.2. Blocking of fields
Since we work in the confinement region of the theory, the physical states are
expected to be bound states, which are spatially extended. In order to improve the
projection of any operator onto such states, it is necessary to smear the operators in the
spatial plane so that their extension becomes comparable to the size of the states. This
can be achieved by using ‘‘blocked’’ field variables instead of the elementary ones in
the expressions for the operators. We use blocked link variables at blocking level n as
proposed in w26x,
1
Ui Ž n. Ž x . s
3 ½ UiŽ ny1 . Ž x . UiŽ ny1 . Ž x q iˆ .
"2
q Ý U jŽ ny1 . Ž x . UiŽ ny1 . Ž x q jˆ . UiŽ ny1 . Ž x q iˆq jˆ .
js"1, j/i
=UjŽ ny1 . † Ž x q 2 iˆ . .
5 Ž 6.
For the scalars, we adapt the procedure employed in w23,24x to the case of adjoint fields,
1 2
w Ž n. Ž x . s
5 ½ w Ž ny1 . Ž x . q Ý
js1
UjŽ ny1. Ž x . w Ž ny1. Ž x q ê . UjŽ ny1.† Ž x .
qUjŽ ny1.† Ž x y ê . w Ž ny1. Ž x y ê . UjŽ ny1. Ž x y ê .

5 . Ž 7.
Note that we will take correlations in the 3ˆ direction and so our blocking always remains
within the Ž1,2.-plane to avoid spoiling the transfer matrix of the theory and the
positivity properties of the correlation functions. In addition, we consider blocked scalar
fields w Ž n, j. Ž x . with non-local contributions from one spatial direction j only:
w Ž n , j. Ž x . s 13 w Ž ny1, j . Ž x . q UjŽ ny1. Ž x . w Ž ny1, j. Ž x q ê . UjŽ ny1.† Ž x . .
qUjŽ ny1.† Ž x y ê . w Ž ny1 , j. Ž x y ê . UjŽ ny1. Ž x y ê . 4 , j s 1,2. Ž 8.
2.3. Basic operators
The operators we use in our simulations are constructed from several basic gauge-in-
variant operator types, which we now list before specifying the quantum numbers. We
have operators involving only scalar fields or products of scalar and gauge field
variables,
R Ž x . s Tr Ž w 2 Ž x . . ,
L i Ž x . s Tr Ž w Ž x . Ui Ž x . w Ž x q ˆı . Ui † Ž x . . ,
Aq ı . w Ž x q 2 ˆı . Ui † Ž x q ˆı . Ui † Ž x .
i j Ž x . s Tr Ž Ui Ž x . Ui Ž x q ˆ
=Uj Ž x . w Ž x q ê . Uj† Ž x . . ,
Ay ı . w Ž x q 2 ˆı . Ui † Ž x q ˆı . Ui † Ž x . Uj Ž x . w Ž x q ê .
i j Ž x . s Tr Ž Ui Ž x . Ui Ž x q ˆ
=Uj† Ž x . w Ž x . . . Ž 9.
Other mixed operators are
B < < Ž x . s Tr Ž w Ž x . Ui j Ž x . . ,
2
Tr Ž w Ž x . Ui j Ž x . .
ž
B H Ž x . s Tr T a Ui j Ž x . y w Ž x .
(Tr Ž w 2
Ž x.. / . Ž 10 .
The naming of the operators becomes apparent when considering the continuum limit, in
which B < < Ž x . ; Tr Ž w Ž x . Fi j Ž x .. gives the projection of the field strength along the
scalar field and B H Ž x . is the projection perpendicular to it.
Further, we have loop operators constructed from link variables only,
Ci1=
j
1
Ž x . s Tr Ž Ui Ž x . Uj Ž x q ˆı . Ui † Ž x q ê . Uj† Ž x . . , i , j s 1,2, i / j, Ž 11.
and in addition to the elementary plaquette C 1= 1 , we also consider squares of size 2 = 2
as well as rectangles of size 1 = 2, 1 = 3, 2 = 3. Hence, we have five versions of
operators consisting of closed loops of gauge fields, viz.
Ci1=
j
1
, Ci2=
j
2
, Ci1=
j
2
, Ci1=
j
3
, Ci2=
j
3
. Ž 12 .
Another pure gauge operator useful to probe the confining properties of the theory is the
Polyakov loop along a spatial direction j,
Ly1
PjŽ L. Ž x . s Tr Ł Uj Ž x q meˆ. , j s 1,2. Ž 13 .
ms0
ˆ
Apart from its relevance for confinement, the Polyakov loop also plays an important role
in the study of finite size effects. Although single torelons arising from correlations of
the Polyakov loop PjŽ L. can neither contribute directly to correlations of operators
C 1= 1 , . . . ,C 2= 3, nor to our mixed operators, they may well couple indirectly on the
grounds that they share the quantum numbers with the infinite volume states. Further-
more, torelon–antitorelon pairs are known to give rise to finite-volume effects in
glueball calculations w27x. Apart from studying correlations of Polyakov loops per se,
q q
we have thus constructed single-torelon and torelon-pair operators in the 0q and 2q
channels and studied their correlations as a safeguard against finite-size effects.
2.4. Quantum numbers
Operators with definite quantum number assignments may be constructed from the
above operator types by taking linear combinations with appropriate transformation
properties, i.e. by applying projection operators for the various irreducible representa-
tions of the symmetry group of the lattice slice upon which the operators are con-
structed.
In a three-dimensional Euclidean gauge theory, the slice is a plane with an SO Ž2.
rotational symmetry and discrete Z 2 parity reflection and charge conjugation symme-
tries. Rotations are generated by RŽ u . under which operators of angular momentum j
gain a phase expwi ju x. Parity in two spatial dimensions reflects operators in one
coordinate axis; P:Ž x, y . ™ Ž x,y y .; reflections in other axes are related to this by
rotations. Charge conjugation acts as complex conjugation on the matrix valued fields,
C:Ui Ž x . ™ Ui ) Ž x ., w Ž x . ™ w ) Ž x .. Physical states and operators are characterised by
their eigenvalues under these operations, the quantum numbers J P C . In contrast to the
fundamental Higgs model, there are no C s y1 states in the SUŽ2. pure gauge theory or
the SUŽ2. adjoint Higgs model. In this model SUŽ2. charge conjugation is equivalent to
a global gauge transformation, and hence any gauge-invariant operator is even under this
operation. We therefore omit the index C from now on, implying that all states have
C s q1.
With an adjoint scalar field present, there is an additional Z 2 symmetry correspond-
ing to reflections w ™ yw , whose eigenvalue we call R. Clearly operators containing
an even number of scalar fields, such as Aq i j , will couple to R s q1 states, and those
with odd, such as Ay i j , to R s y1. We thus classify our operators and the states they
couple to by JRP.
On the lattice the symmetry group is broken from SO Ž2. m Z 2 Ž P . m Z 2 Ž R . to the
point group C Õ4 m Z 2 Ž R ., which restricts the allowed rotations to RŽ un . where un s npr2.
Strictly speaking, we should thus classify our states by the irreducible representations of
C Õ4 rather than by J P. Since we are really interested in continuum physics and, as we
shall see, our data reproduce the continuum symmetries within the statistical errors, we
prefer to keep the continuum notation.
The fact that the two-dimensional parity operator reflects only one spatial direction,
and hence the angular momentum operator changes sign under P, has an important
consequence for the spectrum: in 2 q 1 dimensions all states with J / 0 come in
degenerate pairs of opposite parity Žsee, e.g., Ref. w22x.. This statement of ‘‘parity
doubling’’ is based on the continuum rotation group, whereas a square lattice with a
finite volume only admits rotations in units of pr2. Thus, although we should continue
to observe parity doubling of 1q y
R and 1 R on the lattice, the doubling in the J s 2 sector
is lost, and will only be recovered in the infinite volume limit. Comparing the masses of
the 2q y
R and 2 R states is thus a useful measure of whether our lattice is free from
discretisation and finite volume effects. In this study all simulations were done on
lattices large enough that parity doubling was, within statistical errors, in accordance
with continuum, infinite volume expectations.
An interesting feature of the SUŽ2. fundamental Higgs model is the near replication
of the pure gauge theory glueballs within its spectrum. Whereas one might expect the
excitations of the former to be a mixture of gluonic and scalar degrees of freedom, it is
found instead that within the spectrum there are some essentially purely gluonic states.
Given their remarkable decoupling from the scalar fields, it is then unsurprising that the
masses of these match that of the pure gauge model almost exactly w23,24x. One of our
main interests is to study whether the glueball spectrum survives the addition of adjoint
scalar fields similarly unchanged. From the known spectra of the pure gauge theory w22x
and of the fundamental representation Higgs model w23,24x we expect the spin-0 and
spin-2 channels to contain the lowest glueball masses, so in order to carry out the
necessary mixing analysis, we have constructed large bases of different operator types
for these two channels:
q
0q channel:
R: RŽ x .
L: L1 Ž x . q L 2 Ž x .
4
A: Ý R Ž un . Ž Aq12 Ž x . q PAq12 Ž x . .
ns1
B: BH Ž x .
C: symmetric combinations of C 1=1 ,C 2=2 , C 1=2 , C 1=3 , C 2=3
P: P1Ž L. Ž x . q P2Ž L. Ž x .
Pd : P1Ž L. Ž x . P P2Ž L. Ž x .
2 2
T: Ž P1Ž L. Ž x . . q Ž P2Ž L. Ž x . .
q
2q channel:
R: Ž n.
Ry Ž x . ' 12 Tr wŽ n ,1 . Ž x . wŽ n ,1 . Ž x . y Tr wŽ n ,2 . Ž x . wŽ n ,2 . Ž x . 4, n)1
L: L1 Ž x . y L 2 Ž x .
4
A: Ý e i np R Ž un . Ž Aq12 Ž x . q PAq12 Ž x . .
ns1
C: antisymmetric combinations of C 1= 2 , C 1=3 , C 2=3

P: P1Ž L. Ž x . y P2Ž L. Ž x .
2 2
T: Ž P1Ž L. Ž x . . y Ž P2Ž L. Ž x . .
In these expressions RŽ un . rotates the following brackets by un around x, and P denotes
the parity operation. The labels P, Pd refer to single torelon operators, whereas T
couples to torelon pairs in the respective channels.
In principle, similarly extended bases are possible for the other channels as well. This
requires a lot of memory and computer time, and furthermore the glueballs in the other
channels are anticipated to be rather heavy w22–24x, and a mixing analysis would be
difficult. The dynamics of mixing is moreover expected to be independent of the precise
quantum number channel w23,24x, so it should be sufficient to study the 0q q
q and 2q as
detailed examples. For the other channels we therefore limit ourselves to just one or two
operator types B < < andror A. We only give two more explicit examples; the other
combinations follow straightforwardly.
y
0y channel:
4
B: Ý R Ž un . B < < Ž x .
ns1
4
A: Ý R Ž un . Ž Ay12 Ž x . y PAy12 Ž x . .
ns1
q
1y channel:
4
B: Ý e i np r2 R Ž un . B < < Ž x .
ns1
4
A: Ý e i np r2 R Ž un . Ž Ay12 Ž x . q PAy12 Ž x . .
ns1
For all channels we construct operator versions from elementary as well as blocked
fields at various blocking levels. We then take zero momentum sums by averaging over
a timeslice, f Ž t . s 1rL2 Ý x1, x 2 f Ž x ., where f Ž x . denotes a generic operator at some
blocking level.
2.5. Matrix correlators
In order to compute the excitation spectra of states with given quantum numbers, we
construct matrix correlators by measuring all cross correlations between different types
of operators at several blocking levels. The correlation matrix can be diagonalised
numerically following a variational method. For a given set of N operators f i 4 we find
the linear combination that minimises the energy, corresponding to the lightest state. The
first excitation can be found by applying the same procedure to the subspace f i 4X which
is orthogonal to the ground state. This may be continued to higher states so that we end
up with a set of N eigenstates F i , i s 1, . . . , N 4 given by
N
Fi Ž t . s Ý ai k f k Ž t . . Ž 14 .
ks1
The coefficients a i k quantify the overlap of each individual operator f k used in the
simulation onto a particular, approximate mass eigenstate F i . For a complete basis of
operators this procedure is exact. In practice, the quality of the approximation clearly
depends on the number N of original operators and their projection properties. The
eigenstates already determined are removed from the basis for the higher excitations, so
that the basis for the latter gets smaller and the corresponding higher states are
determined less reliably. A more detailed discussion of the calculation and of checks of
its stability can be found in w23,24x, where the same procedure was applied to the
fundamental Higgs model.
In order to distinguish glueballs and scalar bound states, the most elaborate calcula-
q q q
tion is done in the 0q and 2q channels. For 0q our basis consists of N s 32 operators
q
f k g R, L, A, B,C, P,T, Pd 4 , for 2q we have N s 31 and f k g R, L, A,C, P,T 4 , each
operator considered for at least two blocking levels. This typically enables us to extract
y q q y
the four or five lowest lying states in those channels reliably. In the 0y , 2y , 1y , 1y
channels we work with N s 9 operators f k g A, B4 , and in the remaining channels we
have N s 4 and f k g A4 . Correspondingly, in the smaller bases we can only extract the
ground state and, in some cases, the first excitation reliably. Finally, in order to compute
the string tension we have also diagonalised a separate basis of Polyakov loop operators
at N s 4 different blocking levels.
2.6. Simulation and analysis
The Monte Carlo simulation of the lattice action, Eq. Ž4., is performed using the same
algorithm as in previous work w14x. The update of gauge variables employs a combina-
tion of the standard heatbath and over-relaxation algorithms for SUŽ2. w28,29x, whereas
the scalars are updated by a suitable adaption of the algorithm for fundamental scalars
suggested in w30x and employed in w23,24x. In order to account for the quadratic
dependence of the hopping term on the link variables, the link update for the pure gauge
action has to be supplemented by a Metropolis step.
We define a ‘‘compound’’ sweep to consist of a combination of one heatbath and
several over-relaxation updates of the gauge and scalar fields. In this study we
performed 5 over-relaxation steps for every heatbath update. Measurements are taken
after every such compound sweep. We have gathered between 5000 and 30000
measurements, depending on the lattice sizes which range from 24 3 for the coarsest
lattice to 48 3 on the finest one.
All our mass estimates are obtained from measured correlation functions of operators
F i in the diagonalised bases defined in the preceding subsection. Correlated fits over
some interval w t 1 ,t 2 x are performed to the ansatz for the asymptotic behaviour of the
correlation function on a finite lattice, for large T,
²F i† Ž t . F i Ž 0 . : s A i Ž eya M i t q eya M i Ž Tyt . . , Ž 15 .
where i labels the operator, and T denotes the extent of the lattice in the time direction.
In cases where the effective masses did not show plateaux long enough to perform a
correlated fit, we used the effective masses for our mass estimates.
Statistical errors are estimated using a jackknife procedure for which the individual
measurements have been accumulated in bins of 100–250 sweeps.
3. Numerical results
We have performed simulations at temperatures T s 2Tc , T s 4Tc and T s 5Tc

corresponding to three points x, y4 in the metastable confinement phase of the model, as
detailed in Table 1.
The most elaborate test of finite volume effects as well as an explicit continuum
extrapolation for the lowest states are done for T s 5Tc . Experience of the pure gauge
model w22x suggests that to avoid contamination of the spectrum by light torelonic states
the lattice size should be L 0 24 at b s 9. We thus simulate three different lattice
spacings Žthe third one being half of the first. while approximately maintaining this
physical volume Ži.e. Lrb s 24r9, using the lattice spacing as defined by the bare
coupling.. We find all these couplings to lie in the scaling window of the theory which
permits a continuum extrapolation of the low-lying masses. To verify that finite volume
effects are indeed small we also simulate a second, larger volume at the coarsest lattice
spacing.
At T s 2Tc we simulate two lattice spacings and physical volumes, and at T s 4Tc
just one as we are by now confident that lattice effects are small. In these cases
continuum extrapolation is not possible and we thus use the masses measured at the
smallest lattice spacing as approximations of the continuum result.
In order to perform continuum extrapolations it is necessary to have some definition
of the lattice spacing with which to scale the dimensionless lattice masses. The simplest
measure comes from the bare gauge coupling, a s 4rŽ b g 32 .. Such a formula is liable to
quantum corrections, which in some cases translate as large O Ž a. corrections to the
¨ scaling of the masses w31x. A non-perturbative definition of the lattice spacing
naıve
should in principle avoid this, and in pure gauge theory it is customary to scale all lattice
masses in units of the string tension. We thus discuss this quantity first.
3.1. The string tension
As explained previously, a Polyakov loop in a spatial plane couples to a flux loop

state Žor torelon. that winds around the periodic boundaries of the finite volume. The
exponential fall-off of its correlator is related to the mass of the flux loop, according to
Ý ² PjŽ L. Ž x . PjŽ L.† Ž 0. :, eya M P Ž L.t , aMP Ž L . s a 2sL L. Ž 16 .

x
Such a flux loop state can be easily identified through its energy scaling linearly with
the size of the lattice, as seen in the entries labelled by P in Tables 5, 7 and 9. In
contrast to the fundamental Higgs model, there are no fundamental representation matter
fields in the action to screen the colour flux of static charges leading to a breaking of
flux tubes longer than a screening length w32x. On the contrary, a flux tube between
fundamental charges will persist to infinite separation as in the pure gauge theory, and a
string tension can be defined in precisely the same way by the slope of the static
potential at infinite separation1. Accordingly, periodic flux loops winding through the
boundaries of the lattice will not be screened, and the coefficient sL corresponds to the
string tension on a finite volume. An estimate for the string tension in infinite volume is
then provided by the relation w33x
p
a2s` s a 2sL q . Ž 17 .
6 L2
1
To avoid confusion, we note that this is the string tension in the 2q1-dimensional theory, corresponding
to the spatial string tension in the 3q1-dimensional theory at finite temperature.
Table 2
Polyakov loop masses and string tensions at T s 2Tc
T s 2Tc b s9 b s12
Ls 38 Ls 28 Ls 38
aMP Ž L. 0.849 Ž13. 0.627 Ž6. 0.462 Ž4.
a's ` 0.151 Ž1. 0.152 Ž1. 0.112 Ž1.
2
's r g
` 3 0.339 Ž3. 0.341 Ž2. 0.336 Ž2.
We have diagonalised a basis of Polyakov loop operators and the results are summarised
in Tables 2–4. As we shall see with the glueball masses, the string tension is barely
affected by the presence of the scalar fields. It also shows little dependence on the scalar
parameters x, y.
Expressing the string tension in units of the bare coupling, we may Žlinearly.
extrapolate the results at T s 5Tc to the continuum, where we find s` rg 32 s 0.326Ž7.,
(
which compares to s` rg 32 s 0.3353Ž18. in pure gauge theory w22x. That the slope of
(
this extrapolation is small indicates that the difference between using a tree level
definition of the lattice spacing, and a non-perturbative one is slight, and in most cases
any improvements in using the string tension as opposed to the bare coupling for setting
the scale are outweighed by the additional statistical errors on the former.
For this reason, in the following tables we scale and quote masses in units of the bare
continuum coupling to allow rapid comparisons of measurements at different b .
3.2. The mass spectrum
A detailed account of the mass estimates for the lowest states computed on the
various lattices is given in Tables 5–10,
together with the dominant operator content of a given mass eigenstate for the
channels with the largest operator bases.
Let us discuss in detail only the T s 5Tc parameter set, the procedure being entirely
analogous for the others. As an example for the identification of the operator content of
a mass eigenstate, we display the coefficients a i k Žcf. Eq. Ž4.. of the five lowest mass
q
eigenstates in the 0q channel in Fig. 1 for our largest and smallest lattice spacing. The
operator contributions listed in Table 9 are those which have coefficients a i k larger than
Table 3
Polyakov loop masses and string tensions at T s 4Tc
T s 4Tc b s12, Ls 38
aMP Ž L. 0.476 Ž4.
'
a s` 0.114 Ž1.
's ` r g 32 0.342 Ž2.
Table 4
Polyakov loop masses and string tension at T s 5Tc
T s 5Tc b s9 b s12 b s18
Ls 34 Ls 24 Ls 34 Ls 48
aMP Ž L. 0.80 Ž1. 0.554 Ž5. 0.426 Ž4. 0.257 Ž6.
a's ` 0.155 Ž1. 0.155 Ž1. 0.114 Ž1. 0.075 Ž1.
2
's r g
` 3 0.349 Ž2. 0.349 Ž2. 0.342 Ž3. 0.338 Ž4.
0.5. If the state has weak overlap with all operators of our basis and the dominant
coefficients are smaller than 0.5, we indicate this by parentheses in the table. The
ordering of states F i was obtained during the diagonalisation procedure from the
effective mass of the corresponding correlator on the first timeslice.
As in the confinement phase of the fundamental representation Higgs model w23,24x,
we find here a dense spectrum of bound states also. In w23,24x it was found that the
Yang–Mills glueball spectrum is repeated almost identically in the model with funda-
mental scalar fields, with additional bound states of scalars, and little mixing between
them. An interesting question is whether the situation is similar in the case with adjoint
scalar fields. Fig. 1 Žleft. shows that this is indeed so. The ground state is clearly
dominated by scalar operators, whereas the first as well as the fourth excited state have
almost exclusively gluonic contributions and hence are identified as glueballs. Fig. 1
Žright. is the same for our finest lattice, but now with four blocked versions of the
Table 5
q q
Mass estimates and dominant operator contributions in the 0q and 2q channels at T s 2Tc . The dominant
operator types contributing are denoted with Žwithout. parentheses if a i k - Ž ) .0.5
T s 2Tc b s9 b s12
state L2PT s 38 3 L2PT s 28 3 L2PT s 38 3
0q
q Mr g 32 Ops. Mr g 32 Ops. Mr g 32 Ops.
F1 0.747 Ž7. R, L, A 0.743 Ž7. R, L, A 0.744 Ž9. R, L, A
F2 1.629 Ž27. C, B H 1.411 Ž18. P 1.416 Ž15. P
F3 2.007 Ž45. A,Ž R, L. 1.613 Ž23. C, B H 1.623 Ž21. C, B H
F4 1.913 Ž45. P 1.980 Ž36. A,Ž R, L. 1.998 Ž27. A,Ž R, L.
F5 2.22 Ž6. C, B H 2.36 Ž6. C, B H 2.34 Ž4. C,Ž B H .
F6 2.80 Ž13. A 2.47 Ž9. Pd 2.38 Ž5. Pd
F7 2.94Ž17. C,Ž B H . 2.75 Ž11. A 2.79 Ž7. C, B H
F8 3.11 Ž48. Ž P. 2.86 Ž13. C, B H 2.83 Ž7. Ž A.
2q
q Mr g 32 Ops. Mr g 32 Ops. Mr g 32 Ops.
F1 1.94 Ž5. P 1.440 Ž22. P 1.401 Ž15. P
F2 1.92 Ž5. P 1.454 Ž18. P 1.419 Ž18. P
F3 2.27 Ž6. R, L, A 2.24 Ž5. R, L, A 2.30 Ž4. R, L, A
F4 2.61 Ž11. C 2.34 Ž8. Ž P. 2.37 Ž5. Ž P.
F5 2.60 Ž29. Ž P. 2.70 Ž12. C 2.65 Ž5. C
F6 3.15 Ž19. C 3.05 Ž15. C,T 2.98 Ž8. C,T
Table 6
Masses Mr g 32 for the other channels at T s 2Tc
T s 2Tc b s9 b s12
state Ls 38 Ls 28 Ls 38
0y
q – – 3.24 Ž12.
0y
y 1.229 Ž14. 1.220 Ž14. 1.118 Ž12.
0y)
y 2.14 Ž6. 2.14 Ž6. 2.16 Ž4.
1q
q – – 3.04 Ž12.
1y
q 3.00 Ž5. 3.01 Ž5. 2.93 Ž11.
1q
y 2.02 Ž2. 2.01 Ž2. 1.99 Ž3.
1y
y 1.95 Ž5. 1.87 Ž8. 1.92 Ž5.
2q
y 2.41 Ž9. 2.57 Ž9. 2.52 Ž6.
2y
q 2.35 Ž3. 2.36 Ž6. 2.31 Ž5.
operator B H Žcf. Eq. Ž10.. included in addition to the operators used for the left plot. At
first sight this appears to violate the ‘‘non-mixing’’ between scalar and gluonic states as
observed with the smaller basis. Upon expanding and taking the trace in Eq. Ž10.,
however, one readily sees that the operator is in fact a sum of terms of the form
B H ; Tr ŽU 2 . q B <2< q Tr Ž w 2 . B <2< . Of these, the first term mixes with our other loop
operators and has projection onto the glueball states since ŽTr ŽU 2 . has the same
representation content as Tr ŽU .., whereas the other terms project onto higher excitations
Žthey are of higher dimension. and thus die out asymptotically. Hence, the operator B H
asymptotically behaves as a C-type operator which mainly projects onto glueballs, but
very little onto scalar states, as is corroborated in Fig. 1. This figure further shows that
Table 7
q q
T s 4Tc b s12
state L2PT s 38 3
0q
q Mr g 32 Ops.
F1 0.945 Ž9. R, L, A
F2 1.425 Ž15. P
F3 1.626 Ž18. C, B H
F4 2.151 Ž30. A,Ž R, L.
F5 2.33 Ž4. C, B H
F6 2.33 Ž4. Ž P.
F7 2.43 Ž5. Pd
F8 2.84 Ž7. C
2q
q Mr g 32 Ops.
F1 1.428 Ž21. P
F2 1.452 Ž18. P
F3 2.42 Ž4. R, L, A
F4 2.34 Ž5. Ž P.
F5 2.63 Ž6. C
F6 3.02 Ž8. C,T
Table 8
T s 4Tc b s12
state Ls 38
0y
q 3.48 Ž12.
0y
y 1.323 Ž12.
0y)
y 2.30 Ž4.
1q
q 3.47 Ž24.
1y
q 3.05 Ž11.
1q
y 2.00 Ž3.
1y
y 2.05 Ž4.
2q
y 2.57 Ž5.
2y
q 2.45 Ž5.
the decoupling of the gluonic sector is stable under variations of the lattice spacing by a
factor of two. As we shall see, with the lattice spacings considered we are quite close to
the continuum, and the observed effect should persist in the continuum limit. This
q
picture is repeated in the 2q channel.
The spectrum of higher excitations becomes increasingly dense with increasing
q q
energy. For the 0q and 2q channels, the eigenstates beyond F 6 become increasingly
difficult to identify, and their total overlap with the operators in the basis becomes
smaller. Further, they are closer in energy and within our statistical errors appear often
degenerate. An unambiguous resolution of these states would require still larger bases as
well as more statistics. We have therefore chosen not to quote them as reliable results.
Table 9
q q
2 3 2 3 2 3
state L PT s 34 L PT s 24 L PT s 34 L PT s 48 3
2
0q
q Mr g 32 Ops. Mr g 32 Ops. Mr g 32 Ops. Mr g 32 Ops.
F1 1.010 Ž9. R, L, A 1.010 Ž12. R, L, A 1.005 Ž12. R, L, A 1.008 Ž18. R, L, A
F2 1.61 Ž5. C 1.24 Ž3. P 1.26 Ž3. P 1.152 Ž27. P
F3 1.82 Ž7. P 1.64 Ž3. C, B H 1.63 Ž3. C, B H 1.60 Ž5. C, B H
F4 2.19 Ž14. L, A 2.16 Ž11. L, A 2.19 Ž8. A 1.91 Ž5. Pd
F5 2.29 Ž16. C 2.05 Ž13. Pd 2.33 Ž8. C, B H ,ŽT . 2.19 Ž7. A
F6 2.73 Ž15. Ž P. 2.36 Ž18. C, B H 2.25 Ž8. Pd 2.29 Ž9. C,T
2q
q Mr g 32 Ops. Mr g 32 Ops. Mr g 32 Ops. Mr g 32 Ops.
F1 1.795 Ž32. P 1.215 Ž25. P 1.266 Ž21. P 1.143 Ž27. P
F2 1.798 Ž32. P 1.253 Ž25. P 1.274 Ž24. P 1.161 Ž27. P
F3 2.35 Ž8. R, L, A 2.25 Ž10. Ž P. 2.32 Ž4. Ž P. 2.38 Ž5. Ž P.
F4 2.68 Ž11. C 2.45 Ž15. R, L, A 2.43 Ž5. R, L, A 2.51 Ž7. C,T
F5 2.67 Ž11. Ž P. 2.57 Ž11. C,T 2.61 Ž5. C,ŽT . 2.45 Ž8. L, A
F6 2.87 Ž18. C 2.86 Ž11. C,T 2.78 Ž6. C,T 2.72 Ž6. C
F7 3.34 Ž24. R, L, A 3.17 Ž17. R, L, A 3.23 Ž11. L 3.17 Ž10. A
Table 10
state L2PT s 34 3 L2PT s 24 3 L2PT s 34 3 L2PT s 48 3
0y
q – – 3.39 Ž15. –
0y
y 1.352 Ž27. 1.368 Ž28. 1.342 Ž33. 1.340 Ž30.
0y)
y 2.34 Ž8. 2.28 Ž6. 2.33 Ž8. 2.28 Ž5.
1q
q 3.01 Ž37. 3.65 Ž37. 3.52 Ž37. –
1y
q 3.02 Ž26. 3.16 Ž20. 3.20 Ž20. –
1q
y 2.13 Ž3. 2.04 Ž5. 2.06 Ž6. –
1y
y 2.02 Ž6. 2.09 Ž5. 2.06 Ž5. –
2q
y 2.69 Ž4. 2.60 Ž10. 2.64 Ž9. –
2y
q 2.52 Ž10. 2.60 Ž9. 2.54 Ž10. –
y q y
Another word of caution concerns the states 0q , 1q and 1q . As described in Section
2.5, the bases for these states consist exclusively of four blocked versions of the operator
type A. It is conceivable that these operator sets have very poor projection onto some of
the sought ground states, leading to an overestimate of their masses. Indeed, such an
P
effect was encountered in the 1y sectors, as is apparent from comparing the two
different parity states, which should be degenerate. This problem can only be solved
y
with a larger basis including other operator types. For 2q the basis equally consists of
four operators of type A. In this case, however, we observe parity doubling with the
ground state in the 2q q channel see below , and may thus be confident to have good
Ž .
y
projection on the 2q ground state.
At our largest lattice spacing, b s 9, we have performed an explicit check for finite
volume effects. Comparison of the corresponding columns in Table 9 shows that the
masses for those states that do not couple to winding operators Ž P, Pd ,T . are statistically
compatible on L2 P T s 24 3 and L2 P T s 34 3. It appears that already on the smaller
volume most of our results for the spectrum are free of finite size effects, as was the
case in the pure gauge theory. Of the operator eigenvectors projecting onto winding
operators, those purely of type P or Pd may be discounted as mere finite volume
q
Fig. 1. The coefficients a i k of the operators used in the simulation for the five lowest 0q eigenstates for 28
operators at b s9 left , and for 32 operators at b s18 right .
Ž . Ž .
artefacts, they disappear from the spectrum in the infinite volume limit. More difficult
are those states that couple both to non-winding, C or B H , and to winding operators, T,
q
such as the 2q states F 5 and F 6 whose masses are likely to suffer most from finite
volume effects. At b s 9, where we have two volumes, we can resort to the larger
lattice where the states are decoupled from the torelons, and trust that this number is as
close to the infinite volume limit as the other states. On the finer lattices, however, we
have only one volume, and hence it is likely that the corresponding masses are affected
by the admixture of torelons to these states.
As we have discussed, one important signal of finite volume and discretisation effects
is the loss of parity doubling in the spectrum. On all our lattices we find within
statistical errors a degeneracy of the Jq y
R and JR states not only for J s 1 where it is
Ž
expected even on the lattice. but also for J s 2. In the latter case, the relevant
comparison is between non-torelonic states that do not couple to winding operators Žand
thus remain of finite mass in the infinite volume extrapolation..
We have already commented that in extrapolating to the continuum limit there is a
choice in the method of defining the lattice spacing. There is also some freedom in the
functional form of the extrapolation. From the weak coupling expansion of the lattice
action, the lowest order corrections to the continuum limit mass should be linear in the
lattice spacing. Thus if we are sufficiently close to the continuum limit, a straight line fit
to the data can be attempted.
q q
We have performed linear fits for the low-lying 0q and 2q states, using both tree
level and non-perturbative definitions of the lattice spacing. All fits give acceptable
x 2rd.o.f. - 1. In Fig. 2 we show such fits using units of the bare coupling, and it is
q
such a fit that we use for the ground state in the 2q channel. Given the very large errors
on the gradient for linear fits using the string tension to set the scale, we have
insufficient data to decide if the latter, non-perturbative definition of the lattice spacing
improves the scaling behaviour.
q q q
The results for 0q and 2q are given in Table 11. The 2q pure glueball couples to
torelons on our finest lattice, so that its mass value there could not be used for an
extrapolation. We quoted the value for b s 12 instead. We also compare with the
q
Fig. 2. The approach to the continuum of the four lowest 0q states at T s 5Tc . Filled symbols denote
glueballs, open symbols bound states of scalars.
Table 11
Final mass estimates using the continuum extrapolation in the bare coupling of our data at T s 5Tc . Our data
for the glueball are compared with the results obtained in the pure gauge theory w22x
0q
q channel 2q
q channel
gauge-Higgs pure gauge gauge-Higgs pure gauge

scalar glueball glueball scalar glueball glueball
1.002 Ž35. 2.46 Ž21.
1.57 Ž9. 1.58 Ž1. 2.61 Ž5. 2.62 Ž5.
2.22 Ž18.
2.22 Ž25. 2.29 Ž3.
corresponding glueball states in the pure gauge theory and find their masses to deviate at
the percent level at most from those measured in the Higgs model. We conclude that we
have a situation analogous to the fundamental Higgs model, with the pure gauge glueball
spectrum being repeated, with additional bound states of adjoint scalars and little mixing
between them.
As comparison of Tables 11 and 9 elucidates, the difference between the masses
extrapolated to the continuum and those computed at finite lattice spacing are very
small, of the order of a few percent for the lowest states, and mostly covered by the
statistical errors of the values on the finest lattice. We therefore quote the latter for the
other channels as well as our other parameter sets, noting, however, that for the heavier
states this may well amount to an estimated 5% systematic error.
4. Discussion
With the numerical results for various correlation functions at hand, let us now
discuss their implications for the spectrum of static correlation functions in four-dimen-
sional SUŽ2. Yang–Mills theory at finite temperature. In order to express the masses
entirely in units of temperature, we have to make use of the perturbative expression for
the gauge coupling at that temperature scale, Eq. Ž3.. Using this expression we have
g 32 Ž2Tc . f 3.83T, g 32 Ž2Tc . f 3.07T and g 32 Ž5Tc . f 2.89T. The spectrum in these units is
shown in Fig. 3.
4.1. Comparison with four dimensions
The first important question concerns the comparison with a four-dimensional

simulation, in order to assess the quality of dimensional reduction in the temperature
range under study. A detailed analysis of the low-lying modes for SUŽ2. gauge theory
has recently been presented w16x for the temperature range 2Tc - T - 4Tc . Only the
ground states in each quantum number channel are so far available in four dimensions,
however.
To facilitate the comparison we briefly summarise the situation in four dimensions.
There the lattice symmetry group is D4 h m ZŽ2.R z where D4 h is the point group of the
discretised 2 q 1-dimensional space-time slice used to create screening state operators,
Fig. 3. The spectrum of the lowest inverse static correlation lengths at each temperature. Filled symbols denote
glueballs, open symbols bound states of scalars.
and R z is the Euclidean time reversal operator in the notation of w34x. It acts on the time
component of the gauge field as R z A a0 ™ yA a0 . Dimensional reduction converts the A 0
component of the gauge field into an adjoint scalar field, and the scalar reflection
symmetry is the ‘‘daughter’’ of Euclidean time reflection in the original theory.
As the temperature is increased in the four-dimensional theory, the timelike direction
of the slice becomes squeezed, to the point where the slice becomes quasi-two-dimen-
sional, and the effective symmetry group becomes C Õ4 . At this point the spectrum of
screening masses should conform to a pattern predicted by this latter symmetry group, as
has indeed been observed w16x. If the perturbative matching of finite temperature SUŽ2.
pure gauge theory to the SUŽ2. adjoint Higgs model in three dimensions is correct, the
screening masses should furthermore match those observed in this study. Hence we
expect to see the 3 q 1-dimensional D4 h states Aq y
1 and A 2 converging on the 0
q
mass
we measure the quantum numbers R z , R are suppressed here , and the A1 and Aq
Ž . y
2 on
the 0y. The E states should agree with the J s 1. The B1q and By 2 should agree with
the 2q, and B1y and Bq y
2 should have a common mass with the 2 . For a more detailed
account of this symmetry breaking pattern we refer to w16x.
In w16,17x and the above paragraph discussing the mapping of states, the behaviour
under Euclidean time reflection R z has not been specified. This leaves room for
potential ambiguities in comparing the states. For our comparison, we assume that the
lowest states have been extracted in w16x, without explicit specification of their R z-sym-
metry. Correspondingly, for the same spin and parity we compare with the lighter of our
R-states. Further, because of parity doubling, B1y has to be degenerate with B1q in the
continuum limit. In w17x, B1q still exhibits finite size effects, so that we take B1y instead
q
to compare with our 2q state. The four-dimensional data are taken from Table 4 in Ref.
w16x and Table 2 in Ref. w17x.
The comparison of screening masses after these identifications is shown in Fig. 4.
Comparing the temperature dependence of the screening masses, we find agreement for
the qualitative behaviour of the lowest states in all channels. Their temperature depen-
q
dence is weak, with the 0q slightly rising whereas all other ground states drop with T.
Fig. 4. Comparison of screening masses as calculated in four dimensions directly w16x Žopen symbols. and from
the reduced three-dimensional theory used here Žfull symbols.. In the point group notation of Ref. w16x, the
open squares denote Aq y y
1 , diamonds A1 , circles B1 and triangles E .
q
In our simulation we find in addition that the temperature dependence becomes stronger
for the higher excitations, cf. Fig. 3. But even on a quantitative level, we find a
remarkable agreement for the low-lying states. This could not have been anticipated
given that the temperature is rather close to the phase transition, where we know the
effective theory description has to break down. The only notable disagreement between
the measurement in the full and the reduced theory is for the lightest spin-2 screening
mass. But in comparing the heavier screening masses we have to remind ourselves that
the effective theory is obtained by integrating out the non-zero Matsubara modes, whose
smallest is of the order of ; 2p T. The lowest spin-2 screening mass, on the other hand,
is already larger than ; 6T for the temperatures considered, and hence belongs to an
´
energy region where the effective theory loses its validity. In this regime the non-local
terms and terms of higher dimension which have been discarded in the construction of
the effective theory w11x are no longer suppressed, and disagreement for screening
masses of that order and higher is in fact expected.
We conclude that the three-dimensional SUŽ2. adjoint Higgs model in its metastable
symmetric phase is the correct effective theory to describe the static correlation
functions in a four-dimensional SUŽ2. pure gauge theory at finite temperature. The
effective theory gives the correct qualitative picture, and for the screening masses in the
expected range of validity of the effective theory even quantitative results at tempera-
tures as low as T s 2Tc .
4.2. The Debye mass
Of particular interest for non-Abelian plasma physics is the Debye screening mass,
which may be expanded as
Ng 32 m3
m D s mLO
D q ln q c N g 32 q O Ž g 3 T . . Ž 18 .
4p g 32
Table 12
The different contributions to the Debye mass, Eq. Ž18., where Ž1q2. is the sum of the first two terms
m D r g 32 Ž1q2.r g 32 c2 O Ž g 3 T .r g 32
T s 2Tc 1.12 Ž2. 0.296 1.14Ž4. y0.32Ž6.
T s 4Tc 1.32 Ž2. 0.360 1.14Ž4. y0.18Ž6.
T s 5Tc 1.34 Ž3. 0.379 1.14Ž4. y0.18Ž7.
The leading order result is known perturbatively w35–37x to be

1r2
N Nf
mLO
D s m3 s ž q / gT . Ž 19 .
3 6
At next-to leading order ; g 2 T, one can extract perturbatively a logarithm w38,39x, but
c N is entirely non-perturbative and has to be evaluated by lattice simulations. A gauge
invariant, non-perturbative definition for m D has been given in w34x and is based on
Euclidean time reflection, R z . According to the definition in w34x, the Debye mass m D
corresponds to the mass of the lightest state odd under this transformation, i.e. the
y
lightest R s y1 state. According to our measurements this is the 0y ground state, as
obtained from the operator B < < . As discussed in the previous section, our measurement is
entirely consistent with the four-dimensional one, see Fig. 4.
On the other hand, c 2 has been determined directly in recent simulations w40,41x after
integrating out A 0 following a prescription in w34x. There the value c 2 s 1.14Ž4. was
found for the leading correction to the perturbative result 2 .
Taking these data together, we are now in a position to assess the size of the various
contributions in the expansion Eq. Ž18.. These are summarised in Table 12.
The emerging picture is quite interesting: the Debye mass is entirely non-perturba-
tive, with its leading order value being much smaller than the g 2 T correction up to
temperatures of the order of T ; 10 7 Tc w41x. On the other hand, the O Ž g 3 T . corrections
are only 20% at temperatures as low as T s 2Tc , decreasing rapidly to 10% at T s 5Tc .
We thus have to conclude that the scale dominating the physics of electric screening
is the scale of the three-dimensional effective theory g 32 , or equivalently, the soft modes
; g 2 T of the original finite temperature theory, contrary to the naıve
¨ expectation ; gT.
The contributions on this scale are entirely non-perturbative. Contributions of higher
powers in g, on the other hand, can be viewed as small perturbations.
5. Conclusions
We have performed a detailed simulation of the spectrum of the SUŽ2. adjoint Higgs
model in 2 q 1 dimensions at various points in its metastable confinement phase. The
2
We remark that the operator B < < has been simulated previously to determine the Debye mass w42x. By
fitting Eq. Ž18. to the temperature dependence of the mass of B < < , these authors find the term O Ž g 3 T . to be
consistent with zero, and upon subtracting the perturbative piece mLO D determine the next-to-leading order
correction to be c2 s1.58Ž20.. We consider this likely to be an overestimate, as it is two standard deviations
above the results of w40,41x. Adding the perturbative contribution and no other corrections, the resulting m D
likewise deviates from the four-dimensional result of w16x and our present measurements.
choice of the parameter values was motivated by the connection with four-dimensional
SUŽ2. pure gauge theory at finite temperature, for which the adjoint Higgs model
emerges as an effective theory for the static modes after perturbative dimensional
reduction. Our results are relevant both for the study of confinement in three-dimen-
sional gauge theories, as well as for the question whether dimensional reduction is
applicable to SUŽ2. pure gauge theory, and eventually to QCD.
Regarding three-dimensional gauge theories, we found a dense spectrum of bound
states, which consists of an almost unchanged replica of the glueball spectrum of pure
gauge theory, to which additional bound states of adjoint scalars are added. The glueball
and scalar states show very little mixing, and correspondingly the string tension is very
close to its pure gauge value. The pure gauge quantities are furthermore insensitive to
variations of the scalar parameters in the action. All these findings are completely
analogous to the situation in the SUŽ2. fundamental Higgs model w23,24x. We conclude
that the approximate decoupling of the pure gauge sector from the scalar sector is a very
robust phenomenon in three dimensions, for which an explanation is lacking at present.
ˆ as effective theory, we find that dimensional reduction works
Regarding its role
remarkably well down to temperatures of 2Tc , where the lowest screening masses
computed in the effective and the full theories agree quantitatively. This agreement is
rather striking for a temperature so close to the transition. These findings confirm the
SUŽ2. adjoint Higgs model as the correct effective high temperature theory to describe
the thermodynamics of four-dimensional SUŽ2. pure gauge theory. In particular, our
investigation has established unambiguously that the A 0 degrees of freedom for
temperatures not too far above the phase transition constitute the lightest state in the
three-dimensional effective theory, and hence may not be integrated out, as already
¨ hierarchy of scales gT < g 2 T
discussed in w11,41x. It furthermore implies that the naıve
does not translate into a corresponding separation of the dynamics of the A 0 and A i in
this temperature range.
We expect no qualitative changes when moving from SUŽ2. to SUŽ3.. An analogous
investigation for SUŽ3. and three flavours of quarks poses no additional complications
and appears desirable in the light of our results.
Acknowledgements
We thank UKQCD for computer time on one of their workstations in Edinburgh

where part of these computations were performed, and M. Laine for a critical reading of
the manuscript. This work was supported in part by United Kingdom PPARC grant
GRrL22744.
References
w1x P. Ginsparg, Nucl. Phys. B 170 Ž1980. 388.

w2x T. Appelquist, R. Pisarski, Phys. Rev. D 23 Ž1981. 2305.
w3x K. Kajantie, M. Laine, K. Rummukainen, M. Shaposhnikov, Phys. Rev. Lett. 77 Ž1996. 2887. hep-
phr9605028.
w4x M. Gurtler,
¨ E.-M. Ilgenfritz, A. Schiller, Phys. Rev. D 56 Ž1997. 3888. hep-latr9704013.
w5x ´ J. Rank, Nucl. Phys. B ŽProc. Suppl.. 53 Ž1997. 623, hep-latr9608087.

F. Karsch, T. Neuhaus, A. Patkos,
w6x ¨
W. Buchmuller, O. Philipsen, Nucl. Phys. B 443 Ž1995. 47.
w7x ¨
W. Buchmuller, O. Philipsen, Phys. Lett. B 397 Ž1997. 112, hep-phr9612286.
w8x N. Tetradis, Nucl. Phys. B 488 Ž1997. 92. hep-phr9608272.
w9x F. Csikor, Z. Fodor, J. Heitger, Phys. Rev. Lett. 82 Ž1999. 21. hep-phr9809291.
w10x M. Laine, J. High Energy Phys. 9906 Ž1999. 020. hep-phr9903513.
w11x K. Kajantie, M. Laine, K. Rummukainen, M. Shaposhnikov, Nucl. Phys. B 503 Ž1997. 357. hep-
phr9704416.
w12x E. Braaten, A. Nieto, Phys. Rev. Lett. 76 Ž1996. 1417.
w13x E. Braaten, A. Nieto, Phys. Rev. D 53 Ž1996. 3421. hep-phr9510408.
w14x A. Hart, O. Philipsen, M. Teper, J. Stack, Phys. Lett. B 396 Ž1997. 217. hep-latr9612021.
w15x J. Fingberg, U. Heller, F. Karsch, Nucl. Phys. B 392 Ž1993. 493. hep-latr9208012.
w16x S. Datta, S. Gupta, Nucl. Phys. B 534 Ž1998. 392. hep-latr9806034.
w17x S. Datta, S. Gupta, hep-latr9906023.
w18x P. Lacock, D. Miller, T. Reisz, Nucl. Phys. B 369 Ž1992. 501.
w19x F. Karsch, M. Oevers, P. Petreczky, Phys. Lett. B 442 Ž1998. 291.
w20x U. Heller, F. Karsch, J. Rank, Phys. Rev. D 57 Ž1998. 1438.
w21x P. Lacock, D.E. Miller, B. Petersson, T. Reisz, Nucl. Phys. B 418 Ž1994. 3. hep-latr9310014.
w22x M. Teper, Phys. Rev. D 59 Ž1999. 014512. hep-latr9804008.
w23x O. Philipsen, M. Teper, H. Wittig, Nucl. Phys. B 469 Ž1996. 445. hep-latr9602006.
w24x O. Philipsen, M. Teper, H. Wittig, Nucl. Phys. B 528 Ž1998. 379, hep-latr9709145.
w25x M. Laine, Nucl. Phys. B 451 Ž1995. 484. hep-latr9504001.
w26x M. Teper, Phys. Lett. B 187 Ž1987. 345.
w27x C. Michael, J. Phys. G 13 Ž1987. 1001.
w28x K. Fabricius, O. Haan, Phys. Lett. B 143 Ž1984. 459.
w29x A.D. Kennedy, B.J. Pendleton, Phys. Lett. B 156 Ž1985. 393.
w30x B. Bunk, Nucl. Phys. B ŽProc. Suppl.. 42 Ž1995. 566.
w31x G.D. Moore, Nucl. Phys. B 523 Ž1998. 569. hep-latr9709053.
w32x O. Philipsen, H. Wittig, Phys. Rev. Lett. 81 Ž1998. 4056. hep-latr9807020.
w33x P. de Forcrand, G. Schierholz, H. Schneider, M. Teper, Phys. Lett. B 160 Ž1985. 137.
w34x P. Arnold, L. Yaffe, Phys. Rev. D 52 Ž1995. 7208. hep-phr9508280.
w35x E. Shuryak, Zh. Eksp. Teor. Fiz. 74 Ž1978. 408, wSov. Phys. JETP 47 Ž1978. 212x.
w36x J. Kapusta, Nucl. Phys. B 148 Ž1979. 461.
w37x D. Gross, R. Pisarski, L. Yaffe, Rev. Mod. Phys. 53 Ž1981. 43.
w38x A.K. Rebhan, Phys. Rev. D 48 Ž1993. R3967. hep-phr9308232.
w39x Nucl. Phys. B 430 Ž1994. 319. hep-phr9408262.
w40x M. Laine, O. Philipsen, Nucl. Phys. B 523 Ž1998. 267. hep-latr9711022.
w41x M. Laine, O. Philipsen, Phys. Lett. B 459 Ž1999. 259. hep-latr9905004.
w42x K. Kajantie, M. Laine, J. Peisa, A. Rajantie, K. Rummukainen, M. Shaposhnikov, Phys. Rev. Lett. 79
Ž1997. 3130. hep-phr9708207.
Lattice Yang–Mills theory at finite densities of heavy quarks

Kurt Langfeld a,b, Gwansoo Shin a,c
a
School of Physics, Korea Institute for AdÕanced Study, Seoul 130-012, South Korea
b
¨ Theoretische Physik, UniÕersitat
Institut fur ¨ Tubingen,
¨ ¨
D – 72076 Tubingen, Germany
c
Department of Physics and Center for Theoretical Physics, Seoul National UniÕersity, Seoul 151-742,
South Korea
Received 26 July 1999; accepted 14 January 2000
Abstract
SUŽ Nc . Yang–Mills theory is investigated at finite densities of Nf heavy quark flavors. The
calculation of the Žcontinuum. quark determinant in the large-mass limit is performed by analytic
methods and results in an effective gluonic action. This action is then subject to a lattice
representation of the gluon fields and computer simulations. The approach maintains the same
number of quark degrees of freedom as in the continuum formulation and a physical heavy quark
limit Žto be contrasted with the quenched approximation Nf ™ 0.. The proper scaling towards the
continuum limit is manifest. We study the partition function for given values of the chemical
potential as well as the partition function which is projected onto a definite baryon number. First
numerical results for an SUŽ2. gauge theory are presented. We briefly discuss the breaking of the
color-electric string at finite densities and shed light onto the origin of the overlap problem
inherent in the Glasgow approach. q 2000 Elsevier Science B.V. All rights reserved.
PACS: 11.15.Ha; 12.38.Gc

Keywords: Yang–Mills theory; Fermion determinant; Finite density; Finite temperature
1. Introduction
The next generation of particle accelerators ŽRHIC, LHC., which will start operating
at the beginning of the next millennium, will probe the deconfined regime of QCD, the
theory of strong interactions, and might reveal exotic states of hadronic matter which
appear under extreme conditions, i.e. temperature and density. Due to a significant
increase in computational power in the recent past, numerical simulations of lattice QCD
have provided insights into the high temperature and zero density phase and have
predicted a series of interesting phenomena w1x, such as deconfinement and restoration of
chiral symmetry. Unfortunately, an adequate description of finite density hadron matter
PII: S 0 5 5 0 - 3 2 1 3 Ž 0 0 . 0 0 0 3 1 - 6
K. Langfeld, G. Shin r Nuclear Physics B 572 (2000) 266–288 267
is still lacking due to conceptual problems in setting up an appropriate ’’statistical’’

measure which can be handled in computer simulations.
The generic approach to Yang–Mills thermodynamics at finite densities is based on
the introduction of a non-zero chemical potential. In the case of an SUŽ2. gauge group,
the fermion determinant is real and can be included in the probabilistic measure.
Numerical simulations can be performed by using standard algorithms w2,3x, although
this numerical approach consumes a lot of computer time due to the non-local nature of
the action. Recent progress for the case of an SUŽ2. gauge group can be found in w4x. In
the case of an SUŽ3. gauge group, the fermion determinant acquires imaginary parts for
a non-vanishing chemical potential and cannot be considered to be part of the probabilis-
tic measure. The most prominent example to circumvent this conceptual difficulty is the
so-called Glasgow algorithm w5x. There, the fermion determinant is considered to be part
of the correlation function to be calculated, and the probabilistic measure of zero-density
Yang–Mills theory is used to generate the gauge field configurations. However, it turns
out that this approach suffers from the so-called ’’overlap’’ problem implying that for
realistic lattice sizes an unrealistic number of Monte Carlo steps is necessary to achieve
reliable results w6x.
In order to alleviate this problem, the so-called quenched approximation, i.e. the limit
Nf ™ 0, where Nf is the number of quark flavors, greatly reduces the numerical task for
practical calculations. While in the case of real QCD one expects a drastic change in the
hadron density for a chemical potential m s m onset f m Br3 Ž m B is the baryon mass., the
onset value m onset which is extracted from quenched lattice QCD seems to be unnatu-
rally small w7x. Subsequently, it turned out that the quenched approximation, i.e. the limit
Nf ™ 0, of lattice QCD does not meet with the naive expectation that this limit coincides
with the heavy quark limit of real QCD w8x.
The confining properties of Yang–Mills theory in the desired limit where the quark
mass and the chemical potential are simultaneously made large Žheavy quark limit.
while the density of quarks is kept finite and non-zero were investigated in w9–12x.
Using Kogut–Susskind quarks, this limit simplifies the fermion determinant and allows
a significant improvement of the statistics w10x. It was argued that the phase of the quark
determinant only plays a role at finite temperatures w11,12x. A recent breakthrough
w13,14x was achieved by resorting to the heavy quark limit in addition to the use of the
canonical ensemble, i.e. the hadron system of fixed density Žby contrast to the
grand-canonical ensemble of fixed chemical potential.. The canonical approach involves
the grand-canonical partition function with imaginary values of the chemical potential as
first pointed out in w15x. A great simplification arises in this case from the fact that the
fermion determinant is real. The numerical analysis of w13,14x reveals that the deconfine-
ment phase transition becomes a cross-over at finite density Žsee also w10x.. The string
between static quark breaks yielding a constant heavy quark potential at large distances.
In this paper, we present a new approach to Yang–Mills theory at a finite density of
heavy quarks. We shall calculate the continuum fermion determinant for arbitrary
entries for the gluon field in the large mass limit Žrather than in the quenched
approximation Nf ™ 0. using the Schwinger proper-time regularization. The result is a
gauge invariant action of the gluon fields, depending on an UV regulator L, and is
added to the standard action of Yang–Mills theory. The total result can be discretized on
a lattice with lattice spacing a by standard methods and can be used as input for
268 K. Langfeld, G. Shin r Nuclear Physics B 572 (2000) 266–288
computer simulations. In the critical limit L ™ `, a ™ 0, physical quantities become

independent of the regularization scheme, and the results for observables are indepen-
dent of the choice of the technique, i.e. lattice regularization or Schwinger proper-time
regularization of the fermion determinant. The advantages of our approach are as
follows: firstly, the starting point of the calculation is the continuum quark determinant
with the correct number of degrees of freedom. The approach is not plagued by spurious
states, and its heavy mass limit is manifestly the correct QCD limit. Secondly, despite
the fact that the total gluonic action is still non-local, it is simple enough to allow for
fast computer simulations. Thirdly, the correct scaling of physical quantities towards the
continuum limit is manifestly the same as the one proposed by continuum Yang–Mills
theory with quarks included.
The paper is organized as follows. In Section 2 we contrast the heavy quark limit
with the quenched approximation, and discuss the difference in the renormalization
group scaling in either case. The calculation of the Žcontinuum. quark determinant in the
large-mass limit is presented in Section 3. We address the renormalization of the
coupled gluon quark system and discuss the approach of the continuum limit in lattice
simulations of the joint system. At the end of Section 3, we calculate the canonical
partition function describing a system with definite baryon number. First numerical
results for the case of an SUŽ2. gauge group are shown in Section 4. The dependence of
the quark density on the chemical potential for several values of the temperature below
and above the deconfinement temperature is discussed. The conclusions are left to the
final section.
2. Heavy fermions on the lattice
2.1. Setup
Our aim is to address the Yang–Mills theory at finite densities of baryons. For this
purpose, the grand-canonical partition function in Euclidean space-time, i.e.
H ½
Z Ž m . s Dq Dq D Am exp y d 4 x Ž LQ q L YM . ,
H 5 Ž 1.
1
L YM s Fmna w A x Fmna w A x , Ž 2.
4g2
LQ s q Ž x . i Euy m y g mAm y i m g 0 q Ž x . , Ž 3.
serves as a convenient starting point. Thereby, Fmna w A x is the usual field strength tensor
of the gluon field Am , g is the Yang–Mills gauge coupling strength, m is the chemical
potential and m is the quark mass. We will assume Nf quark flavors which are
degenerate in mass. The convention for the g-matrices can be found in appendix A. A
complete gauge fixing is understood in Ž1. with the gauge fixing terms included in the
measure D Am . Below, we will employ the lattice version of the formulation Ž1–3.
implying that we need not address the details of the gauge fixing. An alternative
description of the grand-canonical partition function is obtained by integrating out the

quark fields q Ž x ., q Ž x ., i.e.
Z Ž m . s D Am Det i Euy m y g mAm y i m g 0 exp y d 4 x L YM

H ½ H 5
H ½
s D Am exp yNf Seff w A x Ž L ,m, m . y d 4 x L YM ,
H 5 Ž 4.
where L is an ultra-violet regulator. Our goal will be to calculate Seff w A xŽ L,m, m . for
large values of the quark mass m. The result will be a gauge invariant functional of the
gluon fields Am . The joint action, Seff q Hd 4 x L YM , will then be discretized on a lattice
of spacing a and will be subject of computer simulations. In the critical limit, a ™ 0,
L ™ `, physical observables will be independent of the type of regularization and will
approach the continuum result.
2.2. HeaÕy fermion Õersus quenched limit
One can think of two limits for specifying the heavy quark approximation, i.e.
s 1r2 ,T < L < m Ž quenched limit. Ž 5.
s 1r2 ,T < m < L Ž heavy quark limit . , Ž 6.
where T is the temperature. The string tension s serves in this case as the typical
energy scale of the pure Yang–Mills system. In the case of the quenched limit Ž5., the
heavy quarks are decoupled from the Yang–Mills system. One-loop perturbation theory
Žsee e.g. Ref. w16x. then tells us that the continuum limit is approached via the scaling
24p 2
M 2 a 2 Ž b . s const. exp y ½ 11 Nc2 5
b , Ž 7.
where Nc is the number of colors, b s 2 Ncrg 2 , and M is an arbitrary physical quantity

of energy dimension one.
By contrast, if we would like to associate the impact of charm, bottom and top quark
on the gluonic sector with heavy quark physics, Eq. Ž6. must be considered as the
physical heavy quark limit. In this case, these quark degrees of freedom contribute to the
critical behavior, and one finds
24p 2
2 2
M a Ž b . s const. exp y
½ Nc Ž 11 Nc y 2 Nf . 5
b . Ž 8.
A lattice Monte Carlo simulation of SUŽ Nc . gauge theory with Nf quark flavors must
recover the scaling Ž8. towards the continuum limit. In particular, this scaling must be
obeyed in the physical heavy quark limit Ž6..
Note that the quenched limit Ž5. is formally recovered from Ž8. by taking the limit
Nf ™ 0. The dependence of the bare coupling constant g on the ultra-violet regulator a
provided by Ž8. must not be confused with the behavior of the renormalized coupling
g R Ž s . on the renormalization point s, which for instance enters a renormalization flow
analysis first proposed by Wilson w17,18x. In the latter case and for an energy scale
s - m c , where m c is the charm quark mass, the ’’running’’ of g R Ž s . with s is dictated

by the three light, so-called active, quark flavors w19x.
3. Heavy fermions’ action
3.1. The fermion determinant
The goal of this section is to calculate the fermion determinant

DF :s Det N f Ž iDu y i m g 0 y m . , Dm :s Em q iAm Ž x . , Ž 9.
resorting to a 1rm expansion where m is the fermion mass. We will here only study the
case where the masses of the quark flavors are equal. Nf is the number of quark flavors
and an UV regularization is understood in Ž9.. AmŽ x . is the SUŽ Nc . gauge field and m
the chemical potential. We will use anti-hermitian g-matrices throughout this paper Žsee
Appendix A..
The determinant D Ž9. is a Lorentz scalar in four dimensions and therefore invariant
under a reflection of all its vector entries, i.e. i Em ™ yi Em , Am ™ yAm , m ™ ym.
Exploiting the anti-hermitian property of the g-matrices, we therefore obtain
DF s Det N f r2 Ž iDu y i m g 0 y m .Ž yiDu q i m g 0 y m .
s Det N f r2 Ž P y i m g 0 .Ž P † q i m g 0 . , PsiDu y m. Ž 10 .
This equation shows the familiar result that the quark determinant is real for vanishing
chemical potential. Eq. Ž10. also tells us that one generically expects an imaginary part
of D for real values of the chemical potential while the determinant D is again real for
purely imaginary entries of m Žnote g 0 † s yg 0 ..
The functional determinant DF can be represented as a product of eigenvalues, i.e.
reg 4 N fr2
DF s Ł n ln , Ž 11 .
Ž P y i m g 0 .Ž P † q i m g 0 . c n Ž x . s l n cn Ž x . , Ž 12 .
cn Ž x 0 q 1rT , x . s Ž y1 . cn Ž x 0 , x . . , Ž 13 .
It is convenient for technical reasons Žsee also Refs. w13,14x. to remove the chemical
potential m from the operator by a scale transformation of the spinor c , i.e.
c ™ c X Ž x . s exp ym x 0 4 c Ž x . ,
P P †cnX Ž x . s l n cnX Ž x . , Ž 14 .
X 0 X 0
cn Ž x q 1rT , x . s yexp Ž ymrT . cn Ž x , x . . . Ž 15 .
Introducing the gauge covariant differential Dm :s Em q iAm and the field strength Fmn ,
i.e.
Fmn s yi Dm , Dn , Ž 16 .
we then find
smn
P P † s yDm Dm q m2 q Fmn . Ž 17 .
2
Schwinger’s proper-time method provides a gauge invariant regularization of functional
determinants. In particular, the contribution of the fermion determinant D to the gluonic
action becomes
Nf 4
` dt
yln D s Hd x lim
H tr K Ž t ; x , y . . Ž 18 .
2 x™y 1r L 2 t
The so-called heat kernel K Žt ; x, y . satisfies the equation
E †
K Žt ; x, y. qP P K Žt ; x, y. s0 Ž 19 .
Et
and the boundary conditions Ž t ' x 0 .
K Ž t s 0; x , y . s d Ž x y y . , for x , y g V , Ž 20 .
ny m
K t ; Ž t x qn T, x . , Ž t y qm T, y . s Ž yey m r T . K Žt ; x, y. , Ž 21 .
lim K Ž t ; x , y . s 0, Ž 22 .
x , y™`
where V is the space-time volume of the Žlattice. universe, T is the temperature and
n,m g Z. At the present stage, V is considered to be a cylinder with a periodicity of 1rT
in time direction and infinite extension in spatial directions. Although the source P P †
which enters the heat Eq. Ž19. is hermitian, the desired imaginary parts of Ž18. will
originate Žsee below. from the non-trivial topology of the non-simply connected
space-time manifold1.
In order to derive the systematic expansion in powers of the inverse fermion mass,
i.e. 1rm, we shall adopt the heat-kernel expansion, and resort to the techniques reported
in the important paper by Ebert and Reinhardt w20x. This expansion is generated by the
ansatz
`
K Žt ; x, y. sK0Žt ; x, y. H Žt ; x, y. , H Žt ; x, y. s Ý hj Ž x , y . t j. Ž 23 .
js0
By definition K 0 Žt ; x, y . satisfies the equation

E
K 0 Ž t ; x , y . q w yE 2 q m2 x K 0 Ž t ; x , y . s 0. Ž 24 .
Et
A particular solution to this equation is given by w20x
2 2
1 Ž tx yty qa . q Ž x yy.
K0Žt ; x, y. s
16p 2t 2 ½
exp ym t y 2
4t 5 , Ž 25 .
1
K.L. greatly acknowledges helpful discussions with Holger Gies.
where a is an arbitrary constant. The interacting part, H Žt ; x, y ., of the heat kernel

satisfies the equation
E 1
ž q zm Dm y Dm Dm q 12 smn Fmn H Ž t ; x , y . s 0,
/ Ž 26 .
Et t
where zm s xm y ym q adm 0 . Suppose that H Žt ; x, y . provides a solution to Eq. Ž26..
Since the gluon field satisfies periodic boundary conditions, i.e.
Am Ž t q nrT , x . s Am Ž t , x . , Ž 27 .
this solution possesses a discrete translation invariance, i.e.
H Ž t ; Ž t x q nrT , x . , Ž t y q nrT , y . . s H Ž t ; Ž t x , x . , Ž t y , y . . , n integer. Ž 28 .
A close inspection of Ž26. then shows that
H Ž t ; Ž t x q nrT , x . , Ž t y , y . .
is also a solution of Ž26. if one chooses a s nrT. Equipped with these prerequisites we
arrive at the central result of this section: the heat kernel which satisfies the boundary
conditions Ž21. is given by
n 1
K Ž t ; x , y . s Ý Ž ye m r T .
n 16p 2t 2
2 2
Ž t x y t y q nrT . q Ž x yy.
=exp ym2t y ½ 4t 5
=H Ž t ; Ž t x q nrT , x . , y . . . Ž 29 .
The so-called diagonal parts of the heat coefficients, i.e. lim x ™ y h j Ž x, y ., can be found
in w20x. In order for ensuring proper boundary conditions imposed by finite temperatures,
the full functional dependence of h j Ž x, y . on x, y is needed. The calculation of the
complete heat coefficients for j s 0,1,2 is one of our goals in the present paper. The
explicit calculation is shown in Appendix B. Below, we will only make use of h 0 Ž x, y .
which is given by
h 0 Ž x , y . s P exp yi ½ HC
xy
Am Ž x X . dxmX ,5 Ž 30 .
where C x y is a straight line connecting the points x and y. The result for h1Ž x, y . and
h 2 Ž x, y . can be found in appendix B. Inserting Ž29. into Ž18. while employing the
expansion Ž23. generates the 1rm expansion of the fermionic contribution to the gluonic
action, i.e.
n 1 ` dt n2
S f s yln DFs Nf Ý
n
Ý wye m r T x
js0 8p 2
H1rL 2 t 3yj ½
exp ym2t y
4T 2t 5
= d 4 x tr h j Ž Ž t x q nrT , x . , x . .
H Ž 31 .
For the later discussion, it is convenient to single out the temperature independent part
n s 0. Performing the t-integration we finally obtain
Nf m2
Sf s
8p 2
Ý
js0
½ m4y2 jG j y 2,
ž L2 /H d 4 x tr h j Ž x , x . Ž 32 .
Nf n m2 m2
q
2
Ý wye m r T x
n/0
ž
m 4y2 j I j n 2
4T 2 L2
,
/H 5
d 4 x tr h j Ž Ž t x q nrT , x . , x . , Ž 33 .
where the trace extends over color only, and where G Ž j, x . is the incomplete gamma
function,
`
jy1
G Ž j, x . s Hx ds s eys , Ž 34 .
and where
1 ` ds x
Ij Ž x , y . s
4p 2 Hy s 3yj ½
exp ys y
s 5 . Ž 35 .
Since the limit lim y ™ 0 Ij Ž x, y . does exist, only the term n s 0 Ž32. picks up a
divergence. This observation reflects the familiar fact that only temperature independent
terms are affected by ultra-violet divergences.
3.2. Renormalization
In this subsection, we will study the UV-divergence which emerge in Ž32.. Since in
this equation only the diagonal part of the heat coefficients are involved, one might
resort to the derivation of the diagonal parts presented by Ebert and Reinhardt in w20x.
One finds Žsee also Appendix B.
tr h 0 s Nc , tr h1 s 0, tr h 2 Ž x , x . s 16 tr Fmn Fmn . Ž 36 .
The only non-trivial gluon field dependence is therefore induced by the h j , j 0 2 terms.
Since the limit lim x ™ 0 G Ž j y 2, x . exists for j 0 3, only the term of Ž32. with j s 2
develops a singularity. In fact, one obtains for a large cutoff L
L2
G Ž 0,m2 L2 . s ln q O Ž 1. . Ž 37 .
m2
The divergent part of the fermionic action can therefore be calculated analytically,
Nf L2 1
S fdiv s H d 4 x ln Fmna Fmna . Ž 38 .
32p 2 m2 3
Renormalization is accomplished by absorbing the divergences by an appropriate choice
of the bare coupling strength g, i.e.
1 4
lim Hd x Fmna Fmna q S gdivl u o n q S fdiv ™ finite. Ž 39 .
L™` 4g2
One-loop perturbation theory of Yang–Mills theory without quarks yields
1 Nc 11 L2
y ln ™ finite, Ž 40 .
4g2 32p 2 6 m2
where Nc is the number of colors and m is an arbitrary renormalization scale. Including

Nf quark flavors, we make use of Ž38. and demand
1 Nc 11 L2 Nf 1 L2
y ln q ln ™ finite. Ž 41 .
4g2 32p 2 6 m2 32p 2 3 m2
The renormalization group b-function,
1 dg Ž L .
b Ž g . :s , Ž 42 .
g dln L
can be readily calculated from Ž41.. We recover the well-known result
g2 11 Nc Nf
b Ž g . sy 2 ž y / qO Ž g3.. Ž 43 .
8p 6 3
The only role of the heavy fermion determinant at this level of the large m-expansion is
to generalize the renormalization group b-function of the pure Yang–Mills theory to the
situation with Nf quark flavors included.
We are now in the position to answer the question which raises from Subsection 2.2:
how can the correct scaling towards the continuum limit be observed in lattice
Yang–Mills theory with Nf quarks included? If we neglect for the moment the
finite-temperature corrections Ž33., the inclusion of fermions only affects the parameter
b s 2 Ncrg 2 in front of the Wilson action, i.e.
Nf Nc
b ™ bF s b y 2
ln Ž a 2s . , Ž 44 .
2p 6
where we have used the freedom in choosing the Žfermionic. cutoff for defining
L2 1
2
s , Ž 45 .
m s a2
where a is the lattice spacing. s is the string tension of pure, i.e. Nf s 0, Yang–Mills
theory and serves as reference scale. Let us assume that we calculate some physical
mass M in units of the lattice spacing as function of the only parameter bF . Since the
numerical simulation is carried out with the standard Wilson action Žwith a coefficient
changed from b to bF ., we recover the familiar scaling behavior at large values of b
which is at one-loop level
24p 2
M 2 a 2 Ž bF . s k exp y ½ 11 Nc2 5
bF , Ž 46 .
where k is a numerical constant which must be ‘‘measured’’ by the numerical

simulation. Inserting bF Ž44. in Ž46. elementary manipulations of this equation finally
yield
2 Nf
ks 2 11 Nc y2 N f 24p 2
2
M a sk2
ž /
M2
exp y
½ Nc Ž 11 Nc y 2 Nf . 5
b . Ž 47 .
The crucial observation is that M 2 a 2 exponentially decreases with b while the slope
precisely meets with the expectations for an SUŽ Nc . gauge theory with Nf quark flavors.
The result of this subsection is also of practical importance: performing numerical
simulations with bF as coefficient of the Wilson action and observing the standard
scaling Ž46. automatically implies the correct approach of the continuum limit with Nf
fermions included.
Let us assume that we have calculated two physical observables M1 and M2 and that
we have obtained the proper b-dependence Ž47. for either quantity at asymptotic values
of b . The ratio of both observables then becomes
2 Nf
M12 k1 k 1 M22 11 Nc y2 N f
M22
s
k2 ž /
k 2 M12
.
Using e.g. M2 as reference scale one expresses M1 in terms of the ’’measured’’

quantities k 1 and k 2 , i.e.
k1
M1 s M .
k2 2
We finally note that temperature dependent terms Ž33., which we have not considered
in this subsection, are finite in the limit a ™ 0. We therefore do not expect that these
terms will change the critical behavior for b ™ ` in Ž47..
3.3. Lattice YM theory with finite chemical potential
In this subsection, we will consider the temperature dependent part of the sum Ž33..
The pre-factors of the heat coefficients h j ŽŽ t x q nrt, x ., x . are given by the functions
Ij Ž n2 m2r4T 2 ,m 2rL2 . which stay finite when the UV regulator is removed Ž L ™ `..
Using the heavy quark, but non-quenched limit m2rL2 ™ 0 Žsee discussion in Subsec-
tion 2.2., one finds
m2 m4y2 j ` ds n2 m2
ž
m4y 2 j Ij n 2
4T 2 /
,0 s
4p 2
H0 s 3yj ½
exp ys y
4T 2 s 5
2 m2 T 2 nj nm
s
p n 2 2
Ž 2 mT .
j
K 2y j ž /
T
, Ž 48 .
where the K nŽ x . are the modified Bessel functions of the second kind. The heat
coefficients h j are functionals of the gluonic fields only and carry energy dimension 2 j.
One therefore expects that their order of magnitude is given by L2YM j
where L YM is the
characteristic energy scale of pure Yang–Mills theory. This implies that the sum over j
in Ž48. generates the desired heavy quark expansion in powers of L2YM rmT. In the
following, we will assume that the mass is large enough for a truncation of this sum at
j s 0.
Using Ž30. one might interpret
h0 Ž Ž t x , x . . s : P Ž x . Ž 49 .
as Polyakov line P Ž x . which starts at the space-time point x and which winds around
the torus in time direction finally ending at the same point x. The operator P Ž x . can be
directly translated to the corresponding lattice operator: it is the Žpath-ordered. product
of the link variables UmŽ x . along the m s 4 direction. Expressing the field strength
squared in terms of the plaquette variable Pmn Ž x ., i.e.
1 c c 4
1
4 Fmn Fmn a s Nc 1 y tr Pmn Ž x . , Ž 50 .
Nc
we finally obtain the lattice action S latt of SUŽ2. Yang–Mills theory with Nf heavy
quark flavors
ySlatt s ySF q bF Ý Pmn Ž x . , Ž 51 .
Ž m ) n . x 4
n
m2 T 2 ` Ž y1. nm
ySF s y Ý 2
a4 Ý 2
K2 ž /
x4 p ns1 n T
n
= e n m T tr P n Ž x . q eyn m T tr Ž P † Ž x . . . Ž 52 .
where bF is given by Ž44.. Exploiting the heavy-quark limit, i.e. m 4 T, one uses the
asymptotic expression for Bessel functions,
p
K2Ž x . s ( 2x
eyx 1 q O Ž 1 x . . Ž 53 .
In this limit ySF in Ž52. becomes

n
m2 T 2 T ` Ž y1.
ySF s y Ý
x4 p 3r2
a 4
( 2m
Ý
ns1 n5r2
n
= e n Ž m ym . T tr P n Ž x . q eyn Ž m qm . T tr Ž P † Ž x . . . Ž 54 .
For an SUŽ2. gauge group, this equation is real even at finite values of the chemical
potential as it is tr P. By contrast, one expects imaginary parts for real m and
SUŽ Nc 0 3.. Finally, one observes that for purely imaginary entries m s i n , n real, Eq.
Ž54. is real as expected Žsee discussion in Subsection 3.1..
For m ( m, the sum in Ž54. is rapidly converging, while the sum is only asymptotic
for m 4 m Žfor a discussion of this issue in the context of QED see Ref. w21x.. In the
later case, resorting to different representations of the Bessel function K 2 it is possible
to perform an analytic continuation from the finite sum for m ( m to large values of
m ) m w22x. The outcome of this procedure is that by means of analytic continuation one
can assign a finite and real value to the sum also for m ) m. The lack of an imaginary
part possesses a physical interpretation: the chemical potential m describes the gain in
energy if a particle is added to the system, while an energy E is necessary to produce
the particle. Neglecting binding and confinement effects Žfor this argument only., the
particle must occupy an empty phase space cell and carries a momentum larger than the
Fermi momentum. The loss in energy due to particle production is therefore always
larger than the gain m in energy. The system is stable.
The production of quarks at m ) m is conflicting with quark confinement, which does
not tolerate single quark productions. This interplay between these mechanisms can be
anticipated from Ž54.. One expects a drastic influence of the heavy quark corrections to
the zero density action if
exp ½m 5
ym
T
²< P <: ) 1. Ž 55 .
For small temperatures Žand moderate densities., ²< P <: is small, and the onset of the
density effects is postponed to values m 4 m. On the other hand, at high temperatures
ŽT ) L YM ., ²< P <: is of order one by temperatures effects only, and a significant rise of
density is expected for m f m. From a physical point of view, this rise of density might
be interpreted as single quark productions which become feasible in the deconfinement
region.
In the following, we will study the case where m is only slightly larger than m. This
procedure will allow us to study baryon matter at moderate densities. We hope that in
this case the truncation of the sum at n s 1 already captures the essence of the
asymptotic series Ž54.. The extension of the considerations to larger values of the
chemical potential is an interesting task which is left to future investigations. In the
present paper, we will only perform a consistency check by estimating the n s 2 term.
Within this approximation, we finally obtain
m3r2 T 5r2
ySlatt s Ý bF
x4
½ Ý
m) n
Pmn Ž x . q
'2 p 3r2 a4 e Ž m ym . T tr P Ž x .
qey Ž m qm . T tr P † Ž x .
5 . Ž 56 .
The partition function which describes the interaction of SUŽ2. gauge fields with Nf
heavy fermions is finally expressed as an integral over the link variables UmŽ x ., i.e.
Z Ž m . s DUm Ž x . exp ySlatt 4 .

H Ž 57 .
Eqs. Ž56., Ž57. are the main results of the present paper. Note that the considerations of
heavy fermions induce non-local terms to the effective action of gluons. These terms are
given by the Polyakov line in Ž56.. This non-locality is confined to the temporal
direction while the action is still local in spatial directions. Note further that tr P Ž x . is
real for an SUŽ2. gauge group. In the SUŽ2. case the action in Ž51. can be therefore
easily simulated with a moderate increase in computer time compared with the case of
pure YM theory by resorting to the standard heat-bath algorithm. First results of such a
simulation will be presented below.
Let us finally calculate the baryon density r from the functional integral Ž57.. The
derivative of the partition function with respect to the chemical potential yields
d ln Z Ž m . Nc B
s , Ž 58 .
dm T
where B is the number of Baryons which are present in the Žlattice. universe. Inserting
Ž57. and Ž56. into Ž58., one finds for the baryon density
3r2
Ž mT .
r s BrV s e Ž m ym . T²tr P : y ey Ž m qm . T²tr P † : , Ž 59 .
Nc'2 p 3r2
where V s Ns3 a3 is the volume and where T s 1rNt a was used. Ns and Nt denote the
number of lattice point in spatial and time direction, respectively. ²tr P : is the
expectation value of the Polyakov line. For positive values of the chemical potential and
m f m, thermal excitations of anti-quarks can be neglected. In this case, the second term
on the right-hand side of Ž59. can be neglected, and the density is triggered by the
expectation value of the Polyakov loop.
3.4. Lattice YM theory at fixed baryon density
Endowed with the results of the previous subsection, it is an easy task to apply the
approach of Miller and Redlich w15x for describing the Yang–Mills theory with a
definite value of the baryon density. Their pioneering approach is based on the
observation that the partition function
dn Nc B
Zr Ž B . s Nc H 2p T exp ½ yi n
T 5Ž
Z in . Ž 60 .
describes the Yang–Mills theory at a fixed density provided by B baryons in the

universe. Thereby, ZŽ i n . is the grand-canonical partition function Ž1. with an imaginary
entry as chemical potential. Inserting Ž1. into Ž60., one readily verifies that the
n-integration constrains the parameter B to the baryon number, i.e.
1 3
Bs Hd x² i q Ž x . g 0 q Ž x . :t , Ž 61 .
Nc
where an average over a time period of length 1rT is understood in Ž61.. Note that T
denotes the temperature, and that we use anti-hermitian g-matrices Žsee Appendix A..
One easily repeats the derivation of the heavy quark lattice action Ž51. for the case of
an imaginary chemical potential. The calculation of the eigenmodes of the heat kernel
ŽSubsection 3.1. is unaffected by the choice of an imaginary chemical potential, and one
essentially observes a change of the boundary conditions Ž15. of the quark fields. The
final result of this consideration is that the heavy quark lattice action Ž51. is given in the
case of an imaginary chemical potential by replacing m ™ i n in Ž51..
In accordance with w10,13,14x, one immediately realizes that the heavy quark lattice
action Ž54. is real for an imaginary chemical potential. In agreement with w13,14x, we
also observe that the partition function Ž60. for a given baryon number B is invariant
under a center transformation of the links which belong to a spatial hypercube at a given
time slice. In this case, the Polyakov loop acquires a phase expŽ i 2prNc . which,
however, can be absorbed by a redefinition of the integration variable n ™ n y 2prNc .
By contrast to the case of a finite real value of the chemical potential, the sum over n
in Ž54. is rapidly converging for m s i n as long as m ) T. Calculating the fermionic
contribution exp ySF 4 to the probabilistic weight from Ž54., one observes that the terms
P n Ž x . with a multiple, i.e. n ) 1, winding of the Polyakov loop around the torus are
suppressed by the factor 1rn5r2 in Ž54.. One finds
eyS F f Ł 1 q z e i n r T tr P Ž x . q eyi n r T tr P † Ž x . 4, Ž 62 .
x4
m2 T 2 T
zs
p 3r2
a4 ( 2m
eym r T . Ž 63 .
Inserting Ž62. into Ž60., the Fourier integration can be explicitly performed. The final
result is
Nc B
Zr Ž B . f z Nc B Ý Ł tr P Ž x i . exp Ý bF ž Ý /
Pmn Ž x . , Ž 64 .
x 4 is1 x4 m) n
where x i / x k for i / k holds. Insertions of P Ž x a . P † Ž x b . are suppressed by a factor z 2

and do not contribute to the leading order result Ž64.. From a physical point of view,
these insertions correspond to particle–anti-particle excitations which are suppressed by
a factor of roughly exp y2 mrT 4 in the probabilistic weight. Note that Zr Ž B . is
manifestly center symmetric since only products of Polyakov loops which consist of a
multiple of Nc factors are present in Ž64..
It is instructive to compare the partition function Zr Ž B . of a system of finite baryon
density, i.e. B / 0, with the zero density result Zr Ž B s 0.. One observes that the finite
density partition function is suppressed by a factor z Nc B relative to the zero density
case. This illustrates the overlap problem which rises by resorting to the Glasgow
algorithm w6x. Simulating the partition function of a system with B baryons employing
an algorithm with generates an Monte Carlo ensemble which is based on the zero
density partition function requires a huge amount of statistics to overcome the entropy
factor z Nc B.
It was argued by Svetitsky w24x quite some time ago that the Polyakov line
corresponds to a field configuration with an essential overlap with the wave function of
a static quark. At that time, this approach paved the way for the interpretation of the
Polyakov line expectation value as an order parameter for the deconfinement phase
transition. Our result Ž64. nicely confirms Svetitsky’s considerations on a quantitative
level.
We finally point out that in a recent publication w25x a close relation of the phase of
the Žlattice. fermion determinant and the imaginary parts of the Polyakov loop was
observed. Our results confirm these findings.
4. Numerical simulations
While the derivation of the effective gluonic action in the previous subsections holds
for an SUŽ Nc . gauge group, we will below confine ourselves to the case of an SUŽ2.
gauge group. The partition function Ž57. supplemented with the action
ySlatt s Ý bF Ý Pmn Ž x . q j 12 tr P Ž x . ,
ž / Ž 65 .
x4 mn
2 m3r2 T 5r2
js a 4 w e Ž m ym . T q ey Ž m qm . T x Ž 66 .
'2 p 3r2
can be simulated with the standard heat bath algorithm of Creutz w23x. The non-locality
of the action due to the Polyakov loops only yields a modest increase in computational
time. We adjusted the temperature T of the system by varying the number Nt of lattice
points in time direction, i.e. T s 1rNt a. In order to remove the superficial renormaliza-
tion point, i.e. bF , dependence from physical quantities, we assume that the one-loop
scaling Ž46. is a good approximation for moderate bF values, i.e. bF g w2.1,2.5x. In
order to fix the overall scales, we used
a Ž bF s 2.3 . s 0.16 fm, Ž 67 .
which would correspond to a string tension of s s Ž440 MeV. 2 in pure Yang–Mills
theory Ž Nf s 0.. The density of nuclear matter is roughly given by r f 0.15 fmy3 . This
value corresponds on average to 6 = 10y4 baryons in an elementary cube of size a 3 and
roughly one baryon in a lattice universe of size 12 3.
In practice, it turned out to be convenient to run the simulation for a definite set of
j-values, e.g. j g w0,1r2x, and to calculate the relation between the parameters m , m, T
afterwards. The derivation of Ž65. relies on a truncation of the series Ž54. at n s 1. For
being consistent the inequality
1 ym r T
e 2 m r T q ey2 m r T
e ² < P < :- 1 Ž 68 .
2 5r2 e mrT qe mrT
must be satisfied. In the present paper, we tacitly assume that the truncation at n s 1 is
reasonable if Ž68. is satisfied. Investigations beyond this approximation are left to future
work.
4.1. Quark genesis Õersus quark confinement
In Subsection 3.3, we have seen that in the deconfined phase Ž² P : s O Ž1.. quarks
can be produced in large numbers if the chemical potential exceeds the value of the
quark mass. By contrast, in the ’’confined’’ phase, i.e. ² P : f 0, the production of single
quarks by means of the chemical potential is forbidden and only thermal excitations of
baryons Ži.e. diquarks for the present case of an SUŽ2. gauge theory. are possible. A
large number of baryons is expected to occur for m ) 2 m y b, where m is the quark
mass and b is the binding energy of the diquark system. In order to study this interplay
Fig. 1. The number B of baryons which are present in the lattice universe as a function of the chemical
potential m in units of the heavy quark mass m.
between quark production and confinement we have calculated the number B of baryons
which are present in the lattice universe as a function of the chemical potential for
several temperatures. Pure Yang–Mills theory Ž Nf s 0. possesses a second-order decon-
finement phase transition for T s Tc f 300 MeV. The result of the simulation using a
12 3 = Nt lattice and bF s 2.3 is shown in Fig. 1. For definiteness, we used m s 10's .
We have checked that the inequality Ž68. is satisfied for the parameter ranges producing
Fig. 1. For temperatures below Tc , we clearly observe an onset value m onset larger than
the quark mass. For temperatures larger Tc a significant rise in the baryon number is
observed for m f m. While in the previous case the net baryon number is produced via
baryonic excitations, excitations of single quarks contribute to the net baryon number in
the high temperature phase.
4.2. String breaking at finite density
At zero baryon Žquark. density, a convenient order parameter O of confinement is

constructed from the Polyakov line P Ž x .,
O :s ¦Ý ; x4
PŽ x. . Ž 69 .
Keeping in mind that the Polyakov line reverses its sign under a center transformation,
i.e. P Ž x . ™ yP Ž x ., ² P Ž x .: s 0 is mandatory for a realization of center symmetry. In
this case, quark confinement is inherent w24x and one observes O A 1r 'V , where V is
the space volume. If the temperature exceeds the critical temperature Tc Žf 300 MeV
for a pure SUŽ2. gauge theory., a non-vanishing expectation value of the Polyakov line,
i.e. ² P Ž x .: / 0 and therefore O s O Ž1., signals the spontaneous breakdown of center
symmetry and deconfinement.
While at zero density the color electric string between static quark sources might
extend to arbitrary length, one expects at finite densities a maximum length l 0 of the
color electric string which is controlled by the average distance between the background
quarks. In this case, string breaking occurs and the heavy quark potential saturates at a
quark distance l 0 . A precise definition of a deconfinement order parameter is cumber-
some in the finite density case, and has been under debate for more than twenty years
Žfor recent progress see Ref. w26x..
Although one does not expect a sharp deconfinement phase transition as a function of
temperature at finite densities, the gluonic state might undergo drastic changes which are
triggered by temperature effects and which are signaled by significant Žbut smooth.
changes in observables. To these respects, the behavior of O is of particular interest also
at finite densities. In the present case of an SUŽ2. gauge group Žtr P s tr P † . and a
system of heavy quarks, the density Ž59. is proportional to the expectation value of the
Polyakov loop. One therefore finds
< r < A ²tr P : (² tr P :, Ž 70 .

and concludes that O s O Ž1. as long as r / 0. Our result nicely confirms the observa-
tion in w13,14x that color electric string breaking occurs as soon as the baryon density is
non-zero.
We are led to the following behavior of O as function of the chemical potential and
density, respectively: for temperatures significantly below the critical temperature and
for m - monset , the baryon density is practically zero yielding small values of O. If the
chemical potential m exceeds monset , the drastic rise of the density is accompanied by a
strong increase of O. At large temperatures ŽT ) Tc ., O is non-zero due to temperatures
effects and changes in O due to density effects are moderate in this case Žsee Fig. 2..
Numerous numerical results confirm the qualitative behavior shown in Fig. 2. Quantita-
Fig. 2. Schematic plot: the confinement order parameter Ž69. of pure SUŽ2. gauge theory in the case of a
finite-density system as a function of the baryon density r Žleft panel. and the chemical potential m Žright
panel., respectively.
tive details are not interesting since they strongly depend on the actual choice m 4 's
and on the fine tuning m ™ mq.
5. Conclusions
Our central idea of the present paper is to combine analytic methods for calculating
the fermion determinant with the lattice description for deriving a valuable description
of the Yang–Mills system at finite densities. Removing the ultra-violet regularization
Ž L ™ ` in the case of the quark determinant and a ™ 0 in the case of the gluonic
functional integral., physical quantities are independent of the type of regularization and
approach a unique result. We have shown in Subsection 3.2 how the proper scaling
towards the continuum limit is obtained in the present case of interest. The advantage of
our approach is twofold: firstly, by construction the approach is not plagued with
spurious quark states Žsee e.g. Ref. w8x.. Secondly, the physical heavy mass limit, i.e.
L YM , T < m < L, is manifest.
In the case of the grand-canonical partition function, the chemical potential is chosen
of the order of the quark mass in order to produce significant effects in the quark density
w10x. Our first numerical results for the case of an SUŽ2. gauge group have been
presented in Section 4. Rising the temperature above the deconfinement critical one we
observe a decrease of the onset value of the chemical potential at which a rapid increase
of baryon density is observed. We interpret this result as follows: at low temperatures,
baryon density is generated by the chemical potential only via the production of baryons
Ži.e. diquarks in the case of an SUŽ2. gauge group.. At high temperatures, by contrast,
the excitation of single quarks contributing to the density becomes feasible due to
deconfinement.
In the case of the canonical partition function ZŽ B ., describing a system with baryon
number B, we observe that the quark determinant is expressed in terms of products of
Polyakov loops. In agreement with the findings in w13,14x, the determinant is center
symmetric, and a non-vanishing expectation value of the Polyakov loop at finite
densities occurs via string breaking w13,14x.
Although the calculation of the quark determinant in the present paper is tied to the
Schwinger proper-time approach which generates the large mass expansion, the basic
idea of combining an analytic calculation of the determinant with a subsequent lattice
representation of the gluon fields is quite universal. Estimating fermion determinants by
resorting to different types of approximation schemes has a long history in the literature.
The idea of studying the opposite limit m ™ 0 Žchiral limit. by applying this new idea
seems very appealing to us.
Acknowledgements
We thank Mannque Rho and Dong-Pil Min for helpful discussions and encourage-
ment. KL is indebted to Holger Gies for interesting discussions on fermion determinants
in Schwinger proper-time regularization. We greatly acknowledge the hospitality of
KIAS where large parts of the present work was performed. We thank C.W. Kim,
President of KIAS, who made this collaboration possible.
Appendix A. Notation and conventions
The metric tensor in Minkowski space is

gmn s diag Ž 1,y 1,y 1,y 1 . . Ž A.1 .
We define Euclidean tensors TŽ E . from the tensors in Minkowski space TŽ M . by
TŽmE1. . . . m N n 1 . . . n n
r s
s Ž i . Ž yi . TŽmM1 .. . . m N n 1 . . . n n , Ž A.2 .
where r and s are the numbers of zeros within m 1 . . . m N 4 and n 1 . . . nn4 , respectively.
In particular, we have for the Euclidean time and the Euclidean metric
x Ž0E . s i x Ž0M . , gŽmn
E . s diag Ž y1,y 1,y 1,y 1 . . Ž A.3 .
Covariant and contra-variant vectors in Euclidean space differ by an overall sign. For a
consistent treatment of the symmetries, one is forced to consider the g m matrices as
vectors. Therefore, one is naturally led to anti-hermitian Euclidean matrices via ŽA.2.,
gŽ0E . s igŽ0M . , gŽkE . s gŽkM . . Ž A.4 .
In particular, one finds
†
Ž g ŽmE . . s ygŽmE . , gŽmE . ,gŽnE . 4 s 2 gŽmnE . s y2 dmn . Ž A.5 .
The so-called Wick rotation is performed by considering the Euclidean tensors ŽA.2. as
real fields.
In addition, we define the square of an Euclidean vector field, e.g. Vm , by
V 2 :s VmVm s yVm V m . Ž A.6 .
This implies that V 2 is always a positive quantity Žafter the Wick rotation to Euclidean
space..
Appendix B. The off-diagonal heat coefficients
The aim of this subsection is to calculate the full space-time dependence of the heat
coefficients h k Ž x, y .. We will show that we recover the well known result for the
diagonal part h k Ž x, x .. For this purpose, our starting point is the recursion relation
i
Ž k q zm Dm . h k Ž x , y . s ž D2 q smn Dm , Dn / h ky1 Ž x , y . , Ž B.1 .
2
Fig. 3. Possibilities Ža. and Žb. for calculating the holonomy along the triangle path shown in Ža. Žnon-Abelian
Stokes theorem..
while the equation for h 0 Ž x, y . is given by
zm Dm h 0 Ž x , y . s 0, zm :s xm y ym . Ž B.2 .
We first show that the gauge covariant connection
h 0 Ž x , y . s U Ž x , y . s P exp yi ½ HC
xy
5
Am Ž x X . dxmX , Ž B.3 .
where the path C x y is a straight line connecting the points x and y, provides a solution
to Eq. ŽB.2.. For a proof we calculate the holonomy along the triangle path shown in
Fig. 3. By a comparison of the results obtained by using the equivalent paths Ža. and Žb.
in Fig. 3 at the level O Ž dx ., one finds
1
U † Ž x , y . DmU Ž x , y . s yi H0 dt t z n Fmn Ž zt q y, y . , Ž B.4 .
Fmn Ž x X , y . :s U † Ž x X , y . Fmn Ž x X . U Ž x X , y . . Ž B.5 .
In the Abelian case Fmn Ž x X , y . becomes independent of y and coincides with the
standard field strength Fmn Ž x X .. It is can be easily checked that both sides of ŽB.4.
transform homogeneously under gauge transformations. Using ŽB.5. and the anti-sym-
metry of the tensor Fmn Ž x X , y . under an exchange of the indices m and n , one
immediately observes that zm Dm h 0 Ž x, y . s 0.
In order to calculate the heat coefficients h k Ž x, y ., we decompose
hk Ž x , y . s UŽ x , y . gk Ž x , y . , with g 0 Ž x , y . s 1.
Making extensive use of ŽB.4. and the relation
Dm Ž U Ž x , y . g k Ž x , y . . s Ž Dm U Ž x , y . . g k Ž x , y . q U Ž x , y . Ž Em g k Ž x , y . . ,
one can cast ŽB.1. into a recursion relation for g k Ž x, y ., i.e.

2i 1
zX dxXa Fm a Ž xX , y . Em y zX dxXa Fm a Ž xX , y .
Ž k q zm Em . g k Ž x , y . s ž E 2y H
z C xy
z2
HC
xy
i
= HC zXX dxbXX Fm b Ž xXX , y . y HC zX 2 dxXa EmX Fm a Ž xX , y .
xy
z2 xy
smn i smn
y HC zaX dxXa Fmn Ž xX , y . y HC zX dxXa HC zXX dxbXX
z2 xy
2 z2 xy xy
smn
= Fma Ž xX , y . ,F
Fnb Ž xXX , y . q HC zX 2 dxXa
2 z2 xy
= Ž EmX Fn a Ž xX , y . y EnX Fm a Ž xX , y . . g ky1 Ž x , y . .

/ Ž B.6 .
In particular, the equation for g 1Ž x, y . becomes

1
Ž 1 q zm Em . g 1Ž x , y . s y z 2 zX dxXa Fm a Ž xX , y .
H HC zXX dxbXX Fm b Ž xXX , y .
Cx y xy
i smn
y 2 HC zX 2 dxXa EmX Fm a Ž xX , y . y HC zaX dxXa Fmn Ž xX , y .
z xy
z2 xy
i smn
y HC zX dxXa HC zXX dxbXX Fm a Ž xX , y . ,F
Fnb Ž xXX , y .
2 z2 xy xy
smn
q HC zX 2 dxXa Ž EmX Fn a Ž xX , y . y EnX Fm a Ž xX , y . . . Ž B.7 .
2 z2 xy
Referring to the particular solution of the equation Ž k q zm Em . AŽ x, y . s B Ž x, y . given by

1
AŽ x , y . s k zX ky2 zX m B Ž x , y . dxX m
H Ž B.8 .
z Cx y
we finally obtain
1 1
zXy1 zlX dxlX zXX dxXXa Fm a Ž xXX , y .
g 1Ž x , y . s y H
z C xy
ž zX 2
HC X
x y
i
= HC zXXX dxbXXXFm b Ž xXXX , y . q HC zXX 2 dxXXa EmXXFm a Ž xXX , y .
X
x y zX 2 X
x y
smn i smn
q X2 HC zaXX dxXXa Fmn Ž xXX , y . q X2 HC zXX dxXXa HC zXXX dxbXXX
z X
x y 2z x y
X X
x y
smn
= Fma Ž xXX , y . ,F
Fnb Ž xXXX , y . y HC zXX 2 dxXXa Ž EmXXFn a Ž xXX , y .
2 zX 2 x y
X
yEnXXFm a Ž xXX , y . . .
/ Ž B.9 .
Not showing terms which vanish if the trace over Dirac indices is performed, we find
1
h1 Ž x , y . s yP
P exp yi ½ HC
xy
Am Ž x X . dxmX 5 z
HC
xy
zXy1 zlX dxlX
1
= HC zXX dxXXa Fm a Ž xXX , y . HC zXXX dxbXXX Fm b Ž xXXX , y . q . . . Ž B.10 .
zX 2 X
x y
X
x y
For calculating g 2 Ž x, y ., we introduce

xX s zp q y, xXX s zX t q y s zpt q y, xXXX s zX s q y s zps q y
and rewrite g 1Ž x, y . as
1 2 1 1
g 1Ž x , y . s y H0 p dpH0 t dtH0 s ds z a zb Fm a Ž zpt q y, y . Fm b Ž zps q y, y .
1 1 1 1
yi H0 dpH0 t dt z a Em Fm a Ž zpt q y, y . y smnH dpH t dt Fmn Ž zpt q y, y .
0 0
i smn 1 2 1 1
y
2
H0 p dpH0 t dtH0 s ds z a zb Fm a Ž zpt q y, y . ,F
Fnb Ž zps q y, y .
smn 1 1
q
2
H0 dpH0 t dt z Ž E F a m na Ž zpt q y, y . y En Fm a Ž zpt q y, y . . ,
Ž B.11 .
where Em Fmn Ž zpt q y, y . s Em Fmn Ž x, y .< xs z p tqy . Using ŽB.6., ŽB.8. and ŽB.11., the final
result of a lengthy calculation is
1 1 2 1 1
g2 Ž x , y . s
z2
HC
xy
zgX dxgX y ½ H0 p dpH0 t dtH0 s ds ž 2 F mn Ž zX pt q y, y .
Fmn Ž zX ps q y, y . q 2 zbX ElX Ž Fm l Ž zX pt q y, y . ,F

=F Fm b Ž zX ps q y, y . 4 .
qzaX zbX E X 2Fm a Ž zX pt q y, y . Fm b Ž zX ps q y, y . /

1 1 i 1 2 1 1
q2
ž H0 dpH0 t dtF
F mn Ž zX pt q y, y . q H0 p dpH0 t dtH0 s ds z X X
a zb
2
1 1 1
= Fma Ž zX pt q y, y . ,F
Fnb Ž zX ps q y, y . y H dpH0 t dt z X
a
2 0
2
= Ž EmX Fn a Ž zX pt q y, y . y En Fm a Ž zX pt q y, y . . /5 q..., Ž B.12 .
where we have not shown the traceless terms. In the diagonal limit x ™ y, the formulas
for g 1Ž x, y . and g 2 Ž x, y . greatly simplify and we recover the familiar result
g 1 Ž x , x . s 0,
g 2 Ž x , x . s y 121 F 2 q 18 F 2 s 16 F 2 . Ž B.13 .
References
w1x K. Kanaya, Prog. Theor. Phys. Suppl. 131 Ž1998. 73.

w2x S. Duane, A.D. Kennedy, B.J. Pendleton, D. Roweth, Phys. Lett. B 195 Ž1987. 216.
w3x S. Duane, J.B. Kogut, Nucl. Phys. B 275 Ž1986. 398.
w4x S. Hands, J.B. Kogut, M. Lombardo, S.E. Morrison, Symmetries and spectrum of SUŽ2. lattice gauge
theory at finite chemical potential, hep-latr9902034.
w5x I.M. Barbour, A.J. Bell, M. Bernaschi, G. Salina, A. Vladikas, Nucl. Phys. B 386 Ž1992. 683.
w6x I.M. Barbour, talk presented at Workshop on QCD at Finite Baryon Density: A Complex System with a
Complex Action, Bielefeld, Germany, 27–30 Apr 1998.
w7x I. Barbour, N. Behilil, E. Dagotto, F. Karsch, A. Moreo, M. Stone, H.W. Wyld, Nucl. Phys. B 275 Ž1986.
296.
w8x M.A. Stephanov, Phys. Rev. Lett. 76 Ž1996. 4472.
w9x I. Bender, T. Hashimoto, F. Karsch, V. Linke, A. Nakamura, M. Plewnia, I.O. Stamatescu, W. Wetzel,
Nucl. Phys. B ŽProc. Suppl.. 26 Ž1992. 323.
w10x T.C. Blum, J.E. Hetrick, D. Toussaint, Phys. Rev. Lett. 76 Ž1996. 1019.
w11x R. Aloisio, V. Azcoiti, G. Di Carlo, A. Galante, A.F. Grillo, Phys. Lett. B 453 Ž1999. 275.
w12x R. Aloisio, V. Azcoiti, G. Di Carlo, A. Galante, A.F. Grillo, hep-latr9903004.
w13x O. Kaczmarek, J. Engels, F. Karsch, E. Laermann, Lattice QCD at non-zero baryon number, hep-
latr9905022.
w14x J. Engels, O. Kaczmarek, F. Karsch, E. Laermann, The Quenched limit of lattice QCD at non-zero baryon
number, hep-latr9903030.
w15x D.E. Miller, K. Redlich, Phys. Rev. D 37 Ž1988. 3716.
w16x F.J. Yndurain, Quantum Chromodynamics ŽSpringer, Berlin, 1983..
w17x K.G. Wilson, Phys. Rev. B 4 Ž1971. 3174.
w18x K.G. Wilson, J. Kogut, Phys. Rep. 12 Ž1974. 75.
w19x T. Appelquist, J. Carrazzone, Phys. Rev. D 11 Ž1975. 2865.
w20x D. Ebert, H.D. Reinhardt, Nucl. Phys. B 271 Ž1986. 188.
w21x P. Elmfors, D. Persson, B. Skagerstam, Astropart. Phys. 2 Ž1994. 299.
w22x H. Gies, QED effective action at finite temperature, hep-phr9812436, Phys. Rev. D., in press.
w23x M. Creutz, Phys. Rev. D 21 Ž1980. 2308.
w24x B. Svetitsky, Phys. Rep. 132 Ž1986. 1.
w25x P. de Forcrand, V. Laliena, The role of the Polyakov loop in finite density QCD, hep-latr9907004.
w26x H. Satz, Nucl. Phys. A 642 Ž1998. 130.
Goldstone-mode effects and scaling function for the

three-dimensional O Ž4. model
¨
Jurgen Engels, Tereza Mendes
¨ fur
Fakultat ¨ Physik, UniÕersitat
¨ Bielefeld, D-33615 Bielefeld, Germany
Received 30 November 1999; received in revised form 13 January 2000; accepted 21 January 2000
Abstract
We investigate numerically the three-dimensional O Ž4. model on 24 3 –120 3 lattices as a

function of the magnetic field H. We verify explicitly the singularities induced by Goldstone
modes in the low-temperature phase of the model, and show that they are also observed close to
the critical temperature. Our results are well described by the perturbative form of the model’s
magnetic equation of state, with coefficients determined nonperturbatively from our data. The
resulting expression is used to generate the magnetization’s scaling function parametrically.
q 2000 Elsevier Science B.V. All rights reserved.
PACS: 64.60.C; 75.10.H; 12.38.Lg

Keywords: Goldstone modes; Scaling function; O Ž N . models
1. Introduction
The O Ž N . spin models Žor, more precisely, the O Ž N .-invariant nonlinear s-models.
are defined by
b H s yJ Ý Si P S j y H P Ý Si , Ž 1.
²i , j: i
where i and j are nearest-neighbour sites on a d-dimensional hypercubic lattice, and S i

is an N-component unit vector at site i. The case N s 1 corresponds to the Ising model.
We will consider here only N ) 1. It is convenient to decompose the spin vector S i into
longitudinal Žparallel to the magnetic field H. and transverse components
ˆ q S iH .
S i s SiI H Ž 2.
E-mail addresses: engels@physik.uni-bielefeld.de ŽJ. Engels., mendes@physik.uni-bielefeld.de ŽT. Men-

des..
PII: S 0 5 5 0 - 3 2 1 3 Ž 0 0 . 0 0 0 4 6 - 8
290 J. Engels, T. Mendesr Nuclear Physics B 572 (2000) 289–304
The order parameter of the system, the magnetization M, is then the expectation value
of the lattice average S I of the longitudinal spin components,
1
Ms ¦Ý;V i
SiI s ² S I : . Ž 3.
Two types of susceptibilities are defined. The longitudinal susceptibility is the usual
derivative of the magnetization, whereas the transverse susceptibility corresponds to the
fluctuation per component of the lattice average S H of the transverse spin components,
EM
xL s sV Ž²SI2 :yM 2 . , Ž 4.
EH
1 M
xT s V ²S H 2 : s . Ž 5.
Ny1 H
These models are of general interest in condensed matter physics, but have applica-
tions also in quantum field theory. In particular, the three-dimensional O Ž4. model is of
importance for quantum chromodynamics ŽQCD. with two degenerate light-quark
flavors at finite temperature. If QCD undergoes a second-order chiral transition in the
continuum limit, it is believed to belong to the same universality class as the 3d O Ž4.
model w1–3x. QCD lattice data have therefore been compared to the O Ž4. scaling
function, determined numerically in Ref. w4x. For staggered fermions this comparison is
at present not conclusive w5x, but results for Wilson fermions w6x seem to agree quite well
with the predictions.
O Ž N . models in dimension 2 - d F 4 are predicted to display singularities on the
coexistence line T - Tc , H s 0 due to the presence of massless Goldstone modes Žsee w7x;
a thorough discussion of the Goldstone phenomenon is found in w8x.. In fact, both
susceptibilities are predicted to diverge in this region. The magnetic equation of state is
nevertheless divergence-free, and compatible with these singularities. The equation of
state was calculated up to order e 2 in the e-expansion by Brezin et al. w9x. On the basis
of this expansion it has been argued w10x that these singularities may be easily
observable, since the perturbative coefficient associated with the diverging term is quite
large.
A further consequence of the Goldstone singularities is the appearance of strong
finite-size effects at all T - Tc for H ™ 0. These effects have been studied using chiral
perturbation theory in Refs. w11–13x. Direct numerical evidence of the Goldstone
singularities is however lacking, apart from early simulations of the three-dimensional
O Ž3. model on small lattices w14x, where indications of the predicted behaviour were
found.
The aim of this paper is to verify explicitly the Goldstone singularities, and to
investigate their interplay with the critical behaviour and the effect they have on the
scaling function. We do this by simulating the three-dimensional O Ž4. model in the
presence of an external magnetic field in the low-temperature phase and close to the
critical temperature Tc . For determining the scaling function, we have also simulated at
some high-temperature values. First results of our work have been presented at Lattice’99
w15x. The plan of the paper is as follows. In the next section we review the perturbative
predictions for the magnetization and the susceptibilities at low temperatures, as well as
J. Engels, T. Mendesr Nuclear Physics B 572 (2000) 289–304 291
the analytic results for the magnetic equation of state, which is equivalent to the
magnetization’s scaling function. Our numerical results are discussed in Section 3. The
fits and a parametrization for the scaling function are given in Section 4. A summary
and our conclusions are presented in Section 5.
2. Perturbative predictions and critical behaviour
The continuous symmetry present in the O Ž N . spin models gives rise to the so-called
spin waÕes: slowly varying Žlong-wavelength. spin configurations, whose energies may
be arbitrarily close to the ground-state energy. In two dimensions these modes are
responsible for the absence of spontaneous magnetization, whereas in d ) 2 they are the
massless Goldstone modes associated with the spontaneous breaking of the rotational
symmetry for temperatures below the critical value Tc w16x. For T - Tc the system is in a
broken phase, i.e. the magnetization M ŽT, H . attains a finite value M ŽT,0. at H s 0. To
be definite we assume here H ) 0. As a consequence the transverse susceptibility, which
is directly related to the fluctuation of the Goldstone modes, diverges as Hy1 when
H ™ 0 for all T - Tc . This can be seen immediately from the identity
MŽT,H .
xT s . Ž 6.
H
This expression is a direct consequence of the O Ž N . invariance of the zero-field free
energy, and can be derived as a Ward identity w9x. It is valid for all values of T and H.
A less trivial result w7,10x is that also the longitudinal susceptibility is diverging on
the coexistence curve for 2 - d F 4. The leading term in the perturbative expansion for
2 - d - 4 is H d r2y2 . The predicted divergence in d s 3 is thus
x LŽ T - Tc , H . ; Hy1 r2 . Ž 7.
1r2
This is equivalent to an H -behaviour of the magnetization near the coexistence curve
M Ž T - Tc , H . s M Ž T ,0 . q cH 1r2 . Ž 8.
An interesting question is whether the above expressions still describe the behaviour
close to the critical region T Q Tc . We recall that the critical behaviour is determined by
the singular part of the free energy. Its scaling form in the thermodynamic limit is
f s Ž t ,h . s byd f s Ž b y t t ,b y h h . . Ž 9.
Here we have neglected possible dependencies on irrelevant scaling fields and expo-
nents. The variables t and h are the conveniently normalized reduced temperature
t s ŽT y Tc .rT0 and magnetic field h s HrH0 , and b is a free length rescaling factor.
The relevant exponents yt, h specify all the other critical exponents,
yt s 1rn , y h s 1rnc ; Ž 10 .
nc s nrbd , dn s b Ž 1 q d . , g s b Ž d y 1. . Ž 11 .
yh
Choosing the scale factor b such that b h s 1 and using M s yE f srE H one finds the
equation
M s h1r d fG Ž trh1r bd . , Ž 12 .
where fG is a scaling function. It becomes universal after fixing the normalization

constants H0 and T0 . This scaling function was calculated numerically for the 3d O Ž4.
model by Toussaint w4x and is used in comparison to QCD lattice data w5,6x.
Alternatively, one may choose b y t < t < s 1. This leads to the Widom–Griffiths form of
the equation of state w17x,
ysf Ž x. , Ž 13 .
where
y ' hrM d , x ' trM 1r b . Ž 14 .
It is usual to normalize t and h such that
f Ž 0. s 1 , f Ž y1 . s 0 . Ž 15 .
The scaling forms in Eqs. Ž12. and Ž13. are clearly equivalent. In the following we will
work with form Ž13. and obtain Ž12. from it parametrically in Section 4.
The equation of state Ž13. has been derived by Brezin´ et al. w9x to order e 2 in the
e-expansion, where e s 4 y d. Although diverging terms in x T appear at intermediate
steps of the derivation, they are canceled by diverging x L terms, and the resulting
expression is divergence-free. This expression has been considered by Wallace and Zia
w10x in the limit x ™ y1, i.e. at T - Tc and close to the coexistence curve. In this limit
the function is inverted to give x q 1 as a double expansion in powers of y and y d r2y1 ,
x q 1 s c˜1 y q c˜2 y d r2y1 q d˜1 y 2 q d˜2 y d r2 q d˜3 y dy 2 q PPP . Ž 16 .

The coefficients c˜1 , c˜2 and d˜3 are then obtained from the general expression of w9x. The
above form is motivated by the H-dependence in the e-expansion of x L at low
temperatures w10x.
In Fig. 1 we show the function f Ž x . from w9x, its low-temperature Ž x ™ y1. limit
and f Ž x . from w10x, which is obtained from the inverted form of the low-temperature
Fig. 1. The function y s f Ž x . Žsolid line. from w9x, its approximation for x ™y1 Ždashed line. and the
inverted form Ždotted line. from w10x.
expression, Eq. Ž16.. We see that the low-temperature curve remains close to the general
one for a significant portion of the phase diagram, including the critical point at x s 0,
y s 1 and moving into the high-temperature phase, the region to the right of the critical
point. The ‘‘inverted’’ curve and the low-temperature curve that generated it agree quite
well. ŽWe remark that the process of inverting produces coefficients that are determined
only to order e , while the original low-temperature expression was known to order e 2 ..
As mentioned above, the form Ž16. is equivalent to the Goldstone-singularity form for
x L at low temperatures if one identifies the variable y with the field H. Nevertheless,
the fact that this form may describe the behaviour also at temperatures close to Tc and
higher is not contradictory, since the variable y can only be identified with H if
M ŽT, H s 0. / 0, which happens only at low temperatures. The form Ž16. has explicitly
nonnegative values of y at x f y1, while the original perturbative expression produces
an unphysical negative y in a very small neighborhood of this point w9x. Possible
problems with the form Ž16. are pointed out in w18x, Section 5.
As for the large-x limit Žcorresponding to T ) Tc and small H ., the expected
behaviour is given by Griffiths’s analyticity condition w17x,
`
f Ž x. s Ý a n x gy2Ž ny1. b . Ž 17 .
ns1
None of the curves in Fig. 1 approaches this limit, since the low-temperature curves are
not valid at large x, and the general curve in its original form is known to have
problems in this limit w9x.
The perturbative equation of state has been used in Ref. w3x to produce pictures of the
expected behaviour of the 3d O Ž4. model for a large range of temperatures and magnetic
fields. The authors have employed an interpolation of the function from w9x with the
inverted form from w10x at low temperatures and the Griffiths condition at high
temperatures. When compared to Monte Carlo data for the same model in Ref. w4x, the
perturbative scaling function shows qualitative agreement. In Section 4 we propose a fit
of our Monte Carlo data to the perturbative form of the equation of state, using Ž16..
3. Numerical results
Our simulations are done on lattices with periodic boundary conditions and linear
extensions L s 24, 32, 48, 64, 72, 96 and 120 using the cluster algorithm of Ref. w11x.
We compute the magnetization M and the susceptibilities x L,T at fixed J s 1rT Ži.e. at
fixed T . and varying H. In the appendix our data are given in tabulated form. We note
that, due to the presence of a nonzero field, the magnetization is nonzero on finite
lattices, contrary to what happens for simulations at H s 0, where one is led to consider
the approximate form ²<ÝS i <rV :. The value Jc s 0.93590, obtained in simulations of
the zero-field model w19x, is used in the following.
In Fig. 2 we show our data for the magnetization for low temperatures up to Tc
plotted versus H 1r2 . We have simulated at increasingly larger values of L at fixed
values of J and H in order to eliminate finite-size effects. The finite-size effects for
small H do not disappear as one moves away from Tc , but rather increase. In Fig. 3 we
plot only the results from our largest lattices. The solid lines are fits to the form Ž8., and
294
J. Engels, T. Mendesr Nuclear Physics B 572 (2000) 289–304
Fig. 2. The magnetization versus H 1r 2 in the low-temperature region for fixed J s1.2, 1.0, 0.95 and Jc and different lattice sizes.
Fig. 3. The magnetization as a function of H 1r 2 for fixed J s1.2, 1.1, 1.0, 0.98, 0.95 and Jc , starting with the
highest curve. The size L is denoted as in Fig. 2.
the filled squares at H ) 0 denote the last points included in our fits. It is evident that
the predicted behaviour Žlinear in H 1r2 . holds close to H s 0 for all temperatures
T - Tc considered. The Goldstone-mode effects are therefore observable also rather
close to Tc . The straight-line fits coincide with the measured points in a wide range of H
for low T Žfor J s 1.2 and 1.1 up to H 1r2 s 0.32..
With increasing T the coincidence region gets smaller and vanishes at Tc . In Table 1
we have listed the fit parameters. The value M ŽT,0. obtained from the fits is the
infinite-volume value of the magnetization on the coexistence line. In the neighbourhood
of Tc it should show the usual critical behaviour
b b
M Ž T Q Tc , H s 0 . s B Ž Tc y T . s B Ž 1rJc y 1rJ . . Ž 18 .
Using this simple form, without including any next-to-leading terms, we are able to fit
all points of Table 1 with B s 0.9670Ž5. and the exponent b s 0.3785Ž6., in agreement
with the high-precision zero-field determination in Ref. w20x.
Table 1
Parameters of the fit of the magnetization to M ŽT,0.q cH 1r 2
J s1r T M ŽT,0. Slope c H-range x 2 rdof
0.95 0.20202Ž62. 1.1825Ž132. 0.001–0.003 0.37
0.98 0.30650Ž13. 0.6495Ž26. 0.0005–0.007 0.38
1.00 0.35061Ž15. 0.5245Ž20. 0.001–0.01 0.60
1.10 0.48257Ž29. 0.2940Ž38. 0.0015–0.01 0.54
1.20 0.55911Ž13. 0.2033Ž13. 0.002–0.02 0.92
Fig. 4. The magnetization at Tc . In the left figure Ža., M is plotted versus H 1r d , the line is the fit Ž19., the
solid part shows the range used for the fit. The right plot Žb. shows the finite-size-scaling function Q M from
Eq. Ž20..
As the critical point is reached the H-dependence of the magnetization should change
to satisfy critical scaling. We thus fit the data from the largest lattice sizes at Tc to the
form
M Ž Tc , H . s d c H 1r d . Ž 19 .
As can be seen in Fig. 4a a very good straight-line fit to the largest-L results is possible.
The smaller lattices show however definite finite-size effects. We find the exponent
d s 4.86Ž1., in agreement with w20x, and in addition we obtain the critical amplitude
Fig. 5. The susceptibilities below Tc at J s 0.98. On the left x L is plotted versus Hy1 r2 . The right plot shows
x T as a function of Hy1 . The lines are explained in the text.
d c s 0.715Ž1.. In Fig. 4b the magnetization at Tc is compared with the finite-size-scaling

prediction
M Ž Tc , H ; L . s Lyb r n Q M Ž HL bd r n . , Ž 20 .
using the critical exponents of Ref. w20x. We observe no corrections to scaling, even at
higher H-values. The scaling function Q M is universal. In order to be consistent with
Eq. Ž19. for large z ' HL bd r n , i.e. for finite small H and large L, it must behave as
Q M Ž z . s d c z 1r d . Ž 21 .
This offers a second way to determine the critical amplitude d c , this time exploiting also
the data of the smaller lattices. From a fit in the z-range 20–1000 we find the value
d c s 0.713Ž1., which agrees with our first determination.
In Fig. 5 we show an example Žat J s 0.98. of the different behaviours of x T and x L
at low temperatures. As a test we compare the result for E MrE H Žline. from the M-fits
in Table 1 to the x L-data. Though there are large finite-size effects for small L, the
results for the highest L-values agree nicely with the expected behaviour. A similar test
can be done for x T by showing also the result for MrH Žline., as obtained from the
measurements of the magnetization in Fig. 3. Here as well we find agreement for large
L.
The H-dependencies of the susceptibilities at the critical point are the same. Their
amplitudes differ however by a factor d . Here and in the following we use the values
d s 4.86 and b s 0.38 . From Eqs. Ž4. and Ž6. and from the magnetization at the critical
point, Eq. Ž19., we derive
x L s Ž d crd . H 1r dy1 and x T s d c H 1r dy1 . Ž 22 .

In the left part of Fig. 6 we compare x L and x T at Tc . The lines in the figure are
calculated from Eq. Ž22. and the fit results of Eq. Ž19.. We see again consistency with
Fig. 6. The susceptibilities at Tc Žleft. versus H 1r d y1 and above Tc Žright. versus H at J s 0.90 Žsquares. and
J s 0.80 Žcircles.. The lines are explained in the text.
Fig. 7. The function y s f Ž x . from fits to data at small x Žsolid line. and at large x Ždashed line..
the highest-L data. The right part of Fig. 6 shows two examples of x L and x T for high
temperatures. Both susceptibilities converge to one value x ŽT . for H ™ 0, since no
spontaneous symmetry breaking occurs for T ) Tc . At the higher temperature corre-
sponding to J s 0.80 the two susceptibilities are essentially constant Ži.e. the magnetiza-
tion is linear in H . and equal for a large range in H. However, at J s 0.90 Žthat is
closer to Tc . and finite H-values x T is larger than x L .
Fig. 8. The scaling function fG s Mr h1r d from Eq. Ž12.. The data are shown together with our fit from Eq.
Ž29..
4. The scaling function
The scaling function for the three-dimensional O Ž4. model was determined from a fit
of Monte Carlo data in Ref. w4x. Our goal here is to describe our data using the
perturbative form of the equation of state as discussed in Section 2, but with nonpertur-
bative coefficients, determined from a fit of the data. The original form of the function
y s f Ž x . from w9x is not suitable for such a fit. We thus consider Eq. Ž16., which is
written as a simple series expansion in y. We do not expect this form to describe the
data for all x and y, yet looking at Fig. 1, we hope to cover a significant portion of the
phase diagram for small x. The idea is to interpolate this result with a fit to the large-x
form Ž17..
Our fits are shown, together with our data, in Figs. 7 and 8. We have considered data
from our largest lattices, for inverse temperatures 0.9 F J F 1.0 and magnetic fields
H F 0.01. The normalization constants H0 and T0 , obtained from Eq. Ž15. and our fits in
Section 3 are given by
H0 s 5.08 Ž 3 . , T0 s 1.093 Ž 2 . . Ž 23 .
Table 2
The magnetization for J s1.2,1.1,1.0,0.98 and 0.95
L H M Ž1.2. M Ž1.1. M Ž1.0. M Ž0.98. M Ž0.95.
24 0.0005 0.3821 Ž30. 0.3150 Ž45. 0.2130 Ž22. 0.1749 Ž20. 0.1071 Ž20.
0.0010 0.4733 Ž29. 0.4076 Ž33. 0.2877 Ž17. 0.2516 Ž18. 0.1711 Ž18.
0.0015 0.5099 Ž21. 0.4331 Ž37. 0.3188 Ž14. 0.2838 Ž14. 0.2090 Ž15.
0.0020 0.5285 Ž17. 0.4587 Ž20. 0.3415 Ž11. 0.3025 Ž12. 0.2278 Ž15.
0.0025 0.5378 Ž24. 0.4685 Ž22. 0.3520 Ž13. 0.3152 Ž12. 0.2385 Ž13.
0.0030 0.5464 Ž12. 0.4761 Ž22. 0.3602 Ž08. 0.3253 Ž20. 0.2504 Ž12.
0.0050 0.5606 Ž09. 0.4927 Ž13. 0.3777 Ž08. 0.3453 Ž07. 0.2768 Ž09.
0.0070 0.5682 Ž08. 0.5002 Ž10. 0.3893 Ž06. 0.3558 Ž12. 0.2928 Ž07.
0.0100 0.5753 Ž09. 0.5085 Ž09. 0.3997 Ž04. 0.3687 Ž05. 0.3111 Ž05.
0.0200 0.5870 Ž03. 0.5227 Ž05. 0.4237 Ž03. 0.3959 Ž03. 0.3470 Ž03.
0.0500 0.6049 Ž02. 0.5477 Ž03. 0.4633 Ž02. 0.4414 Ž02. 0.4049 Ž02.
0.1000 0.6238 Ž01. 0.5730 Ž04. 0.5032 Ž01. 0.4860 Ž01. 0.4571 Ž02.
32 0.0005 0.4854 Ž23. 0.4112 Ž41. 0.2923 Ž17. 0.2553 Ž14. 0.1708 Ž15.
0.0010 0.5312 Ž20. 0.4565 Ž22. 0.3377 Ž15. 0.2977 Ž12. 0.2178 Ž12.
0.0015 0.5436 Ž10. 0.4729 Ž16. 0.3533 Ž08. 0.3159 Ž08. 0.2359 Ž09.
0.0020 0.5555 Ž10. 0.4817 Ž17. 0.3625 Ž07. 0.3254 Ž08. 0.2469 Ž09.
0.0025 0.5599 Ž25. 0.4904 Ž08. 0.3692 Ž08. 0.3318 Ž10. 0.2568 Ž07.
0.0030 0.5621 Ž08. 0.4916 Ž10. 0.3731 Ž06. 0.3358 Ž06. 0.2616 Ž08.
0.0050 0.5700 Ž06. 0.5021 Ž06. 0.3847 Ž04. 0.3508 Ž04. 0.2821 Ž04.
0.0070 0.5738 Ž05. 0.5055 Ž06. 0.3929 Ž03. 0.3603 Ž05. 0.2957 Ž04.
0.0100 0.5783 Ž03. 0.5107 Ž04. 0.4029 Ž02. 0.3706 Ž03. 0.3114 Ž03.
0.0200 0.5877 Ž02. 0.5238 Ž03. 0.4240 Ž02. 0.3968 Ž02. 0.3469 Ž02.
0.0500 0.6048 Ž01. 0.5472 Ž02. 0.4636 Ž01. 0.4418 Ž01. 0.4046 Ž01.
0.1000 0.6240 Ž01. 0.5733 Ž02. 0.5031 Ž01. 0.4860 Ž01. 0.4573 Ž01.
48 0.0005 0.5426 Ž17. 0.4735 Ž14. 0.3462 Ž07. 0.3051 Ž08. 0.2144 Ž09.
0.0010 0.5577 Ž10. 0.4839 Ž24. 0.3618 Ž05. 0.3203 Ž05. 0.2346 Ž06.
0.0015 0.5630 Ž05. 0.4902 Ž07. 0.3672 Ž05. 0.3290 Ž04. 0.2457 Ž04.
0.0020 0.5655 Ž05. 0.4925 Ž07. 0.3724 Ž03. 0.3342 Ž03. 0.2537 Ž04.
Table 3
The magnetization for J s1.2,1.1,1.0,0.98 and 0.95, continued
L H M Ž1.2. M Ž1.1. M Ž1.0. M Ž0.98. M Ž0.95.
48 0.0025 0.5679 Ž06. 0.4962 Ž13. 0.3758 Ž04. 0.3384 Ž02. 0.2604 Ž03.
0.0030 0.5691 Ž13. 0.4973 Ž05. 0.3787 Ž02. 0.3416 Ž02. 0.2662 Ž03.
0.0050 0.5731 Ž03. 0.5027 Ž03. 0.3875 Ž02. 0.3526 Ž02. 0.2838 Ž02.
0.0070 0.5758 Ž02. 0.5065 Ž02. 0.3945 Ž01. 0.3607 Ž01. 0.2960 Ž03.
0.0100 0.5793 Ž01. 0.5120 Ž02. 0.4031 Ž01. 0.3710 Ž02. 0.3119 Ž02.
0.0200 0.5879 Ž01. 0.5234 Ž01. 0.4244 Ž01. 0.3965 Ž01. 0.3471 Ž01.
0.0500 0.6049 Ž00. 0.5472 Ž01. 0.4636 Ž01. 0.4419 Ž01. 0.4049 Ž01.
0.1000 0.6239 Ž00. 0.5732 Ž01. 0.5033 Ž01. 0.4859 Ž00. 0.4574 Ž00.
64 0.0005 0.5564 Ž08. 0.4816 Ž08. 0.3573 Ž07. 0.3157 Ž05. 0.2241 Ž04.
0.0010 0.5636 Ž03. 0.4906 Ž04. 0.3656 Ž04. 0.3256 Ž03. 0.2387 Ž04.
0.0015 0.5663 Ž08. 0.4937 Ž04. 0.3701 Ž02. 0.3309 Ž03. 0.2475 Ž02.
0.0020 0.5674 Ž03. 0.4955 Ž03. 0.3739 Ž04. 0.3348 Ž02. 0.2551 Ž02.
0.0025 0.5689 Ž02. 0.4976 Ž03. 0.3767 Ž01. 0.3388 Ž02. 0.2613 Ž02.
0.0030 0.5703 Ž01. 0.4987 Ž02. 0.3795 Ž01. 0.3421 Ž01. 0.2668 Ž02.
0.0050 0.5734 Ž01. 0.5029 Ž05. 0.3876 Ž01. 0.3524 Ž02. 0.2835 Ž01.
0.0070 0.5762 Ž01. 0.5072 Ž01. 0.3945 Ž01. 0.3609 Ž01. 0.2965 Ž01.
72 0.0005 0.5590 Ž06. 0.4853 Ž09. 0.3594 Ž03. 0.3172 Ž03. 0.2259 Ž12.
0.0010 0.5644 Ž03. 0.4916 Ž04. 0.3666 Ž02. 0.3262 Ž02. 0.2386 Ž02.
0.0015 0.5667 Ž02. 0.4935 Ž04. 0.3710 Ž02. 0.3315 Ž02. 0.2478 Ž02.
0.0020 0.5682 Ž02. 0.4959 Ž02. 0.3739 Ž02. 0.3355 Ž04. 0.2550 Ž02.
0.0025 0.5692 Ž02. 0.4973 Ž02. 0.3768 Ž01. 0.3389 Ž02. 0.2610 Ž02.
96 0.0005 0.5623 Ž03. 0.4890 Ž04. 0.3618 Ž02. 0.3204 Ž02. 0.2274 Ž02.
0.0010 0.5659 Ž01. 0.4925 Ž02. 0.3671 Ž02. 0.3269 Ž01. 0.2394 Ž01.
120 0.0005 0.3210 Ž02. 0.2278 Ž02.
0.0010 0.3271 Ž01. 0.2393 Ž03.
We have performed a fit using the three leading terms in Ž16. for small y,
x 1 Ž y . q 1 s Ž c˜1 q d˜3 . y q c˜2 y 1r2 q d˜2 y 3r2 . Ž 24 .

This form was fitted in the interval y1 - x Q 1.5, giving
c˜1 q d˜3 s 0.345 Ž 12 . , c˜2 s 0.6744 Ž 73 . , d˜2 s y0.0232 Ž 49 . . Ž 25 .

The fit describes all the data at T - Tc and also higher, up to x f 5. This confirms
that the expression Ž16. is valid also away from x f y1, as observed in Section 2. We
note the small value of d˜2 . An attempt to include the next power of y leads to a
coefficient that is zero within errors. We also see that our data are not sensitive to
possible logarithmic corrections to Eq. Ž16. as proposed in Ref. w18x. Our coefficients
can be compared to those calculated perturbatively for N s 4 in Ref. w10x,
c˜1 q d˜3 s 0.528 , c˜2 s 0.530 . Ž 26 .

For large x we have done a 2-parameter fit of the behaviour Ž17., in the correspond-
ing form for x in terms of y,
x 2 Ž y . s ay 1r g q by Ž1y2 b .r g . Ž 27 .
Considering data points with y ) 50 Žcorresponding to x R 15. we obtain
a s 1.084 Ž 6 . , b s y0.874 Ž 25 . . Ž 28 .
Expression Ž27. is seen to describe the data for x R 2. We mention that a fit of our data
to the leading term in Griffiths’s condition, using x G 50, yields g s 1.45Ž1., which is in
agreement with w20x.
The small- and large-x curves cover the whole range of values of x remarkably well.
In fact, the two curves are approximately superimposed in the interval 2 Q x Q 8. We
can therefore interpolate smoothly, for example by taking
y 03 y3
x Ž y . s x 1Ž y . q x2 Ž y . Ž 29 .
y 03 q y 3 y 03 q y 3
at y 0 s 10, which corresponds to x f 4. Expression Ž29. is equivalent to the equation of

state Ž13. and to the scaling function fG in Ž12.. In Fig. 8 we show a plot of fG obtained
Table 4
The data at Jc and the susceptibilities at J s 0.98
Jc s 0.9359 J s 0.98
L H M xL DxL xT Dx T xL DxL xT Dx T
24 0.0005 0.0719 Ž18. 128.0 2.7 142.5 1.6 227.0 4.8 347.7 2.7
0.0010 0.1223 Ž18. 80.1 2.5 122.7 1.8 98.7 3.7 244.6 3.4
0.0015 0.1564 Ž16. 50.9 1.9 102.9 1.6 46.9 1.9 191.3 3.3
0.0020 0.1784 Ž24. 38.6 1.9 90.5 1.8 29.0 1.2 151.3 2.8
0.0025 0.1919 Ž12. 26.0 0.9 78.8 1.5 20.2 0.9 125.3 3.0
0.0030 0.2060 Ž11. 21.0 0.8 67.6 1.4 14.4 0.6 104.5 2.5
0.0050 0.2362 Ž08. 11.4 0.4 46.9 1.9 7.3 0.4 66.5 2.0
0.0070 0.2560 Ž09. 7.9 0.3 35.7 0.9 4.9 0.2 53.3 1.5
0.0100 0.2765 Ž05. 6.1 0.2 27.7 0.7 3.8 0.1 39.7 1.2
0.0200 0.3194 Ž04. 3.2 0.1 16.4 0.5 2.2 0.0 19.9 0.6
0.0500 0.3853 Ž02. 1.5 0.0 7.6 0.2 1.2 0.0 9.2 0.4
0.1000 0.4424 Ž01. 0.9 0.0 4.6 0.1 0.7 0.0 4.8 0.1
32 0.0005 0.1118 Ž16. 134.7 4.3 215.9 3.4 156.0 5.5 503.3 7.0
0.0010 0.1558 Ž13. 58.6 2.2 152.8 5.8 50.3 2.5 301.2 7.1
0.0015 0.1768 Ž11. 36.5 1.5 123.8 3.2 23.4 1.6 208.8 5.0
0.0020 0.1922 Ž10. 26.9 1.3 96.8 3.9 15.4 0.7 156.5 4.7
0.0025 0.2050 Ž08. 20.4 0.9 81.9 2.3 10.8 0.5 127.8 5.1
0.0030 0.2147 Ž17. 16.3 0.6 67.2 1.9 10.5 0.6 119.0 3.8
0.0050 0.2398 Ž06. 10.3 0.3 46.0 1.3 5.4 0.2 69.0 1.9
0.0070 0.2582 Ž04. 7.7 0.3 35.2 1.0 4.1 0.1 50.7 1.4
0.0100 0.2778 Ž04. 5.8 0.2 26.9 0.7 3.3 0.1 37.7 1.1
0.0200 0.3201 Ž02. 3.2 0.1 16.0 0.4 2.2 0.1 19.4 0.5
0.0500 0.3852 Ž02. 1.6 0.0 7.3 0.2 1.1 0.0 8.9 0.2
0.1000 0.4427 Ž01. 0.9 0.0 4.5 0.1 0.7 0.0 4.9 0.1
48 0.0005 0.1391 Ž10. 85.2 4.2 278.6 6.8 57.2 3.0 590.0 16.4
0.0010 0.1720 Ž07. 38.8 1.8 165.5 4.6 19.9 1.6 338.9 11.2
0.0015 0.1862 Ž09. 28.8 1.3 126.1 5.4 11.8 0.7 212.4 7.0
0.0020 0.1983 Ž04. 21.1 0.6 93.6 2.5 8.4 1.2 161.2 6.2
Table 5
The data at Jc and the susceptibilities at J s 0.98, continued
Jc s 0.9359 J s 0.98
L H M xL Dx L xT Dx T xL Dx L xT Dx T
48 0.0025 0.2080 Ž05. 17.5 0.6 81.0 2.5 7.0 0.2 124.6 4.2
0.0030 0.2161 Ž04. 15.0 0.4 69.7 1.8 6.4 0.2 113.6 4.0
0.0050 0.2403 Ž03. 9.8 0.3 51.1 1.4 4.7 0.1 68.7 2.3
0.0070 0.2580 Ž02. 7.1 0.2 35.6 1.0 3.8 0.1 51.5 1.8
0.0100 0.2773 Ž02. 5.6 0.2 27.7 0.9 3.2 0.1 38.5 1.8
0.0200 0.3197 Ž01. 3.3 0.1 16.6 0.5 2.1 0.1 19.3 0.5
0.0500 0.3854 Ž01. 1.6 0.0 7.5 0.2 1.2 0.0 9.0 0.3
0.1000 0.4427 Ž01. 0.9 0.0 4.3 0.1 0.7 0.0 4.8 0.1
64 0.0005 0.1463 Ž07. 76.9 4.8 297.5 9.5 31.3 2.0 651.7 23.3
0.0010 0.1722 Ž05. 33.6 1.3 170.7 6.0 13.3 0.8 332.7 13.9
0.0015 0.1875 Ž04. 24.1 0.9 117.5 3.6 9.8 0.6 222.4 14.9
0.0020 0.1990 Ž03. 21.1 0.7 97.3 3.4 8.3 0.3 181.5 7.9
0.0025 0.2085 Ž03. 17.4 0.6 85.6 3.4 6.6 0.2 134.2 4.8
0.0030 0.2163 Ž03. 16.1 0.5 72.4 2.4 6.2 0.2 109.6 5.6
0.0050 0.2405 Ž02. 9.7 0.3 46.5 2.0 4.6 0.1 68.2 4.3
0.0070 0.2578 Ž02. 7.7 0.2 36.6 1.0 4.0 0.1 52.7 2.4
0.0100 0.2772 Ž01. 5.6 0.2 29.0 0.7
72 0.0005 0.1481 Ž06. 70.3 4.0 292.5 9.0 25.8 1.4 690.9 25.0
0.0010 0.1724 Ž03. 34.1 1.2 175.2 5.5 12.5 0.5 329.1 14.2
0.0015 0.1872 Ž03. 26.2 0.9 126.3 4.0 9.1 0.4 222.7 8.6
0.0020 0.1995 Ž02. 20.7 0.7 100.8 3.4 7.4 0.2 159.4 7.3
0.0025 0.2084 Ž02. 17.7 0.6 84.5 2.8 6.4 0.2 135.4 8.4
0.0030 0.2165 Ž02. 14.7 0.5 68.1 2.0
96 0.0005 0.1487 Ž04. 64.5 2.6 302.6 16.4 16.6 2.2 615.6 36.5
0.0010 0.1721 Ž03. 36.2 1.5 181.0 6.2 10.6 0.5 315.3 13.6
0.0015 0.1875 Ž02. 25.5 0.8 128.8 4.6
0.0020 0.1993 Ž02. 20.7 0.7 98.7 3.0
120 0.0005 0.1488 Ž03. 62.8 3.7 292.6 10.4 14.7 0.6 638.6 32.2
0.0010 0.1722 Ž02. 36.5 1.6 171.4 6.9 10.2 0.3 316.5 43.5
parametrically from x Ž y . in Ž29.. The two variables in the plot are simply related to x
and y by
fG s Mrh1r d s yy1 r d , trh1r bd s xyy1 r bd . Ž 30 .

We see a remarkable agreement of our data points with the form suggested by
perturbation theory. With respect to the scaling function in Ref. w4x our function is
slightly higher for large negative trh1r db , due to our more complete elimination of
finite-size effects.
5. Summary and conclusions
We have shown that the Golstone singularities are clearly observable at low tempera-
tures, and also close to Tc . In fact, we are able to use the observed Goldstone-effect
Table 6
The data for J s 0.90 and J s 0.80
J L H M xL xT
0.90 64 0.0005 0.0129 Ž01. 25.21 Ž38. 25.86 Ž28.
0.0010 0.0256 Ž01. 25.13 Ž43. 25.69 Ž27.
0.0015 0.0378 Ž01. 24.53 Ž37. 24.97 Ž23.
48 0.0020 0.0496 Ž02. 23.69 Ž33. 25.02 Ž23.
0.0030 0.0718 Ž02. 21.19 Ž31. 23.80 Ž20.
0.0050 0.1087 Ž01. 16.20 Ž23. 21.74 Ž18.
0.0070 0.1378 Ž01. 12.56 Ž18. 19.56 Ž16.
0.0100 0.1707 Ž01. 9.62 Ž14. 17.12 Ž14.
0.0200 0.2379 Ž01. 4.98 Ž07. 11.62 Ž10.
0.0500 0.3294 Ž00. 2.02 Ž03. 6.68 Ž06.
32 0.1000 0.4016 Ž01. 1.05 Ž02. 4.05 Ž03.
0.80 64 0.0005 0.0019 Ž01. 3.59 Ž13. 3.76 Ž12.
0.0010 0.0038 Ž01. 3.88 Ž14. 3.77 Ž11.
0.0015 0.0056 Ž01. 3.78 Ž11. 3.65 Ž10.
0.0020 0.0075 Ž01. 3.70 Ž10. 3.76 Ž09.
48 0.0030 0.0112 Ž01. 3.80 Ž08. 3.87 Ž07.
0.0050 0.0189 Ž01. 3.66 Ž06. 3.76 Ž05.
0.0070 0.0262 Ž01. 3.70 Ž06. 3.75 Ž05.
0.0100 0.0374 Ž01. 3.68 Ž05. 3.71 Ž04.
0.0200 0.0732 Ž01. 3.54 Ž05. 3.66 Ž03.
0.0500 0.1650 Ž00. 2.61 Ž04. 3.32 Ž03.
0.1000 0.2673 Ž00. 1.60 Ž02. 2.72 Ž02.
behaviour to extrapolate our data to H ™ 0 and obtain the zero-field critical exponent b
in good agreement with w20x. We remark that the same does not happen at high
temperatures: we are not able to get the exponent g from extrapolations using the
constant behaviour of the longitudinal susceptibility Žor the linear behaviour of the
magnetization., since this behaviour is masked close to Tc for the fields H we have
taken into account. At the same H ’s the H 1r2 behaviour is clearly present for all the
T - Tc we consider, showing that the Goldstone effect is dominating the critical one,
except at Tc .
A strong manifestation of the Goldstone behaviour had been conjectured perturba-
tively w10x, based on the size of the coefficient c˜2 in the e-expansion of the equation of
state. We have fitted the perturbative form in Section 4, finding a coefficient that is even
larger than the perturbative one.
The resulting curve for the equation of state describes all the data beautifully, and can
be plotted parametrically for the scaling function.
As a by-product of our work we have determined the critical exponent d s 4.86Ž1. by
a fit of the magnetization at Tc to the critical scaling behaviour as a function of H. In
addition we checked the finite-size-scaling prediction for M. It is remarkable that we
observed in both cases no corrections to scaling.
A similar investigation for the O Ž2. model is currently being done w21x. The main
objective here is to obtain the O Ž2.-scaling function. This would allow for a direct test
of lattice QCD with staggered fermions, because there the remaining chiral symmetry at
finite lattice spacing is O Ž2..
Acknowledgements
We thank Attilio Cucchieri and Frithjof Karsch for helpful suggestions and com-
ments. Our computer code for the cluster algorithm was based on a zero-field program
by Manfred Oevers. This work was supported by the Deutsche Forschungsgemeinschaft
under Grant No. Ka 1198r4-1.
Appendix A
In Tables 2, 3, 4, 5, 6 we present the results of our Monte Carlo simulations.
References
w1x R. Pisarsky, F. Wilczek, Phys. Rev. D 29 Ž1984. 338.

w2x F. Wilczek, J. Mod. Phys. A 7 Ž1992. 3911.
w3x K. Rajagopal, F. Wilczek, Nucl. Phys. B 399 Ž1993. 395.
w4x D. Toussaint, Phys. Rev. D 55 Ž1997. 362.
w5x C. Bernard et al., MILC Collaboration, hep-latr9908008.
w6x A. Ali Khan et al., CP-PACS Collaboration, hep-latr9909075.
w7x J. Zinn-Justin, Quantum Field Theory and Critical Phenomena ŽClarendon Press, Oxford, 1996..
w8x R. Anishetty, R. Basu, N.D. Hari Dass, H.S. Sharatchandra, Int. J. Mod. Phys. A 14 Ž1999. 3467.
w9x ´
E. Brezin, D.J. Wallace, K.G. Wilson, Phys. Rev. B 7 Ž1973. 232.
w10x D.J. Wallace, R.K.P. Zia, Phys. Rev. B 12 Ž1975. 5340.
w11x ´ P. Hasenfratz, J. Nager, F. Niedermayer, Nucl. Phys. B 350 Ž1991. 893.
I. Dimitrovic,
w12x ¨
M. Gockeler, K. Jansen, T. Neuhaus, Phys. Lett. B 273 Ž1991. 450.
w13x ´ J. Nager, K. Jansen, T. Neuhaus, Phys. Lett. B 268 Ž1991. 408.
I. Dimitrovic,
w14x ¨
H. Muller-Krumbhaar, Z. Physik 267 Ž1974. 261.
w15x J. Engels, T. Mendes, hep-latr9909013, to appear in the proceedings of Lattice’99.
w16x V.G. Vaks, A.I. Larkin, S.A. Pikin, Sov. Phys. JETP 26 Ž1968. 647.
w17x R.B. Griffiths, Phys. Rev. 158 Ž1967. 176.
w18x A. Pelissetto, E. Vicari, Nucl. Phys. B 540 Ž1999. 639.
w19x M. Oevers, Diploma thesis, Bielefeld University Ž1996..
w20x K. Kanaya, S. Kaya, Phys. Rev. D 51 Ž1995. 2404.
w21x J. Engels, S. Holtmann, T. Mendes, T. Schulze, in preparation.
Scalar one-loop integrals using the negative-dimension

approach
C. Anastasiou, E.W.N. Glover, C. Oleari
Department of Physics, UniÕersity of Durham, Durham DH1 3LE, England, UK
Received 3 August 1999; accepted 4 October 1999
Abstract
We study massive one-loop integrals by analytically continuing the Feynman integral to

negative dimensions as advocated by Halliday and Ricotta and developed by Suzuki and Schmidt.
We consider n-point one-loop integrals with arbitrary powers of propagators in general dimension
D. For integrals with m mass scales and q external momentum scales, we construct a template
solution valid for all n which allows us to obtain a representation of the graph in terms of a finite
sum of generalised hypergeometric functions with m q q y 1 variables. All solutions for all
possible kinematic regions are given simultaneously, allowing the investigation of different ranges
of variation of mass and momentum scales.
As a first step, we develop the general framework and apply it to massive bubble and vertex
integrals. Of course many of these integrals are well known and we show that the known results
are recovered. To give a concrete new result, we present expressions for the general vertex
integral with one off-shell leg and two internal masses in terms of hypergeometric functions of
two variables that converge in the appropriate kinematic regions. The kinematic singularity
structure of this graph is sufficiently complex to give insight into how the negative-dimension
method operates and gives some hope that more complicated graphs can also be evaluated. q 2000
PACS: 12.38.Bx; 02.90.qp; 12.20.Ds

Keywords: Scalar integrals; Negative-dimension method; Vertex integrals
1. Introduction
Loop integrals play an important role in making precise perturbative predictions in

quantum field theory, in general, and in the Standard Model of particle physics, in
E-mail addresses: Ch.Anastasiou@durham.ac.uk ŽC. Anastasiou., E.W.N.Glover@durham.ac.uk

ŽE.W.N. Glover., Carlo.Oleari@durham.ac.uk ŽC. Oleari..
PII: S 0 5 5 0 - 3 2 1 3 Ž 9 9 . 0 0 6 3 7 - 9
308 C. Anastasiou et al.r Nuclear Physics B 572 (2000) 307–360
particular. As such, a large effort has been expended in developing methods for
evaluating them. The problem is complicated by the appearance of ultraviolet ŽUV. and
infrared ŽIR. singularities, and it has become customary to use dimensional regularisa-
tion w1–4x to extend the dimensionality of the loop integral away from 4 dimensions to
D s 4 y 2 e , to regulate the infrared and ultraviolet singularities.
With the increasing of the number of legs, of the number of mass scales or of the
number of loops, the integrals can be made almost arbitrarily complex and difficult to
solve analytically. Different methods w5–18x have been developed to solve the Feynman
integrals. We mention here only two of them: the integration by parts w8,9x, which works
well for some two-loop vertex diagrams w19x reducing them to simpler known graphs
with different powers of the propagators Žhowever this breaks down for more compli-
cated graphs such as the double box, where irreducible numerators factors are present.
and the Mellin–Barnes integral representation Žsee for example Refs. w20–22x., which
was successfully used by Smirnov w23x to calculate the two-loop box integral. In this
approach, the integral is usually written as multiple contour integrals of G functions and
powers of ratios of the mass scales in the problem. By closing the contour, we obtain an
infinite series of residues at the singular points of the G functions. These series can be
identified as generalised hypergeometric functions, whose convergence properties reflect
the threshold-singularity structure of the integral.
There are several advantages in using hypergeometric functions to represent the
integral. First, these hypergeometric functions often have integral representations them-
selves, in which an expansion in e can be made, yielding expressions in logarithms,
dilogarithms, etc. It seems that, where direct evaluation of the hypergeometric function
in terms of known functions is possible, very compact results are obtained w24–26x.
Second, because the series is convergent and well behaved in a particular region of
phase space, it can be numerically evaluated w27,28x. In fact, each hypergeometric
representation immediately allows an asymptotic expansion of the integral in terms of
ratios of momentum and mass scales. Third, through analytic continuation formulae, the
hypergeometric functions valid in one kinematic domain can be re-expressed in a
different kinematic region.
Not all work has concentrated around D s 4. In fact, a close connection between
tensor loop integrals – those with additional powers of the loop momentum in the
numerator – and higher-dimension scalar integrals Ž D s 6 y 2 e , for example. is well
established w18,29–33x. Furthermore, in 1987, Halliday and Ricotta w34–37x suggested
that it would be useful to calculate the loop integral considering D as a negative
number. Because loop integrals are analytic in the number of dimensions D Žand also in
the powers of the propagators. they proposed to calculate the integral in negative
dimensions and return to positive dimensions, and specifically D s 4 y 2 e , after the
integrations have been performed. As we will discuss more fully later on, integration
over the loop momentum andror the parameters introduced to do the loop integration is
replaced with infinite series, which again can be identified as generalised hypergeomet-
ric functions. Recently this idea has been picked up again by Suzuki and Schmidt who
have evaluated a number of two-loop integrals w38–42x, three-loop integrals w43x,
one-loop tensor integrals w44x as well as the one-loop massive box integral for the
scattering of light by light w45,46x. In this latter case, as well as reproducing the known
hypergeometric-series representations of Ref. w24x, valid in particular kinematic regions,
C. Anastasiou et al.r Nuclear Physics B 572 (2000) 307–360 309
Suzuki and Schmidt simultaneously found hypergeometric solutions valid in other

kinematic domains. Of course, all of these solutions are related by analytic continuation.
However, it is easy to envisage integrals that yield hypergeometric functions where the
analytic continuation formulae are not known a priori. In these cases, having series
expansions directly available in all kinematic regions is useful.
In this paper we wish to explore the negative-dimension approach ŽNDIM. further. In
particular we focus on one-loop integrals with general powers of the propagators and
arbitrary dimension D. There are several reasons for doing this. First, it allows
connection with the general tensor-reduction program based on integration by parts of
Refs. w30–33x. Here the tensor integrals are linear combinations of scalar integrals with
either higher dimension or propagators raised to higher powers. Second, we can imagine
inserting the one-loop results into a two-loop integral by closing up external legs. This is
trivial for most bubble integrals, but more complicated for vertex and box graphs.
Broadhurst w47x has shown that this is possible for the non-trivial two-loop self-energy
graph. Third, it actually simplifies the calculation. As we will show, by keeping the
parameters general, it is easier to identify the regions of convergence of the hypergeo-
metric series and therefore which hypergeometric functions to group together. For
specific values of the parameters, the hypergeometric functions often collapse to simpler
functions. As a first step, we develop the general framework and apply it to massive
bubble and vertex integrals. Of course many of these results are well known. However,
they serve to iron out some of the subtleties of the NDIM approach. To give a concrete
new result, we present expressions for the vertex integral with one off-shell leg and two
internal masses in arbitrary dimension and for general powers of the propagators in
terms of hypergeometric functions of two variables. The kinematic singularity structure
of this graph is sufficiently complex to give insight into how NDIM operates and gives
some hope that more complicated graphs can also be evaluated.
Our paper is organised as follows. In Section 2.2 we first review the theoretical
framework of one-loop integrals with Schwinger parameters and briefly explain the
basic idea of integrating in negative dimensions. We then apply NDIM to construct
template solutions for arbitrary one-loop integrals together with a linear system of
constraints that relates the powers of the propagators in the loop integral to the
summation variables. The system of constraints has many solutions and each one must
be inserted into the template solution, yielding a sum over fewer variables that can be
identified as a generalised hypergeometric function. This method gives simultaneously
all the solutions in all the possible kinematic regions. The approach is illustrated for the
massive bubble integral where we show how to recover the known results. We discuss
how the form of the solution in different kinematical regimes is dictated by the
convergence properties of the hypergeometric functions and the structure of the system.
In Section 3 we consider one-loop triangles and give the form of the template solution
and the system of constraints with arbitrary powers of the propagators, internal masses,
external legs off-shell and for general D. We apply this result to triangle integrals with
three scales and give expressions valid in the various kinematic regions appropriate to
the vertex integral. Results are given in the form of hypergeometric functions of one and
two variables, which are defined in Appendix A. For specific choices of D and the
propagator powers, these functions can be evaluated as logarithms and dilogarithms
using the integral representations that are also provided in the appendix together with a
Fig. 1. The generic one-loop graph.
list of hypergeometric identities that often simplify the results. Finally, our method is
summarized in Section 4.
2. Theoretical framework
The generic n-point one-loop integral in D-dimensional Minkowski space with loop
momentum k is given by
d Dk 1
InD Ž n i 4 ; Q i2 4 , Mi2 4 . s H ip , Ž 2.1 .
D r2
An1 1 . . . Ann n
where, as indicated in Fig. 1, the external momenta k i are all incoming so that
Ý nis1 k im s 0 and the propagators have the form
A1 s k 2 y M12 q i0,
2
iy1
Ai s k q
ž Ý
js1
kj
/ y Mi2 q i0, i / 1, Ž 2.2 .
Mi being the mass of the ith propagator. The external momentum scales are indicated
with Q i2 4 . For standard integrals, the powers n i to which each propagator is raised are
usually unity. However, we wish, where possible, to leave the powers as general as
possible. As discussed earlier, this may have some advantages in evaluating two-loop
integrals where often one-loop integrals with arbitrary powers can be inserted into the
second loop integration.
To evaluate this integral, we introduce a Schwinger parameter x i for each propagator
Žnoting that A i - 0 after Wick rotation to Euclidean space. so that
n
1 Ž y1. i `
Ani i
s H dx x n iy1 exp Ž x i A i . ,
G Ž ni . 0 i i
Ž 2.3 .
and we can rewrite Eq. Ž2.1. as

dDk n
InD Q i2
Ž ni4 ; 4 , Mi2 4 . s H D xH ip D r2 exp Ý x i A i ž is1
/ , Ž 2.4 .
where we have used the shorthand

n 1 `
s
H D x s Ž y1. is1 0 ž
Ł G Ž n . H dx x
i
i
n iy1
i / , Ž 2.5 .
with
n
ss Ý ni . Ž 2.6 .
is1
The Gaussian integral over the loop momentum can be solved in a straightforward way,
and using the Minkowski space relation
dDk 1
H ip D r2
exp Ž a k 2 . s , D r2 Ž 2.7 .
a
we have the usual Minkowski space result
1
InD Ž n i 4 ; Q i2 4 , Mi2 4 . s D x D r2 exp Ž QrP
H P . exp Ž yM
M.. Ž 2.8 .
P
The quantities P and M are given by
n
Ps Ý xi , Ž 2.9 .
is1
n
Ms Ý x i Mi2 , Ž 2.10 .
is1
while Q may be simply read off from the Feynman diagram

jy1 2 q
ny1 n
Qs Ý Ý
is1 jsiq1
xi x j
žÝ /
ksi
kk s Ý Qi .
is1
Ž 2.11 .
Each of the q terms in Q is indicated with Qi , and is obtained by cutting the loop
diagram into two across propagators a and b and constructing the four-momentum Q im
on each side of the cut: Qi s x a x b Q i2 . For example, the one-loop bubble graph shown in
Fig. 2 has two propagators Ž n s 2., so that P s x 1 q x 2 and M s x 1 M12 q x 2 M22 . Q is
Fig. 2. The one-loop vacuum graph.

obtained by examining the momentum flowing across the only possible cut Ž q s 1.:
Q s Q1 s x 1 x 2 Q12 , with Q12 s k 12 .
2.1. The negatiÕe-dimension approach
The crucial point in the negative-dimension approach is that the Gaussian integral
Ž2.7. is an analytic function of the space-time dimension. Hence it is possible to consider
D - 0 and to make the definition w34,35x
dDk n
H ip D r2 Ž k 2 . s n! dnq D
2 ,0 Ž 2.12 .
for positive values of n. We see that by expanding the exponential in Ž2.7. and inserting
the definition Ž2.12., after the exchange of the integration with respect to the summation
dDk ` an dDk n 1
H ip D r2
exp Ž a k 2 . s Ý H D r2 Ž k2 . s D r2
, Ž 2.13 .
ns0 n! ip a
we recover the original result, provided that D is both negative and even Žso that the
Kronecker d can be satisfied and the contribution with n s yDr2 selected from the
sum.. We note that, with this definition, negative-dimensional integrals can be shown to
obey the necessary translation properties w34,35x.
2.2. The general case: different masses
For the one-loop integrals we are interested in here, we follow the approach
suggested by Suzuki and Schmidt w38–46x and view Eqs. Ž2.4. and Ž2.8. as existing in
negative dimensions. Making the same series expansion of the exponential as above, Eq.
Ž2.4. becomes
ni
` dDk n Ž x i Ai .
InD Q i2
Ž ni4 ; 4 , Mi2 4 . sH D x Ý H ip D r2 Ł
n1 , . . . , n n s0 is1 ni !
` n x in i
s Dx
H Ý InD Ž yn1 , . . . ,y n n ; Q i2 4 , Mi2 4 . Ł ,
n1 , . . . , n n s0 is1 ni !
Ž 2.14 .
where the n i are positive integers. The target loop integral is an infinite sum of
Žintegrals over the Schwinger parameters of. loop integrals with negative powers of the
propagators.
Likewise, we expand the exponentials in Eq. Ž2.8.
D
m
` Q n Pyny 2 ` Ž yM
M.
InD Q i2
Ž ni4 ; 4 , Mi2 4 . sH D x Ý Ý , Ž 2.15 .
ns0 n! ms0 m!
and introduce the integers q1 , . . . ,q q , p 1 , . . . , pn and m1 , . . . ,m n to make multinomial

expansions of Q , P and M respectively
n
` Q1q1 Qqq q
Q s Ý ... Ž q1 q . . . qqq . !,
q1 , . . . , q q s0 q1 ! qq !
D ` x 1p 1 x np n
Pyny 2 s Ý ... Ž p1 q . . . qpn . !,
p1 , . . . , p n s0 p1 ! pn !
m1 mn
m
`
Ž yx 1 M12 . Ž yx n Mn2 .
Ž yM
M. s Ý ... Ž m1 q . . . qm n . !, Ž 2.16 .
m1 , . . . , m n s0 m1 ! mn !
subject to the constraints

q n n
D
Ý qi s n, Ý pi s yn y 2 and Ý m i s m. Ž 2.17 .
is1 is1 is1
Altogether, Eqs. Ž2.15. and Ž2.16. give

` Q1q1 . . . Qqq q x 1p 1 . . . x np n
InD Ž n i 4 ; Q i2 4 , Mi2 4 . s D x H Ý
p1 , . . . , p n s0
q 1 ! . . . q q ! p 1 ! . . . pn !
q1 , . . . , q q s0
m1 , . . . , m n s0
m1 mn
Ž yx 1 M12 . Ž yx n Mn2 .
= ... Ž p1 q . . . qpn . !,
m1 ! mn !
Ž 2.18 .
with the constraints expressed by Eq. Ž2.17..
We recall that each of the Qi is a bilinear in the Schwinger parameters, so that the
target loop integral is now an infinite sum of powers of the scales of the process Žwith
each of the Mi2 and the Q i2 raised to a different summation variable. integrated over the
Schwinger parameters.
Eqs. Ž2.14. and Ž2.18. are two different expressions for the same quantity: InD .
However, rather than performing the integrals over the x i ’s, we use the fact that the x i ’s
are independent parameters, so that the integrands themselves must be equivalent:
` n x in i
Ý InD Ž yn1 , . . . ,y n n ; Q i2 4 , Mi2 4 . Ł
n1 , . . . , n n s0 is1 ni !
m1 mn
` Q1q1 . . . Qqq q x 1p 1 . . . x np n Ž yx 1 M12 . Ž yx n Mn2 .
s Ý ...
p1 , . . . , p n s0
q 1 ! . . . q q ! p 1 ! . . . pn ! m1 ! mn !
q1 , . . . , q q s0
m1 , . . . , m n s0
= Ž p 1 q . . . qpn . !. Ž 2.19 .
Comparing term by term the left-hand-side Žlhs. with the right-hand-side Žrhs. allows us
n1 nn
to read off the value of InD. In fact, the coefficient of the term xy
1 . . . xy
n in the lhs of
Eq. Ž2.19., where the n i are negative integers, is given by
n 1
I2D Ž n 1 , . . . , nn .
ž Ł
is1 G Ž 1 y n i . / . Ž 2.20 .
n1 nn
This term is equal to the coefficient of the term xy
1 . . . xy
n in the rhs of Eq. Ž2.19..
Writing a general expression is not possible, since the Qi are process dependent.
Nevertheless, we can extract the momentum scale Q i2 from each of the Qi and find the
coefficient of this term to be
`
q1 qq m1 mn
Ý Ž Q12 . . . . Ž Q q2 . Ž yM12 . . . . Ž yMn2 .
p1 , . . . , p n s0
q1 , . . . , q q s0
m1 , . . . , m n s0
n q n
1 1
=
žŁ is1 G Ž 1 q m i . G Ž 1 q pi . /žŁ is1 G Ž 1 q qi . /ž G 1q Ý
ks1
/
pk , Ž 2.21 .
subject to the n Žprocess-dependent. constraints that ensure that the power of x i on the
lhs Žyn i . is equal to the power of x i on the rhs, which is generally a combination of
integers of the summation.
By adding together the first two expressions in Eq. Ž2.17., we obtain an additional
constraint, that is
D
p 1 q . . . qpn q q1 q . . . qq q s y . Ž 2.22 .
2
Equating Eqs. Ž2.20. and Ž2.21., we obtain an expression for the loop integral with
negative powers of the propagators in negative dimensions
`
q1 qq m1 mn
InD Ž n i 4 ; Q i2 4 , Mi2 4 . ' Ý Ž Q12 . . . . Ž Q q2 . Ž yM12 . . . . Ž yMn2 .
p1 , . . . , p n s0
q1 , . . . , q q s0
m1 , . . . , m n s0
n G Ž1 y ni .
=
ž Ł
is1 G Ž 1 q m i . G Ž 1 q pi . /
q n
1
=
žŁ is1 G Ž 1 q qi . /žG 1q Ý
ks1
/
pk . Ž 2.23 .
Eq. Ž2.23., together with the constraints, is the main result of this paper. The loop
integral is written directly as an infinite sum. Given that Q can be read off directly from
the Feynman graph, so can the precise form of Eq. Ž2.23. as well as the system of
constraints. Of course, strictly speaking we have assumed that both n i and Dr2 are
negative integers and we must be careful in interpreting this result in the physically
interesting domain where the n i and D are all positive. However, this is relatively
straightforward and in the following sections we show how quite general results for
one-loop massive bubbles and triangles can be obtained.
Example 1. To give an explicit example of how Eq. Ž2.23. and the system of constraints
appear, we consider the one-loop bubble with different masses. Eqs. Ž2.14. and Ž2.18.
become
I2D Ž n 1 , n 2 ;Q12 , M12 , M22 .

` x 1n1 x 2n 2
s DxH Ý I2D Ž yn1 ,y n 2 ;Q12 , M12 , M22 .n
n1 , n 2s0 1! n2 !
q1 m1 m2
`
Ž x 1 x 2 Q12 . x 1p 1 x 2p 2 Ž yx 1 M12 . Ž yx 2 M22 .
s DxH Ý
p1 , p 2 , q 1 , m 1 , m 2s0 q1 ! p 1 ! p 2 ! m1 ! m 2 !
= Ž p 1 q p 2 . !, Ž 2.24 .
n1 n2
so that, by selecting powers of xy
1 and xy
2 , we find Žsee Eq. Ž2.23..
I2D Ž n 1 , n 2 ;Q12 , M12 , M22 .

`
q1 m1 m2
s Ý Ž Q12 . Ž yM12 . Ž yM22 .
p1 , p 2 , q 1 , m 1 , m 2s0
G Ž 1 y n 1 . G Ž 1 y n 2 . G Ž 1 q p1 q p 2 .
= , Ž 2.25 .
G Ž 1 q m1 . G Ž 1 q m 2 . G Ž 1 q p 1 . G Ž 1 q p 2 . G Ž 1 q q1 .
together with the system of constraints
q1 q p 1 q m1 s yn 1 ,
q1 q p 2 q m 2 s yn 2 ,
D
q1 q p 1 q p 2 s y . Ž 2.26 .
2
In Section 2.4, we will show how this particular system can be solved to give results for
the bubble integral in positive dimensions D, with arbitrary positive powers of the
propagators.
2.3. The general form of the solutions
In general, for an n-point one-loop integral with q external momentum scales and m
mass scales, there will be Ž n q q q m. summation variables and Ž n q 1. constraints.
Altogether we expect Ž n q q q m.!rŽ n q 1.!rŽ q q m y 1.! possible solutions Žsome of
which will be eliminated by the specific form of the system of constraints.. It is easy to
see that these solutions span physically different kinematic regions Ždepending on the
powers of the kinematic scales. and the summations will only converge in the appropri-
ate kinematic domain. We expect that solutions in one kinematic region should be
analytically linked to those in other domains.
Each solution of the system of constraints, once inserted into the template of Eq.
Ž2.23., has the following generic form:
P RE = S U M , Ž 2.27 .
where we have introduced the following notation:
– S U M is the sum over the terms that contain unconstrained indices of

summation. Instead of dealing with G functions, we have formed Pochhammer
symbols, defined as
G Ž z q n.
Ž z ,n . ' , Ž 2.28 .
G Ž z.
because they are the most suitable way to write generalized hypergeometric
functions. For example, in the case where there is only one remaining summa-
tion variable n, then S U M takes the form
` Ž a1 ,n . . . . Ž a N ,n . xn
S U M; Ý , Ž 2.29 .
ns0 Ž b 1 ,n . . . . Ž bNy1 ,n . n!
where x is the ratio of kinematic scales. The variables a i and bi are linear in
the n i and D and do not depend on the summation variables. To put S U M in
this form, it is often convenient to use the identity Žsee Eq. Ž2.32..
n 1
Ž z ,y n . s Ž y1. . Ž 2.30 .
Ž 1 y z ,n .
In most cases, S U M can be directly identified as a generalized hypergeomet-
ric function, in the region of convergence of the series. In general, these
hypergeometric functions are analytic and may be evaluated at positive values
of D and n i .
– The prefactor P R E contains all the rest of the terms that are not included in
S U M . More precisely, it is a product of external scales raised to fixed powers,
and G functions that do not depend on the summation variables. These may be
produced either directly from the particular solution of the system, or in the
generation of the Pochhammer symbols.
In the general case of an n-point one-loop integral with q external momentum
scales and m mass scales, inspection of Eq. Ž2.23. dictates that we produce:
– n G functions with argument Ž1 y n i .,

– Ž n q m q q . factorials of the summation variables in the denominator,
– one G function in the numerator: G Ž 1 q Ý nks 1 p k . .
Applying the Ž n q 1. constraints leaves Ž m q q y 1. factorials of the remaining uncon-

strained summation variables and produces an additional Ž n q 1. G functions in the
denominator and one in the numerator, as Pochhammer symbols are formed using Eq.
Ž2.28.. Altogether there will be Ž n q 2. Pochhammer symbols in S U M , while P R E
will be a ratio with Ž n q 1. G functions in both numerator and denominator. In both
S U M and P R E , the number of functions may be reduced if there are cancellations

between numerator and denominator.
For physical loop integrals with positive powers of propagators, we need to evaluate
P R E at positive values of the n i and positive D. A problem is immediately obvious:
the numerator of P R E contains G Ž1 y n i ., so that, for positive integer values n i , it
appears that we need to evaluate the G functions for negative arguments, where they are
singular. However, P R E is an analytic function and these singularities cancel between
the numerator and denominator.
In fact, it can be easily shown that, starting from the identity
G Ž z q 1. s z G Ž z . , Ž 2.31 .
we have
GŽ z . yn GŽ n q 1 y z .
s Ž y1 . , Ž 2.32 .
GŽ z y n. GŽ 1 y z .
where z is a real Žor complex. number, and n is a positive integer.
In the product of G functions in the numerator and denominator of the P R E term,
we can make an iterated use of the identity Ž2.32., provided we treat Dr2 as an integer,
as we have already done in the multinomial expansion. We can then rewrite the
G-function prefactor in a more amenable way by flipping all of the G functions from
numerator to denominator and vice versa
nq 1
nq1 G Ž ai . Ý Ž b y a . nq1
i i
G Ž 1 y bi .
is 1
Ł s Ž y1 . Ł , Ž 2.33 .
is1 G Ž b i . is1 G Ž 1 y a i .
where the index i runs over all Ž n q 1. G functions in the numerator and denominator
of P R E . In addition, it can be shown that
nq1 D
Ý Ž bi y a i . s 2 , Ž 2.34 .
is1
which is independent of the n i .
2.3.1. An example: the massless bubble

Returning to the example of the one-loop self-energy diagram introduced in Section
2.2, and setting the masses of the internal lines to zero, M1 s M2 s 0 Žwhich is
equivalent to terminating the series in m1 and m 2 at the first term., we obtain the
simpler system of constraints Žsee Eq. Ž2.26. with m1 s m 2 s 0.
q1 q p 1 s yn 1 ,
q1 q p 2 s yn 2 ,
D
q1 q p 1 q p 2 s y . Ž 2.35 .
2
Since m s 0, q s 1 and n s 2, we expect that the Ž n q 1. s 3 constraints exactly

determine the Ž m q q q n. s 3 variables. In this case S U M s 1 and the result is
entirely given by the prefactor P R E . Solving this system yields
D
q1 s yn1 yn2 ,
2
D
p1 s n 2 y ,
2
D
p2 s n 1 y .
2
Inserting these values directly into Eq. Ž2.25. with m1 s m 2 s 0 we find
I2D Ž n 1 , n 2 ;Q12 ,0,0 . s P R E
GŽ 1 y n 1 . GŽ 1 y n 2 . GŽ 1 q n 1 q n 2 y D .
s
D D D
ž
G 1qn1 y
2
G 1qn2y /ž2 /ž
G 1q yn1 yn2
2 /
D
y n 1y n 2
2
=Ž Q12 . . Ž 2.36 .
As expected, there are Ž n q 1. G functions in both numerator and denominator and
furthermore the arguments satisfy Eq. Ž2.34.. We therefore apply Eq. Ž2.33. and find
I2D Ž n 1 , n 2 ;Q12 ,0,0 .

D D D
s Ž y1 .
D
2
G ž 2
yn1 G /ž 2 /ž
yn2 G n1 qn2 y
2 /Ž Q12 .
D
2
y n 1y n 2
, Ž 2.37 .
GŽ n 1 . GŽ n 2 . GŽ D y n 1 y n 2 .
where n 1 and n 2 are positive and which agrees with the known result straightforwardly
obtained using Feynman parameters.
2.4. MassiÕe bubble integrals
We want now to give a detailed description of how to build the solutions starting
from the general form Ž2.23. for the loop integral and from the system of constraints,
and we want to discuss how the solutions of the system of constraints need to be
combined to give a meaningful result.
We will refer to a precise example to make things clearer: the bubble integral with
different masses in the propagators of Eq. Ž2.25.. Particular cases with n 1 s n 2 s 1 are
important in electroweak renormalization and have been known for some time Žsee for
example Ref. w3x.. The more general cases with n 1 / n 2 / 1 have been studied by Boos
and Davydychev w20–22x using the Mellin–Barnes integral representations.
As discussed at the end of Section 2.2, in the case where the two masses are non-zero
and different, there are Ž m q n q q . s 5 summation variables: two for the propagator
masses Ž m1 and m 2 ., two for the expansion of P Ž p 1 and p 2 . and one for the external
momentum scale Ž q1 ..
The Ž n q 1. s 3 constraints are given in Eq. Ž2.26.. There are a maximum of
5!r3!r2!s 10 possible solutions, one for each of the ways in which we can choose
three variables among the five, and solve the system with respect to these triplets. In this
case, there is no solution if we try to solve the system for p 1 ,q1 ,m 2 4 or p 2 ,q1 ,m14 , so
that we have only eight solutions.
Each of the eight solutions corresponds to different values of the integer summation
variables and we insert each of them into the general expression for the propagator
integral, Eq. Ž2.25.. For example, solving for p 1 , p 2 ,q14 , yields
D
p1 s n 2 q m 2 y ,
2
D
p 2 s n 1 q m1 y ,
2
D
q1 s y n 1 y n 2 y m1 y m 2 , Ž 2.38 .
2
and the contribution of this solution to the integral Ž2.25. is
D
` y n 1y n 2 y m 1y m 2
2 m1 m2 G Ž 1y n 1 . G Ž 1y n 2 .
I2m 1 , m 24 s Ý Ž Q12 . Ž y M12 . Ž y M22 .
m1 , m 2 s 0
G Ž 1q m1 . G Ž 1q m 2 .
G Ž 1q n 1 q n 2 y Dq m1 q m 2 .
= D D D
,
ž
G 1q n 2 y
2
q m2 / ž / ž
G 1q n 1 y q m1 G 1q y n 1 y n 2 y m1 y m 2
2 2 /
Ž 2.39 .
where we have labelled the integral with respect to the indices of summation and we
have dropped the functional dependence of I2D , for ease of notation.
As discussed in Section 2.2, we now form the Pochhammer symbols, and we make
use of the Eq. Ž2.30. to flip the Pochhammer symbol in the denominator with negative
indices of summation, to obtain
D
2
y n 1y n 2 GŽ 1 y n 1 . GŽ 1 y n 2 . GŽ 1 q n 1 q n 2 y D .
I2m 1 , m 2 4 s Q12
Ž . D D D
ž
G 1qn2y
2
G 1qn1 y /ž 2
G 1q yn1 yn2
2 /ž /
D
= Ý
` Ž 1 q n 1 q n 2 y D,m1 q m 2 . n 1 q n 2 y ž 2
,m1 q m 2 /
D D
m1 , m 2s0
ž 1qn2y
2
,m 2 /ž 1qn1 y
2
,m1 /
m1 m2
Ž M12rQ12 . Ž M22rQ12 .
= , Ž 2.40 .
m1 ! m2 !
so that we can recognize the general form of Eq. Ž2.27.: the first line of the rhs is P R E
while the second is S U M . By flipping the G functions in the prefactor term P R E ,
using Eq. Ž2.33., we get
I2m 1 , m 2 4
D D D
s Ž y1 .
D
2
Ž Q12 .
D
2
y n 1y n 2 G ž 2
yn1 G /ž
2 /ž
yn2 G n1 qn2 y
2 /
GŽ n 1 . GŽ n 2 . GŽ D y n 1 y n 2 .
D D D M12 M22
ž
=F4 1 q n 1 q n 2 y D, n 1 q n 2 y
2
,1 q n 1 y
2
,1 q n 2 y , ,
2 Q12 Q12
,
/
Ž 2.41 .
where we have used the definition of Appell’s F4 function given in Eq. ŽA.6..
In the same way, we can obtain the other seven solutions:
D
D
2
I2 p 1 , m 14 s Ž y1 . Ž Q12 .
y n 1
Ž yM22 .
D
2
yn 2 G n 2 y
2 ž /
GŽ n 2 .
D D D M12 M22
ž
=F4 1 q n 1 y
2
, n 1 ,1 q n 1 y
2
,1 q
D
2
yn2 , ,
Q12 Q12
,
/
I2 p 2 , m 2 4 s Ž y1 .
D
2
Ž Q12 .
yn 2
Ž yM12 .
D
2
yn 1 G n1 y ž 2 /
GŽ n 1 .
D D D M12 M22
ž
=F4 1 q n 2 y
2
, n 2 ,1 q
2
y n 1 ,1 q n 2 y , ,
2 Q12 Q12
,
/
D D D
D yn 1 yn 2
2 y 2 2
I2 p 1 , p 2 4 s Ž y1. Q12
Ž . 2
Ž yM12 . Ž yM22 .
D D D
=
ž
G n1 y
2 /žG n2y
2 /ž /G
2
G Ž n 1 . G Ž n 2 . G Ž 0.
D D D M12 M22
=F4 1,
ž 2
,1 q
2
y n 1 ,1 q
2
yn2 , ,
Q12 Q12
D
,
D
/
I2q1 , m 2 4 s Ž y1 .
D
2
Ž yM12 .
D
2
y n 1y n 2 ž
G n1 qn2 y
2 /ž G
2
yn2 /
D
GŽ n 1 . G ž / 2
D D D Q12 M22
ž
=F4 n 1 q n 2 y
2
,n 2 ,
2
,1 q n 2 y
2
,
M12
,
M12 / ,
I2 p 1 , q14 s Ž y1 .
D
2
Ž yM12 .
yn 1
Ž yM22 .
D
2
yn 2 ž
G n2y
2 /
GŽ n 2 .
D D D Q12 M22
=F4 n 1 ,
ž , ,1 q y n 2 , 2 , 2 ,
2 2 2 M1 M1 /
D D
I2q1 , m 14 s Ž y1.
D
2
Ž yM22 .
D
2
y n 1y n 2 ž
G n1 qn2 y
2 /ž
G
2
yn1 /
D
GŽ n 2 . G ž / 2
D D D M12 Q12
ž
=F4 n 1 q n 2 y
2
, n 1 ,1 q n 1 y , , ,
2 2 M22 M22
,
/
D
I2 p 2 , q14 s Ž y1 .
D
2
Ž yM12 .
D
2
yn 1
Ž yM22 .
yn 2 ž
G n1 y
2 /
GŽ n 1 .
D D D M12 Q12
=F4 n 2 ,
ž 2
,1 q
2
yn1, , ,
2 M22 M22
.
/ Ž 2.42 .
Solution I2 p 1 , p 2 4 deserves a comment. Before any flipping of the G functions

between numerator and denominator, the prefactor P R E , has a G Ž 1 . in the numerator,
due to the fact that the Appell’s F4 function has its first argument equal to 1. Flipping
this G Ž 1 . according to Eq. Ž2.33. generates a G Ž 0 . in the denominator, so that this
solution is to be considered to be equal to 0.
2.4.1. Identification of the groups of solutions using the conÕergence regions

The Appell’s F4 Ž a , b ,g ,g X , x, y ., defined in Eq. ŽA.6., is convergent only if Žsee
Table 1.
<'x < q < y < - 1,

' Ž 2.43 .
so that we can form three different groups, according to the kinematic region of
convergence of the series
I2D Ž n 1 , n 2 ;Q12 , M12 , M22 . s I2m 1 , m 24 q I2 p 2 , m 24 q I2 p 1 , m 14 ( ( (

if M12 q M22 - Q12 ,
I2D Ž n 1 , n 2 ;Q12 , M12 , M22 . s I2q 1 , m 24 q I2 p 1 , q 14 if (Q q(M -(M

2
1
2
2
2
1 , Ž 2.44 .
I2D Ž n 1 , n 2 ;Q12 , M12 , M22 . s I2q 1 , m 14 q I2 p 2 , q 14 if (Q q(M -(M
2
1
2
1
2
2 .
We note that from the convergence properties of the F4 , if it was not eliminated by the
zero in the prefactor, I2 p 1 , p 2 4 would belong to the first kinematic region, M12 (
q (
M22 - ( Q12 .
In this way, using NDIM we have simultaneously obtained all the different forms of
the hypergeometric functions that express the integral I2D for different kinematic regions
of M12 , M22 and Q12 . These results for I2D agree with those obtained with the
Mellin–Barnes method w20–22x.
It must be noted that we can go from one kinematic region to the other, just by
applying formula ŽA.51. of the analytic continuation of the F4 function. As stated at the
end of the previous section, the appearance of G Ž 0 . in the denominator, that occurs
during the process of analytic continuation, just kills that term.
2.4.2. Identification of the groups of solutions using the system of constraints

We would like to address here a different method to form the groups of solutions.
This is based only on considerations of the system of constraints, and more precisely on
the sign of the summed indices of the series pi ,qi ,m i 4 , without any knowledge of the
region of convergence of the specific series.
For this purpose, the actual value of n i and of D in the system Ž2.26. is irrelevant,
because it only modifies the sign of a finite number of the summation variables of the
series. For example, solution Ž2.38., obtained solving the system with respect of the two
indices m1 ,m 2 4 , tells us that the ‘‘bulk’’ sign of p 1 is equal to that of m 2 , because for
m 2 sufficiently large, the contribution of n 2 y D2 is no longer important. The same thing
happens for p 2 and q1 , whose ‘‘bulk’’ sign is equal to that of m1 and ym1 y m 2 ,
respectively.
For this reason, instead of considering the full inhomogeneous system, we consider
the homogeneous one, obtained by setting n i s 0 and D s 0. The system Ž2.26. for the
massive bubble then becomes
q1 q p 1 q m1 s 0,
q1 q p 2 q m 2 s 0,
q1 q p 1 q p 2 s 0. Ž 2.45 .
We would like to stress the fact that the last equation has always the same form, since
this is the constraint expressed by Eq. Ž2.22..
We can now build a table of signs for pi , qi and m i . The last equation gives rise to
one of the following cases:
)0 -0
p1 , p 2 q1
p 1 , q1 p2
p 2 , q1 p1
p1 p 2 , q1
p2 p 1 , q1
q1 p1 , q2
Using the last equation of Ž2.45. to eliminate q1 from the other two equations of the
system Ž2.45., we have
m1 s p 2 , m 2 s p1 , Ž 2.46 .
so that we can complete the previous table in the following way:
)0 -0
p 1 , p 2 , m1 , m 2 q1
p 1 , q1 , m 2 p 2 , m1
p 2 , q1 , m1 p1 , m 2
where the last three lines have been neglected, being equal and opposite to the first three
ones.
We know that the summation indices of the series must be positive integers and we
can therefore imagine solving the system Ž2.26. with respect to any pair of variables that
are simultaneously positive. The table provides us with this information and we can
directly read from the table which integers are simultaneously positive and use them to
form a group of solutions with similar properties by selecting all possible pairs of
summation indices from the list of positive indices.
Starting from the first row of the table, and considering the identities Ž2.46., that
embody the fact that we cannot solve the system with respect to the pairs of indices
p 1 ,m 2 4 and p 2 ,m14 , because they are linearly dependent, we can form the following
subgroups:
p 1 , p 2 , m1 , m 2 ´ p 1 , p 2 4 , p 1 ,m1 4 , p 2 ,m 2 4 , m1 ,m 2 4 . Ž 2.47 .
The same thing can be done with the other two rows of the table:
p 1 , q1 , m 2 ´ p 1 ,q1 4 , q1 ,m 2 4 , Ž 2.48 .
p 2 , q1 , m1 ´ p 2 ,q1 4 , q1 ,m1 4 . Ž 2.49 .
These are exactly the groups obtained by adding solutions according to their region of
convergence Žsee Eq. Ž2.44.., once we consider the fact that I2 p 1 , p 2 4 s 0.
This method gives the correct groups only for the cases where the homogeneous
system can be solved without any ambiguity. There are examples, and we will meet one
in Section 3.2.2, where the sign of some indices of the series are undetermined, because
they have a dependence on other indices of the type
p1 s p 2 q p 3 ,
where p 2 and p 3 have opposite sign. In this case, we cannot say if p 1 is positive or
negative.
Up to now, we do not have a way to deal with these cases directly from the system of
constraints, and we leave the task of further investigating this issue to future works.
2.4.3. The limiting case: M1 / 0, M2 s 0

We conclude this section by considering some extreme cases. First we consider the
limit of one massless propagator in the self-energy diagram. We can compute this
integral in two different ways.
Ž1. We can start with the system Ž2.26. with m 2 s 0: we have four variables and
three constraints, so that we end up with a single-index series, that turns out to
be a Gauss’ hypergeometric 2 F1 function Žsee Eq. ŽA.1...
Ž2. We can simply take the limit for M2 ™ 0 of the general expressions Ž2.41. and
Ž2.42.. We can apply this limit only to the solutions that are convergent in the
new kinematic regions, and we cannot take the limit for solutions I2q1 , m 14 and
I2 p 2 , q14 because they are defined only for Q12 q M12 - M22 . ( ( (
The expression for F4 Ž a , b ,g ,g X , x,0. is easily obtained from its definition ŽA.6. with the
second summation series collapsing to its first term
F4 Ž a , b ,g ,g X , x ,0 . s2 F1 Ž a , b ,g , x . . Ž 2.50 .
Both procedures give the same result.

If M12 - Q12
I2D Ž n 1 , n 2 ;Q12 , M12 ,0 .
D D
y n 1y n 2
2 2
s I2m 14 q I2 p 2 4s Ž y1. Ž . Q12
D D D
=
ž
G n1 qn2 y
2 2 /ž G
2
yn1 G /ž yn2 /
GŽ n 1 . GŽ n 2 . GŽ D y n 1 y n 2 .
D D M12
ž
= 2 F1 1 q n 1 q n 2 y D, n 1 q n 2 y
2
,1 q n 1 y ,
2 Q12 /
D
q Ž y1 .
D
2
Ž Q12 .
yn 2
Ž yM12 .
D
2
yn 1 ž
G n1 y
2 /
GŽ n 1 .
D D M12
ž
= 2 F1 n 2 ,1 q n 2 y
2
,1 q
2
yn1,
Q12 / ; Ž 2.51 .
if Q12 - M12
I2D Ž n 1 , n 2 ;Q12 , M12 ,0 .

D D
s I2q14s Ž y1 .
D
2
Ž yM12 .
D
2
y n 1y n 2 ž
G n1 qn2 y
2 /ž
G
2
yn2 /
D
GŽ n 1 . G ž /
2
D D Q12
ž
= 2 F1 n 1 q n 2 y
2
,n 2 ,
2
,
M12 / . Ž 2.52 .
2.4.4. The limiting case: Q12 ™ 0

Similarly, we can take the limit of the general self-energy diagram where the external
momentum scale vanishes. Once again, we can either return to the system Ž2.26. with
one fewer variable Ž q1 s 0. or we just take the Q12 ™ 0 limit of the general result Ž2.44.
(
in the appropriate kinematic regions: Q12 q M12 - M22 or Q12 q M22 - M12 . ( ( ( ( (
Both procedures yield the same result:
If M1 ) M2 :
I2D Ž n 1 , n 2 ;0, M12 , M22 .

D D
s I2m 2 4 q I3 p 14s Ž y1 .

D
2
Ž yM12 .
D
2
y n 1y n 2 ž
G n1 qn2 y
2 /žG
2
yn2 /
D
GŽ n 1 . G ž / 2
D D M22
ž
= 2 F1 n 2 , n 1 q n 2 y
2
,1 q n 2 y ,
2 M12 /
D
q Ž y1 .
D
2
Ž yM12 .
yn 1
Ž yM22 .
D
2
yn 2 ž
G n2y
2 /
GŽ n 2 .
D D M22
ž
= 2 F1 n 1 ,
2
,1 q
2
yn2 ,
M12 / , Ž 2.53 .
with the result for M2 ) M1 obtained by the exchanges M1 l M2 and n 1 l n 2 .

Provided that we do not violate the validity of the kinematic regions, we can take the
subsequent limits of the energy scales. For example, we can safely take the M2 ™ 0
limit for the solution where M2 - M1. In this case, only the first term survives and we
obtain the familiar result
D D
I2D Ž n 1 , n 2 ;0, M12 ,0 . s Ž y1. Ž

D
2
yM12 .
D
2
y n 1y n 2 ž
G n1 qn2 y
2 /ž G
2
yn2 / .
D
GŽ n 1 . G ž / 2
Ž 2.54 .
2.5. The special case: all masses equal
For the special case where each propagator has the same mass, Eq. Ž2.10. becomes
n
Ms Ý xi M 2 s PM 2 , Ž 2.55 .
is1
and we have an important simplification. As before, we expand the exponentials in Eq.

Ž2.8. and make multinomial expansions of Q and P
D m
` ` Q n Pyny 2 qm Ž yM 2 .
InD Q i2 2
Ž ni4 ; 4 , M . s H D x Ý Ý
ns0 ms0 n! m!
m
` Q1q1 . . . Qqq q x 1p 1 . . . x np n Ž yM 2 .
s HDx Ý q 1 ! . . . q q ! p 1 ! . . . pn ! m!
p1 , . . . , p n s0
q1 , . . . , q q s0
ms0
= Ž p 1 q . . . qpn . !, Ž 2.56 .
subject to the constraints
q n D
Ý qi s n and Ý pi s yn y 2
q m. Ž 2.57 .
is1 is1
ni
Equating Eqs. Ž2.14. and Ž2.56. and, once again, identifying powers of xy
i , we obtain
an expression for the loop integral with negative powers of the propagators in negative
dimensions for all masses equal
`
q1 qq m
InD Ž n i 4 ; Q i2 4 , M 2 . s Ý Ž Q12 . . . . Ž Q q2 . Ž yM 2 .
p1 , . . . , p n s0
q1 , . . . , q q s0
ms0
n G Ž1 y ni . q
1 ž
G 1q
ks1
Ý pk
/
=
ž Ł
is1 G Ž 1 q pi . /ž Ł
is1 G Ž 1 q q i . / GŽ 1 q m.
,
Ž 2.58 .
subject to n constraints that each of the powers of x i match up correctly. However, the
constraint that matches up the powers of Q and P, obtained by summing the two
expressions in Eq. Ž2.57., is now
D
p 1 q . . . qpn q q1 q . . . qq q s y q m, Ž 2.59 .
2
rather than Eq. Ž2.22..
We see that there are Ž n q q q 1. summation variables and Ž n q 1. constraints,
leaving q remaining summations. We note that the structure of the solution is precisely
as for the unequal-mass case and is treated in the same way by constructing the sum
over Pochhammer symbols S U M and the G function prefactor P R E .
Example 2. to give an explicit example, we consider the self-energy correction to the

propagator integral with equal masses. There are Ž n q q q 1. s 4 summation variables
with Ž n q 1. s 3 constraints. In this case, the template solution Eq. Ž2.58. is given by
I2D Ž n 1 , n 2 ;Q12 , M 2 , M 2 .
` G Ž 1 y n 1 . G Ž 1 y n 2 . G Ž 1 q p1 q p 2 .
q1 m
s Ý Ž Q12 . Ž yM 2 . G Ž 1 q p 1 . G Ž 1 q p 2 . G Ž 1 q q1 . G Ž 1 q m .
,
p1 , p 2 , q 1 , ms0
Ž 2.60 .
while, matching the powers of x i gives the system of constraints
q1 q p 1 s yn 1 ,
q1 q p 2 s yn 2 ,
D
q1 q p 1 q p 2 s y q m. Ž 2.61 .
2
There are four summation variables Ž p 1 , p 2 , q1 and m. and three constraints, and we
obtain four series solutions, with only one index of summation.
Defining s s n 1 q n 2 Žsee Eq. Ž2.6.., we have:
If Q12 ) 4 M 2
I2D Ž n 1 , n 2 ;Q12 , M 2 , M 2 .
D D
y n 1y n 2
2 2
s I2m4 q I2 p 14 q I2 p 2 4s Ž y1 . Ž Q12 .
D D D
=
ž
G sy
2 /ž G
2
yn1 G
2 /ž
yn2 /
GŽ n 1 . GŽ n 2 . GŽ D y n 1 y n 2 .
s D 1 s D D D D 4M 2
= 3 F2 1 q
ž 2
y , q y , s y ,1 q n 1 y ,1 q n 2 y , 2
2 2 2 2 2 2 2 Q1 /
D
q Ž y1 .
D
2
Ž Q12 .
yn 1
Ž yM 2 .
D
2
yn 2 ž
G n2y
2 /
GŽ n 2 .
n1 n2 1 n1 n2 D 4M 2
ž
= 3 F2 n 1 ,1 q
2
y , q
2 2 2
y
2
,1 q n 1 y n 2 ,1 q y n 2 , 2
2 Q1 /
D
q Ž y1 .
D
2
Ž Q12 .
yn 2
Ž yM 2 .
D
2
yn 1 ž
G n1 y
2 /
GŽ n 1 .
n1 1 n2 n2n1 D 4M 2
ž
= 3 F2 n 2 ,1 q
2
y , q
2 2 2
y
2
,1 q n 2 y n 1 ,1 q y n 1 , 2 ;
2 Q1 /
Ž 2.62 .
if Q12 - 4 M 2
D
I2D Ž n 1 , n 2 ;Q12 , M 2 , M 2 . s I2q14s Ž y1 .

D
2
Ž yM 2 .
D
2
yn 1 y n 2 ž
G sy
2 /
GŽ s .
D s 1 q s Q12
ž
= 3 F2 n 1 , n 2 , s y , ,
2 2 2
,
4M 2
.
/ Ž 2.63 .
In forming the Pochhammer symbols we have made use of the following duplication
formula:
z z 1
Ž z ,2 n . s 4 n ,nž /ž q ,n . / Ž 2.64 .
2 2 2
The procedure described in Section 2.4.2 on how the solutions group together to give the
correct answer in a particular kinematic region, is straightforward. In fact, the solution of
the homogeneous counterpart of the system of constraints Ž2.61. is
p 1 s p 2 s m s yq1 . Ž 2.65 .
We can then form groups for the indices of summation of the series out of combinations
of indices with the same sign, that is
p1 4 , p 2 4 , m4 or q1 4 . Ž 2.66 .
Not surprisingly these were the groups of solutions formed by considering the conver-
gence properties in Eqs. Ž2.62. and Ž2.63. and reproduce the known result w20–22x.
3. Massive vertex integrals
We now turn to massive-triangle integrals where each propagator can have a different
mass and each external leg can be off-shell. The propagators and momenta are labelled
as in Fig. 3. Throughout this section, the number of propagators n is equal to three and
in the most general case, we have
P s x1 q x 2 q x 3 ,
Q s x 2 x 3 Q12 q x 3 x 1 Q22 q x 1 x 2 Q 32 ,
M s x 1 M12 q x 2 M22 q x 3 M32 , Ž 3.1 .
where Q i2 s k i2 .
Fig. 3. The one-loop vertex diagram.

Based on the discussion of Section 2.2, the NDIM method provides a generic solution
with Ž n q q q m. s 9 summation variables and Ž n q 1. s 4 constraints, with the tem-
plate solution given by
I3D Ž n 1 , n 2 , n 3 ;Q12 ,Q22 ,Q 32 , M12 , M22 , M32 .

`
q1 q2 q3 m1 m2 m3
' Ý Ž Q12 . Ž Q22 . Ž Q32 . Ž yM12 . Ž yM22 . Ž yM32 .
p1 , . . . , p 3s0
q1 , . . . , q 3s0
m1 , . . . , m 3s0
3 G Ž1 y ni .
=
ž Ł
is1 G Ž 1 q m i . G Ž 1 q pi . G Ž 1 q q i . / G Ž 1 q p1 q p 2 q p 3 . , Ž 3.2 .
while the system of four constraints is

q2 q q3 q p 1 q m1 s yn 1 ,
q1 q q3 q p 2 q m 2 s yn 2 ,
q1 q q2 q p 3 q m 3 s yn 3 ,
D
p 1 q p 2 q p 3 q q1 q q 2 q q 3 s y . Ž 3.3 .
2
The number of possible solutions satisfying this system is 9!r4!r5!s 126 of which 45
are eliminated by the particular nature of the system leaving 81. As usual, insertion of
these solutions into the template yields contributions of the general form Ž2.27., where
S U M is a product of Pochhammer symbols and ratios of energy scales summed over
the five remaining variables. The prefactor P R E vanishes in a further 12 instances,
leaving 69 solutions which are distributed among the various kinematic regions.
At present, the technology for dealing with five-fold sums Žand their integral
representations. is not sufficiently developed to handle the completely general case. For
the remainder of this section, we therefore concentrate on particular cases of the vertex
integral where some of the energy scales vanish, leading to either single or double sums
which have been well studied.
3.1. Massless propagators: M1 s M2 s M3 s 0
We first consider the special case where all of the internal lines are massless, so
m1 s m 2 s m 3 s 0 in Eq. Ž3.3., leaving six summation variables. Of the 6!r4!r2!s 15
possible solutions of this system, three are eliminated by the system, leaving twelve. The
three-mass triangle is an extremely symmetric system and there are three allowed phase
space regions:
region I: (Q ) (Q q (Q
2
1
2
2
2
3 ,
region II: (Q ) (Q q (Q
2
2
2
1
2
3 , Ž 3.4 .
region III: (Q ) (Q q (Q
2
3
2
1
2
2 ,
with each region bounded by the external phase-space constraints

D3 Ž Q12 ,Q22 ,Q32 . ) 0, Ž 3.5 .
where
D3 Ž x , y, z . s x 2 q y 2 q z 2 y 2 xy y 2 yz y 2 zx. Ž 3.6 .
The twelve solutions populate the three kinematic regions equally, four in each, and are
easily identified as belonging to a particular region by studying either the convergence
properties of the double sum or by considering the system, as in Section 2.4.2. For
example, the four solutions belonging to region I Ž q1 negative, q2 and q3 positive. are
those where the summation variables include pairs in the set p 2 , p 3 ,q2 ,q34 , that is
p 2 , p 34 , p 2 ,q34 , q2 , p 34 and q2 ,q34 , where p 2 ,q2 4 and p 3 ,q34 have been eliminated
by the system.
As usual, each solution is inserted into Eq. Ž3.2. and treated according to the
procedure described in Section 2.2: the summation variables are converted into
Pochhammer symbols; the G-function prefactor is flipped using Eq. Ž2.33. and the
remaining summations converted into generalised hypergeometric functions. In each
case, we identify Appell’s F4 Ž a , b ,g ,g X , x , y . function Žsee Eq. ŽA.6.., which, accord-
ing to the convergence criteria of Table 1, is well defined when 'x q y - 1, precisely '
matching on to the physically allowed phase space.
Summing the four solutions we find, in the region Q12 ) Q 22 q Q 32 , ( ( (
I3D Ž n 1 , n 2 , n 3 ;Q12 ,Q22 ,Q 32 ,0,0,0 .
D D
y n 1y n 2y n 3
2 2
s I3q 2 , q 34 q I3 p 2 , q 34 q I3 p 3 , q 2 4 q I3 p 2 , p 34s Ž y1. Q12
Ž .
D D D
=
G ž 2
yn1 yn2 G /ž
2
yn1 yn3 G sy /ž 2 /
GŽ n 2 . GŽ n 3 . GŽ D y s .
D D Q22 Q32 D
ž
=F4 n 1 , s y ,1 q n 1 q n 3 y ,1 q n 1 q n 2 y , 2 , 2
2 2 2 Q1 Q1 /
D D
y n 1y n 3
2 yn 2 2
q Ž y1 . Q12
Ž . Q22
Ž .
D D D
=
G ž 2
yn1 yn2 G n1 qn3 y /ž 2 /ž
G
2
yn3 /
GŽ n 1 . GŽ n 3 . GŽ D y s .
D D D Q22 Q 32
=F4 n 2 ,
ž 2
y n 3 ,1 q
2
y n 1 y n 3 ,1 q n 1 q n 2 y , ,
2 Q12 Q12 /
D D
y n 1y n 2
2 yn 3 2
q Ž y1 . Q12
Ž . Q32
Ž .
D D D
=
G ž 2
yn1 yn3 G n1 qn2 y /ž 2 /žG
2
yn2 /
GŽ n 1 . GŽ n 2 . GŽ D y s .
D D D Q 22 Q32
=F4 n 3 ,
ž 2
y n 2 ,1 q n 1 q n 3 y
2
,1 q
2
yn1yn2 , ,
Q12 Q12 /
D D D
D y n 1y n 3 y n 1y n 2
2 n 1y 2 2
q Ž y1 . Q12
Ž . 2 Q 22
Ž . Q 32
Ž .
D D D
=
ž
G n1 qn2 y
2 /žG n1 qn3 y
2 /žG
2
yn1 /
GŽ n 1 . GŽ n 2 . GŽ n 3 .
D D Q22 Q 32 D
ž
=F4 D y s , y n 1 ,1 q y n 1 y n 3 ,1 q y n 1 y n 2 , 2 , 2 ,
2 2 2 Q1 Q1 /
Ž 3.7 .
which agrees with that obtained by Boos and Davydychev w20–22x using the Mellin–
Barnes integral representation. Similar results are obtained for the other two kinematic
regions, either by directly summing the solutions valid in that region Žpairs from
p 1 , p 3 ,q1 ,q34 or p 1 , p 2 ,q1 ,q2 4 , with p 1 ,q14 , p 2 ,q2 4 and p 3 ,q34 excluded by the
system. or by analytic continuation of the Appell’s F4 function using formula ŽA.51..
Note that if one of the n i vanishes Žequivalent to propagator i shrinking to a point.,
only a single term remains. For example, if n 1 s 0, only the first term of Eq. Ž3.7.
survives Žthe others being killed by 1rG Ž 0 . ., and the Appell’s function collapses to
F4 Ž 0, b ,g ,g X , x , y . s 1,
as can be seen from the definition ŽA.6., yielding
I3D Ž 0, n 2 , n 3 ;Q12 ,Q22 ,Q 32 ,0,0,0 . s I2D Ž n 2 , n 3 ;Q12 ,0,0 . , Ž 3.8 .

as it should.
We can obtain some other interesting limits if we set to zero one or two external
invariants.
Ž1. One light-like external momentum: if the ith external leg is light-like Ž Q i2 s 0.,
we can return to the general case and solve the system Ž3.3. with qi s 0, or we
can take the appropriate limit of the general solution. These limits can be safely
made provided that we start from a valid kinematic region. In the region of
( (
validity of Eq. Ž3.7., that is Q12 ) Q 22 q Q 32 , we can surely take the limits(
for Q22 ™ 0 or Q 32 ™ 0. In this last case, for example, the last two terms in Eq.
Ž3.7. vanish, while the first two terms collapse to Gaussian hypergeometric
functions, according to Eq. ŽA.20., yielding, in the region Q12 ) Q 22 ,
I3D Ž n 1 , n 2 , n 3 ;Q12 ,Q22 ,0,0,0,0 .
D D
y n 1y n 2y n 3
2 2
s I3q 2 4 q I3 p 2 4s Ž y1. Q12
Ž .
D D D
=
G ž 2
yn1 yn2 G /ž 2 /ž
yn1 yn3 G n1 qn2 qn3 y
2 /
GŽ n 2 . GŽ n 3 . GŽ D y n 1 y n 2 y n 3 .
D D Q 22
ž
=2 F1 n 1 , s y
2
,1 q n 1 q n 3 y ,
2 Q12 /
D D
y n 1y n 3
2 yn 2 2
q Ž y1 . Q12
Ž . Q22
Ž .
D D D
=
G ž 2
yn1 yn2 G /ž 2 /ž
yn3 G n1 qn3y
2 /
G Ž n1. G Ž n3. G Ž Dyn1 yn2 yn3.
D D Q22
=2 F1 n 2 ,
ž 2
y n 3 ,1 q
2
yn1yn3,
Q12 / . Ž 3.9 .
Analogous results valid in the region Q 22 ) Q12 can be obtained either by

starting from the expression for I3D in region II, that is Q22 ) Q12 q Q32 , or ( ( (
via analytic continuation of Eq. Ž3.9., according to Eq. ŽA.49..
Ž2. Two light-like external momenta: in a similar way, we can obtain the result for
two light-like external momenta, Q 32 s Q 22 s 0 for example, by simultaneously
taking both Q 22 and Q32 ™ 0 in Eq. Ž3.7.. Only the first term in Eq. Ž3.7.
survives, and the Appell’s function collapses to
F4 Ž a , b ,g ,g X ,0,0 . s 1, Ž 3.10 .
yielding
I3D Ž n 1 , n 2 , n 3 ;Q12 ,0,0,0,0,0 .

D D D
s Ž y1 .
D
2
Ž Q12 .
D
2
ys G ž 2
yn1yn2 G /ž 2
yn1yn3 G sy /ž 2 / ,
GŽ n 2 . GŽ n 3 . GŽ D y s .
Ž 3.11 .
which again agrees with the known result straightforwardly obtained using
Feynman parameters. Alternatively, we could have returned to the general
system Ž3.3., where, with only the Q12 scale Ž q2 s q3 s m1 s m 2 s m 3 s 0., we

would have had Ž n q q q m. s 4 summation variables and Ž n q 1. s 4 con-
straints.
3.2. Two massiÕe propagators and one off-shell leg
We now turn to triangle integrals with two internal mass scales and one external
scale. These have Ž n q q q m. s 6 summation variables and Ž n q 1. s 4 constraints and
are therefore described by double sums.
3.2.1. M1 s 0, Q22 s Q 32 s 0
In this case, the system of constraints is obtained by setting m1 s q2 s q3 s 0 in Eq.
Ž3.3.. The first constraint is simply p 1 s yn 1 and there are only 8 solutions for the
system. As usual, each solution is inserted into Eq. Ž3.2. and treated accordingly to the
procedure of Section 2.2.
The solutions can be grouped either by studying the physical thresholds of the
integral Žor the convergence properties of the series. or by considering the system, as in
Section 2.4.2.
The threshold for the production of two massive propagators on-shell, Q12 s M2 q (
M3 , becomes evident when we inspect the hypergeometric functions: four solutions are
( ( (
convergent above threshold, Q12 ) M22 q M32 , while the other solutions equally
populate the regions ( ) (M M32 2
2 q (Q 2
1 and (M ) (M
q Q12 . 2
2
2
3 (
Consideration of the system reveals that the first group of solutions are pairs from the
set m 2 ,m 3 , p 2 , p 34 and the other groups formed are from the sets q1 , p 3 ,m 2 4 and
q1 , p 2 ,m 3 4 respectively. The apparent overlap between the groups, solutions formed
from the pairs m 2 , p 3 4 and m 3 , p 2 4 are excluded by the system. In each case, we
identify Appell’s F4 function, whose convergence properties match onto the anticipated
regions.
We find:
( (
If Q12 ) M22 q M32 (
I3D Ž n 1 , n 2 , n 3 ;Q12 ,0,0,0, M22 , M32 .
D D
y n 1y n 2y n 3
2 2
s I3m 2 , m 34 q I3m 2 , p 2 4 q I3m 3 , p 34 q I3 p 2 , p 34s Ž y1 . Ž Q12 .
D D D
=
ž
G n1 qn2 qn3 y
2 /ž
G
2
yn1 yn2 G
2 /ž yn1 yn3 /
D D M22 M32 D
ž
=F4 1 q s y D, s y ,1 q n 1 q n 2 y ,1 q n 1 q n 3 y , 2 , 2
2 2 2 Q1 Q1 /
D D
q Ž y1 .
D
2
Ž Q12 .
yn 2
Ž yM32 .
D
2
y n 1y n 3 ž
G n1 qn3 y
2 /ž
G
2
yn1 yn2 /
D
GŽ n 3 . G ž /2
yn2
D D D M22 M32
ž
=F4 1 q n 2 y
2
, n 2 ,1 q n 1 q n 2 y
2
,1 q
2
yn1 yn3 , ,
Q12 Q12 /
D D
q Ž y1 .
D
2
Ž Q12 .
yn 3
Ž yM22 .
D
2
y n 1y n 2 ž
G n1 qn2 y
2 /ž
G
2
yn1 yn3 /
D
GŽ n 2 . G ž /2
yn3
D D D M22 M32
ž
=F4 1 q n 3 y
2
, n 3 ,1 q
2
y n 1 y n 2 ,1 q n 1 q n 3 y , ,
2 Q12 Q12 /
D D D
D y n 1y n 2 y n 1y n 3
2 n 1y 2 2
q Ž y1 . Q12
Ž . 2
Ž yM22 . Ž yM32 .
D D D
=
ž
G n1 qn2 y
2 /ž G n1 qn3y
2 /žG
2
yn1 /
GŽ n 1 . GŽ n 2 . GŽ n 3 .
D D D M22 M32
ž
=F4 1 y n 1 ,
2
y n 1 ,1 q
2
y n 1 y n 2 ,1 q
2
yn1 yn3 , ,
Q12 Q12
;
/
Ž 3.12 .
( (
if M32 ) Q12 q M22 (
I3D Ž n 1 , n 2 , n 3 ;Q12 ,0,0,0, M22 , M32 .
D D
y n 1y n 2y n 3
2 2
s I3m 2 , q14 q I3 p 3 , q14s Ž y1. Ž yM32 .
D D
=
ž
G n1 qn2 qn3 y
2 /ž G
2
yn1 yn2 /
D
GŽ n 3 . G ž / 2
D D D M22 Q12
ž
=F4 n 1 q n 2 q n 3 y
2
, n 2 ,1 q n 1 q n 2 y , , ,
2 2 M32 M32 /
D D
q Ž y1 .
D
2
Ž yM22 .
D
2
y n 1y n 2
Ž yM32 .
yn 3 ž
G n1 qn2 y
2 /ž
G
2
yn1 /
D
GŽ n 2 . G ž /
2
D D D M22 Q12
=F4
ž 2
y n 1 , n 3 ,1 q
2
yn1 yn2 , , ,
2 M32 M32
,
/ Ž 3.13 .
( ( (
while the result for M22 ) M32 q Q12 is obtained by the exchanges M2 l M3 ,
n 2 l n 3 in Eq. Ž3.13..
We can check that these expressions are valid in certain limits.
Checks
– The n 1 ™ 0 limit: I3D ( 0,n 2 ,n 3 ;Q12 ,0,0,0,M22 ,M32 )

Pinching out the first propagator, the first three terms in Ž3.12. and both terms
in Ž3.13. survive, yielding the general bubble integral of Eq. Ž2.44., in the
respective kinematic regions,
I3D Ž 0, n 2 , n 3 ;Q12 ,0,0,0, M22 , M32 . s I2D Ž n 2 , n 3 ;Q12 , M22 , M32 . . Ž 3.14 .
– The n 2 ™ 0 limit: ( I3D )
n 1 ,0,n 3 ;Q12 ,0,0,0,M22 ,M32
This limit should produce a self-energy integral with no external momentum
and a single internal mass M3 . This is indeed the case: only the second term in
Ž3.12. and the first term of Ž3.13. survive, each yielding the same result of Eq.
Ž2.54..
– The M2 ™ 0 limit: I3D ( n 1 ,n 2 ,n 3 ;Q12 ,0,0,0,0,M32 )
Here the real production threshold occurs at Q12 s M32 and in this limit, Eq.
Ž3.12. provides the Q12 ) M32 result, while Eq. Ž3.13. gives the expression for
M32 ) Q12 . The Appell functions again collapse to form Gaussian hypergeomet-
ric functions Žsee Eq. ŽA.21.., and we find:
(
If Q12 ) M32 (
I3D Ž n 1 , n 2 , n 3 ;Q12 ,0,0,0,0, M32 .
D D
y n 1y n 2y n 3
2 2
s I3m 34 q I3 p 34s Ž y1. Q12
Ž .
D D D
=
G syž 2 /žG
2
yn1 yn2 G
2 /ž yn1yn3 /
GŽ n 2 . GŽ n 3 . GŽ D y s .
D D M32
ž
= 2 F1 1 q s y D, s y
2
,1 q n 1 q n 3 y ,
2 Q12 /
D D
y n 1y n 3
2 yn 2 2
q Ž y1 . Q12
Ž . Ž yM32 .
D D
=
ž
G n1 qn3 y
2 /ž G
2
yn1 yn2 /
D
GŽ n 3 . G ž / 2
yn2
D D M32
ž
= 2 F1 n 2 ,1 q n 2 y
2
,1 q
2
yn1 yn3 ,
Q12 / ; Ž 3.15 .
(
if M32 ) Q12(
I3D Ž n 1 , n 2 , n 3 ;Q12 ,0,0,0,0, M32 .

D D
s I3q14s Ž y1.
D
2
Ž yM32 .
D
2
y n 1y n 2y n 3 G sy ž 2 /ž G
2
yn1 yn2 /
D
GŽ n 3 . G ž / 2
D D Q12
ž
= 2 F1 n 2 , s y , , 2 .
2 2 M3 / Ž 3.16 .
This latter result agrees with that obtained by taking the limit Q22 ™ 0, Q32 ™ 0
in the general result given by Boos and Davydychev w20–22x for a triangle loop
integral with a single massive propagator.
3.2.2. M3 s 0, Q22 s Q 32 s 0
We now consider the triangle graph where M3 s 0 and Q22 s Q 32 s 0. Although this
graph is not usually present in Standard Model processes, the analysis of this graph turns
out to be rather more subtle than the preceding triangle integrals and we will therefore
describe it in more detail. Inspection of the singularities present in the loop integral via
the Landau equations reveals that threshold singularities occur at M22 s Q12 q M12 . We
expect that this equality will provide the appropriate boundaries of regions of conver-
gence when considering the convergence properties of the generalised hypergeometric
functions. Furthermore, since the convergence properties of these functions only depend
on the absolute value of ratios of scales, we expect that the reflections, M12 q M22 s Q12
and M12 s M22 q Q12 , will also form boundaries in the large Q12 and M12 regions,
respectively. We also expect that, in certain limits, the solutions match onto the
kinematic regions relevant for simpler integrals. For example, in the limit M1 ™ 0, the
discussion of the previous section informs us that the solutions divide according to
whether or not Q12 ) M22 . Similarly as Q12 ™ 0, there should be a threshold at M1 s M2 .
Fig. 4. The kinematic regions for the one-loop triangle with Q22 s Q32 s M32 s 0. The solid line shows the
threshold in the Landau surface at M22 s Q12 q M12 , together with the reflections M12 q M22 s Q12 and
M12 s M22 q Q12 . The reflections are relevant for the convergence properties of the hypergeometric functions
which only involve the absolute values of ratios of the scales. The dashed lines show the boundaries M12 s M22
and M22 s Q12 .
In anticipation, we therefore divide the kinematic regions up as follows:

region I: M22 ) Q12 q M12 ,
region II Ž a . : Q12 ) M12 q M22 and M12 ) M22 ,
region II Ž b . : Q12 ) M12 q M22 and M22 ) M12 , Ž 3.17 .
region III Ž a . : M12 ) Q12 q M22 and Q12 ) M22 ,
region III Ž b . : M12 ) Q12 q M22 and M22 ) Q12 ,
as shown in Fig. 4.
With this set of scales, the system is given by Eq. Ž3.3. with m 3 s q2 s q3 s 0. As
usual, we construct the solutions by solving the system, inserting the solutions into Eq.
Ž3.2. and following the procedure outlined in Section 2.2. Labelling each solution by the
summation variables and using the definitions of the hypergeometric functions of
Section A.1, we find
D D
I3m 1 , q14 s Ž y1 .
D
2
Ž yM22 .
D
2
y n 1y n 2y n 3 ž
G n1 qn2 qn3 y
2 /ž
G
2
yn1yn3 /
D
GŽ n 2 . G ž /
2
D D D M12 Q12
ž
=F2 n 1 q n 2 q n 3 y
2
, n 1 , n 3 ,1 q n 1 q n 3 y , , ,
2 2 M22 M22
,
/
D D
I3 p 2 , q14 s Ž y1 .
D
2
Ž yM12 .
D
2
y n 1y n 3
Ž yM22 .
yn 2 ž
G n1 qn3 y
2 /ž G
2
yn3 /
D
GŽ n 1 . G ž / 2
D D D M12 Q12
ž
=F2 n 2 ,
2
y n 3 , n 3 ,1 q
2
yn1 yn3,
2
,
M22
,
M22 / ,
D D
y n 1y n 2y n 3
2 2
I3m 1 , m 2 4s Ž y1. Q12
Ž .
D D D
=
ž
G n1 qn2 qn3 y
2 2 /ž G
2
yn1 yn2 G /ž yn1 yn3 /
D D D M12 M22
ž
=S1 1 q s y D, s y
2
, n 1 ,1 q n 1 q n 2 y
2
,1 q n 1 q n 3 y
2
,y ,
Q12 Q12
,
/
D D
y n 1y n 3
2 yn 2 2
I3m 2 , p 2 4s Ž y1. Q12
Ž . Ž yM12 .
D D
=
ž
G n1 qn3 y
2 / GŽ n 3 y n 2 . G ž 2
yn3 /
D
GŽ n 1 . GŽ n 3 . G ž 2
yn2 /
D D M12 M22 D
ž
=S1 1 q n 2 y , n 2 , y n 3 ,1 q n 2 y n 3 ,1 q y n 1 y n 3 ,y 2 , 2 ,
2 2 2 Q1 Q1 /
D D
y n 1y n 2
2 yn 3 2
I3m 2 , p 34 s Ž y1. Q12
Ž . Ž yM12 .
D D
=
ž
G n1 qn2 y
2 / GŽ n 2 y n 3 . G ž 2
yn2 /
D
GŽ n 1 . GŽ n 2 . G ž 2
yn3 /
D D D M22 M12
ž
=S2 n 2 y n 3 , n 1 q n 2 y
2
,1 q n 3 y
2
, n 3 ,1 q n 2 y , ,
2 M12 Q12
,
/
D
I3 p 1 , p 34 s Ž y1 .
D
2
Ž Q12 .
yn 3
Ž yM12 .
yn 1
Ž yM22 .
D
2
yn 2 ž
G n2y
2 /
GŽ n 2 .
D D D M22 M22
ž
=F3 n 1 ,1 q n 3 y , y n 3 , n 3 ,1 q y n 2 , 2 , 2 ,
2 2 2 M1 Q1 /
D D
y n 1y n 2
2 yn 3 2
I3m 1 , p 34 s Ž y1. Q12
Ž . Ž yM22 .
D D
=
ž
G n1 qn2 y
2 /ž G
2
yn1 yn3 /
D
GŽ n 2 . G ž / 2
yn3
D D D M12 M22
ž
=H2 n 1 q n 2 y
2
, n 1 ,1 q n 3 y
2
, n 3 ,1 q n 1 q n 3 y , 2 ,y 2 ,
2 M2 Q1 /
D D
y n 1y n 3
2 yn 3 2 n 3y n 2
I3 p 2 , p 34 s Ž y1. Q12
Ž . Ž yM12 . Ž yM22 .
D
=
ž
G n1 qn3 y
2 /
GŽ n 2 y n 3 .
GŽ n 1 . GŽ n 2 .
D D D M12 M22
ž
=H2 n 2 y n 3 ,
2
y n 3 ,1 q n 3 y
2
, n 3 ,1 q
2
yn1 yn3 ,
M22
,y
Q12 / ,
D D
y n 2y n 3
2 2 yn 1
I3m 2 , p 14 s Ž y1 . Ž Q12 . Ž yM12 .
D D D
=
ž
G n2qn3y
2 2 /ž G
2
yn2 G /ž yn3 /
GŽ n 2 . GŽ n 3 . GŽ D y n 2 y n 3 .
D D D M22 Q12
ž
=S2 n 2 q n 3 y
2
,1 q n 2 q n 3 y D, n 1 ,
2
y n 3 ,1 q n 2 y , ,
2 Q12 M12
,
/
D D
I3m 2 , q14 s Ž y1 .
D
2
Ž yM12 .
D
2
y n 1y n 2y n 3 ž
G n1 qn2 qn3 y
2 /ž
G
2
yn2yn3 /
D
GŽ n 1 . G ž / 2
D D D Q12 M22
ž
=S1 n 2 , n 1 q n 2 q n 3 y
2
, n 3 ,1 q n 2 q n 3 y , ,y 2 , 2 ,
2 2 M1 M1 /
D D
I3 p 1 , q14 s Ž y1 .
D
2
Ž yM12 .
yn 1
Ž yM22 .
D
2
y n 2y n 3 ž
G n2qn3y
2 /ž G
2
yn3 /
D
GŽ n 2 . G ž / 2
D D D Q12 M22
ž
=H2 n 2 q n 3 y
2
,n 3 ,n 1 ,
2
yn3, , 2 ,y 2 .
2 M2 M1 / Ž 3.18 .
We now need to study the convergence properties of these solutions. For example, by
inspecting Table 1 in Appendix A.1, we see that the function S1Ž . . . , x, y . is convergent
when < x < q < y < - 1. This implies that solution I3m 1 , m 2 4 is convergent when
yM12 M22
q - 1, Ž 3.19 .
Q12 Q12
or, in other words,
M12 q M22 - Q12 , Ž 3.20 .
independently of whether M1 is larger than M2 or not. This series therefore converges
in both regions IIŽa. and IIŽb..
On the other hand, I3m 1 , p 34 converges when
M12 Q12 M12 M22
y q ) 1 and - 1 and - 1, Ž 3.21 .
M22 M22 M22 Q12
or, alternatively,
M12 q M22 - Q12 and M22 ) M12 , Ž 3.22 .
which corresponds to region IIŽb. only.
Applying the convergence criteria to each of the eleven solutions, we find that they
are distributed as follows:
In region I
I3D Ž n 1 , n 2 , n 3 ;Q12 ,0,0, M12 , M22 ,0 . s I3m 1 , q14 q I3 p 2 , q14 ; Ž 3.23 .
in region IIŽa.
I3D Ž n 1 , n 2 , n 3 ;Q12 ,0,0, M12 , M22 ,0 . s I3m 1 , m 2 4 q I3m 2 , p 2 4 q I3m 2 , p 34 q I3 p 1 , p 34 ; Ž 3.24 .
in region IIŽb.
I3D Ž n 1 , n 2 , n 3 ;Q12 ,0,0, M12 , M22 ,0 . s I3m 1 , m 2 4 q I3m 2 , p 2 4 q I3m 1 , p 34 q I3 p 2 , p 34 ; Ž 3.25 .
in region IIIŽa.
I3D Ž n 1 , n 2 , n 3 ;Q12 ,0,0, M12 , M22 ,0 . s I3m 2 , p 14 q I3m 2 , q14 q I3 p 1 , p 34 ; Ž 3.26 .
in region IIIŽb.
I3D Ž n 1 , n 2 , n 3 ;Q12 ,0,0, M12 , M22 ,0 . s I3m 2 , q14 q I3 p 1 , q14 . Ž 3.27 .
We see that in region IIŽa., two of the solutions Ž I3m 2 , p 2 4 . and Ž I3m 2 , p 34 . contain
dangerous G functions when n 2 s n 3 . These divergences usually indicate the region of
a logarithmic analytic continuation and can be regulated by letting n 2 s n 3 q d ,
cancelling the divergence, and then setting d ™ 0. Similarly, the two divergent contribu-
tions in region IIŽb. Ž I3m 2 , p 2 4 . and Ž I3 p 2 , p 34 . also cancel in this limit.
We have performed several checks of the correctness of this assignment into groups.
Checks
– Analytic continuation
Applying the analytic continuation formulae given in Appendix A, we can see
that the solutions are connected to each other. For example, applying Eq.
ŽA.54. to the F2 functions in region I produces the H2 and S1 functions of

region IIŽb.. Similarly, Eqs. ŽA.56. and ŽA.53. transform the S1 and H2
solutions of region IIIŽb. into the F2 functions of region I.
– The n 1 s n 2 s n 3 s 1 limit: I3D (1,1,1;Q12 ,0,0,M12 ,M22 ,0)
All the groups give the correct answer when all the propagators are set equal to
one.
As an example, we consider region IIŽb., so that we can explicitly show the
cancellation of the d poles. We fix n 1 s n 3 s 1, n 2 s 1 q d and D s 4 y 2 e .
For these choices of the parameters, the hypergeometric functions simplify
using the identities given in Section A.3.2, and we find
2 2e
Ne G Ž 1 y e . Q12 M12
I3m 1 , m 2 4 s y
e 2 GŽ 1 y 2 e . ž Q12 y M22 / ž 2
F1 1,2 e ,1 q e ,
M22 y Q12 / ,
Ž 3.28 .
2e
Ne GŽ 1 y e . GŽ 1 y d . Ž Q12 . ye y d
I3m 2 , p 2 4 s e Ž Q12 q M12 y M22 . ,
e GŽ 1 y e y d . d Ž yM12 .
Ž 3.29 .
e
Ne M22 Q12
I3m 1 , p 34 s
e 2 M22 y M12 ž /
yM22
M12 M22
ž
=F2 1,1, e , e q 1,1 y e ,
M12 y M22 Q12
,
/ , Ž 3.30 .
yd e
Ne Ž y1. Q12 M22 y M12
I3 p 2 , p 34 s y
e d Ž M22 y M12 . d ž / žyM12 2
F1 1, e ,1 y d ,
Q12 / ,
Ž 3.31 .
where we have defined
y1 y e
Ne s G Ž 1 q e . Ž yQ12 . . Ž 3.32 .
Using Eq. ŽA.50., we can rewrite I3 p 2 , p 34 in the following way:
e
Ne Q12 Q12 Q12 q M12 y M22
I3 p 2 , p 34 s y
e2 ž /
yM12 M22 y M12 2 ž
F1 1,1,1 q e ,
M12 y M22 /
yd
Ne G Ž 1 q e q d . Ž y1 . GŽ 1 y d .
y
Ž dqe . GŽ 1 q e . d
2e
Ž Q12 . ye y d
= e Ž Q12 q M12 y M22 . , Ž 3.33 .
Ž yM12 .
where we have safely put d s 0 in the first line, since this is a finite quantity in
d.
We see that the poles in d clearly cancel between I3m 2 , p 2 4 and I3 p 2 , p 34 Žsee Eqs.
Ž3.29. and Ž3.33.., leaving a finite remainder that is straightforwardly obtained
by Taylor expansion about d s 0.
Up to now, we have not required e to be small and expressions Ž3.28. – Ž3.33.
are valid in arbitrary dimension D.
If we make the usual expansion for e ™ 0 we recover the result
I3D Ž 1,1,1;Q12 ,0,0, M12 , M22 ,0 .

Q12 q M12 y M22 M22
s Ne Li 2
ž M12 / ž
y Li 2 1 y
M12 / q OŽ e . , Ž 3.34 .
where we have used a series expansion for the integral representation of the
functions 2 F1 and F2 , given in Eqs. ŽA.26. and ŽA.29., and where Q12 ™ Q12 q
i0, to recover the correct prescription in the Feynman integrals. We describe the
details of the e expansion in Appendix A.2.1. Expression Ž3.34. is finite in e ,
as it should be, having no soft or collinear singularities, despite the fact that the
individual contributions contain poles in e .
– The n 1 ™ 0 limit: I3D ( 0,n 2 ,n 3 ;Q12 ,0,0,M12 ,M22 ,0 )
If we set n 1 ™ 0, we produce a one-mass Ž M2 . bubble integral with external
scale Q12 and internal propagators raised to the powers n 2 and n 3 .
In the different regions Ž3.17. and for the different groups of Eqs. Ž3.23. – Ž3.27.,
we have
I: M22 ) Q12 and M22 ) M12 ´ I3D m 1 , q 14
n 1 s 0 sI3 n 1s 0
II Ž a . : Q12 ) M12 ) M22 ´ I3D n 1 s 0 sI3m 1 , m 24 n 1 s 0qI3 p 1 , p 34 n 1 s 0

II Ž b . : Q12 ) M22 ) M12 ´ I3D m 1 , m 2 4
n 1 s 0 sI3
m 1 , p 34
n 1 s 0 qI3 n 1s 0
III Ž a . : M12 ) Q12 ) M22 ´ I3D n 1 s 0 sI3m 2 , p 14 n 1 s 0 qI3 p 1 , p 34 n 1 s 0

III Ž b . : M12 ) M22 ) Q12 ´ I3D p 1 , q 14
n 1 s 0 sI3 n 1s 0 ,
Ž 3.35 .
where we have used the shorthand notation
I3D D
n 1 s0 s I3 Ž 0,n 2 ,n 3 ;Q12 ,0,0, M12 , M22 ,0 . , Ž 3.36 .
– and where the missing terms have been killed by the G Ž 0 . in the denominator.
It is straightforward to evaluate the different solutions when n 1 s 0. In fact,
taking I3m 1 , q14 as example, we can use the reduction formula
F2 Ž a ,0, b X ,g ,g X , x , y . s2 F1 Ž a , b X ,g X , y . , Ž 3.37 .
to recover
I3D Ž 0, n 2 , n 3 ;Q12 ,0,0, M12 , M22 ,0 . s I2D Ž n 2 , n 3 ;Q12 , M22 ,0 . Ž 3.38 .
in region I, that is Eq. Ž2.52.. The same thing happens to the solution in region
IIIŽb..
The other part of the bubble integral valid when Q12 ) M22 is produced by the
other solutions in the region IIŽa., IIŽb. and IIIŽa., and agrees with Eq. Ž2.51..
– The n 3 ™ ,0 limit: I3D ( n 1 ,n 2 ,0;Q12 ,0,0,M12 ,M22 ,0 )
Likewise, we can set n 3 ™ 0 producing a two-mass bubble Ž M1 and M2 . with
external scale Q12 s 0, for which the result is given in Eq. Ž2.53.. We could
repeat the reasoning made for the previous case, and build a table of surviving
solutions, analogous to Ž3.35..
For example, the two terms in Eq. Ž3.23. collapse to form the correct Gauss’
hypergeometric functions when M22 ) M12 and M22 ) Q12 . Similarly, the result
when Q12 ) M22 ) M12 is produced by the third and fourth term of Eq. Ž3.25.,
for region IIŽb..
– The M1 ™ 0 limit: I3D ( n 1 ,n 2 ,n 3 ;Q12 ,0,0,0,M22 ,0 )
Here we expect to reproduce the result for the triangle integral given in Eqs.
Ž3.15. and Ž3.16., with the exchanges M3 l M2 and n 3 l n 2 . Clearly in
regions IIŽa., IIIŽa. and IIIŽb., it is inappropriate to take this limit, since
M12 ) M22 . In fact, if we just go ahead and apply the limit blindly to the
solutions for regions IIIŽa. and IIIŽb., we just obtain zero.
On the other hand, in regions I and IIŽb. it does make sense to send M1 ™ 0
since M1 is allowed to be the smallest scale present:
I: M22 ) Q12 and M22 ) M12 ´ I3D m 1 , q 14

M 1 s 0 sI3 M 1s 0
II Ž b . : Q12 ) M22 ) M12 ´ I3D M 1 s 0 sI3m 1 , m 2 4 M 1 s 0qI3m 1 , p 34 M 1 s 0 ,
Ž 3.39 .
with the shorthand notation
I3D D
M 1 s0 s I3 Ž n 1 ,n 2 ,n 3 ;Q12 ,0,0,0, M22 ,0 . . Ž 3.40 .
Again th hypergeometric functions collapse to Gauss’ 2 F1 functions Žsee Eqs.
ŽA.17., ŽA.11. and ŽA.23..
F2 Ž a , b , b X ,g ,g X ,0, y . s2 F1 Ž a , b X ,g X , y . ,
S1 Ž a , a X , b ,g , d ,0, y . s2 F1 Ž a , a X ,g , y . ,
H2 Ž a , b ,g , d , e ,0, y . s2 F1 Ž g , d ,1 y a ,y y . ,
and we recover, in region I, the result of Eq. Ž3.16., and in region IIŽb., the
expected result Ž3.15. for M22 - Q12 .
– The M2 ™ 0 limit: I3D Ž n 1 , n 2 , n 3 ;Q12 ,0,0, M12 ,0,0 .
Taking the limit M2 ™ 0 provides us with the integral relevant for the exchange
of a heavy particle in the decay into two light particles.
As usual, we could merely return to the system and, by setting m 2 s 0, solve it
afresh: there are now Ž m q q q n. s 5 variables and still Ž n q 1. s 4 con-
straints leaving five single-sum solutions. However, it is simpler to take the
M2 ™ 0 limit in the appropriate regions: IIŽa. and IIIŽa., as can be seen from
Eq. Ž3.17..
We then obtain
II Ž a . : Q12 ) M12 ) M22 ´ I3D M2s 0 s I3m 1 , m 2 4 q I3m 2 , p 24 q I3m 2 , p 34 M2s 0
III Ž a . : M12 ) Q12 ) M22 ´ I3D M 2 s 0 s I3m 2 , p 14 q I3m 2 , q 14 M2s 0,
Ž 3.41 .
where
I3D D
M 2 s0 s I3 Ž n 1 ,n 2 ,n 3 ;Q12 ,0,0, M12 ,0,0 . . Ž 3.42 .
The hypergeometric functions collapse to 3 F2 functions, according to Eqs.
ŽA.10. and ŽA.13.
S1 Ž a , a X , b ,g , d , x ,0 . s3 F2 Ž a , a X , b ,g , d , x . ,
S2 Ž a , a X , b , b X ,g ,0, y . s3 F2 Ž 1 y g , b , b X ,1 y a ,1 y a X ,y y . ,
and we obtain:
If Q12 ) M12 , region IIŽa.
I3D Ž n 1 , n 2 , n 3 ;Q12 ,0,0, M12 ,0,0 .

D D
y n 1y n 2y n 3
2 2
s I3m 14 q I3 p 2 4 q I3 p 34 s Ž y1. Q12
Ž .
D D D
=
ž
G n1 qn2 qn3 y
2 2/ž G yn1 yn2 G
2 /ž yn1 yn3 /
D D
ž
=3 F2 n 1 ,1 q s y D, s y
2
,1 q n 1 q n 2 y
2
,
D D
D M12 2 yn 2 2
y n 1y n 3
1qn1 qn3 y
2
,y
Q12 / q Ž y1 . Ž Q12 . Ž yM12 .
D D
=
GŽ n 3 y n 2 . G ž 2 /ž
yn3 G n1qn3 y
2 /
D
GŽ n 1 . GŽ n 3 . G ž / 2
yn2
D D
=3 F2 n 2 , ž 2
y n 3 ,1 q n 2 y
2
,1 q n 2 y n 3 ,
D D
D M12 2 yn 3 2
y n 1y n 2
1q
2
y n 1 y n 3 ,y
Q12 / q Ž y1 . Q12
Ž . Ž yM12 .
D D
=
GŽ n 2 y n 3 . G ž 2 /ž
yn2 G n1qn2 y
2 /
D
GŽ n 1 . GŽ n 2 . G ž / 2
yn3
D D
=3 F2 n 3 ,ž 2
y n 2 ,1 q n 3 y
2
,1 q n 3 y n 2 ,
D M12
1q
2
y n 1 y n 2 ,y
Q12 / ; Ž 3.43 .
if M12 ) Q12 , region IIIŽa.
I3D Ž n 1 , n 2 , n 3 ;Q12 ,0,0, M12 ,0,0 .

D D
y n 1y n 2y n 3
2 2
s I3q14 q I3 p 14 s Ž y1. Ž yM12 .
D D
=
ž
G n1 qn2 qn3 y
2 /ž G
2
yn2yn3 /
D
GŽ n 1 . G ž / 2
D D D Q12
ž
= 3 F2 n 2 , n 3 , n 1 q n 2 q n 3 y
D
2
,1 q n 3 q n 2 y , ,y 2
2 2 M1 /
D
y n 2y n 3
2 2 yn 1
q Ž y1 . Q12
Ž . Ž yM12 .
D D D
=
ž
G n2qn3y
2 /ž2
G
2
yn2 G /ž yn3 /
GŽ n 2 . GŽ n 3 . GŽ D y n 2 y n 3 .
D D D Q12
ž
= 3 F2 n 1 ,
2
yn2 ,
2
y n 3 , D y n 2 y n 3 ,1 q
2
y n 2 y n 3 ,y
M12 / .
Ž 3.44 .
We can further check these results by setting one of the n i ™ 0 to form bubble
integrals or by taking one of the limits M1 ™ 0 or Q12 ™ 0. In each case, we recover the
correct results presented in the earlier sections
Discussion of system of constraints

As we have already anticipated in Section 2.4.2, the homogeneous counterpart of the
system of constraints Ž3.3., with q2 s q3 s m 3 s 0, is not uniquely solvable. We can
build the table of signs for pi , qi and m i :
)0 -0 uncertain
p 2 , q1 , m1 p1 , p 3 , m 2
p 2 , p 3 , m1 p 1 , q1 m2
p 1 , q1 p 2 , p 3 , m1 m2
p1 , p 3 , m 2 p 2 , q1 , m1
but this does not determine the complete list of groups, even if some of them Žfirst and
second row, corresponding to region I and region IIŽb.. are correctly predicted by the
table Žif we arbitrarily assume that m 2 in the second row has a positive sign..
3.3. Equal-mass propagators: M1 s M2 s M3 s M
In the special case of M1 s M2 s M3 s M we use the modified form given in Section

2.5. For the most general case of unequal off-shell external legs, there are Ž n q q q 1. s 7
summation variables and Ž n q 1. s 4 constraints. The template solution is easily ob-
tained from Eq. Ž2.58. and the system of constraints is given by
q2 q q3 q p 1 s yn 1 ,
q1 q q3 q p 2 s yn 2 ,
q1 q q2 q p 3 s yn 3 ,
D
p 1 q p 2 q p 3 q q1 q q 2 q q 3 s y q m. Ž 3.45 .
2
3.3.1. Q 22 s Q32 s 0
In this case, there are five summation variables Ž p 1 , p 2 , p 3 , q1 and m., and four
constraints, and the system admits four solutions out of the possible five.
Based on two-particle cuts of the diagram, we see that there is a threshold at
Q12 s 4 M 2 corresponding to producing propagators 2 and 3 on-shell. Of the four
solutions, three converge when Q12 ) 4 M 2 , while the remaining solution converges
when Q12 - 4 M 2 . Recalling s s n 1 q n 2 q n 3 Žsee Eq. Ž2.6.., we have:
If Q12 ) 4 M 2
I3D Ž n 1 , n 2 , n 3 ;Q12 ,0,0, M 2 , M 2 , M 2 .
D D
ys
2 2
s I3m4 q I3 p 2 4 q I3 p 34s Ž y1. Ž . Q12
D D D
=
ž
G sy
2 /ž G
2
yn1 yn2 G
2 /ž yn1 yn3 /
GŽ n 2 . GŽ n 3 . GŽ D y s .
D 1 1 D
= 3 F2 s y
ž 2
,1 q
2
Ž s y D . , Ž 1 q s y D . ,1 q n 1 q n 2 y
2 2
,
D 4M 2
1qn1 qn3 y , 2
2 Q1 /
D
q Ž y1 .
D
2
Ž Q12 .
yn 2
Ž yM 2 .
D
2
y n 1y n 3 ž
GŽ n 3 y n 2 . G n 1 q n 3 y
2 /
GŽ n 3 . GŽ n 1 y n 2 q n 3 .
1 1
ž
= 3 F2 n 2 ,1 q
2
Ž n 2 y n 1 y n 3 . , Ž 1 y n 1 q n 2 y n 3 . ,1 q n 2 y n 3 ,
2
D D
D 4M 2 2 yn 3 2
y n 1y n 2
1q
2
yn1 yn3 ,
Q12 / q Ž y1 . Ž Q12 . Ž yM 2 .
D
ž
GŽ n 2 y n 3 . G n 1 q n 2 y
2 / 1
=
GŽ n 2 . GŽ n 1 q n 2 y n 3 . 3 F2
ž n 3 ,1 q
2
Ž n3 yn1yn2 . ,
1 D 4M 2
2
Ž 1 y n 1 q n 3 y n 2 . ,1 q n 3 y n 2 ,1 q
2
yn1yn2 ,
Q12 / ; Ž 3.46 .
if Q12 - 4 M 2
I3D Ž n 1 , n 2 , n 3 ;Q12 ,0,0, M 2 , M 2 , M 2 .

D
s I3q14s Ž y1 .
D
2
Ž yM 2 .
D
2
ys ž
G sy
2 /
GŽ s .
D s 1 q s Q12
ž
= 3 F2 n 2 , n 3 , s y , ,
2 2 2
,
4M 2
,
/ Ž 3.47 .
where we have made use of formula Ž2.64.. This latter result agrees with that obtained
by taking the limit Q22 ™ 0, Q32 ™ 0 in the general result given by Boos and Davydychev
w20–22x for a triangle loop integral with three off-shell legs and a single mass M
running round the loop. Eq. Ž3.46. appears to be a new result.
We note that, taking the limit n 1 ™ 0 in Eqs Ž3.46. and Ž3.47., we reproduce the
expected equal-mass bubble integral of Eqs. Ž2.62. and Ž2.63.
I3D Ž 0, n 2 , n 3 ;Q12 ,0,0, M 2 , M 2 , M 2 . s I2D Ž n 2 , n 3 ;Q12 , M 2 , M 2 . , Ž 3.48 .
while taking M ™ 0, only the first term in Eq. Ž3.46. survives, yielding Eq. Ž3.11..
However, we observe that there are dangerous G functions in the second and third
lines when n 2 s n 3 . Therefore, to evaluate the integral when n 2 s n 3 s n , we introduce
an additional regulator d such that n 2 s n q d and n 3 s n . As in the previous section,
because the result does not depend on d , the limit d ™ 0 can be safely taken after the
singularities have been cancelled.
4. Conclusions
Finally let us summarize what we have accomplished in this paper. Changing the
number of dimensions D to evaluate loop integrals is well established and relies on the
analytic properties of loop integrals. We have used this property to extend the number of
dimensions to negative values as suggested by Halliday and Ricotta. As discussed at
length in Section 2.2, treating D as a negative even integer allows a multinomial
expansion of the integrand in intermediate steps and, by expanding before and after loop
integration we can identify the loop integral as an infinite series, together with
constraints on the summation variables. The number of summation parameters is equal
to the number of legs n plus the number of energy scales Ž m q q . in the loop, while
there are n q 1 constraints, which can be read off from the Feynman graph. The form of
the series is specified for arbitrary one-loop integrals and forms a template series into
which specific solutions of the system of constraints are inserted. In this way, integration
over the parameters is replaced with infinite sums. In each case we immediately identify
generalised hypergeometric functions and show how to assemble the complete result
valid in a particular kinematic region by considering the convergence properties of the
hypergeometric functions. The procedure is as follows:
Ž1. Write down P, M and Q of Eqs. Ž2.9., Ž2.10. and Ž2.11. by inspection of the
graph.
Ž2. Write down the template solution for the particular process. In other words,
copy out Eq. Ž2.23. inserting the correct mass scales and number of terms in
P, M and Q.
Ž3. Construct the system of constraints by counting powers of x i in Eqs. Ž2.20. and
Ž2.21..
Ž4. Solve the system of constraints and insert each solution into the template
solution, one at a time.
Ž5. Construct Pochhammer symbols and identify the generalised hypergeometric
function.
Ž6. Flip all of the G functions in prefactor according to Eq. Ž2.33..
Ž7. Group the solutions according to their regions of convergence.
Ž8. Evaluate the hypergeometric functions for the specific parameters of interest.
Steps Ž1. – Ž6. are very straightforward and easily achieved with a computer program.
Step Ž7. requires a little more thought, though for the cases we have studied here the
convergence regions were easy to identify. To make the procedure useful for phe-
nomenological studies, it is necessary to evaluate the hypergeometric functions for
specific values of the parameters. In particular, an integral representation is required.
The mathematical literature for Euler integral representations of hypergeometric func-
tions with more than two or three variables is quite sparse, and it may be necessary to
select a complex integral representation for more complicated functions.
More interesting is the application of NDIM to integrals with more than one loop.
Suzuki and Schmidt w38–43x have made some steps in this direction, although the
integrals they have considered are largely of the one-loop insertion type. Given that
quite powerful results for one-loop integrals are achieved so easily and generally, we
expect that NDIM can play a role in simplifying the task of calculating two Žor more.
loop integrals that are necessary to make more precise perturbative predictions within
the Standard Model.
Acknowledgements
We thank J.V. Armitage for stimulating discussions and insights into the field of
hypergeometric functions and P. Watson and M. Zimmer for useful conversations. We
acknowledge the assistance of D. Broadhurst and A. Davydychev regarding generalised
hypergeometric functions. C.A. acknowledges the financial support of the Greek govern-
ment and C.O. acknowledges the financial support of the INFN.
Appendix A. Hypergeometric definitions and identities
The purpose of this appendix is to give sufficient information to evaluate the general
loop integrals presented in Sections 2.2 and 3. In Section A.1 we give the definitions of
the hypergeometric functions as a series together with their regions of convergence.
Integral representations are provided in Section A.2 together with a description of how
to evaluate the integrals in the general case. For specific choices of the n i the general
hypergeometric functions often simplify and some useful identities and analytic-con-
tinuation formulae are collected in Section A.3.
A.1. Series representations
The hypergeometric functions of one variable are sums of Pochhammer symbols over
a single summation parameter m, like, for example,
` Ž a ,m . Ž b ,m . x m
2 F1 Ž a , b ,g , x . s Ý , Ž A.1 .
ms0 Ž g ,m . m!
X X
` Ž a ,m . Ž b ,m . Ž b X ,m . x m
3 F2 Ž a , b , b ,g ,g , x . s Ý , Ž A.2 .
ms0 Ž g ,m . Ž g X ,m . m!
which are convergent when < x < - 1.
We also meet hypergeometric functions of two variables which can be written as
sums over the integers m and n: Fi , i s 1, . . . ,4 are the Appell functions, H2 a Horn
function and S1 and S2 generalised Kampe´ de Feriet´ functions:
X
` Ž a ,m q n . Ž b ,m . Ž b X ,n . x m y n
F1 Ž a , b , b ,g , x , y . s Ý , Ž A.3 .
m, ns0 Ž g ,m q n . m! n!
X X
` Ž a ,m q n . Ž b ,m . Ž b X ,n . x m y n
F2 Ž a , b , b ,g ,g , x , y . s Ý , Ž A.4 .
m, ns0 Ž g ,m . Ž g X ,n . m! n!
X X
` Ž a ,m . Ž a X ,n . Ž b ,m . Ž b X ,n . x m y n
F3 Ž a , a , b , b ,g , x , y . s Ý , Ž A.5 .
m, ns0 Ž g ,m q n . m! n!
X
` Ž a ,m q n . Ž b ,m q n . x m y n
F4 Ž a , b ,g ,g , x , y . s Ý , Ž A.6 .
m, ns0 Ž g ,m . Ž g X ,n . m! n!
Table 1
Convergence regions for some hypergeometric functions of two variables
Function Convergence criteria
F1 , F3 < x < -1, < y < -1
F2 , S1 < x <q < y < -1
F4 '< x < q'< y < -1
H 2 , S2 y < x <q1r < y < )1, < x < -1, < y < -1
` Ž a ,m y n . Ž b ,m . Ž g ,n . Ž g X ,n . x m y n
H2 Ž a , b ,g ,g X , d , x , y . s Ý ,
m, ns0 Ž d ,m . m! n!
Ž A.7 .
X
` Ž a ,m q n . Ž a ,m q n . Ž b ,m . x m y n
S1 Ž a , a X , b ,g , d , x , y . s Ý , Ž A.8 .
m, ns0 Ž g ,m q n . Ž d ,m . m! n!
` Ž a ,m y n . Ž a X ,m y n . Ž b ,n . Ž b X ,n . x m y n
S2 Ž a , a X , b , b X ,g , x , y . s Ý .
m, ns0 Ž g ,m y n . m! n!
Ž A.9 .
These series converge according to the criteria collected in Table 1, The domain of
convergence of the Appell and Horn functions are well known. That one for S1 and S2
may be worked out using Horns general theory of convergence w48x.
When one of the arguments vanishes, then the hypergeometric function collapses in a
straightforward way. For example, if y s 0 in Eq. ŽA.8., then only the first term of the
series in n contributes and we are left with the relation
S1 Ž a , a X , b ,g , d , x ,0 . s3 F2 Ž a , a X , b ,g , d , x . . Ž A.10 .
Similarly, we have
S1 Ž a , a X , b ,g , d ,0, y . s2 F1 Ž a , a X ,g , y . , Ž A.11 .
X X X
S2 Ž a , a , b ,g ,g , x ,0 . s2 F1 Ž a , a ,g , x . , Ž A.12 .
S2 Ž a , a X , b , b X ,g ,0, y . s3 F2 Ž 1 y g , b , b X ,1 y a ,1 y a X ,y y . , Ž A.13 .
F1 Ž a , b , b X ,g , x ,0 . s2 F1 Ž a , b ,g , x . , Ž A.14 .
F1 Ž a , b , b X ,g ,0, y . s2 F1 Ž a , b X ,g , y . , Ž A.15 .
F2 Ž a , b , b X ,g ,g X , x ,0 . s2 F1 Ž a , b ,g , x . , Ž A.16 .
F2 Ž a , b , b X ,g ,g X ,0, y . s2 F1 Ž a , b X ,g X , y . , Ž A.17 .
F3 Ž a , a X , b , b X ,g , x ,0 . s2 F1 Ž a , b ,g , x . , Ž A.18 .
F3 Ž a , a X , b , b X ,g ,0, y . s2 F1 Ž a X , b X ,g , y . , Ž A.19 .
F4 Ž a , b ,g ,g X , x ,0 . s2 F1 Ž a , b ,g , x . , Ž A.20 .
F4 Ž a , b ,g ,g X ,0, y . s2 F1 Ž a , b ,g X , y . , Ž A.21 .
H2 Ž a , b ,g ,g X , d , x ,0 . s2 F1 Ž a , b , d , x . , Ž A.22 .
H2 Ž a , b ,g ,g X , d ,0, y . s2 F1 Ž g ,g X ,1 y a ,y y . . Ž A.23 .
When one of the parameters vanishes producing a Pochhammer Ž0,n., then the series
in n also terminates. If we have Ž0,m q n. then both series terminate. For example
F2 Ž a ,0, b X ,g ,g X , x , y . s2 F1 Ž a , b X ,g X , y . , Ž A.24 .
X
F1 Ž 0, b , b ,g , x , y . s 1. Ž A.25 .
A.2. Integral representations
Euler integral representations of the hypergeometric series of one and two variables
are well known w49,50x and we list them here. We know of no integral representation for
the H2 function and for the closely related S2 function.
GŽ g . 1 b y1 g y b y1 ya
2 F1 Ž a , b ,g , x . s = H0 du u Ž 1 y u. Ž 1 y ux . ,
GŽ b . GŽ g y b .
R Ž b . ) 0, R Ž g y b . ) 0; Ž A.26 .
GŽ d . GŽ e . 1 1 by1 g y1
3 F2 Ž a , b ,g , d , e , x . s H duH0 dÕ u Õ
GŽ b . GŽ d y b . GŽ g . GŽ e y g . 0
d y b y1 e y g y1 ya
=Ž 1 y u. Ž1yÕ. Ž 1 y uÕx . ,
R Ž b . ) 0, R Ž d y b . ) 0, R Ž g . ) 0, R Ž e y g . ) 0; Ž A.27 .
GŽ g . 1 gy a y1
F1 Ž a , b , b X ,g , x , y . s H du u a y1
Ž 1 y u.
GŽ a . GŽ g y a . 0
X
yb yb
= Ž 1 y ux . Ž 1 y uy . ,
R Ž a . ) 0, R Ž g y a . ) 0; Ž A.28 .
F2 Ž a , b , b X ,g ,g X , x , y .
GŽ g . GŽ g X .
s
GŽ b . GŽ b X . GŽ g y b . GŽ g X y b X .
1 1 b y1 X g y b y1 g Xy b Xy1 ya
= H0 duH0 dÕ u Õ b y1 Ž 1 y u . Ž1yÕ. Ž 1 y ux y Õy . ,
R Ž b . ) 0, R Ž b X . ) 0, R Ž g y b . ) 0, R Ž g X y b X . ) 0; Ž A.29 .
F3 Ž a , a X , b , b X ,g , x , y .
GŽ g . uqÕ(1 by1 X
s X HHu00, Õ00du dÕ u
X Õ b y1
GŽ b . GŽ b . GŽ g y b y b .
X
gy b y b Xy1 ya ya
=Ž 1 y u y Õ . Ž 1 y ux . Ž 1 y Õy . ,
R Ž b . ) 0, R Ž b X . ) 0, R Ž g y b y b X . ) 0; Ž A.30 .
F4 Ž a , b ,g ,g X , x Ž 1 y y . , y Ž 1 y x . .
GŽ g . GŽ g X . 1 1 ay1
s H0 duH0 dÕ u Õ by1
GŽ a . GŽ b . GŽ g y a . GŽ g X y b .
gy a y1 g Xyb y1 a y g y g Xq1 b y g y g Xq1
=Ž 1 y u. Ž1yÕ. Ž 1 y ux . Ž 1 y Õy .
gq g Xya y b Xy1
= Ž 1 y ux y Õy . ,
R Ž a . ) 0, R Ž b . ) 0, R Ž g y a . ) 0, R Ž g X y b . ) 0; Ž A.31 .
S1 Ž a , a X , b ,g , d , x , y .
GŽ g . 1 gy a y1
s H0 du u a y1
Ž 1 y u. F2 Ž a X , b ,1, d ,1,ux ,uy .
GŽ a . GŽ g y a .
GŽ g . GŽ d .
s
GŽ a . GŽ g y a . GŽ b . GŽ d y b .
1 1 X
a y1 g y a y1 d y b y1 ya
= H0 duH0 dÕ u Õ by1 Ž 1 y u . Ž1yÕ. Ž 1 y uÕx y uy . ,
R Ž a . ) 0, R Ž g y a . ) 0, R Ž b . ) 0, R Ž d y b . ) 0. Ž A.32 .
A.2.1. Example of explicit eÕaluation of an integral representation

In working out the integral representation for hypergeometric functions in D s 4 y 2 e
dimensions, we often have to deal with the e expansion of integrals of the form
1
IŽ x. s H0 du d Ž u . f Ž u . , Ž A.33 .
y1 q be
d Ž u . s uy1q a e Ž 1 y u . , Ž A.34 .
where a and b are real numbers and f Ž u. is a smooth function in the domain
0 ( u ( 1: in particular, it is finite at the boundary points.
The procedure to deal with this kind of integrals is quite standard. The integral has a
pole in e when the integration variable u approaches either of the end points. We
concentrate first on the point u s 0, and we rewrite the integral in such a way to expose
the pole in e
1 1
IŽ x. s H0 du d Ž u . f Ž 0. qH0 du d Ž u . f Ž u . y f Ž 0 . s Iw1x q Iw2x . Ž A.35 .
The integral Iw1x can be easily done

GŽ a e . GŽ b e . f Ž 0. a q b G Ž 1 q a e . G Ž 1 q b e .
Iw1x s f Ž 0 . s , Ž A.36 .
GŽ Ž a q b . e . e ab GŽ 1 q Ž a q b . e .
and the integrand of Iw2x is now finite in the limit u ™ 0. In fact, we can make a Taylor
expansion
u2
f Ž u . y f Ž 0 . s uf X Ž 0 . q f XX Ž 0 . q . . . ' u g Ž u . , Ž A.37 .
2!
and write Iw2x as
1 1 ae y1 q be
Iw2x s H0 du d Ž u . u g Ž u . sH0 du u Ž 1 y u. g Ž u. . Ž A.38 .
We repeat now the same steps done for Eq. ŽA.35. with respect to the point u s 1, to
obtain
1 ae y1 q be 1 ae y1q be
Iw2x s H0 du u Ž 1 y u. g Ž 1. q H0 du u Ž 1 y u. g Ž u . y g Ž 1.
s Iw3x q Iw4x . Ž A.39 .

The integral Iw3x gives
GŽ 1 q a e . GŽ b e . f Ž 1. y f Ž 0. G Ž 1 q a e . G Ž 1 q b e .
Iw3x s g Ž 1 . s ,
GŽ 1 q Ž a q b . e . be GŽ 1 q Ž a q b . e .
Ž A.40 .
while Iw4x is finite at u ™ 1
1 ae be
Iw4x s H0 du u Ž 1 y u. hŽ u. , g Ž u . y g Ž 1. ' Ž 1 y u . h Ž u . , Ž A.41 .
and can be solved with an e expansion of the integrand. Adding all the contributions
together we have
1 GŽ 1 q a e . GŽ 1 q b e .
IŽ x. s b f Ž 0. q a f Ž 1.
a be GŽ 1 q Ž a q b . e .
1 ae be
q H0 du u Ž 1 y u. hŽ u. , Ž A.42 .
where
1
hŽ u. s Ž f Ž u . y Ž 1 y u . f Ž 0. y u f Ž 1. . . Ž A.43 .
uŽ 1 y u.
In the case where we have two integration variables, the procedure outlined above
can be re-iterated in a straightforward manner. To illustrate the procedure, we evaluate
explicitly the F2 functions of Eq. Ž3.30. to O Ž e 2 . .
The integral representation for F2 Žsee Eq. ŽA.29.. is given by
e 2 GŽ 1 y e .
F2 Ž 1,1, e , e q 1,1 y e , x , y . s IŽ x, y. , Ž A.44 .
GŽ 1 q e . GŽ 1 y 2 e .
where
1
IŽ x, y. s H0 du dÕ d Ž u,Õ . f Ž u,Õ . Ž A.45 .
and
y1 q e y2 e
d Ž u,Õ . s Õy1q e Ž 1 y u . Ž1yÕ. ,
y1
f Ž u,Õ . s Ž 1 y ux y Õy . ,
and I Ž x, y . must be computed to O Ž e 0 . . In order to expose the poles Žsee Eq. ŽA.35..,
we add and subtract the value of the finite function f Ž u,Õ ., computed at the boundary
points, in the following way:
1
IŽ x, y. s H0 du dÕ d Ž u,Õ . f Ž 1,0 . q f Ž u,0 . y f Ž 1,0 . q f Ž 1,Õ . y f Ž 1,0 .
q f Ž u,Õ . y f Ž u,0 . y f Ž 1,Õ . q f Ž 1,0 . 4

s Iw1x q Iw2x q Iw3x q Iw4x . Ž A.46 .
We are now in a position to evaluate the single contributions in the square brackets. In
fact
y1 1 y1q e 1 y1q e y2 e
Iw1x s Ž 1 y x . H0 du Ž 1 y u . H0 dÕ Õ Ž1yÕ.
y1
GŽ 1 q e . GŽ 1 y 2 e .
s Ž1yx . ,
e 2 GŽ 1 y e .
e
yx G Ž 1 q e . G Ž 1 y 2 e . 1 Ž 1 y u.
Iw2x s H0 du Ž 1 y ux . ,
1yx e GŽ 1 y e .
y1 y2 e
Ž1yx . 1 Õe Ž1yÕ.
Iw3x s H0 dÕ ,
e 1 y x y Õy
xy 1 e y2 e Ž Õy q ux q x y 2 .
Iw4x s H du dÕ Ž 1 y u . Õe Ž1yÕ. .
1yx 0 Ž 1 y ux . Ž 1 y x y Õy . Ž 1 y Õy y ux .
Ž A.47 .
The remaining integrals are finite in the limit e ™ 0, so that we can make a Taylor
expansion to O Ž e . for the integrands of Iw2x and Iw3x , and we can put directly e s 0 in
Iw4x. Recalling the definition of the dilogarithm function
x log Ž 1 y z .
Li 2 Ž x . s y H0 dz x(1 , Ž A.48 .
z
it is straightforward to carry out the last integrations and express the result in terms of
Li 2 functions, as done in Eq. Ž3.34..
A.3. Identities amongst the hypergeometric functions
There are three kinds of identities that relate hypergeometric functions. First there are
analytic continuations which connect functions in different regions of convergence.
Second are reduction formula which allow the functions to be expressed as simpler
series for certain values of the parameters. Finally there are transformations which relate
the same functions with different arguments.
A.3.1. Analytic continuation formulae

Here we give only those analytic continuation properties that relate the argument and
inverse argument. Gauss’ hypergeometric function has the following analytic continua-
tion properties Žsee for example Ref. w50x.:
ya GŽ g . GŽ b y a . 1
2 F1 Ž a , b ,g , z . s Ž yz .
GŽ b . GŽ g y a . 2 F1 ž a ,1 q a y g ,1 q a y b ,
z /
yb
GŽ g . GŽ a y b . 1
q Ž yz .
GŽ a . GŽ g y b . 2 F1 ž b ,1 q b y g ,1 q b y a ,
z / ,
<arg Ž yz . < - p , Ž A.49 .

GŽ g . GŽ g y a y b .
2 F1 Ž a , b ,g , z . s zya
GŽ g y a . GŽ g y b .
1
ž
= 2 F1 a ,1 q a y g ,1 q a q b y g ,1 y
z /
gy a y b GŽ g . GŽ a q b y g .
q z ay g Ž 1 y z .
GŽ a . GŽ b .
1
ž
= 2 F1 g y a ,1 y a ,1 q g y a y b ,1 y
z / ,
<arg Ž z . < - p , <arg Ž 1 y z . < - p . Ž A.50 .

The corresponding analytic continuation of the hypergeometric functions with two
variables are summarised in Table 2. There are many possible analytic continuations;
however, we list only those that are relevant to link the groups of solutions for the
Table 2
Analytic continuation for the hypergeometric functions of two variables
Function Continued in terms of
F4 Ž x, y . F4 Ž x r y,1r y ., F4 Ž yr x,1r x .
F3 Ž x, y . H2 Ž1r x,y y ., H2 Ž1r y,y x ., F2 Ž1r x,1r y .
H2 Ž x, y . F2 Ž x,y1r y .
F2 Ž x, y . S1Žy yr x,1r x ., H2 Ž y,y1r x ., S1Žy x r y,1r y ., H2 Ž x,y1r y .
H2 Ž x, y . F3 Ž1r x,y y ., S2 Ž1r x,y xy .
S1Ž x, y . F2 Žy x r y,1r y .
S2 Ž x, y . H2 Ž1r x,y xy .
two-mass triangle integral discussed in Section 3.2, that is the connections between the
Appell and Horn functions. The others are easily derived by summing the series with
respect to one of the summation variables to obtain an 2 F1 , applying Eq. ŽA.49.,
rewriting Gauss’ hypergeometric function as a series and reidentifying the double series.
We see that these functions appear to form a group.
F4 Ž a , b ,g ,g X , x , y .
GŽ g X . GŽ b y a . ya x 1
F4 a , a q 1 y g X ,g , a q 1 y b ,
s X
GŽ g y a . GŽ b .
Ž yy . ž ,
y y /
GŽ g X . GŽ a y b . yb
x 1
F4 b , b q 1 y g X ,g , b q 1 y a ,
q X
GŽ g y b . GŽ a .
Ž yy . ž ,
y y
,
/
Ž A.51 .
F3 Ž a , a X , b , b X ,g , x , y .
GŽ b y a . GŽ g . ya 1
s Ž yx . H2 a q 1 y g , a , a X , b X , a q 1 y b ,
ž ,y y /
GŽ g y a . GŽ b . x
GŽ a y b . GŽ g . yb
1
q Ž yx . H2 b q 1 y g , b , a X , b X , b q 1 y a ,
ž ,y y , /
GŽ g y b . GŽ a . x
Ž A.52 .
H2 Ž a , b ,g ,g X , d , x , y .
GŽ g X y g . GŽ 1 y a . yg 1
F2 a q g , b ,g , d ,g q 1 y g X , x ,y
s
GŽ 1 y a y g . GŽ g . X Ž y. ž y /
GŽ g y g X . GŽ 1 y a . yg
X 1
F2 a q g X , b ,g X , d ,g X q 1 y g , x ,y
q X
GŽ 1 y a y g . GŽ g .
Ž y. ž y / ,
Ž A.53 .
F2 Ž a , b , b X ,g ,g X , x , y .
GŽ b y a . GŽ g . ya y 1
s Ž yx . S1 a , a q 1 y g , b X , a q 1 y b ,g X ,y
ž , /
GŽ g y a . GŽ b . x x
GŽ a y b . GŽ g . yb
1
q Ž yx . H2 a y b , b X , b , b q 1 y g ,g X , y,y
ž / ,
GŽ g y b . GŽ a . x
Ž A.54 .
H2 Ž a , b ,g ,g X , d , x , y .
GŽ b y a . GŽ d . ya 1
s Ž yx . S2 a , a q 1 y d ,g X ,g , a q 1 y b ,
ž ,y xy /
GŽ d y a . GŽ b . x
GŽ a y b . GŽ d . yb
1
q Ž yx . F3 b ,g X , b q 1 y d ,g , b q 1 y a ,
ž ,y y , /
GŽ d y b . GŽ a . x
Ž A.55 .
S1 Ž a , a X , b ,g , d , x , y .
GŽ a X y a . GŽ g . ya x 1
F2 a , b , a q 1 y g , d , a q 1 y a X ,y
s
GŽ g y a . GŽ a . X Ž yy . ž ,
y y /
GŽ a y a X . GŽ g . ya
X x 1
F2 a X , b , a X q 1 y g , d , a X q 1 y a ,y
q X
GŽ g y a . GŽ a .
Ž yy . ž ,
y y
,
/
Ž A.56 .
S2 Ž a , a X , b , b X ,g , x , y .
GŽ a X y a . GŽ g . ya 1
s X Ž yx . H2 a , a q 1 y g , b X , b , a q 1 y a X ,
ž ,y xy /
GŽ g y a . GŽ a . x
X
GŽ a y a . GŽ g . ya
X
q X Ž yx .
GŽ g y a . GŽ a .
1
=H2 a X , a X q 1 y g , b X , b , a X q 1 y a ,
ž /
,y xy . Ž A.57 .
x
A.3.2. Reduction formulae

The F4 functions describing the massive bubble and the off-shell massless triangle
have the following reduction formulae which leave a single remaining Euler integral at
most w50,51x
x y
ž
F4 a , b ,g , b ,y
Ž1yx . Ž1yy.
,y
Ž1yx . Ž1yy. /
a a
s Ž 1 y x . Ž 1 y y . F1 Ž a ,g y b ,1 q a y g ,g , x , xy . , Ž A.58 .
x y
ž
F4 a , b , a , b ,y
Ž1yx . Ž1yy.
,y
Ž1yx . Ž1yy. /
y1 b a
s Ž 1 y xy . Ž1yx . Ž1yy. , Ž A.59 .
x y
ž
F4 a , b , b , b ,y
Ž1yx . Ž1yy.
,y
Ž1yx . Ž1yy. /
a a
s Ž 1 y x . Ž 1 y y . 2 F1 Ž a ,1 q a y b , b , xy . , Ž A.60 .
x y
ž
F4 a , b ,1 q a y b , b ,y
Ž1yx . Ž1yy.
,y
Ž1yx . Ž1yy. /
a x Ž1yy.
ž
s Ž 1 y y . 2 F1 a , b ,1 q a y b ,y
1yx / . Ž A.61 .
Similar reductions for the other functions of two variables are

ya xyy
F1 Ž a , b , b X , b q b X , x , y . s Ž 1 y y . 2 F1 a , b , b q b X , ž / , Ž A.62 .
1yy
yb
X x
F2 Ž a , b , b X ,g , a , x , y . s Ž 1 y y . F1 b , a y b X , b X ,g , x ,
ž / , Ž A.63 .
1yy
yb
y
F2 Ž a , b , b X , a ,g X , x , y . s Ž 1 y x . F1 b X , b , a y b ,g X ,
ž ,y , / Ž A.64 .
1yx
ya y
F2 Ž a , b , b X , b ,g X , x , y . s Ž 1 y x . 2 F1 ža b g , X
, X, / , Ž A.65 .
1yx
yb yb
X xy
F2 Ž a , b , b X , a , a , x , y . s Ž 1 y x . Ž 1 y y . 2 F1 b , b X , a , ž / ,
Ž1yx . Ž1yy.
Ž A.66 .
by a yb
F2 Ž a , b , b X , a , b X , x , y . s Ž 1 y y . Ž1yxyy. , Ž A.67 .
ya
F2 Ž a , b , b X , b , b X , x , y . s Ž 1 y x y y . , Ž A.68 .
yb
X y
F3 Ž a ,g y a , b , b X ,g , x , y . s Ž 1 y y . F1 a , b , b X ,g , x ,
ž / , Ž A.69 .
yy1
aq b y g
F3 Ž a ,g y a , b ,g y b ,g , x , y . s Ž 1 y y . 2 F1 Ž a , b ,g , x q y y xy . , Ž A.70 .
while for certain values of the parameters the H2 function reduces to an F2 or F1
yb
x
H2 Ž a , b ,g , d y a , d , x , y . s Ž 1 y x . ž
F2 d y a , b ,g , d ,1 y a ,y
1yx
,y y , /
Ž A.71 .
yg xy
H2 Ž a , b ,g ,1 y a , d , x , y . s Ž 1 q y . ž
F1 b , a ,g , d , x ,
1qy / . Ž A.72 .
The S1 and S2 functions we have introduced are less well known. From the integral
representation or manipulating the series using Eq. ŽA.76. we find
ya
X x
S1 Ž a , a X , b , a , d , x , y . s Ž 1 y y . 2 F1 a X , b , d ,
ž / , Ž A.73 .
1yy
S1 Ž a , a X , b ,g , b , x , y . s2 F1 Ž a , a X ,g , x q y . , Ž A.74 .
X
ya
S2 Ž a , a X , b , b X , a , x , y . s Ž 1 y x . 2 F1 Ž b , b X ,1 y a X ,y y Ž 1 y x . . . Ž A.75 .
A.3.3. Transformation formulae

A useful formula connecting Gauss’ hypergeometric function to itself is
ya z
ž
2 F1 Ž a , b ,g , z . s Ž 1 y z . 2 F1 a ,g y b ,g ,
zy1 /
yb
z
ž
s Ž 1 y z . 2 F1 g y a , b ,g ,
zy1 /
gy a y b
sŽ1yz . 2 F1 Ž g y a ,g y b ,g , z . . Ž A.76 .
Using the above results, then
yb yb
X x y
F1 Ž a , b , b X ,g , x , y . s Ž 1 y x . Ž1yy. F1 g y a , b , b X ,g ,
ž , /
xy1 yy1
ya x xyy
s Ž1yx . F1 a ,g y b y b X , b X ,g ,
ž , /
xy1 xy1
ya yyx y
s Ž 1 y y . F1 a , b ,g y b y b X ,g ,
ž , , /
yy1 yy1
Ž A.77 .
F2 Ž a , b , b X ,g ,g X , x , y .
ya x y
s Ž1yx . F2 a ,g y b , b X ,g ,g X ,
ž , /
xy1 1yx
ya x y
s Ž 1 y y . F2 a , b ,g X y b X ,g ,g X ,
ž , /
1yy yy1
ya x y
s Ž1yxyy. F2 a ,g y b ,g X y b X ,g ,g X ,
ž , / . Ž A.78 .
xqyy1 xqyy1
References
w1x C.G. Bollini, J.J. Giambiagi, Nuovo Cim. B 12 Ž1972. 20.

w2x J.F. Ashmore, Lett. Nuovo Cim. 4 Ž1972. 289.
w3x G. ’t Hooft, M. Veltman, Nucl. Phys. B 44 Ž1972. 189.
w4x See for example, J. Collins, Renormalisation ŽCambridge Univ. Press, Cambridge, 1984..
w5x G.’t Hooft, M. Veltman, Nucl. Phys. B 153 Ž1979. 365.
w6x K.G. Chetyrkin, F.V. Tkachov, A new approach to evaluation of multiloop Feynman integrals, INR
Preprint R-0118, Institute of Nuclear research, Moscow Ž1979..
w7x A.E. Terrano, Phys. Lett. B 93 Ž1980. 424.
w8x K.G. Chetyrkin, A.L. Kataev, F.V. Tkachov, Nucl. Phys. B 174 Ž1980. 345.
w9x K.G. Chetyrkin, F.V. Tkachov, Nucl. Phys. B 192 Ž1981. 159.
w10x N.I. Ussyukina, Teor. Mat. Fiz. 54 Ž1983. 124.
w11x V.V. Belokurov, N.I. Ussyukina, J. Phys. A 16 Ž1983. 2811.
w12x D.I. Kazakov, Teor. Mat. Fiz. 58 Ž1984. 345.
w13x D.I. Kazakov, A.V. Kotikov, Teor. Mat. Fiz. 73 Ž1987. 348.
w14x G.J. van Oldenborgh, J.A.M. Vermaseren, Z. Phys. C 46 Ž1990. 425.
w15x A.V. Kotikov, Phys. Lett. B 254 Ž1991. 158.

w18x Z. Bern, L. Dixon, D.A. Kosower, Nucl. Phys. B 412 Ž1994. 751.
w19x G. Kramer, B. Lampe, J. Math. Phys. 28 Ž1987. 945.
w20x E.E. Boos, A.I. Davydychev, Theor. Math. Phys. 89 Ž1991. 1052.
w21x A.I. Davydychev, J. Math. Phys. 32 Ž1991. 1052.
w22x A.I. Davydychev, J. Math. Phys. 33 Ž1992. 358.
w23x V.A. Smirnov, Phys. Lett. B 460 Ž1999. 397.
w24x A.I. Davydychev, Proc. Quarks-92, QCD161: A42:1992, World Scientific, Singapore, Ž1993., hep-
phr9307323.
w25x M.A. Sanchis-Lozano, On the connection between generalized hypergeometric functions and diloga-
rithms, hep-phr9511322.
w26x L.G. Cabral-Rosetti, M.A. Sanchis-Lozano, Generalized hypergeometric functions and the evaluation of
one loop integrals in Feynman diagrams, hep-phr9809213.
w27x ¨
S. Bauberger, F.A. Berends, M. Bohm, M. Buza, Nucl. Phys. B 434 Ž1995. 383.
w28x F.A. Berends, M. Buza, M. Bohm,¨ R. Scharf, Z. Phys. C 63 Ž1994. 227.
w29x J.M. Campbell, E.W.N. Glover, D.J. Miller, Nucl. Phys. B 498 Ž1997. 397.
w30x O.V. Tarasov, Phys. Rev. D 54 Ž1996. 6479.
w31x O.V. Tarasov, Nucl. Phys. B 480 Ž1996. 397.
w32x O.V. Tarasov, Acta Phys. Polon. B 29 Ž1998. 2655.
w33x J. Fleisher, F. Jegerlehner, O.V. Tarasov, Nucl. Phys. B 566 Ž2000. 423.
w34x I.G. Halliday, R.M. Ricotta, Phys. Lett. B 193 Ž1987. 241.
w35x R.M. Ricotta, ‘Negative dimensions in field theory’, J.J. Giambiagi Festschrift, ed. H. Falomir, p. 350
Ž1990..
w36x G.V. Dunne, I.G. Halliday, Phys. Lett. B 193 Ž1987. 247.
w37x G.V. Dunne, I.G. Halliday, Nucl. Phys. B 308 Ž1998. 589.
w38x A.T. Suzuki, A.G.M. Schmidt, J. High Energy Phys. 9709 Ž1997. 002.
w39x A.T. Suzuki, A.G.M. Schmidt, Eur. Phys. J. C 5 Ž1998. 175.
w40x A.T. Suzuki, A.G.M. Schmidt, Solutions for a massless off-shell two-loop three-point vertex, hep-
thr9712104.
w41x A.T. Suzuki, A.G.M. Schmidt, Phys. Rev. D 58 Ž1998. 047701.
w42x A.T. Suzuki, A.G.M. Schmidt, J. Phys. A 31 Ž1998. 8023.
w43x A.T. Suzuki, A.G.M. Schmidt, Negative-dimensional approach for scalar two-loop three point and
three-loop two point integrals, hep-thr9904195.
w44x A.T. Suzuki, A.G.M. Schmidt, Eur. Phys. J. C 10 Ž1999. 357.
w45x A.T. Suzuki, A.G.M. Schmidt, Negative-dimensional integration for massive four point functions. 1. The
standard solutions, hep-thr9707187.
w46x A.T. Suzuki, A.G.M. Schmidt, Negative-dimensional integration for massive four-point functions. II.
New solutions, hep-thr9709167.
w47x D. Broadhurst, Phys. Lett. B 197 Ž1987. 179.
w48x H. Exton, Multiple Hypergeometric Functions and Applications ŽEllis, Horwood, 1976..
w49x P. Appell, J. Kampe´ de Feriet,
´ ´ ´
Fonctions Hypergeometriques ´
et Hyperspheriques, Polynomes D’Hermite
ŽGauthiers-Villars, Paris, 1926..
w50x ´
A. Erdelyi, W. Magnus, F. Oberhettinger, F. Tricomi, Higher Transcendental Functions, Vol. I
ŽMcGraw-Hill, New York, 1953..
w51x W.N. Bailey, Generalised Hypergeometric Series ŽCambridge Univ. Press, Cambridge, 1966..
Reduction formalism for dimensionally regulated one-loop

N-point integrals
T. Binoth a , J.Ph. Guillet a , G. Heinrich a,b
a 1
´
Laboratoire d’Annecy-Le-Vieux de Physique Theorique LAPTH, Chemin de BelleÕue, B.P. 110, F-74941,
Annecy-le-Vieux, France
b
Department of Mathematics, UniÕersity of Guelph, Guelph, Ontario N1G 2W1, Canada
Received 7 December 1999; received in revised form 26 January 2000; accepted 31 January 2000
Abstract
We consider one-loop scalar and tensor integrals with an arbitrary number of external legs
relevant for multi-parton processes in massless theories. We present a procedure to reduce N-point
scalar functions with generic 4-dimensional external momenta to box integrals in Ž4 y 2 e .
dimensions. We derive a formula valid for arbitrary N and give an explicit expression for N s 6.
Further a tensor reduction method for N-point tensor integrals is presented. We prove that
generically higher dimensional integrals contribute only to order e for N G 5. The tensor reduction
can be solved iteratively such that any tensor integral is expressible in terms of scalar integrals.
Explicit formulas are given up to N s 6. q 2000 Elsevier Science B.V. All rights reserved.
PACS: 12.38.Bx
Keywords: One-loop; Tensor-reduction; Multi-parton reactions
1. Introduction
At future colliders multi-particlerjet final states will become more and more domi-
nant. A precise theoretical description of the QCD reactions is desirable in order to have
a better control on the backgrounds for various search experiments. Especially at hadron
colliders lots of multi-jet data will be collected which have to be confronted with
theoretical predictions. This motivates the accurate calculation of multi-parton reactions.
For tree level calculations the construction of matrix elements with large numbers of
particles in the final state is well established w1–3x. However, tree level results are very
unstable with respect to variations of the renormalization and factorization scales. As
1
´ a` l’Universite´ de Savoie.
URM 5108 du CNRS, associee
PII: S 0 5 5 0 - 3 2 1 3 Ž 0 0 . 0 0 0 4 0 - 7
362 T. Binoth et al.r Nuclear Physics B 572 (2000) 361–386
2 ™ N parton cross sections behave as a SN, at least next-to-leading order precision is

necessary to stabilize the predictions. The knowledge of the respective one-loop
amplitudes is thus mandatory. For 2 ™ 3 parton scattering NLO contributions have been
calculated in Refs. w4–7x, for eq ey™ 4 jets see w8–11x.
The main technical difficulties for constructing amplitudes consist in the treatment of
the occurring N-point scalar and tensor integrals. Because of infrared ŽIR. divergencies
the standard methods of w12–14x are not directly applicable. The authors of w5,6x used a
formalism working in Feynman parameter space. The generation of parameter integrals
with nontrivial numerators connected to tensor integrals was done by differentiation
techniques. The formalism produced formally higher dimensional integrals which in the
end canceled out. This cancellation had to be shown by explicit calculation. In this paper
we present a proof that this cancellation mechanism is true for arbitrary N. Furthermore,
a scalar reduction formula was derived in Ref. w5,6x which relates scalar integrals in
different space-time dimensions to each other. For 4-dimensional external momenta the
formula could not be shown to hold true in general. The problem stemmed from the
presence of singular matrices. We rederive the formula and treat the necessary inver-
sions of these matrices by using so-called pseudo-inverse matrices. To keep the external
momenta in 4 dimensions is advantageous because it allows one to use helicity
techniques w15x.
A formula relating tensor integrals to scalar integrals with shifted space-time dimen-
sions has been derived in Ref. w16x. This work was extended in Ref. w17,18x, where
scalar integrals were used as a generating functional for tensor integrals. The reduction
of the higher dimensional scalar integrals works in basically the same way as in Ref.
w5,6x with the same problems in the case of 4-dimensional kinematics. The method of
w17,18x has been designed for massive integrals and the applicability to the massless case
has not been worked out. The authors state that a regulator mass is needed for infrared
divergencies in the case of 4-dimensional kinematics.
A generalization of the Passarino–Veltman techniques dealing with the problem of
vanishing Gram determinants has been developed in Ref. w19,20x. The authors show
explicitly how to reduce box tensor integrals to scalar integrals. Another approach which
uses helicity methods for the reduction of tensor integrals to scalar integrals has been
discussed in Refs. w21–23x.
The complete reduction to scalar integrals is also possible in our tensor reduction
scheme. Our formalism is in a sense the generalization of the Passarino–Veltman
methods used for the calculation of electroweak radiative corrections w24x to the
massless case.
The paper is organized as follows. In Section 2 we rederive a reduction formula for
scalar integrals. We concentrate on the case of 4-dimensional kinematics from the start
and show that any N-point scalar integral with N G 6 is a linear combination of
pentagon integrals which in turn are combinations of box integrals plus terms of O Ž e ..
We give an explicit formula for the 6-point function with all external legs on-shell. In
Section 3 we derive our tensor reduction formalism which combines Passarino–Velt-
man-like methods with Feynman parameter space techniques. We prove that for generic
4-dimensional kinematics all higher dimensional N-point functions drop out for arbitrary
N G 5 and generalize the reduction methods of w5,6,25x to arbitrary N. We construct a
hierarchy of tensor formulas up to N s 6 and rank F N which can be solved by
T. Binoth et al.r Nuclear Physics B 572 (2000) 361–386 363
iteration. The explicit expressions and the derivation of some basic formulas are given in
the appendix.
2. Reduction formula for massless scalar integrals
In this section we will first derive a scalar reduction formula valid for an arbitrary
number of external legs. Then we will use these formulas to derive explicit representa-
tions for scalar integrals up to N s 6.
2.1. DeriÕation
Consider a one-loop, scalar N-point function with massless propagators for N G 4. If

all legs are off-shell the integral is finite and can be treated in four dimensions w14x. If at
least one external leg is massless it is infrared divergent and needs a regulator. In the
framework of dimensional regularization, 4-dimensional methods are not applicable
anymore. We work in n s 4 y 2 e dimensions in the following with the external
momenta kept in four dimensions.
With the momentum flows as indicated in Fig. 1, we define the propagator momenta
as ql s k y r l with r l s p l q r ly1 for l from 1 to N and r 0 s rN . Momentum conserva-
tion allows us to choose rN s 0.
The corresponding analytic expression in momentum and Feynman parameter space
is
N
INn Ž R . s d k
H N
1 N
s Ž y1 . G Ž N y nr2 . H0 d
`
N
z
ž
d 1y Ý zl
ls1
Ny nr2
/ . Ž 1.
2 Ž zPSPz.
Ł ls1 q l
Herein R s Ž r 1 , . . . ,rN ., d k s d n krŽ ip n r2 .. The kinematic information is contained in

the matrix S which is related to the Gram matrix G by
2
Sk l s y 12 Ž r l y r k . s 12 Ž G k l y Õ l y Õ k . ,
Gk l s 2 rk P rl , Õk s G k kr2 , k ,l s 1, . . . , N . Ž 2.
Although it is well known w5,6,17,18,25x that the N-point integral can be split into a
finite, Ž6 y 2 e .-dimensional integral and a part with less external legs containing the
infrared poles, we want to present our derivation which later allows us to deal with the
problem of vanishing Gram determinants in a transparent manner. As an ansatz we write
Ž1. as a sum of Žone-propagator. reduced diagrams and a remainder,
N
N
1 y Ý ls1 b l ql2
Ý b q2
ls1 l l
INn s Idiv q Ifin s d k
H N q dkH N . Ž 3.
Ł ls1 ql2 Ł ls1 ql2
Fig. 1. N-point graph.
We want to show in the following that one can find coefficients b l such that Ifin
N
contains no IR poles. In standard Feynman parametrization one gets with D s Ý ls1 z l ql2 ,
N
N 1 y Ý ls1 b l ql2
`
Ifin s G Ž N . H d N z d 1 y Ý is1 z i
ž / Hd k , Ž 4.
0 DN
and after a shift k s k˜ q Ý ls1
N
z l rl ,
N
N 1 y Ý ls1 b l q̃l2
`
Ifin s G Ž N . H d N z d 1 y Ý ls1 z l
ž / Hdk˜ N
, Ž 5.
0
Ž k̃ 2 y M 2 .
with
N
M 2szPSPz , q˜ j s k˜ y Ý Ž d l j y z l . rl . Ž 6.
ls1
Now the term in square brackets in Eq. Ž5. can be written as

N N N
1 y Ý is1 bi q˜i2 s y Ž k˜ 2 q M 2 . Ý bj q Ý z j Ž 1 q 2 Ž S P b . j . . Ž 7.
js1 js1
If now the equations
Ž S P b . j s y 12 , j s 1, . . . , N , Ž 8.
are fulfilled, the second term on the right-hand side of Ž7. vanishes and one finds
N ` N k˜ 2 q M 2
Ifin s yG Ž N .
ž / Ý bl
ls1
H0 d N
ž
z d 1 y Ý ls1 z l / Hdk˜ Ž k˜ 2 y M 2 .
N
. Ž 9.
Finally, the loop momentum integration gives

N
N
Nq1
n ` ž
d 1 y Ý ls1 z l /
Ifin s
žÝ /
ls1
b l Ž y1 . ž
G Ny1y
2 /Ž N y n y 1. H0 d N
z
ŽM . 2 Ny Ž nq2 .r2
N
sy
žÝ / ls1
b l Ž N y n y 1 . INnq 2 . Ž 10 .
The Ž6 y 2 e .-dimensional integral is IR finite, as can be seen by a power-counting

argument in the corresponding momentum integral.
It remains to solve Eq. Ž8.. In the case of non-exceptional 4-dimensional kinematics,
N
rankŽ S . s minŽ N,6. holds. For N F 6 one has b l s y1r2Ý ks1 Sy1
k l . In the case N s 5,
Ifin contains a factor N y 1 y n which is O e . As is well known, pentagon integrals
Ž . Ž .
are just a sum of box integrals up to a remainder which drops out in phenomenological
applications. In the case N ) 6 and 4-dimensional kinematics, Eq. Ž8. does not have a
unique solution. To clarify this point it is useful to rewrite Eq. Ž8.. Using momentum
conservation, rN s 0, Eq. Ž8. leads to the following equations:
Ny1 Ny1 N
Ý G k l b l s Õ k BN , Ý Õl bl s 1 , BN s Ý bl . Ž 11 .
ls1 ls1 ls1
Herein G is now the Gram-matrix of the vectors r ls1 . . . Ny1. In the case of four
dimensional kinematics it is at most of rank 4 for all N G 5. Generically any four
vectors out of r 1 , . . . ,rNy1 are linearly independent. Such a configuration is called
m
non-exceptional in the following. Choosing four linearly independent vectors Els1, . . . ,4
as a basis of the physical Minkowski space, one can define a coefficient matrix R and a
Gram matrix G˜ made out of these basis vectors,
4
r jm s Ý R l j Elm , G˜jk s 2 Ej P Ek . Ž 12 .
ls1
The Gram matrix G is expressible as G s R T GR. ˜

Based on Ž12. one can construct the most general solution for the case detŽ G . s 0 by
means of pseudo-inverse matrices. This concept will also be useful for the tensor
integrals.
A pseudo-inverse H to G is defined by the conditions HGH s H and GHG s G.
Given a pseudo-inverse matrix H to G, the following statement can be proven. A
solution to the linear equation G P x s y exists if and only if y s GH P y. Then the most
general solution can be written as x s H P y q Ž1 Ny 1 y HG . P u, where the last term –
with u g R Ny 1 arbitrary – is parametrizing the solutions of the homogeneous equation.
As G is symmetric, its pseudo-inverse is uniquely defined w27,28x. The pseudo-inverse
to G is given by
y1 y1
H s R T Ž RR T . G˜y1 Ž RR T . R, Ž 13 .
where R is defined in Ž12.. For N s 5 and non-exceptional external momenta the
vectors r 1 , . . . ,r4 are the natural basis. Then R s 1 4 , G˜ s G, H s Gy1 . Note that the
concept of the pseudo-inverse always allows the inversion of linear equations containing
the Gram matrix, even if the external momenta are exceptional 2 .
If the Gram matrix is not of maximal rank, the condition Õ s GH P Õ for the existence
of a solution of the inhomogeneous equation in Ž11. is never fulfilled. This can be seen
by diagonalizing the symmetric matrix GH by an orthogonal transformation O,
GH s O T DO, where D is a diagonal matrix with elements Ž1,1,1,1,0, . . . ,0.. Now it is
evident that in general O P Õ / D O P Õ. Thus a solution to Eq. Ž11. exists only for the
case BN s 0. The solution of the homogeneous equation spans a Ž N y 5.-dimensional
space which is just the kernel of the Gram matrix. It can be parametrized by Ž N y 5.
vectors U Ž1, . . . , Ny5.. Defining
y1
K s 1 Ny 1 y HG s 1 Ny1 y R T Ž RR T . R, Ž 14 .
one has K P Õ g kerŽ G .. Now one can choose U Ž Ny5. s K P ÕrŽ Õ P K P Õ . parallel to Õ
and the others orthogonal, Õ P U Ž ks1, . . . , Ny6. s 0. A general vector in kerŽ G . is then
Ny 6 Žk.
parametrized by U s Ý ks 1 bk U q a U Ž Ny5. and the general solution is given by
Ny6
Ž KPÕ. iq Ý b k Ui Ž k .
ks1
bi s , i s 1, . . . , N y 1 ,
ÕPKPÕ
N
BN s Ý bk s 0 , Ž 15 .
ks1
where a s 1 is imposed by Õ P b s 1. Thus for N s 6 the solution to Ž11. is unique and

therefore equal to the one defined by the inverse of the matrix S. Eq. Ž15. proves
constructively that for all N G 6, INn can be expressed in terms of Ž N y 1.-point
functions without the higher dimensional remainder term. This is a consequence of the
linear dependence of propagators if the external momenta are 4-dimensional. For the
special case N s 7 this has already been demonstrated in Ref. w5,6x. Obviously one can
choose the b ’s to eliminate Ž N y 6. b’s from the set b 1 , . . . ,bNy14 . Doing so one
observes that INn can be expressed by only 6 Ž N y 1.-point graphs for arbitrary N G 6.
Here we make contact to a result in Refs. w14,26x, where a similar relation has been
derived for the IR finite case in integer dimensions by using 4-dimensional Schouten
identities.
For N F 6 one has detŽ S . / 0 and the following relation Žsee also w5,6x. holds:
N N det Ž G .
Ý bl s y 12 Ý Sy1
lk s y
2 N det Ž S .
, Ž 16 .
ls1 l , ks1
which shows that the vanishing of the finite remainder terms Ž10. is related to the
vanishing of the Gram determinant.
2
Clearly one could also solve Eq. Ž8. by means of the pseudo-inverse to S. With OS defining the
orthogonal transformation which diagonalizes S and the non-zero eigenvalues l1 , . . . , l6 of S, the pseudo-in-
verse to S is given by OSTPDSinvPOS , where DSinv sdiagŽ1r l1 , . . . ,1r l6 ,0, . . . ,0.. Its computation is alge-
braically more involved than the computation of H though.
In summary, with the definition of reduced graphs,

p
n
Ł ms1 ql2 m
INy p ,l 1 , . . . ,l p s d k
H N ,
2
Ł ls1 q l
p
n
Ł ms1 Ž qN2 y ql2 . m
INy p , Nyl 1 , . . . , Nyl p s Hd k N , Ž 17 .
Ł ls1 ql2
the scalar reduction formula for a regular matrix S is
N det Ž G .
INn s y 12 Ý Sy1 n
k l INy1,l q Ž N y n y 1 . INnq2 , det Ž S . / 0 . Ž 18 .
k ,ls1 2 N det Ž S .
In the case of a singular matrix S, as outlined above, we can always achieve a
representation without higher-dimensional integrals,
1 Ny1 Ny6
INn s y
ÕPKPÕ
Ý
ls1
ž Ž KPÕ. lq Ý
ks1
/
b k U lŽ k . INy1,
n
Nyl , det Ž S . s 0 . Ž 19 .
Remember that the Ž N y 6. b ’s are free parameters and that the U Ž ks1, . . . , Ny6. are lying
in the kernel of G and are orthogonal to Õ. The case N s 6 is special in the sense that it
is of the form Ž18. with vanishing higher-dimensional term. By applying the above
formulas iteratively, any N-point function can be reduced to linear combinations of box
integrals plus irrelevant O Ž e . terms in a constructive way.
2.2. Application
We will now discuss the formulas for the cases up to N s 6 in more detail. In these
cases one has detŽ S . / 0 and relation Ž16. holds. Note that the vanishing of the Gram
determinant for N - 6 typically occurs at a border of the respective phase space. In the
case N s 3 Ž4,5. and detŽ G . s 0 the scalar integrals are just sums of 2-point, Ž3-point,4-
point. functions, respectively, an observation made some time ago by Stuart w19,20x and
reflected by Eq. Ž18.. This fact can be used as a guideline to define nontrivial groupings
of Ž N y 1.- and N-point functions which is helpful for numerical purposes w25x. The
other dangerous determinant is detŽ S . which vanishes if N G 7. We note that in this case
our formulas are mathematically well-defined and that for the generation of explicit
expressions one just has to apply formula Ž19.. Due to the freedom to choose the
parameters b , the representation of the N-point function in terms of reduced integrals is
not unique as it is the case for N F 6. This is a reflection of the fact that the respective
reduced integrals are not linearly independent.
As the reduction formulas were derived under the assumption N G 4, we discuss now
the special cases N s 1,2,3 first. We will represent the kinematic information in terms of
the matrix S. The relation to the Gram matrix is defined in Eq. Ž2..
Cases N s 1,2
As massless tadpole integrals are zero in dimensional regularization, I1n s 0. The case
N s 2 is trivial in the sense that reduced graphs are tadpoles. The formula simply gives a
relation between n- and Ž n q 2.-dimensional two-point functions. If p 1 is the external

momentum with p 12 s m12 , one obtains
Ž n y 1.
I2n s 2 I2nq2 . Ž 20 .
m12
For vanishing Gram determinant, i.e. m12 s 0, I2n s 0 in dimensional regularization.
Case N s 3
The matrix S generally looks like Ž p k2 s m2k .
0 m22 m12
S s y 12 m 22
m12
0
m23
m23 .
0 0
Note that in the cases of one or two external lines on-shell, S is not of maximal rank. If
Ž 21 .
for example m12 s m22 s 0, or m12 s 0,m 22 / m23 , there exists no solution to Ž8. and no
reduction is possible. If all legs are off-shell one finds a solution which relates an
Ž n q 2.-dimensional off-shell triangle to an n-dimensional one,
3 det Ž G .
I3n s y 12 Ý Sy1 n
l k I2,l q Ž 2 y n . I3nq2 . Ž 22 .
l, ks1 8det Ž S .
We will see later that the Ž n q 2.-dimensional off-shell triangle appears in the tensor
reduction of rank G 3 tensor 4-point functions.
As any N-point integral in Ž4 y 2 e . dimensions can be reduced to triangles plus a
finite remainder, we see that they are the atoms of any scalar reduction formula.
Moreover, at the one-loop level they define an IR counterterm structure to any N-point
function3.
The explicit expressions for the three types of triangle graphs may be found for
example in Refs. w25,29x.
Case N s 4
With Ž p 1 q p 2 . 2 s s, Ž p 2 q p 3 . 2 s t, one has
0 m22 t m12
Ssy 1
2
m22
t
m12
0
m 23
s
m23
0
m 24
s
m24
0
0 .
S is of maximal rank for all on-shell cases as long as s and t are nonzero. The reduction
Ž 23 .
3
This is actually to be expected from looking at the reduced diagrams Žobtained by shrinking all finite
propagators, if the loop momentum becomes softrcollinear. of the one-loop N-point function corresponding to
solutions of the Landau equations. The respective reduced diagrams are just the reduced diagrams of triangle
graphs.
formula Ž18. applies and relates a 4-point function in n dimensions with triangles and
Ž n q 2.-dimensional 4-point functions,
4 det Ž G .
I4n s y 12 Ý Sy1 n
l k I3,l q Ž 3 y n . I4nq2 . Ž 24 .
l, ks1 16det Ž S .
For n s 4 y 2 e the triangles carry all the infrared poles whereas the Ž6 y 2 e .-
dimensional part is a finite remainder. It can be calculated directly by setting e s 0.
Explicit expressions for the box integrals can be found in Refs. w5,6,25x.
Case N s 5
With Ž pj q pjq1 . 2 s s j, jq1 Ž j mod 5. one has
0 m 22 s23 s51 m12

m22 0 m 23 s34 s12
S s y 12
s23
s51
m12
m 23
s34
s12
0
m24
s45
m 24
0
m25
s45 .
m 25
0
0
For N s 5 the higher dimensional integrals come with a prefactor Ž4 y n. and thus drop
out in phenomenological applications. One only has to know box integrals,
Ž 25 .
5
I5n s y 12 Ý Sy1 n
l k I4,l q O Ž e . . Ž 26 .
l, ks1
Explicit expressions for the 5-point function with zero or one massive legs may be found
in Refs. w5,6,25x. Other cases are easily constructed by using known representations of
off-shell box-integrals.
Case N s 6
Now the Gram matrix is not invertible anymore for 4-dimensional kinematics
whereas an inverse for S still exists. With Ž pj q pjq1 . 2 s s j, jq1 , Ž pj q pjq1 q pjq2 . 2 s
s j, jq1, jq2 Ž j mod 6., one has
°0 m 22 s23 s234 s61 m12 ¶

m 22 0 m 23 s34 s345 s12
s23 m23 0 m24 s45 s123
S s y 12 . Ž 27 .
s234 s34 m24 0 m 25 s56
s61 s345 s45 m25 0 m 26
¢m 2
1 s12 s123 s56 m26 0 ß
Before going on, it is instructive to clarify the dependences between the Mandelstam
variables which define the matrix S. Since they are directly related to the cuts of the
N-point graph, the following counting holds for N G 4, where M is the number of
off-shell legs, C the number of cuts, D the number of independent Lorentz invariants
built out of the vectors r 1 , . . . ,rN , and Z s C y D the number of constraints:
C s N Ž N y 3 . r2 q M , D s 3 N y 10 q M , Z s Ž N y 4 . Ž N y 5 . r2 .
Ž 28 .
This makes manifest that for N G 6 one encounters subtleties due to the appearing
constraints. Here, the Žnonlinear. constraint det Ž G . s 0 shows up in the fact that
N
Ý ls1 b l s y 12 Ý k,ls1
N
Sy1
k l s 0. The reduction formula for N s 6 reads
6
I6n s y 12 Ý Sy1 n
l k I5,l . Ž 29 .
l, ks1
If all six external legs are on-shell, the occurring pentagon integrals I5,n l have one
external leg off-shell. The explicit expression for the on-shell 6-point function is given
in Appendix A.
3. Reduction of tensor-integrals
In this section we show that any rank L, N-point tensor integral can, by recursion and
scalar reduction, be expressed in terms of scalar box, triangle and two-point integrals.
We prove that for N G 5 and non-exceptional external momenta higher dimensional
integrals drop out.
The rank L, N-point tensor integral is given by
k m1 . . . k m L
INm1 . . . m L s d kH N . Ž 30 .
2
Ł ls1 q l
We omit the superscript n indicating the dimension in the tensor integrals to simplify
the notation. If the dimension is different from Ž4 y 2 e ., it will always be written out
explicitly. After introducing Feynman parameters and making a shift of the loop
momentum as in the last section, the odd powers of the loop momentum in the
numerator can be dropped. Using the standard integral
k m1 . . . k m 2 m
H dk N
Ž k2yM 2 .
N m 1 . . . m 2 m4 m G Ž N y Ž n q 2 m . r2 . yNq Ž nq2 m .r2
s Ž y1 . g ŽPPm . Ž y 12 . G Ž N.
Ž M2. ,
Ž 31 .
one finds the following formula:
INm1 . . . m L
w Lr2 x Ny1
m m 1 . . . m L 4 nq2 m
s Ý Ž y 12 . Ý g ŽPPm . r jP1 . . . r jPLy 2 m IN Ž j1 , . . . , j Ly2 m . .
ms0 j1 , . . . , j Ly2 m s1
Ž 32 .
m ... m 4
w Lr2x is the nearest integer less or equal to Lr2 and gŽPPm. r jP . . . r jP 1 L
stands
1 Ly 2 m
for the sum over all different combinations of L Lorentz indices distributed to m metric
L m Ž
tensors and Ž L y 2 m. r-vectors. These are
2m
Ł ks ž /
1 2 k y 1 terms. A dot appearing
.
as an index at objects inside a square bracket stands for one index out of the set
specified in curly brackets at the outside of the square bracket. X Ž m. or XŽ m. denotes the
product of m terms of X with adequate indices 4 . For example, a tensor object like
PP P m 1 m 2 m 3 m 4 m 5 4
XŽ2. Y with X a symmetric tensor of rank 2 and Y a vector means
m 1 m 2 m 3 m 4 m 54
XPP
Ž 2. Y
P
s Ž X m1 m 2 X m 3 m4 q X m1 m 3 X m 2 m4 q X m1 m4 X m 2 m 3 . Y m5
q 5 permutations.
IND Ž j1 , . . . , j M . are scalar integrals with nontrivial numerators in D dimensions, defined

by
` N z j1 . . . z j M
N
IND Ž j1 , . . . , j M . s Ž y1 . G Ž N y Dr2 . H0 ž
d Nz d 1 y Ý zl
ls1
/ Ž zPSPz.
NyDr2
.
Ž 33 .
Recursion relations for this kind of integrals were given for M F 4 in Ref. w25x. We
derive such relations for general M in Appendix B.
From now on we use standard summation conventions for the indices. Eq. Ž32.
implies that tensor integrals in momentum space are linear combinations of scalar
integrals in different dimensions. We note that formula Ž32. is equivalent to a more
general formula given in Ref. w16x for the case of integer powers of the propagators.
We want to prove now that for non-exceptional momenta and N G 5 the higher
dimensional scalar integrals drop out. To this end we solve Eq. Ž32. for INn Ž j1 , . . . , j L . by
contracting it with 2 L r lm1 1 . . . r lmL L . For the inversion of the Gram matrices we use its
Žpseudo. inverse as defined above. With 2 r l P k s q N2 y ql2 q Õ l one finds
L
INn Ž j1 , . . . , j L . s Ý wPŽ Lyk . HP l 1 . . . HP l k n
j1 , . . . , j L 4 INyk , Nyl 1 , . . . , Nyl k
ks0
w Lr2 x
m Ž m . nq2 m
y Ý Ž y1. HPP IN Ž Ž L y 2 m . indices. j1 , . . . , j L 4 , Ž 34 .
ms1
where w s H P Õ and H is defined in Eq. Ž13.. On the right-hand side of Eq. Ž34.,
differences of reduced integrals appear as defined in Eq. Ž17.. In principle one has to
add to Ž34. also the solutions stemming from the homogeneous equation which are
present if G is not of maximal rank. But as they contain the matrix K s 1 Ny 1 y HG,
they vanish after contraction with the r ’s, i.e. r m P Ž1 Ny 1 y HG . s 0. If G is invertible,
4
The ‘‘power’’ Ž m. may appear as a lower index for convenience of notation if it would interfere with the
dots standing for upper indices. Hence we write XŽPPm. instead of Ž X PP .Ž m..
L
H s Gy1 . The bracket of the k th term of the first sum stands for ž / terms whereas the
K
L m Ž
bracket of the mth term of the second sum stands for
2m
Ł ks 1 2 k y 1 terms.
.
Inserting Ž34. into Ž32., we find that the tensor integral decays into a part containing
ž /
Ž4 y 2 e .-dimensional objects, K Nm1 . . . m L , and a part built out of higher-dimensional
objects, JNm1 . . . m L ,
INm1 . . . m L s K Nm1 . . . m L q JNm1 . . . m L , Ž 35 .
L
m 1 , . . . , m L4 n
K Nm1 . . . m L s Ý WPŽ Lyk . KlP1 . . . KlPk INyk , Nyl 1 , . . . , Nyl k , Ž 36 .
ks0
w Lr2 x
m
JNm 1 . . . m L s Ý Ž y1.
ms1
Ny1 m 1 . . . m L4
PP PP P
= Ý ž Ž gr2. Ž m .y HŽ m. /r j1 . . . r jPLy 2 m
j1 , . . . , j Ly2 m s1
=INnq 2 m Ž j1 , . . . , j Ly2 m . . Ž 37 .
The Lorentz indices are carried by the objects
H mn s r m P H P r n ,
Klm s Ž r m P H . l ,
W msr m P wsKm PÕ . Ž 38 .
We recall that w s H P Õ and Õ l s Gl lr2.
If the r ’s span 4-dimensional Minkowski space, which is generically the case if
N G 5, H mn is just proportional to the metric tensor g 4mn in 4 dimensions,
Ny1 4
H mn s r m P H P r n s Ý rim r jn Hi j s Ý Elm Ekn G˜y1 1 mn
l k s 2 g4 . Ž 39 .
i , js1 l , ks1
Thus, the coefficients of the higher dimensional integrals in Ž37. are proportional to
g Ž m. y g 4Ž m. which is of order e . As by power counting the higher dimensional integrals
are finite objects, it follows that the whole contribution is O Ž e ., if the external momenta
are 4-dimensional. This proves the cancellation of higher dimensional integrals in tensor
reductions for arbitrary N G 5 and non-exceptional external momenta, and we obtain
L
m 1 , . . . , m L4 n
INm1 . . . m L s Ý WPŽ Lyk . KlP1 . . . KlPk INyk , Nyl 1 , . . . , Nyl k q O Ž e . , NG5 .
ks0
Ž 40 .
For N - 5, Eq. Ž39. is not valid since the external momenta cannot span Minkowski
space anymore. Thus one has to calculate the terms JNm1 . . . m L for N - 5. The explicit
expressions are given in Appendix D.
Now we want to rewrite Eq. Ž35. as a recursion formula for arbitrary tensor integrals.
n
To this effect we express the contracted tensor integrals INy k, Nyl 1 , . . . , Nyl k k ) 0 as
Ž .
Ž N y 1.-point tensor integrals which are maximally of rank Ž L y 1.. By using Ž k y 1.

times the relation q N2 y ql2 s 2 r l P k y Õ l , one gets
ky1 2j
n
INy k , Nyl 1 , . . . , Nyl k s Ý Ž y1. ky jy1 j!
ÕPŽ kyjy1 . rPn 1 . . . rPn j In1 . . . n j
l 1 . . . l ky 1 4 Ny1, Nyl k
,
js0
kG1 . Ž 41 .
Insertion of expression Ž41. into Eq. Ž36. leads to the recursion formula,
1 m 1 . . . m L4
K Nm 1 . . . m L s W PK N Ly1 dots 4
L
2 Ž Ly1. m 1 . . . m L4 n . . . n
q KlPHnP1 . . . HnPLy 1 INy1,
1 Ly 1
Nyl . Ž 42 .
L!
The derivation of Ž42. is given in Appendix C.
Since for N G 5 the terms JNm1 . . . m L contribute only to order e , K Nm1 . . . m L in Eq. Ž42.
can be replaced by INm1 . . . m L to obtain the recursion formula for N G 5. For N - 5, one
cannot drop the JNm1 . . . m L terms, such that the general recursion formula reads
1 Ly1 dots 4 m 1 . . . m L 4
INm 1 . . . m L s JNm 1 . . . m L q W P Ž IN y JN .
L
Ž Ly1.
2 m 1 . . . m L4 n . . . n
1 Ly 1
Nyl . Ž 43 .
L!
As a consequence, higher dimensional integrals INnq X 2 m Ž m s 1,2. will appear in the
X
reduction of tensor N-point integrals only with N F 4. All the higher dimensional
integrals can be remapped to n-dimensional integrals with the scalar reduction formulas
Ž20., Ž22. and Ž24.. In the case of exceptional kinematics and N G 5, higher dimensional
Ž N G 5.-point functions can be present.
As some of the reduced integrals on the right-hand side of Ž43. do not contain a
trivial propagator, they are not in the standard form for applying the reduction formula
again. Therefore one has to perform a shift k ™ k q r l in the tensor integral, leading to
m1 . . . m P m1 . . . m P ˆ m1 . . . m P ˆ
INy 1, Nyl Ž R . s INy1 R w N x y INy1
ž / Rw l x ž /
P
m1 . . . mP4 m1 . . . mP ˆ
s Ý r lP Ž k . INy1
Pyk dots 4
Ž Rw l x. y INy1 Rw l x .
ž / Ž 44 .
ks0
The argument vectors are

R s Ž r 1 , . . . ,rN . ,
R̂ w k x s Ž r 1 , . . . ,rˆk , . . . ,rN . ,
R w l x s Ž r lq1 y r l ,r lq2 y r l , . . . ,rNy2ql y r l ,0 . , Ž 45 .
where rˆ means that the respective vector has to be left out. The vector indices are
understood to be taken cyclically symmetric with periodicity N, i.e. rNq l s r l for
l g 1, . . . , N 4 . As we assume rN s 0, the integrals in the last line of Ž44. have again at

least one trivial propagator and are suited for a further reduction step.
A full tensor reduction of a rank L tensor integral is obtained by employing Eq. Ž42.
for N G 5 resp. Eq. Ž43. for N - 5 to do the first reduction step. Then the integrals
n 1 . . . n Ly 1
INy 1, Nyl have to be shifted by using Eq. 44 to end up with expressions of rank
Ž .
F L y 1 to which the same reduction procedure can be applied again. Explicit tensor
reduction formulas for N s 2, . . . ,6 are given in Appendix D.
For N G 5 one can express by recursion any rank L, N-point tensor integral in terms
of Ž4 y 2 e .-dimensional scalar N-point and tensor box integrals, which in turn can be
reduced further down to scalar box, triangle and two-point integrals, taking also into
account the scalar reduction formulas given in the previous section. This defines an
algorithm which can be easily programmed.
4. Conclusion
We have presented reduction formulas for massless N-point scalar and tensor
functions.
We have shown in a constructive manner how scalar N-point functions can be
represented as linear combinations of Ž4 y 2 e .-dimensional scalar Ž N y 1.-point func-
tions for arbitrary N G 5. In particular, we pointed out how to treat the linear equations
which determine the reduction coefficients, b l , for N G 6 in terms of the pseudo-inverse
of the singular Gram matrix. In this way all mathematical operations are valid for
4-dimensional external kinematics throughout the whole reduction procedure. We ap-
plied the reduction formalism to scalar integrals up to N s 6 explicitly and gave an
expression for the on-shell 6-point function.
For N-point tensor integrals we have formulated a reduction scheme for arbitrary N.
We have proven that higher-dimensional integrals always vanish for N G 5 in the case of
non-exceptional kinematics. We derived a recursion formula for the remaining n-dimen-
sional part. For the derivation we used methods a` la Passarino–Veltman, such as
contracting tensor integrals with external vectors and inverting Gram matrices. In the
case of a singular Gram matrix the inversion has to be done with its pseudo-inverse. By
iteration any tensor integral can be expressed as a linear combination of scalar integrals.
Higher dimensional scalar integrals appear only in the reduction of N F 4 tensor
integrals. These higher dimensional integrals are expressible in terms of Ž4 y 2 e .-
dimensional integrals as has been shown explicitly in the discussion of the scalar
reduction formulas. To make the general formalism more user-friendly we gave explicit
expressions up to N s 6 in an appendix.
We also derived a reduction formula in Feynman parameter space in an appendix.
The formula generalizes results in the literature avoiding a projective transformation. As
we gave all the formulas to translate momentum space expressions into Feynman
parameter space expressions, the equivalence between the two approaches is manifest. In
applications this will allow us to employ methods similar to the ones in Ref. w25x to get
numerically stable expressions, a problem we did not address in this paper.
We conclude that the computation of IR divergent one-loop integrals for arbitrary
numbers of legs can be mastered with the reduction formulas presented here. The
iterative structure makes it easy to implement the formalism in algebraic computer

programs. The conceptual problems for the construction of multi-parton one-loop
amplitudes are thus solved.
The generalization of our method to include also massive particles is postponed to a
future publication.
Acknowledgements
We thank Bas Tausk and Christian Schubert for interesting remarks and discussions.
G.H. would like to thank the LAPTH for its hospitality during her stay in June 1999.
This work was supported in part by the EU Fourth Training Programme ‘‘Training
and Mobility of Researchers’’, Network ‘‘Quantum Chromodynamics and the Deep
Structure of Elementary Particles’’, contract FMRX–CT98–0194 ŽDG 12 - MIHT..
Appendix A. The scalar 6-point function
Here we present an explicit formula for the scalar hexagon function with all external
legs on-shell. The reduction formula Ž29. allows us to write it as a sum of six pentagon
integrals with one external leg off-shell.
I6n Ž s12 , s23 , s34 , s45 , s56 , s61 , s123 , s234 , s345 .
6
s y 12 Ý Sy1
k l I5,1 m a s s Ž s lq2,lq3 , s lq3,lq4 , s lq4,lq5 , s lq5,l ,lq1 , s l ,lq1,lq2 ; s l ,lq1 . .
k ,ls1
Ž A.1 .
The function I5,1 m a s s has been given in Ref. w5,6x. We rederived the formula and agree
with it for Euclidean kinematics Žup to a trivial typo 5 ..
The other ingredient, the matrix S, is given explicitly by Eq. Ž27. after setting all
m 2j s 0. The indices in ŽA.1. have to be taken modulo 6, and due to momentum
conservation one has s456 s s123 , s612 s s345 , s561 s s234 . The nine Lorentz invariants are
not independent since they fulfill the constraint detŽ G . s 0. We find
6 rG
I6n s Ý 2 Ž
A k q Bk . q C k q D k ,
ks1 e
ye
Ž ys12 .
A1 s ,
s61 s12 s23 s234
2 ye ye
Ž s12 y s123 . Ž ys12 . y Ž ys123 .
B1 s
s12 s23 s34 s123 E1
2 ye ye
Ž s12 y s345 . Ž ys12 . y Ž ys345 .
q ,
s56 s61 s12 s345 E1
5
The last term in the curly bracket of Eq. Ž5.8. should read w1yŽ m25 s23 r s45 r s51 .ye x.
1 s45 y s345 s45 y s123 s123 s345

C1 s b 1
ž s34 s45 s56
q
s34 s45 E1
q
s45 s56 E1 / žLi 2 1 y /
s12 s45
2
b1 p2 Ž s123 y s12 . s12 s123
y
s34 s45 s56 3
q
s12 s23 s34 s123 E1 ž Li 2 1 y
s123 / ž y Li 2 1 y
s12 /
2
Ž s345 y s12 . s12 s345
q
s56 s61 s12 s345 E1
Li 2 1 y ž s345 / žy Li 2 1 y
s12 / ,
1 1 s12 y s123 s12 y s345

D1 s y
2
b4
ž s61 s12 s23
q
s12 s23 E1
y
s61 s12 E1 /
1 s61 y s234 s61 y s345
qb 3
ž s56 s61 s12
q
s56 s61 E3
y
s61 s12 E3 /
s12 1 1 1 1
=log 2 ž / s61
y
2
b3
ž s56 s61 s12
y
s56 s61 s345
y
s61 s12 s234 /
s12 y s345 s12 y s123 s56 y s234 s56 y s123
qb1 ž s56 s345 E1
y
s34 s123 E1 / ž
q b5
s12 s234 E2
y
s34 s123 E2 /
s12 1 1 s56 y s123 s56 y s234
=log 2 ž / s56
y
2
b2
ž s45 s56 s61
q
s45 s56 E2
q
s56 s61 E2 /
1 s23 y s123 s23 y s234
qb5
ž s12 s23 s34
q
s12 s23 E2
q
s23 s34 E2 /
s123 1 s12 y s123 1 s12 y s345
=log 2 ž / s234
y
2
b1 ž / ž
s34 s123 E1
y b4
s61 s12 s23
q
s61 s12 E1 /
s56 y s123 s23 y s123 s12
qb5 ž s34 s123 E2
q
s12 s23 E2 / ž /
log 2
s123
1 1 1 s61 y s234 s34 y s234

y
2
b3
ž s61 s12 s234
y
s56 s61 s12
y
s56 s61 E3
y
s12 s234 E3 /
1 1 s23 y s234 s56 y s234 s12
qb5
ž s12 s23 s234
y
s12 s23 s34
y
s23 s34 E2
y
s12 s234 E2 / ž /
log 2
s234
1 s12 y s345 1 s12 y s123

y
2
b1 ž s56 s345 E1 / ž
y b4
s61 s12 s23
q
s12 s23 E1 /
s34 y s345 s61 y s345 s12
qb 3 ž s56 s345 E3
q
s61 s12 E3 / ž /
log 2
s345
, Ž A.2 .
with the abbreviations
E1 s E4 s s123 s345 y s12 s45 ,
E2 s E5 s s234 s123 y s23 s56 ,
E3 s E6 s s345 s234 y s34 s61 ,
and
G Ž1qe . G 2 Ž1ye .
rG s .
G Ž1y2e .
The functions A k , Bk ,Ck , D k for k ) 1 are obtained by cyclic permutations of the

indices. The other necessary ingredients are
2 2 2
b 1 s Ž ys345 s56 s23 q s12 s234 s45 s34 y s56 s23 s345 s45 s34 y s45 s234 s345 s123 s34
2
qs34 s45 s61 s123 y s34 s123 s12 s234 s45 y s34 s123 s56 s23 s345 q 2 s34 s45 s23 s12 s56
2 2 2 2
qs34 s123 s345 s234 y s34 s61 s123 y s34 s345 s56 s123 s61 q s345 s56 s123 s234
2 2
ys45 s234 s123 s56 s345 q s45 s234 s12 s56 q s56 s23 s345 s45 y s56 s34 s45 s61 s123
2 2 2 2
q2 s34 s12 s56 s61 s45 y s12 s56 s45 s234 s345 y s12 s234 s45 y s123 s234 s345
2 2
qs123 s56 s23 s345 q 2 s45 s345 s234 s123 s12 y 2 s34 s45 s12 s56 q 2 s45 s34 s123 s56 s345
2
ys45 s34 s61 s123 s12 y s45 s56 s23 s345 s12 q s34 s123 s61 s345 . rF , Ž A.3 .
where the other b k ’s are again obtained by cyclic permutation. Finally,
2 2 2
Fs y s345 s56 s23 q 4 s23 s34 s12 s56 s61 s45 y 2 s23 s34 s345 s56 s123 s61
2 2 2 2
y2 s23 s12 s56 s45 s234 s345 q 2 s23 s345 s56 s123 s234 y s123 s61 s34
2 2 2 2
y2 s34 s12 s123 s234 s61 s45 q 2 s34 s123 s61 s345 s234 y s45 s234 s12
2 2 2 2
ys345 s123 s234 q 2 s12 s123 s234 s45 s345 s 64det Ž S . . Ž A.4 .
Note that in the e-dependent part of the formula no b’s appear. We did not succeed in
finding a more compact form for the part which does not depend on e without spoiling
this nice feature.
The expression for I6n in the form as given above is strictly only correct in the
Euclidean region where all Mandelstam variables are negative. For most of the terms,
the analytic continuation to positive values is defined by simply using the replacement
s ™ s q i d , where s stands here for any of the si j , si jk . No cut will be hit by the
logarithms, the dilogarithms with a single ratio of Mandelstam variables and the
exponentials Žys y i d .ye . Concerning the dilogarithms of a product of ratios, more

care has to be taken. To avoid the crossing of a cut one has to make the replacement
s1 s 2 s1 q i d s 2 q i d s1 q i d s 2 q i d
Li 2 1 yž s 3 s4 / ž
™ Li 2 1 y
s 3 q i d s4 q i d / ž
qh ,
s 3 q i d s4 q i d /
s1 q i d s 2 q i d
=log 1 y
ž s 3 q i d s4 q i d / , Ž A.5 .
h Ž x , y . s log Ž xy . y log Ž x . y log Ž y . . Ž A.6 .
Appendix B. Recursion relation for integrals with Feynman parameters in the

numerator
Here we want to comment on the reduction of Feynman parameter integrals with

nontrivial numerators as defined in Ž33.. In Ref. w25x recursion relations for integrals
with up to 4 Feynman parameters are given. The derivation is based on the approach of
w5,6x which uses a projective transformation w30x. We present an independent derivation
and generalize the formulas to arbitrary N and numbers of Feynman parameters in the
numerator.
Consider the following identity Ž j s 1, . . . , N y 1.:
` E N N
yN qnr2q1
Hy`d N
z
E zj ž Ł u Ž zl . d
ls1
N
ž 1y Ý zk
ks1
/ z l 1 . . . z l pŽ z P S P z .
Ny 1
/ s 0 . Ž B.1 .
Setting rN s 0 and using Ý ks1 z k s 1, one obtains z P S P z s Ý k,ls1 z l G l k z kr2 y

Ny 1
Ý ks 1 Õ k z k . The first step consists in eliminating the d-function by integrating out z N . If
now the derivative acts on u-functions, terms with d-function insertions are produced.
n
The reduced integrals INy 1, j defined in Eq. B.2 below , obtained by pinching the jth
Ž Ž . .
propagator line in an N-point graph, correspond to these d-function insertions. If the
derivative acts on the Feynman parameters, the monomial in the numerator is reduced by
one degree, and if it acts on the z P S P z term it formally decreases the dimension by two
and increases the numerator by Ž G P z . j y Õj . After these operations one reintroduces the
z N -integration with a delta-function insertion. Using the following generalized definition
for pinched scalar integrals with nontrivial numerators:
` N
Ny 1
n
INy 1, j Ž l 1 , . . . ,l p . s Ž y1 . G Ž N y 1 y nr2 . H0 d N
z d 1y
ž Ý zl
ls1
/
z l 1 . . . z l pd Ž z j .
= Ny 1ynr2
, Ž B.2 .
Ž zPSPz.
we find
Ny1 p
Ý Gjl 0 INn Ž l 0 , . . . ,l p . s Ý d jl INnq2 ž l1 , . . . ,lˆk , . . . l p / q INy1,

k
n
N Ž l 1 , . . . ,l p .
l 0s1 ks1
n n
y INy 1, j Ž l 1 , . . . ,l p . q Õ j IN Ž l 1 , . . . ,l p . . Ž B.3 .
Herein lˆk means that the respective index does not appear.
Now we want to sketch the proof for a second equation,
Ny1
n nq2
INy 1, N Ž l 1 , . . . ,l p . s Ž N y n y p y 1 . IN Ž l1 , . . . ,l p . q Ý Õ l 0 INn Ž l 0 , . . . ,l p . .
l 0s1
Ž B.4 .
The proof is done by induction to p. In order to show that the induction start Ž p s 0.
holds true, one directly calculates
n
k2 Ny1
INy 1, N Ž 1 . s d k H N s Ž N y n y 1 . INnq2 Ž 1 . q Ý Õl 0 INn Ž l 0 . . Ž B.5 .
l 0s1
Ł ql2
ls1
For the induction step, one assumes that ŽB.4. is fulfilled for a given p. Viewing the Õj
as independent variables, one differentiates the formula with respect to Õ l pq 1 and finds
the formula for p q 1 and n y 2. As the dimension is arbitrary we can replace n y 2 by
n in the expression. This proves the validity of Eq. ŽB.4. for all p.
Combining ŽB.3. and ŽB.4. one finds Ž j s 1, . . . , N .
N p
2 Ý S jl 0 INn Ž l 0 , . . . ,l p . s Ý d jl INnq2 ž l1 , . . . ,lˆk , . . . ,l p /
k
l 0s1 ks1
q Ž N y n y p y 1 . INnq 2 Ž l 1 , . . . ,l p . y INy1,
n
j Ž l 1 , . . . ,l p . .
Ž B.6 .
In the case N F 6, S is invertible and the final reduction formula reads

p
INn Ž l 0 , . . . ,l p . s 12 Ý Sy1
ks1
nq2
l l IN ž l1 , . . . ,lˆk , . . . ,l p /
0 k
N
q 12 Ý Sy1 nq2
jl Ž N y n y p y 1 . IN
0
Ž l1 , . . . ,l p .
js1
N
y 12 Ý Sy1 n
jl INy1, j Ž l 1 , . . . ,l p . .
0
Ž B.7 .
js1
This is the generalization of the formulas given in Ref. w25x for the case of four Feynman
parameters in the numerator Ž p s 3.. ŽIn their conventions Sy1 i k s hi k a i a k rNn and
Ý i Sy1
i k s g k rNn , which is identical to our S
y1
up to a trivial relabeling of indices.. It
means that all Feynman parameter integrals with numerators can be reduced to ordinary
scalar integrals by iteration.
In the case N ) 6, S is not invertible. The linear dependence of the row Žresp.
column. vectors of S makes it difficult to generalize the Feynman parameter space
based techniques to arbitrary N w5,6,17,18x. In our approach one can use Eq. ŽB.6.
directly in this case, where the inversion of the Gram matrix should be done with its
pseudo-inverse as explained in the main text. As already noted earlier, it is also possible
to work with the pseudo-inverse to S. In any way, as we have proven, it is the Žpseudo.
inverse to the Gram matrix which induces the cancellation of the higher dimensional
terms by virtue of Eq. Ž39..
Finally, with the formulas given above and in the main text, it is possible to translate
expressions in Feynman parameter space into expressions in momentum space and vice
versa for arbitrary N.
Appendix C. Derivation of the recursion formula for tensor reduction
In this appendix we give the derivation of the recursion relation Ž42.,
1 m 1 . . . m L4
K Nm 1 . . . m L s W PK N Ly1 dots 4
L
2 Ž Ly1. m 1 . . . m L4 n . . . n
1 Ly 1
Nyl . Ž C.1 .
L!
To Eq. Ž36.,
L
m 1 . . . m L4 n
K Nm1 . . . m L s W m 1 . . . W m L INn q Ý WPŽ Lyk . KlP1 . . . KlPk INyk , Nyl 1 , . . . , Nyl k ,
ks1
we apply Eq. Ž41.,
ky1 2j
n
INy k , Nyl 1 , . . . , Nyl k s Ý Ž y1. ky jy1 j!
ÕPŽ kyjy1 . rPn 1 . . . rPn j In1 . . . n j
l 1 . . . l ky 1 4 Ny1, Nyl k
,
js0
and get, by taking into account combinatorial factors from contracting the tensor
brackets
L ky1 j
1 Lyjy1 2
K Nm 1 . . . m L s W m 1 . . . W m L INn q Ý k
Ý Ž y1. ky jy1 ž k y j y 1 j! /
ks1 js0
m 1 . . . m L4 n . . . n
= WPŽ Ly jy1 . KlPHnP1 . . . HnPj INy1,
1 j
Nyl . Ž C.2 .
Using the fact that

L ky1 L Ly1 Ly2 L
Ý Ý ak j s Ý Ý u Ž j F k y 1 . a k j s a L , Ly1 q Ý Ý ak j , Ž C.3 .
ks1 js0 ks1 js0 js0 ksjq1
we can separate terms with and without W P-vectors,

ky jy1
L Ly2 2 j Ž y1 . Lyjy1
K Nm1 . . . m L s W m 1 . . . W m L INn q Ý Ý
ks1 js0
u Ž j F k y 1.
j! k ž kyjy1 /
m 1 . . . m L4 n . . . n
P WPŽ Ly jy1 . HnP1 . . . HnPj KlP INy1,
1 j
Ny l
2 Ly 1 m 1 . . . m L4 n . . . n
q HnP1 . . . HnPLy 1 KlP INy1,
1 Ly 1
Nyl . Ž C.4 .
L!
Now one has to rearrange the terms containing W-vectors.

First we write 1rk in ŽC.4. as 1rk s Ž L y k .rŽ Lk . q 1rL. The 1rL part vanishes
since the sum over k just represents Ž1 y 1. Ly jy1 s 0. The Ž L y k .rŽ Lk . part makes
the upper bound of the k-sum to be L y 1. From the tensor bracket a W P can be
factored in a symmetric way,
m 1 . . . m L4
WPŽ Ly jy1 . HnP1 . . . HnPj KlP
W m1 m 2 . . . m L4
s WPŽ Ly jy2 . HnP1 . . . HnPj KlP
Lyjy1
W mL m 1 . . . m Ly 1 4
q PPP q WPŽ Ly jy2 . HnP1 . . . HnPj KlP .
Lyjy1
Hence Eq. ŽC.4. is equivalent to
2 Ly 1 m 1 . . . m L4 n . . . n
K Nm1 . . . m L y HnP1 . . . HnPLy 1 KlP INy1,
1 Ly 1
Nyl
L!
ky jy1
W m1 Ly1 ky1 2 j Ž y1 . Lyjy2
s
L ž W m 2 . . . W m L INn q Ý Ý
ks1 js0 j! k ž kyjy1 /
m 2 . . . m L4 n . . . n
P WPŽ Ly jy2 . HnP1 . . . HnPj KlP INy1,
1 j
Nyl
/
q Ž L y 1 . permutations . Ž C.5 .
Comparing with Eq. ŽC.2. we see that the right-hand side of Eq. ŽC.5. is just
W m1 1 m 1 . . . m L4
K Nm 2 . . . m L q Ž L y 1 . permutationss W PK N Ly1 dots 4 . Ž C.6 .
L L
Combining ŽC.5. and ŽC.6. we obtain the recursion relation ŽC.1..

Appendix D. Explicit reduction for N s 2,3,4,5,6
We now give explicit formulas for tensor integrals up to N s 6, and rank L F N.

They have been derived by applying Eqs. Ž43., Ž44. and Ž37.. The reduction stops
because tadpole integrals of massless propagators are zero, I1 , I1m s 0. This shows that
any tensor integral with an arbitrary number of legs can be expressed in terms of scalar
integrals only. For non-exceptional kinematics higher dimensional integrals INnqX 2m Žms
X
1,2. occur only with N F 4 in the reduction of arbitrary N-point tensor integrals, as has
been explained above. All the higher dimensional integrals can be remapped to
n-dimensional integrals with the scalar reduction formulas Ž20., Ž22. and Ž24.. Explic-
itly, the expressions for the higher dimensional integrals are given by
m1 m 2 m1 m 2
JNs 2,3,4 s Ž y1 . Ž g r2 y H m1 m 2 . INnq 2 ,
PP m 1 m 2 m 34
m1 m 2 m 3 P
JNs 3,4 s Ž y1 . Ž gr2 y H . W INnq2
PP m 1 m 2 m 34
y Ž gr2 y H . KlP nq2
INy1, Nyl ,
PP m1 m 2 m 3 m 44
J4m1 m 2 m 3 m 4 s Ž gr2 y H . Ž 2 . I4nq4
PP m1 m 2 m 3 m44
y Ž gr2 y H . WŽP2 . I4nq2
PP m1 m 2 m 3 m 44
y 12 Ž gr2 y H . W PKlP nq2
I3,4yl
PP m1 m 2 m 3 m44
n , nq2
y Ž gr2 y H . HnP KlP I3,4yl . Ž D.1 .
Note that rank one, Ž n q 2.-dimensional three-point functions appear. As the reduction
rules do not depend on the space-time dimension all the formulas given above are also
valid for higher dimensional tensor integrals. The tensor structure is carried by the
well-defined 4-dimensional objects W m, Klm , H mn given in Eq. Ž38.. Note that metric
tensors occur only in the combination Ž gr2 y H . as coefficients of the higher dimen-
sional integrals INnq 2 m .
For N G 5, the equal signs are only valid up to O Ž e . since the JNm1 . . . m L terms have
n1 . . . n L Ž .
been dropped. The shift of the integrals INy 1, Nyl R has been done explicitly up to
N s 3 in order to give examples for the algorithm defined in Ž44.. How to proceed for
N ) 3 then should be obvious.
The reduction rules can be implemented easily in algebraic manipulation programs.
Ns2
Necessary for a non-vanishing result is that the external momentum is not light-like.
The Gram matrix is trivial. One has W m Ž r . s r mr2, H m 1 m 2 s r m1 r m 2rŽ2 r P r .,
r m1
I2m1 Ž r . s I2n Ž r . ,
2
1
I2m1 m 2 Ž r . s Žnrm 1 r m 2 y r 2 g m 1 m 2 . I2n Ž r . . Ž D.2 .
4 Ž n y 1.
Ns3
We use the short-hand notation for the arguments of the tensors and integrals defined
in Eq. Ž45. in the following: R s w r 1 ,r 2 ,0x, R w1x s w r 2 y r 1 ,0x, Rˆw1x s w r 2 ,0x, etc. The
arguments for the tensors W m, Klm, H mn are not written explicitly, e.g. it is always
understood that W m s W m Ž R ., etc.,
2
I3m Ž R . s W m I3n Ž R . q Ý Klm ž I2n Ž R w l x . y I2n Ž Rˆ w l x . / ,
ls1
2
I3m1 m 2 Ž R . s 12 Ž W m 1 I3m 2 Ž R . q W m 2 I3m 1 Ž R . . q Ý Ž Klm Hnm 1 2 q Klm 2 Hnm 1 .
ls1
= ž r ln I2n Ž R w l x q I2n
. Ž R w l x y I2n
. Ž Rˆ w l x . /
y Ž g m1 m 2r2 y H m1 m 2 . I3nq 2 Ž R . ,
I3m1 m 2 m 3 Ž R . s 13 Ž W m 1 I3m 2 m 3 y J3m 2 m 3 4
W m 2 I3m1 m 3 y J3m1 m 3 4 q W m 3 I3m1 m 2 y J3m1 m 2 4 .

qW
2
m 1 m 2 m 34 n1 n 2 n
q 23 Ý KlPHnP1 HnP2 žr l r l I2 Ž Rw l x.
ls1
qrln 1 I2n 2 Ž R w l x . q r ln 2 I2n 1 Ž R w l x . q I2n 1 n 2 Ž R w l x . y I2n 1 n 2 Ž Rˆ w l x . /

PP m 1 m 2 m 34
y Ž gr2 y H . W P I3nq2 Ž R .
PP m 1 m 2 m 34
y Ž gr2 y H . KlP nq2
I2,3yl Ž R. . Ž D.3 .
Ns4
n1 . . . n L
R s w r 1 ,r 2 ,r 3 ,0x. The integrals I3,4yl are given in unshifted form. The integrals
J4m1 . . . m L
are given in Eq. ŽD.1.,
3
I4m1 Ž R . s W m 1 I4n Ž R . q Ý Klm I3,4yl
n
Ž R. ,
1
ls1
3
m1 m 24 n
I4m1 m 2 Ž R . s 12 Ž W m 1 I4m 2 Ž R . q W m 2 I4m 1 Ž R . . q Ý KlPHnP I3,4yl Ž R.
ls1
q J4m1 m 2 ,
I4m1 m 2 m 3 Ž R . s 13 Ž W m 1 I4m 2 m 3 y J4m 2 m 3 4 q W m 2 I4m 1 m 3 y J4m 1 m 3 4

3
m 1 m 2 m 34 n n
W m 3 I4m 1 m 2 y J4m 1 m 2 4 . q 23
qW Ý KlPHnP1 HnP2 I3,4yl
1 2
Ž R.
ls1
q J4m1 m 2 m 3 ,
I4m1 m 2 m 3 m 4 Ž R . s 14 Ž W m 1 I4m 2 m 3 m 4 y J4m 2 m 3 m 4 4 q W m 2 I4m 1 m 3 m 4 y J4m 1 m 3 m 4 4
W m 3 I4m1 m 2 m4 y J4m1 m 2 m4 4 q W m4 I4m1 m 2 m 3 y J4m1 m 2 m 3 4 .

qW
3
m1 m 2 m 3 m 44 n n n
q 13 Ý KlP HnP1 HnP2 HnP3 I3,4yl
1 2 3
Ž R.
ls1
q J4m1 m 2 m 3 m4 . Ž D.4 .
Ns5
As discussed above, no higher dimensional integrals appear in the reduction of tensor

N-point integrals for N ) 4, as long as the external momenta are non-exceptional. They
occur only implicitly by reduction of 2,3,4-point tensor integrals. Now, R s
w r 1 ,r 2 ,r 3 ,r4 ,0x. The formulas are valid up to O Ž e . since J5 m 4 terms have been dropped,
4
I5m1 Ž R . s W m 1 I5n Ž R . q Ý Klm 1 n
I4,5yl Ž R. ,
ls1
4
m1 m 24 n
I5m1 m 2 Ž R . s 12 Ž W m 1 I5m 2 Ž R . q W m 2 I5m 1 Ž R . . q Ý KlPHnP I4,5yl Ž R. ,
ls1
I5m1 m 2 m 3 Ž R . s 13 Ž W m 1 I5m 2 m 3 q W m 2 I5m 1 m 3 q W m 3 I5m 1 m 2 .

4
m 1 m 2 m 34 n n
q 23 Ý KlPHnP1 HnP2 I4,5yl
1 2
Ž R. ,
ls1
I5m1 m 2 m 3 m 4 Ž R . s 14 Ž W m 1 I5m 2 m 3 m 4 q W m 2 I5m 1 m 3 m 4 q W m 3 I5m 1 m 2 m 4 q W m 4 I5m 1 m 2 m 3 .

4
m1 m 2 m 3 m 44 n n n
q 13 Ý KlPHnP1 HnP2 HnP3 I4,5yl
1 2 3
Ž R. ,
ls1
I5m1 m 2 m 3 m 4 m 5 Ž R . s 15 Ž W m 1 I5m 2 m 3 m 4 m 5 q . . . qW
W m 5 I5m 1 m 2 m 3 m 4 .
4
m 1 m 2 m 3 m 4 m 54 n n n n
q 152 Ý KlPHnP1 HnP2 HnP3 HnP4 I4,5yl
1 2 3 4
Ž R. .
ls1
Ž D.5 .
Ns6
For non-exceptional kinematics any set of four vectors out of r 1 ,r 2 ,r 3 ,r4 ,r54 spans
4-dimensional Minkowski space. We will express the tensor 6-point functions in terms
of scalar 6-point functions and pentagon integrals. The latter then can be reduced further
by using Ž29. and the reduction formulas for N s 5 with the corresponding new
argument vectors,
5
I6m1 Ž R . s W m 1 I6n Ž R . q Ý Klm 1 n
I5,6yl Ž R. ,
ls1
5
m1 m 24 n
I6m1 m 2 Ž R . s 12 Ž W m 1 I6m 2 Ž R . q W m 2 I6m 1 Ž R . . q Ý KlPHnP I5,6yl Ž R. ,
ls1
I6m1 m 2 m 3 Ž R . s 13 Ž W m 1 I6m 2 m 3 q W m 2 I6m 1 m 3 q W m 3 I6m 1 m 2 .

5
m 1 m 2 m 34 n n
q 23 Ý KlPHnP1 HnP2 I5,6yl
1 2
Ž R. ,
ls1
I6m1 m 2 m 3 m 4 Ž R . s 14 Ž W m 1 I6m 2 m 3 m 4 q W m 2 I6m 1 m 3 m 4 q W m 3 I6m 1 m 2 m 4 q W m 4 I6m 1 m 2 m 3 .

5
m1 m 2 m 3 m 44 n n n
q 13 Ý KlPHnP1 HnP2 HnP3 I5,6yl
1 2 3
Ž R. ,
ls1
I6m1 m 2 m 3 m 4 m 5 Ž R . s 15 Ž W m 1 I6m 2 m 3 m 4 m 5 q . . . qW
W m 5 I6m 1 m 2 m 3 m 4 .
5
m 1 m 2 m 3 m 4 m 54 n n n n
q 152 Ý KlPHnP1 HnP2 HnP3 HnP4 I5,6yl
1 2 3 4
Ž R. ,
ls1
I6m1 m 2 m 3 m 4 m 5 m 6 Ž R . s 16 Ž W m 1 I6m 2 m 3 m 4 m 5 m 6 q . . . qW
W m 6 I6m 1 m 2 m 3 m 4 m 5 .
5
m1 m 2 m 3 m4 m 5 m 64
q 452 Ý KlPHnP1 HnP2 HnP3 HnP4 HnP5
ls1
n 1n 2 n 3 n 4 n 5
=I5,6yl Ž R. . Ž D.6 .
References
w1x M. Mangano, S. Parke, Phys. Rep. 200 Ž1991. 301.

w2x P. Draggiotis, R.H.P. Kleiss, C.G. Papadopoulos, Phys. Lett. B 439 Ž1998. 157.
w3x F. Caravaglios, M. Mangano, M. Moretti, Nucl. Phys. B 539 Ž1999. 215.
w4x R.K. Ellis, J. Sexton, Nucl. Phys. B 269 Ž1986. 445.
w5x Z. Bern, L. Dixon, D.A. Kosower, Phys. Lett. B 302 Ž1993. 299.
w7x ´ ´
Z. Kunszt, A. Signer, Z. Trocsanyi, Phys. Lett. B 336 Ž1994. 529.
w8x Z. Bern, L. Dixon, D.A. Kosower, S. Weinzierl, Nucl. Phys. B 489 Ž1997. 3.
w10x J.M. Campbell, E.W.N. Glover, D.J. Miller, Phys. Lett. B 409 Ž1997. 503.
w11x E.W.N. Glover, D.J. Miller, Phys. Lett. B 396 Ž1997. 257.
w12x G. Passarino, M. Veltman, Nucl. Phys. B 160 Ž1979. 151.
w13x G.J. van Oldenbourg, J.A.M. Vermaseren, Z. Phys. C46 Ž1990. 425.
w14x W.L. van Neerven, J.A.M. Vermaseren, Phys. Lett. B 137 Ž1984. 241.
w15x ¨
A. Signer, Ph.D. thesis, Diss. ETH Nr. 11143, Zurich 1995.
w16x A.I. Davydychev, Phys. Lett. B 263 Ž1991. 107.
w17x O.V. Tarasov, Phys. Rev. D 54 Ž1996. 6479.
w18x J. Fleischer, F. Jegerlehner, O.V. Tarasov, hep-phr9907327.
w19x G. Devaraj, R.G. Stuart, Nucl. Phys. B 519 Ž1998. 483.
w20x R.G. Stuart, Comput. Phys. Commun. 48 Ž1988. 367.
w21x R. Pittau, Comput. Phys. Commun. 104 Ž1997. 23.
w22x R. Pittau, Comput. Phys. Commun. 111 Ž1998. 48.
w23x S. Weinzierl, Phys. Lett. B 450 Ž1999. 234.
w24x A. Denner, Fortsch. Phys. 41 Ž1993. 307.
w26x D.B. Melrose, Il Nouvo Cimento 40A Ž1965. 181.
w27x ¨
M. Kocher, Lineare Algebra und Analytische Geometrie, 2nd ed. ŽSpringer, Berlin, 1985..
w28x A. Ben-Israel, Th.N.E. Greville, Generalized Inverses ŽWiley, New York, 1974..
w29x A. Ghinculov, Phys. Lett. B 385 Ž1996. 279.
w30x G.’t Hooft, M. Veltman, Nucl. Phys. B 153 Ž1979. 365.
Action principles, restoration of BRS symmetry and the

renormalization group equation for chiral non-Abelian gauge
theories in dimensional renormalization with a
non-anticommuting g 5
´ D. Sanchez-Ruiz
C.P. Martın, ´
´
Departamento de Fısica ´
Teorica I, UniÕersidad Complutense, 28040 Madrid, Spain
Received 12 May 1999; accepted 3 August 1999
Abstract
The one-loop renormalization of a general chiral gauge theory without scalar and Majorana
fields is fully worked out within Breitenlohner and Maison dimensional renormalization scheme.
The coefficients of the anomalous terms introduced in the Slavnov–Taylor equations by the
minimal subtraction algorithm are calculated and the asymmetric counterterms needed to restore
the BRS symmetry, if the anomaly cancellation conditions are met, are computed. The renormal-
ization group equation and its coefficients are worked out in the anomaly free case. The
computations draw heavily from the existence of action principles and BRS cohomology theory.
q 2000 Elsevier Science B.V. All rights reserved.
PACS: 11.30.Rd
1. Introduction
Dimensional Regularization w1–3x is the standard regularization method in four-di-

mensional perturbative quantum field theory as applied to particle physics. Both its
axiomatics and properties were rigorously established long ago w4–11x and quite a
number of computational techniques based on the method have been developed over the
years w12–18x. Involved multiloop computations of the parameters of the Standard
Model, a must due to the availability of high precision tests of the model in particle
accelerators, have been carried out in the simplest possible setting thanks to dimensional
regularization w19–25x. The success of Dimensional Regularization as a practical regular-
´ ., domingos@eucmos.sim.ucm.es
E-mail addresses: carmelo@elbereth.fis.ucm.es ŽC.P. Martın
´
ŽD. Sanchez-Ruiz..
PII: S 0 5 5 0 - 3 2 1 3 Ž 9 9 . 0 0 4 5 3 - 8
388 ´ D. Sanchez-Ruizr
C.P. Martın, ´ Nuclear Physics B 572 (2000) 387–477
ization method stems from the fact that it preserves the BRS symmetry of vector-like
non-supersymmetric gauge theories without distorting neither the shape of the integrand
of a regularized Feynman diagram nor the properties of the algebraic objects involved
Žno chiral objects being present.. The minimal subtraction scheme, the famous MS
scheme w12x, then leads w3x to a renormalized BRS-invariant theory w4,7x. Counterterms
are generated by multiplicative renormalization of the tree-level Lagrangian, so that b
functions and anomalous dimensions can be computed easily w11x.
It is an empirical fact that electroweak interactions are chiral and hence vector-like
gauge theories does fail to account for them. It turn thus out that chiral gauge theories
are of key importance to understanding Nature. Unfortunately, Dimensional Regulariza-
tion loses its smartness when applied to chiral gauge theories. The algebraic properties
that the matrix g 5 has in four dimensions cannot be maintained without introducing
algebraic inconsistencies as we move away from four dimensions w26x. Hence, within the
framework of Dimensional Regularization, the definition of the object g 5 in ‘‘d
complex dimensions’’ demands a new set of algebraic identities. There exists such a set
of identities, they were introduced in Ref. w4x, following Refs. w3,27–29x, and they entail
that the object g 5 anticommutes no longer with the object gm . The axiomatics of
dimensional regularization so established is the only one which has been shown to be
thoroughly consistent at any order in perturbation theory if cyclicity of the trace is not
given up Žsee Ref. w30x for the non-cyclic trace alternative.. Minimal subtraction w12x of
the singular part of the dimensionally regularized Feynman diagrams, a subtraction
procedure known as minimal dimensional renormalization w4–7x, leads to a renormalized
quantum field theory that satisfies Hepp axioms w35x of renormalization theory. Field
equations, the action principles and Zimmerman–Bonneau identities hold in dimensional
renormalization. This has been rigorously shown in Refs. w4–8,31x and it constitutes one
of the key ingredients of the modern approach to the quantization of gauge theories by
using BRS methods w32–34x.
That Dimensional Regularization can be used along with algebraic BRS techniques is
of the utmost importance, for this regularization method and the minimal subtraction
algorithm that comes with it break, generally speaking, chiral gauge symmetries. This
breaking gives rise to both physical and non-physical anomalies w8,36x. Physical
anomalies in the currents of the Lagrangian impose anomaly cancellation conditions
w39x, lest the theory ceases to make sense as a quantum theory. These anomalies belong
to the cohomology of the Slavnov–Taylor operator that governs the chiral gauge
symmetry at the quantum level and they are not artifacts of the regularization method
nor due to the renormalization scheme employed. Non-physical anomalies correspond to
trivial objects in the cohomology of the BRS operator and, therefore, they can be set to
zero by an appropriate choice of finite counterterms w36x. Non-physical anomalies are
artifacts either of the regularization method, the renormalization method or both.
Notwithstanding the unique status that as a thoroughly consistent framework the
Breitenlohner and Maison scheme enjoys among the dimensional renormalization pre-
scriptions, there is, up to the best of our knowledge, no complete one-loop study of a
general non-Abelian chiral gauge theory, let alone the Standard Model. Most of the
implementations of dimensional regularization in quantum field theories which involve
the matrix g 5 and are anomally free, e.g. the Standard Model, take a fully anticommut-
ing g 5 in ‘‘d complex dimensions’’. This is the so-called ‘‘naive’’ prescription for the
´ D. Sanchez-Ruizr
C.P. Martın, ´ Nuclear Physics B 572 (2000) 387–477 389
object g 5 w37,38x. The purpose of this paper is to remedy this situation by carrying out
such one-loop study for a general chiral gauge theory with neither scalar nor Majorana
fields. The inclusion of scalar fields will be discussed elsewhere.
The layout of this paper is as follow. In Section 2 we give a quick account of the
dimensional renormalization algorithm a´ la Breitenlohner and Maison, the Regularized
and Quantum Action Principles and Bonneau identities. Section 3 is devoted to a
thorough study of the one-loop dimensional renormalization of a general chiral gauge
theory for a simple gauge group. In this section the BRS anomaly is computed for the
first time in the Breitenlohner and Maison framework. Of course, the standard anomaly
cancellation conditions are thus obtained. The beta function and the anomalous dimen-
sions of the theory are computed in this section as well. Again, the one-loop renormal-
ization group equation for the model at hand had never been work out as yet in
Breitenlohner and Maison scheme. In Section 4 we adapt the results obtained in the
previous section to the case of a non-simple gauge group. We also include three
appendices. In Appendices A and B we show how to use the action principles to obtain
the breaking of the Slavnov–Taylor identities. In Appendix C we compare, for a simple
gauge group, the one-loop renormalized chiral gauge theory obtained by means of the
Breitenlohner and Maison scheme with the corresponding one-loop renormalized theory
obtained with the help of a fully anticommuting g 5 in ‘‘d complex dimensions’’.
2. Dimensional renormalization: notation and general results
As explained in the Introduction, we will use the formulation of dimensional

renormalization given by Breitenlohner and Maison w4–6x, because it is a systematic and
consistent procedure Žvalid in the sense of Hepp w35x to all orders in perturbation theory.
in which tools like field equations, action principles and Ward identities can be
rigorously implemented. This applies specially to the treatment of g 5 .
We will also compare Breitenlohner and Maison formulation with the ‘‘naive’’
prescription which sets the object g 5 to anticommute with the object gm w37,38x.
2.1. The Breitenlohner and Maison ‘‘d-dimensional coÕariants’’
Breitenlohner and Maison define the usual d-dimensional ‘‘Lorentz Covariants’’

Ž gmn , pm , gm , etc.. to be formal objects obeying the standard algebraic identities that
they would satisfy in spaces of integral dimension w4x Ž d is a complex number, indices
do not take any value and questions like ‘‘Lorentz invariance’’ are meaningless!..
Besides the ‘‘d-dimensional’’ metrics gmn , they introduce a new one, gˆmn , which can
be considered as a ‘‘Ž d y 4.-dimensional covariant’’. Moreover, the e tensor is consid-
ered to be a ‘‘four-dimensional covariant’’ object Žbecause of the axiom in Eq. Ž57.
below..
The symbols are required to obey Žapart from obvious rules concerning contractions
of indices, addition, multiplication by numbers, commutation of some symbols, etc..:
g mn pn s p m , gˆ mn pn s pˆ m , g mngn s g m , gˆ mngn s gˆ m ,
gmn g n r s gm r , gˆmn g n r s gˆmn gˆ n r s gˆm r , Ž 1.
m m n mn
g m s d, Tr II s 4, g ,g 4 s 2 g II;
4
em1 . . . m 4 en 1 . . . n 4 s y Ý sign p Ł Ž gm inp Ž i. y gˆm inp Ž i. . , Ž 2.
pg S4 is1
where S4 denotes the permutation group of four objects, and II is the unit of the
symbolic algebra of gammas.
With the definitions
g mn s g mn y gˆ mn , g mn pn s p m , g mngn s g m ,
i
g5 s em 1 . . . m 4 g m1 . . . g m 4 Ž 3.
4!
and the assumption of cyclicity of the symbol Tr, the following properties can be proved
algebraically w4x:
gˆmn g n r s 0, pˆ P q s 0, gmn g n r s gm r , gˆ m m s d y 4, g m m s 4,
g m ,g n 4 s 2 g mn II, gˆ m ,gˆ n 4 s 2 gˆ mn II, em 1... m 4 s sign p emp Ž1. . . . mp Ž4. ,
em1 . . . m 4 gˆ m i n i s 0, em 1 . . . m 4 g m i n i s em 1 . . . m 4 g m i n i ;
Tr g m s Tr g 5 s 0, g 52 s II,
Tr g mg ng 5 s Tr g m g ng 5 s Tr gˆ mgˆ ng 5 s Tr g mgˆ ng 5 s 0,
Tr g m1 . . . g m 4g 5 s Tr g m 1 . . . g m 4g 5 s i Tr II em 1 . . . m 4 ; Ž 4.
g 5 ,g m 4 s g 5 ,gˆ m 4 s 2g 5gˆ m s 2gˆ mg 5 ,

g 5 ,g m s g 5 ,g m s 2g 5g m s y2g mg 5 ,
g 5 ,g m 4 s g 5 ,gˆ m s 0. Ž 5.
g mn can be thought of as a projector over the ‘‘four-dimensional space’’ and gˆ mn as
a projector over the ‘‘Ž d y 4.-dimensional’’ one.
All usual formulae Žnot involving g 5 . used for computing traces of strings of gammas
in terms of combinations of the metric remain valid, even when the gammas are hatted
or barred: in these cases one has only to put hats or bars over the corresponding metrics.
Also the trace of an odd number of Žnormal, hatted or barred. g ’s vanish.
Strings of gammas with contracted indices are simplified with the aid of formulae
like
g mg ngm s Ž 2 y d . g n ,
g m g n gm s y2 g n , g mgˆ n gm s y4 gˆ n ,
gˆ mgˆ ngˆm s Ž 6 y d . gˆ n , gˆ m g ngˆm s Ž 4 y d . g n .

´ D. Sanchez-Ruizr
Of course, when g 5 of the e tensor appears, the situation is very different. In the
‘‘naive’’ prescription, one assumes the anticommutatiÕity of g 5 : gm ,g 5 4 s 0, and the
cyclicity of the trace. From these assumptions one obtains w4,11,26,36x
2 d Tr w g 5 x s 0,
2 Ž d y 2 . Tr w gm1 gm 2 g 5 x s 0 if d / 0,
2 Ž d y 4 . Tr w gm1 . . . gm 4 g 5 x s 0 if d / 0,2; Ž 6.
which has the consequence that Trw gm1 . . . gm 4 g 5 x is identically 0 in the dimensionally
regularized theory Ž d / 4.. So, it also vanishes in the dimensionally renormalized
theory, which is incompatible with the true property Žsee Eq. Ž4.. in four dimensions.
We could introduce the symbol e with the requirement that it satisfies Eq. Ž4., but then
the symbol e would be identically 0. Despite this well known fact, people use the
‘‘naive’’ prescription in e.g. the Standard Model, but they do not set e equal to 0 and at
the end of calculations the symbol e is supposed to have its usual meaning as the
Levi-Civita tensor. This is clearly a mathematical inconsistency Žthe axioms are not
compatible. and when mathematical inconsistencies are present, results obtained from
the axioms are ambiguous and cannot be trusted.
Eq. Ž6. is obtained by computing Tr w gm1 . . . gm m g aga g 5 x in two different ways: first,
one contracts the index a without moving around the g 5 and, second, one anticom-
mutes the ga which is next to the g 5 with this very g 5 and puts it on the left thanks to
the cyclicity of the trace; then, one anticommutes it to the right through the rest of g ’s
until the other g a is met, contraction of the index a is now carried out. So, at least the
trace of six g ’s, having two indices contracted, and a g 5 are needed to obtain this
inconsistency.
Here, the ambiguity in the results is a consequence of the choice of the position of the
g 5 . For example, if the g 5 is not moved in any case,
Tr g m1 . . . g m 4g aga g 5 s d Tr g m 1 . . . g m 4g 5 / yTr ga g m 1 . . . g m 4g ag 5
s Ž 8 y d . Tr g m1 . . . g m 4g 5 , Ž 7.
unless the trace of four g ’s and one g 5 Žor the e tensor. is 0.
Notice that the ambiguity will be always proportional to Ž d y 4. or a Ž d y 4. object
like gˆm , but there are poles in Ž d y 4. in the divergent diagrams that will make finite this
ambiguity in the dimensionally renormalized theory, even in one-loop calculations.
Then, why do people go on using the ‘‘naive’’ prescription in calculations of, e.g.
Standard Model? The key is that the ambiguities are claimed to be always proportional
to the coefficient of the Žchiral gauge. anomaly. So, it would appear that one could
freely use the ‘‘naive’’ prescription in theories with cancellation of anomalies Žlike the
Standard Model.. Calculations to low orders in perturbation theory in some models
support this idea but there is not a rigorous proof of it valid to all orders in perturbation
theory. ŽAlso, it seems difficult to build a consistent theory with inconsistent elements..
We shall close this subsection with a few words regarding the construction of Lorentz
covariants due to Wilson and Collins w11,40x. It is worth mentioning it since it furnishes
explicit expressions for the Lorentz covariants satisfying Eqs. Ž1., Ž2., thus showing that
no inconsistencies arise. In this construction gamma ‘‘matrices’’ are represented as
infinite-dimensional objects. The ‘‘matrix’’ g 5 being defined by Eq. Ž3., with em 1 . . . m 4
as defined by Eq. Ž2.. However, we will not use this construction but the algebraic
approach of abstract symbols by Breitenlohner and Maison.
2.2. Minimal dimensional renormalization and the renormalized theory
Dimensional renormalization a´ la Breitenlohner and Maison is carried out by

applying the pole subtraction algorithm w4–7x as given by the forest formula to each
dimensionally regularized 1PI Feynman diagram. This algorithm can also be imple-
mented by adding to the ‘‘d-dimensional’’ Lagrangian Žsee below. counterterms whose
coefficients have poles at d s 4. Both the set of dimensionally regularized Feynman
diagrams, which conforms what is referred to as the regularized theory, and the
renormalized 1PI functional obtained from them are best established in the following
way:
Ži. Write a classical, i.e. order zero in ", action, S0 , in d space-time dimensions Ž d
being a generic positive integer. which yield the four-dimensional action of the
theory when the formal limit d ™ 4 is taken. The ‘‘d-dimensional’’ covariants
in S0 are defined as in the previous subsection. We shall assume as in Ref. w4x
that each free term, Hd d x 12 f Df , in S0 is such that Dy1 is the same algebraic
expression as in four dimensions, although it is expressed in terms of ‘‘d-di-
mensional’’ covariants. f denotes a collection of bosonic or fermionic fields.
i"Dy1 gives the free propagator. S0 will be referred to as the Dimensional
Regularization classical action. We shall call the ‘‘d-dimensional’’ space-time
used above to set the Dimensional Regularization classical action, S0 , the
‘‘d-dimensional’’ space-time of Dimensional Regularization.
Žii. Use the Dimensional Regularization classical action, S0 , along with standard
path integrals textbook techniques, formally applied, to obtain the set of
‘‘d-dimensional’’ Feynman rules stemming from it. These Feynman rules will
lead to the collection of Feynman diagrams in the d-dimensional space-time of
Dimensional Regularization, d being a generic positive integer, which is to be
turned into the dimensionally regularized Feynman diagrams by using the
algorithms in Refs. w4–6,11x. This set of dimensionally regularized Feynman
diagrams defines the regularized theory.
Žiii. Introduce the Dimensional Regularization generating functional Z DReg w J; K ; lx:
i
Z DR eg w J ; K ; l x s H Df exp ½ ž
"
S 0 w f ; K ; l x q d d x Ji Ž x . f i Ž x .
H /5 ,
Ž 8.
where the right-hand side of Eq. Ž8. is defined as the formal power expansion
in ", K and J given by the Feynman diagrams obtained by using the
‘‘d-dimensional’’ Feynman rules previously established. The symbols K and l
denote, respectively, any external field, usually coupled linearly to composite
operators, and any parameter occurring in the action. Let us stress that Eq. Ž8.
is merely a symbol which denotes the whole set of Feynman diagrams in the
‘‘d-dimensional’’ space-time of Dimensional Regularization, d being a generic
positive integer, which are converted into dimensionally regularized Feynman
´ D. Sanchez-Ruizr
diagrams by analytic continuation in d. However, this symbol is very useful

due to the Regularized Action Principle w4–6x discussed below. Indeed, the
standard formal manipulations of the path integral, e.g variations of fields and
differentiation with respect to parameters, that in four dimensions lead to
equations of motion, Ward identities and so on, can be formally performed in
the path integral in Eq. Ž8.; thus leading to formal functional equations
involving Z DR eg . These formal manipulations are mathematically well defined
if expressed in terms of the dimensionally regularized Feynman diagrams
arising from the right-hand side of Eq. Ž8., and the formal functional equations
they lead to are also mathematically sound if they are considered as symbols
denoting in a condensed manner the equations verified by the corresponding
dimensionally regularized Feynman diagrams. These equations are the dimen-
sionally regularized counterparts of the four-dimensional equations of motion,
Ward identities, etc. It is also useful to introduce the Dimensional Regulariza-
tion 1PI functional, G DR eg w f ; K ; lx, which can be obtained from Eq. Ž8. through
formal Legendre transform; a procedure well defined in terms of regularized
Feynman diagrams. Again, the Regularized Action Principle guarantees that the
formal functional equations verified by G DR eg w f ; K ; lx make sense mathemati-
cally when expressed in terms of dimensionally regularized Feynman diagrams.
We shall refer to the formal equations satisfied by Z DReg and G DReg as
dimensionally regularized equations, keeping always in mind the previous
discussion.
Živ. If there are classical symmetries in the four-dimensional classical theory that
should hold in the quantum theory, one generalizes next the corresponding field
transformations to the d-dimensional space-time of Dimensional Regulariza-
tion. Objects in this d-dimensional space-time should be defined according to
the algebra of ‘‘d-dimensional’’ covariants given in the previous subsection.
The field transformations in the d-dimensional space-time thus obtained should
yield, in the limit d ™ 4, the classical four-dimensional transformations. Gener-
ally speaking, the transformations in the d-dimensional space-time of Dimen-
sional Regularization do not leave invariant the Dimensional Regularization
classical action S0 . This lack of invariance will make the Dimensional Regular-
ization generating functionals, Z DR eg and G DReg , satisfy anomalous Ward
identities, which can be derived Žsee Appendix A. by performing formal
manipulations of the path integral in Eq. Ž8.. Again, the Regularized Action
Principle guarantees that both these formal manipulations and the identities they
lead to have a well defined mathematical meaning when expressed in terms of
dimensionally regularized Feynman diagrams. The dimensionally regularized
identities so obtained are of enormous help in the computation of the anoma-
lous breaking terms of the dimensionally renormalized four-dimensional theory
Žsee the next subsection., which otherwise would have to be computed by
evaluating the complete dimensionally renormalized 1PI functional.
Žv. The minimal subtraction algorithm of Refs. w4,7x is applied next to every
dimensionally regularized Feynman diagram coming from the 1PI functional
G DR eg w f ; K ; lx. The minimally renormalized 1PI functional Gminren w f ; K ; l; m x
Ž m stands for the Dimensional Regularization scale. is obtained by taking the
limit d ™ 4, first, and then setting every hatted object to zero in every
subtracted 1PI diagram. The minimal subtraction algorithm amounts to subtract-
ing the pole at d s 4 from every diagram once subdivergences have been taken
care of, and it can be formulated in terms of singular, at d s 4, counterterms w4x
added to the Dimensional Regularization classical action S0 . These singular
counterterms, which can be read from the forest formula w4,11x, are local
polynomials of the fields and their derivatives in the d-dimensional, with
d / 4, space-time of Dimensional Regularization. The coefficients of these
polynomials are the principal part at d s 4 of a certain meromorphic function
of complex d. The singular counterterms in question give rise to new ‘‘d-di-
mensional’’ vertices, which in turn yield new Feynman diagrams that cancel,
after dimensionally regularized, the singular behaviour of the dimensionally
regularized diagrams provided by S0 .
According to Ref. w7x the dimensional regularization scale, m , is introduced by
replacing every loop momentum measure d d prŽ2p . d with m4ydd d prŽ2p . d
before applying the subtraction algorithm. If one follows the procedure of
singular counterterms, the previous replacement should be made regardless the
diagram involves singular counterterms. The introduction of the scale m will
render dimensionally homogeneous the Laurent expansion around d s 4 of a
given dimensionally regularized diagram.
Unfortunately, if there is a symmetry we wanted preserved at the quantum
level, the renormalized functional Gminren w f ; K ; l; m x would not do, since it
would not define, in general, a quantum theory having such a symmetry: the
regularization and also the subtraction process may break the symmetry.
However, the situation is not hopeless. Algebraic BRS renormalization along
with the Quantum Action Principle w32x comes in our aid: if the anomaly
cancellation conditions w39x are met, the anomalous breaking terms, called
spurious, can be cancelled at each order in " by adding appropriate countert-
erms to the four-dimensional classical action of the theory.
Suppose that n is the order in the " expansion at which a non-anomalous symmetry
is broken for the first time in the dimensionally renormalized theory. Then,
Žvi. Compute the breaking, which will be a local four-dimensional functional, with
the help of the action principles and the Bonneau identities w7–9x. This can be
easily done Žsee the next subsection. by taking as action, not the classical one,
S0 , but a new action S Ž n. s S0 q Ssct Ž n. Ž n.
, where Ssct denotes the set of singular
counterterms needed to minimally renormalize the theory at order " n , and then
redo steps Ži. to Žv.. Indeed, if d / 4, the Regularized Action Principle still
holds for the Dimensional Regularization generating functionals, Z DR eg w J; K ; lx
and G DR eg w f ; K ; lx, defined for S Ž n., and, hence, we can take advantage of the
dimensionally regularized equations, in the sense explained in Žii., to obtain the
renormalized equations verified by Gminren w f ; K ; l; m x.
Next, extract from the breaking the four-dimensional finite counterterms needed
to restore the symmetry, generalize them to d space-time dimensions with the
´ D. Sanchez-Ruizr
help of the algebra of d-dimensional covariants Žwe shall denote this general-
Ž n. .
ization by Sfct , and add them to the action S Ž n. to obtain yet another new
Ž n. Ž n. Ž n. Ž n.
action S DR eg s S0 q Ssct q Sfct . SDReg will be called the Dimensional Regular-
Ž n.
ization action. Furnished with this new action SDR eg , establish then new
Ž n.
perturbation theory: redo steps Ži. to Žv. upon replacing S0 with S DReg . The
new renormalized 1PI functional, Gren w f ; K ; l; m x, will satisfy up to order " n
the Ward identities that govern the given symmetry at the quantum level. Move
on to Žvi. and compute the breaking at order " nq 1, and so on and so forth.
Žvii. Use the Bonneau w7,8x identities to obtain the renormalization group equation at
every order in the perturbative expansion for the theory being analyzed. Thus
the beta functions and anomalous dimensions of the theory will be evaluated.
Let us close this subsection by making some remarks on the generalization to d

space-time dimensions, i.e. to the d-dimensional space-time of Dimensional Regulariza-
tion, of the four-dimensional classical action, the four-dimensional counterterms needed
to restore non-anomalous symmetries and the four-dimensional symmetry transforma-
tions. If the classical action of the theory involves objects whose properties depend on
the dimension of space-time Že.g. g 5 , the Levi-Civita symbol, . . . . there is no canonical
Dimensional Regularization classical action, S0 , in d space-time dimensions. Any local
functional which formally go to the four-dimensional classical action as d ™ 4 would
do, provided the free terms of the former lead to the propagators described in Ži.. For
instance, let k 1 and k 2 be a couple of positive real numbers, then, the four-dimensional
metric and gamma matrices in the interaction part of the four-dimensional classical
action can be replaced with gmn q k 1 gˆmn and gm q k 2 gˆm , respectively, to obtain an
admissible biparametric family of Dimensional Regularization classical actions. Each
member of this family yields a particular regularization of the theory, this situation is
somewhat reminiscent of the lattice regularization method. Generally speaking the
difference between two admissible ‘‘d-dimensional’’ Lagrangians will always be a local
‘‘evanescent’’ operator vanishing 1 as the coupling constants go to zero. Analogously, in
the type of theories under scrutiny, namely, chiral gauge theories, there is no canonical
generalization to d-dimensional space-time of Dimensional Regularization of the four-
dimensional finite counterterms needed to restore a non-anomalous broken symmetry.
Again, two such generalizations of a given four-dimensional finite counterterm which
agree up to order " ny 1 will differ in a ‘‘d-dimensional’’ integrated ‘‘evanescent’’
operator of order " n. However, the dependence on the choice of generalization will
show in the renormalized theory at order " nq 1, never at order " n. Last, but not least,
from the point of view of renormalization, there is no canonical generalization to d
space-time dimensions of the symmetry transformations of the fields for theories which
are not vector-like. Again, two such generalizations differ in an ‘‘evanescent’’ operator.
However, this arbitrariness in the choice of Dimensional Regularization classical action,
‘‘d-dimensional’’ finite counterterms and symmetry transformations is useful since one
can play around with it so as to simplify the form of the symmetry breaking contribu-
tions. Finally, for vector-like theories such as QCD, there is, of course, a canonical
1
An ‘‘evanescent’’ operator is an operator whose tree-level contribution vanishes as d™4.
choice of Dimensional Regularization classical action and vector-like transformations:

the usual one w11x.
2.3. Regularized and quantum action principles
As we said in Subsection 2.2, one of the main characteristics of the dimensional

renormalization procedure is that action principles are precisely stated and proved w4–6x.
These principles are most efficiently expressed in terms of the Dimensional Regulariza-
tion generating functional, Z DR eg w J; K ; lx, for the Greens function given by the Gell-
Mann–Low series and the dimensional renormalization 1PI functional, Gren w f ; K ; l; m x,
obtained from it Žsee Subsection 2.2..
Let us take as Dimensional Regularization action Žsee Žvi. in Subsection 2.2.
Ž n.
SDR eg s Sfree w f ; l x q Sint w f ; K ; l x , Ž 9.
where f denotes a collection of commuting or anticommuting quantum fields, K set of
commuting or anti-commuting external fields, l is a generic parameter, and
Sfree w f ; l x s S0 free w f ; l x ,
Ž n. Ž n.
Sint w f ; K ; l x s S0 int w f ; K ; l x q Ssct w f ; K ; l x q Sfct w f ; K ; lx .
We next introduce
SINT w f , J , K , l x s Sint w f , K , l x q d d x Ji Ž x . f i Ž x .
H
and define, following Eq. Ž8., the functional Z DR eg
i i
Z DR eg w J ; K ; l x s H Df exp ½ ž
"
Žn.
SDReg q Ji f i
H /5 ¦ ½
' exp
"
S INT w f ; J ; K ; l x 5;0
.
Ž 10 .
The symbol exp in the previous equation stands for its formal power series and the
symbol ² . . . : 0 denotes the usual vacuum expectation value
i
² . . . : 0 s Df . . . exp
H S ½
" 0 free
w f ; lx , 5
defined by gaussian integration, which gives a formal power series in " and the external
fields J and K. Each coefficient of this series is a sum of ‘‘d-dimensional’’ Feynman
diagrams. Every ‘‘d-dimensional’’ Feynman diagram can be converted into a meromor-
phic function of d by promoting d to a complex variable and understanding the
‘‘d-dimensional’’ covariants as in Subsection 2.1.
Now, the Regularized Action Principle states that the following three functional
equations hold in the dimensionally regularized theory:
Ž1. Arbitrary polynomial variations of the quantized fields f , df Ž x . s

du Ž x . P Ž f Ž x .. leave Z DR eg invariant and
i
¦
d Z DReg w J , K x ' d Ž Sfree q S INT . exp ½ "
S INT w f , J , K , l x 5;
0
s 0, Ž 11 .
where the variations are the linear parts in du .

´ D. Sanchez-Ruizr
Ž2. Variations of external fields EŽ x . ' Ž K Ž x ., J Ž x .. give

d SINT i d Z DReg w J , K , l x
¦ d EŽ x .
exp ½ "
S INT w f , J , K , l x 5;
0
s yi"
d EŽ x .
. Ž 12 .
Ž3. Variations of parameters give
E Ž Sfree q SINT . i E Z DReg w J , K , l x
¦ El
exp ½ "
S INT w f , J , K , l x 5;
0
s yi"
El
.
Ž 13 .
Let us explain in what sense the functional Eqs. Ž11., Ž12. and Ž13. are mathemati-
cally meaningful. First, these equations are to be understood diagramatically only. The
formal expansion in powers of ", K and J of both sides of each equation leads to an
infinite set of equations involving only ‘‘d-dimensional’’ Feynman diagrams. It is this
infinite set of diagramatic equations in the d-dimensional space-time of Dimensional
Regularization what is taken as the definition of the corresponding functional equation.
The Regularized Action Principle w4,7x states that this very set of diagramatic equations
still holds if every formal ‘‘d-dimensional’’ Feynman diagram in them is replaced with
its dimensionally regularized counterpart, with the scale m introduced as explained in
Subsection 2.2, Žv..2 A mathematically sound set of equations is thus obtained. Of
course, it is easier to handle the functional equations above than their dimensionally
regularized diagramatic definition. Furthermore, this functional equations can be de-
duced by formal path integral manipulations, which in turn can be given a diagramatic
definition valid within the Dimensional Regularization method a´ la Breitenlohner and
Maison. Let us finally note that the limit d ™ 4 of dimensionally regularized the Green
functions coming from the right-hand side of Eqs. Ž12. and Ž13. does exist at each order
Ž n.
in the " expansion up to order n. Notice that we have included in the action S DReg ,
n
defined in Eq. Ž9., the appropriate singular counterterms up to order " . Hence, Eqs.
Ž12. and Ž13. have a well-defined d ™ 4 limit up to order " n and can be used to
compute the variations with EŽ x . and l of the renormalized functional Zren w J; K ; lx up
to order " n.
The Regularized Action Principle given by Eqs. Ž11., Ž12. and Ž13. leads w4,7,8,11x
to the Quantum Action Principle Žsee Ref. w41x and references therein. which states that
the following functional equations hold for the dimensionally renormalized theory:
QAP1
Žn.
EGren E Ž S0 q Sfct .
sN P Gren . Ž 14 .
El El
QAP2
Žn.
dGren d Ž S0 q Sfct .
sN P Gren . Ž 15 .
d EŽ x . d EŽ x .
2
The scale m is introduced so as to render dimensionally homogeneous the Laurent expansions around
ds 4.
QAP3 Let f X Ž x . s P Ž f Ž x .. denote a linear transformation of f Ž x . whose coeffi-

cients do not depend explicitly on d, then
Žn.
dGren d Ž S0 q Sfct .
fX Ž x . s N fX Ž x . P Gren . Ž 16 .
df Ž x . df Ž x .
QAP4 If df Ž x . is a non-linear field transformation in the four-dimensional space-time
and K Ž x . is the external field coupled to it, then
dGren dGren
s N O Ž x . P Gren , Ž 17 .
d K Ž x . df Ž x .
where the symbols Nw . . . x and Nw . . . x P Gren denote normal product as defined in Refs.
w7,8,11x and its insertion in the renormalized 1PI functional, respectively. Nw O Ž x .x is a
Ž n.
local normal product of ultraviolet dimension 4 y dimŽ f . q dimŽ df .. S0 and Sfct have
been defined in Ži. and Žv. of Subsection 2.2.
Eqs. Ž14. – Ž17. have been shown to hold by using the forest formula w4,5,7,8x and the
singular counterterms algorithm w11x, respectively. Of course, both proofs are equivalent.
The Quantum Action Principle is of the utmost importance to the renormalization of
BRS symmetries w32x, for it guarantees Žsee QAP1 and QAP2. that the breaking of such
symmetry is given by the insertion into the dimensionally renormalized 1PI Gren of a
certain integrated normal operator, Nw O x s Hd 4 x Nw O Ž x .x, where Nw O Ž x .x is a local
normal operator of ultraviolet dimension 4 y dimŽF . q dimŽ sF . and ghost number 1.
Indeed, the following equation holds for Gren w w ,F ; KF x
dGren dGren dGren
SS Ž Gren . ' d 4 x Ž sw .
H q s D breaking . Ž 18 .
dw d KF dF
with D breaking s Nw O x P Gren . In the previous equation w and F stand for fields which
undergo linear and non-linear BRS transformations, respectively. As it is customary sw
denotes a BRS transformation. KF stands for the external field coupled to the non-linear
BRS transformation sF in the four-dimensional classical action w32x.
Now, if finite counterterms have been added to the four-dimensional classical action
so that the BRS symmetry is preserved up to order " ny 1 Ži.e. the first non-vanishing
contribution to the right-hand side of Eq. Ž18. is order " n ., Eq. Ž18. reads
Ž n.
SS Ž Gren . s N w O x q O Ž " nq1 . ,
where Nw O xŽ n. is the contribution to Nw O x of order " n. We have taken into account that
Nw O x P Gren s Nw O xŽ n. q OŽ " Nw O xŽ n. ., since by assumption Nw O x P Gren s OŽ " n ..
Next, if Nw O xŽ n. s y" n b Sfct, n Ži.e. Nw O xŽ n. is b-exact. for some four-dimensional
integrated local functional of the fields, the BRS symmetry can be restored, up to order
" n, by adding to the four-dimensional classical action the finite counterterm Sfct, n w32x.
Here b denotes the linearized BRS operator:
Ž 0.
dGren d
b s s q d4 x
H ½ dF Ž x . d KF Ž x . 5 ,
and Gren
Ž0.
is the order " 0 contribution to Gren , i.e. the BRS invariant classical
´ D. Sanchez-Ruizr
Ž n.
four-dimensional action. Then, we set Sfct s Ý nms1 " m Sfct, m . Of course, if Nw O xŽ n. is not
b-exact the theory is anomalous at order " n.
The computation of the right-hand side of Eq. Ž18., D breaking , is the issue we will
discuss now. One can always compute the complete 1PI functional Gren at each order in
", insert it in the left-hand side of Eq. Ž18., work out the functional derivatives and thus
obtain the right-hand side. This method is, of course, very impractical. The Regularized
Action principle, as given in Eqs. Ž11. and Ž12., provides us with a more efficient way to
compute the symmetry breaking term of the BRS symmetry. Indeed, in Appendix A we
show that the following functional equation hold, in the same sense as Eqs. Ž11. and
Ž12., for the 1PI formal functional G DR eg
dG DR eg dG DReg dG DReg
SS d Ž G DReg . ' d d x Ž s d w .
H q s D P G DReg q Dct P G DReg
dw d KF dF
d SctŽ n . dG DReg
q dd x
H P G DR eg , Ž 19 .
d KF Ž x . dF Ž x .
where G DR eg is the 1PI functional computed with the Dimensional Regularization
Ž n.
action, SDR eg , defined at the beginning of the current subsection. The symbol s d denotes
the generalization to d dimensions of the four-dimensional BRS transformations Žsee
Subsection 2.2 Živ... The operators D and Dct are given by D s s d S0 and Dct s s d SctŽ n.,
respectively. Here, S0 s S0 free q S0 int and SctŽ n. s Sfct
Ž n. Ž n. Ž
q Ssct see Eq. Ž9. and the para-
graph below it for definitions..
Let us recall that G DR eg is defined as a formal series expansion in powers of " and
the fields w , F and K f . The coefficients of this power series are 1PI Green functions of
the theory at a given order in ". Taking into account that G DR eg contains the
contributions coming from the singular counterterms up to order " n, denoted by Ssct Ž n.
,
one concludes that the dimensionally regularized 1PI Green functions obtained from
G DReg are finite in the limit d ™ 4 up to order " n. These 1PI Green functions are the
dimensionally renormalized 1PI Green functions upon introduction of the Dimensional
Regularization scale m Žsee Subsection 2.2... Hence, up to order " n , one may formally
write
LIM d ™ 4 G DReg w w ,F ; KF ; m x s Gren w w ,F ; KF ; m x . Ž 20 .
In the previous equation, LIM d ™ 4 is defined by the following process: one takes first
the limit d ™ 4 of the dimensionally regularized 1PI Green functions and then sets to
zero every hatted object Ži.e. Ž d y 4.-dimensional covariant.. Recall that the Dimen-
sional Regularization scale, m , is introduced by hand to render dimensionally homoge-
neous the regularized Green functions upon Laurent expansion around d s 4 w7x.
Now, by taking LIM d ™ 4 of the left-hand side of Eq. Ž19. after discarding any
contribution of order greater than " n , and bearing in mind Eq. Ž20., one obtains the
left-hand side of Eq. Ž18. up to order " n. Hence, D breaking in the latter equation is given
by
½
D breaking s LIM d ™ 4 D P G DReg q Dct P G DReg
d SctŽ n . dG DReg
q dd x
H d KF Ž x .
P G DR eg
dF Ž x . 5 . Ž 21 .
Of course, the previous equation only makes sense up to order " n. The order "
contribution to Eq. Ž21. reads particularly simple since s d S0 is an evanescent operator
Žthere is no O Ž " 0 . contribution to Nw D x thereby. and the lowest order contribution to Sct
is order ":
Ž1. Ž1. Ž1 .
D Ž1.
breaking s LIM d ™ 4 ½ D P G DReg singular
q bd Ssct 5
q N w D x P Gren Ž1.
q b Sfct ,
Ž 22 .
where the superscript Ž1. denotes contributions of order " and D is equal to the
evanescent operator s d S0 . In Eq. Ž22., bd stands for the linearized BRS operator in d
dimensions:
d S0 d
bd s s d q d d x
H ½
dF Ž x . d KF Ž x .
. 5
The symbol w D P G DR eg xŽ1.
singular denotes the expression, singular at d s 4, that it is
subtracted from D P G DR eg to obtain wNw D x P Gren xŽ1. according to the forest formula w4,7x.
Hence,
Ž1 . Ž1 . Ž1 .
D P G DR eg s D P G DReg singular
q N w D x P Gren q vanishing terms. Ž 23 .
By ‘‘ vanishing terms’’ we mean contributions that vanish as d is send to 4 first and then
every hatted object is set to zero.
There are some comments regarding Eq. Ž22. which we would like to make. First, if
Ž1.
the theory is not anomalous, the finite counterterms in Sfct can be chosen so that
D breaking s 0 w32x. Second, there are two possible sources of symmetry breaking in Eq.
Ž1.
Ž22.. One is the regularization method; in particular, the Dimensional Regularization

classical action, S0 , one began with. The other is the minimal subtraction algorithm Ži.e.
the renormalization algorithm. we are employing. That the regularization method breaks
the symmetry shows itself in the contributions w D P G DR eg xŽ1. w w x
singular and N D P Gren
xŽ1..
Ž1.
The subtraction algorithm gives rise to the contribution bd Ssct , which need not be zero
as we shall show later on. Third, the ‘‘limit’’
Ž1. Ž1.
LIM d ™ 4 ½ D P G DReg singular
q bd Ssct 5
need not be zero either, as we shall discuss in Section 3. Of course, this ‘‘limit’’ yields a
finite local four-dimensional functional of the fields and their derivatives. Fourth, the
operator D s s d S0 being evanescent ushers in the techniques introduced by Bonneau
w7,8x to express normal products of evanescent operators, also called anomalous normal
products, in terms of a basis of standard Žnon-evanescent operators. normal products.
We shall address this subject in the next subsection. Suffice it to say that expansion of
the normal product of an evanescent operator in terms of ‘‘standard’’ normal products
has coefficients that are series expansions in " with no OŽ " . term. Hence, there is no
breaking of the BRS symmetry at order " 0 and wNw D x P Gren xŽ1. is, in agreement with the
Quantum Action Principle, a sum of standard local operators.
We shall finally recall that the equation of motion holds in dimensional renormaliza-
tion w4,7,11x:
Žn. Ž` , n .
dGren d Ž S0 q Sfct . d SDReg
df Ž x .
sN
df Ž x .
P Gren s LIM d ™ 4 ½ df Ž x . 5
P G DReg , Ž 24 .
´ D. Sanchez-Ruizr
Ž n. Ž`, n.
where Sfct and S DReg are defined in Subsection 2.2 and Eq. Ž9.. The superscript ` tell us
that the singular counterterms needed to minimally renormalize the theory at any order
in " have been included; however, one has added finite counterterms only up to order
" n. Eq. Ž24. can obtained from QAP3 ŽEq. Ž16.. by setting f X Ž x . to 1. Alternatively,
Eq. Ž11., for P Ž f Ž x .. s 1, leads to Eq. Ž24..
The use of both the equation of motion of the ghost field and the equation of motion
of the auxiliary field coupled to the gauge fixing condition simplifies the renormaliza-
tion of theories with BRS symmetries.
2.4. Expression of anomalous terms with the aid of Bonneau identities
The computation of the insertions of normal products of evanescent operators, also

called anomalous normal products, are, in dimensional renormalization, of key impor-
tance to the calculation of the symmetry breaking terms in Ward identities w7,11x. We
shall see in the next section that wNw D x P Gren xŽ1. Žsee Eq. Ž22.. yields both the essential
and the spurious BRS anomalies for the regularized theory we will consider. The
purpose of the current subsection is to summarily recall that the normal product of an
evanescent operator can be expressed as linear combinations of normal products of
standard Žnon-evanescent. operators. Further details can be found in Refs. w7–9x.
In renormalized perturbation theory, any set of insertions of quantum operators such
that their classical approximations form a basis in the linear space of classical operators
of dimensions bounded by D is also a basis in the linear space of insertions, with
dimension bounded by the same D, of quantum operators w32x. So, in dimensional
renormalization, if O i 4 form such a basis in the space of classical operators, then
Nw O i x P Gren 4 , where O i is any of the possible generalizations to d dimensions of the
corresponding classical four-dimensional operator, is a basis in the space of quantum
insertions. We will call standard operators the operators in the ‘‘d-dimensional’’
space-time of Dimensional Regularization which are generalizations of the classical
operators in four dimensions. These standard operators are obtained with the help of the
algebra ‘‘d-dimensional’’ covariants and they are non-evanescent. Hence, Nw O i x P Gren is
an insertion of an standard normal product.
It is thus expected that anomalous normal products, i.e. normal products of non-
evanescent operators minimally subtracted should be decomposable in terms of some
basis of standard normal products, minimally subtracted again. This was shown by
Bonneau w9x who, remarkably, proved a linear system of Zimmerman-like identities
which expresses a decomposition of any anomalous normal product on all possible
normal products of monomials of operators, including standard and anomalous ones.
If there is only a scalar field, the Bonneau identities have the form
N gˆmr O mr Ž x . P Gren
r
4 4yn Ž yi .
sy Ý
n
Ý
r
Ý
i1 . . . i r 4
1(i j(n
½ r.s.p.
r!
Er
=
E pim1 1 . . . E pimr r
¦f˜ Ž p . . . . f˜ Ž p . N
1 n gˇm r O mr ; 1PI
ˇ
p i'g'0 5
1 n
=N
n!
Ł
ks1
½ž Ł
a ri ask 4
/
Ema f 5 Ž x . P Gren , Ž 25 .
where the tilde indicates Fourier transformed fields, the bar means that only the minimal
subtraction of the subdivergences has been done, r.s.p. stands for ‘‘residue of the simple
pole in ´ s 4 y d’’ Žthis is the reason why the global sign y1 occurs: gˆ m m s y´ . and
the = symbol means that the tensorial structure . . . 4 really appears inside the normal
product. Of course, any colour or pure number factor can be taken out of the normal
product, but note that gmn Nw O mn x s Nw gmn O mn x / Nw gmn O mn x, the difference being the
anomalous normal product Nw gˆmn O mn x, which need not vanish.
Notice that the sum in n is a sum in the number of fields in the monomial and the
sum in r, a sum in the number of derivatives in the mononial.
The tensor gˇ is a new one, which has been introduced in order to simplify the
calculations. Its properties are
gˇmn s gˇmn , gˇm m s 1,
gˇmr g r n s gˇm r gˆ r n s gˇmn ,
gˇmr g rm s 0, N gˇmn O mn s gˇmn N w O mn x .
Eq. Ž25. is easily generalized when there are several fields, not necessarily scalar,
involved. One just sums over all kind of fields, taking special care of the symmetry
factors and the fermionic signs. We shall give below the generalization of Eq. Ž25. that
is relevant to our computations in Subsection 3.5.
Due to the fact that all subdivergences have been previously subtracted, the barred
1PI function of Eq. Ž25. has a polynomial singular part. We are interested in the
coefficients of the simple pole. These coefficients will be in general combinations of
metrics, constants and colour factors. In vector-like gauge theories with the usual
regularizations, its tensorial structure will involve only usual metrics, which combined
with the tensor indices of the normal product will get the expansions of the anomalous
normal product in terms of a basis of standard monomial normal products. However, if
in the regularized theory or in the calculations some hatted objects appear, then the
Bonneau identities ŽEq. Ž25.. would express any anomalous normal product in terms of a
collection of standard Ž M i . and also evanescent Ž Mˆ j . monomial normal products. That
is
N gˆmr O mr Ž x . P Gren s Ý a i N w M i x Ž x . P Gren q Ý aˆ j N w Mˆ j x Ž x . P Gren .

i j
Therefore, the rest of independent anomalous monomials should be also expanded by a

similar formula, then getting a system of identities.
But notice that the left-hand side of the Bonneau identities and also the coefficients
given by the r.s.p. of the right-hand side are both of order "1, in the least. Therefore, the
Bonneau identities are not a trivial system but a linear system whose unique solution
´ D. Sanchez-Ruizr
give the desired expansion of any anomalous operator in terms of a quantum basis of
standard insertions:
N gˆmr O mr Ž x . P Gren s Ý qi N w M i x Ž x . P Gren .

i
Of course, the coefficients qi are formal series in ".

Clearly, at lowest order the linear system is decoupled and has an easy interpretation:
the loops with the anomalous insertion are replaced with sums of tree diagrams
corresponding to insertion of standard operators, but with coefficients of order "1. This
will be most relevant to our computations below. In general, for the calculation of qi at
order " m , it is needed the coefficients a i up to order " m and the coefficients aˆ i up to
order "Ž my1..
3. Chiral non-Abelian Yang–Mills theories: Simple gauge groups
This section is devoted to the study of the one-loop dimensional renormalization of a

general Chiral non-Abelian gauge theory with neither Majorana nor scalar fields. We
shall assume for the time being that the gauge group is a compact simple Lie group and
leave for Section 4 the case of a general compact group.
Let us give first some definitions and display some properties:
1 y g5 1 q g5
PL s , PR s , PL2 s PL ,
2 2
PR2 s PR , PL PR s PR P L s 0,
c L s PL c , c R s PR c , c L s c PR , c R s c PL ,
and also the special properties of the symbols in the algebra of covariants:
PR g m PR s PR gˆ m PR s gˆ m PR / 0,
P L g m PL s PL gˆ m P L s gˆ m PL / 0,
P L g m PR s P L g m PR s g m PR s P L g m ,
PR g m P L s PR g m P L s g m P L s PR g m ,
P L g m P L s PR g m PR s 0 s P L gˆ m PR s PR gˆ m P L . Ž 26 .
3.1. The classical action
The BRS invariant classical four-dimensional action is

Scl s Sinv q Sgf q Sext , Ž 27 .
with
1 i l i l
Sinv s d 4 x y
H 2
Tr Fmn F mn q u L cq
cD cX D
u RcX,
4g 2 2
a
Sgf s d 4 x Tr
H B 2 q Tr B Ž Em A m . y Tr v E m =m v ,
2
Sext s d 4 x Tr r m sAm q Tr z s v q Lsc q Lsc q Rsc X q Rsc X ,

H
where Am s Ama t a, v , v , B, r and z take values on the Lie algebra of a compact simple
Lie group G. t a are the generators of G in a given finite-dimensional representation,
normalized so that Tr wt at b x s d a b. We thus have wt a,t b x s ic a b c t c, c a b c being com-
pletely antisymmetric, which defines the adjoint representation ŽTAa . i j s yi c ai j with a
certain normalization Tr w TAa TAb x s TA d a b , ŽTAe TAe . i j s CA d i j , TA s CA . The following
definitions will be used in the sequel:
Fmn s Fmna t a s Em An y En Am y i Am , An ,
=m f a s Em f a q c a b c Amb f c , f being a Lie algebra valued object,
D L m c s Ž Em y iAma T La P L . c ,
D R m c X s Ž Em y iAma TRa PR . c X .
c Ž c X . represents a collection of left-handed Žright-handed. fermionic multiplets carry-
ing finite representations, TLa ŽTRa . of the group generators. The following equations hold
T La ,T Lb s ic a b c TLc , TRa ,TRb s ic a b c TRc ,
Tr TLa T Lb s TL d a b , Tr TRa TRb s TR d ab ,

TLe TLe s C L , TRe TRe s C R .
We also introduce the shorthand notation:
Tr TLa1 . . . T La n ' T La1 . . . a n , Tr TRa1 . . . TRa n ' TRa1 . . . a n ,
TLa1 . . . a n q TRa1 . . . a n ' TLqR
a1 . . . a n
, TLa1 . . . a n y TRa1 . . . a n ' TLyR
a1 . . . a n
,
Tr L w f 1 . . . fn x ' f 1a1 . . . fna n Tr TLa1 . . . T La n ,
Tr R w f 1 . . . fn x ' f 1a1 . . . fna n Tr TRa1 . . . TRa n .
The appropriate index labelling different fermions will be understood throughout this
paper: every left handed multiplet can yield a different, say, T L , C L . . .
For the action we have chosen, the free boson propagator is Žin momentum space.
yi a k mk n
g2 d a b g mn y 1 y ž /
k 2 qie g2 k2
yi k mk n a k mkn
sg2 d ab ž g mn y / q .
k 2 qie k2 g2 k2
´ D. Sanchez-Ruizr
Table 1
Ghost number and dimension of the fields. In the last row, q1 Ž-1. means that the symbol commutes
Žanticommutes.
s Am c ,c X v v B rm z L, R
Ghost number 1 0 0 1 y1 0 y1 y2 y1
Dimension 0 1 3r2 0 2 2 3 4 5r2
Commutat. y1 q y1 y1 y1 q1 y1 q1 q1
Because in perturbative calculations the combination arg 2 will often appear, we denote
it with a X . Therefore, the Feynman gauge here is defined by a s g 2 or a X s 1.3
Scl is left invariant in four dimensions by the BRS transformations:
sc s i v a TLa P L c , sc s i c TLa PR v a ,
sc X s i v a TRa PR c X , sc X s i c X TRa PL v a ,
sAm s =m v , sv s i v 2 , s v s B, sB s 0,
s r m s 0, sz s 0, sL s sL s sR s sR s 0. Ž 28 .
This is due to the anticommutativity of g 5 in four dimensions, which allow us to
write Sinv in the gauge invariant form
1 i l i l
Sinv s d 4 x y
H 2
Fmn F mn q u L cL q
cL D c RX D
u R c RX
4g 2 2
i l i l
q c R Eu c R q c LX Eu c LX .
2 2
The action preserves the ghost number. Its value together with the dimensions and the
commutativity for the different fields are shown in Table 1.
The four-dimensional BRS linear operator b is
d Scl d d Scl d d Scl d

b ' s q d 4 x Tr
H ½ d Am dr m q Tr
dv dz
q
dc d L
d Scl d d Scl d d Scl d

q X
dc d R
q
dc d L
q
dc X d R 5 , Ž 29 .
which satisfies b 2 s 0 because is the linearization of the BRS operator and the classical
action satisfy the BRS identities w32x. The symbol s has been defined in Eq. Ž28..
3 X X X X
If the fields are rescaled by g: A™ gA , v ™ g v , v ™ gy1 v ; and a ™ g 2a , then the action adopt the
other usual form with g being interpreted as the ‘‘coupling’’ constant. But let us remember that the loop
expansion is an expansion in " rather than in g.
The classical b-invariant combinations are

E Scl 1
Lg s g s Hd 4
x Tr Fmn F mn ,
Eg 2 g2
L A s b P d 4 x r˜ma A a m ,
H Lv s yb P d 4 x z a v a ,
H
i l
LLc s yb P d 4 x LPL c q c PR L s 2 d 4 x
H H c Eu PL c q cg m PL TLa c Ama ,
2
i l
LRc X s yb P d 4 x RPR c X q c X PL R s 2 d 4 x
H H c X Eu PR c X q c Xg m PR TRa c XAma ,
2
i l
LRc s yb P d 4 x LPR c q c P L L s 2 d 4 x
H H c Eu PR c ,
2
i l
LLc X s yb P d 4 x RPL c X q c X PR R s 2 d 4 x
H H c X Eu PL c X , Ž 30 .
2
where r˜ m s r m q E m v . LRc and LLc X will prove not to be useful due to our choice of the
regularized chiral vertex.
It will be shown later that the non-trivial L g is associated with the finite renormaliza-
tions of the coupling constant g, whereas the b-exact terms, Lc , Lc X , L A and Lv give
rise to finite renormalizations of the corresponding wave functions in the space of field
functionals in four-dimensional space-time.
3.2. The dimensional regularization action
We shall now follow Ži. of Subsection 2.2 and set the Dimensional Regularization
classical action S0 . The kinetic terms are uniquely defined, not so the interaction terms.
Notice, for instance, that the Dirac matrix part of the left-handed fermion–gauge-boson
vertex has the following equivalent forms in four dimensions: g m P L s PR g m s PR g m PL .
All these forms are different in the d-dimensional space-time of Dimensional Regular-
ization because of the non-anticommutativity of g 5 . Of course, the generalization of the
interaction to the Dimensional Regularization space is not unique, and any choice is
equally correct, although some choices will be more convenient than others.
We choose the following generalization of Sinv :
1 i l i l
Hd d
x y 2
Tr Fmn F mn q c Eu c q c X Eu c X
4g 2 2
q Ž c PR g m P L TLa c q c X P L g m PR TRa c X . Ama
1 i l i l
s dd x y
H 2
Tr Fmn F mn q c Eu c q c X Eu c X
4g 2 2
q cg m PL TLa c q c X g m PR TRa c X Ama .

ž /
´ D. Sanchez-Ruizr
We next generalize, in an obvious way, to the d-dimensional space-time of Dimen-

sional Regularization the BRS variations Žwhich we will denote by s d and call
Dimensional Regularization BRS variations. and the gauge-fixing terms. The Dimen-
sional Regularization BRS transformations read
s d c s i v a TLa P L c , s d c s i c T La PR v a ,
s d c X s i v a TRa PR c X , s d c X s i c X TRa PL v a ,
s d Am s =m v , sd v s i v 2 , s d v s B, s d B s 0,
s d r m s 0, s d z s 0, s d L s s d L s s d R s s d R s 0. Ž 31 .
Notice that we have chosen the Dimensional Regularization BRS transformations so
that no explicit evanescent operator occurs in them. This will certainly simplify the
computation of the anomalous breaking term in BRS identities for the minimally
renormalized theory. We refer the reader to Appendix B for the discussion of the
computation of the BRS anomalous breaking term when the Dimensional Regularization
BRS transformations involve coefficients that vanish as d ™ 4.
In summary, our Dimensional Regularization classical action, S0 , has the following
expression:
1 i l i l
S0 s d d x y
H 2
Tr Fmn F mn q q c Eu c q c X Eu c X
4g 2 2
q cg m PL TLa c q c X g m PR TRa c X Ama

ž /
a
q d d x Tr
H B 2 q Tr B Ž Em A m . y Tr v E m =m v
2
q d d x Tr r m s d Am q Tr z s d v q Ls d c q Ls d c q Rs d c X q Rs d c X .
H Ž 32 .
The fermionic part of the Dimensional Regularization classical Lagrangian has the
following non-gauge invariant form:
l l l l l
i i i i i
u L cL q
cL D c RX u
D X
R cR q c R Eu c R q c LX Eu c LX q c R Eû c R
2 2 2 2 2
l l l
i i i
q c L Eû c L q c LX Eû c LX q c RX Eû c RX . Ž 33 .
2 2 2
Hence, S0 is not BRS invariant, the breaking, s d S0 , coming from the last four terms of
Eq. Ž33.:
i l i l
s d S0 s s d d d x cgˆ m PL pr m c q c PR gˆ m pr m c
H 2 2
i l i l
q c Xgˆ m PR pr m c X q c X PL gˆ m pr m c
2 2
1 l l
s d d x v a cgˆ mg 5TLa pr m c q Em Ž cgˆ m TLa c . y c Xgˆ mg 5 TRa pr m c X
H ½
2
qEm Ž c Xgˆ m TRa c X . ' Dˆ ' d d x Dˆ Ž x . .

5 H Ž 34 .
The Feynman rule of the insertion of this anomalous breaking is given by Fig. 1. We
denote hereafter the line of the fermion which interact leftŽright.-handedly as LŽR..
Notice that if the fermion representationŽs. were compatible with a CP invariance of
the classical action, then the Dimensional Regularization action with this choice of chiral
vertex would also be formally CP invariant, and the breaking Eq. Ž34. would have a
definite CP value Žq1 if we assign Ž v L ŽR . .X s yŽ v L ŽR . . t , y1 if we assign Ž v L ŽR . .X s
Ž v L ŽR . . t , where v L ŽR . ' v a T L ŽR . and t stands for transposition of the colour matrices..
The breaking is an Žimplicit. ‘‘Ž d y 4.-object’’, i.e. an evanescent operator, and,
clearly, this would be also true for any other choice Dimensional Regularization
classical action. For example, if we had chosen as the regularized interaction
cg m PL TLa c Ama q c Xg m PR TRa c XAma ,
the breaking would have been

1 l l
Hd d
x v a cgˆ mg 5TLa Eu m c q Em Ž cgˆ mg 5TLa c . y c Xgˆ mg 5TRa Eu m c X
½
2
yEm Ž c Xgˆ mg 5 TRa c X . y c a b c v b cgˆ m P L TLa c q c Xgˆ m PR TRa c X Amc

5 ž /
qi v b cgˆ m P L TLc TLb c q c Xgˆ m PR TRc TLb c X Amc ,
ž /
which is certainly more involved. ŽAnd with this vertex, the formal CP invariance of the
Dimensional Regularization classical action for CP invariant classical actions would be
lost..
Remember that in the Breitenlohner and Maison prescription gˆ m anticommutes
whereas g m commutes with g 5 ; so, in general the algebraic manipulations of the strings
of g ’s would be a bit more tedious with the second kind of vertex because g m s g m q
gˆ m.
Obviously, due to the fact that these two vertices are different, the results obtained by
using minimal subtraction would be different – they would be two different renormal-
ization schemes – and they should be related by finite terms. Moreover, in general, it
Fig. 1. Feynman rule of the insertion of the non-integrated breaking Eq. Ž34..
´ D. Sanchez-Ruizr
should be expected that both results are different from the one obtained with the usual
‘‘naive’’ prescription of an anticommuting g 5 every time a fermionic loop occurs. But
the ‘‘hermitian vertex’’ of Ref. w42x is not the only correct one as claimed there, but also
Fig. 2. Feynman rules for the order " singular counterterms. Here, only functions with no external fields are
shown.
any other one differing from it by a evanescent operator of dimension 4 as correctly

remarked in w43x.
The fact that in general the results obtained by minimal subtraction a´ la Breitenlohner
and Maison in theories with cancellation of anomalies do not satisfy the Ward identities
is a drawback from the practical point of view but not a reason to cast doubt on
Breitenlohner and Maison schemes w44x. Indeed, we should always remember that we
have the freedom to add any finite counterterm to restore the identities. And in theories
with anomalies the mathematical rigour of the regularization scheme we are considering
has not been surpassed by any other dimensional regularization prescription .
Furnished with the action S0 given in Eq. Ž32. one next develops a dimensionally
regularized perturbation theory by following steps Ži. to Živ. in Subsection 2.2. This
regularized theory is not invariant under the Dimensional Regularization BRS transfor-
mations in Eq. Ž31.. Indeed, the Dimensional Regularization 1PI functional G 0 obtained
from S0 satisfies anomalous BRS identities, the symmetry breaking terms being given
by the insertion in G 0 of the operator Dˆ in Eq. Ž34..
Fig. 3. The same as for Fig. 2 but with functions involving external fields.
´ D. Sanchez-Ruizr
Since our ultimate goal is to obtain a BRS invariant theory Žif the anomaly
cancellation conditions w39x are met. we shall further proceed and develop a dimension-
ally regularized perturbation theory by using the following action:
Ž n. Ž n.
S DR eg s S0 q Sct , SctŽ n. s Ssct
Ž n. Ž n.
q Sfct , Ž 35 .
instead of the Dimensional Regularization classical action S0 . We have named, in
Ž n.
Subsection 2.2, S Dreg the Dimensional Regularization action Žat order " n .. It is
computed order by order in the perturbative expansion in " by proceeding as in steps Žv.
Ž n.
to Žvii. of Subsection 2.2. Let us recall that Ssct denotes the singular Žat d s 4.
counterterms needed to minimally renormalize the theory dimensionally up to order " n,
Ž n.
whereas Sfct stands for the finite Žat d s 4. counterterms needed to turn the minimally
renormalized theory into a BRS invariant theory up to order " n. In this paper we shall be
concerned only with the renormalized theory at order ". Hence, we will just compute
Ž1. Ž1.
Ssct and Sfct .
Note that we have denoted by G 0 , instead of G DReg , the Dimensional Regularization
1PI functional for S0 to avoid confusion with the Dimensional Regularization 1PI
Ž n.
functional for SDreg , which shall denote by G DReg .
(1)
3.3. The one-loop singular counterterms: Ssct
Let G 0Ž1.
sing be the one-loop singular contribution at d s 4 to G 0 , the latter being the
Dimensional Regularization 1PI functional obtained from S0 . S0 is given in Eq. Ž32..
Then, by definition
Ž1.
Ssct s yG 0Ž1.
sing . Ž 36 .
Ž1.
We show the 1PI functions contributing to Ssct in Figs. 2, 3 and 4.
Let ´ s 4 y d. Then, the functional G 0 sing is given by
Ž1.
" 1
G 0Ž1.
sing s
´ Ž 4p . 2 ½
g 2 yCA 10
3 q Ž 1 y a X . S0 A A
8 T Lq R
q S0 A A y C A 4
3 q 32 Ž 1 y a X . S0 A A A
3 2
8 T Lq R
q S0 A A A y CA y 23 q 2 Ž 1 y a X . S0 A A A A
3 2
8 T Lq R
q S0 A A A A q C L 2 a X S0 cc q C R 2 a X S0 c Xc X
3 2
Ž1yaX .
CA q 2 a X C L S0 cc A
q 2y
ž 2 /
Ž1yaX . X
Ž1yaX .
q 2y
ž 2 / C A q 2 a C R S0 c c A y 1 q
X X
ž 2 /
=CA S0 v v q a X CA S0 v v A q a X CA S0 L sc q S0 R sc X q S0 L sc q S0 R sc X
Ž1yaX .
y 1q
ž 2 / CA S0 rv q aX CA S0 rv A q a X CA S0 zv v 5
" 1 TLq R 4
q 2 Hd d ˆ Am,
x 12 Am I Ž 37 .
´ Ž 4p . 2 3
where the S0 X terms are the corresponding Ž a-independent. terms of the Dimensional
Regularization classical action S0 . The bar above some of them means that all the index
in the term are barred.
Let us define the Dimensional Regularization linearized BRS operator bd as follows:
d S0 d d S0 d d S0 d d S0 d
bd ' s d q d d x Tr
H ½ d Am dr m q Tr
dv dz
q
dc d L
q
dc X d R
d S0 d d S0 d
q
dc d L
q
dc X d R 5 .
Notice that bd2 / 0 since s d S0 / 0. The action of s d on the fields has been defined in Eq.
Fig. 4. One-loop Feynman graphs contributing to the 1PI functions involving external fields.
´ D. Sanchez-Ruizr
Ž31.. If we define in the d-dimensional space-time of Dimensional Regularization the

following integrated field polynomials:
1 E 1
Lg ' 2 Hd d
x Tr Fmn F mn s g S0 , Lg s Hd d
x Tr Fmn F mn ,
2g Eg 2 g2
X
LLc ŽR
ŽX . ' ybd P
.
d d x L Ž R . PL Ž R . c ŽX . q c Ž . PR Ž L . L Ž R .
H ½ 5
i X
l X
s Ž NcL Ž R . y NL Ž R . . S0s2 d d x H ½ 2
c Ž . Eu PL Ž R . c Ž .
X X
qc Ž . g m PL Ž R . TLa Ž R . c Ž . Ama , 5
d d x i c Ž . Eû P L ŽR . c Ž .
X X
LLc ŽR
ŽX . ' Lc Ž . y
. X
H
i X X X X
s 2 dd x
H ½ 2
cŽ . Eu PL ŽR . c Ž . q c Ž . g m PL ŽR .TLaŽR . c Ž .Ama , 5
E
L A ' bd P d d x r˜ma A a m s Tr NA y Nr y NB y Nv q 2 a
H ž / S0 ,
Ea
Lv ' ybd P d d x z a v a s Tr Nv y Nz S0 ,
H
where
d
Nf ' d d x f Ž x .
H , f s Am , rm , B, v , v and z ,
df Ž x .
d d
NcL ŽR . ' d d x Ž PL ŽR . . c . b
H , NcRŽL
X
.
' d d x Ž PRŽL . . c X . b
H ,
dcb dcbX
d d
NcRŽL . ' d d x c PRŽL .
H ž / , NcLXŽR . ' d d x c X PL ŽR .
H ž / Ž 38 .
b
dcb b
dcbX
then, the functional G 0Ž1.

sing in Eq. 37 can be cast into the form
Ž .
1 " 11 1yaX
G 0Ž1.
sing s 2
g2 CA L g q 2 y ž / CA L A q a X CA Lv
Ž 4p . ´ 3 2
1 " 4 TLq R
q 2
g2 y L g q a X C L Lc q a X C R Lc X
Ž 4p . ´ 3 2
1 " TLq R 4 1
q 2 Hd d
x ˆ Am;
Am I Ž 39 .
Ž 4p . ´ 2 3 2
i.e. bd G 0Ž1. Ž ˆmn ..

sing s O g
The second and third lines of the previous equation show explicitly that the standard
‘‘infinite’’ multiplicative renormalization of the coupling constant g and of the fields of
S0 is already lost at the one-loop level.
3.4. The BRS identity and the anomalous symmetry breaking term
The Regularized Action Principle, explained in Subsection 2.3. and Appendix A,

leads to the following BRS identity:
dG DR eg dG DReg dG DReg dG DReg dG DReg
Sd Ž G DReg . ' d d x Tr
H ½ dr m
d Am
q Tr
dz dv
q Tr B
dv
dG DR eg dG DReg dG DReg dG DReg dG DReg dG DReg

q q X q
dL dc dR dc dL dc
dG DR eg dG DReg
q
dR dc X 5 s Dˆ P G DR eg q Dct P G DReg
d SctŽ n . dG DReg
q dd x
H Ý
F
½ d KF Ž x .
P G DR eg
dF Ž x . 5 , Ž 40 .
when applied to the Dimensional Regularization 1PI functional, G DReg , obtained from
Ž n.
the Dimensional Regularization action S DR eg , the latter being given in Eq. 35 . The
Ž .
symbols F and KF in the previous equation stand, respectively, for any field that
undergoes non-linear BRS transformation, i.e Am , v , c , c X , c and c X and the external
field which couple to the corresponding BRS variation, i.e. rm , z , L, R, L, and R. The
operator Dˆ is given in Eq. Ž34.. The symbol Dct is equal to s d SctŽ n., where SctŽ n. and s d
are defined in Eqs. Ž35. and Ž31., respectively. Eq. Ž40. is a particular instance of Eq.
Ž19., and it is obtained by considering the concrete realization of the Dimensional
Regularization BRS transformations given in Eq. Ž31.. This equation is a rigorous
equation, valid to all orders in the expansion in powers of " and fields.
Notice that G DR eg is not BRS invariant since Sd Ž G DReg . does not vanish. Indeed, the
far right-hand side of Eq. Ž40.,
d SctŽ n . dG DReg
D̂ P G DR eg q Dct P G DReg q d d x
H Ý
F
½ d KF Ž x .
P G DReg
dF Ž x . 5 ,
is not zero. This term is the symmetry breaking term of the dimensionally regularized
Ž n.
theory defined by the Dimensional Regularization action SDReg .
Ž n.
We have to ask now about the renormalized counterpart of Eq. Ž40.. Since S DReg
contains the singular counterterms needed to render non-singular at d s 4 the 1PI
functions of the theory up to order " n, Eq. Ž20. defines the renormalized 1PI functional
Gren up to order " n :
LIM d ™ 4 G DReg w w ,F ; KF ; m x s Gren w w ,F ; KF ; m x . Ž 41 .
Let us recall that the limiting process denoted by LIM d ™ 4 is accomplished by taking the
ordinary limit d ™ 4 first and then replacing with zero every hatted object. The
´ D. Sanchez-Ruizr
Dimensional Regularization scale has been introduced by hand as explained in Žv. of

Subsection 2.2.
Now, by particularizing Eqs. Ž18. – Ž20. to the field theory under study, one concludes
that the following equation, which is the renormalized counterpart of Eq. Ž40., holds up
to order " n :
dGren dGren dGren dGren dGren dGren dGren

S Ž Gren . ' d 4 x Tr
H ½ dr m
d Am
q Tr
dz dv
q Tr B
dv
q
dL dc
dGren dGren dGren dGren dGren dGren

q
dR dc X q
dL dc
q
dR dc X 5 s D breaking , Ž 42 .
where Gren is the dimensionally renormalized 1PI functional as defined by Eq. Ž41.. The
BRS symmetry breaking term D breaking is given up to order " n by the following
‘‘limit’’:
½
D breaking s LIM d ™ 4 Dˆ P G DReg q Dct P G DReg
d SctŽ n . dG DReg
q d x
H d
Ý
F
½ d KF Ž x .
P G DR eg
dF Ž x . 55 . Ž 43 .
The fact w32x that if the anomaly cancellation conditions are met w39x the cohomology
of the linearized BRS operator b in Eq. Ž29. is trivial on the space of local polynomials
Ž n.
of ghost number one, guarantees that Sfct can be chosen so that D breaking above vanishes
whatever the value of n. Hence, if the BRS symmetry is non-anomalous, non-symmetric
counterterms can be added to the Dimensional Regularization classical action so that the
resulting renormalized action is BRS invariant: S Ž Gren . s 0. In this paper we shall
Ž1.
compute Sfct explicitly Žsee Subsection 3.6...
Let us close this subsection and express in terms of local operators in four dimensions
the order " contribution, D Ž1.breaking , to the symmetry breaking term given in Eq. 43 .
Ž .
Adapting Eqs. Ž21. and Ž23. to the regularized theory at hand we conclude that
Ž1. Ž1 .
D Ž1.
breaking s LIM d ™ 4 ½ Dˆ P G DReg singular q bd
Ž1.
Ssct q N w Dˆ x P Gren
5 Ž1.
q b Sfct
Ž1 .
s N w Dˆ x P Gren Ž1.
q b Sfct . Ž 44 .
where Dˆ and bd are defined in Eqs. Ž34. and Ž29., respectively. Indeed, we shall show
below that
Ž1 . 1 TLqR 4 " d
D̂ P G DR eg singular
s 2 Hd x
Ž 4p . 2 3 ´
ˆ Em A m a q c a b c I
yv a I ½ ˆ Ama A m b s y bd Ssct
Ž1.
. ž / 5
Fig. 5. 1PI Feynman diagrams needed to compute Eq. Ž50..
The order " singular Žpole. contribution, w Dˆ P G DR eg xŽ1.

singular , to D P G DReg only comes
from the diagrams depicted in Figs. 5 and 6, after replacing Dˇ with Dˆ. These
contributions read
m b a Ž1 . i 2 1
D̂ P G DR eg Ž p. s 2
T LqR d a b p m pˆ 2 ,
Aw singular Ž 4p . 3 ´
mn b c a Ž1 . i 4 1
D̂ P G DR eg A Aw
Ž p1 , p 2 . s 2
ic a b c
3 ´
Ž pˆ 12 y pˆ 22 . g mn . Ž 45 .
singular Ž 4p .
Now, bd L g s bd L A s bd Lv s bd L g s bd Lc ŽX . s 0, and therefore the only non-
vanishing bd variation of G 0Ž1.
sing comes from the third line of Eq. 39 , which in turns is
Ž .
equal to Eq. Ž45.. The fact that Ssct Ž1.
s yG 0Ž1.
sing concludes the proof.
The next task to face is the computation of the order " contribution to the anomalous
insertion Nw Dˆ x P Gren . We shall carry out this calculation in the next section. The
computation is somewhat simplified if we bare in mind that the field B, enforcing the
gauge-fixing condition, has no dynamics Žno interaction vertices. unless forcibly intro-
Ž n. Ž
duced by appropriate choice of Sfct finite countertems.. Hence, it will be very advisable
to impose that the finite counterterms be independent of B, so that the only contribution
to G DR eg involving B is order " 0 and given by S0 in Eq. Ž32.. It is thus clear that the
so-called gauge-fixing equation
dGren
B Ž Gren . ' y Em A m y a B s 0 Ž 46 .
dB
holds for the renormalized theory. Notice that Eq. Ž46. does not clash with restoring the
BRS symmetry since D breaking defined in Eq. Ž43. does not depend on B. Of course, Eq.
Ž46. is the equation of motion of B.
Fig. 6. 1PI Feynman diagrams needed to compute Eqs. Ž51., Ž52. and Ž61..
´ D. Sanchez-Ruizr
Finally, the ghost equation is another interesting equation. It is just the equation of
motion of v :
d d
G Gren ' ½ dv
q Em
drm 5 Gren s 0, Ž 47 .
and is a consequence of Eq. Ž42. and the gauge-fixing equation. Some functional
differentiation and the fact that D breaking does not depend on B lead to Eq. Ž47.. A
further simplification, the ghost equation implies that the functional which satisfy it
depends on rm and v through the combination r˜m s rm q Em v . Finally, it is not difficult
to come to the conclusion that this very simplification applies to each term on the
right-hand side of Eq. Ž43. and SctŽ n. in Eq. Ž35..
3.5. Expansion of the anomalous insertion
The anomalous insertion Nw Dˆ x P Gren has ghost number q1 and it is the insertion into
the 1PI functional Gren of an evanescent integrated polynomial operator of ultraviolet
dimension 4 with neither free Lorentz indices nor free indices for the gauge group. This
anomalous insertion is a functional of c , c X , Am , v Žonly through r˜m ., v , but not of B,
and also of the external fields L, L, R, R, rm Žonly through r˜m . and z . It is thus
compulsory to generalize first Eq. Ž25. so that it includes also Feynman diagrams with
the external fields as vertices.
Due to the explicit power-counting in our model, only diagrams with either no or one
external fields can be divergent. Consider a diagram with the vertex KF sF ŽF denotes
any field which undergoes non-linear BRS transformations.. The Feynman rule in
momentum space of this vertex is an integration over a momentum qX , which is
absorbed in the definition of the Fourier transform of the function with an insertion of an
operator, and the factors "i K˜F Ž qX . times the Feynman rule of the operator insertion sF
at momentum qX . The singular part is also a polynomial in qX , therefore it can be
expanded, together with the rest of momenta, in a finite Taylor series in qX . A factor
qX m 1 . . . qX m s together with K˜ Ž qX . will lead to a Žyi . s Em1 . . . Em s KF Ž x . which multiply
the insertion of the monomial operator obtained with the rest of momenta. Therefore the
Bonneau identities ŽEq. Ž25.. are generalized to:
N w Dˆ x Ž x . P Gren
r
4 dŽJ. Ž yi . Er
sy Ý Ý
ns0 j1 . . . j n 4
Ý Ý
rs0 i1 . . . i r 4
1(i j(n
½ r! E pim1 1 . . . E pimr r
Ž yi" . r.s.p.
= f˜ j1Ž p 1 . . . . f˜ j nŽ pn . N w Dˇ x Ž q s y Ý pi .
¦ ; 1PI
Ks0 p s0, gs0
i ˇ 5
1 1
=N
n!
Ł
ksn
½ž Ł
a ri ask 4
/
Ema f jk 5 Ž x . P Gren
sq t
Ž yi . E sqt
qÝ
F s,t
Ý Ý
i1 . . . i s4
0(sqt( d Ž J ;F . 1(i j(n
½ Ž s q t . ! E pim1 1 . . . E pims sE pnq1
n1 nt
. . . E pnq1
r.s.p.
= f˜ j1Ž p 1 . . . . f˜ j nŽ pn . ;N w sF x Ž pnq1 . N w Dˇ x Ž q s y Ý pi .
¦ ; 1PI
Ks0 p s0, g'0
i ˇ 5
1 1
= Ž En 1 . . . En t KF . Ž x . N
n!
Ł
ksn
½ž Ł
a ri ask 4
/
Ema f jk 5 Ž x . P Gren , Ž 48 .
where J ' j1 , . . . , jn 4 and the indexes jk , k s 1, . . . ,n label the different

types of quantum fields, with the proviso that fields having different values
of gauge group indices are taken as different. The symbols d Ž J ., d Ž J;F .
denote, respectively, the overall degrees of ultraviolet divergence of the
1PI
1PI functions f˜ j1Ž p 1 . . . . f˜ j nŽ pn . N w Dˇ x Ž q s yÝ pi . Ks0 and f˜ j1Ž p 1 . . . . f˜ j nŽ pn . ;
¦ ; ¦ ;
²N w sF x Ž pnq 1 . N w Dˇ x Ž q s yÝ pi . :1PIKs0
. Notice that since our propagators are massless
the only non-vanishing contributions to the left-hand side of Eq. Ž48. come from
r s d Ž J . and s q t s d Ž J;F .. Ł 1ks n means that fields in the product are ordered from
left to right according to decreasing values of k.
The previous formula gives rise to an expansion of Nw Dˆ xŽ x . P Gren in terms of both
standard and evanescent monomials w7,8x Žsee Subsection 2.4... If we denote the
monomials generically by the letter M , the expansion will have the form
N w Dˆ x P Gren s d 4 x N w Dˆ x Ž x . P Grens Ý bia b c . . . N Mia b c . . . P Gren

H
i
q Ý bˆja b c . . . N Mˆja b c . . . P Grens Ý bia b c . . . N Mia b c . . . P Gren

j i
q Ý bˆjX a b c . . . N Mˆja b c . . . P Gren , Ž 49 .

j
with each coefficient being a formal series in " Žstarting at order "1 .. The latin letters
abc stand for the gauge group indices. We shall assume that repeated gauge group
indices are summed over. The monomials in Eq. Ž49. have no free Lorentz index. The
objects Mia b c . . . 4 , called ‘‘barred’’ non-omials, are monomials where all Lorentz
contractions are carried out with g mn . Notice that there is a one-to-one correspondence
between Mia b c . . . 4 and Mia b c . . . 4 . The objects Mîa b c . . . 4 , called ‘‘hatted’’ non-omials,
are evanescent operators.
Notice that in our case the monomials will be integrated local functionals of the fields
and its derivatives with ghost number one and ultraviolet dimension 4.
Expanding in the same way all the normal insertions of the evanescent monomials
Mîa b c . . . on the right-hand side of Eq. Ž49. and solving the system of Bonneau identities
´ D. Sanchez-Ruizr
we would get finally the true expansion of the anomalous insertions in terms of the basis
of standard Žor barred. normal products Žalso called standard quantum insertions.:
N w Dˆ x P Gren s Ý k ia b c . . . N Mia b c . . . P Gren s Ý k ia b c . . . N Mia b c . . . P Gren ,
i i
where the barred monomials simply means again that all the Lorentz contractions are
done with the g tensor.
Let us shorthand Hd d x to H. The list of integrated monomials of ghost number 1,
ultraviolet dimension 4 and no free of Lorentz indices which is relevant to our
computation is the following:
Ža. Monomials with one v and one A.
M1a b ' d d x v a I Em A bm .
H
Žb. Monomials with one v and two A’s and no ´mn a b tensor.
M2a b c ' v a Ž I Amb . Ac m ,

H M3a b c ' v a Ž Em Anb . Ž E mAcn . s M3acb ,
H
M4a b c ' v a Ž Em En A bm . Acn ,
H M5a b c ' v a Ž Em A bm . Ž En Acn . s M5acb ,
H
M6a b c ' v a Ž Em Anb . Ž E nAc m . s M6acb .
H
Žc. Monomials with one v and three A’s and no ´mn a b tensor.
M 7a b c d ' v a Ž Em A bm . Anc A d n s M 7a b d c ,
H
M8a b c d ' v a Ž Em Anb . Ac mA d n .
H
Žd. Monomials with one v and four A’s and no ´mn a b tensor.
M9a b c d e ' v a Amb Ac mAnd A en s M9acb d e s M9a b c e d s M9ad eb c .

H
Since the formulae with fermions c and c X are very similar we collect them in a
way obvious to interpret. In general, we will denote with a roman letter L or R the
fermionic lines or loops in Feynman diagrams, the monomials, their coefficients, . . .
corresponding to the fermions c , whose interaction is left-handed, or to the fermions
fermions c X , whose interaction is right-handed.
Že. Monomials with one v and c , c Ž c X , c X ..

X X
M10a ,iLjŽR . ' v a c iŽ .g m P L ŽR . Em c jŽ . ,
H
X X
M11a ,iLjŽR . ' v a Em c iŽ . g m PL ŽR . Em c jŽ . ,
H ž /
i, j denoting the group indices of the corresponding fermion fields.
Žf. Monomials with one v , one A and c , c Ž c X , c X ..
M12a bL,iŽRj . ' v a c iXg m PL ŽR . c jX Amb .

H
Žg. Monomials with fermions and external fields.
X
M14a bL,iŽRj . ' v av b L i Ž R i . PL ŽR . c jŽ . ,
H
X
M15a bL,iŽRj . ' v av b c iŽ . PRŽL . L j Ž R j . .
H
We have not considered operators like v a c i g m PR Em c j as admissible monomials
since it will not be generated by the Bonneau identities due to the form of our choice for
fermion vertex at order " 0 .
Will shall adopt the notation that the . . . 4 enclosing indices denotes symmetrization
and w . . . x antisymmetrization.
Žh. Monomials with ´mn a b .
M50a b c ' ´mn a b v a Ž E aA bm . Ž E bAcn . s M50acb ,

H
M51a b c d ' ´mnra v a Ž E aA bm . AcnA d r s yM
H M51a b d c ,
M52a b c ' ´mnrl v a A bmAcnA d rA e l s M52aw b c d ex .

H
Ži. Monomials with r˜ or z .
Notice that it is possible to construct other monomials with ghost number q1 and
dimension 4: r˜ma v bv cA d m , r˜ma Ž E mv b . v c, z av bv cv d. The formulae for their coefficients
in the Bonneau expansion are also easily obtained, but we do not write them because in
the one-loop computation all this coefficients turn out to be zero. This is clear because
all the 1PI functions we would have to compute involve at least two loops.
3.5.1. Anomalous monomials, i.e. with some ĝ

Take all the monomials written above and write all monomials obtained by adding
hats to the Lorentz indices in all possible ways. No hatted index should appear
contracted with the ´X mn a b , because this X
contraction
X
vanish. NoticeX that with the
fermionic vertices c Ž . g m P L ŽR .TLaŽR . c Ž .Ama s c Ž . PRŽL . g m P L ŽR .TLaŽR . c Ž .Ama in the Di-
Fig. 7. 1PI Feynman diagrams needed to compute Eqs. Ž53., Ž54. and Ž62..
´ D. Sanchez-Ruizr
Fig. 8. 1PI Feynman diagrams needed to compute Eqs. Ž56. and Ž63..
mensionalX
Regularization
X
classical action S0 , neither v a c i P L gˆ m P L Em c j nor
a m
v c i PR gˆ PR Em c j occur
Ž. Ž.
X
in the Bonneau X
expansion we
X
are considering; however, X
Ž . m a Ž . a aŽ Ž . m a
with
X
the vertex c X
g P L Ž R . T L Ž R . c A m s A m c P R Ž L . g P L Ž R .T L Ž R . c
Ž .
q
c Ž . PL ŽR . gˆ m TLaŽR . PL ŽR . c Ž . ., both do.
It is convenient to use the expansion in terms of barred and hatted monomials Žso the
1PI functions with Dˇ inserted have to be expressed in terms of g and gˆ .. Then, the
formulae of the coefficients for all orders and the result for the first-order read Žsee Figs.
7–12.:
Ža. From 1PI functions with one v and one A ŽFig. 5.:
3
b 1a b s y Ž yi . = coef. in r.s.p. ² v a Ž p 2 ' yp1 . Amb Ž p 1 . N w Dˇ x Ž q s 0 . :1PI
Ks0
of p 1 2 p 1 m
3
1 Ž yi . E 3
sy r.s.p.² v a Ž yp1 . Amb Ž p 1 . N w Dˇ x Ž q s 0 . :1PI
Ks0
48 E p 1n E p 1 n E p 1 m p i'0
1 T L q TR
sy 2
d a b "1 q O Ž " 2 . . Ž 50 .
Ž 4p . 3
Žb. From 1PI functions with one v and two A’s, and no ´mn a b tensor ŽFig. 6.:
b 2a b c , b4ab c 4 s y Ž yi . 2 = coef. in r.s.p.
=² v a Ž p 3 ' yp1 y p 2 . Amb Ž p 1 . Anc Ž p 2 . N w Dˇ x Ž q s 0 . :1PI
Ks0
2
of ½p 1 gmn , p 1 m p 1 n 5 respectively
1
s 2 Ž TL q TR . c a b c y 13 , 23 4 "1 q O Ž " 2 . , Ž 51 .
Ž 4p .
b 3a b c s b 3acb , b5a b cs b5acb , b6a b cs b6acb 4
2
Ž yi .
sy = coef. in r.s.p.
2
=² v a Ž p 3 ' yp1 y p 2 . Amb Ž p 1 . Anc Ž p 2 . N w Dˇ x Ž q s 0 . :1PI
Ks0
of ½ p Pp
1 2 gmn , p 1 m p 2 n , p 1 n p 2 m 5 resp.s0 "1 q O Ž " 2 . . Ž 52 .
Žc. From 1PI functions with one v and three A’s, and no ´mn a b tensor ŽFig. 7.:
Ž yi .
b 7a b c d s b 7a b d cs y = coef. in r.s.p.
2
3
¦ž
= v a y Ý pi Amb Ž p 1 . Anc Ž p 2 . Ard Ž p 3 . N w Dˇ x Ž qs0 .
i
/ ; 1PI
Ks0 of p 1 m gnr
1 1 abcd ac d b ab dc adcb ac b d adbc 1 2

s 2
TLq R q T LqR q T LqR q T LqR y T LqR y T LqR " q O Ž " . ,
Ž 4p . 6
Ž 53 .
b 8a b c d s y Ž yi . = coef. in r.s.p.
3
¦ž i
/ ;
= v a y Ý pi Amb Ž p 1 . Anc Ž p 2 . Ard Ž p 3 . N w Dˇ x Ž q s 0 . 1PI
Ks0
of p 1 n gm r q p 1 r gmn
ž /
1 1 ab dc acb d ac d b adbc abcd adcb 1 2
s 2
TLq R q T LqR q T LqR q T LqR y T LqR y T LqR " q O Ž " . .
Ž 4p . 3
Ž 54 .
Žd. From 1PI functions with one v and four A’s, and no ´mn a b tensor ŽFig. 8.:
1
b 9a b c d e s b 9ab c4 d e4s b 9ad e4 b c4s y = coef. in r.s.p.
8
=² v aAmb Anc Ard Ale N w Dˇ x Ž q s 0 . : 1PI

Ks0 of
gmn s 0 "1 q O Ž " 2 . . Ž 55 .

In principle, due to the presence of g 5 , there should be also evanescent normal
products in the expansion of the anomalous insertion Nw Dˆ x P Gren . Indeed, the coefficient
of HNw v a ˆ Em A bm x is
I
3
y Ž yi . = coef. in r.s.p. ² v a Ž p 2 ' yp1 . Amb Ž p 1 . N w Dˇ x Ž qs0 . :1PI ˆ 1 2 p1 m
Ks0 of p
3
Ž yi . E 3
1
sy r.s.p.² v a Ž yp1 . Amb Ž p 1 . N w Dˇ x Ž q s 0 . :1PI
Ks0 p '0
8 Ž d y 4 . E pˆ 1n E pˆ 1 n E p 1 m i
1 T L q TR
sy 2
d a b "1 q O Ž " 2 . ,
Ž 4p . 3
´ D. Sanchez-Ruizr
However, a simplification occurs, for the coefficient of HNw v a

I Eˆm A bm x is
3
y Ž yi . = coef. in r.s.p. ² v a Ž p 2 ' yp1 . Amb Ž p 1 . N w Dˇ x Ž qs0 . :1PI
Ks0 of p 1
2
pˆ 1 m
3
1 Ž yi . E 3
sy r.s.p.² v a Ž yp1 . Amb Ž p 1 . N w Dˇ x Ž qs0 . :1PI
Ks0
8 Ž d y 4 . E p 1n E p 1 n E p̂ 1 m p i'0
s 0 "1 q O Ž " 2 . ,
and so on. Due to our expansion in barred and hatted objects rather than in standard and
hatted ones, and because of the form of the regularized interaction it turns out that the
rest of anomalous coefficients are 0 in the one-loop approximation. Nevertheless, as we
know from Subsection 2.4, these anomalous coefficients only matter at the next
perturbative order, so this vanishing of the coefficients is only a simplification relevant
to higher order computations.
Že. From 1PI functions with one v and one c , c Ž c X , c X . pair ŽFig. 9.:
a ,i j
b 10 L ŽR .
s y Ž yi . = coef. in r.s.p.
X X
=² v a Ž p 3 ' yp1 y p 2 . cbŽ j. Ž p 2 . caŽ i. Ž p 1 . N w Dˇ x Ž qs0 . :Ks0
1PI
of Ž pu 2 PL ŽR . . a b
1 Ž yi . E
sy r.s.p.
4 P 2 E p 2m
X X
=² v a Ž yp1 y p 2 . cbŽ j. Ž p 2 . caŽ i. Ž p 1 . N w Dˇ x Ž qs0 . :1PI
Ks0 p i'0
P Ž P L ŽR . g m . ba
i E
s Tr r.s.p.
4P2 E p 2m
X X
=² v a Ž yp1 y p 2 . c j Ž . Ž p 2 . c i Ž . Ž p 1 . N w Dˇ x Ž qs0 . :1PI
Ks0 PL Ž R . g m
p i'0
1 CA ij
sy
Ž 4p .
2
g2 ½ž CL Ž R . y
4 / TLa Ž R .
CL Ž R . CA ij
q Ž a X y 1. ž 6
y
4 / T LaŽR . 5 "1 q O Ž " 2 . , Ž 56 .
a ,i j
b 11 L ŽR . s y Ž yi . = coef. in r.s.p.
X X
ˇ x Ž q s 0 . :Ks0
=² va Ž p 3 ' -p 1-p 2 . cbŽ j. Ž p 2 . caŽ .i Ž p 1 . N w D 1PI
of
i E
Ž pu 1 PL ŽR . . a b s 4 P 2 Tr E p 1m
r.s.p.
X X
=² v a Ž yp1 y p 2 . c j Ž . Ž p 2 . c i Ž . Ž p 1 . N w Dˇ x Ž q s 0 . :Ks0
1PI
p i'0
1 CA ij
=PL Ž R . g m s y
Ž 4p .
2
g2 ½ž CL Ž R . y
4 / TLa Ž R .
CL Ž R . CA ij
q Ž a X y 1. ž 6
y
4 / TLaŽR . 5 "1 q O Ž " 2 . . Ž 57 .
Žf. From 1PI functions with one v , one A and one c , c Ž c X , c X . pair ŽFig. 10.:
a b ,i j
b 12 L ŽR .
X X
s y1 = coef. in r.s.p. ² v aAmb cbŽ j.caŽ i. N w Dˇ x Ž q s 0 . :1PI m
Ks0 of Ž g P L ŽR . . a b
1 X X
sy Tr r.s.p.² v aAmb c j Ž . c i Ž . N w Dˇ x Ž q s 0 . :Ks0
1PI
PL Ž R . g m
4P2 p i'0
1 1 q Ž a X y 1.
sy 2
g2 CA i TLa Ž R . ,T Lb Ž R . ij "1 q O Ž " 2 . . Ž 58 .
Ž 4p . 4
Notice that the gauge group structure of the first five diagrams in Fig. 10 involves
three or more gauge group generator matrices. The gauge group structure of each
diagram is rather involved but it turns out that the r.s.p. of each of the first three
diagrams vanishes, that the r.s.p. of the fourth cancels exactly against the fifth, and,
Fig. 9. 1PI Feynman diagrams needed to compute Eqs. Ž56. and Ž57.. ŽNotice that the gauge group structure is
explicitly shown under these and subsequent diagrams..
´ D. Sanchez-Ruizr
thanks to the Jacobi identity, that the result of the last four diagrams combine in pairs to
give Eq. Ž58., where the gauge group structure is simply a commutator. This is
remarkable, because we shall see in Appendix C that the difference between the
one-loop 1PI fermionic vertex computed a` la Breitenlohner and Maison and the same
1PI function evaluated with the ‘‘naive’’ prescription involves only a single gauge group
generator and, hence, it is sensible to expect that the finite counterterm that is needed to
restore at the one-loop level the BRS symmetry in the Breitenlohner and Maison
formalism involved only a generator. The BRS variation, b, of this counterterm yields a
contribution linear in Amb which will match the gauge group structure of the symmetry
breaking diagrams in Fig. 10 thanks to the various cancellations and simplifications just
described.

Žg. From 1PI functions with fermions and external fields ŽFigs. 11 and 12.:
i X X
a b ,i j
b 14 L ŽR . s = coef. in r.s.p. ² v av bcbŽ j. N scaŽ i. N w Dˇ x Ž q s 0 . :1PI
Ks0 of
2"
= Ž PL ŽR . . a b
1 CA
s 2
g 2 1 q Ž a X y 1. TLa Ž R . ,T Lb Ž R . ij "1 q O Ž " 2 . . Ž 59 .
Ž 4p . 8
i X X
a b ,i j
b 15 L ŽR . s = coef. in r.s.p. ² v av bcbŽ j. N scaŽ i. N w Dˇ x Ž q s 0 . :1PI
Ks0 of
2"
= Ž PRŽL . . a b
1 CA
sy 2
g 2 1 q Ž a X y 1. TLa Ž R . ,T Lb Ž R . ij "1 q O Ž " 2 . . Ž 60 .
Ž 4p . 8
As in case Žf., it seems impossible that the gauge group structure of the diagrams be
matched by gauge group structure of a simple one-loop finite counterterm. But the last
pair of diagrams of each 1PI function vanishes, and the other pair of antisymmetric
diagrams conspire with the help of the Jacobi identity to give a simple commutator as
result.
Žh. From 1PI functions with ´mn a b ŽFigs. 6–8.:
These coefficients are very important, because if they are non-zero, give monomials
in the expansion of the breaking that can not be cancelled by finite counterterms, i.e.
they give the essential non-Abelian chiral anomaly.
2
Ž yi .
b50a b c s b50
ac b
2
= v a Ž p 3 ' yp1 y p 2 . Amb Ž p 1 . Anc Ž p 2 . N w Dˇ x Ž q s 0 .

¦ ; 1PI
Ks0 of
=´mn a b p 1 a p 2 b
1 1 abc 1 2
sy 2
d Ly R " qOŽ " . , Ž 61 .
Ž 4p . 3
´ D. Sanchez-Ruizr
abcd ab dc
b51 s yb51
Ž yi .
2
3
¦ž
= v a p4 ' y Ý pi Amb Ž p 1 . Anc Ard N w Dˇ x Ž q s 0 .
i
/ ; 1PI
Ks0 of ´mnra p 1 a
1 1 abcd 1 2
sy 2
DLy R " qOŽ " . , Ž 62 .
Ž 4p . 6
abcde aw b c d ex
b52 s b52
1
sy = coef. in r.s.p. ² v aAmb Anc Ard Ale N w Dˇ x Ž q s 0 . :1PI
Ks0 of ´mnrl
4!
s O Ž "2 . , Ž 63 .
with the definitions
d La b c s Tr T La TLb ,T Lc 4 s d a
L
b c4
,
DLa b c d s yi 3! Tr T La TLw bT L cT L d x s 12 Ž d La b e c ec d q d Lac e c e d b q d La d e c eb c . ,

and the same for d Ra b c and DRa b c d, and d LyR ' d L y d R , DLyR ' DL y DR .
Of course, we write O Ž " 2 . because we have computed at first-order in "; but due to
the Adler–Bardeen theorem for non-Abelian gauge theories w32x, all these coefficient
should be zero at higher orders if there would not be an anomaly at first-order.
Our one-loop computation of these coefficients give the non-Abelian (essential)
chiral gauge anomaly:
4
Hd x ½ b abc
50 N ´mn a b v a Ž E aA m b . Ž E bAn c .
qb51a b c d N ´mn a b v a Ž E aA m b . An cA r d q b52

abcde
N ´mn a b v aA m bAn cA r dA l e 5P
1 1
Gren s y 2
abc
´mn a b d LyR v a Ž E aA m b . Ž E bAn c .
Ž 4p . 3
q 12 ´mnra DLyR v Ž E aA m b . An cA r d 4
abcd a
1 1
q O Ž "2 . s 2
´mnrs v a E m d LyR
abc
Ž E nA r b . As c
Ž 4p . 3
n b rc s d
q 16 DLy
abcd
R A A A 4 q O Ž "2 .
1 2 i
s
Ž 4p . 3
2 ½ ž
´mnrs Tr LyR v E m E nA rAs y
2
AnA rAs /5 q O Ž "2 . .
Remember that in the formulae for the coefficients, a sum over the different left or right
representations is understood. Therefore, the matter representation may be chosen such
that the coefficients of the essential anomaly above vanish.
Finally, as explained at the end of Subsection 2.4, the coefficients k i of the true
expansion in the quantum Žstandard. basis of insertions are equal to the coefficients bi
in the one-loop approximation. Moreover k iŽ1. s k iŽ1..
3.5.2. The expression of the breaking at order "

Putting it all together we get the complete form of the integrated breaking, wNw Dˆ x P
Gren xŽ1., at order " Žthis formula is a four-dimensional one!., which reads
Ž1 .
N w Dˆ x P Gren
1 T L q TR 1 T L q TR
sy 2
d a b M1a b "1 y 2
c a b c M2a b c "1
Ž 4p . 3 Ž 4p . 3
1
q 2 Ž TL q TR . 23 c a b c M4a b c "1
Ž 4p .
1 1 acb d adbc eb c e d a
q 2 6
T Lq R q T LqR y Ž T L q TR . Ž c c q c eb d c ec a . M 7a b c d "1
Ž 4p .
1 1 abcd a d cb eb c e d a
q 2 3
T Lq R q T LqR q Ž T L q TR . Ž c c q c eb a c e d c . M8a b c d "1
Ž 4p .
1 CA CL CA
y 2
g 2 CL y q Ž a X y 1. ž y /Ž M10 L q M11L . "1
Ž 4p . 4 6 4
1 CA CR CA
y 2
g 2 CR y q Ž aX y1 .ž y /Ž M10 R q M11R . "1
Ž 4p . 4 6 4
i 2
1 q Ž a X y 1.
y 2
g CA Ž M12 L y M13L q M12 R y M13R . "1
Ž 4p . 4
i 1 q Ž a X y 1 . CA
q 2
g2 c a b c Ž M14a bLc y M15L
abc
q M14a bRc y M15R
abc
. "1
Ž 4p . 4 8
1 2 i
q 2 Hd 4
x ´mnrs Tr LyR v E m E nA rAs y ž AnA rAs / "1 , Ž 64 .
Ž 4p . 3 2
ij X X
M10 L ŽR . s Ž TLaŽR . . M10a ,iLjŽR . s d 4 x v a c Ž . g m P L ŽR .T LaŽR . Em c Ž . ,
H
ij X X
M11 L ŽR . s Ž TLaŽR . . a ,i j
M11L 4
ŽR . s d x vHa
Em c Ž . g m PL ŽR .TLaŽR . c Ž . ,
ž /
ij X X
M12 L ŽR . s Ž TLaŽR .TLbŽR . . M12abL,iŽRj . s d 4 x v a c Ž . g m PL ŽR .TLaŽR .TLbŽR . c Ž .Amb ,
H
´ D. Sanchez-Ruizr
ij X X
M13 L ŽR . s Ž TLbŽR .TLaŽR . . M12a bL,iŽRj . s d 4 x v a c Ž . g m PL ŽR .TLbŽR .TLaŽR . c Ž .Amb ,
H
ij X
M14a bLcŽR . s Ž TLc ŽR . . M14a bL,iŽRj . s d 4 x v a v b L Ž R . P L ŽR .T Lc ŽR . c Ž . ,
H
ij X
M15a bLcŽR . s Ž TLc ŽR . . a b ,i j
M15L H4 a b Ž. c
ŽR . s d x v v c PRŽL . T L ŽR . L Ž R . ,
and we have used the following equation:

abcd adcb ab dc acb d ac d b adbc
y2TLqR y 2TLqR q T LqR q TLqR q TLqR q TLqR
s Ž TL q TR . Ž c eb c c e d a q c eb a c e d c . . Ž 65 .
Notice that if the classical theory were CP invariant, the combination of monomials
obtained have the same CP definite value as the breaking, as we would have expected
from the fact that Dimensional Regularization respects discrete symmetries. But we have
not assumed any discrete symmetry property when writing the basis of Bonneau
monomials because we work with general representations.
The result in Eq. Ž64. can be obtained by computing the left-hand side of Eq. Ž64. to
the one-loop order, i.e. calculating the renormalized part of all possible one-loop 1PI
diagrams with the insertion Dˆ; their finite part has to be a local term, because when
going to the four-dimensional space all Ž d y 4.-objects are set to be 0, so the only
remaining finite part of such a diagram must come from a polynomial singular part
formed by a four-dimensional object and a contracted Ž d y 4.-object which cancel the
singularity in d y 4. Therefore, being local this one-loop renormalized 1PI functions,
there are operators Žin the four-dimensional space. whose Feynman rules attached with a
"1 give the same result. The sum of this operators is precisely the right-hand side of Eq.
Ž64..
In fact, the one-loop calculation of the coefficients of the Bonneau identities via r.s.p.
of diagrams with the insertion of Dˇ are exactly the same as the finite part of diagrams
with the insertion of Dˆ, excepting some factors in the Bonneau coefficients which
directly give the operators whose Feynman rules lead to the same result.
But the Bonneau identities unravel the problem in higher order computations, and the
formulae of the coefficients given above would be the same Žalthough with new
Feynman rules coming from the finite counterterms we would have added in the
previous recursive step..
(1)
3.6. Restoring the BRS symmetry: the computation of Sfct
We know from algebraic renormalization theory and BRS cohomology w32x that there
exists an integrated local functional of the fields and their derivatives, let us call it X Ž1.,
such that the anomalous contribution wNw Dˆ x P Gren xŽ1., given in Eq. Ž64., can be cast into
the form
Ž1 .
N w Dˆ x P Gren
1 2 i
s 2 Hd 4
x ´mnrs Tr LyR v E m E nA rAs y
ž AnA rAs / "1 q bX Ž1. .
Ž 4p . 3 2
The integrated local functional, X Ž1., has ghost number zero and ultraviolet dimension 4.
Ž1.
Hence, by choosing Sfct in Eq. Ž35. so that it verifies
Ž1.
Sfct s yX Ž1. ,
and recalling that Eqs. Ž44. and Ž42. hold, we conclude that the BRS identity is broken
at order " by the essential non-Abelian gauge anomaly only:
S Ž Gren .
1 2 i
s 2 Hd 4
x ´mnrs Tr LyR v E m E nA rAs y
ž AnA rAs / "1 q O Ž " 2 . .
Ž 4p . 3 2
The previous equation leads in turn to the result we sought for; namely, that if the
fermion representations meet the anomaly cancellation criterion w39x, the BRS symmetry
can be restored Žup to order " . by an appropriate choice of finite counterterms Sfct
Ž1.
.
Ž1. Ž1. Ž1.
Our next task will be to compute Sfct s yX . We shall demand that Sfct do not
depend neither on B nor r˜ . Recall that these fields do not occur in Eq. Ž64..
The more general approach to this computation, similar to our study of the expansion
of wNw Dˆ x P Gren xŽ1., would be
Ž1. Write the complete list of possible monomials of ultraviolet dimension 4, ghost
number 0 and with no free Lorentz indices 4 ;
Ž2. Write the general expression of the finite counterterms as a linear combination
of those monomials with coefficients depending on gauge group indices.
Ž3. Take their b-variation and finally
Ž4. impose that this variation cancels the expansion of the breaking term wNw Dˆ x P
Gren xŽ1., excepting the essential non-Abelian anomaly. This should lead to a
compatible indeterminate system with the coefficients as variables and with a
degree of indeterminacy exactly equal to the number of b-invariant operators in
the classical theory.
But, writing such list and taking its b variation is a tedious task: lots of indices would
come in and the resolution of the final system would appear a bit messy.
It is clear that the gauge group structure of the finite counterterms which would be
needed in the one-loop computation to restore the BRS symmetry is not arbitrary. A
natural ansatz for the gauge group structure of these counterterms is to take the one-loop
gauge group structure which can appear while computing the Feynman diagrams.
Therefore, we write
2
Ž1.
Sfct s a1 HŽ Em A m . q a2 Am I A m q c ic a b c Ž Em Ana . A bmAcn
H H
q Ž d L d La b c q d R d Ra b c . HŽ E
m Ana . A bmAcn q Ž f L TLa b c d q f R TRa b c d .
4
This list is a list in the four-dimensional space-time. Of course, some generalization of them to the
d-dimensional space-time of Dimensional Regularization has to be chosen while writing them in the
Dimensional Regularization action; but the differences, being Ž dy4.-objects, will begin to produce any effect
at the following order in " Žbecause they require loops to yield a non-zero contribution and there is already an
Ž1. .
explicit " in Sfct .
´ D. Sanchez-Ruizr
i l
= A a mAmb AcnAnd q e c a b c c a d e A bmAcnAmd Ane q n L
H H H 2 cg m
PL pr m c
i X l
q nR H2cg m
PR pr m c X q p L cg m P L T La c Ama q p R c Xg m PR TRa c XAma
H H
q H u1L Lsc q u1R Rsc X q u 2 L Lsc q u 2 R Rsc X . Ž 66 .
Notice that because of the reason given above we have not written terms with r or v .
Moreover, we have the combinations of terms given in Eq. Ž30. which are b-in-
variant. Obviously, these combinations are not relevant to our computation. Therefore,
we can reduce the basis of possible finite counterterms to a minimum number taking
into account these invariant counterterms.
L A and Lv contain r˜ m, which we supposed already not to appear in the finite
counterterms, so these invariant terms can not be used to reduce more the number of
finite counterterms. But, using L g we can impose e s 0 and by taking advantange of
Lc ŽX . we can set p L ŽR . s 0, without loss of generality.
Then, there should be a unique solution for the problem of cancellation of spurious
anomalies in terms of the rest of variables a1 , a2 , c, d L ŽR . , f L ŽR . , n L ŽR . , u1L ŽR . and
u 2 L ŽR . .
We recast Eq. Ž66. in the following form:
Ž1.
Sfct s a1 D˜1 q a2 D˜2 q c D˜3 q d L D˜4 L q d R D˜4 R q f L D˜5L q f R D˜5R q n L D˜6 L
q n R D˜6 R q u1L D˜7 L q u1R D˜7 R q u 2 L D˜8 L q u 2 R D˜8R . Ž 67 .

We need the b variations of the D˜’s expanded in terms of the integrated monomials we
called M ’s. Moreover, the coefficients of these expansions should be symmetrized if the
gauge group indices of the corresponding monomials have some symmetry properties.
This is necessary because, for example, M3a b c s M3acb and then not all M3a b c are linearly
independent. The result reads
b D˜1 s 2 d a b M1a b y 2 c a b c M4a b c , b D˜2 s y2 d a b M1a b q 2 c a b c M2a b c ,

i
b D˜3 s yic a b c M4a b c q ic a b c M2a b c y w c eb c c e d a q c eb d c ec a x M7a b c d
2
q i w c e b c c e d a q c e b a c e d c x M8a b c d ,
b D˜4 L ŽR . s yd La bŽRc . M2a b c y d La bŽRc . M3a b c q d La bŽRc . M4a b c q d La bŽRc . M5a b c
y Ž d LebŽRc . c e d a q d LebŽRd . c ec a . M 7a b c d q d LebŽRd . c ec a M8ab c d ,
b D˜5L ŽR . s y TLa bŽ Rc d. q TLacŽ Rd b. q T La bŽ Rd c. q TLa dŽ Rc b. M 7ab c d
y 2 TLacb d ab dc ac d b adbc
Ž R . q TL Ž R . q TL Ž R . q TL Ž R . M8a b c d
y 14 Ž TLmŽcRd .e q TLmŽdRec. q TLmŽcRe d. q TLmŽeRd .c . c m a b

q Ž TLmŽbRd.e q TLmŽdReb. q T LmŽbRe .d q TLmŽeRd .b . c m ac
q Ž TLmŽebR .c q TLmŽbRc .e q TLmŽecb m cb e

R . q TL Ž R . . c
m ad
q Ž TLmŽdRb.c q TLmŽbRc .d q T LmŽdRcb. q T LmŽcbR .d . c m a e M9a b c d e ,

s y 2TLacb d adbc
Ž R . q 2T L Ž R . y Ž T L q TR . Ž c
eb c e d a
c q c eb d c ec a . M 7a b c d
y 2 2TLa bŽ Rc d. q 2TLa dŽ Rc b. q Ž TL q TR . Ž c eb c c e d a q c eb a c e dc . M8ab c d
q 0 M9a b c d e ,
b D˜6 L ŽR . s M10 L ŽR . q M11L ŽR . ,
i
b D˜7 L ŽR . s M11L ŽR . q i M13L ŽR . q c a b c M14a bLcŽR . ,
2
i
b D˜8 L ŽR . s M10 L ŽR . y i M12 L ŽR . y c a b c M15L
abc
ŽR . . Ž 68 .
2
The coefficient in b P D˜5L ŽR . of M9a b c d e turns out to be zero!, which can be proved by
expressing the c a b c ’s as commutators of the corresponding matrices and then using the
cyclicity of the trace. If this coefficient would not be zero the system of equations below
would be incompatible! We have also used the relationship given in Eq. Ž65..
Now we impose that the sum of the terms on the right-hand side of Eq. Ž68. matches
the right-hand side of Eq. Ž64., after removal, of course, of the term carrying the
essential non-Abelian anomaly. The following linear system of equations is thus
obtained:
1 T L q TR
from M1a b : 2 a1 y 2 a 2 s 2
;
Ž 4p . 3
1 T L q TR
from M2a b c : y2 a 2 y i c s y 2
and d L ŽR . s 0;
Ž 4p . 3
1
from M4a b c : 2 a1 q i c s 2 Ž TL q TR . 23 and d L ŽR . s 0;
Ž 4p .
from M5a b c : d L ŽR . s 0, we will not write this unknown anymore;
i 1 T L q TR
from M 7a b c d : y c q T L f L q TR f R s 2
2 Ž 4p . 6
1 1
and y 2 f L ŽR . s y 2
;
Ž 4p . 6
1 T L q TR
from M8a b c d : i c y 2T L f L y 2TL f R s y 2
Ž 4p . 3
1 1
and y 4 f L ŽR . s y 2
;
Ž 4p . 3
f L ŽR .
from M9aLbŽR
cd
.: = 0 s 0;
2
´ D. Sanchez-Ruizr
from M10 L ŽR . : n L ŽR . q u 2 L ŽR .
1 CA CL Ž R . CA
s 2
g 2 CL Ž R . y q Ž a X y 1. ž y / ;
Ž 4p . 4 6 4
from M11 L ŽR . : n L ŽR . q u1L ŽR .

1 CA CL Ž R . CA
s 2
g 2 CL Ž R . y q Ž a X y 1. ž y / ;
Ž 4p . 4 6 4
i CA
from M12 L ŽR . : yi u 2 L ŽR . s 2
g 2 1 q Ž a X y 1. ;
Ž 4p . 4
i CA
from M13 L ŽR . : i u1L ŽR . s y 2
g 2 1 q Ž a X y 1. ;
Ž 4p . 4
i i CA
from M14a bLcŽR . : y u1L ŽR . s 2
g 2 1 q Ž a X y 1. ;
2 Ž 4p . 8
i i CA
from M15a bLcŽR . : u1L ŽR . s y 2
g 2 1 q Ž a X y 1. .
2 Ž 4p . 8
Notice that there are more constraints than unknowns. In fact, several equations appear
repeated and the system has 14 different equations and 13 unknowns, which turns to be
compatible whose unique solution is
1 5 1 1
a1 s 2 Ž T L q TR . , a2 s 2 Ž T L q TR . ,
Ž 4p . 12 Ž 4p . 4
i 1
cs 2 Ž T L q TR . , d L ŽR . s 0,
Ž 4p . 6
1 1 1 CL Ž R .
f L ŽR . s 2
, n L ŽR . s 2
g 2 C L Ž R . q Ž a X y 1. ,
Ž 4p . 12 Ž 4p . 6
1 CA
u1L ŽR . s u 2 L ŽR . s y 2
g2 aX.
Ž 4p . 4
Ž1.
In summary, the following choice of Sfct removes up to order " the spurious
anomaly terms from the BRS equation ŽEq. Ž42..:
"1 5 1
Ž1.
Sfct s Hd d
x
½ Ž 4p .
T q TR .
2 Ž L ŽE Am .
12 m
2
q
4
Am I A m
"1 TL q TR
y 2
c a b c Ž Em Ana . A bmAcn
Ž 4p . 6
"1 TLa b c d q TRa b c d

q 2
Ama A bmAnc A d n
Ž 4p . 12
"1 i l
q 2
g 2 CL q Ž a X y 1 . CLr6 cg m PL pr m c
Ž 4p . 2
"1 i l
q 2
g 2 CR q Ž a X y 1 . CRr6 c Xg m PR pr m c X
Ž 4p . 2
"1 CA
y
Ž 4p .
2
g2
4
1 q Ž a X y 1. ž Lsc q Lsc q Rsc q Rsc /X X
5
q "1 l Ž1. 1 Ž1. 1 Ž1.
X X 1 Ž1. 1 Ž1.
g L g q " lc Lc q " lc Lc q " l A L A q " lv Lv , Ž 69 .
where l Ž1. Ž1. Ž1. Ž1.
g , lc , lc , l A and lv
X Ž1.
are arbitrary coefficients which will determine the
renormalization conditions at the one-loop order and L g , Lc , Lc X , L A , Lv are any of the
generalizations to the d-dimensional space-time of Dimensional Regularization of the
Ž1.
corresponding b-invariant four-dimensional operators. We have written Sfct in the
d-dimensional space of Dimensional Regularization so that diagrammatic computation
can be readily done.
Notice that no terms in v or r are added, excepting the combination r˜ in the
invariant counterterms; therefore the ghost equation holds up to order ".
Notice also that, because of finite counterterms depending of the external fields L, L,
R and R have been added, we have now, e.g. Žset lvŽ1. s 0.,
dGren
/ N w sc x Ž x . P Gren ,
d LŽ x . L' 0
dGren
s N w sc x Ž x . P Gren q "1 u1L
Ž1.
sc Ž x . q O Ž " 2 .
d LŽ x . L' 0
s N sc q "1 u1L
Ž1.
sc Ž x . P Gren q O Ž " 2 . ,
which can be interpreted as a ‘‘non-minimal’’ renormalization of the operator insertion
Žwe always apply minimal subtraction to the regularized diagrams, but the action have
explicit higher order finite counterterms.. Moreover, with l AŽ1. / 0 or lvŽ1. / 0, the
insertion of the non-linear BRS variations of the fields can be renormalized in different
ways, all of them compatible with the BRS identities Ži.e. there are several renormaliza-
tion conditions for that insertions compatible with the identities..
Of course, the finite counterterms given in Eq. Ž69., are strongly dependent on the
choice for the O Ž " 0 . action, S0 Žthe Dimensional Regularization classical action in Eq.
Ž35... If we would have chosen the other mentioned fermionic vertex, the counterterms
would have certainly been completely different Žand more complicated.. But anyway the
procedure would have been equally correct.
Notice that if the classical original theory is CP invariant the finite counterterms of
Eq. Ž69. are also CP invariant. This was expected, due to the comments in the second
paragraph below Eq. Ž34..
´ D. Sanchez-Ruizr
3.7. The renormalization group equation
It is clear that the standard textbook techniques usually employed to derive the
renormalization group equation for dimensionally renormalized vector theories like QED
and QCD are of no use here. The formalism of bare fields, bare coupling constants and
multiplicative renormalization of the classical action breaks down as shows Eq. Ž37..
The generalization of this formalism to include evanescent operators, and the corre-
sponding renormalized coupling constants, becomes a bit awkward when the regularized
theory does not posses the symmetries that should hold in the quantum field theory.
Indeed a host of bare evanescent non-symmetric operators, each bringing in a new
renormalized coupling constant, may enter the game. This is in addition to the finite
non-evanescent non-symmetric counterterms needed to restore the symmetry broken in
the dimensional renormalization process. These finite counterterms should also come
from bare operators. Hence, the construction of an appropriate bare action and the
derivation of the true Ži.e. only involving as many coupling constants and anomalous
dimensions as there are classical couplings and fields. renormalization group equation is
a rather involved task Žfor related information see Refs. w47–49x.. We shall abandon this
quest altogether. Indeed, thanks to the Quantum Action Principle, Bonneau identities
and the formalism of algebraic renormalization there is no need to introduce bare fields
and bare couplings to derive the true renormalization group equation. We only need the
renormalized 1PI functional, Gren , which is symmetric up to the order in " demanded,
obtained by the method explained in this paper.
Let Gren w w ,F , KF ; g, a X , m x be the dimensionally renormalized up to order " n 1PI
functional of our theory, which we will take to be anomaly free all along the current
subsection. Gren depends explicitly on the fields, collectively denoted by w and F , the
external fields, collectively denoted by KF , the coupling constant g, the gauge fixing
parameter a X and the Dimensional Regularization scale m , which is introduced along the
Ž n. Ž
lines laid out in Žv. of Subsection 2.2. We shall assume that finite countertems, Sfct see
Eq. Ž35., has been chosen so that the BRS equation SSŽ Gren . s 0, the gauge-fixing
condition BŽ Gren . s 0 ŽEq. Ž46.., and the ghost equation G Gren s 0 ŽEq. Ž47.. hold up
to order " n.
The renormalization group equation of the theory gives the expansion of the
functional mEGrenrEm in terms of a certain basis of quantum insertions of ultraviolet
dimension 4 and ghost number 0 w8,32x. The coefficients of this expansion, which are
formal power series in ", are the beta functions and anomalous dimensions of the
theory. Since Gren satisfies, up to order " n , the equations mentioned in the previous
paragraph, the elements of basis of insertions we are seeking is not only constrained by
power-counting and ghost number. Indeed, the following three equations hold up to
order " n :
EGren d EGren EGren
SSG ren m s 0, m s 0, Gm s 0. Ž 70 .
Em dB Em Em
Notice that the operator m EmE commutes with the functional operators B, G and that
E EGren
m Ž SS Ž Gren . . s SSG m .
Em ren
Em
SSG ren stands for the linearized BRS operator,

dGren d dGren d dGren d
SSG ren s d 4 x Tr
H ½ dr m
d Am
q Tr
d Am dr m q Tr
dz dv
dGren d dGren dGren d
qTr q Tr B q
dv dz dv d L dc
dGren d dGren d dGren d
q q X q
dc d L d R dc dc X d R
dGren d dGren d d dGren dGren d
q
d L dc
q
dc d L
q
d R dc X q
dc X d R 5 .
The functionals with ultraviolet dimension 4 and ghost number 0 which satisfy up to
order " n the set of equations in Eq. Ž70. form a linear space. Let us construct a basis of
this space. In the classical approximation Ži.e. at order " 0 ., a basis of this space is given
w32x by the b ' SSS -invariant terms L g , LcR , LcL X , LcL , LcR X , L A and Lv , defined as
cl
follows:
E Scl
Lg s g s y2 Ž Scl A A q Scl A A A q Scl A A A . ,
Eg
LRc s NcR Scl s 2 SclR cc , LLc X s NcLX Scl s 2 SclL c Xc X ,
LLc s NcL Scl s 2 SclL cc q 2 Scl cc A , LRc X s NcRX Scl s 2 SclR c Xc X q 2 Scl c Xc X A ,
L A s NA Scl s 2 Scl A A q 3Scl A A A q 4Scl A A A A q Scl cc A q Scl c Xc X A y Scl rv

y Scl v v ,
Lv s Nv Scl s Scl rv q Scl v v q Scl rv A q Scl zv v q Scl L sc q Scl R sc X q Scl L sc
q Scl R sc X , Ž 71 .
where the Scl . . . terms are the corresponding B-independent terms of the classical action
Scl given in Eq. Ž27.. The symmetric differential operators w32x NcR , NcL , NcRX , NcLX ,
NA and Nv in Eq. Ž71. are given by the following identities:
NcL ' NcL q NcL y NL y NL , NcR ' NcR q NcR ,
NcLX ' NcLX q NcLX , NcRX ' NcRX q NcRX y NR y NR ,

E
NA ' Tr Ž NA y Nr y NB y N v . q 2 a ,
Ea
Nv ' Tr Ž Nv y Nz . . Ž 72 .
The field-counting operators NcL ŽR ., NcLX ŽR ., NL , N L , NR , NR , NA , Nv , N v and Nz are
the operators in Eq. Ž38. for d s 4.
´ D. Sanchez-Ruizr
We shall work out now a basis Žup to order " n, of course. of the space of joint
solutions of Eqs. Ž70. with the help of the statement made in the second paragraph of
Subsection 2.4 Žsee Refs. w32x for further details.. Up to order " n, a quantum extension
of the classical basis in Eq. Ž71. is furnished w32x by the action on Gren of the symmetric
differential operators defined above along with symmetric differential operator g Eg .
Indeed, the functionals
EGren
g , NcR Gren , NcLX Gren , NcL Gren , NcRX Gren , NA Gren , Nv Gren ,
Eg
which are to be understood as formal series expansions in ", match the corresponding
functionals in Eq. Ž71. at order " 0 and the symmetric differential operators g Eg , NcR ,
NcL , NcRX , NcLX , NA and Nv are compatible Žthis is why they are symmetric., up to
order " n, with the BRS equation, the gauge-fixing condition and the ghost equation, in
the sense we spell out next.
Let F be the linear space of functionals with ultraviolet dimension 4 and ghost
number 0 which satisfy Žup to order " n . the BRS equation, the gauge-fixing condition
and the ghost equation: SSŽ G . s 0, BŽ G . s 0, G Ž G . s 0, ;G g F. An operator D
acting on F is said to be compatible with the BRS equation, the gauge-fixing condition
and the ghost equation if, by definition, the following set of equations hold:
d
SSG Ž DG . s 0, Ž DG . s 0, G Ž DG . s 0.
dB
In summary, up to order " n, any joint solution of Eqs. Ž70. is a linear combination
with "-dependent coefficients of the following functionals:
EGren
g , NcR Gren , NcLX Gren , NcL Gren , NcRX Gren , NA Gren , Nv Gren . Ž 73 .
Eg
In particular, if m is the arbitrary ‘‘scale’’ introduced by hand for each loop
integration in minimal Dimensional Renormalization Žsee Žv. in Subsection 2.2., then,
mEGrenrEm has an expansion in the quantum basis given above:
E E
m qb g y gcLNcL y gcRX NcRX y gcRNcR y gcLX NcLX y gA NA y gv Nv Gren
Em Eg
s 0. Ž 74 .
This equation holds up to order " n and it is the renormalization group equation of our
theory. Notice that the expansion mEGrenrEm in terms of the basis in Eq. Ž73. is only
possible if the renormalized functional satisfies order by order the BRS identity Žhence,
there is anomaly cancellation, in particular. and the gauge-fixing condition. It is
important to notice that the regularized 1PI functional does not generally satisfy an
equation such as Eq. Ž74..
Let us notice that the coefficients of the expansions of m EGrenrEm in the quantum
basis of insertions in Eq. Ž73. are the beta functions and anomalous dimensions of the
theory. These coefficients are to be understood as formal expansions in ". The first
non-trivial contribution to these expansions is always of order ", since the lowest order
contribution to Gren is the classical action, Scl , and m E SclrEm s 0.
We next proceed to the computation of the beta function and anomalous dimensions
of our theory. This is done w7x by expressing first m EGrenrEm , g EGrenrE g and Nf Gren ,
where Nf denotes any of the differential operators in Eq. Ž72., as insertions in Gren of
linearly independent local normal products of the fields and their derivatives; then, these
insertions are substituted into the left-hand side of the renormalization group equation
ŽEq. Ž74.. and, finally, one solves for b , gcL , gcRX , gcR , gcLX , gA and gv , the set of
equations involving only numbers which results from the linear independence of the
aforementioned local normal products.
That g EGrenrE g and Nf Gren , where Nf denotes any of the differential operators in
Eq. Ž72., can be expressed as insertions in Gren of normal products is a mere
consequence of the Quantum Action Principle: QAP1 and QAP3, given by Eqs. Ž14. and
Ž16., lead to
Žn.
sN P Gren Ž 75 .
Eg Eg
and
Nf Gren s N Ž Nf Ž S0 q SfctŽ n. . . P Gren , Ž 76 .
respectively. Hence, every element of the symmetric quantum basis in Eq. Ž73. is a local
combination of quantum insertions of integrated standard Ži.e. non-evanescent. monomi-
als with ultraviolet dimension 4 and ghost number 0 Žsee Subsection 2.4.. It should not
Ž n.
be overlooked the fact S0 q Sfct in Eqs. Ž75. and Ž76. is the non-singular contribution to
the Dimensional Regularization action, SDŽ n.R e g Žsee Subsection 3.2., employed to obtain
the minimal dimensionally renormalized 1PI functional up to order " n, Gren , by
Ž n.
following the algorithm spell out in Subsection 2.2. The formal functional S0 q Sfct is
therefore an object in the d-dimensional space-time of Dimensional Regularization. If
n ) 2, it would not do to use some other functional, SX , such that Ž S0 q Sfct Ž n. .
y SX s SˆX is
an integrated evanescent operator whose "-expansion has contributions of order " m ,
with m ( n y 2. Recall that the contribution O Ž "Ž ny1. . in SˆX is not be relevant here
since, being an evanescent object, the lowest order contribution coming from its
insertion in Gren is O Ž " n . and local, so that it does not contribute to the renormalization
group equation at order " n ; a similar argument holds for the term of order " n in SˆX .
Now, by substituting Eqs. Ž75. and Ž76. in Eq. Ž74. one obtains the following
equation
Žn.
m
Em
s yb g N
Eg
P Gren q Ý gf N Ž Nf Ž S0 q SfctŽ n . . . P Gren ,
f
Ž 77 .
where f labels the symmetric differential operators in Eq. Ž72.. Hence, to compute the
beta functions and anomalous dimensions of the theory we only need an independent
way of computing m EGrenrEm in terms of the insertions on the right-hand side of Eq.
Ž77.. We shall set up this independent way of computation next.
Unlike for the subtraction algorithm employed in Ref. w4x, in minimal dimensional
renormalization, one can not use Žsee Ref. w7x. the Quantum Action Principle to obtain
an equation for m EGrenrEm analogous to Eqs. Ž75., Ž76.. Indeed, in minimal dimensional
´ D. Sanchez-Ruizr
Ž n. .
renormalization, m is not a parameter of the action Ž S0 q Sfct , but a parameter which is
introduced by hand in each loop-momentum integration Žas a factor m4y d . before
computing the renormalized value of the corresponding Feynman diagram.
The problem of expressing m EGrenrEm as an insertion in Gren of a linear combination
normal product operators was also solved by Bonneau w7x in the framework of rigorous
minimal dimensional renormalization. He obtain a formula similar to Eq. Ž25., i.e. a
Zimmerman-like identity, for a scalar case with no added finite counterterms in the
action. Here we need the generalization to the presence of several type of fields,
including external fields, the presence of hatted objects and also the presence of finite
counterterms up to order " n, in the action Ž S0 q Sfct
Ž n. .
. The needed generalization reads
r
EGren 4 vŽJ. Ž yi . Er
m
Em
s Ý Ý Ý
ns0 j1 , . . . , j n 4 Nl)0
Nl Ý
rs0
Ý
i1 . . . i r 4
1(i j(ny1
½ r.s.p.
r! E pim1 1 . . . E pimr r
= Ž yi" . ² f˜ j1Ž p 1 . . . . f˜ j nŽ pn s y Ý pi . :1PI

Ks0
, Ž Nl .
p is0 5
1 1
=
n!
Hd 4
x N f jn Ł
ksny1
½ž Ł
a ri ask 4
/
Ema f jk 5 Ž x . P Gren
r
v Ž J ;F . Ž yi . Er
qÝ
F
Ý
rs0
Ý
i1 . . . i r 4
1(i j(n
½ r.s.p.
Ž r . ! E pim1 1 . . . E pimr r
=² f˜ j1Ž p 1 . . . . f˜ j nŽ pn . N w sF x Ž pnq1 s y Ý pi . :Ks0

1PI ,Ž N l .
p is0 5
1 1
=
n!
H d 4 x Ž KF Ž x . N Ł
ksn
½ž Ł
a ri ask 4
/
Ema f jk 5 Ž x . P Gren , Ž 78 .
where J s j1 , . . . , jn 4 , and, v Ž J . and v Ž J;F . stand, respectively, for the overall

ultraviolet degree of divergence of ² f˜ j1Ž p 1 . P P f˜ j nŽ pn .:1PIKs 0 and ² f ˜ j1Ž p1 . P P
˜ w xŽ 1PI
f j n pn N sF pnq1 Ks 0 . The bar upon the 1PI Green functions means that all
Ž . .:
subdivergences has been subtracted from the Feynman diagrams which contribute to
them. This is as in Eq. Ž48.. The novel feature here is the presence of the superscript Nl
on the upper right of the Green function, which indicates that only Feynman diagrams
with precisely Nl loops are to be considered. Notice that one then sums over all number
loops upon multiplication by Nl of the Nl-contribution.
The r.s.p. of the subtracted Žsubdivergences only. 1PI Feynman diagrams contributing
to the right-hand side of Eq. Ž78. is a local polynomial in the d-dimensional space of
Dimensional Regularization of external momenta associated with the fields of the
corresponding 1PI function; so, in general, it will contain hatted and barred objects
Žmetrics, momenta, gamma matrices.. Therefore, the formula will generate automatically
also evanescent Žalso called anomalous. insertions. The meaning of = in last formula is
the same as in Eq. Ž25.. Hence, the expansion in Eq. Ž78. can be recast into the form
EGren
m s Ý ri N W i P Gren q Ý rˆj N Wˆ j P Gren , Ž 79 .
Em i j
where W i and Wˆ j denote, respectively, the barred Žnon-evanescent. and hatted

Ževanescent. elements of a basis of integrated monomials with ghost number 0 and
ultraviolet dimension 4. The symbols ri and rˆj in Eq. Ž79. are coefficients defined as
formal expansions in " Žstarting at ", since their computation involves that of the r.s.p.
of the divergent part of the appropriate 1PI function.. The order " m , 0 - m ( n,
contribution to these coefficients is obtained as follows:
– Take a divergent, by power-counting, 1PI function with some number of fields

of definite type Žmaybe including some external field coupled to a BRS
variation..
– Compute the residue of the simple pole Žr.s.p. . of every O Ž " m .-graph
Ž n. .
constructed with the Feynman rules derived from the action Ž S0 q Sfct and
contributing to that 1PI function, with all their subdivergences subtracted.
– Multiply the last quantity by the number of the loops of the graph
– Sum over all graphs contributing to the 1PI function
– Being the sum a local expression, their different terms can be interpreted as
Feynman rules of tree-level integrated insertions of Lorentz invariant Žnormal
or anomalous. composite operators, formed by just the fields of the original 1PI
function and some different combination of metrics, derivatives and gamma
matrices.
– Do the same steps for every divergent 1PI function. Then you will obtain the
expansion Eq. Ž79. with the O Ž " m . coefficients.
Explicit expressions for each ri and rˆj similar to that ones for the expansion of the
anomalous breaking can be given. For example, if
W 2a b s I Ama A bm ,
H Wˆ 21a b s I Am̂a A bm̂ ,
H
Wˆ 22a b s I
ˆ Ama A bm ,
H Wˆ 23a b s I
H ˆ Am̂a A bm̂ ,
then
2
y Ž yi .
r 2a b s Nl = coef. in r.s.p. Ž yi" . ² Ama Ž p 1 . Anb Ž p 2 ' yp1 . :1PI
2
of p 1 2 g mn ,
2
y Ž yi .
ˆ ab
r 21 s Nl = coef.in r.s.p. Ž yi" . ² Ama Ž p 1 . Anb Ž p 2 ' yp1 . :1PI
2
of p 1 2 gˆ mn ,
´ D. Sanchez-Ruizr
2
y Ž yi .
ˆ ab
r 22 s Nl = coef. in r.s.p. Ž yi" . ² Ama Ž p 1 . Anb Ž p 2 ' yp1 . :1PI
2
of pˆ 1 2 g mn ,
2
y Ž yi .
ab
rˆ23 s Nl = coef. in r.s.p. Ž yi" . ² Ama Ž p 1 . Anb Ž p 2 ' yp1 . :1PI
2
of pˆ 1 2 gˆ mn ;
and so on.
Next, by using the Bonneau identities technique we shall express Žsee Subsection 2.4.
every evanescent insertion Nw Wˆ j x P Gren on the right-hand side of Eq. Ž79. as a linear
combination of standard Ži.e. non-evanescent. insertions Nw W i x P Gren :
N Wˆ j P Gren s Ý c ji N W i P Gren . Ž 80 .
i
The coefficients c ji are formal expansions in powers of ", having no order " 0
contribution. The use of the previous equation turns Eq. Ž79. into the following
equation:
EGren
m s Ý ri N W i P Gren , Ž 81 .
Em i
where ri s ri q Ý j rˆj c ji , so that ri s ri at order ". Notice that, because the evanescent
insertions are to be expanded according to the Bonneau identities in Eq. Ž25., the
computation of ri up to order " n only involves the contributions to rˆj up to order
"Ž ny1., whereas the contributions to ri are needed up to order " n.
Now, since g EEg w S0 q Sfct Ž n. x
, Nf w S0 q SfctŽ n. x
, g EEg wŽ S0 y S0 . q Ž Sfct
Ž n. Ž n. .x
y Sfct and
n. n.
Nf wŽ S0 y S0 . q Ž Sfct y Sfct .x are linear combinations of monomials of ultraviolet di-
Ž Ž
mension 4 and ghost number 0, it is advisable to choose the basis of monomials

Wˆ j 4 , which occurs in Eq. Ž79., in such a way that it contains the former set of
W i ,W
monomials. In other words, the basis W i ,W Wˆ j 4 so chosen makes it possible to obtain the
coefficients on the right-hand side of the following equations:
E Žn. Žn.
g S0 q Sfct s Ý wg i W i , Nf S0 q Sfct s Ý wf i W i ,
Eg i i
E
g Ž S0 y S0 . q Ž SfctŽ n . y SfctŽ n . . s Ý wˆ g j Wˆ j ,
Eg j
Nf Ž S0 y S0 . q Ž SfctŽ n . y SfctŽ n . . s Ý wˆf j Wˆ j ,

j
by reading them from the left-hand side of the corresponding equation. The coefficients
wg i , wˆ g i , wf i and wˆf i have expansions in powers of "; these expansions, though, have
now contributions of order " 0 .
It is a trivial exercise to check that Eq. Ž77. can be recast into the form
EGren
m
Em ž /
s Ý yb wg i q Ý gf wf i N W i P Gren
i f
q Ý yb wˆ g i q Ý gf wˆf i N Wˆ j P Gren .
ž /
j f
The substitution of Eq. Ž80. into the last equation gives the following result:
EGren
m
Em i
ž f
/
s Ý yb wg i q Ý gf wf i N W i P Gren , Ž 82 .
where
wg i s wg i q Ý wˆ g j c ji and wf i s wf i q Ý wˆf j c ji ,
j j
where the index j labels the elements of the ‘‘evanescent’’ basis Wˆ j 4 .

Finally, by comparing Eqs. Ž81. and Ž82., one obtains an overdeterminate system of
linear equations whose unknowns are the coefficients b and gf :
ri s yb wg i q Ý gf wf i , Ž 83 .
f
where the index i runs over the elements of the standard Ži.e. with no evanescent
operator. basis W i 4 .
Several remarks about Eq. Ž78. are now in order.
Ž n.
Remark 1. This equation is only correct if the finite counterterms in Sfct are
independent of m. This poses no problem, because the finite counterterms we need to
add to the action to restore the BRS identities do not depend on m , thanks to the fact
that in the Bonneau identities we have to compute only r.s.p.s of 1PI functions with all
subdivergences minimally subtracted and these r.s.p. are always mass and m indepen-
dent.
Remark 2. The counterpart of Eq. Ž78. in Ref. w7x replaces the factor Nl , which counts
the number of loops, with the operator " ErE ". This is because, if no terms of order " m ,
m ) 0 are present in the action, then the powers in " of a graph counts its number of
loops. But now we have added higher order terms in " to the action and so we cannot
use " as a loop meter. This can be superseded by introducing a new parameter l, which
divides the whole action, including the finite counterterms. Then l Nly1 would be the
factor attached to a Nl-loop Feynman diagram. Note that only in the limit l ™ 1 the
finite counterterms restore the BRS identities. Therefore, the factor Nl in Eq. Ž78. is
replaced by lim l ™ 1 1 q l EEl 4 . Thanks to the action principles, the action of the operator
lim l ™ 1 1 q l EEl 4 on a 1PI Green function computed for arbitrary l can be replaced with
Ž n. .x
the insertion of the integrated operator 1 y Nw iŽ S0 q Sfct into that 1PI function at
l s 1. It should be stressed that this is true only if there is no explicit m-dependence in
the finite counterterms.
´ D. Sanchez-Ruizr
Let us move on and compute the beta function and anomalous dimensions of our
theory at order ". At "-order, Eq. Ž83. reads
riŽ1. s yb Ž1. wgŽ0.i q Ý gfŽ1. wfŽ0.i , Ž 84 .

f
for ri s riŽ1. " q O Ž " 2 ., b s b Ž1. " q O Ž " 2 ., wg i s wgŽ0.i q O Ž " ., gfŽ1. s gfŽ1. " q O Ž " 2 .
and wf i s wfŽ0.i q O Ž " .. To obtain riŽ1.4 , we just need a suitable basis of integrated
barred monomials W i 4 with ultraviolet dimension 4 and ghost number 0: the evanescent
monomials Wˆ j 4 are not needed at this order in ". The basis W i 4 is furnished by the
following list of monomials with ghost number 0, dimension 4, free gauge indices and
fully contracted Lorentz indices:
Ža. Monomials with only A’s:
W 1a b ' HŽ E m En A a m . A bn , W 2a b ' HžI A / A a

m
bm
,
W 3a b c ' HŽ E m Ana . A bmAcn , W4a b c ' HŽ E m A a m . Anb Acn ,
W 5a b c ' ´mnra Ž E aA bm . A anAc r ,

H
W6a1 a 2 a 3 a 4 ' Ama1 A a 2 mAna 3 A a 4 n s W6a1 a 2 4 a 3 a 4 4 ,
H
W 7a1 a 2 a 3 a 4 ' ´mnrd A a1 mA a 2 nA a 3 rA a 4 d .
H
Žb. Monomials involving fermions and no external fields:
i m
l
W 8i j ' H2c g i P L pr m c j , W 9i j a ' c i g m P L c j Ama ,
H
i l
W 10i j ' H2c i
X
g m PR pr m c jX , W 11i j a ' c iX g m PR c jX Ama ,
H
i m l
W 12i j ' H2c g i PR pr m c j , W 13i j a ' c i g m PR c j Ama ,
H
i l
W 14i j ' H2c i
X
g m PL pr m c jX , W 15i j a ' c iX g m PL c jX Ama .
H
Žc. Monomials involving ghost and no external fields:
W 16a b ' y v a I v b s
H HŽ E mv
a
. Ž E mv b . ,
W 17a b c ' v a E m Ž v cAmb . ,
H W 18a b c ' v a Ž E mA bm . v c ,
H
W 19a b c d ' v av bAc mAmd ,
H W 20a b c d ' v a v bv cv d .
H
Žd. Monomials involving external fields:
W 21i j a ' v a L i PL c j ,
H W 22i j a ' c i PR v a L j ,
H
W 23i j a ' v a R i PR c jX ,
H W 24i j a ' c iX PL v a R j ,
H
W 25i j a ' v a L i PR c j ,
H W 26i j a ' c i PL v a L j ,
H
W 27i j a ' v a R i PL c jX ,
H W 28i j a ' c iX PR v a R j ,
H
W 29a b ' rma E mv b ,
H W 30a b c ' rma v bAc m ,
H
W 31a b c ' z av bv c .
H
Next, Eq. Ž78. leads to
EGren 31
m
Em
s Ý Ý riŽ1., A i W i A i qO Ž " 2 . , Ž 85 .
is1 A i
where A i denotes the set of gauge indices, W i A i denotes henceforth the counterpart of
the operator W i A i in four-dimensional space-time and the coefficients riŽ1., A i are defined
as follows:
2
Ž yi .
r 1Ž1. a b s coef. in r.s.p. Ž yi . ² Ama Ž p 1 . Anb Ž p 2 ' yp1 . :1PIŽ1. of p 1m p 1n
2
1 2 5 1yaX
s 2
d a b TLq R y CA ž q / ,
Ž 4p . 3 3 2
2
Ž yi .
r 2Ž1. a b s coef. in r.s.p. Ž yi . ² Ama Ž p 1 . Anb Ž p 2 ' yp1 . :1PIŽ1. of p 1 2 g mn
2
1 ab
2 5 1yaX
s
Ž 4p .
2
d TLq R
3
y CA ž 3
q
2 / ,
Ž yi .
r 3Ž1. a b c s coef. in r.s.p. Ž yi . ² Ama Ž p 1 . Anb Ž p 2 . Arc Ž p 3 ' yp1 y p 2 . :1PIŽ1.
3
of p 1n g m r
Ž yi .
q coef. in r.s.p. Ž yi . ² Ama Ž p 1 . Anc Ž p 2 . Arb Ž p 3 ' yp1 y p 2 . :1PIŽ1.
3
of p 1r g mn
1 2 2 2 3Ž 1 y a X .
s 2
Ž 4p . 3
c abc
3
T Lq R y CA
ž 3
q
4 / ,
´ D. Sanchez-Ruizr
Ž yi .
r4Ž1. a b c s coef. in r.s.p. Ž yi . ² Ama Ž p 1 . Anb Ž p 2 . Arc Ž p 3 ' yp1 y p 2 . :1PIŽ1.
3
of p 1m g nr
1 2 2 2 3Ž 1 y a X .
s 2
Ž 4p . 3
c abc
3
T Lq R y CA
ž 3
q
4 / ,
Ž yi .
r5Ž1. a b c s coef. in r.s.p. Ž yi .
3
=² Ama Ž p 1 . Anb Ž p 2 . Arc Ž p 3 ' yp1 y p 2 . :1PIŽ1. of
= p 1a ´mnras0,
r6Ž1. a1 a 2 4 a 3 a 4 4 s r6Ž1. a 3 a 4 4 a1 a 2 4
1
s coef. in r.s.p. Ž yi . ² Ama11 Ama 22 Ama 33 Ama 44 :1PIŽ1. of
8
= g m1 m 2 g m 3 m 4
1 1 4
s 2
T Lq R q 2 CA Ž 13 y Ž 1 y a X . .
Ž 4p . 8 3
= Ž c e a1 a 3 c e a 4 a 2 y c e a1 a 4 c e a 2 a 3 . ,
r 7Ž1. a1 a 2 a 3 a 4 s r 7Ž1. a1 a 2 a 3 a 4 4

1
s coef. in r.s.p. Ž yi . ² Ama11 Ama 22 Ama 33 Ama 44 :1PIŽ1.
4!
of ´m1m 2m 2m 4s0,
i
r 8Ž1. i j s i coef. in r.s.p. ² cb j Ž yp . ca i Ž p . :1PIŽ1. of Ž pu PL . ab s 2
2 g 2a X C L ,
Ž 4p .
r 9Ž1. i j a s i coef. in r.s.p. Ž yi . ² cb j ca i A am :1PIŽ1. of Ž g m PL . ab
1 1yaX
s 2
2 g 2
C L 1 y ž CA q a X C L Ž TLa . i j ,
/
Ž 4p . 4
Ž1. i j
r 10 s i coef. in r.s.p. ² cbX j Ž yp . caX i Ž p . :1PIŽ1. of Ž pu PR . ab
i
s 2
2 g 2a X C R ,
Ž 4p .
Ž1. i j a
r 11 s i coef. in r.s.p. Ž yi . ² cbX j caX i A am :1PIŽ1. of Ž g m PR . ab
1 1yaX
s 2
2
2 g CR 1 y ž CA q a X C R Ž TRa . i j ,
/
Ž 4p . 4
Ž1. i j
r 12 s i coef. in r.s.p. ² cb j Ž yp . ca i Ž p . :1PIŽ1. of Ž pu PR . ab s 0,
Ž1. i j a
r 13 s i coef. in r.s.p. Ž yi . ² cb j ca i A am :1PIŽ1. of Ž g m PR . ab s 0,
Ž1. i j
r 14 s i coef. in r.s.p. ² cbX j Ž yp . caX i Ž p . :1PIŽ1. of Ž pu PL . ab s 0,
Ž1. i j a
r 15 s i coef. in r.s.p. Ž yi . ² cbX j caX i A am :1PIŽ1. of Ž g m PL . ab s 0,
Ž1. a b
r 16 s i coef. in r.s.p. Ž yi . ² v b Ž yp . v a Ž yp . :1PIŽ1. of p 2
1 1yaX
sy 2
CA g 2 1 q
ž / ,
Ž 4p . 2
Ž1. a b c
r 17 s Ž yi . coef. in r.s.p. Ž yi . ² v c Ž yp1 y p 2 . v a Ž p 1 . Amb Ž p 2 . :1PIŽ1. of p 1m
1
s 2
CA g 2a X c a b c ,
Ž 4p .
Ž1. a b c
r 18 s Ž yi . coef. in r.s.p. Ž yi . ² v c Ž yp1 y p 2 . v a Ž p 1 . Amb Ž p 2 . :1PIŽ1.
of p 2ms0,
Ž1. a b c d
r 19 s 12 coef. in r.s.p. Ž yi . ² v b v aAmb And :1PIŽ1. of g mn s 0,
Ž1. a b c d
r 20 s 14 r.s.p. Ž -i . ² vd vc vb va :1PI Ž1. s 0,
Ž1. i j a
r 21 s coef. in r.s.p. ² c j b v a ; sc i a:1PIŽ1. of Ž P L . ab
i
s 2
g 2a X CA Ž T La . i j ,
Ž 4p .
Ž1. i j a
r 22 s coef. in r.s.p. ² v ac i a ; sc j b :1PIŽ1. of Ž PR . ab
i
s 2
g 2a X CA Ž T La . i j ,
Ž 4p .
Ž1. i j a
r 23 s coef. in r.s.p. ² c jXb v a ; sc iXa:1PIŽ1. of Ž PR . ab
i
s 2
g 2a X CA Ž TRa . i j ,
Ž 4p .
Ž1. i j a
r 24 s coef. in r.s.p. ² v ac iXa ; sc jXb :1PIŽ1. of Ž P L . ab
i
s 2
g 2a X CA Ž TRa . i j ,
Ž 4p .
Ž1. i j a
r 25 s coef. in r.s.p. ² c j b v a ; sc i a:1PIŽ1. of Ž PR . ab s 0,
Ž1. i j a
r 26 s coef. in r.s.p. ² v ac i a ; sc j b :1PIŽ1. of Ž P L . ab s 0,
Ž1. i j a
r 27 s coef. in r.s.p. ² c jXb v a ; sc iXa:1PIŽ1. of Ž P L . ab s 0,
´ D. Sanchez-Ruizr
Ž1. i j a
r 28 s coef. in r.s.p. ² v ac iXa ; sc jXb :1PIŽ1. of Ž PR . ab s 0,
Ž1. a b
r 29 s Ž yi . coef. in r.s.p. ² v b Ž p 1 . ; sAma Ž p 2 s yp1 . :1PIŽ1. of p 1m
1 ab 2
1yaX
sy
Ž 4p .
2
d CA g ž 1q
2 / ,
Ž1. a b c
r 30 s Ž yi . coef. in r.s.p. ² v bAmc ; sAna :1PIŽ1. of gmn
1
sy 2
c a b c CA g 2 ,
Ž 4p .
1 1 1
Ž1. a b c
r 31 s r.s.p. ² vc vb ; sva :1PI Ž1. s y c a b c CA g 2 . Ž 86 .
2 2 Ž 4p . 2
To remove much of the redundancy in the linear system in Eq. Ž84. we shall rather
express mEGrenrEm in terms of the functionals in Eq. Ž71.. By using the following
equations:
d ab c ab c
Scl A A s
2 g2
Ž W 1a b y W 2a b . , Scl A A A s y
g2
W 3a b c ,
1
Scl A A A s y Ž c ac e c b d e q c b c e c ad e . W6a b4c d4 ,
8g2
SclL cc s d i j W 8i j , Scl cc A s Ž T La . i j W 9i j ,
SclR c Xc X s d i j W 10i j , Scl c Xc X A s Ž TRa . i j W 11i j ,
Scl v v s d a b W 16a b , Scl v v A s yc a b c W 17a b c ,

Scl L sc s i Ž TLa . i j W 21i j a , Scl L sc s i Ž TLa . i j W 22i j a ,
Scl R sc X s i Ž TRa . i j W 23i j a , Scl R sc X s i Ž TLaŽR . . i j W 24i j a ,
Scl rv s d a b W 29a b , Scl rv A s yc a b c W 30a b c ,
Scl zv v s y 12 c a b c W 31a b c ,
one easily obtains the contribution to mEGrenrEm at first-order in ":
EGren 1 8 TLqR
m s 2
g2 y CA Ž 103 q Ž 1 y a X . . Scl A A
Em Ž 4p . 3 2
1 8 T Lq R
q 2
g2 y CA Ž 43 q 32 Ž 1 y a X . . Scl A A A
Ž 4p . 3 2
1 8 T Lq R
q 2
g2 y CA Ž y 23 q 2 Ž 1 y a X . . Scl A A A A
Ž 4p . 3 2
1 1
q 2
g 2 C L 2 a X SclL cc q 2
g 2 C R 2 a X SclR c Xc X
Ž 4p . Ž 4p .
1 Ž1yaX .
CA q 2 a X C L Scl cc A
q
Ž 4p .
2
g2
ž 2y
2 /
1 Ž1yaX .
CA q 2 a X C R Scl c Xc X A
q
Ž 4p .
2
g 2
ž 2y
2 /
1 Ž1yaX . 1
g 2 a X CA Scl v v A
y
Ž 4p .
2
g 2
ž 1q
2 / CA Scl v v q
Ž 4p .
2
1
q 2
g 2 a X CA Scl L sc q Scl R sc X q Scl L sc q Scl R sc X
Ž 4p .
1 Ž1yaX . 1
g 2 a X CA Scl rv A
y
Ž 4p .
2
g2 1q
ž 2 / CA Scl rv q
Ž 4p .
2
1
q 2
g 2 a X CA Scl zv v qO Ž " 2 . , Ž 87 .
Ž 4p .
where TLq R means sum over all left and right representations Že.g. the number of
‘‘flavours’’. in the theory.
Now, it is clear that Eq. Ž77. reads
EGren E Scl
m s yb Ž1. g q Ý gfŽ1.Nf Scl qO Ž " 2 .
Em Eg f
s yb Ž1. L g q Ý gfŽ1. Lf qO Ž " 2 . , Ž 88 .

f
which in turn can be expressed in terms of the functionals Scl . . . given in Eq. Ž71..
Finally, the fact that the left-hand sides of Eqs. Ž87. and Ž88. should match gives rise
to the following system of equations:
1 8 T Lq R
From Scl A A ™ 2 bŽ1. q 2 gA Ž1. s 2
g2 y CA Ž 103 q Ž 1 y a X . . ,
Ž 4p . 3 2
1 8 TLq R
From Scl A A A ™ 2 bŽ1. q 3 gA Ž1. s 2
g2 y CA Ž 43 q 32 Ž 1 y a X . . ,
Ž 4p . 3 2
1 8 T Lq R
From Scl A A A A ™ 2 bŽ1. q 4 gAŽ1. s 2
g2 y CA Ž y 23 q 2 Ž 1 y a X . . ,
Ž 4p . 3 2
1
From SclL cc ™ 2 gcLŽ1. s 2
g 2 CL 2 a X ,
Ž 4p .
1
From SclR c Xc X ™ 2 gcRX Ž1. s 2
g 2 CR 2 a X ,
Ž 4p .
´ D. Sanchez-Ruizr
1 Ž1yaX .
CA q 2 a X C L ,
From SclL cc A ™ 2 gcLŽ1. q gAŽ1. s
Ž 4p .
2
g2
ž 2y
2 /
1 Ž1yaX .
CA q 2 a X C R ,
From SclR c Xc X A ™ 2 gcRX Ž1. q gAŽ1. s
Ž 4p .
2
g2
ž 2y
2 /
1 Ž1yaX .
From Scl v v ™ ygA q gv s y Ž1. Ž1.
Ž 4p .
2
g 2
ž 1q
2 / CA ,
1
From Scl v v A ™ gv Ž1. s 2
g 2 a X CA ,
Ž 4p .
1
From Scl L sc ™ gv Ž1. s 2
g 2 a X CA ,
Ž 4p .
1
From Scl R sc X ™ gv Ž1. s 2
g 2 a X CA ,
Ž 4p .
1
From Scl L sc ™ gv Ž1. s 2
g 2 a X CA ,
Ž 4p .
1
From Scl R sc X ™ gv Ž1. s 2
g 2 a X CA ,
Ž 4p .
1 Ž1yaX .
From Scl rv ™ ygAŽ1. q gv Ž1. s y
Ž 4p .
2
g2 1q
ž 2 / CA ,
1
From Scl rv A ™ gv Ž1. s 2
g 2 a X CA ,
Ž 4p .
1
From Scl zv v ™ gv Ž1. s 2
g 2 a X CA , Ž 89 .
Ž 4p .
which is a simplified version of Eq. Ž84.. This simplification takes place since the
Ž n.
non-symmetrical finite counterterms, Sfct , needed to restore the BRS symmetry and the
evanescent monomials, Wˆ j , only begin to contribute to the renormalization group
equation at order " 2 . At order " n, n ) 1, it is the linear system in Eq. Ž83. which is to
be dealt with. The system in Eq. Ž89. is compatible and overdeterminate and its solution
reads
1 4 TLq R 11
bŽ1. s g2 ž y C , /
Ž 4p .
2
3 2 3 A
1 1yaX
gAŽ1. s 2
g2 2y ž / CA ,
Ž 4p . 2
1
gvŽ1. s 2
g 2 a X CA ,
Ž 4p .
1 1
gcLŽ1. s 2
g 2 a X CL , gcRX Ž1. s 2
g 2 a X CR . Ž 90 .
Ž 4p . Ž 4p .
The reader should notice that to actually compute the beta function and anomalous
dimensions of the theory one does not need to evaluate the thirty one coefficients in Eq.
Ž85. Žcomputed in Eq. Ž86.. but just a few of them, if appropriately chosen. Indeed, the
computation of, say, r 1Ž1. a b , r 8Ž1. i j, r 10
Ž1. i j Ž1. a b
, r 16 Ž1. ab c
and r 17 would do the job. However,
a thorough check of the formalism demands the computation of the thirty one coeffi-
cients in question. And this we have done.
We have finished the computation of the renormalization group equation at order "
for the theory with classical action in Eq. Ž27.. Our expansions were expansions in ", as
suits the cohomology problems which arise in connection with the BRS symmetry,
rather that in the coupling constant g. Our parametrizations of the wave functions were
such that g only occurs, in the classical action, in the Yang–Mills term, so that g Eg Scl
is the only element Žmodulo multiplication by a constant. of the local cohomology of the
BRS operator b Žsee Eq. Ž29.. over the space of local integrated functionals of ghost
number 0 and ultraviolet dimension 4 ŽLorentz invariance and rigid gauge symmetry are
also assumed.. The other contributions to the renormalization group equation are b-exact
and have to do with the anomalous dimensions of the theory.
It is often the case that one chooses a parametrization of the wave function such that,
at the tree-level, the fermionic and the three-boson vertices carry the coupling factor g,
and the four-boson vertex is proportional to g 2 . The renormalization group equation for
this parametrization of the wave function is easily retrieved from our results. Let us
denote by Gren A , v w , v w , B w , r w , z w , c , c , c X , c X , a w , g x the 1PI functional for this
ww w
new parametrization. Then,

w
Gren Aw , v w , v w , B w , r w , z w , c , c , c X , c X , a w , g
' Gren gAw , g v w , gy1 v w gy1 , gy1 B w , gy1r w , gy1z w , c , c , c X , c X , g 2a w , g
' Gren A, v , v , B, r , z , c , c , c X , c X , a , g ,
where Gren w A, v , v , B, r , z , c , c , c , c X , a , g x is the 1PI functional for the parametrization
of Eq. Ž27.. Now, since
w
EGren EGren 1 w w w w w
g s y NAw Gren y Nv w Gren q Nv w Gren q NB w Gren q Nr w Gren
Eg Eg g
w
w
EGren
q Nz w Gren y2 a w
,
Ea w
the renormalization group Eq. Ž74. becomes Žin terms of the usual combination
a S ' g 2rŽ4p ..
E E E
m qb S a S q da w a w y gcLNcL y gcRX NcRX y gAw NAw y gv w Nv w
Em Ea S Ea w
w
yg v w Nv w y g B w NB w y gr w Nr w y gz w Nz w Gren s 0,
with
bS s 2 b ,
gAw s ygr w s ygv w s yg B w s gA q b ,
´ D. Sanchez-Ruizr
gv w s ygz w s gv q b ,
da w s y2 gAw s y2 Ž gA q b . .
Finally, by using the results in Eq. Ž90., one obtains
a S 2 TLqR 11
b SŽ1. s y C ,
p 3 2 6 A
a S 1 TLqR CA 13 aw
gAw Ž1. s
p 3 2
y
4 6
y
2
, ž /
aS 1 TLqR CA 31 aw
gv w Ž1. s
p 3 2
y
4 ž 6
q
2 / .
4. Chiral Yang–Mills theories: non-simple gauge groups
4.1. The tree-leÕel action
In this section the gauge group will be a compact Lie group which is the direct
product of NSS simple groups and NA Abelian factors. Its Žreal. Lie algebra is a direct
sum: G s G 1 [ . . . [ G N A [ G N A q1 [ . . . [ G N A qNSS of dimension d G s NA q d G N q 1
A
q . . . qd G N q N . The index G, which labels the group factors, will run from 1 to
A SS
NA q NSS , the index A, which labels the Abelian factors, from 1 to NA and the index
SS, which labels the simple factors, from 1 to NSS .
Therefore, there exists a basis for the Lie algebra which can be split and enumerated
as follows:
dG
X a 4 as1 s X 1 , . . . , X N A , X N A q1 , . . . , X N A qd G , . . . 4
1
d G N ASS
N SS
' X 1 , . . . , X N A 4 D j SSs1
ž X aŽSSSS . 4 a S Ss1 /,
and such that the Killing form of the semisimple part is diagonal and positive definite;
so, we choose d ab to lower and rise indices. For any such basis, the structure constants,
given by w X a , X b x s i c a b c X c , are completely antisymmetric, and c a G d e c b G w d e s 0 if
G / G w , c a G d e c b G d e s c a G d G e G c b G d G e G s CAŽG. d a b , with CAŽ A . ' 0 Žclearly, c a A b c s 0..
Let us display the field content of the theory and introduce the Dimensional
Regularization classical action, S0 . We have first the Lie-algebra-valued fields. The pure
Yang–Mills part of the tree-level action in the d-dimensional space of Dimensional
Regularization reads
S YM s d d x y 14 Fmna C a b F b mn ,
H Ž 91 .
where C is a diagonal matrix with
½ gy2 y2
^
y2
`
y2
1 , . . . , g N A , g N A q1 , . . . , g N A q1
_d G N Aq1 times
, . . . , gy2
^
y2
N A qN SS , . . . , g N A qN SS
` _d G N AqN SS times
5
in the diagonal. Hence,
1
S YM s d d xH Ý y 4 g 2 FmnŽG.a G
FaŽG.
G mn
,
G G
ŽG. a G
where Fmn s Em AnŽG. a G y En AmŽG. a G q c a G b G c G AmŽG. b G AnŽG. c G . Obviously, all the remain-
ing Lie algebra-valued fields are indexed in the same way, yielding
1
Sgf s d d x Ba L a b Bb q Ba Ž E mAma . y v a E m=m v a
H2
a G ŽG. a ŽG. m ŽG. a G
s dd x Ý H B GB
a G q Ba G Ž E Am
ŽG.
. y v aŽG.G E m=m v ŽG. aG .
G 2
The free boson propagator now reads
ab
yi k mkn k mkn
ž y1
C
k 2 qie ž g mn
y
k2 / qC L
k2 / ,
i.e. diagonal but no longer a multiple of the identity.

Notice that we have also introduced ghosts to control the Abelian variations. Since
ghosts in Abelian theories do not interact with the rest of fields Ž=m v A s Em v A ., they
can occur in a non-tree-level 1PI diagram only as a part of a vertex insertion ŽBRS
variations and breakings.; so that, by differentiating with respect to v A the Slavnov–
Taylor identities, the usual Abelian Ward identities are easily recovered. But we insist
on keeping the Abelian ghost in order to have an unified notation.
The matter content of the theory will be only non-Majorana fermions. The corre-
sponding fermion fields, c Ž c X ., carrying a left-handed Žright-handed. fully reducible
representation of the gauge group, so that it can be discomposed in a sum of irreducible
representations. From now on, until otherwise stated, we will take both the left-handed
and right-handed fermionic to be irreducible, the general case just involve a sum over all
the irreducible representations.
We shall index the fermion fields carrying the irreducible representation of the gauge
Lie algebra as follows: c i 1 , . . . ,i N SS Ž c X i 1 , . . . ,i N SS ., with i S S running from 1 to d S SŽ d S S X .;
so that the generators of the Lie algebra are given by
NA NA
TLAŽ R . 4 As1 s t LAŽ R . 4 As1 m 1 d ŽX .
1
. . . m 1 d ŽNX . ,
SS
where each t LAŽR . is a real number and

N SS
N A qd dG N
TLa Ž R . 4 asN G
A q1
s1m ½1 d1
X
Ž . m . . . m TLŽ SS . aS S
ŽR .
A
4 aS Ss1 m . . . 1dN qS S X
Ž .
5 ,
SS SSs1
X
where TLŽ SS . aS S
ŽR . g M d S S ŽX . ŽC. are complex matrices
X
of dimension d S S Ž . which furnish an
Ž.
irreducible representation of dimension d S S of the Lie algebra G N A qSS , SS s 1 . . . NSS .
The BRS variations of the fermion fields in the d-dimensional space of Dimensional
Regularization are defined to be
X X
sc Ž . s i v a TLaŽR . PL ŽR . c Ž .
NA
X
s Ý i v A t LAŽR . 1 d1ŽX . m . . . m 1 d ŽNX . PL ŽR . c Ž .
SS
As1
NSS dG N
Aq SS
X
q Ý i Ý v Ž SS . a S S 1 d1ŽX . m . . . m TLŽ SS . aS S
ŽR . m . . . m 1 d ŽNX . PL ŽR . c Ž . ,
SS
SSs1 a S Ss1
and the regularized interaction is introduced in the same way.

´ D. Sanchez-Ruizr
The following results will be useful:
Tr M d ŽX . ŽC . T LŽ SS . a S S Ž SS . b S S
ŽR . TL ŽR . ' TL ŽR . S S d a S S b S S ,
SS
Tr M d ŽX . ŽC . T LŽ SS . aS S
ŽR . s 0,
SS
° NSS
~
ž / Ł d KŽ .
K / SS
X
TL Ž R . S S d a b , if as a SS and bs bSS ,with SSs1, . . . , NSS
Tr TLaŽR .TLbŽR . s NSS
¢ž /
Ł d SŽX . S t La ŽR . t Lb ŽR . , if a( NA and b( NA
SSs 1
0 in the remainder
X
d SŽ .S
2
Ý TLŽ SS . eS S
ŽR . TLŽ SS . eS S
ŽR . ' C LŽ SS .
ŽR . 1 d S S Ž . ,
X
Ž t LAŽR . . ' C LŽ AŽR. . ,
eS S
NA qNSS
Ý TLaŽR . TLaŽR . s Ý ŽR . 1 d 1Ž . m . . . m 1 d ŽN . ' Ý C L ŽR . ,

C LŽG. X X ŽG.
SS
a Gs1 G
NA qNSS
a
Ý TLaŽR . ŽC .y1 b
b T L ŽR . s Ý g G2 C LŽG. g
ŽR . ' C L ŽR . ,
a, b Gs1
NA qNSS
a
Ý TLaŽR . L a b TLbŽR . s Ý a G C LŽG.
ŽR . ' C L ŽR . .
a, b Gs1
4.2. The one-loop singular counterterms
To obtain the one-loop singular counterterms, it is necessary to make in the

non-hatted terms in Eq. Ž37. the following substitutions:
g 2 CA S0 X . . . X Y . . . Y ™ Ý g S2 S CAŽ SS . S0 X . . . X YS S . . . YS S ,
SS
g 2 a X CA S0 X . . . X Y . . . Y ™ Ý g S2 S a X S CAŽ SS . S0 X . . . X YS S . . . YS S ,
SS
g 2 C L ŽR . ™ C Lg ŽR . ,
g 2 C L ŽR . a X ™ C LaŽR . ,
NA NSS
X 2
g 2 TL ŽR . S0 A . . . A ™ Ý
As1
g A2
ž Ł
SSs1
d SŽ . S
/Ž t L ŽR . A
. S0 A A . . . A A
NSS NSS
X
q Ý
SSs1
g N2 A qSS
žŁ /
K/SS
d KŽ . TL ŽR . S S S0 A S S . . . A S S .
Above, the symbol Y stands for any Lie-algebra-valued field. A new type of term should
also be added. This term is the Abelian mixing term, which is given by
NA NSS
" 1 X
´ Ž 4p . 2 Ý
A1 , A2
ž Ł
SSs1 /
d SŽ . S t LAŽR
1 A2
. t L ŽR . d d x y 14 Ž Em AnA1 y En AmA1 .
H
A1/A2
= Ž Em AnA 2 y En AmA 2 . .
The last line of Eq. Ž37., i.e. the hatted contribution, has to be modified in the same
manner.
The reader should bear in mind that g 2 C L ŽR . a X S0 c ŽX .c ŽX . A ™ C LaŽR . ÝG S0 c ŽX .c ŽX . A G .
4.3. Expression of the breaking
To obtain the generalization of Eq. Ž64. to current case, one applies first the same
kind of modifications as in the previous subsection. In addition, the following replace-
ments should be made
NSS NSS
X
TL ŽR . c a b c M4a b c ™ Ý
SSs1
ž K/SS
Ł /
d KŽ . TL ŽR . SS c a S S b S Sc S S M4a S S b S Sc S S ,
and similar substitutions for the terms of the same type as TL ŽR . c eb c c e d aM a b c d.

The traces of the product of three and four generators, which appear for example in
the expression of anomaly, keep its form, but with the understanding that Tr is a matrix
trace over the full fermion representation of the gauge Lie algebra G.
4.4. Restoration of SlaÕnoÕ–Taylor identities
Apart from modifications of the same type as describe above, we have to be careful
with the resolution of the equations involving M1a b Žand also M 7 and M8 , but they
actually do not give rise to any new result.. Now, we substitute in Eq. Ž67. a1 D˜i ™
a1GD˜iG q a1A1 A 2D˜iA1 A 2 , i s 1,2, A1 / A2 , with the obvious meaning.
Apart from the equation in M1a G b G , which will give together with the rest of the
system an unique solution for
NSS NSS
1 5 2 2
a1A s 2
Ž 4p . 12 ž Ł /Ž
SSs1
dS S t LA . q
ž Ł /Ž
SSs1
d S SX t RA . ,
1 5 X
a1S S s 2 ž Ł d /T K L SS q ž Ł d /T K R SS ,
Ž 4p . 12 K/SS K/SS
NSS NSS
1 1 2 2
dXS S
a2A s 2
Ž 4p . 4 ž Ł /Ž
SSs1
dS S t LA . q
ž Ł /Ž
SSs1
t RA . ,
1 1 X
a2S S s 2 ž Ł d /T K L SS q ž Ł d /T K R SS ,
Ž 4p . 4 K/SS K/SS
´ D. Sanchez-Ruizr
we have new equations in M1A1 A 2 and M1A 2 A1 , A1 / A2 :

NSS NSS
1 1
a1A1 A 2 y a 2A1 A 2 s
Ž 4p .
2
3 žŁ /SSs1
d S S t LA1 t LA 2 q
žŁ /
SSs1
d S S X t RA1 t RA 2 ,
NSS NSS
1 1
a1A 2 A1 y a 2A 2 A1 s 2
Ž 4p . 3 žŁ /SSs1
žŁ /
SSs1
d S S X t RA1 t RA 2 ,
whose solution is not unique. This was expected because we have now a new invariant:
FmnA1 F A 2 mn , A1 / A2 , mixing Abelian bosons. We can impose, for example, a2A1 A 2 ' 0.
Therefore, from Eq. Ž69., with the help of the modifications given above Ždo not
forget the new terms., one obtains following result:
NA NSS NSS
"1 2 2 2
dXSS
Ž1.
Sfct s
Ž 4p .
2 HÝ
As1
ž SSs1
Ł d SS
/ ž
Ž t LA . q Ł
SSs1 /Ž t RA . 5
12 Ž E mAmA .
NSS NSS NSS
dXK TR SS
q 1
4 AmA I A Am
q
SSs1
Ý žŁ /
K/SS
d K TL SS q
žŁ /
K/SS
= 5
12 Ž E mAŽmSS . a . Ž E nAŽnSS . a . q
SS SS 1
4 AmŽ SS . a SS I AŽ SS . a SS m
NSS NSS NSS

TL S S TR S S
dXK
y Ý
SSs1
ž Ł
K/SS
dK
/ 6
q
ž Ł
K/SS / 6
=c a SS b SS c SS Ž Em AnŽ SS . a SS . AŽ SS . b SS mAŽ SS .c SS n
Trfull TLa TLb TLc TLd q Trfull TRa TRb TRc TRd
q Ama A bmAnc A d n
12
NA qNSS
i l
q Ý g G2 C LŽG. 1 q Ž a X Ž G . y 1 . r6 cg m PL pr m c
Gs1 2
NA qNSS
i l
q Ý g G2 C RŽG. 1 q Ž a X Ž G . y 1 . r6 c Xg m PR pr m c X
Gs1 2
NSS
CAŽ SS . X X
y Ý g S2 S 1 q Ž a X Ž SS . y 1 . ž Lsc q Lsc q Rsc q Rsc /
SSs1 4
NA qNSS
q "1 Ý Ž lg G
Ž1.
L gŽ G . q l AŽ 1Ž G. . u02 L AŽ G . q lvŽ1.ŽG . LŽG. 1 Ž1 . L 1 Ž1.
v . q " lc Lc q " lc Lc
X RX
Gs1
NA NSS NSS
"1 1
q
Ž 4p .
2
3
Ý
A1 , A2
žŁ /
SSs1
žŁ /
SSs1
d S S X t RA1 t RA 2
A1/A2
= d d x Ž Em A A1 m . Ž Em A A 2 m .
H
NA
1
q "1 Ý l Ž1. d
A1 A 2 d x
H FmnA1 F A 2 mn , Ž 92 .
A1 , A2
2 g A1 g A 2
A1/A2
g G , l AŽG . , lv ŽG . , lc , lc , and each l A1 A 2 1 for A1 - A2 ( NA are arbitrary

where l Ž1. Ž1. Ž1. Ž1. Ž1.
X Ž .
coefficients which will determine the renormalization conditions at the one-loop order
ŽG. L RX
and LŽG. ŽG.
g , L A , Lv Lc , Lc , are any of the generalizations to the d-dimensional
space-time of Dimensional Regularization of the corresponding b-invariant four-dimen-
sional operators which are written down explicitly in next subsection.
4.5. Renormalization group equation
The O Ž " 0 . Lorentz invariant functionals L i with ultraviolet dimension 4 and ghost
number 0 which satisfy the following restrictions:
d
b Li s 0 L s0 , G Li s 0 ,
d Ba i
form a linear space. It is very convenient to have a basis for such space constituted by
elements which are derived by the action of a compatible differential operator over the
classical action, because, as explained in Subsection 3.7, then, a quantum basis can be
properly defined. The expansion of mEm Gren in such a quantum basis, is just the
renormalization group equation. Let us construct a basis for that linear space.
(a) Cohomologically triÕial terms

We define for each field
d
Nfab ' d 4 x f a
H ,
df b
a,b being group indices. These are terms which are b-exact:
E
LaA b ' b P d 4 xAma r˜ bm s NAab y Nrab y Nvab y NBab q 2 L a e
H Scl
EL eb
' NAab Scl ,
Lav b ' yb P d 4 x v az b s Nvab y Nzab Scl ' Nvab Scl ,

H
X
LLc ŽR
ŽX .
.i 4
j ' yb P d x L j R j P L ŽR . c
H ž / Ž .i
s NcLŽXŽR
.j
.i
y NLŽi R. j Scl ,
.j X
RXŽL . j
LRc ŽL
ŽX .
4
H Ž. j j j
i ' yb P d x c i PRŽL . L Ž R . s Nc Ž . i y NLŽ R.i Scl ,
´ D. Sanchez-Ruizr
X
LRc ŽL
ŽX .
.i 4
j ' yb P d x L j R j PRŽL . c
H Ž .i
ž / s NcRŽL
ŽX .
j
.i
Scl ,
.j RXŽL . j
LLc ŽR
ŽX .
4 j j
i ' yb P d x c i P L ŽR . L Ž R . s Nc Ž . i
H Scl ,
but not all are linearly independent:

d i j LLc ŽR i RŽL . j
ŽX . j s d j Lc ŽX . i ,
.i
d i j LRŽL i L ŽR . j
c ŽX . j s d j Lc ŽX . i .
.i
In the expression of L A in terms of field counting operators, it is used the fact that Scl
satisfies the gauge condition and the ghost equation.
Usually, not only the BRS symmetry is imposed to the 1PI functional, but also other
symmetry of the classical action, the rigid symmetry:
a
d erigf a s i TeŽ f . b
f b,
X
where TeŽ f . is the adjoint
X
representation for Am , v , v , rm , z and B; T L ŽR . e for c Ž ., LŽ R .;
and yTL ŽR . e for c , L R .
Ž. Ž .
Then d erig Scl s 0, which has the consequence w d erig ,b x s 0.

Also, our 1PI renormalized functional constructed from Eq. Ž92. with Breitenlohner
and Maison minimal dimensional renormalization realizes this symmetry at one-loop
level.
If the 1PI functional satisfies this new requirement, the set of admissible b-invariants
terms is obviously smaller. Thanks to proposition 5.9 of Ref. w32x, the trivial subset
formed by terms which are b-exact and rigidly symmetric is fully determined by
studying the restricted cohomology, i.e. by studying the set of terms which are
b-variations of rigidly symmetric terms.
Some of them are obvious generalizations of the trivial elements of Eq. Ž30.:
LŽASS . s b P d 4 x r˜ amS SŽ SS . AmŽ SS . a S S ,

H LŽvSS . s yb P d 4 x z aŽSSS
H S v
. Ž SS . a S S
,
i l
LLc s yb P d 4 x LPL c q c PR L s 2 d 4 x
H H c Eu PL c q cg m PL TLa c Ama ,
2
i l
LRc X s yb P d 4 x RPR c X q c X PL R s 2 d 4 x
H H c X Eu PR c X q c Xg m PR TRa c XAma ,
2
i l
LRc s yb P d 4 x LPR c q c P L L s 2 d 4 x
H H c Eu PR c ,
2
i l
LLc X s yb P d 4 x RPL c X q c X PR R s 2 d 4 x
H H c X Eu PL c X ,
2
where SS runs from 1 to NSS .
The rest are terms which are b-exact but involve possibly different Abelian compo-
nents:
LA
A
1 A2 s b r˜ A1 m AmA 2
H
Ž y1.
s d4 x 2
H 2
FmnA 2 F A1 mn q cg m t LA1 PL c AmA 2 q c Xg m t RA1 PR c XAmA 2
4gA 1
y r˜mA1E mv A 2 ,
LvA1 A 2 s yb z A1 v A 2
H
s d 4 x r˜mA1E mv A 2 Li v A 2 t LA1 PL c q c PR t LA1 i v A 2 L q Ri v A 2 t RA1 PR c X
H ½
qc X PL t RA1 i v A 2 R . 5
Notice that similar non-diagonal terms are not allowed for non-Abelian components
because of rigid invariance Ž L A A
1 A 2 and LvA1 A 2 are trivially rigid invariant even if
A1 / A2 . .
LA
A
1 A 2 and L A1 A 2 are related with the following multiplicative field renormalizations:
v
A A
AmA1 s Ž ZA . A21 Amw A 2 , BA s Ž Zy1 2 w
A . A B A2 ,
A A
v A1 s Ž Zv . A21 v w A 2 , v A1 s Ž Zy1 2 w
A . A1 v A 2 ,
A
wm A
r Am1 s Ž Zy1 2
A . A1 r A 2 , z A1 s Ž Zvy1 . A21 z Aw2 ,
A A X X
L A 1 A 2 s Ž ZA . 1 A1X Ž ZA . A22X Lw A 1 A 2
But they will not be needed in renormalization group computations, because in
minimal subtraction of loop diagrams this kind of terms can not appear due to the fact
that they involve Abelian ghosts.
(b) Elements of the cohomology

Imposing the rigid invariance, we have the generalization of the unique element of
the cohomology of the simple group case:
E Scl 1 4 ŽG. a G ŽG. mn
LŽG.
g s gG s Hd x Fmn Fa G ,
E gG 2 g G2
and the new elements Žfor A1 / A2 .:
1
L A1 A 2 s d 4 x
H FmnA1 F A 2 mn
2 g A 1 g A2
with A1 , A2 ( NA . Note that L A A s L g A s g AEg A Scl .

These terms come from the renormalization of the Abelian components of the matrix
C of Eq. Ž91.. At tree level, it is a diagonal matrix, but, more generally, it can be
changed to a non-diagonal matrix in the Abelian part while the right-hand side of Eq.
Ž91. continue being a rigidly invariant term:
A1X A2X wX X
C A1 A 2 s Ž ZC . A 1 A2 CA 1 A2
.
We define
E L A1 A 2
LCA1 A 2 s Scl s y .
E C A1 A 2 2
´ D. Sanchez-Ruizr
Therefore, Eq. Ž74. becomes

NA qNSS
E E
m q Ý b G gG y gcLNcL y gcRX NcRX y gcRNcR y gcLX NcLX
Em G E gG
NSS NSS NA
E
y Ý gASSNAŽ SS . y Ý gvSSN vŽ SS . y Ý gCA1 A 2 Gren s 0,
SSs1 SSs1 A1 , A2
E C A1 A 2
A1/A2
which at order " reads

NA qNSS
EGren
m s y Ý b G Ž1. LŽG.
g q gc
L Ž1. L
Lc q gcRX Ž1. LRc X
Em Gs1
NSS NSS
q Ý gASS Ž1. LŽASS . q Ý gvSS Ž1. LŽvSS .
SSs1 SSs1
NA
q Ý gCA1 A 2 Ž1. LCA1 A 2 qO Ž " 2 . .
A1 , A2
A1/A2
Eq. Ž88. is thus generalized.

Eq. Ž87. should be modified by applying the rules stated above and adding the new
term:
1 8 NA NSS
1
q 2 Ý
Ž 4p . 3 A , A SSs1
Ł
1 2
ž
d SŽX . S t LAŽR
1 A2
. t L ŽR .
4 /
d 4 x y FmnA1 F A 2 mn Scl A A H
A1/A2
Therefore,
NSS NSS
1 2 2 2
dXSS
b A Ž1. s
Ž 4p .
2
2
gA
3 ž Ł
SSs1
NSS
d SS
/ Ž t LA . q
ž SSs1
Ł
NSS
/Ž t RA . ,
1 2 11
b SS Ž1. s
Ž 4p .
2
g S2 S
ž ž
3
Ł
K/SS
X SS
/
d K TL SS q
ž Ł
K/SS
dXK TR SS y
/ 3 /
CAŽ SS . ,
1 1ya
gASSŽ1. s
Ž 4p .
2 ž
g S2 S 2 y
2 / CAŽ SS . ,
1
gvSSŽ1. s 2
g S2 S a X SS
CAŽ SS . ,
Ž 4p .
NA qNSS N A qNSS
1 1
gcLŽ1. s 2 Ý g G2 a X G C LŽG. , gcRX Ž1. s 2 Ý g G2 a X G C RŽG. ,
Ž 4p . Gs1 Ž 4p . Gs1
1 4
gCA1 A 2 Ž1. s 2 Ž t LA t LA q t RA t RA . ,
1 2 1 2
Ž 4p . 3
where a X G ' a Grg G2 , 1 ( A, A1 , A2 ( NA , 1 ( SS ( NSS .
5. Conclusions
We have shown by performing explicit thorough one-loop computations that Dimen-

sional Regularization a´ la Breitenlohner and Maison can be used blindly to carry out
calculations in chiral gauge theories. Up to best of our knowledge it is the first time that
such a complete one-loop study has been carried out. It shows that the renormalization
method at hand has all the properties that on general grounds were expected. Notice that
here the cancellation of anomalies is not used to try to hide any inconsistency. Even
when there is not cancellation of anomalies, the renormalized theory is finite and
unambiguous: in particular, the BRS anomaly is obtained without doing any tinkering.
This is in stark contrast to the ‘‘naive’’ approach, where cancellation of anomalies is
invoked to dangerously try to hide some inconsistencies.
We would like to stress the fact that it would not have been feasible to carry out the
computations made in this paper without the help of the action principles and the BRS
cohomolgy theory. This makes manifest the importance of these two theoretical tools in
practical computations in chiral gauge theories.
Acknowledgements
Financial support from the Universidad Complutense de Madrid under grant PR156/
97-7164 and the Consejerıa ´ de Educacion
´ y Cultura of the Junta de Comunidades de
Castilla-La Mancha are acknowledged.
Appendix A. Slavnov–Taylor identities
This appendix is devoted to establishing Eq. Ž19. by using the Regularized Action
Principle as stated in Eqs. Ž11. and Ž12..
Ž n.
Let SDR eg be given by
Sfree w w ,F x q Sint w w ,F ; KF x ,
where
Sfree w w ;F x s S0 free w w ;F x ,
Sint w w ,F ; KF ; l x s S0 int w w ,F x q d d x KF Ž x . s dF Ž x . q SctŽ n. w w ,F ; KF x .

H
The symbols w and F stand, respectively, for sets of fields which transforms linearly
and non-linearly under arbitrary BRS transformations. The generalization to ‘‘d-dimen-
sions’’ of the BRS transformations are denoted by s d f Ž x . and s dF Ž x .. We define
s d KF Ž x . s 0
We next define
SINT w ,F ; Jw , JF , KF s Sint w w ,F ; KF x q d d x Ž Jw Ž x . w Ž x . q JF Ž x . F Ž x . . .
H
´ D. Sanchez-Ruizr
Following Eq. Ž10., the generating functional Z DR eg is introduced next:
i
Z DR eg Jw , JF , KF s exp ¦½ "
SINT w ,F ; Jw , JF , KF 5; 0
,
where the symbol ² . . . : 0 is defined as in the paragraph below Eq. Ž10..

Let D and Dct de defined as follows:
D s s d S0 , Dct s s d SctŽ n. ,
where S0 s S0 free w w ;F x q S0 int w w ,F x q Hd d x KF Ž x . s dF Ž x ..

ŽF .
We now introduce SINT for reasons that will become clear later:
ŽF .
SINT s SINT w ,F ; Jw , JF , KF q Saux ,
where
d SctŽ n .
H ž
Saux s d d x F1 Ž x . D Ž x . q F2 Ž x . Dct Ž x . q F3F Ž x .
d KF Ž x . / ,
and D s Hd d x DŽ x . and Dct s Hd d x Dct Ž x .. Fi , i s 1,2 and 3 are external fields. We

define s dFi Ž x . to be zero for any external field Fi . Then, we introduce the generating
functional ZFDR eg as follows:
i
ZFDR eg Jw , JF , KF ,F1 ,F2 ,F3F s exp ¦½ "
SFINT w ,F ; Jw , JF , KF ,Fi 5;0
,
where ² . . . : 0 is defined as for Z DReg .

It is obvious that the following equation holds in Dimensional Regularization:
Z DR eg Jw , JF , KF s ZFDReg Jw , JF , KF ,F1 s 0,F2 s 0,F3F s 0 .
The Regularized Action Principle Žsee Eq. Ž11.. implies that the following equation
holds in Dimensional Regularization:
w
s d ZFDReg ' d d x ² D Ž x . q Dct Ž x . q Ž y1 . Jw Ž x . sw Ž x .
H ž
F
q Ž y1 . JF Ž x . sF Ž x . q terms proportional to Fi /
i
=exp ½ "
SF
INT w w ,F , Jw , JF , K F ,Fi 5 : 0s0. Ž A.1 .
Next, by using Eq. Ž12., one easily obtains that
i " d ZFDR eg d ZFDReg

² Jw Ž x . sw Ž x . exp ½ " 5
SFINT : 0 s
i
Jw Ž x . aww
d Jw Ž x .
q awF
d JF Ž x .
Ž A.2 .
and
i d Ž SF y Sct y Saux .
² JF Ž x . sF Ž x . exp ½ "
SFINT 5 : 0 s JF Ž x . ²
d KF Ž x .
:0
" d ZFDReg " d ZFDReg

s JF Ž x . y JF Ž x .
i d KF Ž x . i dF3F Ž x .
q terms proportional to Fi . Ž A.3 .

In equation ŽA.2. the linear BRS transformations s d w are given by s d w s aww w q awF F .
Now, by substituting Eqs. ŽA.2. and ŽA.3. in Eq. ŽA.1., one arrives
w " d ZFDR eg d ZFDReg

Hd d
x Ž y1 .
i ž
Jw Ž x . aww
d Jw Ž x .
q awF
d JF Ž x . /
F " d ZFDReg d ZFDReg " d ZF
DReg " d ZF
DReg
q Ž y1 .
i
JF Ž x .
ž KF Ž x . .
y
dF3F Ž x . / q
i dF1 Ž x .
q
i dF2 Ž x .
q terms proportional to Fi s 0. Ž A.4 .
Let us introduce as usual the generating functionals ZF w

c DReg Jw , JF ; K F ,Fi
x and
F w w ,F ; KF ,Fi x:
G DReg
ı
ZFDR eg Jw , JF ; KF ,Fi s exp ž ZFc DReg Jw , JF ; KF ,Fi / ,
"
F F d
G DR eg w w ,F ; K F ,Fi x s Z c DReg Jw , JF ; K F ,Fi y d x Ž Jw Ž x . w Ž x . H
qJF Ž x . F Ž x . . . Ž A.5 .
where the functionals in the previous equations are to be understood as formal power
series in " and the fields Žboth external and quantum.. By taking advantage of Eq. ŽA.5.
F
eg w ,F ; K F ,Fi :
one turns Eq. ŽA.4. into an equation for the 1PI functional G DR w x
F F F F
dG DR eg dG DReg dG DReg dG DReg
H dd x y
dw Ž x .
Ž aww w Ž x . q awF F Ž x . . y dF Ž x . ž d KF Ž x .
y
dF3F Ž x . /
F F
dG DR eg dG DReg
q q q term prop. to Fi s 0. Ž A.6 .
dF1 Ž x . dF3F Ž x .
Finally, by setting Fi Ž x . s 0, ; i, in Eq. ŽA.6. and taking into account that

F
G DR eg w w ,F ; KF x s G DReg w w ,F ; KF ,Fi s 0 x
´ D. Sanchez-Ruizr
one obtains Eq. Ž19. from Eq. ŽA.6.:
d
dG DR eg dG DReg dG DReg
Hd x Ž sd w . q
dw Ž x . d KF Ž x . dF Ž x .
s d d x D Ž x . P G DR eg q Dct Ž x . P G DReg
H
d Sct dG DReg
q
ž d KF Ž x .
P G DR eg
/ dF Ž x .
. Ž A.7 .
Notice that
F F
dG DR eg dG DReg dG DReg d Sct
y s y P G DReg ,
d KF Ž x . dF3F Ž x. F is0
d KF Ž x . d KF Ž x .
F
dG DR eg
s D Ž x . P G DR eg ,
dF1 Ž x . F is0
F
dG DR eg
s Dct Ž x . P G DReg ,
dF2 Ž x . F is0
for Eq. Ž12. holds.
Appendix B. Slavnov–Taylor identities for ´-dependent BRS
This appendix is devoted to the study of the contributions to the right-hand side of
Eq. Ž21. coming from BRS variations which depend explicitly on ´ s 4 y d. We shall
see that, unlike for the BRS transformations in Eq. Ž31., the variation under these new
BRS transformations of the singular counterterms yield a finite contribution to the
right-hand side of Eq. Ž21..
To introduce an explicit ´ dependence on the Dimensional Regularization BRS
transformations, we shall replace the BRS transformation of Am given in Eq. Ž31. with
sXd Am s =m v q C Ž d y 4.=m v . We shall assume, however, that sXd acts on the rest of the
fields as s d does. Hence, sXd defines BRS transformations with explicit ´ dependence.
Although we have changed the Dimensional Regularization BRS transformations, we
shall not take a new Dimensional Regularization classical action, which still will be S0
in Eq. Ž32.. Thus, we will not modify the minimally renormalized 1PI functional Gminren ,
so that we will have the effect of the new Dimensional Regularization BRS transforma-
tions isolated.
The sXd variation of S0 is given by
sXd S0 s Dˆ q Df´ q Dgf´ q Dr´ .

Here, Dˆ s s d S0 is the breaking given in Eq. Ž34. and

a
Df´ s y´ C d d x cg m PL TLa c q c X g m TRa c X
H ž / Ž= v .
m ,
a b
Dgf´ s y´ C d d x B a E m Ž =m v . yc a b c E m v a Ž =m v . v c ,
H
b
Dr´ s q´ C d d x c a b c r a m Ž =m v . v c
H
are the new breakings Žwhich are also evanescent operators.. Notice that sXd2 Ama / 0,
whereas s d2 Ama s 0.
Now, let G 0 denote dimensionally regularized 1PI functional obtained from S0 , then,
regularized Slavnov–Taylor equation for the sXd-transformations reads
SSXd Ž G 0 . s Dˆ q Df´ q Dgf´ q Dr´ P G 0 , Ž B.1 .
where SSXd is the Slavnov–Taylor operator for sXd . Notice that
dG 0 dG 0 dG 0 dG 0 dG 0
H ½
SSXd Ž G 0 . ' d d x Ž 1 y C´ . Tr
dr m
d Am
q Tr
dz dv
q Tr B
dv
dG 0 dG 0 dG 0 dG 0 dG 0 dG 0 dG 0 dG 0
q
d L dc
q
d R dc
d L dc X q
d R dc X
. q 5
The next issue to address is the derivation of the Žanomalous. Slavnov–Taylor identity
satisfied by the minimal dimensional renormalization 1PI functional, Gminren . If we,
naively, just replace in Eq. ŽB.1. every regularized object with their minimally renormal-
ized counterpart, we will obtain the following equation:
SS Ž Gminren . s N Dˆ q Df´ q Dgf
´
q Dr´ P Gminren . Ž B.2 .
This equation does not hold, though, since the coefficients of the regularized Slavnov–
Taylor equation, Eq. ŽB.1., do have an explicit ´ dependence. Indeed, at the one-loop
level the following identity holds:
Ž1 .
SS Ž Gminren . s N w Dˆ x P Gminren q O Ž "2 . . Ž B.3 .
This has been shown in Subsection 3.4. Now, let us substitute the previous equation in
Eq. ŽB.2.. We thus come to the conclusion that Nw Df´ q Dgf ´
q Dr´ x P Gminren should
vanish at the one-loop level. Which is nonsense, for a diagramatic computation shows
that
Ž1 .
N Df´ q Dgf
´
q Dr´ P Gminren
y" d Scl cc A
s
Ž 4p .
2 ½
C g 2 Ž a X CA q 2 a X C L . Hd 4
x
d Ama
Em v a
d Scl cc A
q Ž 3 y 32 Ž 1 y a X . . CA q 2 a X C L Hd 4
x c a b c Amb v c
d Ama
d Scl c Xc X A
q Ž a X CA q 2 a X C R . Hd 4
x Em v a
d Ama
´ D. Sanchez-Ruizr
3 d Scl c Xc X A
q 3y ž Ž 1 y a X . CA q 2 a X C R
/ Hd 4
x c a b c Amb v c
2 d Ama
3 d Scl v v A d Scl rv A
Ž 1 y a X . CA Hd 4 x
y
2 ž d Ama
q
d Ama / Em v a
d Scl v v A d Scl rv A
q2 a X CA Hd 4
x
ž d Ama
q
d Ama / c a b c Amb v c
1
q 1qž Ž 1 y a X . CA Hd 4 x Ž Em B a . Ž E mv a .
/
2
ya X CA 4
Hd x Ž E B . c m
a abc
Amb v c .
5 Ž B.4 .
Hence, a contribution to the right-hand side of Eq. ŽB.2. is missing. This missing
contribution should apparently be furnished by the second and third terms on the
right-hand side of Eq. Ž21., which is a consequence of Eq. ŽA.7.. But, it is essential to
remember the assumption made in Appendix A in order to properly deduce Eq. ŽA.7.:
the dependence in KF in the starting dimensional regularized action should be HKF sXdF ,
if sXd is the variation under study.
Therefore, due to the fact that the dependence in rm of the chosen action is just
Hrm s d A m, an extra factor of 1 y C´ is needed in Eq. ŽA.3., when studying sXd
variations:
i F
¦ JAm a Ž x . sXd Ama Ž x . exp S
" INT 0 ½ 5;
i
s Ž1yC ´ . ¦ JAm a Ž x . sXd Ama Ž x . exp ½ "
SFINT 5; 0
" d ZFDReg d ZFDReg

s Ž1yC ´ .
i
JAm a Ž x .
ž drma x Ž .
y
dF3A mA x Ž . /
q terms proportional to Fi .
Then, Eq. ŽA.7. becomes
d Sct dG DReg
SSXd Ž G DReg . s sXd Ž S0 q Sct . P G DReg q d d x H Ý
F/A
ž d KF Ž x .
P G DReg
/ dF Ž x .
d Sct dG DReg
q Ž 1 y C´ .
ž drma Ž x.
P G DR eg
/ d Ama Ž x .
,
and the correct modification of Eq. ŽB.2. is

Ž1 .
SS Ž Gminren . s N Dˆ q Df´ q Dgf´ q Dr´ P Gminren
a dF d Scl d F
q C d4 x
H Ž =m v . q qO Ž " 2 . . Ž B.5 .
d Ama d Ama dr am
Here, F s LIM d ™ 4 ´G 0Ž1.

sing 4 , with G 0 sing as given by Eqs. 36 and 37 . Notice that
Ž1. Ž . Ž .
the second term on the right-hand side of ŽB.5. cancels the contribution displayed in Eq.
ŽB.4.. Eq. ŽB.3. is thus recovered.
Of course, if we want to use Eqs. Ž19., Ž21. and Ž22. – with sXd instead of s d – we
have to choose the following new Dimensional Regularization classical action:
SX0 s S0 y d d x rma s d A a m q d d x rma sXd A a m ,

ž H / H
which cause the breaking
sXd SX0 s Dˆ q Df´ q Dgf´ q Ž 1 y C´ . Dr´ ,
X
and a new perturbative renormalized 1PI functional Gminren .
Therefore, in this case, from Eq. Ž19., Ž21. and Ž22., we conclude that the Žanoma-
lous. one-loop renormalized Slavnov–Taylor identity reads is
Appendix C. Comparison between ‘‘naive’’ and Breitenlohner and Maison minimal

dimensional renormalization schemes
In this appendix we shall make a thorough comparison of the minimally renormalized

one-loop 1PI functionals obtained within the ‘‘naive’’ dimensional renormalization
prescription, on the one hand, and the Breitenlohner and Maison Dimensional Regular-
ization scheme, on the other. The quantum field theory under scrutiny will be the theory
already studied in Section 3. Obviously, at the one-loop level, differences only arise in
diagrams involving fermions. We will denote these contributions with the subscript F.
All along this appendix, the subscripts BM and ‘‘naive’’ will indicate that the corre-
sponding quantity has been evaluated by using the Breitenlohner and Maison scheme
and the ‘‘naive’’ prescription, respectively.
Let us begin with the self-energy of the gauge field. Here, only the fermionic loops
may give rise to differences between the two calculational techniques. There are terms
with one g 5 and four g ’s in the trace, but their contribution explicitly vanish, the reason
being the contraction of two g ’s of the antisymmetic trace with the same momentum q.
The ‘‘naive’’ prescription does not yield ambiguous the results in this instance. These
results read
mn , a b
1 1 g 5 1 yp 2 q i e
G̃ FŽ1.A Anaive Ž p. s T q TR .
2 Ž L
y q y ln qOŽ ´ .
4p 3´ 6 18 6 4pm2
=d a b Ž p m p n y p 2 g mn . ,
for the regularized theory and
mn , a b
1 g 5 1 yp 2 q i e
G̃ FRAŽ1.
Anaive Ž p . s T q TR . y q
2 Ž L
y ln
4p 6 18 6 4pm2
=d a b Ž p m p n y p 2 g mn . ,
for the renormalized Green function. It should be understood a sum over the different
fermionic multiplets: Ý nks
L T k q Ý n R T r . Notice that both results are transverse, so that
1 L rs1 R
the corresponding Slavnov–Taylor identity holds.
´ D. Sanchez-Ruizr
The calculation within the Breitenlohner and Maison framework is also straightfor-
ward and it yields the following regularized expression:
G˜FŽ1.A ABM
mn , a b
Ž p.
1 1 g 5 1 yp 2 q i e
sd ab
½ 4p 2
Ž T L q TR .
3´
y
6
q
18
y
6
ln
4pm2
1 1
= Ž p m p n y p 2 g mn . y 2 Ž T L q TR . g mn p 2
4p 12
1 Ž T L q TR . 1 g 5 1 yp 2 q i e 1
y
4p 2
2 3´
y
6
q
18
y
6
ln
4pm 2
q
6
g mn p̂ 2
5
qOŽ ´ . . Ž C.1 .
Notice that there is a singular contribution at ´ s 0 whose residue is a an ‘‘evanes-
cent’’ polynomial of the momentum. Minimal subtraction has to be applied also to this
contribution; in the language of ultraviolet divergent counterterms: counterterms with
hatted objects are also needed! This is important for the consistency of calculations in
the next orders w47–49x.
The renormalized 1PI fermionic contribution to the gauge field two-point function
defined by minimal subtraction a´ la Breitenlohner and Maison reads
1 Ž T L q TR .
G˜FRAŽ1.
ABM
mn , a b
Ž p . s G˜FRAŽ1.Anaive
mn , a b
Ž p. y d ab 2
p 2 g mn ,
4p 12
which differs from the ‘‘naive’’ result in a non-transversal quantity, so that it is clear
that this result do not satisfy the Slavnov–Taylor identity.
But, we learned from Eq. Ž40. that the Slavnov–Taylor identities for the regularized
theory a´ la Breitenlohner and Maison have a breaking, so instead of transversality the
following equation should hold
yipm GÃŽ1.
ABM
mn , a b
Ž p . s GÃŽ1. n ,ab
v ; Dˆ BM Ž p;0 . Ž C.2 .
Žbecause the insertion of the breaking is integrated, it is an insertion at zero momentum..
The right-hand side of Eq. ŽC.2. is just the sum of the 1PI diagrams in Fig. 5 with Dˆ
instead of Dˇ and p instead of p 1. Their value is
i Ž T L q TR . 1 g 5 1 yp 2 q i e
G̃AŽ1. m ,ab
v ; Dˆ BM Ž p;q . s d
ab
½ 4p 2 2 3´
y
6
q
18
y
6
ln
4pm2
i Ž T L q TR .
= p m pˆ 2 q
4p 2
12 5
p m p 2 qO Ž ´ . , Ž C.3 .
as it should be. Indeed, Eqs. ŽC.1. and ŽC.3. satisfy Ž O Ž ´ .. Eq. ŽC.2.. The contribution
from the sum of diagrams with bosonic and ghost loops is obviously transversal and thus
it is of no bearing here.
From Eq. Ž42. we get the renormalized counterpart to Eq. ŽC.2.:

yipm GÃRABM
Ž1. mn , a b
Ž p . s GÃRv Ž1. n ,ab
; N w Dˆ xBM Ž p;0 . ,
which is also satisfied because minimal subtraction applied to Eq. ŽC.3. yields
m ,ab
i Ž T L q TR .
G̃ARv Ž1.
; N w Dˆ xBM Ž p;q . s d
ab
2
p m p2 .
4p 12
We have checked thus Eq. Ž42. by explicit computation of both its sides.
The following results will be needed as well
1
G˜ccR BM
Ž1. a i , b j
Ž p . y G˜ccR naive
Ž1. a i , b j
Ž p. s 2
g 2 1 q Ž a X y 1 . 23 C L w pu P L x ab d i j ,
Ž 4p .
GÃRccŽ1.BMn , a i , b j Ž q, p . y GÃRccŽ1.naive
n ,a i,b j
Ž q, p .
1 CA
sy 2
g 2 CL y
ž / 2 q Ž a X y 1 . 56 g m PL ab TLa ij
Ž 4p . 2
1
qy 2
g 2 CA 1 q Ž a X y 1 . 125 g m PL ab TLa ij
Ž 4p .
1
sy 2
g 2 C L 2 q Ž a X y 1 . 56 g m PL ab TLa i j.
Ž 4p .
The first term of the sum corresponds to the QED-like diagram and the second term to
the diagram with the three boson vertex.
Let us move on to the computation of the most interesting one-loop 1PI functions to
compare, namely, the three and four point 1PI functions for the gauge field. In both
these cases the ‘‘naive’’ prescription leads to ambiguities.
The fermionic diagrams which contribute to the one-loop three boson vertex are
shown in Fig. 13. The contribution from the diagram drawn there will be denoted with
mnr Ž
iL ab c k 1 ,k 2 ,k 3 . The full contribution will thus read
.
i G˜FŽ1.A mnr mnr nm r mnr

A A a b c Ž k 1 ,k 2 ,k 3 . ' iL a b c Ž k 1 ,k 2 ,k 3 . q iL b ac Ž k 2 ,k 1 ,k 3 . q iR a b c Ž k 1 ,k 2 ,k 3 .
q iRnm r
b ac Ž k 2 ,k 1 ,k 3 . .
We shall focus first on the calculation of iL amnr mnr

b c q iR a b c . The other diagrams are
obtained simply by exchanging Ž k 1 , m ,a. with Ž k 2 , n ,b .. Of course, the three momenta
are not independent: for example, k 3 s yk 2 y k 1.
Fig. 13. The one-loop 1PI Feynman diagrams with three bosonic legs.
´ D. Sanchez-Ruizr
The ‘‘naive’’ and Breitenlohner and Maison dimensionally regularized integrals take
the form
iL amnr mnr
b c Ž k 1 ,k 2 ,k 3 . q iR a b c Ž k 1 ,k 2 ,k 3 .
ddk Namnr
bc
s m´ H Ž 2p . d 2 2 2
,
Ž k y k1 q i e . Ž k q i e . Ž k q k 2 q i e .
where the numerator depends on the regularization algorithm:
Namnr
bc
abc
naive s T L Ž ku y ku 1 . g m PL kug n PL Ž ku q ku 2 . g r PL
Tr
q TRa b c Tr Ž ku y ku 1 . g m PR ku g n PR Ž ku q ku 2 . g r PR ,
Namnr abc
b c BM s T L Tr Ž ku y ku 1 . g m PL ku g n PL Ž ku q ku 2 . g r P L
q TRa b c Tr Ž ku y ku 1 . g m PR ku g n PR Ž ku q ku 2 . g r PR .
Now, we anticommute the g 5’s of the projectors in the ‘‘naive’’ expression towards
the right to get
TRa b c q TLa b c
Namnr
b c1
naive
s Tr Ž ku y ku 1 . g m kug n Ž ku q ku 2 . g r
2
TRa b c y TLa b c
q
2
Tr Ž ku y ku 1 . g m kug n Ž ku q ku 2 . g r g 5 . Ž C.4 .
Notice that when the integration over the k’s is done a trace over six gammas, two of
them contracted, and a g 5 occurs. This is precisely the source of the ambiguity
displayed by Eq. Ž7.! Indeed, if we had decided to leave the matrix g 5 in another place,
the minimally renormalized results should differ in a local quantity: there is a divergent
loop present, and the ambiguity of Eq. Ž7. is proportional to ´ s 4 y d. This is the
reason why we have labelled the numerator with the subscript 1.
Therefore, we can write
TRa b c q TLab c
iL amnr mnr
b c Ž k 1 ,k 2 ,k 3 . q iR a b c Ž k 1 ,k 2 ,k 3 . s
mnr
T VVV Ž k 1 ,k 2 ,k 3 .
2
TRa b c y TLa b c mnr
q T VV A Ž k 1 ,k 2 ,k 3 . , Ž C.5 .
2
where T VV V stands for the Abelian Green function Ži.e. without gauge group matrices.
mnr Ž
and with three vector-like vertices and T VV A k 1 ,k 2 ,k 3 denotes the Abelian diagram with
.
r
an axial vertex g g 5 in the leg carrying momentum k 3 .
The Breitenlohner and Maison counterpart to Eq. ŽC.4. is Žwe have used Eq. Ž26.. the
following:
TRa b c q TLa b c
Namnr
bc BM s Tr ž ku y ku / g
1
m
ku g n ku q ku 2 g r
ž /
2
TRa b c y T La b c m
q Tr ž kuyku / g
1 ku g n kuqku2 g r g 5 .
ž /
2
It is easy to calculate the difference between the renormalized result obtained with the
help of Eq. ŽC.4. Ž do not moÕe around the g5 ‘‘matrix’’ . and the renormalized result
obtained from Eq. ŽC.5.: the only contribution to this difference comes from the
renormalization of the dimensionally regularized integral whose numerator is given by
TRa b c q TLa b c
Tr kû g m kû g n ku q ku 2 g r qTr kû g m ku g n kû g r
2 ½ ž /
TRa b c y T La b c
ku y ku 1 g m kû g n kû g r Tr kû g m kû g n ku q ku 2 g r g 5
qTr ž / 5 q
2 ½ ž /
qTr kû g m ku g n kû g r g 5 qTr kû g n kû g r g 5

ž ku y ku / g 1
m
5 .
The value of that difference is thus the following:
iL amnr R mnr R mnr R

b cBM Ž k 1 ,k 2 ,k 3 . q iR a b cBM Ž k 1 ,k 2 ,k 3 . y iL a b cnaive 1Ž k 1 ,k 2 ,k 3 .
yiR amnr R
b cnaive 1Ž k 1 ,k 2 ,k 3 .
r m
i TRa b c q TLa b c mn
Ž k1 y k 2 . nr
Ž k1 q 2 k 2 .
s 2
4 g qg
Ž 4p . 2 3 3
n
Ž 2 k1 q k 2 . i TRa b c y TLa b c Ž ku 1 y ku 2 .
yg m r q 2
Tr g mg ng rg 5 .
3 Ž 4p . 2 3
Ž C.6 .
Notice that the second term of the previous equation is not cyclic, i.e. if we make the
change Ž k 1 , m ,a. ™ Ž k 2 , n ,b ., Ž k 2 , n ,b . ™ Ž k 3 , r ,c ., Ž k 3 , r ,c . ™ Ž k 1 , m ,a. the term
changes. This seems to be incompatible with having a trace which is cyclic Žor with the
fact that we can start to read a closed fermion loop wherever we want.. Of course, if a
prescription for Feynman rules is consistent, the renormalized result should be also
cyclic. This is the case for the Breitenlohner and Maison result, as can be checked
explicitly. But in the ‘‘naive’’ prescription case, if we insist on an anticommuting g 5 ,
the cyclicity of the trace is to be abandoned, lest the trace of four gammas and one g 5
vanishes. The lack of cyclicity of Eq. ŽC.6. is due, of course, to the inconsistency of the
‘‘naive’’ prescription with the cyclicity of the trace.
One could insist on keeping an anticommuting g 5 and cyclicity of the trace. But,
then, as we know from Eq. Ž6., the trace of four gammas and a g 5 should unavoidably
vanish. This would not do since it would set to zero any dimensionally regularized
Feynman integral that is finite by power-counting at d s 4 and whose Dirac algebra
yields a contribution proportional to the trace of four gammas and one g 5 .
´ D. Sanchez-Ruizr
The cyclicity of the diagram is also needed for obtaining a Bose symmetric 1PI
function after summing the crossed diagram. Taking into account this crossed diagram
the difference between the complete 1PI functions in both schemes is
G˜ Ž1. mnr R
a b cBM Ž k 1 ,k 2 ,k 3 . y G
˜ Ž1. mnr R
a b cnaive 1Ž k 1 ,k 2 ,k 3 .
1 r
s 2 3
2
i c a b c Ž T L q TR . g mn Ž k 1 y k 2 .
Ž 4p .
m n
qg nr Ž k 2 y k 3 . y g m r Ž k 3 y k 1 .
1 1
q 2 6
Tr TRa ,TRb 4 TRc yTr TLa ,T Lb 4 TLc
4
Ž . p
=Tr Ž ku 2 y ku 1 . g mg ng rg 5 . Ž C.7 .
The second term on the right-hand side of the previous equation is not Bose symmetric
unless it is zero.
Notice that the problem of the inconsistency of the ‘‘naive’’ prescription can be
swept under the carpet by choosing the matter content so that TrTRa ,TRb 4TRc y TrTLa ,T Lb 4TLc
Ž' d Ra b c y d La b c ' d RyL . vanish, i.e. if there is cancellation of anomalies. Moreover, we
have compared the unambiguous Breitenlohner and Maison computation with the
‘‘naive’’ result obtained by putting the g 5 after the g r Žsee Eq. ŽC.4... If we had put it
after, say, the g m, the new difference with the result a´ la Breitenlohner and Maison has
exactly the same first term as the difference in Eq. ŽC.7. but its second term is the
second contribution in the right-hand side of Eq. ŽC.7., upon having changed cyclically
the indices and momenta of the latter. That is, the one-loop difference between two
possible ‘‘naiÕe’’ calculations is a term inÕolÕing a product of an epsilon tensor and the
bca abc a b c45
coefficient d Ry L s d RyL s d RyL . This justifies the statement that if there is cancella-
tion of anomalies there is no ambiguity in the one-loop ‘‘naiÕe’’ computation of the
three boson Õertex. In Ref. w45x, it was correctly claimed that the one-loop essential
anomaly could be also obtained with the ‘‘naive’’ prescription for the Abelian case from
a VVA triangle diagram by simply not moving around the unique g 5 which appear in
the traces, computing explicitly the dimensional integrals and contracting, at the end of
the day, the final expression with the momentum in the axial vertex. In fact, with our
notation
i
k 3r T VVA
mnr
Ž k 1 ,k 2 ,k 3 . s y 2
Tr ku 1 ku 2 g mg ng 5 , Ž C.8 .
Ž 4p .
which, with the crossed diagram, gives correctly the Abelian chiral anomaly. Of course,
also is checked explicitly that
k 1m T VVA
mnr
Ž k 1 ,k 2 ,k 3 . s k 2n T VVA
mnr
Ž k 1 ,k 2 ,k 3 . s 0.
It is widely known that a zero so, incorrect. result can be obtained by performing
Ž
shifts in the integration variables and using the cyclicity of the trace and the anticommu-
tatiÕity of g5 . Therefore, the correctness of the results above seems to be due to the fact
that in the explicit calculations no use of the anticommutativity of g 5 have been made.
5
. . . 4 enclosing indices means symmetrization and w . . . x antisymmetrization.
Unfortunately, for this restricted -the g 5 is not moved around- ‘‘naive’’ prescription
action principles cannot be usefully applied Žthe action is not invariant if the g 5 is not
anticommuted. and, more importantly, in local gauge chiral theories, such as the model
studied here, triangle diagrams VAA and AAA also appear. If one decides not to
anticommute the g 5’s before the subtraction is made, one cannot tell before hand
whether the result will satisfy the Slavnov–Taylor identities with the correct essential
breaking, unless explicit check is made order by order. And if one naively anticommutes
the g 5’s, then one will obtain ambiguous results Žwhich turns out to be safe in one-loop
calculations if there is cancellation of anomalies, but it is not clear at all what happens at
higher orders..
Of course, the Breitenlohner and Maison scheme is a consistent method and has none
of these problems. From Eq. Ž42. we get the explicit identity which involves GÃRAŽ1. A :
m
yi Ž k q p . GÃRAŽ1. bca
A nrm Ž k , p . y i c
bce
Ž 2 k q p . r gmn y Ž k q 2 p . n g rm
q Ž p y k . m gnr G˜wR;Nw
Ž1. a e
s Ax m Ž ;k q p . q c
e b a ˜ R Ž1.c e
GA A rn Ž p . q c ec a GÃRAŽ1. bnre Ž k .
y 1rg 2 p 2 PTrm Ž p . GÃw

R Ž1. b ac
;Nw s Ax nm Ž k ; p .
y 1rg 2 k 2 PTrm Ž k . GÃw

R Ž1.c a b
;Nw s Ax rm Ž p;k .
s GÃRAwŽ1.;Nw
bca
Dˆ x nr Ž k , p;0 . ,
where PTmn Ž k . is the transversal Lorentz projector g mn y Ž k m k nrk 2 ..

The diagrams without fermions satisfy the Slavnov–Taylor identity without breaking;
therefore,
yi Ž k q p . m G˜FRAŽ1. bca
A A nrm Ž k , p . q c
eb a ˜ R Ž1.c e
G F A A rn Ž p . q c ec a G˜FRAŽ1. be
A nr Ž k .
s GÃRAwŽ1.;Nw
bca
Dˆ x nr Ž k , p;0 . . Ž C.9 .
The explicit calculation of the right-hand side of the last equation gives a result which
contains spurious anomalies, i.e. which can be cancelled by adding finite counterterms to
the action, this we have already seen in Subsection 3.6, and a part which is the correct
essential anomaly:
1 1
G̃ARAwŽ1.;Nw
bca
Dˆ x nr
BM
Ž k , p;0 . s y 2 3 Ž p 2 y k 2 . c b c a gnr Ž TR q TL .
Ž 4p .
1 2
y 2 3 Ž kn kr y pn pr . c b c a Ž TR q TL .
Ž 4p .
bca
1 1 d Ry L
q 2
Tr ku pu gn gr .
Ž 4p . 3 2
Notice that even if there d Ry L s 0 there are spurious anomalies in the calculation.
´ D. Sanchez-Ruizr
For the ‘‘naive’’ prescription, by using Eq. ŽC.5. and anticommuting the g5’s, it is
possible to write
TR q TL nrm
G̃ FRAŽ1. b c a naive
A A nrm Ž k , p. s i cbca T VVV Ž k , p,y k y p .
2
bca
d RyL nrm
q T VV A Ž k , p,y k y p . .
2
Therefore, the left-hand side of Eq. ŽC.9. in the ‘‘naive’’ prescription Žanticommuting
g 5 . reads
TR q T L nrm nr nr
cbca Ž k q p . m T VVV Ž k , p,y k y p . y T VV Ž p . q T VV Ž k.4
2
bca
d RyL nrm
y i Ž k q p . m T VVA Ž k , p,y k y p . .
2
But the first term between brackets vanish in the ‘‘naive’’ prescription without incurring
into any inconsistency -no g 5 is involved. Hence, we arrive at the conclusion that
bca
d RyL nrmm
l.h.s. of Eq. Ž C.9 . s i Ž yk y p . T VV A Ž k , p,y k y p . , Ž C.10 .
2
in the ‘‘naive’’ dimensional regularization prescription Žanticommuting g 5 .. Next, one
bca
could think of applying Eq. ŽC.8. to obtain an ‘‘anomaly’’ 1 dRy L Tr w ku pu gn gr x, which
Ž4 p . 2 2
is different from the correct Breitenlohner and Maison result. Moreover, Eq. ŽC.10. was
obtained by putting the one g 5 that is left after the anticommutation process is done in
the position m. If we had put it in the position n , we would not have obtained any
anomaly at all. ŽThe ‘‘anomaly’’ obtained with the first choice for a place for the g 5 is
three times the correct Breitenlohner and Maison result and three is the number of
choices.. This shows clearly that the ‘‘naive’’ prescription cannot give consistently the
one-loop anomaly in chiral local gauge theories. Therefore, the claim that ‘‘we do not
need discuss the graphs containing the AAA triangle since these behave in the same
manner as the AVV triangle’’ of Ref. w46x seems to be not well justified.
Let us move on to computation of the one-loop four gauge field vertex. This vertex
also contributes to the anomaly. We have already shown in Subsection 3.5 that the
Breitenlohner and Maison scheme gives the correct relative factors between the anomaly
from the three gauge field vertex and from four gauge field vertex.
The one-loop 1PI four gauge field function is given by the sum of the diagram
depicted in Fig. 14 and all the diagrams which are obtained from the first by performing
all the possible permutations of three legs.
Fig. 14. One-loop 1PI Feynman diagrams with four bosonic legs.
Proceeding as for the three boson vertex case, the difference between the renormal-
ized minimal Breitenlohner and Maison result for the diagrams of Fig. 14 and the
‘‘naive’’ result defined with the by putting the g 5 after the g l – the result depends on
the position of the g 5 –
i
y 2
Ž 4p .
abcd ab dc ac d b adcb acb d adbc
2 T Lq TLqR TLqR TLqR TLqR TLqR
=
3 ½ g mn g rl 3
2
R
q3
2
q3
2
q3
2
y5
2
y5
2
abcd acb d adbc adcb ab dc ac d b
TLq R T LqR TLqR T LqR TLqR T LqR
qg m l g nr 3 q3 q3 q3 y5 y5
2 2 2 2 2 2
acb d ac d b adbc ab dc abcd adcb
TLq TLqR TLqR T LqR T LqR T LqR
qg m r g nl 3
2
R
q3
2
q3
2
q3
2
y5
2
y5
2 5
abcd ac d b adbc acb d ab dc
i TRy L TRyL TRyL TRyL TRyL
y
Ž 4p .
1
2 2
Tr g mg ng rg l g 5 ½ 2
q
2
q
2
y
2
y
2
adcb
TRyL
y
2 5 .
Notice that the first term between brackets is cyclic but that the second one is, in fact,
aw b c d x
3 TRy L , i.e. antisymmetric in the last three indices but not in the four indices, so, again,
is not cyclic.
Then, the complete difference, taking into account all diagrams, is
a , b 4 acb d adbc
i 2 TLq R c,d4 TLqR TLqR
y 2
Ž 4p . 3 ½ g mn g rl 3
2
y5
2
y5
2
a , d 4 ab dc ac d b
TLq R c,b4 TLqR T LqR
qg m l g nr 3 y5 y5
2 2 2
a ,c 4 abcd adcb
TLq R b,d4 TLqR TLqR
qg mr
g nl
3
2
y5
2
y5
2 5
i
q 2 2
1
Tr g mg ng rg l g 5 DRa b c d y DLa b c d 4 ,
Ž 4p .
a , b4 c , d4
T Lq R s Tr TLa ,T Lb 4 TLc ,T Ld 4 q Tr TRa ,TRb 4 TRc ,TRd 4 ,
DLa b c d s yi 3! T La TLw b TLc TLd x s 12 Ž d La b e c ec d q d Lac e c e d b q d La d e c eb c . s DLaw b c d x ,
and similarly DRa b c d.
Again, the first term between brackets is a local term which can be checked to be just
proportional to the Feynman rules of Tr Rq L Am A mAn An and Tr RqL Am An A mAn ;
´ D. Sanchez-Ruizr
whereas the second term is not Bose symmetric and reflects the inconsistency of the
‘‘naive’’ prescription. Also, we can use these result to show that the ambiguities in the
‘‘naiÕe’’ calculation of the one-loop four-boson Õertex is proportional to the epsilon
tensor and to the coefficient DRa b c d y DLa b c d, which vanish if there is cancellation of
anomalies. For example, the one-loop renormalized difference between the ‘‘naive’’
calculation with the g 5 after g lP L ŽR . and the ‘‘naive’’ calculation with the g 5 after
g m PL ŽR . is ir8p 2´mnrl Ž DRa b c d y DLa b c d q DRb c d a y DLb c d a ..
We have exhibited by explicit computation the many inconsistencies that plague the
computations leading to a renormalized one-loop chiral gauge theory if the ‘‘naive’’
Dimensional Regularization prescription is used. The classical action of the theory was
given in Subsection 3.1. To close this appendix we shall show that in spite of these
inconsistencies, and if the chiral theory is anomaly free- a compulsory constraint-, the
renormalized BRS invariant chiral theories obtained by means of the ‘‘naive’’ dimen-
sional regularization prescription, on the one hand, and Breitenlohner and Maison
Dimensional Regularization procedure, on the other, are equivalent at the one-loop level,
in the sense that they are related by finite BRS symmetric renormalizations of the fields
and couplings of the theory.
As we have previously compared the Breitenlohner and Maison renormalized theory
with the ‘‘naive’’ renormalized theory when no finite BRS-asymmetric counterterms
have been added to the former to restore the BRS symmetry, the comparison between
the two renormalized theories, when the BRS-asymmetric counterterms have been added
to the minimal Breitenlohner and Masion renormalized theory, is very easy.
Due to the fact that, in the situation of cancellation of anomalies, the ‘‘naive’’
minimal dimensionally renormalized theory satisfy the Slavnov–Taylor identities at
order "1, the finite counterterms relating the two renormalized theories should appear in
b-invariant combinations. And this is what nicely happen, upon inclusion of finite BRS
asymmetric counterterms, the Breitenlohner and Maison renormalized theory equals the
‘‘naive’’ renormalized theory if the coefficients – l Ž1. Ž1. Ž1. Ž1. Ž1.
g , lc , lc , l A and lv – of the
X
BRS-symmetric counterterms in Eq. Ž69. are chosen to be given by

1 1
l Ž1.
g sy 2 24
5
g2, lcŽ1.ŽX . s 2
g 2 C L ŽR . 1 q Ž a X y 1 . 125 ,
Ž 4p . Ž 4p .
l AŽ1. s lvŽ1. s 0.
This corresponds, of course, to a mass-independent multiplicative finite renormalization
of the fields and parameters of the effective action. Hence, the beta function and
anomalous dimensions of both renormalized theories are the same at the one-loop level.
The finite renormalizations we have just mentioned read
A s Aw , B s Bw , vsvw, vsvw,
rsrw, zsz w, asa w,
1
gs 1q
ž Ž 4p .
2 48
5
/
g w 2 "1 g w ,
1
g w 2 C L 1 q Ž a X y 1 . 125 "1 P L c w ,
cs 1q
ž Ž 4p .
2 /
1
g w 2 C L 1 q Ž a X y 1 . 125 "1 PR ,
csc w 1q
ž Ž 4p .
2 /
1
cXs 1q g w 2 C R 1 q Ž a X y 1 . 125 "1 PR c X w ,
ž Ž 4p .
2 /
1
cXscX w 1q g w 2 C R 1 q Ž a X y 1 . 125 "1 PL ,
ž Ž 4p .
2 /
1
g w 2 C L 1 q Ž a X y 1 . 125 "1 P L ,
L s Lw 1 q
ž Ž 4p .
2 /
1
g w 2 C L 1 q Ž a X y 1 . 125 "1 PR Lw ,
Ls 1q
ž Ž 4p .
2 /
1
g w 2 C R 1 q Ž a X y 1 . 125 "1 PR ,
R s Rw 1 q
ž Ž 4p .
2 /
1
g w 2 C R 1 q Ž a X y 1 . 125 "1 P L R w ,
Rs 1q
ž Ž 4p .
2 /
where the fields and constants with a star are the renormalized fields and constants in
the ‘‘naive’’ minimal Dimensional Regularization prescription and the fields and
constant without stars stand for the corresponding renormalized quantities in renormal-
ized theory obtained by means of the Breitenlohner and Maison scheme Žminimal
subtraction plus addition of the finite counterterms in Eq. Ž69., with l Ž1. Ž1. Ž1. Ž1.
g , lc , lc , l A
X
Ž1.
and lv set to zero .
.
References
w1x J. Ashmore, Lett. Nuovo Cimento 4 Ž1972. 289.

w2x C.G. Bollini, J.J. Giambiagi, Phys. Lett. B 40 Ž1972. 566.
w3x G. ’t Hooft, M. Veltman, Nucl. Phys. B 44 Ž1972. 189.
w4x P. Breitenlohner, D. Maison, Comm. Math. Phys. 52 Ž1977. 11.
w7x G. Bonneau, Nucl. Phys. B 167 Ž1980. 261.
w10x G. Leibbrandt, Rev. Mod. Phys. 47 Ž1975. 849.
w11x J. Collins, Renormalization ŽCambridge Univ. Press, Cambridge, 1984..
w12x G. ’t Hooft, Nucl. Phys. B 61 Ž1973. 455.
w13x G. ’t Hooft, Nucl. Phys. B 62 Ž1973. 444.
w14x O.V. Tarasov, A.A. Vladimirov, A.N. Zharkov, Phys. Lett. B 93 Ž1980..
w15x F.V. Tachkov, Phys. Lett. B 100 Ž1981. 65.
w16x D.I. Kazakov, Phys. Lett. B 133 Ž1983. 406.
w18x ¨
L. Brucher, J. Franzkowski, D. Kreimer, Oneloop 2.0, hep-thr9709209.
w19x M.E. Machacek, M.T. Vaughn, Nucl. Phys. B 222 Ž1983. 83.
´ D. Sanchez-Ruizr
w20x M.E. Machacek, M.T. Vaughn, Nucl. Phys. B 236 Ž1984. 221.
w21x I. Jack, D.R.T. Jones, Nucl. Phys. B 249 Ž1985. 472.
w22x C. Ford, I. Jack, D.R.T. Jones, Nucl. Phys. B 387 Ž1992. 373.
w23x C. Ford, D.R.T. Jones, P.W. Stephenson, M.B. Einhorn, Nucl. Phys. B 395 Ž1993. 17.
w24x T. Van Ritbergen, J.A.M. Vermaseren, Phys. Lett. B 400 Ž1997. 379.
w25x B.A. Kniehl, Int. J. Mod. Phys. A 10 Ž1995. 443, and references therein.
w26x G. Bonneau, Phys. Lett. B 96 Ž1980. 148.
w27x D.A. Akyeampong, R. Delbourgo, Nuovo Cimento A 17 Ž1973. 578.
w30x D. Kreimer, Phys. Lett. B 237 Ž1990. 59, The role of g 5 in Dimensional Regularization, hep-phr9401354.
w31x J.C. Collins, Nucl. Phys. B 92 Ž1975. 477.
w32x O. Piguet, S. Sorella, Algebraic Renormalization ŽSpringer, Berlin, 1995., and references therein.
w33x F. Brandt, Commun. Math. Phys. 190 Ž1997. 459.
w34x E. Krauss, Renormalization of the Electroweak Standard Model to all orders, hep-thr9709154, to appear
in Ann. Phys.
w35x K. Hepp, Renormalization theory in statistical mechanics and quantum field theory, in: Les Houches XX
1970 ŽGordon and Breach, New York, 1971..
w36x G. Bonneau, Int. J. Mod. Phys. A 5 Ž1990. 3831.
w37x W.A. Bardeen, R. Gastmans, B. Lautrup, Nucl. Phys. B 46 Ž1972. 319.
w38x M. Chanowitz, M. Furman, I. Hinchliffe, Nucl. Phys. B 159 Ž1979. 225.
w39x L. Alvarez-Gaume, ´ E. Witten, Nucl. Phys. B 234 Ž1984. 269.
w40x K. Wilson, Phys. Rev. D 7 Ž1973. 2924.
w41x O. Piguet, A. Rouet, Phys. Rep. 76 Ž1981. 1.
w42x ¨
J.G. Korner, N. Nasrallah, K. Schilcher, Phys. Rev. D 41 Ž1990. 888.
w43x R. Ferrari, A. Le Yaouanc, L. Oliver, J.C. Raynal, Phys. Rev. D 52 Ž1995. 3036.
w44x ¨
A. Barroso, M.A. Doncheski, H. Grotch, J.G. Korner, K. Schilcher, Phys. Lett. B 261 Ž1991. 123.
w45x S. Gottlieb, J.T. Donohue, Phys. Rev. D 20 Ž1979. 3378.
w46x B.A. Ovrut, Nucl. Phys. B 213 Ž1983. 241.
w47x M. Bos, Ann. Phys. 181 Ž1988. 197.
w48x H. Osborn, Ann. Phys. 200 Ž1990. 1.
w49x I. Jack, D.R.T. Jones, K.L. Roberts, Z. Phys. C 63 Ž1994. 151.
Spectral sum rules of the Dirac operator and partially quenched

chiral condensates
P.H. Damgaard, K. Splittorff
The Niels Bohr Institute, BlegdamsÕej 17, DK-2100 Copenhagen Ø, Denmark
Received 23 December 1999; accepted 11 January 2000
Abstract
Exploiting Virasoro constraints on the effective finite-volume partition function, we derive

generalized Leutwyler–Smilga spectral sum rules of the Dirac operator to high order. By
introducing NÕ fermion species of equal masses, we next use the Virasoro constraints to compute
two Žlow-mass and large-mass. expansions of the partially quenched chiral condensate through the
replica method of letting NÕ ™ 0. The low-mass expansion can only be pushed to a certain finite
order due to de Wit–’t Hooft poles, but the large-mass expansion can be carried through to
arbitrarily high order. Results agree exactly with earlier results obtained through both Random
Matrix Theory and the supersymmetric method. q 2000 Elsevier Science B.V. All rights reserved.
1. Introduction
The exact spectral sum rules for the Dirac operator in finite-volume gauge theories
derived by Leutwyler and Smilga w1x and later by Smilga and Verbaarschot w2x have
sparked a huge activity in the field. It has turned out that there is an exact relation
between universality classes in Random Matrix Theories and the three main ways in
which chiral symmetry can break spontaneously, depending on the gauge group and the
representation of the fermion fields w3–10x. The Random Matrix Theory formulation
brings a completely different set of techniques into play, and much has been learned
from this approach. One of the main efforts recently has nevertheless been to see how all
the results obtained from Random Matrix Theory can be derived directly from the
effective partition function. One of the first steps in this direction was a set of very
compact relations that expressed microscopic spectral correlators Žincluding the micro-
scopic spectral density itself. directly in terms of effective partition functions with
additional quark species w11–13x. More recently, many of these results have been
derived directly from the effective field theory w14–16x. The idea has been to compute
the Žpartially. quenched chiral condensate and higher chiral susceptibilities through an
appropriately extended effective Lagrangian, and from this derive the microscopic
PII: S 0 5 5 0 - 3 2 1 3 Ž 0 0 . 0 0 0 2 2 - 5
P.H. Damgaard, K. Splittorffr Nuclear Physics B 572 (2000) 478–498 479
spectral correlators. As the technique of Refs. w14–16x makes use of additional quark
species, of which half are bosonic and the others are fermionic, this extends the flavour
symmetry to a super Lie group at intermediate steps. For this reason it is commonly
known as the supersymmetric method Žalthough it, as applied, has nothing to do with
space-time supersymmetry..
The purpose of this paper is to explore another technique that can be used to derive
the partially quenched chiral condensate. This is based on the so-called replica method1
in which one extends the theory with NÕ additional fermionic species, and takes the
limit NÕ ™ 0 at the end of the calculation. The replica method is known in other
contexts to be often problematical w17x Žsee, however, also the recent discussion on this
issue w18–22x., but we shall find no fundamental difficulties with this technique in the
present context.
Let us first outline how the replica method can be used to compute partially quenched
averages in QCD. Consider the QCD partition function with Nf physical quark fields
and NÕ additional quark fields of degenerate masses m Õ . We restrict ourselves to gauge
field sectors of fixed topological charge n , which we for simplicity from now on take to
be non-negative:
Nf Nf
NÕ
ZnŽ N fqN Õ . s
ž /
Ł mnf
fs1
m ÕN Õ n w dA x n detX Ž iDu y m Õ .
H Ł detX Ž iDu y m f . eyS
fs1
YM w Ax
Ž 1.1 .
,
where the determinants are taken over non-zero-modes only. This can be considered as
an average over gauge fields Žand physical fermions, here already integrated out. of NÕ
identical replicas of the fermionic partition function
4
ZÕ ' d c d c exp
H Hd x c Ž iDu y m Õ . c , Ž 1.2 .
in the sense that

Nf Nf
NÕ
ZnŽ N fqN Õ . s
ž /
Ł mnf
fs1
H w dA x Ł
fs1
X
det Ž iDu y m . w Z x
n f Õ eyS YM . Ž 1.3 .
In condensed matter physics this Žunnormalized. average is conventionally denoted by a

bar:
N
ZnŽ N fqN Õ . s w ZÕ x Õ . Ž 1.4 .
The subscript Õ will throughout denote ‘‘ valence’’, and the additional quarks are thus
valence quarks, a terminology borrowed from lattice gauge theory. Obviously, if we let
NÕ s 0 we simply recover the original QCD partition function. For Nf q NÕ not too large,
the theory is presumed to undergo, in the chiral limit of massless quarks, spontaneous
chiral symmetry breaking according to the usual pattern SUŽ Nf q NÕ . = SUŽ Nf q NÕ . ™
SUŽ Nf q NÕ .. Taking the limit NÕ ™ 0 must therefore incorporate some kind of analytic
continuation in n of the Lie group SUŽ n..
1
Or replica ‘‘trick’’, a name that suggests magic and trickery. We prefer the more neutral terminology.
480 P.H. Damgaard, K. Splittorffr Nuclear Physics B 572 (2000) 478–498
For any fixed NÕ we can compute observables in the extended theory by viewing the
partition function as the generating function of the n-point functions of cc . We shall
here focus on just the chiral condensate itself, and only on the chiral condensate of the
additional fermionic copies. As we will take NÕ ™ 0, this becomes the partially
quenched condensate in the theory with Nf physical fermions. We define this partially
quenched Žmass-dependent. chiral condensate by
Sn Ž m Õ , m 4 . 1 E
' lim ln ZnŽ N fqN Õ . , Ž 1.5 .
S NÕ™0 NÕ Em Õ
where S is the physical infinite-volume chiral condensate in the theory with Nf
fermions, m j ' m j VS and similarly m Õ ' m Õ VS . Treating NÕ as a parameter that is not
restricted to be integer Žwhich is permitted by the representation of the partition function
in Eq. Ž1.1.., we can make a Taylor expansion in NÕ :
E
ZnŽ N fqN Õ . s ZnŽ N f . qNÕ ZnŽ N fqN Õ . q... Ž 1.6 .
E NÕ N Õ s0
and thus write, equivalently,2

Sn Ž m Õ , m 4 . y1 E E
s ZnŽ N f . ZnŽ N fqN Õ . . Ž 1.7 .
S E NÕ Em Õ N Õ s0
The partially quenched chiral condensate is a particularly convenient quantity for lattice
gauge theory simulations, and comparisons with theory have already been made both in
the topologically trivial sector w23x and, very recently, in sectors of non-vanishing
topological charge n w24–27x. We note here that higher-order partially quenched chiral
susceptibilities, which are also readily studied by Monte Carlo simulations, can be
derived in a completely analogous manner. For a chiral k-point function one simply
needs to introduce NÕ1 , . . . , NÕ k sets of different additional species Žeach of NÕ i ’th
degenerate masses m Õ i ., and take the combined limit of all NÕ i ™ 0 in the end, after
having performed the required differentiations.
In this paper we will show how the replica method can be applied to the effective
finite-volume partition function, and in this way we shall derive analytical series
expansions for the partially quenched chiral condensate. As a by-product of the analysis
we will get, for free, generalized Leutwyler–Smilga sum rules to very high order. The
plan of the paper is as follows. In Section 2 we review the technique for deriving
high-order expansions of the effective partition function on the basis of an iterative
sequence of partition function constraints. From this expansion we immediately derive a
long list of generalized spectral sum rules of the Dirac operator. In Section 3 we turn to
the replica method, where we first exploit the same partition function constraints to
derive a small-mass expansion for the partially quenched chiral condensate. We also
show how a different set of partition function constraints can be used to derive a very
high order expansion in large masses. In Section 4 we point out that the usual
Leutwyler–Smilga spectral sum rules, which by definition are taken with respect to the
2
One can explicitly check that the two derivatives commute in our case.
massless theory, have a natural generalization where the spectral sums are taken with
respect to the theory with massive fermions. We derive a series of such massive spectral
sum rules, perturbatively expanded in the physical masses. In Section 5 we discuss some
generalizations to different patterns of spontaneous chiral symmetry breaking, and point
out why the present series expansions cannot be used to derive the microscopic spectral
density of the Dirac operator itself. In Section 6 we briefly list the main conclusions.
2. Spectral sum rules from Virasoro constraints
Spectral sum rules of the Dirac operator are derived by comparing the full QCD
partition function with the effective low-energy partition function. In the large-volume
scaling region V < 1rmp4 , this effective partition function is to leading order w1x
n 1
ZnŽ N f . s HUgU Ž N .dU Ž detU . exp 2 Tr Ž M U † q U M † . , Ž 2.1 .
f
where M is the quark mass matrix rescaled by the space-time volume V times the value
S of the chiral condensate in the chiral limit Žand at u s 0.. We will in fact always take
this to be diagonal of entries m i s m i S V. In this section we shall restrict ourselves to Nf
Žphysical. fermions with no additional quark species, and NÕ will thus here be taken to
be zero from the very beginning. The integration in Eq. Ž2.1. is over the coset of chiral
symmetry breaking, here extended from SUŽ Nf . to UŽ Nf . due to the projection on a
sector of fixed topological charge n . The integral is known in closed form w28,29x,
det A Ž m 4 .
ZnŽ N f . Ž m 4 . s Ž 2.2 .
D Ž m2 4 .
where the Nf = Nf matrix A is given by
A Ž m 4 . i j s m ijy1 Inqjy1 Ž m i . , Ž 2.3 .
InŽ x . is a modified Bessel function, and the denominator is given by the Vandermonde
determinant of rescaled masses:
Nf
2 jy1
DŽ m 4 . ' Ł Ž m2i y m2j . s det i , j Ž m2i . . Ž 2.4 .
i)j
Although the effective partition function is known explicitly, Eq. Ž2.2. is not in a
form suitable for the derivation of spectral sum rules of the Dirac operator. To get it into
such a form, it is convenient to start with the case n s 0. In that case the left and right
invariance of the Haar measure under unitary transformations shows that the partition
function depends only on the combination M † M . A crucial observation of Ref. w30,31x
is that the partition function in fact satisfies an infinite set of constraint equations that
can be used to determine it uniquely. The precise form of these constraints depend on
the chosen variables, and one can use two different sets:
1 k
tq
k' Tr Ž M † M . , k s 1,2, . . . , Ž 2.5 .
4kk
which is a suitable set for small-mass expansions, and

2 2 kq1
ty
k 'y Tr Ž Ž M † M . yŽ 2 kq1 . r2 . , k s 0,1, . . . , Ž 2.6 .
2kq1
which is a suitable set for large-mass expansions Žthe coefficients in front have been
inserted for later convenience.. Spectral sum rules of the Dirac operator are derived by
means of a small-mass expansion, and we shall therefore begin with the description in
terms of the variables 3 tq
k.
Defining, for n 0 1,
E ` E ny1 E2
Lnq ' Nf q Ý ktq
k q Ý , Ž 2.7 .
E tq
n ks0 E tq
nqk ks1 E tq q
k E t nyk
the partition function Ž2.10. is found to satisfy w31x

LnqZ 0Ž N f . s dn ,1 Z0Ž N f . . Ž 2.8 .
These constraints are consistent in the sense that they satisfy the classical Virasoro
algebra
Lnq , Lmq s Ž n y m . Lnqm
q
, Ž 2.9 .
so that they do not generate new constraints beyond the infinite tower of The Lnq ’s.
constraints are also complete: They determine the partition function uniquely, given the
boundary condition that Z0Ž N f . s 1 for all tq
k s 0.
The small-mass power series for Z0Ž N f . can conveniently be chosen w31x
k 1 tq q
1 . . . k M tM
Z0Ž N f . s 1 q Ý Ý CN f Ž k 4 . , Ž 2.10 .
M 1(k 1 . . . (k M Ž k 1 q . . . qk M . !
where use has been made of the boundary condition to determine the zeroth order
coefficient.
The coefficients CN fŽ k 4. are determined uniquely by the Virasoro constraints Ž2.8..
They factorize into a polynomial part, CˆN fŽ k 4., and a singular part w31x:
K Ž k 4.y1
y1
CN f Ž k 4 . ' CˆN f Ž k 4 . Ł Ž Nf2 y l 2 . , Ž 2.11 .
ls0
ˆ in what follows.
where K Ž k 4. ' Ý i k i . The singular parts will play an important role
Originally noted by de Wit and ’t Hooft in the context of strong-coupling expansions in
lattice gauge theory w32x, these poles are in fact innocuous as far as the expansion of the
partition function itself is concerned w33x. What happens is that for Žinteger and
non-vanishing!. values of N the poles only occur in terms that are linearly dependent,
and therefore should be combined. After combining them, all poles cancel. This
phenomenon becomes obvious when we consider the explicit expansion below.
The coefficients CˆN fŽ k 4. have already been computed recursively in Ref. w31x to high
order, enough to determine the partition function to 8th order in the masses. The
3
Since M † M is a hermitian matrix of size Nf = Nf , only Nf of these variables are independent; this is of
no consequence for the subsequent analysis.
restriction to the n s 0 sector can easily be lifted due to flaÕour-topology duality, which
states that
n
ZnŽ N f . Ž M , M † . s det Ž M . Z0Ž N fq n . Ž M M † . Ž 2.12 .
where Z0Ž N fqn . Ž M M † . is the n s 0 partition function extended with n massless quarks
w35x. Introducing N ' Nf q NÕ q n Žhere with NÕ s 0. we have
1 1 2
ž
ZnŽ N f . s det n Ž M . 1 q
4N
Tr M † M q
32 Ž N y 1 . 2 Ž Tr Ž M †M . .
1 2 1 3
y 2
Tr Ž M † M . q Tr Ž M † M .
ž /
32 N Ž N y 1 . 96 N Ž N y 1 . Ž N 2 y 4 .
2
1 2
y Tr Ž M † M . Tr Ž M † M .ž /
128 Ž N y 1 . Ž N 2 y 4 .
2
N 2y2 3
q 2 2 Ž Tr Ž M †M . .
384N Ž N y 1 . Ž N y 4 .
5 4
y Tr Ž M † M .
ž /
1024N Ž N y 1 . Ž N 2 y 4 . Ž N 2 y 9 .
2
2 N 2y3 3
q 2 2 2 2
Tr Ž M † M . Tr Ž M † M . ž /
768 N Ž N y 1 . Ž N y 4 . Ž N y 9 .
N 2q6 † 2 2
q 2 2
2048 N Ž N y 1 . Ž N y 4 . Ž N 2 2
y 9.
ž Tr Ž M M. /
1 2
y 2 2 Ž Tr Ž M † M . . Tr Ž Ž M † M 2 . .
1024N Ž N y 1 . Ž N y 9 .
N4y8 N 2q6 4
q
6144N 2 Ž N 2 y 1 . Ž N 2 y 4 . Ž N 2 y 9 .
Ž Tr Ž M †M . . q . . . .
/ Ž 2.13 .
We have checked that this expansion is correct to the order given by expanding, for
fixed Nf , the closed expression Ž2.2. up to that order. ŽNote that it is not at all simple to
rearrange the closed expression Ž2.2. in terms of an expansion in just powers of traces of
the mass matrix; the rearrangements are different for each value of Nf .. In this
expansion we directly verify the cancellation of de Wit–’t Hooft poles for any of the
finite, integer, values involved.
With this high-order expansion of the effective partition function, we can now easily
derive a long series of spectral sum rules, following the method of Leutwyler and
Smilga w1x. Because the method is described in detail in Ref. w1x, we shall not give
details of how these sum rules are extracted, but only quote the results. We rescale as
usual the eigenvalues l i of the Dirac operator according to z i ' l i S V. Expanding the
original partition function Ž1.1. around the massless theory, and comparing with the
above expansion up to 8th order, we then find
1 1 1
¦Ý ;
zn)0 z n
2
s
4 N
, Ž 2.14 .
1 1 1
¦Ý ;
zn)0 z n
4
s
16 N Ž N 2 y 1 .
, Ž 2.15 .
1 1 1
¦Ý ;
zn)0 z n
6
s
32 N Ž N y 1 . Ž N 2 y 4 .
2
, Ž 2.16 .
1 5 1
¦Ý ;
zn)0 z n
8
s
256 N Ž N y 1 . Ž N 2 y 4 . Ž N 2 y 9 .
2
. Ž 2.17 .
2
1 1 1
¦ž Ý / ;
zn)0 z n
2
3
s
16 N y 1 2
, Ž 2.18 .
1 1 N 2y2
¦ž Ý / ;
zn)0 zn2
2
s
64 N Ž N 2 y 1 . Ž N 2 y 4 .
, Ž 2.19 .
1 1 N 2q6
¦ž Ý / ;
zn)0 zn4
s
256 N 2 Ž N 2 y 1 . Ž N 2 y 4 . Ž N 2 y 9 .
, Ž 2.20 .
1 1 1
¦Ý ; 2 4
zm , z n)0 z m z n
s
64 Ž N y 1 . Ž N 2 y 4 .
2
, Ž 2.21 .
1 1 2 N 2y3
¦Ý ;
zm , z n)0 zm2 zn6
s
256 N 2 Ž N 2 y 1 . Ž N 2 y 4 . Ž N 2 y 9 .
, Ž 2.22 .
1 1 1
¦Ý ; 2 2 4
z k , z m , z n)0 z k z m z n
s
256 N Ž N y 1 . Ž N 2 y 9 .
2
, Ž 2.23 .
1 1 N4y8 N 2q6
¦ Ý ;
z k , z l , z m , z n)0 z k2z l 2zm2 zn2
s
256 N 2 Ž N 2 y 1 . Ž N 2 y 4 . Ž N 2 y 9 .
. Ž 2.24 .
Of these, the relations Ž2.14., Ž2.15. and Ž2.18. have been derived previously w1x Žsee
also Ref. w34x.. We note that the DeWit–‘t Hooft poles make these sum rules divergent
for an increasingly larger number of integer N-values. This is due to infrared diver-
gences near z ; 0. Just as these poles cancel among different terms in the small-mass
expansion, one can form analogous combinations of sum rules that are infrared finite.
For example, the de Wit–’t Hooft poles at N s 1 associated with the terms of order m4
in Eq. Ž2.13. cancel by combining the third and fourth terms in that expansion. This
directly translates into a cancellation between two eigenvalue sum rules from our list,
2
1 1 1 1
¦ž / ; ¦ ;
Ý
zn)0 z n
2
y Ý
z n)0 z n
4
s
16 N Ž N q 1 .
, Ž 2.25 .
a cancellation that was already noticed by Leutwyler and Smilga w1x. Similarly, while
sum rules Ž2.16., Ž2.19. and Ž2.21. are individually singular at both N s 1 and N s 2,
the combination
3
1 1 1 1 1
2¦ Ý ; ¦ž Ý / ; ¦ Ý ;
zn)0 z n
6
q
z n)0 z n
2
y3 2 4
z m , z n)0 z m z n
s
64 N Ž N q 1 . Ž N q 2 .
Ž 2.26 .
6.
is finite Žthis is six times the O Ž1rl contribution to the partition function.. A similar
phenomenon persists to all orders.
The massless spectral sum rules Ž2.14.-Ž2.24. can of course be compared with the
chiral Random Matrix Theory result. This is done by averaging the inverse moments
according to the universal massless microscopic spectral density w4–7x
<z < 2
rSŽ N f , n . Ž z . s JN fq n Ž z . y JN fq ny1 Ž z . JN fq nq1 Ž z . , Ž 2.27 .
2
e.g.,
1 ` rS Ž z .
¦ ; Ý
zn)0 z n
2k
s Hy`dz z 2k
. Ž 2.28 .
We have explicitly checked a number of the above sum rules, always finding completely
agreement with the Random Matrix Theory results.
3. The partially quenched chiral condensate

Let us now extend the effective theory with NÕ valence quark fields. We are
interested in obtaining the partially quenched chiral condensate, not higher order
susceptibilities, and therefore take the valence quark masses as degenerate and of size
m Õ s m Õ S V. As mentioned above, the extended partition function is also the generating
functional for the n-point functions of cc Žquenched or unquenched.; the partially
quenched chiral condensate is obtained through the replica method. Since the partition
function for n s 0 only depends on M † M and the valence quarks have degenerate
masses it is natural to introduce a mass matrix of the physical quarks M so that
Tr Ž M † M . s Tr Ž M 2 . q NÕ m2Õ . Ž 3.1 .
In short, for the definition of the partially quenched chiral condensate we can take
Sn Ž m Õ , m 4 . 1 E
' lim ln ZnŽ N fqN Õ . . Ž 3.2 .
S NÕ™0 NÕ Em Õ
The subscript on S refers to the topological sector, as usual. As we have seen above, it
is not particularly helpful to know the exact analytical expression for the Žextended.
effective partition function ZnŽ N fqN Õ . if one wishes to derive spectral sum rules for the
Dirac operator. There a small-mass expansion is required by definition. As for finding
the partially quenched chiral condensate using the expression Ž3.2., no expansion in
either large or small masses is required in principle. However, the closed analytical
formula Ž2.2. is unfortunately not directly suitable from our point of view. This is
because the NÕ-dependence enters in a quite non-trivial way through the size of the
matrix whose determinant needs to be taken. Because there is no simple extension of the
analytical expression Ž2.2. outside integer values of NÕ , we restrict ourselves here to

series expansions. We shall in fact be able to carry such expansions through in both the
large-mass and small-mass regions.
3.1. The small-mass expansion
We begin with the small-mass expansion, because here we already have the needed
expansion at hand Žsee Eq. Ž2.13... Our first observation is that NÕ enters in a manner
that allows for an analytic continuation once we make use of Eq. Ž3.1.. We can thus
proceed with the replica method.
Inserting the expansion Ž2.13. into Ž3.2. we obtain the partially quenched chiral
condensate to 7 th order in the masses Žtruncating the expansion at this order is a matter
of choice; we do it because it corresponds to consistently expanding both in the valence
quark mass m Õ and the physical fermion masses m i to the same order.:
2
Sn Ž m Õ , m 4 . n 1 Tr M 2 Ž Tr M 2 .
s q mÕ q q
S mÕ 2N 8 N 2 Ž N 2 y 1. 8 N 3 Ž N 2 y 1. Ž N 2 y 4.
Tr M 4
y
16 N 2 Ž N 2 y 1 . Ž N 2 y 4 .
15Tr M 6
q
384N 2 Ž N 2 y 1 . Ž N 2 y 4 . Ž N 2 y 9 .
3
3 Ž 2 N 2 y 3 . Ž Tr M 2 .
q 2
32 N 4 Ž N 2 y 1 . Ž N 2 y 4 . Ž N 2 y 9 .
3 Ž 3 y 2 N 2 . Tr M 2 Tr M 4
q 2
q...
32 N 3 Ž N 2 y 1 . Ž N 2 y 4 . Ž N 2 y 9 .
1 Tr M 2
y m3Õ q
8 N Ž N 2 y 1. 8 N 2 Ž N 2 y 1. Ž N 2 y 4.
2
Ž 6 N 2 y 9 . Ž Tr M 2 .
q 2
32 N 3 Ž N 2 y 1 . Ž N 2 y 4 . Ž N 2 y 9 .
3 Ž 3 N 2 y 7 . Tr M 4
y 2
q...
128 N 2 Ž N 2 y 1 . Ž N 2 y 4 . Ž N 2 y 9 .
1
q m5Õ
16 N Ž N y 1 . Ž N 2 y 4 .
2
15Tr M 2
q q...
128 N 2 Ž N 2 y 1 . Ž N 2 y 4 . Ž N 2 y 9 .
5
y m7Õ q... , Ž 3.3 .
128 N Ž N y 1 . Ž N 2 y 4 . Ž N 2 y 9 .
2
where N s Nf q n Žsince the NÕ ™ 0 limit has already been taken.. Note that the
‘‘topological’’ term nrm Õ comes out trivially from the Ždet M . n-factor in front, as it
should.
Is this expansion correct? There are by now two independent ways to check it:
Comparing with Random Matrix Theory, and comparing with the supersymmetric
method of partially quenched Lagrangians. Before proceeding with such comparisons,
let us however first note an obvious feature of the above expansion. In contrast to the
small-mass expansion of the partition function itself Ž2.13., there is clearly not going to
be a cancellation of the de Wit–’t Hooft poles in the expansion of the partially quenched
chiral condensate. This may appear surprising, since the chiral condensate is just
determined from the partition function through, for example, a formula like Ž1.5.. If the
de Wit–’t Hooft poles cancel in the partition function, how can they not cancel after
taking the m Õ-derivative and the limit NÕ ™ 0 as in Eq. Ž1.5.? The answer is simple. To
be able to take the quenched limit of NÕ ™ 0 we must not cross any singularities that can
obstruct the analytical continuation. This is, however, precisely what happens for those
terms in the expansion Ž3.3. that, for given N, are singular. To give an example, let us
consider the m3Õ-term in the expansion Ž3.3.. For simplicity, let us also consider the case
where all physical masses vanish. Then there is only one term:
m3Õ
. Ž 3.4 .
8 N Ž N 2 y 1.
This term is singular for N s 1 Žand, of course, also in the more trivial case of N s 0.,
and therefore cannot possibly represent the partially quenched chiral condensate to order
m3Õ in the N s 1 theory. Let us therefore trace what happens if we set Nf q n s 1 from
the outset. The only term in the expansion Ž2.13. that gives an O Ž m3Õ . contribution to
Sn Ž m Õ . in this massless case is Ž N ' Nf q NÕ q n .
1 2 NÕ m4Õ m4Õ
Tr Ž M † M . s s , Ž 3.5 .
32 N Ž N 2 y 1 . 32 N Ž N 2 y 1 . 32 Ž NÕ q 1 . Ž NÕ q 2 .
which precisely in this Nf q n s 1 case is not linear in NÕ . The factor 1rNÕ in the
expression for the quenched condensate Ž3.2. will in this case remain uncancelled, and
an analytic continuation to NÕ s 0 is prohibited by the singularity. The term Ž3.4. in the
expansion Ž3.3. is therefore incorrect precisely in the case Nf q n s 1 Žbut it is valid for
all Nf q n ) 1.. This holds also for all higher orders in the expansion. The small-mass
expansion for Sn is thus given to a finite number of terms. For example, in the massless
Nf q n s 3 theory we obtain
Sn Ž m Õ , m 4 . n mÕ m3Õ m5Õ
s q y q q..., Nf q n s 3, Ž 3.6 .
S mÕ 6 192 1920
where the higher-order terms in m Õ cannot be purely power-like. The same phenomenon
occurs for larger integer values of N, where we encounter the singularity at higher
orders. We thus conclude that the small-mass expansion Ž3.3. is valid up to the first term
where a pole appears. The obstruction to an all-order expansion of Sn Ž m Õ , m i 4. is from
this point of view uncancelled de Wit–’t Hooft poles. The expansion can of course be
pushed to very high orders by considering Nf q n sufficiently large. We stress again that
the expansion is perfectly meaningful up to the order at which the first uncancelled de
Wit–’t Hooft pole appears, and that it is completely understood why the expansion
cannot be pushed beyond this order. It is not a peculiar and unwanted artifact of the
replica method, but a real feature of the small-mass expansion of the partially quenched
chiral condensate.
With this caveat in mind, we can now check the expansion Ž3.3. against both the
supersymmetric method and results from Random Matrix Theory. For instance, both
methods yield a fully quenched chiral condensate of the form w14–16,23x
Sn Ž m Õ . n
s m Õ In Ž m Õ . Kn Ž m Õ . q Inq1 Ž m Õ . Kny1 Ž m Õ . q . Ž 3.7 .
S mÕ
where InŽ x . and K nŽ x . are modified Bessel functions. Looking at the small-m Õ
expansions of K nŽ m Õ ., one notices that they are only purely power-like up to order
n y 2 Žafter which a logarithmic term must be included.. The small-mass expansion of
the closed expression Ž3.7. therefore cannot possibly match to all orders the purely
power-like expansion that would seem to follow from the small-mass expansion of the
partition function in Eq. Ž2.13.. And we find that the agreement of the two expansions is
exact precisely up to the order at which our small-expansion Ž3.3. ceases to be valid.
The appearance, at higher orders, of logarithmic terms in the expansion explicitly
confirms the statement above: there is simply no power-law small-mass expansion of the
chiral condensate beyond that given order. It is quite remarkable that the normally
cancelling de Wit–’t Hooft poles precisely obstruct the small-mass expansion at just the
right order.
We have also checked that the double-expansions in both m Õ and physical fermion
masses m i are correct precisely up to the order at which Eq. Ž3.3. is valid. We have done
this by comparing the small-mass expansions of the partially quenched chiral condensate
in the Nf s 1 theory with the expansions of the closed expression
Sn Ž m Õ , m .
s m Õ Inq1 Ž m Õ . Knq1 Ž m Õ . q Inq2 Ž m Õ . Kn Ž m Õ .
S
n Kn Ž m Õ . m Õ In Ž m Õ . Inq1 Ž m . y m In Ž m . Inq1 Ž m Õ .
q q2m ,
mÕ In Ž m . m2Õ y m2
Ž 3.8 .
which has been derived by the supersymmetric method w14–16x Žand which can also
easily been seen to agree with the Random Matrix Theory result.. Again we find perfect
agreement with all terms up to the order at which our expansion Ž3.3. ceases to be valid,
i.e. up to the first uncancelled de Wit–’t Hooft pole.
3.2. The large-mass expansion
Having succeeded in deriving the low-mass expansion for the partially quenched
chiral condensate, we next turn to the opposite expansion, i.e. for large masses. The
advantage of this expansion is that it can be pushed to arbitrarily high order, for any
value of Nf .
Recall that the effective partition function Ž2.1. only depends on the combination
M † M for n s 0. This combination is hermitian and there are thus only Nf degrees of
freedom in Z0Ž N f ., represented for instance by the eigenvalues m 2i of M † M . In terms of
m i the expansion variables of Eq. Ž2.6. read
Nf
2 2 kq1 kq1.
ty
k 'y Ý myŽ2
i . Ž 3.9 .
2kq1 is1
The idea is now to find suitable partition function constraints that will enable us to
solve for the partition function in an expansion in the variables ty k . This is not
completely straightforward, but as shown first by Gross and Newman w30x, it is possible
to recover a complete set of Virasoro constraints in the variables ty
k if one first extracts
a simple prefactor from the partition function. The factorization is as follows w30x Žsee
also Ref. w31x for the generalization to other types of matrix integrals.:
Nf
y1 r2
Ým b
Z0Ž N f . s Ł Ž 12 m a q 12 m b . e b
Y0Ž N f . Ž ty
k 4 .. Ž 3.10 .
a, b
The exponential prefactor indicates that an expansion of the partition function entirely in
terms of the ty k ’s simply is not possible. This is also obvious if we consider the original
group integral in a saddle-point approximation around large M : the leading behaviour is
indeed exponentially large. Moreover, one suspects from the form Ž3.10. which is
essentially a saddle-point expansion that the resulting series expansion we will derive
below is at best asymptotic. This is in contrast to the previously considered small-mass
expansion, which we believe is convergent.
The remaining factor Y0Ž N f . Ž ty
k
4. in Eq. Ž3.10. turns out to be annihilated by an
infinite set of Virasoro operators. Explicitly, by defining
` E 1 E
L 0y ' Ý Ž k q 12 . tyk y
q q y, Ž 3.11 .
ks0 E tk 16 E t0
` E 1 n E2 E
Lny ' Ý Ž k q 12 . tyk E ty q Ý y y
q
E ty
, n01 Ž 3.12 .
ks0 kq n 4 ks1 E t ky1 E t nyk n
the function Y0Ž N f . Ž ty

k
4. of Ž3.10. is found to satisfy w30x
LnyY 0Ž N f . s 0, n 0 0. Ž 3.13 .
One readily verifies that these constraints indeed also fulfill the commutation relations
of classical Virasoro generators Ž2.9., and thus form a consistent set of constraints.
These constraints are also complete in that they determine the partition function
uniquely, given the boundary condition Y0Ž N f . s 1 for ty
k s 0, i.e. in the limit of infinite
masses. While these constraint equations were already established in Ref. w30x, they have
not previously been used to derive a systematic expansion for the partition function. We
do this by copying the procedure of the small-mass expansion. That is, we expand the
unknown function Y0Ž N f . Ž ty
k
4. as follows:
`
Y0Ž N f . Ž ty
k 4 . '1q Ý c k tyk q Ý c k 1 , k 2 ty y
k1 t k 2 q . . . Ž 3.14 .
ks0 0(k 1(k 2
We next solve for the coefficients c k 1 , . . . , k n iteratively. In fact, this is in many ways
simpler than in the small-mass expansion in that we in this case easily can derive closed
analytical expressions to all orders. For example, we find the following simple formula
for the first strings of coefficients:
n y1 8 m Ž 2 k q 1 . q 8 j q 1
c 0,0, . . . ,0, k , . . . , k s
nq1 m
ž Ł
js1 jq1 16 / c 0, k , . . . , k , .
m
Ž 3.15 .
Using this formula, and others similar, it is straightforward to derive very high order
expansions for Y0Ž N f . Ž ty
k
4., and therefore also for the partition function itself. For
example, to 7th order in the masses we find a rather formidable-looking expression Žthat
is easily pushed to much higher orders.:
Nf
y1r2 1 y1 r2
Z0Ž N f . s Ł Ž 12 m a q 12 m b . exp Ý mb 1q
8
Tr Ž M † M .
a, b b
9 † y1 r2 2 51 † y1r2 3
q
128
ž Tr Ž M M. / q
1024
ž Tr Ž M M. /
3 y3 r2 1275 y1r2 3
q Tr Ž M † M . q ž Tr Ž M †
M. /
128 32768
75 † y1 r2 † y3r2 8415 † y1r2 5
q
1024
ž Tr Ž M M. / ž Tr Ž M M. / q 262144 ž Tr Ž M M. /
2475 † y1 r2 2 † y3r2
q
16384
ž Tr Ž M M. / ž Tr Ž M M. /
45 y5 r2 115005 y1r2 6
q Tr Ž M † M . q ž Tr Ž M †
M. /
1024 4194304
33825 † y1 r2 3 † y3r2
q
131072
6075 † y3 r2 2 1845 † y1r2 † y5r2
q
98304
ž Tr Ž M M. / q
8192
805035 † y1 r2 7 1657425 † y1r2 4
q
33554432
ž Tr Ž M M. / q
4194304
ž Tr Ž M M. /
y3 r2 99225 y1r2 y3r2 2
= Tr Ž M † M .
ž / q 262144 ž Tr Ž M †
M. / ž Tr Ž M †
M. /
90405 † y1 r2 2 † y5r2
q
131072
7875 y7 r2
q Tr Ž M † M . q... Ž 3.16 .
32768
In writing Ž2.6. and Ž3.9. we have implicitly assumed that M † M is invertible. This
means that there must be no zero-eigenvalues in the mass matrix, and we thus by default
cannot consider any number of massless fermions in this large-mass expansion. In
particular, the concept of flavour-topology duality, by which a topological charge n is
implemented by adding n massless fermions to the n s 0 partition function, cannot be
applied in the framework of a large-mass expansion. Because the Virasoro constraints
Ž3.11. and Ž3.12. apply only to the case n s 0 we know of no analogous way to
implement this large-mass expansion outside of the n s 0 sector.
Introducing NÕ mass-degenerate valence quarks into the large-mass expansion of the
effective partition function, and noting that in this case
y1 r2 y1
Tr Ž M † M . s Tr Ž M . q NÕ my1 , Ž 3.17 .
we again observe that NÕ can be regarded as a continuous parameter. The replica
approach can therefore again be applied.
Using the definition Ž3.2. and a good handful of simple algebra we find the partially
quenched chiral condensate for large masses:
Nf 2
S0 Ž mÕ , m4 . 1 1 1 1 1 1 1
S
s1y Ý
is1 mi q mÕ ž / y
m2Õ 8
q
8
Tr
M
q
8 ž ž // Tr
M
3 4
9 1 1 1 1 1
q
128
Trž / ž ž // ž
M3
q
8
Tr
M
q
8
Tr ž //
M
5
9 1 1 1 1
q
32
Tr ž / ž / ž ž //
M3
Tr
M
q
8
Tr
M
2
45 1 1 225 1
q
64
Tr ž /ž ž //
M 3 ž Tr
M
q
1024
Tr
M5 / q...
2 3
1 9 27 1 27 1 45 1
y
m4Õ 128 ž / ž ž
q
128
Tr
M
q
64
Tr
M // q
64ž ž //Tr
M
189 1 1 225 1125 1

q
512
Trž / M 3
q... y
m6Õ 1024
q
1024
Tr ž / M
q... .
Ž 3.18 .
This expansion agrees with the asymptotic expansion of the analytical expressions for
S Ž m Õ , m4. found in w14–16x for Nf s 0 and Nf s 1.
4. Massive spectral sum rules
The spectral sum rules presented in Section 2 concern inverse moments of the Dirac
eigenvalues averaged with respect to the massless theory. These massless spectral sum
rules are, however, only special cases of massive spectral sum rules obtained by
averaging with respect to the massive theory. In fact, massive spectral sum rules appear
quite naturally in lattice simulations. Such massive spectral sum rules are conventionally
taken w3,36x to be of the form
` 1
H0 dz r Ž z ; m 4 . Łj
S i nj ,
Ž z 2 q m2j .
for given integer n j ’s. Here masses and Dirac operator eigenvalues are treated on equal
footing in the denominators. However, we could equally well define massive spectral
sum rules by
1 ` rS Ž z ; m i 4 .
¦Ý ;
z j)0 zj 2n
s H0 dz z 2n
, Ž 4.1 .
which are just the usual inverse moments, but evaluated in the massive theory. Here and
below the brackets denote the average with respect to the massive and partially
quenched theory in a topological gauge field sector of charge n . In order to derive such
massive spectral sum rules we return to the small-mass expansion of the partially
quenched condensate Ž3.3..
Inserting the original partition function Ž1.1. into the definition Ž3.2. of the partially
quenched chiral condensate one has
Sn Ž m Õ , m 4 . 1 n
S
s 2 mÕ ¦ Ý 2 2
zn)0 z n q m Õ
; q
mÕ
. Ž 4.2 .
We can now find a useful general relation between the partially quenched chiral
condensate, and Žmassive. spectral sum rules. The connection is simple: the sum rules of
inverse moments zy2 n of the eigenvalues will be convergent up to a given value of n,
which we denote by k. Let us therefore expand the denominator of Eq. Ž4.2. in partial
fractions up to this maximal value k:
k
1 1 m2Õ Ž y1. m2Õ k
s y q...q Ž 4.3 .
zn2 q m2Õ zn2 zn4 zn2 k Ž zn2 q m2Õ .
which means that
Sn Ž m Õ , m 4 . 1 1
S
s2 ¦Ý ; ¦Ý ;
zn)0 z n
2
mÕ y 2
z n)0 zn4
m3Õ q . . .
k
2 Ž y1 . m2Õ kq1 n
q
¦ Ý
zn)0 z n
2k
Ž zn2 q m2Õ ;
.
q
mÕ
. Ž 4.4 .
We can thus simply read off a whole string of massive spectral sum rules from the
coefficients in the m Õ-expansion of the partially quenched chiral condensate Ž3.3.:
2
1 1 Tr M 2 Ž Tr M 2 .
¦Ý ;
zn)0 zn2
s
4N
q
16 N 2 Ž N 2 y 1 .
q
16 N 3 Ž N 2 y 1 . Ž N 2 y 4 .
2
Tr Ž M 2 .
ž /
y
32 N Ž N y 1 . Ž N 2 y 4 .
2 2
3
15Tr Ž M 2 .
ž /
q
768 N Ž N y 1 . Ž N y 4 . Ž N 2 y 9 .
2 2 2
3
3 Ž 2 N 2 y 3 . Ž Tr M 2 .
q 2
64 N 4 Ž N 2 y 1 . Ž N 2 y 4 . Ž N 2 y 9 .
2
3 Ž 3 y 2 N 2 . Tr M 2 Tr Ž M 2 . ž /
q 2
q... , Ž 4.5 .
3 2 2 2
64 N Ž N y 1 . Ž N y 4 . Ž N y 9 .
1 1 Tr M 2
¦ ;Ý
zn)0 zn4
s
16 N Ž N 2 y 1 .
q
16 N 2 Ž N 2 y 1 . Ž N 2 y 4 .
2
Ž 6 N 2 y 9 . Ž Tr M 2 .
q 2
64N 3 Ž N 2 y 1 . Ž N 2 y 4 . Ž N 2 y 9 .
2
3 Ž 3 N 2 y 7 . Tr Ž M 2 . ž /
y 2
q..., Ž 4.6 .
256 N Ž N y 1 . Ž N y 4 . Ž N 2 y 9 .
2 2 2
1 1 15Tr M 2
¦Ý ;
zn)0 zn6
s
32 N Ž N 2 y 1 . Ž N 2 y 4 .
q
256 N 2 Ž N 2 y 1 . Ž N 2 y 4 . Ž N 2 y 9 .
q..., Ž 4.7 .
1 5
¦Ý ;
zn)0 z n
8
s
256 N Ž N y 1 . Ž N 2 y 4 . Ž N 2 y 9 .
2
q... Ž 4.8 .
The resulting sum rules are here only given as a perturbative expansion in the physical
fermion masses, in contrast to the massless spectral sum rules which are exact. One
notices that trivially the ‘‘diagonal’’ massless spectral sum rules Ž2.14.-Ž2.17. are
reproduced by taking all m i s 0. As discussed already in Section 2, the above expan-
sions are valid for N larger than the biggest integer poles. This is completely
understandable from the present point of view, as it corresponds to performing the
expansion Ž4.3. in partial fractions only up to the point where the spectral sums of all
terms still converge. The ‘‘non-diagonal’’ massive sum rules can also be calculated
within this framework. One simply needs to break the mass-degeneracy of the valence
quarks and consider general partially quenched n-point correlators of cc , as discussed

in the introduction.
These new massive spectral sum rules can easily be compared with the predictions
from Random Matrix Theory. What we need is to expand the microscopic spectral
density rSŽ N f , n . Ž z , m i 4. in powers of Tr M 2 , and then insert this expansion in
Sn Ž m Õ , m i 4 . ` rSŽ N f , n . Ž z , m 4 . n
s 2 mÕ H0 dz 2
q . Ž 4.9 .
S z q m2Õ mÕ
For example, for Nf s 1 the microscopic spectral density for a massive fermion in a
gauge field sector of arbitrary topological index n is given by w36x Žsee also Refs.
w9,37x.:
<z < 2
< z < m2 Jn Ž z .
rSŽ n . Ž z ; m . s Jnq1 Ž z . y Jn Ž z . Jnq2 Ž z . q
2 2 Ž n q 1. Ž z 2 q m2 .
Inq2 Ž m .
= Jn Ž z . q Jnq2 Ž z .
In Ž m .
<z < 2
Jn Ž z . Jnq2 Ž z . 2
s Jnq1 Ž z . y Jn Ž z . Jnq2 Ž z . q m
2 2 Ž n q 1. < z <
1 Jn Ž z . Jnq2 Ž z .
q
2 Ž n q 1. ž 4 Ž n q 1. Ž n q 2. < z <
y
< z <3 / Jn Ž z . m4 q . . .
Ž 4.10 .
Inserting this expansion into Ž4.9., we find that the relevant integrals become analyti-
cally doable when n is taken large enough to make the integrals converge. In this way
we have explicitly confirmed a number of terms in the above expansions of the massive
spectral sum rules.
5. Generalizations and outlook
The small and large mass expansions for the partially quenched chiral condensate can
of course in principle be extended to the two other major universality classes, corre-
sponding to SUŽ2. gauge group with fermions in the fundamental representation, and
SUŽ Nc 0 2. with fermions in the adjoint representation. The challenge is here to find a
convenient method that permits high-order expansions. In our present case all simplifica-
tions arose from the observation that the effective partition function is annihilated by
two sets of Virasoro constraints, which in turn is rooted in the fact that this effective
partition function is a t-function of an integrable KP hierarchy. The effective partition
functions of the two other universality classes have not been nearly as well studied,
although there are reasons to believe that they are what is known as ‘‘Pfaffian
t-functions’’ Žsee, e.g., Ref. w38x.. It may thus be possible to derive analogous partition
function constraints, which will permit high-order expansions with little labor. In this
connection we also make the following observations. The Virasoro constraints Ž2.8. and
Ž3.13. both follow from the simple differential equation
E 2 Z0Ž N f .
s d i j 14 Z0Ž N f . Ž 5.1 .
E Mi k E Mk†j
by changing variables to the pertinent expansion parameters. This differential equation
by itself is a trivial consequence of the fact that the integration manifold is UŽ Nf q NÕ .,
as it just amounts to inserting the unitarity condition U †U s 1 in the integrand of the
effective partition function Ž2.1.. Imposing two boundary conditions is thus sufficient to
determine the partition function uniquely. In the case at hand it is the choice Z0Ž N f . Ž tq
k
s 0. s 1 and Y0Ž N f . Ž ty Ž Nf .Ž
k s 0 s 1, together with the required property Z0
. M, M †. s
Z0Ž N f . Ž M † M . that provide the two conditions.
Except for a change of the proportionality constant, here 1r4, the differential
equation Ž5.1. also holds for the effective partition functions of the two other universal-
ity classes mentioned above, as well as for the one of QCD 3 . It is also instructive to
notice that the restriction to a sector of fixed topological charge does not affect the
differential equation. The boundary conditions imposed on the differential equation,
however, separate the solutions, i.e., the partition functions. For instance, before the
projection onto a fixed topological sector the partition function studied here depends on
detŽ M . and detŽ M † . as well as on M M †.
Of course we can already at this stage obtain the partially quenched chiral conden-
sates to lowest non-trivial orders for the two remaining ensembles by simply making use
of the same small-mass expansions of the partition function that were used to derive the
first sum rules. This is a trivial exercise: from Ref. w2x we learn that the effective
partition function for SUŽ2. gauge group and Nf q NÕ fermions in the fundamental
representation is
n 1
Zn s Pf Ž M˜ . Tr M˜ † M˜q . . . ,
ž 1q
8 Ž 2 Nf q 2 NÕ q n y 1 . / Ž 5.2 .
where, now M˜ is an antisymmetric Ž2 Nf q 2 NÕ . = Ž2 Nf q 2 NÕ . matrix with the usual

mass matrices qM M and yM M placed in the two off-diagonal blocks, and zeros in the two
diagonal blocks. Using again Eqs. Ž3.1. and Ž3.2. we find the partially quenched chiral
condensate to this order:
Sn Ž m Õ . n mÕ
s q q..., Ž 5.3 .
S mÕ 2 Ž 2 Nf q n y 1 .
which one can confirm matches the first term in the expansion of the result obtained
from Random Matrix Theory Žsee Ref. w24x.. Note that this expansion has its first
uncancelled de Wit–’t Hooft pole at 2 Nf q n s 1, and the above term in the expansion is
thus valid for 2 Nf q n ) 1.
Similarly, for the universality class corresponding to Nf q NÕ adjoint fermions and
arbitrary gauge group SUŽ Nc . we see from Ref. w2x that the partition function expansion
is
n 1
Zn s det Ž M .
ž 1q
2 Ž Nf q NÕ q 2 n q 1 . /
Tr M † M q . . . , Ž 5.4 .
where n s n Nc is an integer. Using the present replica method this leads to a partially
quenched chiral condensate of
Sn Ž m Õ . n mÕ
s q q..., Ž 5.5 .
S mÕ Nf q 2 n q 1
which one again can check matches the result from Random Matrix Theory w24–27x.
There is no de Wit–’t Hooft pole in this case, and thus no restriction on the validity of
this first term.
Since one can obtain perturbative solutions to the partially quenched chiral conden-
sate in all three universality classes, a natural question concerns the microscopic spectral
density of the Dirac operator rS Ž z ; m i 4.. This density is given by the discontinuity of
the partially quenched chiral condensate across the cut on the imaginary axis w14–16x:
1 1
rS Ž z ; m i 4 . s Disc m Õ s i z Sn Ž m Õ , mi 4 . s lim Sn Ž i z q e , m i 4 .
2p 2p e™0
ySn Ž i z y e , m i 4 . . Ž 5.6 .
ŽThis identification holds when one considers Sn Ž m Õ , m i 4. as a function of a real mass
m Õ , and then replaces m Õ ™ i z " e .. It now seems straightforward to insert our
small-mass and large-mass series expansions for Sn Ž m Õ , m i 4., and derive corresponding
series expansions for the spectral density rS Ž z ; m i 4.. But this is not possible. Let us first
consider the small-mass expansion of Eq. Ž3.3.. As we emphasized in Subsection 3.1,
this expansion is valid up to the first de Wit–’t Hooft pole. Except for the topological
term nrm Õ which always trivially yields the correct d-function contribution to the
microscopic spectral density, this power-series expansion of a finite number of terms has
no cut across the imaginary axis. This may seem surprising, since the microscopic
spectral density rS Ž z ; m i 4. does have a simple and well-defined power-series expansion.
The explanation is, however, simple. Because of the de Wit–’t Hooft poles, we can
derive the small-mass expansion of Sn Ž m Õ , m i 4. up to Žand including. order m2Õ N fq2 ny1.
But the perturbatively expanded rS Ž z ; m i 4. Žsee, e.g. the massless case Ž2.27.. starts
only at order z 2 N fq2 nq1. The microscopic spectral density is always precisely one step
ahead of the order to which we can push the small-mass expansion of the partially
quenched chiral condensate! In hindsight, a phenomenon like this had to occur, since the
pure power series implied by the small-mass expansion cannot give rise to any
discontinuity across the imaginary axis. So the de Wit–’t Hooft poles in fact precisely
save what would otherwise be a paradoxical situation.
The asymptotic large-mass expansion of Subsection 3.2 cannot be used to extract the
microscopic spectral density either. Here the reason is very different. Recall that the
asymptotic expansion we have derived in Subsection 3.2 is suitable m Õ on the positive
real axis. For example, for Nf s 1 our expansion simply coincides, up to an irrelevant
normalization factor, with the asymptotic expansion of the Bessel function I0 Ž m .:
k
em ` Ž y1. G Ž k q 1r2 .
Z0Ž N fs1. s I0 Ž m. ; Ý k Ž 5.7 .
'2pm ks0 Ž 2 m . k! G Ž yk q 1r2.
This asymptotic expansion is correct, but it neglects exponentially suppressed terms of
order eym r m and lower. Because of this, the asymptotic expansion Ž5.7. does not
'
reproduce the asymptotic expansion for J0 Ž m . after rotation to the imaginary axis Žthe
exponentially suppressed terms then become of magnitude comparable to those kept in
Ž5.7., and that expansion is therefore no longer correct near the imaginary axis.. We
conclude that neither of the two expansions we have considered here are suitable for
deriving the microscopic spectral density.
6. Conclusions
Using the Virasoro constraints on the effective partition function we have obtained
small-mass and large-mass expansions for the QCD partition function valid in the
scaling region V < 1rmp4 . In these expansions we can treat the number of valence
quarks NÕ as a continuous parameter. This form is thus suited for a calculation of the
partially quenched chiral condensate using the replica method. Two series of spectral
sum rules for the QCD Dirac operator follow from the small-mass expansion. One series
extends the Leutwyler–Smilga sum rules in a simple way to very high Žhere 8th. order,
while the other series is new: it computes the spectral sum rules in the massive theory,
and we have given series expansions in the physical masses for these new sum rules. In
all cases we have checked, we have found complete agreement with earlier results based
either on Random Matrix Theory or the supersymmetric technique. The replica method
constitutes a new and independent derivation of these results.
With the expansions of the partition function at hand we have also derived the
small-mass and large-mass expansions of the partially quenched chiral condensate. In all
cases we have checked there is again complete agreement with earlier analytical
predictions based on the two other methods.
We have here restricted ourselves to small-mass and large-mass series solutions only
because they provide the simplest framework in which to employ the replica method.
This is not an inherent restriction, though. It is thus an open challenge to extend the
method beyond these two series expansions.
Acknowledgements
¨
We thank M. Luscher and H.B. Nielsen for an early conversation. This work was
supported in part by EU TMR grant no. ERBFMRXCT97-0122.
References
w1x H. Leutwyler, A. Smilga, Phys. Rev. D 46 Ž1992. 5607.

w2x A. Smilga, J.J.M. Verbaarschot, Phys. Rev. D 51 Ž1995. 829.
w3x E.V. Shuryak, J.J.M. Verbaarschot, Nucl. Phys. A 560 Ž1993. 306.
w4x J.J.M. Verbaarschot, I. Zahed, Phys. Rev. Lett. 70 Ž1993. 3852.
w6x J.J.M. Verbaarschot, Nucl. Phys. B 426 Ž1994. 559.
w7x J.J.M. Verbaarschot, Phys. Rev. Lett. 72 Ž1994. 2531.
w8x G. Akemann, P.H. Damgaard, U. Magnea, S. Nishigaki, Nucl. Phys. B 487 Ž1997. 721.
w9x P.H. Damgaard, S.M. Nishigaki, Nucl. Phys. B 518 Ž1998. 495.
w10x M.K. Şener, J.J.M. Verbaarschot, Phys. Rev. Lett. 81 Ž1998. 248.
w11x P.H. Damgaard, Phys. Lett. B 424 Ž1998. 322.
w12x G. Akemann, P.H. Damgaard, Nucl. Phys. B 519 Ž1998. 682.
w13x G. Akemann, P.H. Damgaard, Phys. Lett. B 432 Ž1998. 390. hep-thr9910190.
w14x J.C. Osborn, D. Toublan, J.J.M. Verbaarschot, Nucl. Phys. B 540 Ž1999. 317.
w15x P.H. Damgaard, J.C. Osborn, D. Toublan, J.J.M. Verbaarschot, Nucl. Phys. B 547 Ž1999. 305.
w16x D. Toublan, J.J.M. Verbaarschot, hep-thr9904199.
w17x J.J.M. Verbaarschot, M.R. Zirnbauer, J. Phys. A 18 Ž1985. 1093.
w18x A. Kamenev, M. Mezard, cond-matr9901110.
w19x A. Kamenev, M. Mezard, cond-matr9903001.
w20x I.V. Yurkevich, I.V. Lerner, cond-matr9903025.
w21x M.R. Zirnbauer, cond-matr9903338.
w22x E. Kanzieper, cond-matr9908130.
w23x J.J.M. Verbaarschot, Phys. Lett. B 368 Ž1996. 137.
w24x P.H. Damgaard, R.G. Edwards, U.M. Heller, R. Narayanan, hep-latr9907016.
w25x P.H. Damgaard, R.G. Edwards, U.M. Heller, R. Narayanan, hep-latr9909009.
w26x P. Hernandez, K. Jansen, L. Lellouch, hep-latr9907022.
w27x P. Hernandez, K. Jansen, L. Lellouch, hep-latr9909026.
w28x R. Brower, P. Rossi, C.-I. Tan, Nucl. Phys. B 190 Ž1981. 699.
w29x A.D. Jackson, M.K. Şener, J.J.M. Verbaarschot, Phys. Lett. B 387 Ž1996. 355.
w30x D.J. Gross, M.J. Newman, Nucl. Phys. B 380 Ž1992. 168.
w31x A. Mironov, A. Morozov, G.W. Semenoff, Int. J. Mod. Phys. A 11 Ž1996. 5031.
w32x B. de Wit, G.’t Hooft, Phys. Lett. B 69 Ž1977. 61.
w33x I. Bars, J. Math. Phys. 21 Ž1980. 2678.
w34x J.J.M. Verbaarschot, Phys. Lett. B 329 Ž1994. 351.
w35x J. Verbaarschot, Nato Lectures Cambridge 1997 on Confinement, duality, and non-perturbative aspects of
QCD pp. 343–378, hep-thr9710114. T. Akuzawa, M. Wadati, J. Phys. Soc. Jpn. 67 Ž1998. 2151.
w36x P.H. Damgaard, Phys. Lett. B 425 Ž1998. 151.
w37x T. Wilke, T. Guhr, T. Wettig, Phys. Rev. D 57 Ž1998. 6486.
w38x J. van der Leur, solv-intr9909028.
On the geometry of non-trivially embedded branes

Ruben Minasian a,b, Dimitrios Tsimpis a
a
Department of Physics, Yale UniÕersity, New HaÕen, CT 06520, USA
b
´
Centre de Physique Theorique, Ecole Polytechnique, F-91128 Palaiseau, France
Abstract
We present a formal supersymmetric solution of type IIB supergravity generalizing previously

known solutions corresponding to D3-branes to geometries without an orthogonal split between
parallel and transverse directions. The metric is given implicitly as one with respect to which a
certain connection is compatible. The case of the deformed conifold is discussed in detail. q 2000
PACS: 11.25.-w; 04.65.qe
1. Introduction and motivation
The AdSrCFT correspondence in cases with lower supersymmetries involving the

two- and three-dimensional AdS spaces is much less understood than the others w1x. In
particular, the near-horizon geometry of black strings in simple five-dimensional super-
gravity Žobtained by compactification of M-theory on a Calabi–Yau threefold X . is
given by AdS 2 = S 2 . Here we meet a little puzzle discussed also in w2x. The spectrum of
the supergravity theory is defined by H 2 Ž X,Z., and when considering a magnetic string
solution, one naturally writes down a string coupled to Ž h11Ž X . y 1. vectors and a
graviphoton. However, as pointed out in w3x, the theory on the string is governed by a
much larger lattice defined by H 2 Ž P,Z., where P is the four-cycle around which the
M-theory fivebrane is wrapped. In particular, this accounts for a very large entropy. This
information about the cycle is not reflected in the KK spectrum though. As seen in w4x,
the target space of the dual Ž0,4. conformal field theory is factorized. The coupling to
supergravity is governed mostly by the so called universal sector, while the numerous
modes making up the theory on the two-dimensional world-sheet are in the entropic
sector. Apparently, the usual procedure of compactifying M-theory and then looking for
PII: S 0 5 5 0 - 3 2 1 3 Ž 0 0 . 0 0 0 3 5 - 3
500 R. Minasian, D. Tsimpisr Nuclear Physics B 572 (2000) 499–513
the solutions misses the deformations of the cycle. In doing so we simply find a solution
corresponding to the universal sector 1 , and schematically we can write
w compactification, solutionx s entropic sector.
Finding a complete solution with all the modes seems to be extremely hard. Instead, we
try to give a description of the 10d geometry corresponding to the situation in w6x: a
D-brane wrapping a non-trivial cycle in an internal Calabi–Yau space, shrinking to
zero-size at certain points of the moduli space thus turning into a massless BPS black
hole from the four-dimensional non-compact space point of view 2 .
D3-branes on conifold points have been discussed extensively in the AdS literature
recently w7–10x. However, only the situation where the branes are transverse to a CY
manifold, and can be thought of as space-time-filling, has been addressed there. We will
be concerned with a rather different geometrical setup, the case of a D3 wrapping a
3-cycle in an internal CY threefold.
In Sections 2 and 3 we will outline a procedure for finding formal supersymmetric
solutions under some general assumptions about the metric in the absence of D3.
Unfortunately, the metric of the solution is only given implicitly as one with respect to
which a certain connection is compatible.
As a specific case, we will consider type IIB supergravity on a non-compact
Calabi–Yau threefold which we will call the deformed conifold C Ž ´ .. In doing this we
are able to avoid the question of extra modes since in our limiting case the three-fold is
approximated by T ) S 3 and S 3 is rigid. The deformed conifold, which will be described
in the following in some detail, is topologically a 6d cone over an S 2 = S 3 base, whose
apex is replaced by an S 3.
Before turning to the concrete example of D3 on a shrinking S 3 cycle in Section 5,
we discuss the geometry of the singular and deformed conifolds of w11x. Using the
machinery of coset-space geometry w12,13x, we are able to give an explicit form of the
deformed conifold metric presented implicitly in w11x. A possible generalization to M2
and M5 cases is discussed in Section 6.
2. Branes on cycles
We present a method for constructing formal supersymmetric solutions of type IIB

supergravity corresponding to D3-branes, which generalizes previously known solutions
to geometries without an orthogonal split between directions parallel and transverse to
the brane. In particular, we do not assume that the D3 is space-time-filling but our
method covers this situation as a special case.
In the presence of the D3 the 10d geometry will get deformed to account for the back
reaction due to the brane. This back reaction is captured by the warping factor, a
function of the coordinates transverse to the brane. There is considerable amount of
literature on the subject of supergravity solutions corresponding to branes Žsee e.g. Refs.
1
The full entropy is recovered by loop corrections to supergravity w5x.
2
Note that N D3-branes do not form a bound state on the conifold. To ensure the reliability of supergravity
we are tacitly assuming that N is large.
R. Minasian, D. Tsimpisr Nuclear Physics B 572 (2000) 499–513 501
w14,15x.. These solutions assume that in the absence of D3 there is an orthogonal split of
the 10d metric along parallel and transverse to the brane directions:
ds 2 s hmn dx mdx n q g m n Ž y . dy m dy n , Ž 2.1 .
where hmn is the Žflat. metric on the cycle around which the D3 will wrap Žwhich in that
case is the whole 4d Minkowski space-time. and g m nŽ y . is the transverse metric. In the
presence of the D3, the geometry is modified,
ds 2 s D H Ž y . hmn dx mdx n q D < < Ž y . g m n Ž y . dy m dy n , Ž 2.2 .
where D H Ž y ., D < < Ž y . are the warping factors Žwhich turn out to be related. and are
functions only of the coordinates transverse to the brane.
Here we will not assume that there is the ‘‘nice’’ split of the form Ž2.1.. Instead we
will take the 10d metric in the absence of D3 to be of the form
5
a
ds 2 s gmn Ž x , u . dx m dx n q f 2 Ž U . dU 2 q Ý Ž e Ž x ,u ,U . m dx
m
as1
a i 2
qe Ž x , u ,U . i du . , Ž 2.3 .
where x m ; m s 0, . . . ,34 are the coordinates parametrizing a non-trivially embedded
cycle C , U is a ‘‘radial’’ coordinate in the transverse space and u i ;i, . . . ,54 are
‘‘angular’’ coordinates in the transverse space. The metric on C is gmn and it does not
depend on U. Note that eŽ x, u ,U . a m encodes the deviations from orthogonally split
geometries. We will further assume that
a a
e Ž x , u ,U s 0 . mse Ž x ,u ,U s 0 . i s 0, Ž 2.4 .
so that C is at U s 0. If we wish, we may consider U as a collective label for a set of
‘‘radial ’’ coordinates U1 ,U2 , . . . which enter the metric Ž2.3. as f 12 ŽU1 . dU12 q
f 22 ŽU2 . dU22 q . . .
For simplicity we will drop the u dependence of gmn in the following. All our
arguments of Section 3 go through for u-dependent gmn as well. Let us remark that if we
keep the u dependence, it appears that in the flat D3 limit we may recover warped
AdS 5 =w S 5 products. The possibility of such supergravity vacua has been known for
some time w16,17x, however these have not appeared as brane near-horizon limits so far.
This discussion may provide a brane realization of such vacua.
The metric of Ž2.3. is of some generality. A trivial example would be the case where
the 10d metric is that of a direct sum of 4d Minkowski plus a 6d cone, with U being the
distance Žin the ten-dimensional sense. from the apex. Another example, which will be
discussed in the following, is the geometry of the deformed conifold near the S 3 at the
apex.
3. The solution
In this section we will discuss how does Ž2.3. change in the presence of a D3 along
C . As already emphasized, the solution will be given implicitly: the metric in the
presence of D3 is the metric with respect to which the connection of Eq. Ž3.8. below, is
metric compatible. This statement makes sense since for every torsion-free connection
with SO Ž N . holonomy Žwhere N is the dimension of the manifold. there is an
essentially unique metric with respect to which the connection is compatible. We can see
this as follows: take the metric at a given point to be some constant symmetric N = N
matrix. Parallel-transport it using the connection to define the metric at any other point.
The absence of inconsistencies under parallel transport along closed loops is equivalent
to the requirement of SO Ž N . holonomy. The metric constructed in this way is defined
up to rigid GLŽ N,R. coordinate transformations.
We want to warn the reader that as we do not have the charge distribution explicitly,
nothing excludes the possibility that this solution corresponds to a completely ‘‘smeared’’
D3. In that case it would be wrong to think of the D3 as wrapping C . It is more correct
to say that C will be identified with the horizon in the presence of the D3 3.
3.1. ConÕentions
m , n are curved indices for the directions along the D3.

m, n are curved indices for the coordinates transverse to the D3 including U.
i, j are curved indices for the coordinates transverse to the D3 excluding U.
M, N are ten-dimensional curved indices.
a , b are flat indices for the directions along the D3.
a, b are flat indices corresponding to the directions transverse to the D3 excluding U.
v is the flat index corresponding to U.
A, B are ten-dimensional flat indices.
3.2. The solution
It will be useful to give the coframe version of Ž2.3.:

3 5
2
ds 2 s Ý Ž ea . 2 q Ý Ž ea. 2 q Ž ev . , Ž 3.1 .
as0 as1
where
a
ea seŽ x . m v
m dx , ev seŽU . U dU s f Ž U . dU,
a m a
e a s e Ž x , u ,U . m dx q e Ž x , u ,U . i du
i
. Ž 3.2 .
The fields of IIB are the graviton g M N , a complex scalar t parametrizing an
SLŽ2,R.rUŽ1. coset space, a pair of two-forms BM1,2N which form an SLŽ2,R. doublet, a
self-dual four-form AŽ4. with field strength F Ž5., and two complex-Weyl fermions: a
gravitino c M and a dilatino l.
3
We would like to thank G. Moore for explaining this to us.
The supersymmetry transformations are parametrized by a complex-Weyl spinor e

and, in a background with all fields set to zero except for the graviton and the four-form,
only the gravitino transformation is not identically zero,
i A1 . . . A 5
de cA s D Ž V . A e q G FA 1 . . . A 5 GA e , Ž 3.3 .
4 = 5!
where
D Ž V . A s EA q 14 V A BC
GB C , EA s e A ME M Ž 3.4 .
and V A B is the connection one-form corresponding to the 10d metric in the presence of
the D3.
Our ansatz for the five-form is given in terms of one function of U, C ŽU .
Fa 0 a 1a 2 a 3 v s á 0 a 1a 2 a 3 C Ž U . , Fa1 a 2 a 3 a 4 a 5 s ´ a1 a 2 a 3 a 4 a 5 v
CŽU . Ž 3.5 .
with all other components equal to zero. We then find
A1 . . . A 5
G FA 1 . . . A 5 Ga s yi5!C Ž U . G v Ga Ž r Ž6. q r Ž4. . ,
A1 . . . A 5
G FA 1 . . . A 5 Ga s i5!C Ž U . G v Ga Ž r Ž6. q r Ž4. . , Ž 3.6 .
where
r Ž4. :s i G as0 . . . G as3 , r Ž6. :s yi G as 1 . . . G as 5 Ž 3.7 .
are the ‘‘parallel’’ and ‘‘transverse’’ chirality operators. Let v A B be the connection
one-form associated to the metric Ž3.1.. Our ansatz for the 10d metric in the presence of
the D3 will be given implicitly by requiring that it be associated to the connection V A B
given by
V A B s v A B q e AE B ln D Ž U . Ž A . y e B E A ln D Ž U . Ž B . , Ž 3.8 .
where
DŽ U . < < Asa
D Ž U . Ž A. s ½ DŽ U . H Asa
Ž 3.9 .
are two warping factors.

In w7x a relation identical to Ž3.8. holds, arising from a rescaling of the vielbeins
e ™ DŽ A. e A. Our case however is very different as it does not imply that the associated
A
metrics are related by such a rescaling. The reason is that, as follows from Ž3.1., Ž3.2.,
the connection v A B generally has non-zero components of the form v a a mixing
transverse with parallel directions.
We will now assume that
r Ž4.e s r Ž6.e s e Ž 3.10 .
the susy transformations then read
0 s D Ž v . a q Ga v Ž E U ln D < < q 12 C . e ,
0 s D Ž v . a y 12 Cd av q Gav Ž E U ln D H y 12 C . e . Ž 3.11 .
Setting
1
C Ž U . s 2 E U ln D Ž U . H , DŽ U . Hs , e s D Ž U . < < eˆ Ž 3.12 .
DŽ U . < <
Eq. Ž3.11. reduces to
D Ž v . A eˆ s 0, Ž 3.13 .
i.e. the solution preserves some supersymmetry provided the geometry Ž3.1. in the
absence of D3 admits a covariantly constant spinor. The integrability of Ž3.13. is
equivalent to the requirement of Ricci-flatness for the geometry in the absence of D3.
Ric Ž v . A B s 0. Ž 3.14 .
To check the consistency of our ansatz with Einstein equations
1
Ric Ž V . A B s F A 1 . . . A 4 FB A 1 . . . A 4 Ž 3.15 .
4 = 4! A
we simply substitute Ž3.8. into Ž3.15. taking Ž3.14. into account. The result is w7,18x
U 2
D Ž v . U D Ž v . D H Ž U . s 0. Ž 3.16 .
This is the condition that D H ŽU . 2 is harmonic. In proving the above we used the fact
that
v v a s 0. Ž 3.17 .
We can see this as follows. From Ž3.1. we get
dU . s de v s yv v a e a y v v a e a ,
v
0 sdŽ eŽU . U Ž 3.18 .
therefore the only possibly non-zero components of v v a are of the form va v
a. On
the other hand,
de a s yv a v e v q . . . s yva a
v eanev q . . . Ž 3.19 .
a a
But since E U e s 0, de cannot have a piece proportional to dU and we conclude that
v v a s 0.
4. Geometry of conifolds
The main purpose of this section is to provide a specific example of the geometrical
setup under which our solution of the previous section is valid. Indeed we show Žsee
Subsection 4.5. that the near-horizon limit of the deformed conifold is a particular case
to which our method applies.
None of this section is new. The results are in principle contained in previous works
w11–13x. However, we want to draw the attention of the reader to two points. Eq. Ž4.15.
below, contains a term Žthe one proportional to B . which is usually omitted from
discussions in the literature related to T 1,1 spaces. However, this term appears naturally
in the metric of the deformed conifold, explicitly presented in Eq. Ž4.35.. In Section 5 of
Ref. w11x the metric is given implicitly in terms of seven variables Žone radius and six
Euler angles. in a form which makes it difficult to distinguish six independent ones. For
these reasons we think the discussion of this section is useful.
4.1. T p ,q spaces
A cone Cdq 1 in d q 1 space-time dimensions over a d-dimensional base X d is given

by
ds 2 s d r 2 q r 2 g i j dx i dx j . Ž 4.1 .
The cone Cdq 1 has the property that the vector ErEr is conformally killing. The metric
g i j ; i, j s 1, . . . ,d determines the geometry of the base. Cdq1 is Ricci-flat iff X d is
Einstein with cosmological constant Ž d y 1. and is irregular at r s 0 unless X d s S d.
Our situation corresponds to d s 5 and we will consider the base to be a T 1,1 space.
A T 1,1 space is a particular example of T p, q spaces w19x. These can be thought of as
UŽ1. fibrations over S 2 = S 2 . Let 0 ( f i ( 2p , 0 ( u i ( p , i s 1,2 parametrize the two
S 2 and let 0 ( c i ( 4p be the coordinate on the UŽ1. fiber. The line element of T p, q is
then given by
2
ds 2 s l1 Ž d c q pcos u 1 d f 1 q qcos u 2 d f 2 . q l2 Ž sin2u 1 d f 12 q du 12 .
q l 3 Ž sin2u 2 d f 22 q du 22 . , Ž 4.2 .
where the first term on the r.h.s. is the vertical displacement along the fibre and the other
two terms are the line elements on the S 2 ’s. By ‘‘forgetting’’ one of the S 2 ’s the T 1,1
space can be thought of as an S 3 fibration over S 2 – the base being the S 2 we ‘‘forget’’
and the fibre being a UŽ1. fibration over the other S 2 . This fibration is actually trivial 4
and therefore T 1,1 is topologically S 2 = S 3.
If the following algebraic conditions are met,
2 2 2 2
p2 l1 q2 l1 l1 p2 l1 l1 q2 l1
Ll1 s
2 ž /
l2
q
2 ž /
l3
s
l2
y
2 ž /
l2
s
l3
y
2 ž /
l3
, Ž 4.3 .
Eq. Ž4.2. describes an Einstein manifold of cosmological constant L.

In the case p s q s 1 and L s 4 the cone over T 1,1 is Ricci-flat and Ž4.3. implies
l1 s 19 , l2 s l3 s 16 . Ž 4.4 .
The spaces T p, q also have a coset description as SUŽ2. = SUŽ2.rUŽ1., which is
another way to see the S 2 = S 3 topology. It will pay off to make a digression on the
geometry of coset spaces which will eventually help us give a useful description of the
deformed conifold. For a more comprehensive account one should consult the literature
w12,13x.
4
There are various way to see this. See for example w8x.
4.2. The geometry of coset spaces
In this section we use techniques of coset spaces to give a generalization of the metric
Ž4.2.. This generalized metric for T 1,1 will appear naturally in the following when we
discuss the deformed conifold.
Consider a Lie group G and a subgroup H g G generated by Hi ;i s 1, . . . ,dim H 4
such that G is generated by Hi , Ea ;a s 1, . . . ,dimG y dim H 4 and
Hi , H j s c i j k H k , w Hi , E a x s c i a b Eb , w E a , Eb x s c a b d E d q c a b i Hi .
Ž 4.5 .
We call left cosets the elements of the form g P H, g g G. To parametrize the coset we
choose a particular group element in each coset which we call the coset representatiÕe
LŽ f a .. Here f a , a s 1, . . . ,dimG y dim H 4 are coordinates on GrH. The element
Ly1Ea L, where Ea :s ErEf a , is in the Lie algebra of G. The expansion
Ly1Ea L s eaa Ea q vai Hi Ž 4.6 .

defines the vielbein eaa Ž f . and the H-connections vai Ž f .. Coset manifolds GrH have
at least an isometry GX = N Ž H .rH, where the latter factor is the normalizer Žthe largest
subgroup of G of which H is a normal subgroup. and G s GX = ŽUŽ1. factors common to
G and to N Ž H .rH ..
A metric on GrH preserving the isometries is given by
g ab s h a b eaa ebb , Ž 4.7 .
where h a b is an H-invariant tensor
c i a c h cb q c i b c h ac s 0. Ž 4.8 .
As an application, we can reproduce the metric Ž4.2. for p s q s 1 as follows.
Parametrizing using Euler angles, the group element of SUŽ2. = SUŽ2. can be written as
X X X X
e i s 3 f 1 r2 e i s 2 u 1 r2 e i s 3 f 2 r2 e i s 2 u 2 r2 e iŽ s 3qs 3 .Ž c 1qc 2 .r4 e iŽ s 3ys 3 .Ž c 1yc 2 .r4 , Ž 4.9 .
where si 4 and siX 4 obey the algebra of the Pauli matrices with w s , s X x s 0. We take the
coset representative to be
X X X
L s e i s 3 f 1 r2 e i s 2 u 1 r2 e i s 3 f 2 r2 e i s 2 u 2 r2 e iŽ s 3qs 3 . c r2 . Ž 4.10 .
Let
i i X
E1,2 s s 1,2 , E3 ,4 s s 1,2 ,
2 2
i i
E5 s Ž s 3 q s 3X . , Hs Ž s 3 y s 3X . . Ž 4.11 .
4 4
From Ž4.6. we get

e 1 s ysin u 1 d f 1 ,
e 2 s du 1 ,
e 3 s cos c sin u 2 d f 2 y sin c du 2 ,
e 4 s sin c sin u 2 d f 2 q cos c du 2 ,
e 5 s d c q cos u 1 d f 1 q cos u 2 d f 2 . Ž 4.12 .
b
Since the coset representation c i a of H Žcf. Ž4.5.. is block diagonal
ys 2 0 0
i
ci a b s
2
0
0
s2
0
0
0
0
the most general H-invariant tensor h a b is of the form
Ž 4.13 .
A1 2 B1 2 0

h a b s B1 2
0
C1 2
0 D
0
0 .
Substituting into Ž4.7. we get the T 1,1 metric

Ž 4.14 .
2 2 2 2 2
ds 2 s D Ž e 5 . q A Ž e 1 . q Ž e 2 . q C Ž e 3 . q Ž e 4 . q 2 B Ž e 1 e 3 q e 2 e 4 .
ž / ž /
2
s D Ž d c q cos u 1 d f 1 q cos u 2 d f 2 . q A Ž sin2u 1 d f 12 q du 12 .
q C Ž sin2u 2 d f 22 q du 22 . q 2 B cos c Ž du 1 du 2 y d f 1 d f 2 sin u 1 sin u 2 .
qsin c Ž sin u 1 d f 1 du 2 q sin u 2 d f 2 du 1 . . Ž 4.15 .
Note that for B / 0 the metric above cannot be Einstein. Indeed one finds for the a s 2,
b s 4 component of the Ricci tensor R a b
yB 2 Ž cos2 u 1 y cos2 u 2 . csc 2u 1csc 2u 2
R 24 s 3r2 3r2
. Ž 4.16 .
8Ž A y B . Ž AqB.
For B s 0 the metric reduces to the standard metric Ž4.2. with p s q s 1.
For simplicity we will consider the case A s C. It is useful to make a redefinition
1 1
12 y 12 0
'2 '2
e a ™ g a :s P a b e b ,
In this basis Ž4.15. becomes diagonal

P :s
'2
1
0
12
1
'2
0
12 0
1
0 . Ž 4.17 .
2 2 2 2 2
ds 2 s D Ž g 5 . q Ž A q B . Ž g3. q Ž g4 . q Ž AyB. Ž g1. q Ž g2. . Ž 4.18 .
We remark that the first two terms describe a squashed S 3. Indeed, if we define
T :s L1 s 1 L†2 s 1 , Ž 4.19 .
we find
2 2 2
d V 32 :s 12 Tr Ž dT †dT . s 12 Ž g 5 . q Ž g 3 . q Ž g 4 . . Ž 4.20 .
Since T is an SUŽ2. matrix, the metric above is the standard round sphere metric.
Comparing with Ž4.18. we conclude that the first two terms are the line element of a
Žsquashed. S 3. Moreover the last two terms
2 2
ds22 :s Ž g 1 . q Ž g 2 . Ž 4.21 .
should describe a surface which is topologically an S 2 fibred over the squashed S 3. The
fact that T 1,1 can be viewed as either an S 3 fibration over S 2 or an S 2 fibration over S 3
is because it is a trivial fibration.
4.3. The Ricci-flat cone oÕer T 1,1 as a Kahler

¨ manifold
Before we come to the description of the deformed conifold let us summarize some
of the results of w11x. In particular we will see how the Ricci-flat cone over T 1,1 can be
thought of as a singular, non-compact Calabi–Yau threefold.
Consider the cone in C 4 given by
4
2
Ý Ž w A . s 0. Ž 4.22 .
As1
Let us define a radial coordinate r by

r 2 s tr WW † , Ž 4.23 .
1 i 4
Ž w si q iw 1 2 .. The base of the cone is described by
where W:s
'w2x
detW s 0, r 2 s constant. Ž 4.24 .
¨
A Kahler ¨
metric deriving from an SUŽ2. = SUŽ2.-invariant Kahler potential K Ž r 2 .
reads
XX X
dsC2 s <tr W †dW < 2 K Ž r 2 . q tr Ž dW †dW . K Ž r 2 . . Ž 4.25 .
¨
Ricci-flatness determines the Kahler potential to be proportional to r 4r3.
Let us parametrize Ž4.24. as
W s r Z, Z s L1 P Z Ž0. P L†2 , Ž 4.26 .
where L i , i s 1,2 are SUŽ2. matrices. In terms of Euler angles
uj uj
cos e i Žc jq f j . r2 ysin eyi Žc jy f j . r2
Lj s
sin
2
uj
2
e i Ž c jy f j . r2
cos
uj
2
2
e yi Ž c jq f j . r2
0 Ž 4.27 .
and
Z Ž0. s 12 Ž s 1 q i s 2 . . Ž 4.28 .
Substituting into Ž4.25. Žafter a redefinition of the radial coordinate and by setting
c :s c 1 q c 2 . we find
2
dsC2 s Ž d r . q ds X2 Ž 4.29 .
with ds X2 the T 1,1 metric previously given in Ž4.2., for the unique choice of constants
l1 s 1r9, l2 s l3 s 1r6 which give a Ricci-flat metric on the cone. The fact that the
base is T 1,1 can also be seen directly from Ž4.26. by noting that there is a transitive
SUŽ2. = SUŽ2. action with a UŽ1. stabilizer embedded symmetrically in the two SUŽ2.
factors
e iu 0
L1 ™ L1U, L2 ™ L2 U † , U:s
ž 0 yi u
e / g U Ž 1. . Ž 4.30 .
4.4. The deformed conifold: smooth non-compact CY3
In this section we will describe the deformation of the conifold. The apex is replaced
by an S 3. Insisting on Ricci-flatness we get a smooth non-compact CY3 . The r s
constant/ ´ surfaces are still T 1,1 spaces whose geometry is described by the general-
ized metric Ž4.15.. We also examine the geometry near the apex.
Consider deforming Ž4.22. to
detW s y´ 2r2. Ž 4.31 .

We can again define a radial coordinate as in Ž4.23. but now r is bounded below by ´ .
We can parametrize the r s constant surfaces by
W´ s r Z´ , Z´ s L1 P Z´Ž0. P L†2 , Ž 4.32 .

where L i ’s are as before and
e2 e2 e2
Z´Ž0. s ž 0
b
a
0
, / as
1
2 ž( 1q
r2
q ( /
1y
r2
, bs
2r2
ay1 .
Ž 4.33 .
For r / e there is again a transitive SUŽ2. = SUŽ2. action with a UŽ1. stabilizer. For
r s e however, the stabilizer is enhanced to the whole of SUŽ2.
L1 ™ L1U, L2 ™ L 2 s 1Us 1 , Ž 4.34 .

where U g SUŽ2.. The surfaces r s constant are again T 1,1 spaces except for the
surface r s e which is an S 3.
The metric is obtained by substituting W given by Ž4.32. into Ž4.25.. The result is
2
2 1 e4
2 r 2g X y g . q Ž r 2g X y g .
ds 2 s Ž d r .
r 2 Ž
4 ž 1y
r4 /
2
qg Ž d c q cos u 1 d f 1 q cos u 2 d f 2 .
g
q
4
Ž sin2u 1 d f 12 q du 12 q sin2u 2 d f 22 q du 22 .
´2
qg cos c Ž du 1 du 2 y d f 1 d f 2 sin u 1 sin u 2 .
2r2
qsin c Ž sin u 1 d f 1 du 2 q sin u 2 d f 2 du 1 . , Ž 4.35 .
X X X
where g :s r 2 K Ž r 2 . and g :s g Ž r 2 .. The metric on the r s constant slice is of the
generalized form Ž4.15..
Requiring Ricci-flatness and correct asymptotic behaviour of the metric Ž4.35. leads
to the following differential equation:
X
r 2 Ž r 4 y ´ 4 .Ž g 3 . q 3 ´ 4g 3 y 2 r 8 s 0. Ž 4.36 .
The general solution reads
1r3
´4 y1
gs cq
ž 2
Ž sinh2t y 2t . / Ž tanht . , Ž 4.37 .
where c is a constant. In the limit rr´ ™ ` the constant c can be dropped and the
solution asymptotes its ‘‘cone’’ value. For c / 0, g diverges at r s ´ . From now on we
set c to zero.
4.5. The near-horizon limit r ™ ´
Let us examine in more detail the limit r ™ ´ . We make a change of variables

dsry´ Ž 4.38 .
so that
0 ( d - `. Ž 4.39 .
We find that the metric Ž4.35. takes the form
12 2
ds 2 ; R´2 5 Ž dy . q d V 32 q y ds22 , Ž 4.40 .
where
d
y :s ™0 Ž 4.41 .
´
and
1r6
1 2´ 4
R´ :s
'2 ž /3
Ž 4.42 .
is the radius of the S 3 on the apex, d V 32 is the round sphere element defined in Ž4.20.
and ds22 was defined in Ž4.21. and describes a fibre of S 2 topology.
5. D3 on S 3
We will now consider the near-horizon limit in the ten-dimensional sense r ; 0,

r ; ´ . The ten-dimensional metric in the absence of D3 is of the form Ž2.3., and our
solution-generating method applies, only near the S 3 at the apex of the deformed
conifold. We therefore want to emphasize that we only have a metric Žimplicitly. for the
geometry near the horizon. Presumably there is a complete solution, corresponding to a
D3 whose near-horizon limit coincides with the one given here but we were not able to
obtain it. Of course the remark at the beginning of Section 3 applies here as well: as we
do not have the charge distribution explicitly, nothing excludes the possibility that this
solution corresponds to a completely ‘‘smeared’’ D3.
We define a ten-dimensional radial coordinate U and an angle u by
5
U
r s Ucos u , ys ( 12
R´
sin u , Ž 5.1 .
where y was defined in Ž4.41.. Taking Ž4.40. into account, we see that the metric is of
the form Ž3.1. as advertised
5
ds 2 s ydt 2 q R´2 d V 32 q dU 2 q U 2 du 2 q U 2 cos 2u d V 22 q ( 12 R´ Usin u ds22 ,
Ž 5.2 .
where the first line contains the directions parallel to the D3 and the second line contains
the transverse geometry. As already remarked below Ž4.21., ds22 is the line element of
an S 2 fibred over an S 3. The vielbein e˜ A of the 10d metric in the presence of D3 has to
be compatible with the connection given in Ž3.8. as already explained. In particular, if
we define
h A :s e˜ A y e A Ž 5.3 .
we have
D Ž v . h A q e A n dln DŽ A. q e A n h BE B ln DŽ A. q e B n h BE A ln DŽ B . s 0. Ž 5.4 .
The geometry in the presence of D3 is given by Eqs. Ž5.4., Ž3.16., but we will not be
more explicit here. However, the case of the shrinking cycle limit ´ ™ 0 Ž R´ ™ 0. of the
near-horizon geometry and the case of the flat D3 limit can be analyzed explicitly.
5.1. The ´ ™ 0 limit of the near-horizon geometry
As we see from Ž4.40., in the y ™ 0, ´ ™ 0 Ž R´ ™ 0. limit the near-horizon geometry

becomes effectively four-dimensional and is split orthogonally between the directions
parallel to the brane Žthe time direction. and the transverse directions. Eq. Ž3.16. reduces
to the harmonic condition in three spatial dimensions and we recover AdS 2 = S 2 as
usual.
5.2. The flat D3 limit
We can examine the limit where the D3 is seen as flat. To be more precise, let us go
back to Subsection 4.5 and make a change of coordinates
5
u:s ( 12 y R´ . Ž 5.5 .
In terms of u the metric reads
5
ds 2 ; du 2 q R´2 d V 32 q ( 12 uR´ ds22 . Ž 5.6 .
The above approximation is of course valid in the limit y ™ 0. Moreover, the D3 will
appear flat in the regime R´ 4 u ) c, where c is some positive constant. These
conditions are equivalent to
´ 4 d ) c = ´ 1r3 , ´ ™ `. Ž 5.7 .
It is straightforward to check that all curvatures vanish in this limit. As d ™ ` we are
moving away from the brane and we expect to recover ten-dimensional Minkowski
space as a solution. Indeed in this case we easily see that D ™ constant.
6. M-branes, M pq spaces and T-duality
We conclude with a brief discussion on higher-dimensional conifolds. Indeed one can

also consider conifolds with seven-dimensional bases in M-theory – as has been done
extensively in the eighties in compactifications of the eleven-dimensional supergravity
w20x but also very recently. We shall concentrate only on the example that is directly
related to our previous discussion, namely M 10 Žand its ‘‘T-dual’’ M 01 .. Just as T p, q
these are constructed as UŽ1. bundles w21,22x:
pq pq0
S5 = S3
M sM s . Ž 6.1 .
U Ž 1.
Since odd spheres can be thought of as UŽ1. bundles over projective spaces, the
factoring leads to identification of the two fibers in Ž6.1. and as a result M p q can be
thought of as a UŽ1. bundle over CP 2 = S 2 with the topology depending on the ratio of
p and q Žhomotopically all these spaces are the same.. While in general the isometry
group is SUŽ3. = SUŽ2. = UŽ1., we are interested in cases with larger symmetry -
SO Ž6. = SO Ž3. for M 10 s S 5 = S 2 and SUŽ3. = SUŽ2. = SUŽ2., for M 01 s CP 2 = S 3.
A cone over M 10 , C Ž M 10 ., can be almost everywhere described by a quadric in C 6
with two real planes intersecting it. Trying to resolve the singularity, we end up
replacing the apex of the cone either by S 2 factor of by S 5. Differently from the
previous case where only the deformation of C ŽT 1,1 . Žfinite size S 3 . was of interest for
us here we get a ‘‘duality’’ between the factors – when putting M-theory on the cone,
we can get a three-dimensional back hole either by wrapping M 2 on S 2 , or by the dual
procedure of wrapping M5 on S 5.
Similarly, a two-dimensional black hole can be constructed by wrapping D3 on the
shrinking S 3 in C Ž M 01 .. A circle compactification of the previous case accompanied
with T-duality should relate this to the M2 and M5 discussed above. Indeed, as known
w23,24x, T-duality untwists UŽ1. bundles interchanging the bases of the two cones. All
these cases have no supersymmetry preserved since both M 10 and M 01 admit no Killing
spinors.
Acknowledgements
We would like to thank Gregory Moore for suggesting to us the problem of D3 on a

conifold as a more tractable example of a brane on a non-trivially embedded cycle, for
suggestions on a preliminary version of this paper and for numerous useful discussions.
Helpful discussions with V. Moncrief, G. Zuckerman are also gratefully acknowledged.
This work is supported by DOE grant DE-FG02-92ER40704.
References
w2x J. de Boer, Nucl. Phys. B 548 Ž1999. 139. hep-thr9806104.
w3x J. Maldacena, A. Strominger, E. Witten, J. High Energy Phys. 12 Ž1997. 002. hep-thr9711053.
w4x R. Minasian, G. Moore, D. Tsimpis, Calabi–Yau black holes and Ž0,4. sigma models, hep-thr9904217.
w5x K. Behrndt, G. Lopes Cardoso, B. de Wit, R. Kallosh, D. Lust, T. Mohaupt, Nucl. Phys. B 488 Ž1997.
236. hep-thr9610105.
w6x A. Strominger, Nucl. Phys. B 451 Ž1995. 96. hep-thr9504090.
w7x A. Kehagias, Phys. Lett. B 435 Ž1998. 337. hep-thr9805131.
w8x I.R. Klebanov, E. Witten, Nucl. Phys. B 536 Ž1998. 199. hep-thr9807080.
w9x B.S. Acharya, J.M. Figueroa-O’Farrill, C.M. Hull, B. Spence, Branes at conical singularities and
holography, hep-thr9808014.
w10x D.R. Morrison, M.R. Plesser, Non-spherical horizons. 1, hep-thr9810201.
w11x P. Candelas, X.C. de la Ossa, Nucl. Phys. B 342 Ž1990. 246.
w12x P. Van Nieuwenhuizen, General Theory of Coset Manifolds and Antisymmetric Tensors Applied to
Kaluza–Klein Supergravity, in: Supersymmetry and Supergravity ’84, Proceedings of the Trieste Spring
School Ž1984..
w13x L. Castellani, R. D’ Auria, P. Fre, Supergravity and superstrings: A Geometric perspective ŽWorld
Scientific, Singapore, 1991..
w14x G.T. Horowitz, A. Strominger, Nucl. Phys. B 360 Ž1991. 197.
w15x M.J. Duff, R.R. Khuri, J.X. Lu, Phys. Rep. 259 Ž1995. 213. hep-thr9412184.
w16x B. de Wit, H. Nicolai, Phys. Lett. B 148 Ž1984. 60.
w17x P. van Nieuwenhuizen, N.P. Warner, Commun. Math. Phys. 99 Ž1985. 141.
w18x M.J. Duff, J.X. Lu, Phys. Lett. B 273 Ž1991. 409.
w19x D.N. Page, C.N. Pope, Phys. Lett. B 144 Ž1984. 346.
w20x M.J. Duff, B.E. Nilsson, C.N. Pope, Phys. Rep. 130 Ž1986. 1.
w22x L. Castellani, L.J. Romans, N.P. Warner, Nucl. Phys. B 241 Ž1984. 429.
w23x M.J. Duff, H. Lu, C.N. Pope, Phys. Lett. B 409 Ž1997. 136. hep-thr9704186.
w24x M.J. Duff, H. Lu, C.N. Pope, Nucl. Phys. B 532 Ž1998. 181. hep-thr9803061.
Nuclear Physics B 572 wFSx Ž2000. 517–534
Conformal field theory and the exact solution of the BCS

Hamiltonian
´ Sierra
German
´
Instituto de Matematicas ´
y Fısica Fundamental, C.S.I.C., Madrid, Spain
Abstract
We propose a connection between conformal field theory ŽCFT. and the exact solution and
integrability of the reduced BCS model of superconductivity. The relevant CFT is given by the
SUŽ2. k-WZW model in the singular limit when the level k goes to y2. This theory has to be
perturbed by an operator proportional to the inverse of the BCS coupling constant. Using the free
field realization of this perturbed Wess–Zumino–Witten model, we derive the exact Richardson’s
wave function and the integrals of motion of the reduced BCS model in the saddle point
approximation. The construction is reminiscent of the CFT approach to the Fractional Quantum
Hall effect. q 2000 Elsevier Science B.V. All rights reserved.
PACS: 11.25.Hf; 74.20.z; 04.20.Jb

Keywords: Conformal field theory; Exactly integrable models; Superconductivity
1. Introduction
The BCS theory has been used for decades to describe the superconducting properties
of ‘‘low Tc’’ metallic materials w1x. The starting point of the theory is a Hamiltonian
which describes the attractive interaction between the electrons in well defined energy
levels. The grand canonical BCS wave function gives a very accurate solution of the
BCS Hamiltonian in the limit where the number of electrons, Ne , is very large. The BCS
wave function may also be projected to a fixed number of electrons, Ne , giving
essentially the same physics when Ne 4 1. However, for small values of Ne , one has to
use exact analytical or numerical methods to obtain reliable results. The study of small
fixed-Ne superconductivity has a long story which goes back to an old question posed by
Anderson as to what is the smaller size of a metallic particle to remain superconducting
w2x. Recent experiments w3,4x involving aluminium grains with nanometer size have
inspired a number of theoretical works where Anderson’s question is reconsidered
w5–17x.
PII: S 0 5 5 0 - 3 2 1 3 Ž 0 0 . 0 0 0 3 6 - 5
518 G. Sierra r Nuclear Physics B 572 [FS] (2000) 517–534
In this paper we shall be concerned with the exact analytic solution of the reduced
BCS Hamiltonian proposed by Richardson in a series of papers between 1963 and 1977
w18–27x. This solution emerged in the framework of Nuclear Physics and has passed
unnoticed by most of the physics community until the recent upheaval in ultrasmall
metallic grains. A closely related work is that of Cambiaggio, Rivas and Saraceno ŽCRS.
who proved recently the integrability of the reduced BCS Hamiltonian without recourse
to Richardson’s solution w28x. These authors found a set of integrals of motion whose
number equals that of the degrees of freedom of the system. Our aim is to show that
Richardson’s solution, together with its integrability, can be naturally understood in the
framework of conformal field theory ŽCFT.. In other words, we shall propose a
correspondence between CFT and the BCS theory which gives a neat picture of the
Richardson’s wave function and the conserved quantities found by CRS, which are in
that sense unified in a common framework. Our work is reminiscent to the application of
CFT to the Laughlin wave function w29x of the Fractional Quantum Hall effect ŽFQHE.
w30–35x.
The organization of the paper is as follows. In Sections 2 and 3 we present brief
reviews on the exact solution of the reduced BCS Hamiltonian and the free field
realization of the SUŽ2. k-WZW model. In Section 4 we derive the exact Richardson’s
solution and we show integrability of the BCS model using the free field realization of
the SUŽ2. k-WZW model. In Section 5 we explain the BCSrCFT connection in a second
quantized language. In Section 6 we explore the analogies and differences between the
CFT approaches to BCS and the FQHE. Finally, in Section 7 we state our conclusions
and prospects of future work.
2. Review of the exact solution of the BCS model
The reduced BCS model is defined by the Hamiltonian w1,5–17x

H BCS s Ý ´ js c†js c js y gd Ý c†jq c†jy c jXy c jXq , Ž 1.
X
j, s s" j, j
where c j," Ž c†j," . is an electron destruction Žcreation. operator in the time-reversed states
< j," : with energies ´ j , d is the mean level spacing and g is the BCS dimensionless
coupling constant. The sums in Ž1. run over a set of V doubly degenerate energy levels
´ j Ž j s 1, . . . , V .. We shall assume in this paper that the energy levels are all distinct, i.e.
´ i / ´ j for i / j. The Hamiltonian Ž1. is really a simplified version of the reduced BCS
Hamiltonian where all couplings have been set equal to a single one, namely g. This is
the model that is commonly used to described ultrasmall grains, which describes the
scattering of pairs of electrons between discrete energy levels that come in time-reversed
states. Hereafter we shall refer to Ž1. simply as the BCS Hamiltonian.
As mentioned in the Introduction, Richardson had long ago solved this model exactly
for an arbitrary set of levels, ´ j , not necessarily all distinct w18–27x. To simplify matters,
we shall assume that there are not singly occupied electronic levels. As can be seen from
Ž1., these levels decouple from the rest of the system; they are said to be blocked,
contributing only with their energy ´ j to the total energy E . The above simplification
implies that every energy level j is either empty Ži.e. <vac:., or occupied by a pair of
G. Sierra r Nuclear Physics B 572 [FS] (2000) 517–534 519
electrons Ži.e. c†j,q c†j,y <vac:.. Denote the total number of electrons pairs by N. Then of
course N ( V . The most studied case in the literature corresponds to the half-filled
situation, where the number of electrons, Ne s 2 N, is equal to the number of levels V
w5–17x. In the absence of interaction Ži.e. g s 0., all the pairs occupy the lowest energy
levels forming a Fermi sea. The pairing interaction promotes the pairs to higher energies
and eventually, for large values of N, all the levels are pair correlated, giving rise to
superconductivity w1x.
2.1. Richardson’s solution
In order to describe Richardson’s solution one defines the hard-core boson operators
bj s c j,y c j,q , b†j s c†j,q c†j,y , Nj s b†j bj , Ž 2.
which satisfy the commutation relations
bj ,b†jX s d j, jX Ž 1 y 2 Nj . . Ž 3.
The Hamiltonian Ž1. can then be written as
H BCS s Ý 2 ´ j b†j bj y g Ý b†j bjX , Ž 4.
X
j j, j
where we have set d s 1 Ži.e. all the energies are measured in units of d .. Richardson
showed that the eigenstates of this Hamiltonian with N pairs have the Žunnormalized.
product form w18–21x Žfor a direct proof of the results of this subsection see Ref. w36x.
N V 1
< N :R s Ł Bn <vac: , Bn s Ý b†j , Ž 5.
ns1 js1 2 ´ j y en
where the parameters en Ž n s 1, . . . , N . are, in general, complex solutions of the N
coupled algebraic equations
1 N 2 V 1
q Ý s Ý . Ž 6.
g m s1 Ž/ n .
em y en js1 2 ´ j y en
The energy of these states is given by the sum of the auxiliary parameters en , i.e.
N
EŽ N. s Ý en . Ž 7.
ns1
The ground state of H BCS is given by the solution of Eqs. Ž6. which gives the lowest
value of E Ž N .. The Žnormalized. states Ž5. can also be written as w23x
C
< N :R s Ý c R Ž j1 , . . . , j N . b†j1 . . . b†j N <vac: , Ž 8.
'N ! j1 , . . . , j N
where the sum excludes double occupancy of pair states and the wave function c takes
the form
N 1
c R Ž j1 , . . . , j N . s Ý Ł . Ž 9.
P ks1 2 ´ jk y e P k
The sum in Ž9. runs over all the permutations, P, of 1, . . . , N. The constant C in Ž8.
guarantees the normalization of the state w23x Ži.e. R ² N < N :R s 1.; its expression will be
given in Section 4.
A well known fact about the BCS Hamiltonian is that it is equivalent to that of a XY
model with long range couplings and a ‘‘position dependent’’ magnetic field propor-
tional to ´ j . To see this let us represent the hard-core boson operators Ž2. in terms of the
Pauli matrices as follows:
bj s sq
j , b†j s sy
j , Nj s 12 Ž 1 y sz . Ž 10 .
in which case the Hamiltonian Ž4. becomes
H BCS s HX Y q Ý ´ j q g Ž Vr2 y N . ,
j
g
HX Y s y Ý 2 ´ j t j0 y Ž Tq Tyq Ty Tq . , Ž 11 .
j 2
where the matrices
V
T as Ý t ja Ž a s 0,q,y . ,
js1
t j0 s 12 sj z , tq q
j s sj , ty y
j s sj Ž 12 .
satisfy the SUŽ2. algebra
w T a ,T b x s f ca b T b , fqq0s fy0ys y1, f 0qy s 2 Ž 13 .
whose Casimir is given by
T P T s T 0 T 0 q 12 Ž Tq Tyq Ty Tq . . Ž 14 .
2.2. Integrability of the BCS Hamiltonian
From the existence of an exact analytic solution of H BCS , one may expect that H BCS
should be integrable. Indeed CRS found the integrals of motion w28x1
V tiPt j
R i s yt i0 y g Ý Ž i s 1, . . . , V . , Ž 15 .
jŽ/i .
í y ´j
where the denominator does not blow up since we are assuming non-degenerate energy
levels. Integrability amounts to the equations
w H BCS , R i x s R i , R j s 0 Ž i , j s 1, . . . , V . . Ž 16 .
Denote the eigenvalue of R j acting on the state Ž8. by l j , namely
R j < N :R s l j < N :R . Ž 17 .
CRS, seemingly unaware of Richardson’s solution, did not give an expression of l j in
1
There is a sign difference between Eq. Ž15. and the one used by CRS in w28x. This is due to our convention
for t i0 , which is opposite from that of CRS.
their work. However, they did show that HX Y given in Eq. Ž11. can be expressed in
terms of the operators R i as
2
HX Y s Ý 2 ´ j R j q g ž /
Ý Rj y 34 g V . Ž 18 .
j j
Hence, combining Ž17. and Ž18. one can find the eigenvalues of HX Y
2
EX Y s Ý 2 ´ j l j q g ž /
Ý lj y 34 g V Ž 19 .
j j
and in turn those of EBCS by recourse to Eq. Ž11.. We shall show in Section 4 that l j
has the simple expression
1 N 1 1 V 1
li s y q g
2 ž Ý
ns1 2 ´ i y en
y
4
Ý
js1 Ž /i . í y ´j / . Ž 20 .
One can check this result by deriving the energy Ž7. from Eqs. Ž11., Ž19. and Ž20..
This ends the presentation of the exact solution of the BCS Hamiltonian. The
existence of an underlying analytic structure reminiscent to that of a CFT is apparent
from Eqs. Ž9., Ž15. and Ž20.. Indeed, the aforementioned equations contain factors of the
form 1rŽ z y zX . where z and zX stands for either 2 ´ j or en . Terms of this sort arise
quite naturally as correlators Ži.e. ² AŽ z . B Ž zX .: s 1rŽ z y zX .. of chiral primary fields
AŽ z . and B Ž z . in diverse CFT’s. The problem is to identify which CFT explains all the
features presented so far in a unified manner. We shall argue that the solution of this
problem is given by the Wess–Zumino–Witten ŽWZW. model based on the affine
Kac–Moody group SUŽ2. k in the limit where the level k goes to y2. The proof of this
result requires standard tools of CFT and, more precisely, the free field or Coulomb Gas
ŽCG. representation of the WZW model.
3. Review of the free field representation of the SU(2)k-WZW model
The material presented in this section is standard in the CFT literature w37,38x.
Nevertheless, we have included it for the benefit of readers that are not experts in CFT.
This will allows us to highlight the main tools we shall use in later sections. We shall
follow closely Ref. w39,40x.
The WZW model is an interacting theory which nevertheless admits a description in
terms of free fields w41–45x. The correlators and conformal blocks can then be easily
calculated as integrals of vacuum expectation values of vertex operators. This gives an
integral representation of the conformal blocks which satisfy automatically the Knizh-
nik–Zamolodchikov ŽKZ. equations w46x. In the case of the SUŽ2. k-WZW model the free
fields are a b y g system with conformal weights 1 and 0, respectively, and a boson
field w which satisfy the following operator product expansion ŽOPE. w39–45x:
1
b Ž z . g Ž w . s yg Ž z . b Ž w . s ,
zyw
w Ž z . w Ž w . s yln Ž z y w . . Ž 21 .
The WZW currents J a Ž z .Ž a s 0," . can be expressed in terms of these fields as

Žhereafter, normal order of operators will be implicitly assumed.
Jqs i b ,
i
J 0sy Ew y bg ,
2 a0
i
Jys i bg 2 q gEw y k Eg , Ž 22 .
a0
which satisfy the OPE’s
kr2 1
J aŽ z. J bŽ w. s 2
q ab q f ca b J c Ž w . q reg. terms, Ž 23 .
Ž zyw. zyw
where f ca b are the SUŽ2. structure constants defined in Eq. Ž13., and q 00 s 1, qqys
qyqs 2. The level of the WZW model, k, is related to the ‘‘charge’’ a 0 by
1
kq2s . Ž 24 .
2 a 02
If k is a positive integer, the WZW model is a Rational Conformal Field Theory
ŽRCFT. with k q 1 primary fields labelled by the total spin, j s 0,1r2, . . . ,kr2. In our
CFT approach to BCS, we shall need to consider the limit where k ™ y2, which
corresponds to taking a 0 ™ `. This is a singular limit which takes us away from the
rational WZW models. Actually, the case when k is exactly y2 is mathematically
interesting due to its relation to the singular hyperplanes in the representation theory of
affine Kac–Moody algebras w47,48x. For non-positive integer values of k, we can still
define the theory by the free field representation given above.
The Sugawara energy–momentum tensor TSug of the WZW model is given by the
sum of the energy momentum tensors of the b – g system and the bosonic field w
2
TSug s bEg y 12 Ž Ew . q i a 0 E 2w . Ž 25 .
The central extension c of the Virasoro algebra generated by the modes of TSug Žs
Ý n L n zyn y2 . is
3k
c s 3 y 12 a 02 s , Ž 26 .
kq2
where the b y g system contributes with 2 and the field w contributes with 1 y 12 a 02 .
In Ž26. we have used the relation Ž24., which for integer k’s gives the well known value
of the Virasoro central charge of the SUŽ2. k-WZW model. In the limit Ž k q 2. ™ 0 the
central extension c diverges. In order to get a meaningful theory one has to scale the
Virasoro operators as L˜ n s lim k ™y2 Ž k q 2. L n . In that limit the Virasoro algebra
becomes
L˜ n , L˜ m s 0, Ž 27 .
which suggests some sort of integrability. In fact the commutativity of the Virasoro
operators L˜ n has been used to study the representation theory of the SUŽ2. ksy2
Kac–Moody algebra for the proof of the Kac–Kazhdan conjecture w47x concerning
character formulas Žsee Ref. w48x for references..
The primary fields Fmj Ž z . of the WZW model are labelled by the total spin
j s 0,1r2, . . . and the third component of the spin m s j, . . . ,y j. Their free field
representation is given by
Fmj Ž z . s g jym Ž z . Va jŽ z . , Ž 28 .
Va jŽ z . s e i a j w Ž z . , a j s y2 a 0 j
and have a conformal weight D j given entirely by that of the vertex operator Va j ,
namely
j Ž j q 1.
D j s 12 a j Ž a j y 2 a 0 . s . Ž 29 .
kq2
In the free field representation of a CFT, every primary field has a conjugate version
besides its ‘‘direct representative’’, which is needed for the computation of correlators
w49,50x. In the case of the WZW model, the conjugate of the primary field with m s j is
F˜ jj Ž z . s b sq2 j Ž z . V2 a 0 Ž sqj. Ž z . , s s y Ž k q 1. . Ž 30 .
The corresponding equation for m - j is much more complicated, and it is a sum of
terms where the difference between b fields and g fields is given by s q j q m.
A particular case of Ž30. is when j s 0, which corresponds to the conjugate field of
the identity
I˜Ž z . s b s Ž z . V2 a 0 s Ž z . . Ž 31 .
A consequence of Ž31. is that the expectation values of operators should satisfy the
following charge neutrality conditions:
Nb y Ng s s, Ý a i s 2 a 0 s, s s y Ž k q 1. , Ž 32 .
i
where Nb and Ng is the number of b and g fields in the correlator, and a i are the
charges of the vertex operators made of the field w . Eq. Ž32. means that there is a
background charge y2 a 0 s in the boson sector and a charge ys in the b – g sector,
which need to be neutralize for the correlator to be non-vanishing. The latter properties
can alternatively be attributed to the out vacuum which have charges y2 a 0 and ys in
the w and b – g sectors, respectively.
The remaining ingredient of the free field representation is provided by the so-called
screening charge
Qs ECdu S Ž u . , S Ž u . s b Ž u . V2 a 0 Ž u . , Ž 33.
whose basic property is that it commutes with the SUŽ2. current algebra and the
Virasoro operators. In Eq. Ž33. and below, du is meant to contain the factor 1rŽ2p i. to
take care of the factor 2p i that comes out in the residue formula. Using the vertex
representations Ž28. and Ž30. of the primary fields, together with the screening charge
Ž33., one can compute the conformal blocks of the WZW model. Conformal blocks are
the chiral building blocks of correlators. The latter are obtained by combining the
holomorphic and the anti-holomorphic conformal blocks and imposing monodromy
invariance. In the WZW model a conformal block c WZ W Ž z 1 , . . . , zV q1 . involving V q 1

Vq1 Vq1
primary fields Ž jk ,m k .4ks1 , inserted at the positions z k 4ks1 , can be associated with
the SUŽ2. tensor product decomposition
V
j1 m . . . m jV ™ jV q1 , Ý m k s mVq1 , Ž 34 .
ks1
where Ž jVq1 ,mV q1 . appears as an outgoing state. The free field expression of the
conformal blocks is given by
c WZ W Ž z 1 , . . . , zV q1 .
¦
s Fmj11 Ž z 1 . . . . FmjVV Ž zV . F˜mjVVqq11 Ž zV q1 . EC du S Ž u . . . . EC du
1
1 1
N
NS ;
Ž uN . , Ž 35 .
where F˜mjVVq1
q1 is the conjugate of the outgoing state Ž j
V q1 ,mV q1 , and the screening
.
charges are integrated along the contours C1 , . . . ,CN . The charge neutrality conditions
Ž32. applied to Ž35. yield
V V
Ns Ý jk y jV q1 , mV q1 s Ý mk , Ž 36 .
ks1 ks1
which agree with the Clebsch–Gordan decomposition Ž34.. The case when N s 0
corresponds to the maximal allowed value of jVq1 s Ý V ks1 j k . On the other hand, if V
is even, the minimal value of jVq1 is zero, which requires N s Ý V ks1 j k screening
charges. Hence, the different choices of the screening charges and contours give rise to
all possible conformal blocks. In this manner the free field representation provides
integral solutions of the KZ equations satisfied by the conformal blocks Ž35.. The KZ
equations are w46x
E Vq1 tiPt j
ž k
E zi
y Ý js1
Ž /i .
zi y z j / c WZW Ž z 1 , . . . , zV q1 . , Ž 37 .
where k s Ž k q 2.r2 and t i are the SUŽ2. matrices in the ji representation acting at the
ith site.
4. CFT representation of the exact solution of BCS
Our first aim is to obtain Richardson’s wave function Ž9. using the free field
representation of the WZW model.
4.1. Richardson’s waÕe function
The starting point is the pseudospin version of the BCS model introduced in Section
2, according to which an empty energy level ´ i Ž i s 1, . . . , V . has spin m i s 1r2 while
and occupied level has spin m i s y1r2. This suggests to rewrite Richardson’s wave
function as
N 1
cmR1 , . . . , m V Ž z 1 , . . . , zV ;e1 , . . . ,e N . s Ý Ł , Ž 38 .
P ks1 zlk y e P k
where z i s 2 ´ i and l k is defined by the condition m l k s y1r2. The SUŽ2. quantum

numbers m i of Ž38. satisfy
V V
Ý mi s 2 y N. Ž 39 .
is1
Let us compare now the conformal block Ž35. and the wave function Ž38.. If we take
jk s 1r2 Ž k s 1, . . . , V . in Ž35. and use Ž39., we are led to the identifications
V
jVq1 s mV q1 s y N, Ž 40 .
2
which requires N ( Vr2. Hence from a formal point of view, we can regard Richard-
son’s wave function as a conformal block involving V primary fields Fm1r2 j
Ž z j ., located
at the positions z j s 2 ´ j , and a primary field F˜mjVVq1
q1 Ž z
V q1
. whose position we shall
place at `. The last ingredient we need, in order to reproduce Richardson’s wave
function using CFT tools, is to find the role played by the BCS coupling constant g. We
shall see below that g is associated to the operator
i a0
Vg s exp y
ž g
EC dz z Ew Ž z .
g
/ . Ž 41 .
Our claim is that Richardson’s wave function Ž38. is given, up to a proportionality

factor, by the limit
cmR Ž z,e . A lim cmCG Ž z . , Ž 42 .
a 0™`
where the Coulomb Gas wave function c CG is given by the following expectation value
¦
cmCG Ž z . s VgFm 11 2 Ž z 1 . . . . Fm1 V2 Ž zV . F˜ jjVVqq11 Ž ` . EC du S Ž u . . . . EC du
1
1 1
N
NS ;
Ž uN . .
Ž 43 .
Except for the presence of the operator Vg , Eq. Ž43. coincides with the conformal block
Ž35.. Let us now prove Eq. Ž42.. First of all, using the free field representation of the
primary fields, one can write Ž43. as
w
cmCG Ž z . s EC du 1... EC du Nc Ž z,u . cmbg Ž z,u . , Ž 44 .
1 N
where
V N
¦
c w s Vg Ł Vy a 0Ž z i .
is1
Ł V2 a Ž un . V2 a Ž sqj
n s1
0 0 Vq1 . ;
Ž `. , Ž 45 .
V 1 N
ym i
cmbg s ¦ Łg
is1
2
Ž zi . Ł b Ž un . b sq2 j Ž `. g 2 j Ž `.
n s1
Vq1 Vq1
; .
Using Eqs. Ž21. and the Wick theorem, one can show that c bg is, up to a sign, equal to
Richardson’s wave function, namely
N
cmR Ž z,e . s Ž y1 . cmbg Ž z,e . . Ž 46 .
In this equation we have chosen the positions of the screening operators u i equal to
the Richardson’s parameters e i . However, in Eq. Ž44. one must integrate over the u’s.
The job of the limit a 0 ™ ` is to set u i s e i . Let us see how this happens. First of all
the contribution of the vertex operators can be computed using the formula
ai a j
² Vg Va Ž w 1 . . . . Va Ž wM . : s Ł Ž wi y wj .
1 M
Ł eya 0 a jw j r g , Ž 47 .
i-j j
where the contour C g encircles all the coordinates wj . The factor c w then becomes
2 2 2 2
c f Ž z,u . s Ł e a 0 z j r g Ł ey2 a 0 un r g Ł Ž z i y z j . a 0 Ł Ž un y um . 4a 0
j n i-j n- m
2
= Ł Ž z i y un . y2 a 0 . Ž 48 .
i ,n
In the limit a 0 ™ `, the integral Ž44. can be computed using the saddle point method.
Indeed, writing Ž48. as
2
c w Ž z,u . s eya 0 UŽz ,u . , Ž 49 .
where
V N V N
U s y Ý ln Ž z i y z j . y 4 Ý ln Ž un y um . q 2 Ý Ý ln Ž z i y un .
i-j n- m is1 n s1
1 V N
q
g ž y Ý z i q 2 Ý un
is1 n s1
/ , Ž 50 .
the stationary solutions of U are given by the solutions of Richardson’s equation Ž6.,
namely
EU 1 N 2 V 1
0s
ž /
E un umsem
™
g
q Ý
m s1 Ž/ n .
em y en
s Ý
js1 z j y en
. Ž 51 .
Under these conditions the saddle point value of the integral Ž44. is
1 2
cmCG Ž z . ; N 1r2
ey a 0 UŽz ,e. cmbg Ž z,e . , Ž 52 .
1r2
Ž 2p a 0 . Ž det A .
where A is the N = N hessian matrix defined as
1 EU
A m ,n s y , Ž 53 .
2 E um E un
use
1 2 2
An , n s Ý 2
yÝX 2
, Am , n s 2
, Ž m/n . .
i Ž z i y en . n Ž en y em . Ž en y em .
Eqs. Ž46., Ž51. and Ž52. constitute the proof of Ž42.. Moreover, the factor Ždet A.y1 r2
in Ž52. turns out to coincide with the normalization constant C appearing in the
normalized state Ž8. w23x, namely
1r2
C s 1r Ž det A . . Ž 54 .
Let us make some comments about the results presented so far.

v Richardson has observed that Eqs. Ž6. could be derived as the stationary
configurations for a set of electrostatic charges in 2D w27x. The electrostatic potential of
these charges is given by a 02 U or rather by the sum a 02 ŽU q U ) .. Our CFT derivation
of the exact BCS solution shows that the value of the charges are ya 0 for the energy
levels, 2 a 0 for the screening ones, and a charge 2 a 0 Ž s q jV q1 . placed at infinity. The
sum of all these charges neutralizes the background charge y2 a 0 s. The stationary
conditions are set by the limit a 0 ™ `, which is equivalent to the limit k ™ y2.
v The operator Vg , which is needed in order to get the term 1rg in the Richardson’s
equations, breaks conformal invariance in an explicit manner. This is why c CG is not,
strictly speaking, a conformal block of the WZW model. In the spirit of Perturbed
Conformal Field Theory w51,52x we could say that the WZW model has been perturbed
by the chiral operator y i a 0 Edz z Ew which is equal to ya 0 ay1rg, where ay1 is the
g
n s y1 mode of the field Ew Ži Ew Ž z . s Ý n a n zyn y1 ..
v The perturbative renormalization group ŽRG. analysis of the BCS model yields
that the coupling constant g ) 0 flows to large values in the infrared regime, leading to
the superconducting instability of the Fermi sea w53x. In our approach the fixed point of
this RG flow is described by the SUŽ2. ks y2-WZW model. The 1rg perturbation takes
us away from this fixed point, breaking conformal invariance. In that sense, we are
dealing with the strong coupling version of BCS, which is valid for all values of the
BCS coupling constant g, since there are not phase transitions as we go from weak to
strong coupling.
v So far, we have assumed for simplicity that the pair degeneracy of the energy
levels ´ j is unity, which corresponds in the WZW model to primary fields with spin
jk s 1r2. The construction can be straightforwardly generalized to levels with higher
degeneracy, d k s 1,2, . . . , in which case the associated primary fields have spin jk s
d kr2. In this case the Richardson equations read w26,27x
1 N 2 V dj
q Ý s Ý , Ž 55 .
g m s1 Ž/ n .
em y en js1 z j y en
and they can be derived using the free field representation explained above in terms of
the fields Fmjkk Ž z k . with jk s d kr2.
4.2. The integrability of BCS from CFT
At this stage it is clear that Eqs. Ž15. and Ž17. must be related to the KZ equation
Ž37.. After all, in the limit a 0 ™ ` the CG wave function Ž43. coincides with the WZW
wave function. Let us show this in detail. The KZ equation Ž37. with zVq1 s ` and the
definition Ž15. can be written as
E 1 1
ž k
E zi
q
2g
Ri q
2g /
t i0 c WZW s 0. Ž 56 .
Let us define a new wave function c

c WZ W s e H X Y r2 g kc . Ž 57 .
Using the commutativity of R i and HX Y and the fact that

E
H s yt i0 Ž 58 .
E zi X Y
Eq. Ž56. becomes
1 E
ž 2g
Ri qk
E zi / c s 0. Ž 59 .
This equation is completely equivalent to the KZ equation Imposing that c diagonal-

izes R i with eigenvalue l i , c becomes proportional to the Richardson solution, or
rather to c CG . Hence Eq. Ž49. yields the following expression for l i
g EU
li s , Ž 60 .
2 E zi
which in turn leads to the formula Ž20.. In other words, c WZ W , c CG and c R agree up
to overall factors in the limit when a 0 ™ `. It is interesting to observe that the factor
relating the WZW and Richardson wave functions has indeed a conformal structure
reflected solely in the presence of logarithms, namely
2
c WZ W s exp a 02 ½ ž g
EX Y y U
/5 c R, Ž 61 .
2 EX Y
y U s Ý ln Ž z i y z j . q 4 Ý ln Ž un y um . y 2 Ý ln Ž z i y un . .
g i-j n- m i ,n
The Coulomb energy U contains in a disguised manner both Richardson’s equations

and the constants of motion l i .
5. The BCS r CFT correspondence
In Section 4 we have shown the closed relationship between the BCS and the CFT
wave functions. In this section we want to investigate this relation at the second
quantized level. In a quantum field theory a generic N-body wave function C Ž x 1 , . . . , x N .
is usually constructed from the overlap of an N-body state <C : with the eigenstates
< x 1 , . . . , x N : created by the action of the field operator cˆ Ž x . acting on the Fock vacuum
<0:, namely
C Ž x 1 , . . . , x N . s ²C < x 1 , . . . , x N : < x 1 , . . . , x N : s cˆ Ž x 1 . . . . cˆ Ž x N . <0: .
Ž 62 .
An interesting example of this formalism is provided by the CFT interpretation of the
Laughlin wave function w29x of the Fractional Quantum Hall effect ŽFQHE., first
proposed by Fubini w30–35x. In that case cˆ Ž x . is a vertex operator of a single boson
governed by a c s 1 CFT. One can also make use of an array of bosons.
In the spirit of the FQHErCFT correspondence, one can interpret the Coulomb Gas
wave function Ž43. as
cmCG Ž z . s ² c CG < z 1 ,m1 , . . . , zV ,mV : , Ž 63 .
where
V
< z 1 ,m1 , . . . , zV ,mV : s Ł Fm 1 2 Ž z i . <0: , Ž 64 .
i
is1
N
² c CG < s ² a 0 Ž V y 2 N . < Vg Ł E du S Ž u . .
n n
n s1 Cn
The out vacuum ² a 0 Ž V y 2 N .< has been defined as the action of the operator F˜mjVVq1 q1 on
the out vacuum of the WZW model, and has a charge a 0 Ž V y 2 N . in the w sector and
no charge in the b – g sector. The different states ² c CG < correspond to different choices
of the integration contours Cn of the screening operators SŽ un ., which in CFT yield
different conformal blocks. For example, the ground state is obtained by choosing the
n th -contour Ž n s 1, . . . , N . to run from zn to infinity, where we assume that the energy
levels are ordered in increasing order, i.e. z 1 - z 2 - . . . - zV . The excited states
correspond to other contour choices. The total number of eigenstates of H BCS with N
V
pairs and V levels, which is given by the combinatorial number
the total number of contour choices.
N ž /
, coincides with
Eqs. Ž63. and Ž64. provide the basic correspondences between the exact solution of
BCS and CFT, which can be extended to other instances. We collect them in Table 1
and make some comments below.
v In analogy to Eqs. Ž64. one could try to define the bra state² z 1 ,m1 , . . . , zV ,mV <
using the conjugate operators F˜m1 2Ž z .. However, this state would have a large charge
which would be difficult to compensate for. Similarly, the ket state < c CG : is not easy to
construct since the screening operator SŽ z . does not have a conjugate version. A
possibility would be to use the second screening operator, given by S˜Ž z . s
Table 1
The BCSrCFT correspondence
BCS CFT
Pair energy level WZW Primary field
Pair degeneracy Ž d k . Total spin Ž jk s d k r2.
Eigenstates of H BC S Conformal blocks
Richardson’s eqs Saddle point conditions
Integrability ŽCRS. KZ eqs.
g s` Ž g finite. WZW ŽPerturbed WZW.
Cooper pair operator Screening operator
Phase stiffness A a 02
Empty, occupied level spin up, down
N † < V
Ł ks 1 bj k vac R
: Ł is1 Fm1r2
i
Ž z i .<0:
²N< ² c CG <
cR c bg
bj ,b†j , 12 y Nj Ez dzJq Ž z ., Jy Ž z ., J 0 Ž z .
j
C 1r'det A
g E UC
li
2 E zi
Ri y gL˜Žy1
i.
HX Y y gL˜ 0
b sy 1 expŽyi w Ž z .ra 0 . w39,40x. This operator appears in the construction of fractional

spin representations which are related upon a certain reduction to the minimal models. In
the limit a 0 ™ ` we see that S˜Ž z . converges to the identity. Further work is needed to
clarify this issue.
v In Section 3 we defined the Virasoro operators L˜ n as the singular limit
lim k ™y2 Ž k q 2. L n . From Eq. Ž59. we see that R i could be identified with the action of
ygL˜Žy1 i.
on c CG , where L˜Žy1i.
s ErE z i . Similarly, Eq. Ž18. implies that HX Y could be
identified, up to constants, with yg Ž k q 2.Ý j z j ErE z j and thus with ygL˜ 0 . Upon these
identifications, the integrability of the BCS model, given by Eqs. Ž16., becomes
equivalent to the commutativity Ž27. of the Virasoro operators L˜ 0 and L˜Žy1 i.
. We may
expect the existence of another CRS-like integrals of motion associated to the Virasoro
operators L˜ nŽ n ( y2..
v The Richardson’s state R ² N < corresponds, in the CFT formulation, to the state
² c CG <, which is the product of N screening operators acting on the out vacuum
² a 0 Ž V y 2 N .<. This correspondence suggests that in the grand canonical Žg.c.. ensem-
ble Žwhere the number of pairs is not fixed. the corresponding state should be given by
² cgCG < ² <

.c . s 0 exp žE /
dz S Ž z . . Ž 65 .
ŽNote that we have assumed the half filled condition V s 2 N.. This state is similar to
the BCS state given by w1x
² BCS < s ²vac <exp žÝg b /,k k Ž 66 .

k
where g k is the ratio Õ kru k of the BCS variational parameters. In this sense, the
screening charge EdzSŽ z . is the CFT version of the Cooper pair operator Ý k g k b k . In
CFT it has been argued w49,50x that the screening operators can be exponentiated into
the action and that their number is fixed upon imposing the charge neutrality conditions
on correlators. This CFT exponentiation corresponds to working in the grand canonical
ensemble in BCS.
v In the previous item we argued that SŽ z . s b Ž z .expŽ2 a 0 i w Ž z .. is the CFT
analogue of g k b k . On the other hand, b k corresponds to the contour integration of
JqŽ z . s i b Ž z . around the pair energy z k s 2 ´ k . Hence it is natural to associate the BCS
variational parameter g k with expŽ2 a 0 i w Ž z ... This means that 2 a 0 w Ž z . can be
associated to the phase of the superconducting order parameter. Shifting w Ž z . by a
constant leads to an overall phase shift of the BCS order parameter. This correspondence
yields an insight about the physical meaning of a 02 , which seems to be related to the
phase stiffness or the superfluid density n s . Indeed, if we identify the phase of the
superconducting order parameter u with a 0 w then the Lagrangian of w becomes that of
a continuum XY model for u , with a 02 playing the role of the superfluid density.
Actually a 02 appears in the denominator of the Lagrangian while n s appears in the
numerator. However, recall that we are working in the energy space so things are
inverted. The identification of u with a 0 w is also consistent with the fact that both
variables are defined modulo 2p . The limit a 0 ™ ` therefore corresponds to the limit of
very large phase stiffness which in fact leads to the standard BCS theory, where the
phase of the superconducting order parameter is rigid and plays no role in fixing the
critical temperature or other observables w54–56x. As was shown by Richardson w27x,

Eqs. Ž6. reduce in the bulk limit N ™ ` to the BCS gap equation and hence the state Ž8.
becomes the fixed N projection of the mean field g.c. BCS state. Finite values of a 0
should lead to non-mean field theories with the phase u playing a dynamical role. It is
rather intriguing that models of this sort have already been proposed by several authors
for an explanation of high-Tc superconductivity w54–56x.
6. Comparison between BCS and the FQHE
In Sections 5 and 6 we noticed some analogies between the CFT approaches to BCS
and the FQHE. Let us consider them in some more detail. A common feature is the
Coulomb Gas treatment. In the FQHE the CG is associated to the Laughlin wave
function of Ne electrons at filling factor n s 1rm, w29x
Ne 1
m 2
c LŽ w 1 , . . . ,wNe . s Ł Ž wi y wj . ey 4 Ý < w l < . Ž 67 .
i-j l
The norm of Ž67. can be seen as a classical probability distribution eyb UL of a

two-dimensional one-component plasma at fictitious temperature b s 1rm and potential
energy UL where w29x
Ne
m
UL s y2 m2 Ý ln < wi y wj < q
2 l
Ý < wl < 2. Ž 68 .
i-j
The particles with charge m repel each other with a logarithmic interaction, and they are
attracted to the origin by an uniform neutralizing background charge with density
r s 1rŽ2p l B2 ., where l B is the magnetic length, which has been set equal to one in
Ž67. and Ž68.. For small values of mŽs 3,5, . . . . the electrons form a liquid with uniform
density r e s 1rŽ2p m l B2 . which neutralizes the background charge. However, for large
values of m, Quantum Monte Carlo studies have shown that the Laughlin liquid
becomes a solid Ži.e. a Wigner crystal. where the positions of the charges are localized.
The comparison between the wave function c w given in Eq. Ž48. and the Laughlin
wave function c L suggests a formal identification of the electron positions wj with the
screening positions u j rather than with the pair energies levels z j . The reason is that
both the u’s and the w’s are subject to integration, while the z’s are held fixed.
Following this analogy, we may establish the relations
kq2
m s 4a 02 or n s , Ž 69 .
2
according to which the freezing of the screening charges in the limit a 0 ™ ` would
parallel the Wigner crystal structure of the FQHE when m is large. On the contrary, for
finite values of a 0 the screening charges, which are essentially Cooper pairs, would
delocalize becoming a sort of liquid. The discussion at the end of Section 5 suggests that
this liquid should arise from the fluctuations of the phase of the superconducting order
parameter.
Besides these analogies, the Laughlin and BCS Coulomb Gas models differ in the
nature of the background charge. In the Laughlin case the charge is two-dimensional,
while in the BCS case the linear terms appearing in U Žsee Eq. Ž50.. can be attributed to
a linear uniform density r l A a 0rg placed at infinity. The latter density creates a
uniform electric field Ex ŽU q U ) . along the x s Ž z q z ) .r2 axis. Another difference is
that the BCS theory is not really conformal invariant for finite values of g while the
Laughlin state has gapless edge excitations described by CFT w57x.
7. Conclusions and prospects

In this paper we have established a closed relationship between the exact solution of
the BCS Hamiltonian and the Coulomb Gas version of the SUŽ2. k-WZW model in the
singular limit when k ™ y2. The Richardson’s wave function comes from the b y g
chiral correlators, while the Richardson’s equations and the normalization factor of the
state arises from the saddle point evaluation of the chiral boson correlators. The BCS
coupling constant g enters the construction as a perturbation of the WZW model and
breaks conformal invariance. The integrability of the BCS model is related to that of the
WZW model through the KZ equations, which has lead us to an expression of the
integrals of motion of the BCS model found by Cambiaggio, Rivas and Saraceno. We
have proposed a BCSrCFT correspondence which, in many respects, parallels the CFT
interpretation of the Fractional Quantum Hall effect. We have conjectured that the
singular limit a 02 ™ ` amounts in physical terms to the limit of very strong phase
stiffness, which leads to the mean field BCS theory. Finite a 0 generalizations of the
BCS model may correspond to non-mean field theories where the phase of the
superconducting order parameter should have a dynamical role as in some models of
high-Tc superconductivity w54–56x.
Besides giving new insights into the exact solution of the BCS model, the CFT
approach may also help in solving some problems as the computation of observables
with the Richardson’s exact solution. The finite temperature BCS model is also an
interesting problem which one may try to address with CFT tools.
Finally, the BCSrCFT approach can be straightforwardly generalized to any WZW
model based on an affine Kac–Moody algebra G k , where k is the level and G is a
semi-simple Lie group or supergroup. As in the SUŽ2. case, one can use the free field
realization of these models w41–45,58–60x. For the G k-WZW model the singular limit is
given by k q h ™ 0, where h is the dual Coxeter number of G w48x. The charge and spin
independent pairing Hamiltonians studied by Richardson w26,27x in the context on
Nuclear Physics probably belong to this category of models.
Note added
After completion of this work we have been informed by A. Belavin about some
related work by Babujian w61,62x, where he applies the Bethe ansatz and the Knizhnik–
Zamolodchikov equations to Gaudin magnets w63x. Gaudin’s model is given essentially
by the g ™ ` limit of the reduced BCS model. In fact Gaudin’s Hamiltonians can be
identified with lim g ™` R jrg, where R j are the CRS conserved quantities define in Eq.
Ž15..
Acknowledgements
I want to thank especially J. Dukelsky for many discussions on the subject. I also
´
want to acknowledge conversations with A. Belavin, E.H. Kim, M.A. Martin-Delgado,
A. Ramallo and J. von Delft. This work was supported by the DGES Spanish grant
PB97-1190.
References
w1x J. Bardeen, L.N. Cooper, J.R. Schrieffer, Phys. Rev. 108 Ž1957. 1175.
w2x P.W. Anderson, J. Phys. Chem. Solids 11 Ž1959. 28.
w3x D.C. Ralph, C.T. Black, M. Tinkham, Phys. Rev. Lett. 76 Ž1996. 688.
w4x D.C. Ralph, C.T. Black, M. Tinkham, Phys. Rev. Lett. 78 Ž1997. 4087.
w5x J. von Delft et al., Phys. Rev. Lett. 77 Ž1996. 3189.
w6x F. Braun et al., Phys. Rev. Lett. 79 Ž1997. 921.
w7x F. Braun, J. von Delft, Phys. Rev. B 59 Ž1999. 9527.
w8x R.A. Smith, V. Ambegaokar, Phys. Rev. Lett. 77 Ž1996. 4962.
w9x K.A. Matveev, A.I. Larkin, Phys. Rev. Lett. 78 Ž1997. 3749.
w10x A. Mastellone, G. Falci, R. Fazio, Phys. Rev. Lett. 80 Ž1998. 4542.
w11x S.D. Berger, B.I. Halperin, Phys. Rev. B 58 Ž1998. 5213.
w12x F. Braun, J. von Delft, Phys. Rev. Lett. 81 Ž1998. 4712.
w13x J. Dukelsky, G. Sierra, Phys. Rev. Lett. 83 Ž1999. 172.
w14x J. Dukelsky, G. Sierra, cond-matr9906166 Žto appear in Phys. Rev. B..
w15x J. Dukelsky, P. Schuck, Phys. Lett. B 464 Ž1999. 164.
w16x F. Braun, J. von Delft, Advances in Solid State Physics, ed. B. Kramer ŽVieweg, Braunschweig, 1999. p.
341 Žcond-matr9907402..
w17x G. Sierra, J. Dukelsky, G.G. Dussel, J. von Delf, F. Braun, cond-matr9909015.
w18x R.W. Richardson, Phys. Lett. 3 Ž1963. 277.
w20x R.W. Richardson, N. Sherman, Nucl. Phys. 52 Ž1964. 221.
w21x R.W. Richardson, N. Sherman, Nucl. Phys. 52 Ž1964. 253.
w23x R.W. Richardson, J. Math. Phys. 6 Ž1965. 1034.
w24x R.W. Richardson, Phys. Rev. 141 Ž1966. 949.
w27x R.W. Richardson, J. Math. Phys. 18 Ž1977. 1802.
w28x M.C. Cambiaggio, A.M.F. Rivas, M. Saraceno, Nucl. Phys. A 624 Ž1997. 157.
w29x R. B Laughlin, Phys. Rev. Lett. 50 Ž1983. 1395.
w30x S. Fubini, Mod. Phys. Lett. A 6 Ž1991. 347.
w31x ¨
S. Fubini, C.A. Lutken, Mod. Phys. Lett. A 6 Ž1991. 487.
w32x G.V. Dunne, A. Lerda, C.A. Trugenberger, Mod. Phys. Lett. A 6 Ž1991. 2819.
w33x C. Cristofano, G. Maiella, R. Musto, F. Nicomedi, Mod. Phys. Lett. A 6 Ž1991. 1779.
w34x C. Cristofano, G. Maiella, R. Musto, F. Nicomedi, Mod. Phys. Lett. A 6 Ž1991. 2985.
w35x A. Cappelli, G.V. Dunne, C.A. Trugenberger, G.R. Zemba, Nucl. Phys. B 398 Ž1993. 531.
w36x J. von Delft, F. Braun, cond-matr9911058.
w37x A. Belavin, A.M. Polyakov, A. Zamolodchikov, Nucl. Phys. B 241 Ž1984. 33.
w38x For a review see P. Ginsparg, Les Houches Lectures 1988 ŽNorth-Holland, Amsterdam, 1990..
w39x Vl.S. Dotsenko, Nucl. Phys. B 338 Ž1990. 747.
w40x Vl.S. Dotsenko, Nucl. Phys. B 358 Ž1991. 547.
w41x M. Wakimoto, Comm. Math. Phys. 104 Ž1986. 605.
w42x A.B. Zamolodchikov, talk given at Montreal Ž1988., unpublished.
w43x A. Gerasimov, A. Marshakov, A. Morozov, M. Olshanetskii, S. Shatashvili, Int. J. Mod. Phys. A 5 Ž1990.
2495.
w44x B.L. Feigin, E.V. Frenkel, Comm. Math. Phys. 128 Ž1990. 161.
w45x P. Bouwknegt, J. McCarthy, K. Pilch, Phys. Lett. B 234 Ž1990. 297.
w46x V.I. Knizhnik, A.B. Zamolodchikov, Nucl. Phys. B 247 Ž1984. 83.
w47x V.G. Kac, D.A. Kazhdan, Adv. Math. 34 Ž1979. 97.
w48x B.L. Feigin, E.V. Frenkel, in: Memorial Volume for V. Knizhnik ŽWorld Scientific, Singapore, 1990..
w49x Vl.S. Dotsenko, V.A. Fateev, Nucl. Phys. B 240 Ž1984. 312.
w50x Vl.S. Dotsenko, V.A. Fateev, Nucl. Phys. B 251 Ž1985. 691.
w51x A.B. Zamolodchikov, Int. J. Mod. Phys. A 3 Ž1988. 743.
w52x A.B. Zamolodchikov, Int. J. Mod. Phys. A 4 Ž1989. 4235.
w53x R. Shankar, Rev. Mod. Phys. 66 Ž1994. 129.
w54x S. Doniach, M. Inui, Phys. Rev. B 41 Ž1990. 6668.
w55x V.J. Emery, S.A. Kivelson, Nature 374 Ž1995. 434.
w56x V.J. Emery, S.A. Kivelson, Phys. Rev. Lett. 74 Ž1995. 3253.
w57x X.G. Wen, Phys. Rev. Lett. 64 Ž1990. 2206.
w58x V.V. Schechtman, A.N. Varchenko, Invent. Math. 106 Ž1991. 139.
w59x A. Matsuo, Commun. Math. Phys. 134 Ž1990. 65.
w60x H. Awata, A. Tsuchiya, Y. Yamada, Nucl. Phys. B 365 Ž1991. 680.
w61x H.M. Babujian, J. Phys. 26 Ž1993. 6981.
w62x H.M. Babujian, R. Flume, hep-thr9310110.
w63x M. Gaudin, J. Physique 37 Ž1976. 1087.
Fermionic mapping for eigenvalue correlation functions of

weakly non-Hermitian symplectic ensemble
M.B. Hastings
Physics Department, Jadwin Hall, Princeton, NJ 08544, USA
Received 10 August 1999; received in revised form 20 October 1999; accepted 28 January 2000
Abstract
The eigenvalues of an arbitrary quaternionic matrix have a joint probability distribution

function first derived by Ginibre. We derive the j.p.d. for the weakly non-Hermitian version of this
problem and then show that there exists a mapping of this system onto a fermionic field theory.
This mapping is used to integrate over the positions of the eigenvalues and obtain eigenvalue
density as well as all higher correlation functions for both the strongly and weakly non-Hermitian
cases. q 2000 Elsevier Science B.V. All rights reserved.
PACS: 02.50; 11.10; 71.20

Keywords: Non-Hermitian random matrices; Eigenvalue correlation functions
1. Introduction
Several ensembles of non-Hermitian matrices were given by Ginibre w1x. These are
the ensembles of matrices with arbitrary real, complex, or quaternionic entries. Ginibre
gave joint probability distributions for the eigenvalues for the complex and quaternionic
cases, and succeeded in obtaining correlation functions in the complex case, while
correlation functions for the quaternionic case were found later w3x. The purpose of this
paper is to extend the correlation functions for the quaternionic problem to the weakly
non-Hermitian case w9x, as well as to introduce a fermionic mapping to simplify the
computation of these correlation functions. Further, the mapping also permits us to
derive the 4-point, and higher, correlation functions, that were only conjectured before.
Although Ginibre’s ensembles are interesting in themselves, they are also closely
connected with the chiral random matrix ensembles that appear in QCD w5,6x and some
condensed matter systems w7,8x. Knowing the eigenvalue correlation functions in the
E-mail address: hastings@feynman.princeton.edu ŽM.B. Hastings..
PII: S 0 5 5 0 - 3 2 1 3 Ž 0 0 . 0 0 0 4 3 - 2
536 M.B. Hastingsr Nuclear Physics B 572 [FS] (2000) 535–546
non-Hermitian ensemble, one can easily determine correlation functions in the corre-
sponding chiral ensemble. Further, the weakly non-Hermitian w9x versions of these
ensembles are of interest in open quantum systems; there exists a study using supersym-
metric techniques of the eigenvalue distribution in the weakly non-Hermitian version of
the ensemble considered in this paper w2x.
One interesting feature to observe in the two-level correlation function is the
crossover from a non-monotonic correlation function with algebraic tails in the limit of
very weak non-Hermiticity to a monotonically decaying correlation function with
Gaussian tails in the limit of strong non-Hermiticity.
Another interesting property that the symplectic non-Hermitian ensemble exhibits is a
depletion of the eigenvalue density near the real axis; this could be guessed at by
looking at the joint probability distribution Žj.p.d.. derived originally by Ginibre Žgiven
in Eq. Ž2. below.. The depletion was also found numerically w4x.
Consider an arbitrary N-by-N matrix of quaternions. This is equivalent to a 2 N-by-2 N
matrix M with complex entries. Let M be chosen from an ensemble of such matrices
with Gaussian weight
1
†
P Ž M . s ey 2 Tr Ž M M.
dM. Ž 1.
This defines the strongly non-Hermitian ensemble. The eigenvalues of M come in
complex conjugate pairs; for every eigenvalue z s x q iy, there is an eigenvalue
z s x y iy. Let the matrix M have eigenvalues z i , z i , i s 1... N. Then the j.p.d. of the
eigenvalues is given, up to a constant factor, by
1
N!
Ł eyz z < z i y z i < 2 Ł Ž z i y z j . Ž z i y z j .Ž z i y z j . Ž z i y z j . Ł d z i d z i ,
i i
Ž 2.
i i-j i
where d z d z s 2d x d y.
In Section 2, a fermionic mapping is introduced to write Eq. Ž2. as a correlation
function in a fermionic field theory. The mapping is then used to calculate the
eigenvalue density. In Section 3, we introduce the weakly non-Hermitian ensemble and
calculate eigenvalue density for that ensemble. In Section 4, multi-eigenvalue correlation
functions are calculated for both strongly and weakly non-Hermitian cases. The calcula-
tions in Section 3 and 4 are simple extensions of the calculation given in Section 2. For
this reason, the calculation in Section 2 is given in the most detail, while the other
calculations are sketched.
For the strongly non-Hermitian case, the main results are Eqs. Ž20., Ž26. for the
eigenvalue density, and Eq. Ž44. for the two-level correlation function. For the weakly
non-Hermitian case, the main results are Eq. Ž42. for the eigenvalue density and Eqs.
Ž48., Ž49. for the two-level correlation function.
2. Fermionic mapping and Green function
In this section we develop the fermionic mapping for the j.p.d. of the strongly
non-Hermitian ensemble. First, we write Eq. Ž2. as a correlation function in a fermionic
field theory. Then, for convenience we shift to radial coordinates, making a conformal
M.B. Hastingsr Nuclear Physics B 572 [FS] (2000) 535–546 537
transformation. Finally, we integrate over all but one of the z i to obtain the eigenvalue
density. The integral over the z i is done inside the correlation function; only after doing
the integral is the correlation function evaluated. This amounts to commuting the order
of doing the integral and evaluating the correlation function, and is the essential trick
used in this section. In Section 4 we will demonstrate how to obtain multi-level
correlation functions by a simple extension of the procedure of this section.
2.1. Fermionic mapping
First, let us show that Eq. Ž2. can be written as a correlation function in a
two-dimensional fermionic field theory. A similar fermionic mapping was demonstrated
previously for the Hermitian orthogonal and symplectic ensembles w10x. Let the field
c Ž x . have the action
S s 12 d z d z c † Ž z . Ec Ž z . .
H Ž 3.
Note that we are using only one chirality of fermionic field. Consider a correlation
function of this field, such as
2N
¦ Ł c Ž a i . c † Ž bi .
is1
; . Ž 4.
This correlation function is equal to

2N 2N
ž i-j
2N
/ž
Ł Ž a j y a i . Ł Ž b j y bi .
i-j / . Ž 5.
žŁŽ i, j
a j y bi .
/
j
Let us consider a specific choice of bj , with bj s Le 2 p i . In the limit L ™ `, we find

2N
that Eq. Ž5. reduces to

2N
2
Ly2 N yN
Ł Ž a j y ai . . Ž 6.
i-j
Comparing this to Eq. Ž2., we realize that Eq. Ž2. can be written as
1 2N N
N!
lim L2 N
L™`
2
qN
¦ž Ł c † Ž bj .
js1 /ž Ł UŽ zj , zj .c Ž zj . c Ž zj . d zj d zj
jsi
;/ , Ž 7.
j
where bj s Le 2 p i and UŽ z, z . s eyz z Ž z y z .. When integrating over z i , the limit

2N
L ™ ` must be taken before doing the integral over z i .

Now we will make a conformal transformation to radial coordinates. Write z s e w
and z s e w . Let w s t q i u . The action for the fermionic field is unchanged under this
transformation, but we must change Eq. Ž7. as the field c has nonvanishing scaling
dimension and conformal spin. Eq. Ž7. gets replaced by
1 2N N
N! L™`
lim L2 N ¦ž
2
Ł c † Ž Õj .
js1 /ž Ł e t U Ž e w ,e w . c Ž wj . c Ž wj . dwj dwj
js1
j j j
;/ , Ž 8.
j
where Õj s log bi L q 2p i 2 N .
Now, we will introduce Fourier transforms for the creation and annihilation operators.
We will write c Ž wi . s Ý k e k w i aŽ k . and c † s Ý k eyk w i a† Ž k .. In the limit L ™ `, the
only states involved in Eq. Ž8. are those with k s 1r2,3r2,5r2, . . . , 2 N y 1r2. If there
are excitations in states with higher k, they will vanish in the large L limit.
Then, we can rewrite Eq. Ž8., up to factors of order unity, as
2 Ny1r2 N
1
N! ¦ž Ł
ks1r2
a† Ž k .
/ Ł Ý a Ž k . a Ž kX . e w k e w k
js1
ž k,k
X
j j
X
e t j U Ž e w j ,e w j . dwj dwj ;/ . Ž 9.
In Eq. Ž9., consider integrating over wj ,wj for some given set of j s 1, . . . , M. We
will do the integral inside the correlation function. The integral
M
X X
wjk wjk
HŁ ž Ý aŽ k . aŽ k . e e e t j U Ž e w j ,e w j . dwj dwj / Ž 10 .
js1 X
k,k
is equal to
OM , Ž 11 .
where the operator O is defined by
O s Ý a Ž k . a Ž k q 1 . 4p Ž k q 1r2 . ! . Ž 12 .
k
Therefore, if we integrate over all eigenvalues in Eq. Ž9., we obtain

2 Ny1r2
1
Zs
N! ¦ž Ł
ks1r2 /;
a† Ž k . O N s Ł Ž 4p Ž k q 1r2 . ! . ,
k
Ž 13 .
where the product extends over k s 1r2,5r2,9r2, . . . , 2 N y 3r2.
2.2. EigenÕalue density
To calculate the density of eigenvalues ŽFig. 1., we must integrate over all, except
one, of the coordinate pairs wi ,wi in Eq. Ž9.. Using Eq. Ž11., and normalizing with Eq.
Ž13., we wish to compute
2 Ny1r2
U Ž e w ,e w . e t
m,m
Ý e mw e m w
X
X
¦ž ks1r2
Ł a† Ž k . O Ny1 a Ž m . a Ž mX . dw dw
/ ; . Ž 14 .
Ž N y 1 . !Z
Fig. 1. Eigenvalue density.
It may be verified that the correlation function appearing in the sum of Eq. Ž14. is
nonvanishing only if either m y 1r2 is even, mX y 1r2 is odd, and m - mX , or if
mX y 1r2 is even, m y 1r2 is odd, and mX - m. In the first case, with m - mX , the
contribution to Eq. Ž14. is
1 1 X
Ł Ł Ž l q 1r2. !eyz z'zz Ž z y z . z m z m dw dw ,
4p k Ž k q 1r2 . ! l
Ž 15 .
where the product over k extends over

k s 1r2,5r2,9r2,... Ž 16 .
and the product over l extends over
l s 1r2,5r2,9r2... m y 2,m q 1,m q 3, . . . , mX y 2,mX q 1,
mX q 3, . . . , 2 N y 1r2 . Ž 17 .
This is equal to
1 1 X
X eyz z'zz Ž z y z . z m z m dw dw . Ž 18 .
4p Ž m y 1r2 . !! Ž m y 1r2 . !!
In the second case, with mX - m, the result is
1 1 X
y X eyz z'zz Ž z y z . z m z m dw dw . Ž 19 .
4p Ž m y 1r2 . !! Ž m y 1r2 . !!
We can obtain the eigenvalue density r Ž z, z . by adding Eqs. Ž18., Ž19. and summing
over m,mX . Shifting m and mX by one-half, and changing from dw dw to d z z w, we find
that the final result for the eigenvalue density r Ž z, z . d z d z is
1
r Ž z, z. d z d zs eyz z Ž z y z . G Ž z , z . d z d z , Ž 20 .
4p
where the Green function GŽ z, z . is given by
1 X X
GŽ z , z . s Ý Ž z mz m yz mz m . . Ž 21 .
X X
m-m ; ms0,2,4,...; m s1,3,5,... m!!mX !!
2.3. Discussion
Let us now look at the properties of Eq. Ž20.. We will discuss in turn the
normalization of the density; the way the density depends on x and y separately, where
z s x q iy; an integral representation for the density; the circular law; and the depletion
of density near the real axis.
First, consider the normalization of the single particle density. It is automatic from
the above derivation that the eigenvalue density is properly normalized, although one
must be careful about defining the normalization depending on whether one is counting
total number of eigenvalues or total number of pairs of eigenvalues. The normalization
is defined such that Hr Ž z, z .d z d z s N.
Next, writing z s x q iy and z s x y iy, one can show by differentiating the power
series in Eq. Ž21. that Ex GŽ x q iy, x y iy . s E q2 E GŽ x q iy, x y iy . s 2 xGŽ x q iy, x y iy .,
for large N. This implies that
2
G Ž x q iy, x y iy . s e x f Ž y . Ž 22 .
2
and therefore r s 41p 2 yieyy f Ž y ., for some function f, so the interesting properties of
the eigenvalue density are contained in f Ž y .. Later we will discuss the properties of
f Ž y . for small y, and show that there is a depletion of the density of eigenvalues near
the real axis.
Using Eqs. Ž20., Ž21., Ž22., we can derive an integral representation for r . We can
use Eq. Ž22. to write
G Ž z , z . s e z z G Ž z y z ,0 . . Ž 23 .
Then Eq. Ž21. implies that
1 m
G Ž z y z ,0 . s Ý Ž zyz. . Ž 24 .
ms1,3,5,... m!!
This is equal to
` zyz 2
H0 i sin ž i /
t eyt r2
dt . Ž 25 .
Using this integral representation in Eq. Ž20., we find that

1 `
yt 2 r2
r Ž z, z. s 2y H0 sin Ž 2 yt . e dt . Ž 26 .
4p
Let us now consider the circular law. For large y, Eq. Ž26. reduces to
1
r Ž z, z. d z d z™ dzdz. Ž 27 .
4p
So, the density tends to a constant for large y. However, this integral representation is
valid only for N infinite; for finite N, the density tends to a constant only within a disc
of radius '2 N , and vanishes outside the disc. This is the well-known circular law
w1,11,12x. The vanishing of the density outside the disc is easy to see from the power
series representation. For finite N the highest power of Ž zz . appearing in Eq. Ž21. is
roughly 2 N and so r will be exponentially small for Ž zz . ) 2 N.
Note that the total density in the disc is correct. The area of a disc of radius R is
2p R 2 , where we are using the measure d z d z s 2d x d y. The density is 41p . So, the
number of particles in a disc of radius '2 N is indeed N, as desired.
For small y, we find that r is reduced below the expected result. Such a reduction
was found numerically before w4x. In the figure, we graph the eigenvalue density as a
function of y, for x s 0, for a system of 100 particles.
3. Weakly non-Hermitian case
Now we will consider the weakly non-Hermitian version of the ensemble given
above. In the weakly non-Hermitian random matrix ensemble, we again consider an
arbitrary N-by-N matrix of quaternions, H, but use a different Gaussian weight. Let
H s Hh q Ha , where the Hh is Hermitian and Ha is anti-Hermitian. Then, we chose the
matrix H with Gaussian weight
1
N Tr Ž H h†H h .y 12 N 2 aTr Ž H a†H a .
ey 2
, Ž 28 .
where a is some constant. In the large a limit, this reduces to the Gaussian Symplectic
Ensemble. For finite a, the weight in Eq. Ž28. is chosen to make sure that the imaginary
part of the eigenvalues scales with N in the same way as the level spacing.
If the matrix H is chosen with weight given by Eq. Ž28., then the j.p.d. of Eq. Ž2.
gets replaced by
1 2 2
a y i2 <
Ł eyN x yN i z i y z i < 2 Ł Ž z i y z j . Ž z i y z j .Ž z i y z j . Ž z i y z j . Ł d z i d z i .
N! i i-j i
Ž 29 .
The only difference in the weakly non-Hermitian case is that the Gaussian function of
2 2
eigenvalue position eyz i z i is replaced by eyx i yNa y i .
We have not found Eq. Ž29. previously in the literature. This equation can be derived
most easily as follows: write an N-by-N matrix of quaternions, H, as
H s Xy1 TX , Ž 30 .
y1 †
where X is a quaternion matrix such that X s X , and T is an upper triangular matrix
of quaternions. This procedure is a Schur decomposition, and is possible since the field
of quaternions, like the field of complex numbers, is algebraically closed.
The eigenvalues, z i , can be obtained from the diagonal elements of T ; each diagonal
element of T is a quaternion, which is associated with a pair of complex conjugate
eigenvalues z i , z i . If a given diagonal element of T is Ti s A q Bi q Cj q Dk, then
z i s A " i'B 2 q C 2 q D 2 .
The Jacobian associated with this change of variables is Ł i - j Ž z i y z j .Ž z i y z j ..
Further,
1
N Tr Ž H h†H h .y 12 N 2 aTr Ž H a†H a . 1
N Tr Ž T h† T h .y 12 N 2 aTr Ž Ta† Ta .
ey 2
s ey 2
, Ž 31 .
where Th ,Ta are Hermitian and anti-Hermitian parts of T. The integral over the elements
of T above the diagonal can be done trivially as this integral is Gaussian. The integral
over the diagonal elements of T includes a Gaussian factor and a factor from the
Jacobian. This integral is exactly the integral over the j.p.d. of Eq. Ž29..
Given Eq. Ž29., we could follow the procedure of the previous section. However, we
would run into some difficulties which are purely technical. The problem is that, while
in the strongly non-Hermitian case the eigenvalue density is independent of x, it is not
independent of x in the weakly non-Hermitian case. This makes the power series
expansion very awkward. We will find it convenient to change to a different geometry,
given in Eq. Ž32. below, for the weakly non-Hermitian case. Let me again stress that the
reason for choosing a different geometry is purely technical, to simplify the math.
An analogous simplification is often used in the Hermitian ensembles. For example,
consider the Gaussian Symplectic Ensemble in the large N limit. The eigenvalue density
is a function of energy, but if one appropriately scales all energies by the local level
spacing, it is simpler to obtain correlation functions from the Circular Symplectic
Ensemble w13x.
Let us introduce new coordinates, z s f q ir and z s f y ir, where f is periodic
with period 2p . Let us replace Eq. Ž29. by
2
1 ya N 2 r i2
Ž e i zi y e i zi .
N!
Łe ey2 i f i
i
Ž e i zi y e i zj . Ž e i zi y e i zj . Ž e i zi y e i zj . Ž e i zi y e i zj .
=Ł Ł d zi d zi . Ž 32 .
i-j ey2 i f iy2 i f j i
This describes a system of N pairs of levels, with average level spacing 2prN. The
imaginary part of the level is of order 1rN, so it is of order the level spacing. For large
a this reduces to the Circular Symplectic Ensemble. For finite a and large N, we expect
that the ensemble of Eq. Ž32. reproduces the behavior of the ensemble of Eq. Ž29. within
a small neighborhood of some given energy, just as the Circular Symplectic Ensemble
reproduces the results of the Gaussian Symplectic Ensemble within a neighborhood of a
given energy.
The next step is to write Eq. Ž32. as a correlation function in a fermionic field theory.
We will introduce N creation operators at r s q` and N creation operators at r s y`.
We find that the desired correlation function is
1 N N
N!
lim e N
L™`
2
¦ž
L
/ž /ž
Ł c † Ž bj . Ł c † Ž c j . Ł U Ž r j . c Ž z j . c Ž z j . d z i d z i
js1 js1 j
;/ ,
Ž 33 .
j j 2 2
where bj s 2pN q iL and c j s 2 Np y iL and UŽ r j . s eya N r j Ž e r j y eyr j .. For large N, we
2 2
can write UŽ r j . s eya N r j 2 r j .
Now, we will introduce Fourier modes for the creation and annihilation operators,
writing c Ž z . s Ý k e i k z aŽ k . and c † Ž z . s Ý k eyi k z aŽ k .. In the limit L ™ `, the only
states involved in Eq. Ž33. are those with k s yN q 1r2,y N q 3r2, . . . , N y 1r2. If
there are excitations in states with higher k, they will vanish in the large L limit. Then,
Eq. Ž33. can be written as
Ny1r2
1
N! L™`
lim ¦ž Ł
ksyNq1r2
a† Ž k .
/ j
ž k,k
X
j
X
Ł Ý a Ž k . a Ž kX . e i k z e i k z U Ž r j . d z i d z i
j
;/ .
Ž 34 .
As in the previous section, we will integrate over some set of z j , for j s 1... M, inside
the correlation function. The integral
M
X X
ik zj ik zj
HŁ
js1
ž Ý aŽ k . aŽ k . e
k,k
X
e U Ž rj . d z j d z j / Ž 35 .
is equal to
OwM , Ž 36 .
where the operator Ow is defined by
k2
p 3r2
aN2
Ow s 8 ž /
aN 2
Ý
k
ž ke a Ž k . a Ž yk . . / Ž 37 .
So, if we integrate over all coordinates z, z in Eq. Ž34., we obtain

Ny1r2 Ny1r2 k2
1 p 3r2
Zs
N! ¦ž Ł
ksyNq1r2
†
/ ;
a Ž k . Ž Ow .
N
s Ł
ks1r2 ž 16 ž /
aN 2
ke aN2
/ . Ž 38 .
To obtain the eigenvalue density, we must integrate over all but one of the coordinates
in Eq. Ž34.. Using Eq. Ž36., and normalizing with Eq. Ž38., we obtain
Ny1r2
UŽ r .
m,m
Ý X
X
e i m ze i m z ¦ž Ł
ksyNq1r2
a† Ž k . Ž Ow .
/ Ny 1
;
a Ž m . a Ž mX . d z d z
Ž 39 .
Ž N y 1 . !Z
The correlation function in Eq. Ž39. is nonvanishing only if m s ymX . Eq. Ž39. is equal
to
Ny1r2
UŽ r . Ý e i mŽ zyz . Gw Ž m . d z d z , Ž 40 .
msyNq1r2
where
1 p y3 r2 m2
Gw Ž m . s ž / 2
ey a N 2 . Ž 41 .
16 m aN
In the large N limit, we can simplify Eq. Ž40. by introducing scaled coordinates. Let
us introduce k s mrN and let us also scale z by a factor of N so that f now runs from
0 to 2p N. Then we can replace the sum by an integral and obtain
1 p y3 r2
2
r Ž f ,r . d f d r s ž / reya r Gw Ž z , z . d f d r , Ž 42 .
4 a
where
1 i kŽ zyz . yk 2 r a
1
Gw Ž z , z . s Hy1e e dk . Ž 43 .
k
As in the previous section, the proper normalization of the above result is automatic
from the derivation. It is possible to show that Eq. Ž42. is equivalent to Eq. Ž26. in the
limit of very small a. The qualitative feature of a depletion of eigenvalues near the real
axis is the same for weak and strong non-Hermiticity. Eq. Ž42. may be compared to the
results of the SUSY calculation w2x, and found to agree, with some differences in
notation between the two calculations.
4. Multi-point correlation functions
The calculation of multi-level correlation functions is quite easy. In Eqs. Ž9., Ž34., we
must integrate over all except for two, three, or more, of the coordinate pairs wi ,wi .
Since the system is a non-interacting fermion system, the multi-point Green functions
can be expressed very simply in terms of the Green function Ž21., using Wick’s
theorem. This permits the two-point correlation function to be easily generalized to a
multi-point correlation function, as conjectured previously w3x. We will not show this in
detail, but simply sketch the results, first for the strongly non-Hermitian case and then
for the weakly non-Hermitian case.
4.1. Strongly non-Hermitian case
First let us examine the strongly non-Hermitian case, generalizing the results of
Section 2. Consider the two-level correlation function, the probability to find a pair of
levels at position z, z given that there is another pair at position zX , zX . Then, the
two-level correlation function is
2
1 X X
ž / eyz zyz z Ž z y z . Ž zX y zX . Ž G Ž z , z . G Ž zX , zX . y G Ž z , zX . G Ž zX , z .
4p
qG Ž z , zX . G Ž zX , z . . d z d z d zX d zX . Ž 44 .
This is just an application of Wick’s theorem.
Let us consider the behavior of Eq. Ž44. in the limit when both z and z are far from
the real axis. Without loss of generality, assume that ReŽ zX .X )X
ReŽ z .. Then, use the
integral representation of the Green function to rewrite eyz zyz z GŽ z, zX .GŽ zX , z . as
X 2 ` X 2 X 2 X 2r 2 0 X
Ž z yz . t yt 2 r2
ey< zyz < žH e Ž zyz . t eyt r2
d t y 12 e Ž zyz . r2
/ž 1
2 e Ž z yz . y Hyè e /
dt .
0
Ž 45 .
We can find a similar representation for GŽ z, zX .GŽ zX , z .. Now, in the limit with z and
zX both far from the real axis, then either ImŽ z y zX . is large or ImŽ z y zXX. is large. In the
2
first case, Eq. Ž45. is exponentially small because of the factor of ey< zyz < . In the second
X 2 X 2
case, the integrals over t can beX performed in this limit, while e Ž zyz . r2 and e Ž z yz . r2
2 1
are small. The integral, H0è Ž zyz .t eyt r2 d t, is equal to zX y Ž X
z , for large Im z y z ; here
.
X
we rely on the fact that ReŽ z y z . ) 0.
So, up to exponentially small terms, Eq. Ž45. is equal to
X 2 1 1
ey< zyz < X X . Ž 46 .
zyz z yz
Also, in this limit, if Eq. Ž46. is not exponentially small, then z y1 zX zX y
1 1
X
1
z s zy z z y z .
X
X X
Inserting this result, and similar results for G z, z G z , z , back into Eq. 44 we find
Ž . Ž . Ž .
that, for both z and zX far from the real axis, the two-level correlation function is equal
to
2
1 X 2 X 2
ž /Ž 1 y ey< zyz < y ey< zyz < . d z d z d zX d zX . Ž 47 .

4p
This is essentially the same as the two-level correlation function found in the complex
non-Hermitian case w1x.
For z and zX near the real axis, I have examined the behavior of Eq. Ž44.
numerically. If ImŽ z . s ImŽ zX ., then the correlation function is a monotonically decay-
ing function of ReŽ z y zX ., with no signs of any oscillation. The correlation function is
exponentially small if both z and zX are near the real axis.
4.2. Weakly non-Hermitian case
Now let us consider the two-level correlation function in the weakly non-Hermitian
case, the probability to find one pair of levels z, z given that there is another pair at
zX , zX . As before, we must integrate over all except for two of the eigenvalue coordinates.
Using the scaled coordinates, we find that the two-level correlation function is given by
1 p y3
2 X2
ž / eya r ya r
rr X Ž Gw Ž z , z . Gw Ž zX , zX . y Gw Ž z , zX . Gw Ž zX , z .
16 a
qGw Ž z , zX . Gw Ž zX , zX . . d f d r d f X d r X . Ž 48 .
As in the previous subsection, this is just an application of Wick’s theorem.
To examine the behavior of the two-level correlation function, let us integrate over
r,r X in Eq. Ž48., to be left with a function of f y f X . The result is
2
1 1 1 Ž k q kX . X 2 X X X
2
y 2 Hy1 X eyŽ kyk . rŽ2 a. i kŽ f y f . i k Ž f y f .
e e d k d kX . Ž 49 .
4p 32p kk
In the limit a ™ `, the integral over k,kX in the above expression can be performed to
yield
1 1 2 <k< <k< X
log Ž < k < y 1 . ei k Ž f y f
4p 2
y
4p 2 Hy2 ž 1y
2
y
4
.
/ dk . Ž 50 .
Eq. Ž50. is the known result for the correlation function in the Circular Symplectic
Ensemble. It is a non-monotonic function, algebraically decaying for large f . For
sufficiently small a, Eq. Ž49. will describe a monotonically decaying function of

f X y f , but for fixed, nonvanishing a, the function will always decay algebraically for
large f .
5. Conclusion
In conclusion, we have given a simple fermionic mapping for determining the

correlation functions of the non-Hermitian symplectic ensemble. Although the eigen-
value density was found previously using SUSY, the present derivation is simpler and
can be more easily extended to the two-level correlation function. The two-level
correlation function in the strongly non-Hermitian case was found to be similar to that
for the ensemble of arbitrary complex matrices. In the weakly non-Hermitian case, the
two-level correlation function exhibits an interesting crossover as a function of a.
References
w1x J. Ginibre, J. Math. Phys. 6 Ž1965. 440.

w2x A.V. Kolesnikov, K.B. Efetov, preprint cond-matr9809173.
w3x M.L. Mehta, Random Matrices, 1st ed. ŽAcademic Press, New York, 1967..
w4x M.A. Halasz, J.C. Osborn, J.J.M. Verbaarschot, Phys. Rev. D 56 Ž1997. 7069.
w6x J.J.M. Verbaarschot, Phys. Rev. Lett. 72 Ž1994. 2531.
w7x R. Gade, Nucl. Phys. B 398 Ž1993. 499.
w8x T. Nagao, K. Slevin, Phys. Rev. Lett. 70 Ž1993. 635.
w9x Y.V. Fyodorov, B.A. Khoruzhenko, H.-J. Sommers, Phys. Lett. A 226 Ž1997. 46.
w10x M.B. Hastings, preprint cond-matr9612122.
w11x L. Girko, Theor. Probab. Appl. 29 Ž1985. 694.
w12x L. Girko, Theor. Probab. Appl. 30 Ž1986. 677.
w13x F.J. Dyson, J. Math. Phys. 3 Ž1962. 140.
Exclusion statistics in conformal field theory and the UCPF for

WZW models
Peter Bouwknegt, Leung Chim, David Ridout 1
Department of Physics and Mathematical Physics, UniÕersity of Adelaide, Adelaide, SA 5005, Australia
Received 23 March 1999; received in revised form 30 November 1999; accepted 10 December 1999
Dedicated to the memory of Prof. H.S. Green
Abstract
In this paper we further elaborate on the notion of fractional exclusion statistics, as introduced
by Haldane, in two-dimensional conformal field theory, and its connection to the Universal Chiral
Partition Function as defined by McCoy and collaborators. We will argue that in general, besides
the pseudo-particles introduced recently by Guruswamy and Schoutens, one needs additional ‘null
quasi-particles’ to account for the null-states in the quasi-particle Fock space. We illustrate this in
several examples of WZW-models. q 2000 Elsevier Science B.V. All rights reserved.
1. Introduction
Elementary excitations in low dimensional quantum many-body systems may exhibit

‘fractional statistics’ generalizing the usual Bose and Fermi statistics. In such cases the
single particle states available to an excitation may depend on the entire particle content
of the multi-particle state. To handle such systems Haldane w1x introduced a particular
form of, so-called, ‘fractional exclusion statistics’ where the statistical interactions are
encoded into a matrix Ga b . The thermodynamics of an ideal gas of particles satisfying
Haldane’s fractional exclusion statistics was subsequently worked out in a series of
papers w2–10x.
Fractional exclusion statistics arises naturally in quasi-particle descriptions of two-di-
mensional Conformal Field Theories ŽCFTs.. Here quasi-particles correspond to inter-
twiners ŽChiral Vertex Operators or CVOs. between the various representations of the
chiral algebra and the Žchiral. spectrum is constructed by repeated application of the
modes of a preferred set of CVOs on the vacuum. Inspired by w11–13x, such a basis was
1
Present address: Department of Mathematics, University of Western Australia, Nedlands WA 6907,
Austalia.
PII: S 0 5 5 0 - 3 2 1 3 Ž 9 9 . 0 0 7 7 7 - 4
548 P. Bouwknegt et al.r Nuclear Physics B 572 [FS] (2000) 547–573
$
first constructed for the Žs l 2. k 0 1 WZW models w14,15x in terms of a j s 1r2 spinon
field. This basis in particular illuminates the appearance $
of fundamental spinons in
spin-S integrable spin chains whose effective CFT is an Žs l 2. ks2 S WZW model w16–20x.
The virtue of the quasi-particle approach to CFT is that there is a simple method,
developed in w21x, to compute the thermodynamical properties of the quasi-particles and
expose their ‘fractional exclusion statistics’. This method involves truncating the
quasi-particle basis in momentum space and finding recursion relations for the associ-
ated finitized chiral characters. From the recursion relations one immediately derives the
Žtotal. single-level grand partition function ltot for the quasi-particles and hence their
statistical properties. A large number of examples were subsequently worked out w
w21–26xx. Particularly interesting applications of this approach include the study of the
fractional exclusion statistics of the edge quasi-particle excitations over abelian quantum
Hall states w25x. $
In many cases, including Žs l 2. ks1 WZW models w1,21x and Z k parafermions w24,26x,
it was discovered that the exclusion statistics of these CFT quasi-particles is indeed of
the type introduced by Haldane. All these examples involved quasi-particles with abelian
braid statistics, corresponding in the CFT to intertwiners ŽCVOs. with a unique fusion
path. The corresponding statistics is referred to as abelian exclusion statistics. At the
same time it was obvious that quasi-particles with non-abelian exclusion statistics,
corresponding to non-abelian braid group representations, do not satisfy exclusion
statistics of the type originally envisaged in w1–9x. Recently, however, it was realized
w27x that these cases could be incorporated into Haldane’s scheme as well by allowing
for pseudo-particles, i.e. particles that do not carry$
any bare mass or energy. In
particular, the non-abelian exclusion statistics of s l 2 k ) 1 spinons Žfor k s 2,3,4. and
Ž .
generalized fermions in minimal models Mkq 2 of CFT Žfor k s 1,2,3. was shown to
agree with the pseudo-particle generalization of Haldane statistics w27x. These pseudo-
particles are closely related to the pseudo-particles that arise in the Thermodynamic
Bethe Ansatz ŽTBA. for models with non-diagonal scattering Žsee, for example, Refs.
w28–38x..
In a parallel development much progress has been made in the last few years in the
analytic calculation of the character formulas directly from a statistical mechanics
approach. These works generally involve the classification of all the eigenvalues of the
transfer matrix and the computation of their finite-size corrections. This was first carried
out by the Stony Brook group by solving the Bethe-Ansatz type equations w39–42x, and
was followed by the work of w43–45x which deals directly with the functional relations
for the eigenvalues. Typically these calculations lead to the so-called ‘fermionic’ Žor
quasi-particle. type expressions for the characters of the representations of the chiral
algebra. One can again identify the quasi-particles in these fermionic characters, and
they seem to be related to the particle spectrum appearing in certain integrable
perturbations of the underlying CFT w46–54x.
Based on the many known examples Žsee, e.g., Refs. w15,52–74x and references
therein., McCoy et al. Žsee, in particular, Ref. w75x. conjectured that all CFT characters
can be written in the so-called ‘Universal Chiral Partition Function’ ŽUCPF. form,
which can be interpreted as the grand partition function for a system of chiral particles
with fugacities, and whose single particle momenta satisfy certain fermionic counting
rules. Actually, it was noted a few years earlier by the Stony Brook group w76x that such
P. Bouwknegt et al.r Nuclear Physics B 572 [FS] (2000) 547–573 549
counting rules are very similiar to Haldane’s exclusion statistics. The relation of
exclusion statistics to models solvable by the Thermodynamic Bethe Ansatz was also
noticed by Bernard and Wu w77x. Thus it became natural to conjecture that the
quasi-particles underlying the UCPF satisfy Haldane exclusion statistics with a statistical
interaction matrix Ga b given by the bilinear form matrix entering the UCPF. This was
successfully demonstrated in a number of cases corresponding to abelian $
exclusion
statistics w24–26,75,78x, but it was realized w26x, and confirmed for Žs l 2. k ) 1 spinons
and generalized fermions in w27x, that the most general form of the UCPF involves
quasi-particles with non-abelian exclusion statistics.
In this paper we will further elaborate on exclusion statistics in CFT, and the
connection with the UCPF. In particular, based on the results of w79,80x, we will argue
that in general, to write characters of WZW models in UCPF form, one needs to
introduce besides the pseudo-particles yet another kind of quasi-particles. These particles
are composites of the basic quasi-particles in the sense that their chemical potentials,
charges and other quantum numbers are given in terms of those of their constituents, and
account for the null-states in the quasi-particle Fock spaces. We will refer to these as
null-particles. They can contribute to the UCPF either with strictly positive or with
alternating signs.
The new results in this paper are as follows. In Sections 2 and 3, we generalize
previous discussions of quasi-particles by incorporating null-particles in the scheme
from the outset. We will show that both Haldane’s approach and the UCPF approach
lead to the same effective central charge. In Sections 4 and 5 we discuss various
examples of WZW models, corresponding to both abelian and non-abelian exclusion
statistics. We compare the exclusion statistics defined by the interaction matrix Ga b in
the UCPF with the results obtained from the recursion relation approach and find
complete agreement in all cases. We end with some conclusions.
2. Exclusion statistics with pseudo- and null-particles
Exclusion statistics, as introduced by Haldane w1x Žand generalized to the multi-com-

ponent case in w10x 2 ., is based on the idea that the number of accessible states dŽ a, k . for
a particle of species a and momentum k depends on the occupation number NŽ a, k . of all
other particles though a statistical interaction matrix gŽ a, k .Ž b, k X . by
D dŽ a , k . s y Ý gŽ a , k .Ž b , k X . D NŽ b , k X . . Ž 2.1 .
Ž b , kX .
It follows that the total number of states W Ž NŽ a, k .4., at fixed occupation numbers
NŽ a, k .4 is given by
D Ž0a , k . q NŽ a , k . y 1 y Ý g Ž a , k . Ž b , k X . NŽ b , k X .
W Ž NŽ a , k . 4 . s Ł
Ž a, k . Ž b , kX .
NŽ a , k .
where DŽ0a, k . is the total number of states available to particles of species a with
0 , Ž 2.2 .
momentum k when there are no particles in the system. Thus a gas of particles
2
For a different approach to exclusion statistics with internal degrees of freedom, see Ref. w81x.
satisfying the above exclusion statistics would have a grand canonical partition function
given by
NŽ a,k .
Zs Ý ž Ł Žt . / W Ž N
a Ža,k. 4 . exp Ý ž N Ž a , k . Ž m a y e a0 Ž k . . rk B T ,
/
NŽ a , k .4 Ž a,k . Ž a,k .
Ž 2.3 .
where e a0 Ž k . and m a are, respectively, the bare energy and chemical potential of the
particle of species a. In the sequel we shall specialize to the case of a one-dimensional
ideal gas where the particle interaction is localized in momentum space and encoded in a
finite matrix Ga b , i.e.
gŽ a , k .Ž b , k X . s d k , k X Ga b . Ž 2.4 .
We have allowed for particles that contribute to the partition function Ž2.3. with
alternating signs, i.e. ta s y1, as opposed to the ‘real particles’ with ta s 1. We will
see that they occur naturally in the quasi-particle description of certain conformal field
theories.3 We also partition the full set of particle species S into a set of ‘physical
particles’ S ph , and a set of ‘pseudo-particles’ S ps . The pseudo-particles do not carry
any bare mass or energy Ži.e. e a0 Ž k . s 0., but have the unique role of exchanging
internal degrees of freedom Žcolor. between the physical particles. As remarked in the
introduction, in the TBA literature pseudo-particles arise in models with non-diagonal
scattering Žsee, for example, Refs. w28–38x.. Pseudo-particles were recently introduced
in Haldane’s framework in w27x. It also seems that they have been anticipated in w10x
where the case of one physical particle and several pseudo-particles was referred to as a
hierarchical basis.
In the thermodynamic limit where the system size M ™ ` with finite fixed tempera-
ture T ) 0, a saddle-point approach to the partition function Ž2.3. yields the following
most probable 1-particle distribution function w2–9x
E
naŽ k . s za log l a Ž z . , Ž 2.5 .
E za 0
z b s t b e b Ž m b y e b Ž k ..
where l a is determined by
la y 1
ž la /Ł b
lGb a b s z a , Ž 2.6 .
and z a s ta expŽ b Ž m a y e a0 Ž k .... From a TBA point of view, l a s 1 q expŽybe a .,

where e a is the dressed energy for species a. One could proceed further, generalizing
the computation in w75x, by using n a to calculate the internal energy per unit length.
Alternatively, we could work with the expression for the free energy obtained in w2–10x
F s yk B T Ý DŽ0a , k . log l a . Ž 2.7 .
Ža,k.
3
Of course, the alternating sign can be absorbed in the exponent by adding an imaginary part to the
chemical potential.
For a gas with linear dispersion relation, i.e.
Ž Õ < k < , M D kr2p . for a g S ph ,

Ž e a0 Ž k . , DŽ0a , k . . s ½ Ž 0,0 . for a g S ps ,
Ž 2.8 .
where we assumed the speed of sound Õ is independent of the species, we obtain
k B2 T 2 M ya dz a
Fsy Ý H0 log l a Ž z . , Ž 2.9 .
Õ ag S ph
za
where ya s ta e bm a is the fugacity of species a. Thus the specific heat per unit length Žat
constant fugacity. for a one-dimensional ideal gas with exclusion statistics Ž2.1. is
p 2 k B2 T
Cs c˜ , Ž 2.10 .
3Õ
with
p2 ya dz a
c˜ s Ý H0 log l a Ž z . . Ž 2.11 .
6 ag S ph
za
We have written the specific heat in a form where c˜ admits an interpretation as the
effective central charge for systems with conformal symmetry.
The integral Ž2.11. may be evaluated along the lines of w26x Žsimilar computations are
of course well-known from related TBA equations, cf. Refs. w30–33,36–38,81–85x. and
leads to 4
p2 1 y ja
ž / 6
c˜ Ž y . s Ý L Ž j a . y L Ž ha . y 12 log ya log
a
ž 1 y ha / , Ž 2.12 .
where Ž j a ,ha . are solutions to the equations

Gab Gab
j a s ya Ł Ž 1 y j b . , ha s ya sa Ł Ž 1 y h b . , Ž 2.13 .
b b
where sa s 0 Ž sa s 1. for a g S ph Ž a g S ps ., and LŽ x . is Rogers’ dilogarithm defined

by
log z log Ž 1 y z .
L Ž x . s y 12 HC
0, x
dz
ž 1yz
q
z / , Ž 2.14 .
where log z Žfor z / 0. signifies the logarithm in the branch yp - arg z ( p and C0, x
is a contour in C from 0 to x which does not go across the branch cuts of log z and
logŽ1 y z .. Thus, in contrast to the case with no pseudo-particles w26x, the presence of
the pseudo-particles induces subtraction terms in the effective central charge Ž2.12..
4
Here we have assumed that Ga b is symmetric.
3. Exclusion statistics in conformal field theory
Suppose we have a two-dimensional conformal field theory with chiral algebra A, a

set of A-modules Vi , labeled by some index set i g I , and intertwiners ŽCVOs.
j j
fa ž /Ž . z s Ý fa ž / z nqŽ D iXy D iy D j . , Ž 3.1 .
iX i ngZ iX i yny Ž D iXy D i .
where Di denotes the conformal dimension of Vi and aX s 1, . . . ,dim Vj . The number of

intertwiners i = j ™ iX is given by the fusion rules Ni j i . In a quasi-particle approach to
conformal field theory the Žchiral. spectrum is constructed by repeated application of the
modes of a preferred set of CVOs on the vacuum <0:, i.e. a set of quasi-particle states of
type
jN j2 j1
f aN
ž i N i Ny1 / yn N y D Ž N .
. . . f a2
ž /
i 2 i1 yn 2 y D Ž2 .
f a1
ž /
i1 0 yn 1 y D Ž1 .
<0: , Ž 3.2 .
where DŽ k . s Di k y Di ky 1, constitute a basis of the A-module Vi . This basis is overcom-

plete unless we put restrictions on the mode sequences Ž n1 , . . . ,n N .. These restrictions
are obtained both from the braiding and fusion relations satisfied by the intertwiners
Ž3.1. as well as by possible null-states in the Fock space of intertwiners and may depend
on the fusion path Ž0,i 1 ,i 2 , . . . ,i N ..
It has been observed in a variety of approaches – TBA approaches, integrable spin
chains and also in the context of conformal field theory – that the degrees of freedom
contained in Ž3.2. can be separated into physical excitations and pseudo-particle
excitations. Loosely speaking, the physical excitations correspond to excitations over
some reference fusion path, while the pseudo-particles create ‘excited fusion paths’.
While this separation might, strictly speaking, not hold in the conformal field theory, it
may hold in some crystal limit Žcf. Ref. w86x. which is sufficient as far as the discussion
of the partition function is concerned. Thus, if the quasi-particles in a conformal field
theory are described by Haldane exclusion statistics, we need to distinguish two kinds of
particles: ‘pseudo-particles’ with vanishing bare energy and ‘physical particles’ with an
infinite range of energy levels separated by integers, i.e. we have the dispersion relation
Ž2.8..
Let us consider the partition function ch i Ž y;q ., i.e. the character of the A-module Vi .
If we assume that the quasi-particle interaction is purely statistical according to Ž2.1.,
and that l a of Ž2.6. can be interpreted as the single quasi-particle grand partition
function, the character will have the following approximate form in the thermodynamic
limit 5 :
ch i Ž y ;q . ; Ł l a Ž ya q D a . Ł Ł l a Ž ya q D ql . . a
Ž 3.3 .
ag S ps ag S ph l00
Of course the expression Ž3.3. is not meant to be exact but only valid, in general, in the
thermodynamic limit. We have chosen to write it in a discrete form, rather than in an
integral form like Ž2.9., to emphasize the discrete energy spectrum of the CFT. Also, the
5
The modular parameter q is related to the quantum spin chain quantities by q sexpŽy2p Õr Mk B T ., thus
M ™` Žat fixed T ) 0. corresponds to q ™1y.
product over a and l will be subject to restrictions depending on the sector i. Cases
where the CFT characters Žor sums thereof. do admit exact factorizations of the form
Ž3.3. have recently been studied in w87x.
The character ch i Ž y;q . may be expanded as a power series in q,
ch i Ž y ;q . s Ý aN Ž y . q N . Ž 3.4 .
N00
It is well known, of course, that modular transformations relate the asymptotic

behaviour of a N Ž y ., for N 4 0, to the specific heat Ž2.10.. For definiteness, let us see
how this works out using Ž3.3.. Asymptotically, we may approximate
1 dq
aN Ž y . s E Nq1
ch i Ž y ;q .
2p i q
1
s
2p i
Edq exp Žy Ž N q 1. log q q log ch Ž y ;q . . i
y1
ya dz a
E ž
; dqexp y Ž N q 1 . log q y Ž log q . Ý
ag S ph
H0 za /
log l a Ž z . .
Ž 3.5 .
In the last step, we have omitted all terms that do not contribute to the leading N
behaviour of a N Ž y .. The integral can be evaluated using a saddle-point approximation,
and we find
ceff Ž y . N
log a N Ž y . ; 2p ( 6
, Ž 3.6 .
with
p2 ya dz a
ceff Ž y . s Ý H0 log l a Ž z . . Ž 3.7 .
6 ag S ph
za
From Ž3.6. we can identify ceff Ž y . with the effective central charge of the partition
function Ž3.3. as shown in, e.g., Refs. w88,89x. Note that Ž2.11. and Ž3.7. indeed imply
that we can also identify ceff Ž y . with the c˜ computed in Section 2.
For future use, note that if all the chemical potentials m a , for a g S ph , are given in
terms of a single chemical potential m as m a s l a m , then we may write Ž3.7. as
p2 y dz
ceff Ž y . s H0 log ltot Ž z . , Ž 3.8 .
6 z
where
la
ltot Ž z . s Ł la Ž z . . Ž 3.9 .
ag S ph
The central charge c s ceff q Dmin of the conformal field theory follows from Ž2.12.
and Ž2.13. in the limit of vanishing chemical potentials, i.e. ya s ta , where we need to
take those solutions to Ž2.13. that satisfy 0 ( j a ,ha ( 1 for ta s 1 and y1 ( j a ,ha ( 0
for ta s y1. For the latter case, the imaginary part of the corresponding dilogarithm is
precisely canceled by the logarithm term in Ž2.12.. Indeed, for x - 0,
pi 1 yx
LŽ x . y
2
log Ž 1 y x . s L ž 1yx / y L Ž 1 . s yL ž 1yx / . Ž 3.10 .
An interesting development over the last few years has been the derivation of
quasi-particle type character formulas for the modules of chiral algebras Žsee, e.g., Refs.
w15,52–74x.. This work has led to the conjecture that all conformal field theory
characters can be written in the so-called ‘Universal Chiral Partition Function’ ŽUCPF.
form Žsee, in particular, Ref. w75x.
u
ch i Ž y ;q . s Ý žŁ /q
a
yam a
1
2 mPGPmy 12 APm
Ł
a
žŽ 1 y G. P m q
2 / a
m1 ,.., m n 00 ma
restrictions
Ž 3.11 .
where G is an n = n matrix and A and u are certain n-vectors. Both A and u as well as
the restrictions on the summations over the quasi-particle numbers m a will in general
depend on the sector i, while G will be independent of i. Furthermore, we have defined
Ž q. m n
m s , Ž q . n s Ł Ž1 y q k . . Ž 3.12 .
n Ž q . n Ž q . myn ks1
It has been conjectured by various groups Žsee, in particular, Ref. w75x. that the
quasi-particles underlying Ž3.11. satisfy Haldane exclusion statistics with statistical
interaction matrix given by the same matrix G as the one entering Ž3.11.. We will refer
to this conjecture as the ‘UCPF-exclusion statistics’ conjecture.
A convincing piece of evidence in support of this conjecture is that the asymptotics of
the character Ž3.11. Žin the thermodynamic limit q ™ 1y. is given by exactly the same
formulas Ž2.12. and Ž2.13. provided one chooses sa s 0 for u a s `, while sa s 1 for
u a - `. The asymptotic form of the character Ž3.11. for ya s 1 was derived in w89x Žsee
also Refs. w88,90,91x.. The present result is a straightforward generalization of these
derivations. As a word of caution we would like to add that the meaning of ‘Žbare.
energy’ in the Haldane–Wu approach is different from that in the UCPF. We have not
attempted to find a complete dictionary between the two approaches.
To prove the conjecture beyond the comparison of thermodynamics requires an exact
computation of the partition function starting from first principles, i.e. Eq. Ž2.1., as has
been done for g-ons w78x. Alternatively, it has been argued that the analytic continuation
of Ž2.12. to the covering space of C _ 0,14 contains information about all the excita-
tions in the spectrum. This idea has been successfully applied to some minimal
$
models
of conformal field$ theory w43x, Žgeneralized. parafermions w92,93x, Žs l n. ks1 WZW
models w94x and Žs l 2. k 0 1 WZW models w95x and might be put on a more rigorous
footing.
Exclusion statistics in conformal field theory can be studied by a method based on
recursion relations for truncated characters w21x. Truncated characters PLŽ i. Ž y;q . are
defined by taking the partition function of those states Ž3.2. where all the modes satisfy
n i q DŽ i . ( L. Thus, for large L, we will have Žcf. Ž3.3..
PLŽ i. Ž y ;q . ; Ł l a Ž ya . Ł Ł l a Ž ya q l . , Ž 3.13 .
ag S ps ag S ph 0(l(L
where the products are subject to certain restrictions depending on the sector a. Thus
Ž i. Ž i.
PLq 1 Ž y ;q . rPL Ž y ;q . ; Ł l a Ž ya q L . , as L ™ ` . Ž 3.14 .
ag S ph
More generally, if some of the physical particles a g S ph are composites of l a more

fundamental particles, then their modes will be cut off at l a L and we find
Ž i. Ž i. l
PLq 1 Ž y ;q . rPL Ž y ;q . ; Ł l a Ž ya q L . a s ltot Ž ya q L . , Ž 3.15 .
ph
ag S
where ltot is defined in Ž3.9.. Therefore, from recursion relations for the truncated
characters PLŽ i. Ž y;q . in the large L limit, one derives algebraic equations for the total
1-particle partitions functions ltot Ž y ., which can be compared to the ltot Ž y . arising
from a solution of Ž2.6., with a statistical interaction matrix as suggested by the UCPF
formula Ž3.11.. This procedure was carried out, and agreement was found, in several
cases including g-ons Žthe one-component case of Ž3.11.. w75,78x and several multi-
component cases w21,24–26x. All these cases involve only physical particles Ž u a s `.,
i.e. correspond to intertwiners with a unique fusion path, and the corresponding statistics
was therefore called ‘abelian exclusion statistics’. From Ž2.6. it is clear that the absence
of pseudo-particles always leads to small x expansions of the form l aŽ x . s 1 q x a q
O Ž x 2 .. In w22,26x it was observed, however, that generally we have expansions of the
form l aŽ x . s 1 q a a x a q O Ž x 2 ., where a a is the largest eigenvalue of the fusion
matrix of the quasi-particle a w26x. The exclusion statistics corresponding to the more
general case, a a / 1, was named ‘non-abelian exclusion statistics’. It was recently
recognized that non-abelian exclusion statistics can be accounted for in the Haldane
approach by incorporating pseudo-particles w27x. The UCPF-exclusion statistics
$
conjec-
ture was subsequently confirmed in various non-abelian cases, namely Žs l 2. k ) 1 WZW-
models and generalized fermions in minimal models of CFT w27x.
The main purpose of the remainder of this paper is to verify the UCPF-exclusion
statistics conjecture in various other, rather non-trivial, WZW-examples Žboth abelian
and non-abelian. for which UCPF characters have been found recently w79,80x. In all
examples we find complete agreement, thus supporting the conjecture above. To find
agreement, however, we have had to slightly extend the definition of the UCPF form to
account for certain null-particles, i.e. quasi-particles which are composites of the basic
quasi-particles Žsuch that their chemical potentials and other quantum numbers are given
in terms of those of their constituents., which have the purpose of removing the
null-states in the quasi-particle Fock spaces.
4. Towards the UCPF for WZW models
The following questions naturally arise. Given a conformal field theory, how does
one identify a set of quasi-particles in terms of which the characters take the UCPF
form, and how is the statistical interaction matrix G determined in terms of the
conformal field theory data Žchiral algebra, modules, fusion rules, conformal dimen-
sions, etc..? In this section we will give a partial answer for WZW models.
Thus far, only $isolated cases of UCPF
$
characters for affine
$
Lie algebras were known.
These included Žs l 2. ks1 w13,14x, Žs l 2. k ) 1 w15x and Žs l n. ks1 w74,96x.6 Recently an
algorithm was given which, in principle, can be used to obtain affine Lie algebra
characters in UCPF form for any affine Lie algebra gˆ and at arbitrary level w79,80x. Here
we will briefly explain the idea, we refer to Refs. w79,80x for the technical details. In
Section 5 we discuss some examples, examine the exclusion statistics, and make a
comparison to the results of Ref. w26x.
l
Let g be a simple finite-dimensional Lie algebra of rank l , let L i 4is1 be the set of
fundamental weights and LŽ L i . the corresponding finite-dimensional irreducible repre-
sentations. As our set of quasi-particles we take the intertwiners7
Li
fa ž /
LX L
Ž z. , a s 1, . . . ,dim L Ž L i . , Ž 4.1 .
corresponding to all fundamental representations LŽ L i . and between all possible X

ĝ
modules Žgiven$ by L and LX . at level-k, as determined by the fusion rules NL L i L . For
example, for s l 3 we take intertwiners transforming in both the 3 as well as the 3
representation of s l 3 . Next, we need to decouple the pseudo-particle excitations,
representing the excited fusion paths with respect to a reference fusion path, from the
physical excitations. The sums over pseudo-particle excitations are well known from
RSOS and spin chain models and yield, up to a factor, level-k restricted modified
Ž k .Ž
Hall-Littlewood polynomials Mlm y;q . Žor Kostka–Foulkes polynomials in the case of
s l n . for which, in some cases, UCPF expressions are known Žsee, e.g., Refs. w89,91,97–
99x for y s 1.. On the other hand, the physical excitations would yield a ŽFock space.
contribution to the character given by
1
MPBPM
2
q
Ý yiM i
ž Ł / Ł Ł Ž q. , Ž 4.2 .
i m Žai.
Ý mŽai. s Mi i a
a
where Mi is the total number of intertwiners in LŽ L i . and we have specialized to the

case where all particles in LŽ L i . have the same fugacity yi . The bilinear form matrix B
is given, in the case of simply laced Lie algebras, by the inverse Cartan matrix
Ay1
i j s L i , L j of g and arises from the mutual locality of the basic vertex operators
Ž .
expŽ i L i P f .. For non-simply laced Lie algebra these vertex operators have to be
corrected by fermionic operators to account for the difference between 12 < L i < 2 and the
conformal dimension Di s Ž L i , L i q 2 r .rŽh kq 1. of Vi at level k s 1 – the corre-
6
In principle one can get quasi-particle affine Lie algebra characters by taking limits of the W-algebra
minimal model characters Ži.e. coset characters. of, e.g., Ref. w89x. These will however involve an infinite
number of pseudo-particles and are not of the type considered here.
7
While in some level-1 cases it is possible to give the character in a UCPF form using less quasi-particles
than the ones discussed here, these formulas do not seem to generalize to level k )1 just by the inclusion of
additional pseudo-particles. We refer to Section 5.6 for an example.
sponding B easily follows from $

Refs. w100,101x. Combining these two ingredients gives
the required result for, e.g., Žs l 2. k 0 1 Žsee Section 5.1.. Unfortunately, in general this is
not the whole story as we have not yet incorporated the possible null-states in the
physical quasi-particle Fock space. It turns out, as discussed in w79,80x, that these can be
accounted for in UCPF form by interpreting the Fock space modulo null-states as the
coordinate ring of an affinized projective variety and applying standard techniques from
algebraic geometry. Besides deforming the exponent 12 M P B P M in Ž4.2. by a term
1
2 m P Q P m, this will in general involve the addition of null quasi-particles, correspond-
ing to certain composites of the basic physical quasi-particles Žhence, in particular, their
chemical potentials are fixed in terms of those of their constituents..
The final answer for the UCPF is then of the form
1 Žk.
chl Ž y ;q . s Ý l Mlm Ž y ;q . Mm Ž y ;q . , Ž 4.3 .
msM 1 L 1q . . . qM l L l
Ł Ž q. M i
is1
where
1
mPQPm
l 2
q
Mm Ž y ;q . s
žŁ / Ý žŁ
is1
Ž q . Mi
m a
yam a / Ł Ž q.
a
ma
. Ž 4.4 .
Ž k .Ž .
The factor containing B in Ž4.2. has been absorbed in Mlm q . Indeed, it has been
conjectured Žand proven in some cases. w98,99,102,103x that the affine Lie algebra
Ž k .Ž
characters chlŽ y;q . are indeed of the form Ž4.3., where Mlm y;q . and MmŽ y;q . are,
respectively, the level-k restricted and unrestricted Hall–Littlewood polynomials.
The procedure leading to Ž4.4. is not unique, however, and various equivalent UCPFs
with different null quasi-particle contents may be given Žsee the examples in Section 5..8
The equality between the characters with the various null-state subtractions is based
on the successive application of the following two identities Žsee, e.g., Ref. w96x for an
elementary proof.:
1 q Ž Mym.Ž Nym.
s Ý Ž q. Ž q. Ž q. Ž 4.5 .
Ž q. M Ž q. N m00 m Mym Nym
and
1
m Ž my1 .
MN 2
q m q
s Ý Ž y1. Ž q . Ž q . Ž q . , Ž 4.6 .
Ž q. M Ž q. N m00 m Mym Nym
which are, in a sense, ‘inverses’ of each other. Both identities are specializations of the
q-Chu–Vandermonde identity for the basic hypergeometric series 2 f 1 Žsee, e.g., Ref.
8
An interesting example in the context of minimal models of conformal field theory was recently discussed
in 52–54x.
w
w104x..9 Both are intimately related to, and in fact constitute a proof of, the five-term
identity for Rogers’ dilogarithm
x Ž1yy. yŽ1yx .
L Ž x . q L Ž y . s L Ž xy . q L
ž 1 y xy / ž
qL
1 y xy / . Ž 4.7 .
Indeed, by comparing ceff Ž y . for the asymptotics of two characters related by a single
application of either Ž4.5. or Ž4.6. one discovers Ž4.7.. Denoting the solutions of Ž2.13.
for the corresponding variables on the left-hand side of Ž4.5. and Ž4.6. by Ž j 1 , j 2 . and on
the right-hand side by Ž j 1X , j 2X , j 3X . we find that for Ž4.5. they are related by
j 1Ž 1 y j 2 . j2Ž1yj1.
j 1X s , j 2X s , j 3X s j 1 j 2 , Ž 4.8 .
1yj1 j2 1yj1 j2
while for Ž4.6. we find the inverse relations
j 1X Ž 1 y j 2X . j 2X Ž 1 y j 1X . j 1X j 2X
j1s , j2s , j 3X s y . Ž 4.9 .
1 y j 1X j 2X 1 y j 1X j 2X 1 y j 1X j 2X
With the use of Ž2.12. and Ž3.10. both lead to Ž4.7.. The fact that the characters of the
various null-particle formulations always seem to be related by either Ž4.5. or Ž4.6. can
be taken as further evidence for the conjecture that, loosely speaking, dilogarithm
identities are always accessible by means of the five-term identity Ž‘Goncharov’s
conjecture’, see Ref. w91x..
Having obtained the WZW characters in a UCPF form we can read off the statistical
interaction matrix G and verify whether the alleged exclusion statistics defined by Ž2.6.
indeed agrees with the exclusion statistics derived by the recursion method in w26x. We
will carry out this procedure in several non-trivial examples ŽSection 5. and find
agreement in all cases.
In Ref. w26x one of the authors and K. Schoutens conjectured that the recursion
relations for the truncated characters PLŽ i. Ž q . of the level-1 WZW models were, in all
cases, solved Župto an overall q-factor . by modified Hall–Littlewood polynomials
MlŽ L,i.Ž q ., with argument q ™ qy1 , and l a function of both i and L. Thus, these
modified Hall–Littlewood polynomials have to approach the WZW characters ch i Ž q . in
the limit L ™ `. This observation leads to the conclusion that, asymptotically, MlŽ L,i.Ž q .
is given by an expression like Ž3.13.. Extrapolating this reasoning a bit further, using the
fact that MlŽ q ., for l s M1 L1 q . . . qM l L l , is precisely the TBA-limit of the
partition function of an integrable spin chain with Mi spins in the minimal Uq Ž ĝ .
affinization Wi of the fundamental representation LŽ L i . whose elementary excitations
are precisely the quasi-particles Ž4.1. w105x, we are led to the conjecture that, asymptoti-
cally,
l Mi
NŽ l .
Ml Ž q . ; q Ł Ł zi Ž qyl . , i
Ž 4.10 .
is1 l is1
9
We are grateful to Ole Warnaar for pointing this out to us.
where N Ž l. is such that MlŽ q . s const q O Ž q . and the z i are expressed in terms of the
l a according to the fusion rules, i.e. according to which composites of the quasi-par-
L
ticles f a X i Ž z . make up the sectors j. For s l n we would have, more explicitly,
ž /
L L
Bi j
zi Ž x . s Ł
j
ž Ł l˜
a
Žj.
a Ž x. / , Ž 4.11 .
where Bi j is the inverse Cartan matrix of s l n and l˜Žaj. are the solutions lŽaj. to Ž2.6. for
the physical particles corresponding to Ž4.1. dressed with the l’s for the composite
null-particles which contain that particle. Explicit formulas for z i will be given in the
examples of Section 5. Note however that the modified Hall–Littlewood polynomial is a
q-deformation of the character of the tensor product of Mi-fold copies of Wi . Thus, a
consistency check on the assertion Ž4.10. is that
z i Ž 1 . s dim Wi . Ž 4.12 .
We will verify this in the examples. In fact, the $
analogous statement seems to be true in
higher level cases as well as suggested by the Žs l 2. k example in Section 5.1.
The fact that limits of modified
$
Hall–Littlewood Žor Kostka–Foulkes. polynomials
lead to WZW characters for Žs l n. k was first conjectured in w91x and subsequently
proven, in special cases, in w98,99,106x.
5. Examples
$
5.1. s l 2, leÕel-k
$
The Žnon-abelian. exclusion statistics for the case Žs l 2. k has been extensively
discussed in w23,26,27,107x, where it was shown, among other things, that the solutions
to Eq. Ž2.6. indeed agree with the expressions obtained from the recursion approach Žat
least for small k .. Here we suffice by making a few additional remarks.$
At level-k there are k q 1 integrable highest weight modules$
of s l 2 labeled by
Žtwice. the s l 2 spin, 2 j s 0,1, . . . ,k. The character of Žs l 2. k 0 1 can be written as
w15,86,108x
1 1
2
ch j Ž z ;q . s Ý Ý qy 4 m 1 q 4 mPA kPm
mq, m y00 m 2 , . . . , m k 00
mqqm ysm 1 restrictions
k
1 uj 1
1
=Ł
as2
žž 1y
2 /
A k Pmq
ma
2 / a
Ž q . mq Ž q . my
z 2
Ž m qym y .
, Ž 5.1 .
where A k is the Cartan matrix of A k ( s l kq1 , and Žu j .a s d a,2 j . The restrictions are
such that all entries in the q-binomials are integers. This character is obviously of the
UCPF form Ž3.11. with
.
° 1
2
1
2
.
. y 12 ¶
.
1 1 . y 12
2 2 .
... ... ... ... ... ...
Gs . , Ž 5.2 .
y 12 y 12 .
.
.
. 1
A ky 1
. 2
¢ .
.
.
ß
where the entries of G correspond to the summation variables mq,my,m 2 , . . . ,m k 4 in
Ž5.1.. In particular uqs uys ` while u a - ` for a s 2, . . . ,k.
We find the following solution to Ž2.13. for yqs yys 1
2
1 1
jqs jys 1 y
kq1
, ja s 1 y ž kq2ya / , a s 2, . . . ,k ,
p 2
sin
hqs hys 0 ,
leading to
ha s 1 y
0 sin
kq2
pa
kq2
, a s 2, . . . ,k , Ž 5.3 .
p2
ž /6
c s Ý Ž L Ž j a . y L Ž ha . .
a
k
s L Ž jq . q L Ž jy . q Ý Ž L Ž 1 y ha . y L Ž 1 y j a . .
as2
p 2
k k 1 k sin
s2 L ž kq1 / y
p
ÝL
as1
ž /
2
a 2
q ÝL
as1 0 0
sin
kq2
pa
kq2
k sin p2 3k
s Ý
as1
L
0 0
sin
kq2
pa
kq2
s
6 ž kq2
as required. Moreover, as shown in w21x for k s 1 and w27x for k s 2, . . . ,4, the solution
/ , Ž 5.4 .
to Ž2.6. agrees with the one found by the recursion method w23,26x. Also note that from
Ž5.3. it follows that the expression
z s Ž lq ly . 1 2 , Ž 5.5 .
at x s 1 is given by z s k q 1, which is consistent with the interpretation of the

quasi-particles as the excitations of an SUŽ2. spin chain with 2 S s k.
$
5.2. s l 3, leÕel-1
$
The affine Lie algebra s l 3, at level-k s 1, has three integrable representations
corresponding to the singlet 1, the vector 3 s LŽ L1 . and the conjugate vector 3 s LŽ L 2 .
of s l 3 . As discussed in Section 2, for the quasi-particles we take the intertwiners f a Ž z .
and f a Ž z . transforming in, respectively, the 3 and 3 representations. Since at level
k s 1 the fusion path is unique, there will be no pseudo-particles. However, the
quasi-particle Fock space will contain null-states as a consequence of the null-field
Ý a : f a Ž z . f a Ž z .:. The most natural way of incorporating this null-field is by introducing
one null-particle with t s y1. $
The following character formula for the integrable
highest weight modules of Žs l 3. ks1 was found in w74x:
1
MPBPM
2
ch i Ž y ;q . s Ý y 1M 1 y 2M 2 q
M1 , M 200
M1q2 M 2'i mod 3
1
m Ž my1 .
2
m q 1
= Ý Ž y1. , Ž 5.6 .
ma, m a, m
Ž q. m Ł Ž q. m Ž q. m
a a
a
Ý m a q m s M1
Ý m a q m s M2
$
where i s 1,2,3, labels the different integrable modules of Žs l 3. ks1 and
2 1
3 3
Bs
ž /
1
3
2
3
Ž 5.7 .
is the inverse Cartan matrix of s l 3 . This is indeed of the UCPF form Ž3.11. with
. .
° 2
3
2
3
2
3
.
.
1
3
1
3
1
3
.
. 1 ¶
. .
2 2 2 . 1 1 1 . 1
3 3 3 . 3 3 3 .
. .
2 2 2 . 1 1 1 . 1
3 3 3 . 3 3 3 .
... ... ... ... ... ... ... ... ...
. .
G s 13 1 1 . 2 2 2 . 1 Ž 5.8 .
3 3 . 3 3 3 .
. .
1 1 1 . 2 2 2 . 1
3 3 3 . 3 3 3 .
. .
1 1 1 . 2 2 2 . 1
3 3 3 . 3 3 3 .
... ... ... ... ... ... ... ... ...
¢ 1 1 1
.
.
. 1 1 1
.
.
. 3 ß
and
t s 1,1,1 < 1,1,1 < y1 4 ,
u s `,`,` < `,`,` < ` 4 . Ž 5.9 .
As remarked in Section 2, the fugacity of the null particle is given in terms of that of its
constituents as yy1 y 2 . The central charge Ž2.12. works out correctly, as c s 2, with
j a4 s 13 , 13 , 13 < 13 , 13 , 13 < y 18 4 .
To compare the exclusion statistics based on the statistical interaction matrix Ž5.8.
with the results of w21,26x we have to solve Ž2.6. with
l a 4 s l1 , l2 , l3 < l1 , l2 , l3 < m 4 ,
z a4 s x , x , x < x 2 , x 2 , x 2 < y x 34 . Ž 5.10 .
We find
1q2 x x
l ' l1 s l2 s l3 s s1q ,
1qx 1qx
1qxqx2 x2
l ' l1 s l2 s l3 s s1q ,
1qx 1qx
3
Ž1qx . x3
ms s 1 y . Ž 5.11 .
Ž1q2 x . Ž1qxqx2 . Ž1q2 x . Ž1qxqx2 .
This indeed implies
2 3
ltot Ž x . s Ž Ł l a .Ž Ł l a . m3 s Ž 1 q x q x 2 . , Ž 5.12 .
w
in accordance with the results of Refs. 21,26 . x
Alternatively, we might also incorporate the effect of the null-field by slightly
changing the statistics of the physical particles f 1 and f 1. In the characters this
amounts to applying Ž4.6.. This yields
1
2
MP BPM q m1 m1
ch i Ž y ;q . s Ý y 1M 1 y 2M 2 q Ý ,
M 1 , M 2 00 ma, m a Ł Ž q. m Ž q. m
a a
a
M1q2 M 2'i mod 3
Ý m a s M1 ,
Ý m a s M2
Ž 5.13 .
which is of the UCPF form with
.
° 2
3
2
3
2
3
.
.
4
3
1
3
1
3
¶
.
2 2 2 . 1 1 1
3 3 3 . 3 3 3
.
2 2 2 . 1 1 1
3 3 3 . 3 3 3
Gs ... ... ... ... ... ... ... . Ž 5.14 .

.
4 1 1 . 2 2 2
3 3 3 . 3 3 3
.
1 1 1 . 2 2 2
3 3 3 . 3 3 3
¢ 1
3
1
3
1
3
.
.
.
2
3
2
3
2
3
ß
The corresponding solutions to Ž2.6. are now given by

lX1 s lm , lX2 s lX3 s l ,
lX1 s lm , lX2 s lX3 s l , Ž 5.15 .
where l, l and m are as in Ž5.11.. In other words, changing the statistics of the physical
particles f 1 and f 1 precisely corresponds to dressing these particles by the null-particle
in the previous formulation. Again we find that ltot s ŽŁ lXa .ŽŁ lXa . is given by Ž5.12..
$
5.3. s l 4, leÕel-1
$
The affine Lie algebra s l 4 at level k s 1 has four integrable highest weight modules,
corresponding to the singlet 1, the vector LŽ L1 . s 4, the rank-2 anti-symmetric tensor
LŽ L2 . s 6 and the conjugate vector LŽ L3 . s 4. The UCPF form of the characters,
corresponding to quasi-particles Žintertwiners. transforming in the 4, 6 and 4, was
obtained in w79,80x. To incorporate the null-states in the quasi-particle Fock space we
need to deform both the inverse Cartan matrix of s l 4
3 1 1
4 2 4
Bs
0
1
2
1
4
1
1
2
1
2
3
4
Ž 5.16 .
as well as introduce one additional null-particle Žcorresponding to the composite of two

6 particles..
The analogue of the s l 3 expression Ž5.14. is given by Ž3.11. with
Ž 5.17 .
and A a s 0, u a s ` and ta s 1 for all a. The solution to Ž2.13. is given by

j a 4 s 17 , 29 , 14 , 14 < 101 , 81 , 71 , 71 , 81 , 101 < 41 , 41 , 92 , 71 < 811 4 , Ž 5.18 .
and leads to
6
cs ž /Ý
p2 a
LŽ ja . s 3 , Ž 5.19 .
as it should. The solution of Ž2.6. with

l a 4 s l1 , l2 , l3 , l4 < l12 , l13 , l14 , l23 , l24 , l34 < l123 , l124 , l134 , l234 < m 4 ,
z a 4 s ^x ,` ½ x < x 2 ,`
. . . , _^ 2 < 3
. . . , x_^
4
x ,` 3 < 4
. . . , x_
6
x
4
5 Ž 5.20 .
is given in Appendix A, and leads to

3 4 1 2 1 4
z 1 ' Ž Ł li . Ž Ł li j . Ž Ł li jk . m s 1 q 3 x ,
1 2 1 2
z 2 ' Ž Ł l i . Ž Ł l i j .Ž Ł l i jk . m2 s 1 q 2 x q 3 x 2 ,
1 4 1 2 3 4
l1tot4 s z 3 ' Ž Ł l i . Ž Ł l i j . Ž Ł l i jk . m s 1 q x q x 2 q x 3 . Ž 5.21 .
Note that the expression for l tot is in complete
$
agreement with the results of w21,26x,
confirming that the exclusion statistics of Žs l 4. ks1 is indeed described by a statistical
interaction matrix Ž5.17., while the z i Ž x s 1. are in agreement with Ž4.12..
$
5.4. s l n, leÕel k 0 1
Obtaining results for for s l n , n 0 5, at level-1, using the algorithm described in

w79,80x becomes extremely cumbersome. No complete results are known, but prelimi-
nary investigations suggest
i
nyiy1yk
zi Ž x . s Ý ž xk , / ltot s znny1 , Ž 5.22 .
ks0
k
where z i is defined in Ž4.11., such that indeed
i
nyiy1yk n
zi Ž 1. s
ks0
Ý ž k
s
i / ž/
s dim L Ž L i . . Ž 5.23 .
As explained in Section 4, once the results for the level k s 1 UCPF characters are
known, one can immediately obtain the level k ) 1 characters by correcting for the
level-k pseudo-particles as in Ž4.3..
$
5.5. s o 5, leÕel-1
$
The affine Lie algebra s o 5, at level k s 1, has three integrable highest weight
representations, corresponding to the singlet 1, the vector
$
Õ s 5 s LŽ L1 . and the spinor
s s 4 s L L2 of s o 5 . The UCPF form of the s o 5 1 characters is obtained by
Ž . Ž .
combining the results of w26,79,80,103x. In w103x the character was given in terms of
Žrestricted. s o 5 Kostka polynomials and a recipe was given to compute the restricted
Kostka polynomial. Explicit expressions for the restricted Kostka polynomial Žcorre-
sponding to the pseudo-particle part of the character. were given in w26x while in w79,80x
the UCPF form of the physical particles was found. The final result requires one
pseudo-particle, physical particles transforming in the 5 and 4 of s o 5 and one
null-particle Žcorresponding to the composite of two 5 particles.. The characters are
given by Ž3.11. with
° 1
.
.
. 0 0 0 0 0
.
.
. y 12 y 12 y 12 y 12
.
.
. 0 ¶
... ... ... ... ... ... ... ... ... ... ... ... ... ...
. . .
0 . 1 1 1 1 2 . 1 1 3 3 . 2
. . 2 2 2 2 .
. . .
0 . 1 1 1 2 1 . 1 1 1 3 . 2
. . 2 2 2 2 .
. . .
0 . 1 1 1 1 1 . 1 1 1 3 . 2
. . 2 2 2 2 .
. . .
0 . 1 2 1 1 1 . 1 1 1 1 . 2
. . 2 2 2 2 .
. . .
0 . 2 1 1 1 1 . 1 1 1 1 . 2
Gs . . 2 2 2 2 . ,
... ... ... ... ... ... ... ... ... ... ... ... ... ...
. . .
y 12 . 1 1 1 1 1 . 3 3 3 3 . 1
. 2 2 2 2 2 . 4 4 4 4 .
. . .
y 1 . 1 1 1 1 1 . 3 3 3 3 . 1
2 . 2 2 2 2 2 . 4 4 4 4 .
. . .
y 12 . 3 1 1 1 1 . 3 3 3 3 . 2
. 2 2 2 2 2 . 4 4 4 4 .
. . .
y 1 . 3 3 3 1 1 . 3 3 3 3 . 2
2 . 2 2 2 2 2 . 4 4 4 4 .
... ... ... ... ... ... ... ... ... ... ... ... ... ...
¢ 0
.
.
. 2 2 2 2 2
.
.
. 1 1 2 2
.
.
. 4 ß
Ž 5.24 .
which is a deformation of the matrix
1
1 2
Bs
ž /1
2
3
4
, Ž 5.25 .
entering Ž4.2.. Furthermore, A a s 0 and ta s 1 for all a g S , and
½` _^ ` _ 5
< `, . . . ,` < `, . . . ,` <`
us 0^
5 4
for 1 and Õ,
½ ^ ` _^ ` _ 5
u s 1 < `, . . . ,` < `, . . . ,` <`
5 4
for s, Ž 5.26 .
while there are also some restrictions on the summation over the m a’s Žsee Ref. w26x..
Note that this case corresponds to order k s 2 non-abelian exclusion statistics in the
sense of w27x as far as the coupling of the pseudo-particle to the physical spinor-particles
are concerned. The physical vector-particles have a unique fusion rule and therefore do
not couple to the pseudo-particle.
Eqs. Ž2.13. have the solution
j a 4 s 11 < 1 1 5 7 11 < 4 4 16 8 < 1
16 12 , 8 , 33 , 40 , 60 11 , 11 , 49 , 33 49 4 ,
ha 4 s 21 < 0,0,0,0,0 < 0,0,0,0 < 0 4 Ž 5.27 .

leading to
6
cs ž / ÝŽ Ž L j a . y L Ž ha . . s 3 y 12 s 52 , Ž 5.28 .
p2 a
as it should. Moreover, we have verified that the total 1-particle partition function l tot
2
s Ž Ł i l i . Ž Ł a la . m4 , resulting from the solution of Ž2.6. with
l a 4 s l < l1 , l2 , l0 , l2 , l1 < lqq , lqy , lyq , lyy < m 4 ,
½
< x2, . . . , x2 < x, . . . , x < x4 ,
z a4 s 1^ ` _^ ` _
5 4
5 Ž 5.29 .
satisfies, up to at least O Ž x 11 ., the equation
3 2 2
ltot y Ž 2 q 3 x 2 . ltot q Ž 3 x 2 y 1 . Ž x 2 y 1 . l tot
1 2
y x 2 Ž x 2 y 1. s 0 , Ž 5.30 .
derived in w26,109x from the recursion approach Žsee Appendix B for the explicit
solution up to O Ž x 11 ...
In addition, from Ž5.27., we obtain that the expressions for
1 2 1 2
ltot s z 1 s Ž Ł l i .Ž Ł la . m2 ,
1 2 1 4
z 2 s l1 2
Ž Ł l i . Ž Ł la . m, Ž 5.31 .
at x s 1 are given by, respectively, z 1 s 5 and z 2 s 4. Again this is in complete
agreement with Ž4.12.. The results in this section might prove to be useful with regards
to certain quasi-particle excitations Ž‘non-abelian electrons’. in SO Ž5. superspin regimes
for strongly correlated electrons on a two-leg ladder w109x. $
The UCPF and corresponding exclusion statistics for higher level s o 5 modules can
be worked out using the results of w103x.
$
5.6. s l n, leÕel-1, reÕisited
$
For Žs l n. ks1 it is also possible to give a description purely in terms of quasi-particles
Ž‘spinons’. f a transforming in the n-dimensional vector representation n. In this case
the null-field will be of the form : f 1 Ž z . . . . f n Ž z .:. The corresponding character formula
was found in w74x
1 1 1
2 2
m Ž my1 . Ž Ým y n ŽÝm . .
a a
2 2
ma q q
yÝ
m
ch i Ž y ;q . s Ý Ý Ž y1. .
m a00 m00 Ž q. m Ł Ž q . m ym a
a
Ý m a'i mod n
Ž 5.32 .
It can be brought in the UCPF form by shifting the summation variables m a ™ m a q m

Žfor y / 1.. Then,
.
.
.
.
d a b y 1n . 0
Gs

... ...
0
...
.
.
.
.
...
.
.
.
...
1
.
0Ž 5.33 .
The corresponding equations Ž2.6. with l a4 s l1 , . . . , l n < m4 , and z a4 s x, . . . , x < y x n 4

have the solution
1
l1 s . . . s l n s , ms1yxn , Ž 5.34 .
1yx
so that indeed
n
ltot s Ž l1 . . . l n . m n s Ž 1 q x q . . . qx ny1 . . Ž 5.35 .
It is also possible to write Ž5.32. in terms of a non-alternating sum by repeated
application of Ž4.6. and Ž4.5.. Besides the n $ spinons this requires n y 2 additional
null-particles for s l n . Here give the result
$
for s l 3 Žsee Refs. w79,80x for the origin of
this formula and the generalization to s l n .
1
mPGPm
m q2 m q2
ch i Ž y ;q . s Ý y Ý
a
, Ž 5.36 .
m , m00
a
Ž q. m Ž q. m Ž q. m Ž q. m1 2 3
with
.
2 2
y 13 . 1
3 3 . 3
.
2 2
y 13 . 1
Gs y1
3
3
...
1
3
leading to the solution

y 13
...
3
1
3
2
3
...
1
3
.
.
.
.
...
.
.
.
3
1
3
...
2
3
0 , Ž 5.37 .
1qxqx2
l1 s l2 s 1 q x , l3 s 1 q x q x 2 , ms , Ž 5.38 .
1qx
and again confirming
3
ltot s Ž l1 l2 l3 . m2 s Ž 1 q x q x 2 . . Ž 5.39 .
In contrast to the UCPF formulas in section 4.2 and 4.3, it does not appear that the
formula Ž5.32. has a straightforward generalization to levels k ) 1.
6. Conclusions
In this work we have to tried to reconcile Haldane’s notion of exclusion statistics w1x
with the Stony Brook group’s proposal of a Universal Chiral Partition Function form for
all Žchiral. characters of two-dimensional conformal field theories w75x. The main new
result of this paper is that, besides the pseudo-particles of w27x, in general this requires
yet another kind of particles, so-called null-particles. In support of the conjectured
relation between Haldane statistics and the UCPF, we have shown that an ideal gas of
physical, pseudo- and null-particles, with linear dispersion relations, in the thermody-
namic limit exhibits the same effective central charge as the UCPF. It would of course
be most desirable to extend this comparison to the different sectors of the UCPF and
gain an understanding of the restrictions that enter the sum.
The UCPF was put forward to structuralize the form of the characters of CFT. By
indicating how the characters of affine Lie algebras may be written in the UCPF form by
introducing null-particles we have obtained further support for the alleged ‘universality’
of the UCPF.
To demonstrate this method we have discussed various examples of UCPFs for
WZW-models and the associated exclusion statistics and found agreement with previous
results, computed by the recursion method w21,26x, in all cases.
Acknowledgements
We would like to thank Sathya Guruswamy and Kareljan Schoutens for discussions,
useful remarks, and for making their manuscript w27x available to us prior to publication.
P.B. is supported by a QEII research fellowship from the Australian Research Council
and D.R. was supported by a University of Adelaide Faculty of Science summer
scholarship.
Appendix A. Solution for s l 4
The explicit solution of Ž2.6. with Ž5.20. is given by

1q3 xq3 x2 x
l1 s 2
s1q ,
1q2 xq3 x 1q2 xq3 x2
2
Ž1q2 x . x Ž1qx .
l2 s 2
s1q ,
1q3 xq3 x 1q3 xq3 x2
1q3 x x
l3 s s1q ,
1q2 x 1q2 x
1q3 x x
l4 s s1q ,
1q2 x 1q2 x
Ž1q2 xq2 x2 . Ž1q3 xq2 x2 q2 x3. x2 Ž1q3 x .
l12 s 2
s 2
,
Ž1q2 x . Ž1qxqx2 qx3. Ž1q2 x . Ž1qxqx2 qx3.
2
Ž1qx . Ž1q2 xq3 x2 . x2 Ž1q2 x .
l13 s s 1 q ,
Ž1qxqx2 . Ž1q3 xq3 x2 . Ž1qxqx2 . Ž1q3 xq3 x2 .
1q2 xq3 x2qx3 x2
l14 s s1q ,
Ž1qx . Ž1qxqx2 . Ž1qx . Ž1qxqx2 .
1q3 xq3 x2 x2
l23 s s1q ,
Ž1qx . Ž1q2 x . Ž1qx . Ž1q2 x .
2
Ž1qx . Ž1q2 xq3 x2 . x2 Ž1qxqx2 .
l24 s s 1 q ,
Ž1q2 x . Ž1q2 xq3 x2 qx3. Ž1q2 x . Ž1q2 xq3 x2 qx3.
Ž1q2 xq2 x2 . Ž1q3 xq2 x2 q2 x3. x2 Ž1qxqx2 qx3.
l34 s 2
s1q 2
,
Ž1q3 x . Ž1qxqx2 . Ž1q3 x . Ž1qxqx2 .
1qxqx2qx3 x3
l123 s s1q ,
1qxqx2 1qxqx2
1qxqx2qx3 x3
l124 s s1q ,
1qxqx2 1qxqx2
2
Ž1qxqx2 . x3Ž1qx .
l134 s s1q ,
1q2 xq3 x2qx3 1q2 xq3 x2qx3
1q2 xq3 x2qx3 x3
l234 s s1q ,
1q2 xq3 x2 1q2 xq3 x2
2 2
Ž1q2 x . Ž1qxqx2 .
ms
Ž1qx . Ž1q2 xq2 x2 . Ž1q3 xq2 x2 q2 x3.
x4
s1q . Ž A.1 .
Ž1qx . Ž1q2 xq2 x2 . Ž1q3 xq2 x2 q2 x3.
Appendix B. Approximate solution for s o 5
Up to O Ž x 11 . the solution of Ž2.6. with Ž5.29. is given by
l s 2 q 4 y y 8 y 2 q 18 y 3 y 48 y 4 q 303 5 6 8505 7
2 y y 544 y q 4 y
y 8768 y 8 q 1198427
32 y 9 y 163968 y 10 q O Ž y 11 . ,
l1 s 1 q 2 y 2 y 16 y 3 q 108 y 4 y 696 y 5 q 4408 y 6 y 27702 y 7 q 173424 y 8
y 1083451 y 9 q 6760800 y 10 q O Ž y 11 . ,
l2 s 1 q 2 y 2 y 12 y 3 q 64 y 4 y 334 y 5 q 1736 y 6 y 18053

2 y
7
q 47008 y 8 y 980991
4 y 9 q 1281696 y 10 q O Ž y 11 . ,
l0 s 1 q 2 y 2 y 12 y 3 q 68 y 4 y 374 y 5 q 2024 y 6 y 21709
2 y
7
q 57904 y 8 y 1231627
4 y 9 q 1634080 y 10 q O Ž y 11 . ,
l2 s 1 q 2 y 2 y 8 y 3 q 28 y 4 y 92 y 5 q 272 y 6 y 619 y 7 q 160 y 8 q 21685
2 y
9
y 100320 y 10 q O Ž y 11 . ,
l1 s 1 q 2 y 2 y 8 y 3 q 28 y 4 y 84 y 5 q 152 y 6 q 569 y 7
y 9616 y 8 q 166483
2 y 9 y 601248 y 10 q O Ž y 11 . ,
lqqs 1 q 2 y y 6 y 2 q 25 y 3 y 116 y 4 q 2255 5 6
4 y y 2808 y q
113577 7
8 y
y 72496 y 8 q 23858843
64 y 9 y 1926944 y 10 q O Ž y 11 . ,
lqys 1 q 2 y y 6 y 2 q 25 y 3 y 116 y 4 q 2255 5 6
4 y y 2808 y q
113577 7
8 y
y 72496 y 8 q 23858843
64 y 9 y 1926944 y 10 q O Ž y 11 . ,
lyqs 1 q 2 y y 6 y 2 q 21 y 3 y 80 y 4 q 1255 5 6 39117 7
4 y y 1240 y q 8 y
y 19104 y 8 q 4700235
64 y 9 y 275296 y 10 q O Ž y 11 . ,
lyys 1 q 2 y y 6 y 2 q 13 y 3 y 24 y 4 q 151 5 6 363 7
4 y y 40 y y 8 y
q 576 y 8 y 203605 9 10 11
64 y q 15264 y q O Ž y . ,
m s 1 q 4 y 4 y 48 y 5 q 400 y 6 y 2872 y 7 q 19072 y 8 y 120906 y 9

q 743936 y 10 q O Ž y 11 . , Ž B.1 .
where y s xr '2 . This leads to
1 2
l1tot2 s ž Łl / ž Łl /
i a m
i a
s 1 q 4 y q 2 y 2 q 2 y 3 y 8 y 4 q 632 y 5 y 128 y 6 q 2145

4 y
7
y 2304 y 8 q 323323 9 10 11
32 y y 45056 y q O Ž y . . Ž B.2 .
References
w1x D. Haldane, Phys. Rev. Lett. 67 Ž1991. 937.

w2x Y.-S. Wu, Phys. Rev. Lett. 73 Ž1994. 922.
w3x C. Nayak, F. Wilczek, Phys. Rev. Lett. 73 Ž1994. 2740.
w4x S.B. Isakov, Mod. Phys. Lett. B 8 Ž1994. 319.
w5x ` de Veigy, S. Ouvry, Phys. Rev. Lett. 72 Ž1994. 600.
A. Dasnieres
w6x ` de Veigy, S. Ouvry, Mod. Phys. Lett. B 9 Ž1995. 271. cond-matr9411036.
A. Dasnieres
w7x A.K. Rajagopal, Phys. Rev. Lett. 74 Ž1995. 1048.
w8x Y.-S. Wu, Y. Yu, Phys. Rev. Lett. 75 Ž1995. 890.
w9x S.B. Isakov, D.P. Arovas, J. Myrheim, A.P. Polychronakos, Phys. Lett. A 212 Ž1996. 299. cond-
matr9601108.
w10x T. Fukui, N. Kawakami, Phys. Rev. B 51 Ž1995. 5239. cond-matr9408015.

w11x F. Haldane, Phys. Rev. Lett. 66 Ž1991. 1529.
w12x F. Haldane, Z. Ha, J. Talstra, D. Bernard, V. Pasquier, Phys. Rev. Lett. 69 Ž1992. 2021.
w13x D. Bernard, V. Pasquier, D. Serban, Nucl. Phys. B 428 Ž1994. 612. hep-thr9404050.
w14x P. Bouwknegt, A. Ludwig, K. Schoutens, Phys. Lett. B 338 Ž1994. 448. hep-thr9406020.
w15x P. Bouwknegt, A. Ludwig, K. Schoutens, Phys. Lett. B 359 Ž1995. 304. hep-thr9412108.
w16x L. Takhtajan, Phys. Lett. A 87 Ž1982. 479.
w17x L. Faddeev, N. Reshetikhin, Ann. Phys. 167 Ž1986. 227.
w18x I. Affleck, F. Haldane, Phys. Rev. B 36 Ž1987. 5291.
w19x N. Reshetikhin, J. Phys. A 24 Ž1991. 3299.
w20x P. Fendley, Phys. Rev. Lett. 71 Ž1993. 2845. cond-matr9304031.
w21x K. Schoutens, Phys. Rev. Lett. 79 Ž1997. 2608. cond-matr9706166.
w22x K. Schoutens, Phys. Rev. Lett. 81 Ž1998. 15704. cond-matr9803169.
w23x H. Frahm, M. Stahlsmeier, Phys. Lett. A 250 Ž1998. 293. cond-matr9803381.
w24x J. Gaite, Nucl. Phys. B 525 Ž1998. 627. hep-thr9804025.
w25x R. van Elburg, K. Schoutens, Phys. Rev. B 58 Ž1998. 15704. cond-matr9801272.
w26x P. Bouwknegt, K. Schoutens, Nucl. Phys. B 547 Ž1999. 501. hep-thr9810113.
w27x S. Guruswamy, K. Schoutens, Nucl. Phys. B 556 Ž1999. 501. cond-matr9903045.
w28x C.N. Yang, C.O. Yang, J. Math. Phys. 10 Ž1969. 1115.
w29x B. Sutherland, Phys. Rev. Lett. 20 Ž1967. 98.
w30x Al.B. Zamolodchikov, Nucl. Phys. B 358 Ž1991. 497.
w33x P. Fendley, H. Saleur, Al.B. Zamolodchikov, Int. J. Mod. Phys. A 8 Ž1993. 5751.
w34x F. Ravanini, Phys. Lett. B 282 Ž1992. 73. hep-thr9202020.
w35x F.R. Ravanini, R. Tateo, A. Valleriani, Int. J. Mod. Phys. A 8 Ž1993. 1707. hep-thr9207040.
w36x V.V. Bazhanov, N.Y. Reshetikhin, Int. J. Mod. Phys. A 4 Ž1989. 115.
w37x V.V. Bazhanov, N.Y. Reshetikhin, J. Phys. A 23 Ž1990. 1477.
w38x V.V. Bazhanov, N.Y. Reshetikhin, Prog. Theor. Phys. Suppl. 102 Ž1990. 301.
w39x R. Kedem, B.M. McCoy, J. Stat. Phys. 71 Ž1994. 865. hep-thr9210146.
w40x G. Albertini, S. Dasmahapatra, B. McCoy, Int. J. Mod. Phys. A ŽSuppl. 1A. 7 Ž1992. 1.
w41x G. Albertini, S. Dasmahapatra, B. McCoy, Phys. Lett. A 170 Ž1992. 397.
w42x S. Dasmahapatra, R. Kedem, B.M. McCoy, E. Melzer, J. Stat. Phys. 74 Ž1994. 239. hep-thr9304150.
w43x D. O’Brien, P. Pearce, S.O. Warnaar, Nucl. Phys. B 501 Ž1997. 773.
w44x V. Bazhanov, S. Lukyanov, A. Zamolodchikov, Comm. Math. Phys. 177 Ž1996. 381. hep-thr9412229.
w45x V. Bazhanov, S. Lukyanov, A. Zamolodchikov, Comm. Math. Phys. 190 Ž1997. 247. hep-thr9604044.
w46x A.B. Zamolodchikov, Adv. Stud. in Pure Math. 19 Ž1989. 641.
w47x D. Bernard, A. Le Clair, Nucl. Phys. B 340 Ž1990. 721.
w48x C. Ahn, D. Bernard, A. Le Clair, Nucl. Phys. B 346 Ž1990. 409.
w49x F.A. Smirnov, Nucl. Phys. B 337 Ž1990. 156.
w50x F.A. Smirnov, Int. J. Mod. Phys. A 6 Ž1991. 1407.
w51x G. Mussardo, Phys. Rep. 218 Ž1992. 215.
w52x A. Berkovich, B. McCoy, P. Pearce, Nucl. Phys. B 519 Ž1998. 597.
w53x A. Berkovich, B. McCoy, The perturbation w 2,1 of the M Ž p, pq1. models of conformal field theory
and related polynomial character identities, math. QAr9809066.
w54x S.O. Warnaar, q-Trinomial identities, math. QAr9810018.
w55x J. Lepowski, M. Primc, Structure of the standard modules for the affine Lie algebra AŽ1. 1 , Contemp.
Math. 46, ŽAmer. Math. Soc., Providence, 1985..
w56x R. Kedem, T.R. Klassen, B.M. McCoy, E. Melzer, Phys. Lett. B 304 Ž1993. 263. hep-thr9211102.
w57x R. Kedem, T.R. Klassen, B.M. McCoy, E. Melzer, Phys. Lett. B 307 Ž1993. 68. hep-thr9301046.
w58x S. Dasmahapatra, R. Kedem, T.R. Klassen, B.M. McCoy, E. Melzer, Int. J. Mod. Phys. B 7 Ž1993.
3617. hep-thr9303013.
w59x E. Melzer, Int. J. Mod. Phys. A 9 Ž1994. 1115. hep-thr9305114.
w60x A. Berkovich, Nucl. Phys. B 431 Ž1994. 315. hep-thr9403073.
w61x A. Berkovich, B.M. McCoy, Lett. Math. Phys. 37 Ž1996. 49. hep-thr9412030.
w62x S.O. Warnaar, J. Stat. Phys. 82 Ž1996. 657. hep-thr9501134.

w63x S.O. Warnaar, J. Stat. Phys. 84 Ž1996. 49. hep-thr9508079.
w64x A. Berkovich, B.M. McCoy, A. Schilling, Comm. Math. Phys. 191 Ž1998. 325. q-algr9607020.
w65x O. Foda, Y.-H. Quano, Int. J. Mod. Phys. A 12 Ž1997. 1651. hep-thr9408086.
w66x O. Foda, K.S.M. Lee, T.A. Welsh, Int. J. Mod. Phys. A 13 Ž1998. 4967. q-algr9710025.
w67x O. Foda, T. Welsh, Melzer’s identities revisited, math. QAr9811156.
w68x A. Schilling, Nucl. Phys. B 459 Ž1996. 393. hep-thr9508050.
w69x A. Schilling, S.O. Warnaar, The Ramanujan Journal 2 Ž1998. 459. q-algr9701007.
w70x B.L. Feigin, A.V. Stoyanovsky, Quasi-particle models for the representations of Lie algebras and
geometry of flag manifolds, hep-thr9308079.
w71x B.L. Feigin, A.V. Stoyanovsky, Funct. Anal. Appl. 28 Ž1994. 55.
w72x G. Georgiev, J. Pure Appl. Algebra 112 Ž1996. 247. hep-thr9412054.
w73x G. Georgiev, Combinatorial constructions of modules for infinite-dimensional Lie algebras, II.
Parafermionic space, q-algr9504024.
w74x P. Bouwknegt, K. Schoutens, Nucl. Phys. B 482 Ž1996. 345. hep-thr9607064.
w75x A. Berkovich, B. McCoy, The universal chiral partition function for exclusion statistics, hep-thr9808013.
w76x R. Kedem, B. McCoy, E. Melzer, in: Recent progress in Statistical Mechanics and Quantum Field
Theory, ed. P. Bouwknegt et al. ŽWorld Scientific, Singapore, 1995., hep-thr9304056.
w77x D. Bernard, Y.-S. Wu, A note on statistical interactions and the thermodynamic Bethe Ansatz, in: New
developments in integrable systems and long-range interaction models, Nankai Lecture Notes on
Mathematical Physics ŽWorld Scientific, Singapore, 1994., cond-matr9404025.
w78x K. Hikami, Phys. Lett. A 205 Ž1995. 364.
w79x P. Bouwknegt, q-Identities and affinized projective varieties, I. Quadratic monomial ideals, math-
phr9902010 Žto be published in Cond. Math. Phys...
w80x P. Bouwknegt, N. Halmagyi, q-Identities and affinized projective varieties, II. Flag varieties, math-
phr9903033 Žto be published in Cond. Math. Phys...
w81x A. Bytsko, A. Fring, Nucl. Phys. B 532 Ž1998. 588. hep-thr9803005.
w83x T. Klassen, E. Melzer, Nucl. Phys. B 338 Ž1990. 485.
w86x A. Nakayashiki, Y. Yamada, Comm. Math. Phys. 178 Ž1996. 179. hep-thr9504052.
w87x A. Bytsko, A. Fring, Factorized combinations of Virasoro characters, hep-thr9809001.
w88x W. Nahm, A. Recknagel, M. Terhoeven, Mod. Phys. Lett. A 8 Ž1993. 1835. hep-thr9211034.
w89x S. Dasmahapatra, R. Kedem, T. Klassen, B. McCoy, E. Melzer, Int. J. Mod. Phys. B 7 Ž1993. 3617.
hep-thr9303013.
w90x B. Richmond, G. Szekeres, J. Austral. Math. Soc. ŽSeries A. 31 Ž1981. 362.
w91x A. Kirillov, Prog. Theor. Phys. Suppl. 118 Ž1995. 61. hep-thr9408113.
w92x A. Kuniba, T. Nakanishi, Mod. Phys. Lett. A 7 Ž1992. 3487. hep-thr9206034.
w93x A. Kuniba, T. Nakanishi, J. Suzuki, Mod. Phys. Lett. A 8 Ž1993. 1649. hep-thr9301018.
w94x J. Suzuki, J. Phys. A 31 Ž1998. 6887. cond-matr9805242.
w95x J. Suzuki, Spinons in magnetic chains of arbitrary spins at finite temperatures, cond-matr9807076.
$
w96x P. Bouwknegt, K. Schoutens, Spinon decomposition and Yangian structure of s l n modules, in:
Geometric Analysis and Lie Theory in Mathematics and Physics, Australian Mathematical Society
Lecture Series 11, ed. A.L. Carey, M.K. Murray ŽCambridge University Press, Cambridge, 1997.,
q-algr9703021.
w97x A. Berkovich, B. McCoy, A. Schilling, Comm. Math. Phys. 191 Ž1998. 325. q-algr9607020.
w98x G. Hatayama, A. Kirillov, A. Kuniba, M. Okado, T. Takagi, Y. Yamada, Nucl. Phys. B 536 Ž1999. 575.
math. QAr9802085.
w99x G. Hatayama, A. Kuniba, M. Okado, T. Takagi, Y. Yamada, Remarks on fermionic formula, math.
QAr9812022.
w100x P. Goddard, W. Nahm, D. Olive, A. Schwimmer, Comm. Math. Phys. 107 Ž1986. 179.
w101x D. Bernard, J. Thierry-Mieg, Comm. Math. Phys. 111 Ž1987. 181.
w102x A. Nakayashiki, Y. Yamada, On spinon character formulas, in: Frontiers in Quantum Field Theories, ed.
H. Itoyama et al. ŽWorld Scientific, Singapore, 1996..
w103x Y. Yamada, On q-Clebsch Gordan rules and the spinon character formulas for affine C2Ž1. algebra,
q-algr9702019.
w104x G. Gasper, M. Rahman, Basic hypergeometric series, Encycl. of Math. and its Appl. 35 ŽCambridge
Univ. Press, Cambridge, 1990..
w105x A. Kirillov, N. Reshetikhin, J. Sov. Math. 41 Ž1988. 925.
w106x A. Nakayashiki, Y. Yamada, Selecta Math. ŽN.S.. 3 Ž1997. 547. q-algr9512027.
w107x T. Fukui, N. Kawakami, S.-K. Yang, Spin-S generalization of fractional exclusion statistics, cond-
matr9507143.
w108x T. Arakawa, T. Nakanishi, K. Oshima, A. Tsuchiya, Comm. Math. Phys. 181 Ž1996. 157. q-algr9507025.
w109x P. Bouwknegt, K. Schoutens, Phys. Rev. Lett. 82 Ž1999. 2757. cond-matr9805232.
Correlation functions of the XYZ model with a boundary

Yuji Hara
Institute of Physics, Graduate School of Arts and Sciences, UniÕersity of Tokyo, Tokyo 153-8902, Japan
Received 8 November 1999; accepted 15 December 1999
Abstract
Integral formulae for the correlation functions of the XYZ model with a boundary are
calculated by mapping the model to the bosonized boundary SOS model. The boundary K-matrix
considered here coincides with the known general solution of the boundary Yang–Baxter
equation. For the case of a diagonal K-matrix, our formulae reproduce the one-point function
previously obtained by solving the boundary version of the quantum Knizhnik–Zamolodchikov
equation. q 2000 Elsevier Science B.V. All rights reserved.
PACS: 75.10.Jm; 02.20.Tw; 05.50.qq

PACS: Solvable lattice models; Boundary; Vertex operator; Bosonization
1. Introduction
The XYZ model, the eight-vertex model and the ABF model are representative
examples of solvable lattice models in statistical mechanics w2x. In addition to the exact
results, e.g. the free energy, the one-point function, relations with many areas of
mathematics and physics have been studied: conformal field theory, integrable quantum
field theory, theory of Lie algebras and its representation theory. In the last years, the
boundary problems of the solvable lattice models have been investigated intensively.
Under a special boundary condition the models with boundaries are also solvable w21x. In
this case the boundary condition is described by the so-called K-matrix and the
integrability of the model is ensured by the boundary version of the Yang–Baxter
equation which includes the K-matrix in addition to the R-matrix.
There are several ways to solve the models. Among them the vertex operator
approach is the one which allows us to get deep insight into the symmetry of the model
and to obtain integral formulae for the correlation functions. This approach is first
launched on the XXZ model w15x. In this case the symmetry of the model is described
E-mail address: ss77070@komaba.ecc.u-tokyo.ac.jp ŽY. Hara..
PII: S 0 5 5 0 - 3 2 1 3 Ž 9 9 . 0 0 7 8 7 - 7
Y. Hara r Nuclear Physics B 572 [FS] (2000) 574–608 575
$
with the quantum affine algebra Uq Žs l2 .. Representation theoretical counter parts of the
space of the states, the transfer matrix and the creation operator of the elementary
excitation are the
$
level one highest modules, type I vertex operators and type II vertex
operators of Uq Žs l2 . respectively. By bosonizing the algebra, the modules and the vertex
operators, the Hamiltonian is diagonalized and the formulae of the correlation functions
are obtained. The same recipe is applied to the ABF model and its symmetry is
described by the deformed Virasoro algebra w19x. As for the boundary problems, the
half-infinite XXZ model with a diagonal K-matrix is studied by this approach in w13x.
The vacuum states are constructed in Fock spaces and the correlation functions are
obtained. For the boundary ABF model, diagonalization of the transfer matrix is carried
out in Ref. w20x.
On the other hand, for the XYZ model, the bosonization was, until recently, not
known. Difference equations of quantum Knizhnik–Zamolodchikov type are used to get
the correlation functions. The bulk case is discussed in w16x and the boundary case can
be found in w14x for a diagonal K-matrix. The bosonization is achieved in w17,18x by
mapping the model to the ABF model without restriction Žthe SOS model. through the
face-vertex correspondence. The correlation functions of the XYZ model are expressed
in terms of those of the SOS model with a non-local insertion called the tail operator.
The Baxter–Kelland formula for the one-point function is reproduced as a special case.
In this paper, we apply the bosonization scheme of w17x to the boundary problem of
the XYZ model. The K-matrix discussed here is obtained from the diagonal K-matrix of
the ABF model through the face-vertex correspondence. It has off-diagonal elements and
coincides with that of w12x which is a general solution of the boundary Yang–Baxter
equation. We obtain the integral formulae for the correlation functions for this general
boundary condition. And as a special case of the formulae the results of Refs. w13,14x are
reproduced.
In Section 2 we formulate the boundary XYZ model. In Section 3, we recall the
results of w20x. Modification to the unrestricted case is almost trivial. Section 4 is the
main part. After discussing the face-vertex correspondence, we give the formulae for the
correlation functions. In Appendix C, we detail the correspondence of our K-matrix and
the one of w12x.
2. The boundary XYZ model
2.1. The bulk weights and lattice Õertex operators
First we fix the convention of the R-matrix ŽFig. 1.. Originally the R-matrix was
found by Baxter as the local Boltzmann weight of the eight-vertex model w2x. Here we
Fig. 1. Graphical representation of R and K-matrix.

576 Y. Hara r Nuclear Physics B 572 [FS] (2000) 574–608
yy qq
follow Ref. w17x in which elements RŽ u.qqs RŽ u.yy are neglected compared to the one
in w2x. The elements are defined as
qq yy
a Ž u . s R Ž u . qq s R Ž u . yy
e 2e r e 2e r e 2e r
s yi k Ž u . R 0 Ž u . q4 i ž p
;i
p / ž
q4 i
p
u;i
p / ž
q1 i
p
Ž 1 y u . ;i
p / ,
qy yq
b Ž u . s R Ž u . qy s R Ž u . yq
;i
p / ž
q1 i
p
u;i
p / ž
q4 i
p
Ž 1 y u . ;i
p / ,
yq qy
c Ž u . s R Ž u . qy s R Ž u . yq
;i
p / ž
q4 i
p
u;i
p / ž
q4 i
p
Ž 1 y u . ;i
p / ,
yy qq
d Ž u . s R Ž u . qq s R Ž u . yy
;i
p / ž q1 i
p
u;i
p / ž
q1 i
p
Ž 1 y u . ;i
p / .
Ž 2.1 .
The notation of the theta functions are given in Appendix A. We use the parameters e
and r and consider the so-called principal regime
e ) 0, r ) 1, y1 - u - 1. Ž 2.2 .
We also use the following parameters:
x s eye , psx2r, zsx2u.
The common factor k Ž u. R 0 Ž u. is so chosen that the partition function per site is equal
to one,
ry 1 r
y2 y1 y1 y1
k Ž u . s zy 2r x 1y 2 Ž x 2 r ; x 2 r . ` Ž x 4 r ; x 4 r . ` Ž x 2zy1 ; x 2 r . ` Ž x 2 ry2z ; x 2 r . ` ,
Ž 2.3 .
ry 1
2r
rŽ z .
R0 Ž u. s z , Ž 2.4 .
r Ž zy1 .
x 4 z4 x 2rz4
rŽ z. s , Ž 2.5 .
x 2 z 4 x 2 rq2 z 4
z4 s Ž z ; x 4 , x 2r . `, Ž 2.6 .
`
Ž z ; p 1 , . . . , pN . ` s Ł Ž 1 y zp1n . . . pNn . .
1 N
Ž 2.7 .
n1 , . . . , n N s0
The R-matrix satisfies the following relations:
Ži. Yang–Baxter equation

R 12 Ž u1 y u 2 . R 13 Ž u1 y u 3 . R 23 Ž u 2 y u 3 .
s R 23 Ž u 2 y u 3 . R 13 Ž u1 y u 3 . R 12 Ž u1 y u 2 . , Ž 2.8 .
Žii. Unitarity relation
R 12 Ž u1 y u 2 . R 21 Ž u 2 y u1 . s id, Ž 2.9 .
Žiii. Crossing relation
X
R´´1X1´´X22 Ž 1 y u . s R´´X22 y
y´ 1
´1 Ž u. . Ž 2.10 .
We will call the half-infinite transfer matrix f´ Ž u. the ‘‘lattice vertex operator of
vertex type’’ and f´) Ž u. its ‘‘dual’’ ŽFig. 2. w16x. A vertex-path is a semi-infinite
sequence Ž . . . , pŽ2., pŽ1..,Õ Ž n. g q,y 4 . We denote by H i , i s 0 or 1 the eigenspace
of the corner transfer matrix w2x in the NE quadrant spanned by paths such that
pŽ n. s Žy1. ny i for n 4 1. Then the lattice vertex operators act on these spaces as
f´ Ž u . , f´) Ž u . : H i ™ H 1y i . Ž 2.11 .
All these are unrigorous picture but for the$
XXZ model, the model has the symmetry
described by the affine quantum$group Uq Žs l2 . and H i are identified with the level one
highest weight modules of Uq Žs l2 . w15x. And the lattice vertex operators are identified
with the intertwiners $
for these modules. For the eight-vertex model, the elliptic affine
quantum group Aq, p Ž s l2 . is expected to play the same role w10,11x.
2.2. The boundary XYZ model
The boundary of the model is formulated with a K-matrix which satisfies the
following equations w21x ŽFig. 1.:
Živ. Boundary Yang–Baxter equation

K 2 Ž u 2 . R 21 Ž u1 q u 2 . K 1 Ž u1 . R 12 Ž u1 y u 2 .
s R 21 Ž u1 y u 2 . K 1 Ž u1 . R 12 Ž u1 q u 2 . K 2 Ž u 2 . , Ž 2.12 .
Žv. Boundary unitarity relation
K Ž u . K Ž yu . s id, Ž 2.13 .
Žvi. Boundary crossing relation
X X
K ab Ž 1 y u . s b , ya
Ý Ryb a
a Ž 2 y 2 u. K b Ž u. .
X X
X
aX,b
The explicit form of the K-matrix will be given in Section 4.2.

Then the transfer matrix for the boundary eight-vertex model TB Ž u. is defined as
TB Ž u . s Ý f´) Ž yu . K´´ Ž u . f´ Ž u . .
X X Ž 2.14 .
´ ,´X
Fig. 2. Lattice vertex operators of vertex type.
The boundary Yang–Baxter equation ensures the integrability of this model

TB Ž u . ,TB Ž Õ . s 0. Ž 2.15 .
The Hamiltonian of the XYZ model with a boundary is defined as
1 `
x
HB s y Ý Ž Ž 1 y G . skq1 s kx q Ž 1 q G . s kq1
y
s ky
2 ks1
z z x y z
qDs kq 1 sk . q h x s1 q h y s1 q h z s1 , Ž 2.16 .
and it is connected to the boundary eight-vertex model as
p snh Ž 2 e Krp ,k . d
HBren s y z TB Ž u . , Ž 2.17 .
4K dz zs1
where
G s k snh2 Ž 2 e Krp ,k . , D s ycnh Ž 2 e Krp ,k . dnh Ž 2 e Krp ,k . , Ž 2.18 .
h x s 1x q h y s 1y s hq sq y
1 q hy s 1 ,
d y
hqs z Kq Ž u. ,
dz zs1
d q
hys z Ky Ž u. , Ž 2.19 .
dz zs1
d q y
h z sz Ž Kq Ž u . y Ky Ž u. . . Ž 2.20 .
dz zs1
See Appendix A for the elliptic functions snh, cnh, dnh and the elliptic modulus k, the
half-period magnitude K.
We assume that the boundary transfer matrix TB Ž u. acts on the well-defined
infinite-dimensional subspace of . . . m C 2 m C 2 m C 2 and denote it by HBi , i s 0 or 1.
The space HBi is the eigenspace of TB Ž u. spanned by paths such that pŽ n. s Žy1. ny i
for n 4 1. Then
TB Ž u . : HBi ™ HBi . Ž 2.21 .
Contrary to the bulk problems mentioned in Subsection 2.1, affine quantum algebras are
not the symmetry of the models anymore and the representation theoretical meaning of
the space HBi is not clarified yet even for the XXZ model w13x.
We assume the existence of the lowest energy state in each sector < i :B g HBi at least
for the regions
G s 0, D ™ y`, h z arbitrary, h x ,h y small,
and
G, Darbitrary, h z small, h x ,h y small.
When G s 0, D ™ y`, h x s h y s 0. The state is given by a path

<0:B s Ž . . . ,y,q,y . , <1:B s Ž . . . ,q,y,q . , Ž 2.22 .
and non-zero h z resolves the degeneration of the ground states. If we apply small h x ,h y ,
say d = h x , then we get for HB0
<0:B s Ž . . . ,y,q,y . y d Ž . . . ,y,q,q . Ž 2.23 .
and there is a finite gap between the next lowest energy state Ž . . . ,y,q,q . q
d Ž . . . ,y,q,y .. The case of the XXZ model with a diagonal K-matrix is discussed
with q-expansion in w13x.
3. The boundary SOS model
With some modifications we recall the result of Ref. w20x which concerns the
boundary problem for the ABF model with a diagonal K-matrix. In Ref. w20x, by
following Refs. w13,19x, the boundary transfer matrix TBŽ k . Ž u. is bosonized and the
eigenstates of TBŽ k . Ž u. are constructed in Fock spaces.
3.1. The SOS model
The local Boltzmann weight of bulk type is given as follows w1x. We follow the
convention of Ref. w17x ŽFig. 3..
a a"1
W ž a"1 a"2 /
u s R0 Ž u. , Ž 3.1 .
a a"1 w 1 xw a " u x
W ž a"1 a
u s R0 Ž u./w a xw 1 y u x
, Ž 3.2 .
a a"1 w a " 1 xw u x
W ž a.1 a
u s R0 Ž u./w a xw 1 y u x
, Ž 3.3 .
u2
yu
w ux sx r
Qx 2 r Ž x 2 u . , Qp Ž z . s Ž z ; p . ` Ž pzy1 ; p . ` Ž p; p . ` , Ž 3.4 .
R 0 Ž u. is given in Ž2.4.. The site variable a called height is an integer and satisfies the
admissibility: for variables a1 ,a 2 at adjacent sites
< a1 y a 2 < s 1. Ž 3.5 .
Fig. 3. Graphical representation of the bulk and boundary Boltzmann weights of the SOS model.
The difference between our formulation and Ref. w20x is that we do not impose the
restriction
1 ( a ( r y 1. Ž 3.6 .
We consider the so-called regime III:
e ) 0, r 0 1, 0 - u - 1. Ž 3.7 .
The Boltzmann weight satisfies the following relations:
Ži. Star-triangle relation

f g g d f e
ÝW
g
ž a b
uyÕ W
b /ž c
u W
g/ž d
Õ
/
a g f e e d
s ÝW
g
ž b c /ž
Õ W
a g /ž
u W g c uyÕ , / Ž 3.8 .
Žii. Unitarity relation

d g d c
ÝW
g
ž a b
u W
g /ž b /
y u s d ac , Ž 3.9 .
Žiii. Crossing relation
d c w ax a d
W ž a b
u s /
w bx
W
b ž c
1yu . / Ž 3.10 .
X
The X lattice vertex operators of face type Žhalf infinite transfer matrices. f Ž u. nn ,
f u n , nX s n " 1 are defined as in Fig. 4 w19x. A face-path is a semi-infinite sequence
) Ž .n
of integers Ž . . . ,a 2 ,a1 .. Among them, a Ž l,k .-path is the one having central heights k
and the boundary heights Ž l,l q 1., that is Ž . . . ,l q 1,l,l q 1,l, . . . ,k .. We denote by Hl, k
Fig. 4. Lattice vertex operators of face type.

the eigenspace of the corner transfer matrix w1x in the NW quadrant spanned by
Ž l,k .-paths. Then
kq ´ kq ´
f Ž u. k ,f ) Ž u. k : Hl , k ™ Hl , kq ´ . Ž 3.11 .
From graphical arguments, the following relations can be derived:
Ži. Commutation relation

a b a g a g
f Ž u 2 . b f Ž u1 . c s Ý W
g
ž b c 2 /
u y u1 f Ž u1 . g f Ž u 2 . c , Ž 3.12 .
Žii. Duality
kq ´
1 kq ´
f ) Ž u. k s f Ž 1 y u. k , Ž 3.13 .
wkx
Žiii. Inversion relation
a g a b
Ý f ) Ž u . g f Ž u . a s 1, f Ž u . b f ) Ž u . c s d ac . Ž 3.14 .
g
3.2. The boundary SOS model
The boundary weight: the K-matrix is given as a solution of
Živ. Reflection equation

c f c d g e
ÝW
f,g
ž b a
uyÕ W
f /ž g a /ž /ž /
uqÕ K f u K d g Õ
c d c f e g
s ÝW
f,g
ž f e
uyÕ W
b /ž g /ž /ž /
uqÕ K f g u K b Õ .
a
Ž 3.15 .
In Ref. w20x, the diagonal solution found by w6x is used ŽFig. 3.

k
K kq1ž k
u / s
w c q u xw k q c y u x
. Ž 3.16 .
k w c y u xw k q c q u x
K ky1ž k
u /
There are two regions A and B depending on the parameter c,
region A: x 2 c s yx 2 b , y1 - b - 1,
region B: x 2 c s x 2 b , y1 - b - 1, Ž 3.17 .
and u is restricted to satisfy
0 - u - < b < - 1. Ž 3.18 .
Then the unnormalized boundary transfer matrix is defined as

k k kq ´
TBŽ k . Ž u . s Ý f ) Ž yu . kq ´ K k q ´ž /
u f Ž u. k , Ž 3.19 .
´s"1
k
and we will denote its eigenspace spanned by Ž l,k .-paths as Hl,Bk . The reflection
equation Ž3.15. implies the integrability of the model,
TBŽ k . Ž u . ,TBŽ k . Ž Õ . s 0.
The ground state of the model depends on c as
k
ž
K kq1
k
u / ½s
) 1 if b ) 0,
Ž 3.20 .
k - 1 if b - 0.
ž
K ky1 u
k /
That is, the ground state belongs to Hk,Bk , Hky1, B
k for b ) 0, b - 0, respectively.
The normalization of the K-matrix is so chosen that the largest eigenvalue of the
boundary transfer matrix TBŽ k . Ž u. is 1. It also depends on the positivity of b. For
b ) 0 it is given by
Ž c. k
K) kq1ž u s hŽ)k . Ž u . ,
/ Ž 3.21 .
k
Ž c. k w c y u xw k q c q u x
K) ky1ž u s hŽ)k . Ž u .
/ , Ž 3.22 .
k w c q u xw k q c y u x
ry 1y2 k Žk. Žk.
2r
f Ž z . p) Ž z . p) Ž x 2zy1 .
hŽ)k . Ž u. s z Žk.
, Ž 3.23 .
f Ž zy1 . p) Ž zy1 . p)Ž k . Ž x 2z .
Ž x 2 rz 2 ; x 8 , x 2 r . ` Ž x 8z 2 ; x 8 , x 2 r . `
fŽ z . s , Ž 3.24 .
Ž x 6z 2 ; x 8 , x 2 r . ` Ž x 2q2 rz 2 ; x 8 , x 2 r . `
Žk.
x 2Ž1qc.z 4 x 2Ž rycykq1.z 4
p) Žz .s , Ž 3.25 .
x 2Ž ryc.z 4 x 2Ž cqk .z 4
and for b - 0
Ž c. k w c q u xw k q c y u x
K- kq1ž u s hŽ-k . Ž u .
/ , Ž 3.26 .
k w c y u xw k q c q u x
Ž c. k
K- ky1ž u s hŽ-k . Ž u . ,
/ Ž 3.27 .
k
2 ky1yr Žk. Žk.
2r
f Ž z . p- Ž z . p- Ž x 2zy1 .
hŽ-k . Ž u. s z Žk.
, Ž 3.28 .
f Ž zy1 . p- Ž zy1 . p-Ž k . Ž x 2z .
Žk.
x 2Ž1yc.z 4 x 2Ž cqkq1.z 4
p- Žz .s . Ž 3.29 .
x 2Ž rqc.z 4 x 2Ž rycyk .z 4
There are two more relations satisfied by the K-matrix:

Žv. Boundary unitarity relation
k k
Ž c.
Kc kXž / ž Ž c.
u Kc kX y u s 1, / Ž 3.30 .
k k
Žvi. Boundary crossing relation
k w kXX x kX k k
Ž c.
Kc kXž 1yu s Ý / W ž Ž c.
/ ž
y2uq2 Kc kXX u .
/
k k sk"1 w k x
XX k kXX k
Ž 3.31 .
For definiteness, we write down the boundary transfer matrix in the normalized
form:1
k k kq ´
TBŽ k . Ž u . s Ý Ž c.
f ) Ž yu . kq ´ K c kq´ ž /
u f Ž u. k . Ž 3.32 .
´s"1
k
3.3. Bosonization
Bosonization of the objects introduced in the previous sections is achieved with the
following oscillators w19x:
w n x 3x w r ) n x x
w bn , bm x s dnqm ,0 , Ž 3.33 .
n w 2 n x x w rn x x
w2 n x x
an s b , Ž 3.34 .
w nx x n
w P ,iQ x s 1, Ž 3.35 .
n yn
x yx
w nx x s , Ž 3.36 .
x y xy1
r ) s r y 1, Ž 3.37 .
and the Fock space Fl, k and its dual Fl,)k are defined as
Fl , k s C w by1 , by2 , . . . x < l,k : , Ž 3.38 .
bn < l,k : s 0, for n ) 0, Ž 3.39 .
r r)
P < l,k : s yl
ž ( 2r)
qk ( / 2r
< l,k : , Ž 3.40 .
Fl ,)k s ² l,k <C w b 1 , b 2 , . . . x , Ž 3.41 .
1
Eq. Ž2.17. of Ref. w20x is misprinted and the arguments of TBŽ k . should be exchanged.
² l,k < bn s 0, for n - 0, Ž 3.42 .

r r)
² l,k < P s ² l,k < yl
ž ( 2r)
( /
qk
2r
, Ž 3.43 .
² l,k < l,k : s 1. Ž 3.44 .

For Hl,Bk we assume
HlB, k ; Hl , k ( Fl , k . Ž 3.45 .
Bosonic vertex operators are given as
´K
F´ Ž u . s F´ Ž u . , Ž 3.46 .
wKx
r) r) r)
bn ( 2r
iQ ( 2r
Pq
4r
Fy Ž u . s :exp y
ž Ý
n/0 w n x x
zyn
/ := e z , Ž 3.47 .
dz 1 w u y u1 y 1r2 q K x
Fq Ž u . s EC 2p i z Fy Ž u . xy Ž u1 . , Ž 3.48 .
1 w u y u1 q 1r2x
an
žÝ w x / ( (
2r) 2r) r)
iQ y Pq
xy Ž u . s :exp zyn := ey r z r r , Ž 3.49 .
n/0 n x
K < F l , k s k = id F l , k , Ž 3.50 .
xy Ž u . : Fl , k ™ Fl , ky2 , Ž 3.51 .
Fy Ž u . : Fl , k ™ Fl , kq1 , Ž 3.52 .
Fq Ž u . : Fl , k ™ Fl , ky1 . Ž 3.53 .
xy Ž u. is a screening current.
The commutation relation, duality and inversion relation for F´ Ž u. are as follows:
F´ 2Ž u 2 . F´ 1Ž u1 .
K K q ´ 1X
s Ý
´ X1 , ´ X2
´ X1q ´ X2s ´ 1q ´ 2
W
ž Kq´2 Kq´1 q´2 2 /
u y u1 F´ X1 Ž u1 . F´ X2 Ž u 2 . , Ž 3.54 .
K
F´) Ž u . s ´Fy ´ Ž u y 1 . w K x Ž y . , Ž 3.55 .
g Ý F´) Ž u . F´ Ž u . s 1, Ž 3.56 .
´s"
g F´ Ž u . F´)X Ž u . s d´´ X , Ž 3.57 .

2 2 rq2
r) x 4 x 4
g s xy 2 r Ž x 2 . ` Ž x 2 r . ` 4 2 rq4
. Ž 3.58 .
x 4 x 4
Hence we identify the lattice vertex operators and bosonic operators as

k
f Ž u . kq ´ s F´ Ž yu . , Ž 3.59 .
kq ´
f ) Ž u. k s F´) Ž yu . . Ž 3.60 .
Then the boundary transfer matrix in the bosonic language is
k
TBŽ k . Ž u . s g Ý Ž c.
F´) Ž u . K c ž ky´ /
u F´ Ž yu . . Ž 3.61 .
´s"
k
Remark 1. F´ Ž u.,F´) Ž u. serve as vertex operators for Ref. w17x Xwhere Eq. Ž3.15. of
Ref. w17x is modified as g Ý nX Ž n y nX .w nX xŽy. nq 1F Ž u y 1. nnXF Ž u. nn s 1 and fixing the
parameter m s 1.
3.4. Boundary Õacuum states
The maximal eigenvectors of TBŽ k . Ž u. are called the boundary vacuum vectors. Under
the normalization of the K-matrix Ž3.21., Ž3.22., Ž3.26., Ž3.27., they satisfy
TBŽ k . Ž u . < k ,k :cB s < k ,k :cB g Fk , k , for b ) 0, Ž 3.62 .
TBŽ k . Ž u . < k y 1,k :cB s < k y 1,k :cB g Fky1, k , for b - 0, Ž 3.63 .
c
< Žk.
B k ,k TB
² Ž u. sBc ² k ,k < g Fk), k for b ) 0, Ž 3.64 .
c
< Žk.
B k y 1,k TB
² Ž u . sBc ² k y 1,k < g Fky1,
)
k for b - 0. Ž 3.65 .
It is easily shown with Ž3.14. that these are equivalent to
k
Ž c.
Kc ž
ky´ u F´ Ž yu . < k y i ,k :cB s F´ Ž u . < k y i ,k :cB ,
/ Ž 3.66 .
k
c k
² < )
B k y i ,k F´
Ž c.
Ž u. K c k y ´ u scB ² k y i ,k <F´) Ž yu . .
ž / Ž 3.67 .
k
Their explicit form is obtained by solving the equations w20x
c,k
< k y i ,k :cB s e F i < k y i ,k : , Ž 3.68 .
1
Fic , k s y Ý 2
k m bym q Ý bym Dmc ,,ik , Ž 3.69 .
2 m)0 m)0
c,k c,k Dmc ,,ik

eyF i bm e F i s bm y bym q , Ž 3.70 .
km
c,k c,k
eyF i bym e F i s bym , Ž 3.71 .
q q
Ž k y 1. m x Ž r y 2 c y k . m x w mr2x x w rmr2x x
Dmc ,,0k s y
w mx x w r ) mx x
y um
ž w m x x w r ) mr2x x / , Ž 3.72 .
Dmc ,,1k s Dmc ,,0k < k ™ kyr , Ž 3.73 .

c ™ cqr
and
c c,k
< ² < Gi ,
B k y i ,k s k y i ,k e
² Ž 3.74 .
1
Gic , k s y Ý x 4 mk m bm 2 q Ý bm Emc ,,ik , Ž 3.75 .
2 m)0 m)0
c,k c,k Emc ,,ik

eyG i bym e G i s bym y x 4 mbm q , Ž 3.76 .
km
c,k c,k
eyG i bm e G i s bm , Ž 3.77 .
q q
Ž r y k y 1. m x Ž 2 c q k . m x w mr2x x w rmr2x x
Emc ,,1k s x 2 m
w mx x w r ) mx x
2m
q x um
ž w m x x w r ) mr2x x / ,
Ž 3.78 .
Emc ,,0k s Emc ,,1k < k ™ kqr , Ž 3.79 .

c ™ cyr
where
x if m is even;
um s ½ 0 otherwise.
Ž 3.80 .
q
w m x x s x k q xyk . Ž 3.81 .
Note that, as a linear combination of paths, the boundary vacuum states coincide for the
restricted SOS and unrestricted SOS model.
3.5. AlternatiÕe bosonization
In Section 4 we need another bosonization such that
HlB, k ; Hl , k ( Fyl ,y k . Ž 3.82 .

It is given as follows. Vertex operators are realized as
kq ´
f Ž u. k s F´ Ž yu . , Ž 3.83 .
k
f ) Ž u . kq ´ s F´) Ž yu . . Ž 3.84 .
As for the boundary vacuum states, the conditions Ž3.67. are changed as
k
Ž c.
Kc ž
kq´ u F´ Ž yu . <k y i ,k:cB s F´ Ž u . <k y i ,k:cB ,
/ Ž 3.85 .
k
c k
² < )
B k y i ,k F´
Ž c.
Ž u. K c k q ´ u scB ²k y i ,k<F´) Ž yu . ,
ž / Ž 3.86 .
k
and the states are given by

c,k
< k y i ,k:cB s e F i < y k q i ,y k : , Ž 3.87 .
Fic , k s Fic , k < D nc ,,ik ™ D nc ,,ik Ž 3.88 .
c ™yc
Dnc,i, k s Dnc,1yi
,k <
, Ž 3.89 .
k ™ ryk
c c,k
< ² < Gi ,
B k y i ,k s y k q i ,y k e
² Ž 3.90 .
Gic , k s Gic , k < E nc ,, ik ™ E nc ,,ik , Ž 3.91 .
c ™yc
Enc ,i, k s Enc ,1yi
,k < . Ž 3.92 .
k ™ ryk
4. Correlation functions of the boundary XYZ model
In this section we consider the correlation functions for the sector i s 0 of HBi . The
case i s 1 can be handled in the same way.
4.1. Face-Õertex correspondence of the bulk weights
We recall the face-vertex correspondence of the bulk weights w3–5,17x. The intertwin-
ing vectors are given as 2
n
X 1 Ž nX y n . u q nX ip
tq Ž u . n s
'2 q3 ž 2r
;
2e r / , Ž 4.1 .
n
n
X Ž y. Ž nX y n . u q nX ip
ty Ž u . s n
'2 q0 ž 2r
X
;
2e r / ,
X
Ž 4.2 .
X
where nX s n " 1. Its conjugate, t´) Ž u. nn , primed, t´X Ž u. nn , and primed conjugate, t´X ) Ž u. nn ,
are given by ŽFig. 5.
n
X
n nX y n n
X
t´) Ž u . n s Ž y . C 2 ty ´ Ž u y 1 . n , Ž 4.3 .
w n xw u x
n
X w nX x
w ux n
X
t´X Ž u . n s t´ Ž u y 2 . n , Ž 4.4 .
w u y 1x w n x
n
X w u y 1 xw n x ) n
X
t´X ) Ž u . n s X t´ Ž u y 2 . n , Ž 4.5 .
w u xw n x
p
Cs ( er
e e r r4 . Ž 4.6 .
2
We fix the parameter ms1 in w17x.
Fig. 5. Graphical representation of intertwining vectors.
They satisfy the following relations ŽFig. 6.:

X
n n s n
Ý t´) Ž u . n t´ Ž u . nXX s dnX nXX , Ý t´)X Ž u . n t´ Ž u . s s d´ X ´ ,
´s" ssn"1
XX
n X n n s
Ý t´) Ž u . n t´
X Ž u . s dnX nXX ,
n Ý t´)X Ž u . s t´X Ž u . n s d´ X ´ ,
´s" ssn"1
X
n X n s n
Ý t´X ) Ž u. n t´ Ž u. n s d nX nXX , Ý t´X )X Ž u . n t´X Ž u . s s d´ X ´ . Ž 4.7 .
´s" ssn"1
The basic face-vertex correspondence is ŽFig. 7.

´X ´X n
X
s
X
Ý R Ž u y Õ . ´ 11 ´ 22 t´ X1 Ž u 0 y u . sX t´ X2 Ž u 0 y Õ . n
´ X1 , ´ X2 s "
nX sX u y Õ ,
X
n s
s Ý t´ Ž u 0 y Õ . s t´ Ž u 0 y u . nW
sgZ
2 1 ž s n / Ž 4.8 .
and from this identity and Ž4.7. we get the variants, for instance
s n ´ ´
Ý t´)X2 Ž u 0 y Õ . nX t´)X1 Ž u 0 y u . s R Ž u y Õ . ´ X11 ´ X22
´ X1 , ´ X2 s "
X
sX u y Õ t ) Ž u y u . s X t ) Ž u y Õ . nX ,
X
s Ý W n ž ´1 0 /
n ´2 0 Ž 4.9 .
X s n s
s gZ
X
n ´X ´ s
Ý t´X )X2 Ž u 0 y Õ . s R Ž u y Õ . ´ 11 ´ X22 t´X X1 Ž u 0 y u . n
´ X1 , ´ X2 s "
X X
s
X
s gZ
1
n
Ý t´X Ž u 0 y u . s W X
ž ns
X
n u y Õ tX ) Ž u y Õ . s .
sX ´2 0 / n Ž 4.10 .
Note that the principal regime of the XYZ model Ž2.2. is mapped to the regime III of the
SOS model Ž3.7..
Fig. 6. Graphical representation of Ž4.7..

Fig. 7. The face-vertex correspondence of bulk weights Ž4.8..
4.2. Face-Õertex correspondence of K-matrix
We make a K-matrix of vertex type K Ž u;c,l,u 0 . from the K-matrix of Section 3 as

Žsee Fig. 8.
´ l lq n l
K Ž u;c,l,u 0 . ´ X s Ý t´) Ž u 0 y u . lq n t´X X Ž u 0 q u . l Ž c.
K- ž
lqn u . / Ž 4.11 .
ns"1
l
This type of face-vertex correspondence was found in w8x. It can be easily verified by
graphical arguments that this K-matrix satisfies the boundary Yang–Baxter equation
Ž2.12.. As we will see below, through this face-vertex correspondence of the K-matrix
and those of the R-matrix, the sector i s 0 of the boundary XYZ model is mapped to the
region b - 0 of the boundary SOS model. And this fixes the normalization of the
K-matrix of the SOS model in Ž4.11..
The explicit forms of the elements are
´ C 2 w u0 q u x hŽ-l . Ž u . ´
K Ž u;c,l,u 0 . ´ X s K Ž u;c,l,u 0 . ´ X ,
2 w u 0 y u xw u 0 q u y 1 xw l x w c y u xw l q c q u x
Ž 4.12 .
q
K Ž u;c,l,u 0 .q
u q d q l ip u y d q l ip
s yq 0 ž 2r
;
2e r
q3
2r/ ž ;
2e r /
w c q u xw l q c y u x
u q d y l ip u y d y l ip
q q0 ž 2r
;
2e r
q3
2r / ž ;
2e r /
w c y u xw l q c q u x ,
y
K Ž u;c,l,u 0 .y
u q d q l ip u y d q l ip
s q3 ž 2r
;
2e r
q0 / ž
2r
;
2e r /
w c q u xw l q c y u x
u q d y l ip u y d y l ip
y q3 ž 2r
;
2e r
q0 / ž
2r
;
2e r /
w c y u xw l q c q u x ,
Fig. 8. The face-vertex correspondence of the K-matrix Ž4.11..

q
K Ž u;c,l,u 0 .y
lq1
uqdql ip uydql ip
s Ž y. q0 ž 2r
;
2e r / ž
q0
2r
;
2e r /w c q u xw l q c y u x
l
uqdyl ip uydyl ip
q Ž y. q0 ž 2r
;
2e r / ž q0
2r
;
2e r /w c y u xw l q c q u x ,
y
K Ž u;c,l,u 0 .q
l
uqdql ip uydql ip
s Ž y. q3 ž 2r
;
2e r / ž q3
2r
;
2e r /w c q u xw l q c y u x
lq1
uqdyl ip uydyl ip
q Ž y. q3 ž 2r
;
2e r / ž
q3
2r
;
2e r /w c y u xw l q c q u x ,
Ž 4.13 .
d s yu 0 q 1. Ž 4.14 .
We show below that the diagonal K-matrix of Ref. w14x is a special case of this
K Ž u;c,l,u 0 .. Our convention of parameters is that of Refs. w17,20x and the correspon-
dence with Ref. w14x is as follows. We attach the subscript ‘‘d’’ to those of Ref. w14x.
e ld
x s yq d , s ,
p 2 Kd
x 2 r s pd Ž elliptic nome. ,
kskd Ž elliptic modulus. ,
KsKd Ž half-period magnitude. ,
eu ud
z s zy2
d , s ,
p 2 Kd
R Ž u . < dŽ u.™ydŽ u. s R d Ž z d . . Ž 4.15 .
Then the diagonal K-matrix is as follows:
p ud
1 2Kd
K Ž zd . s Kˆ d Ž z d ;r d . , zd s e , Ž 4.16 .
f Ž z d ;r .
snh Ž hd q u d .
K̂ Ž z d ;r d . s
snh Ž hd y u d .
1
0 , r d s ep h d Kd
, for 0 - r d - 1, Ž 4.17 .
snh Ž hd q iK d q u d .
K̂ Ž z d ;r d . s
snh Ž hd q iK d y u d .
1
0 , r d s yep h d Kd
, for y1 - r d - 0.
Ž 4.18 .
The parameter r d is for the magnetic field h z and y1 - r d - 1 corresponds to the sector
i s 0. The parameter hd satisfies
p
0 - u d - yhd - l d m 0 - u - y hd - 1. Ž 4.19 .
2 Kde
Setting
d 1 ip r
s q , ls y c, Ž 4.20 .
r 2 2e r 2
the off-diagonal elements vanish and the diagonal elements are
1 ip ip
q
K Ž u;c,l,u 0 .q s yC 2 e
p2
4e r
q1 ž ;
2 er
q 2 0;/ ž
2e r /
1 ip
q12
ž ; /
4 2e r
u ip c ip u y c ip u q c ip
=q 2 ž ;
r 2e r / ž q2 ;
r er
q2 ; / ž
2r 2e r
q1 ;
2r 2e r
, / ž /
Ž 4.21 .
1 ip ip
y
K Ž u;c,l,u 0 .y s C 2 e
p2
4e r
q1 ž ;
2 er
q 2 0; / ž
2e r /
1 ip
q12 ; ž /
4 2e r
u ip c ip u y c ip u q c ip
=q 2 ž ;
r 2e r / ž q2 ;
r er
q1 ; / ž
2r 2e r
q2 ;
2r 2e r
. / ž /
Ž 4.22 .
From this we have
uyc ip uqc ip
KŽ
q
u;c,l,u 0 q
y
.
sy
q2
2r 2e r ž2r 2e r
; / ž
q1 ; /
KŽ u;c,l,u 0 y . u y c ip u q c ip
q1 ; q2 ž ;
2r 2e r / ž 2r 2e r /
e i2 e r e i2 e r
sy
q0 ž ip
e
Ž u yc. ;
p / ž
q1
ip
e
Ž u qc. ;
p /
q1 ž pŽi
u y c . ;i2 e rp q 0 / ž pŽ i
u q c . ;i2 e rp /
2 Ke
sy
snh u d q ž p
c;k d / . Ž 4.23 .
2 Ke
snh u d y ž p
c;k d /
Therefore the parameter for the magnetic field hd is identified with c as

2 Kde
hd q iK d s c for y1 - r d - 0, Ž 4.24 .
p
2 Kde
hd s c for 0 - r d - 1. Ž 4.25 .
p
Thus, from Ž4.19., the region considered in w14x is mapped to the regions of the SOS
model Ž3.17. as
ip
y1 - r d - 0 ™ 0 - u - yc q - 1, Ž i.e. b - 0of region A . , Ž 4.26 .
2e
0 - r d - 1 ™ 0 - u - yb - 1, Ž i.e. b - 0of region B . , Ž 4.27 .
K Ž u;c,l,u 0 . does not satisfy the boundary crossing relation and K d Ž z d ;r d . and
K Ž u;c,l s 2r y c,u 0 s 1 y 2r y 2ipe . differ by over all factor. Therefore we should change
the intertwining vectors as
X X
n n
tńew Ž u . n s f Ž u . t´ Ž u . n ,
X
n 1 X
n
t´) new Ž u . n s t´) Ž u . n ,
f Ž u.
X X
n n
t´X new Ž u . s f Ž u . t´X Ž u . n ,
n
X
n 1 X
n
t´) new Ž u . n s t´) Ž u . n , Ž 4.28 .
f Ž u.
where f Ž u. satisfies
f Ž u0 y 1 q u. f Ž u0 q u. w u 0 y u q 1x
s , Ž 4.29 .
f Ž u0 q 1 y u. f Ž u0 y u. w u0 q u x
l lq n l
K d Ž z ;r d . s Ý t´) new Ž u 0 y u . lq n t´X new
X Ž u0 q u. l Ž c.
K- ž lqn u /
ns"1
l
r ip
s
ž
f 1y
2
y
2e
qu / ž K u;c,l s
r
y c,u 0 s 1 y
r
y
ip
/ . Ž 4.30 .
r ip 2 2 2e
ž
f 1y
2
y
2e
yu /
But these new intertwining vectors satisfy the relations Ž4.7. as before and in the
formulae for the correlation function Ž4.36., Ž4.42. the vectors always consist such a pair
that the factor f Ž u.’s cancel. Therefore we can substitute the vectors dressed with f Ž u.
for those not dressed.
It can be verified that this K Ž u;c,l,u 0 . coincides with K g Ž u g ; j g , l g , m g .: the general
solution of the boundary Yang–Baxter equation obtained in w12x. ŽWe use the subscript
g for w12x.. The relations between the parameters c,l,u 0 4 l j g , l g , m g 4 are highly
complicated and discussed in Appendix C.
4.3. Correlation functions
In this section we give the integral formulae for the correlation functions of the
boundary XYZ model. For clarity we mainly discuss the special case of one-point
functions, i.e. the boundary magnetization, but generalization to the N-point case is
straightforward though cumbersome.
4.3.1. Basic idea

r X
We denote the matrix unit operator acting on the site r as E´´ e.g.
r
E ".s sr " , r
Eqq r
y Eyys sr z .
Our strategy is represented as the following naive equation
B
²0 < E´1X ´ <0:B
s Ž The partition function of Fig. 9 . s lim
X
²0 < f´)X Ž j X . f´ Ž j . <0:B
j , j ™1 B
s lim
X
Ž Fig . 10 . s lim
X
Ž Fig . 11.
j , j ™1 j , j ™1
c X
s s
s lim
X Ý ² l y 1,l < f ) Ž ÕX . lsX f Ž Õ . s LX Ž u 0 . l < l
j , j ™1 X
s, s B
X
l s
y 1,l :cB = t´)X Ž u 0 y Õ . sX t´ Ž u 0 y Õ . s , Ž 4.31 .
where
X
jsx2Õ, j Xsx2Õ . Ž 4.32 .
The first equality comes from Ž2.17. and the same reasoning as in the bulk case w2,15x.
The second and the third equalities are argued in Section 3 of Ref. w14x. They argued the
Fig. 9. The boundary eight-vertex model. Lattice extends to the north, the west and the south.
Fig. 10. The boundary eight-vertex model corresponding to the boundary magnetization.
equivalence of the two lattices of Figs. 10 and Fig. 11 and made the following
identification:
<0:B ; Ž Upper half lattice of Fig. 9 . , Ž 4.33 .
B
²0 < ; Ž Lower half lattice of Fig. 9 . . Ž 4.34 .
Fig. 11. The boundary SOS model corresponding to the boundary magnetization. Summations including
intertwining vectors are emphasized with black dots for clarity.
The fourth equality is from the equivalence of two lattices in Fig. 12. Iteration of
face-vertex correspondences of the local Boltzmann weights on one lattice yields the
other. This is the boundary version of w17x.
The fifth equality comes from a similar argument for the second and third above.
Thus the one-point function for the XYZ model is equivalent to the two-point
function of the boundary SOS model with the insertion of the tail operator LX Ž u. nm . Here
m s n y 2 k Ž k g Z. because the number of vertex operators is even in Ž4.31.. Graphical
arguments show that the tail operator satisfies the commutation relation
X X X
LX Ž u 0 . s f ) Ž u . n s Ý LX snX s u y u f ) Ž u . nX LX Ž u . s ,
n s
nX 0 s 0 n Ž 4.35 .
X
s
X
sX Õ s
X
LX n
n n
Ý t´X Ž Õ . sX t´) Ž Õ . s . Ž 4.36 .
s n
´s"1
To bosonize LX Ž u 0 . nny 2 k we have to consider the two cases k ( 0 and k 0 0 separately
Fig. 12. Face-vertex correspondence of finite size lattice.

w17x. In the former case, we use the bosonization of Sections 3.3 and 3.4, and the tail
operator is given by
ny 2 k y2 k y2 k K
LX Ž u . n s LX Ž u . s L Ž yu . Ž y. Ž k g N. , Ž 4.37 .
y2 k
w Ky2k x
Ž y. Ky1 k ,
Ž .
LŽ u. s Xyk Ž u . Ž 4.38 .
wKx
dz 1 w u y u1 y 1r2 q K x
Xy Ž u . s E 2p i z xy Ž u1 . . Ž 4.39 .
1 w u y u1 q 1r2x
Note that
Fq Ž u . s Fy Ž u . Xy Ž u . .
The commutation relation on a Fock space is
y2 k
LX Ž u 0 . F´) Ž yu .
Kq2ky´ Kq2k X
y2 ky ´ q ´
s Ý LX u 0 y u F´)X Ž yu . LX Ž u 0 . . Ž 4.40 .
´ Xs"
Ky´X K
For k - 0, Ž4.38. is meaningless and we use the bosonization of Section 3.5 and
ny 2 k 2k 2k K
LX Ž u . n s LX Ž u . s L Ž yu . Ž y . Ž yk g N . . Ž 4.41 .
X
Normalizing L Ž u. nm is not necessary for our purpose, see e.g. Eq. Ž4.45..
4.3.2. K-matrix of general type

We evaluate the correlation with two types of bosonization in Section 3 as
c X
s s
B
²0 < f´)X Ž j . f´ Ž j . <0:B s ÝB ² l y 1,l < f ) Ž Õ . lsX f Ž Õ . s LX Ž u 0 . l < l y 1,l :cB
X
s s l"1
X
s s s "1
X
l s
= t´)X Ž u 0 y Õ . sX t´ Ž u 0 y Õ . s
lc
s g Ž y . B ² l y 1,l <Fq) Ž yÕ . Fq Ž yÕ . < l
l ly1
y 1,l :cB = t´)X Ž u 0 y Õ . ly1 t´ Ž u 0 y Õ . l
q gcB ² l y 1,l <Fq) Ž yÕ . Fy Ž yÕ . LXy2 Ž u 0 . < l

l ly1 lc
y 1,l :cB = t´)X Ž u 0 y Õ . ly1 t´ Ž u 0 y Õ . ly2 q g Ž y . B
=²l y 1,l<Fq) Ž yÕ . Fq Ž yÕ . <l y 1,l:cB

l lq1
=t´)X Ž u 0 y Õ . lq1 t´ Ž u 0 y Õ . l
q gcB ²l y 1,l<Fq) Ž yÕ . Fy Ž yÕ . LXy2 Ž u 0 . <l y 1,l:cB

l lq1
= t´)X Ž u 0 y Õ . lq1 t´ Ž u 0 y Õ . lq2 . Ž 4.42 .
The boundary vacuum expectation values in this formula are one-fold integrals and the
essential part of the integrands is
c
<
B k y 1,k :Fy
² Ž yÕ y 1 . Fy Ž yÕ . xy Ž u1 . : < k y 1,k :cB
c
< :c
B k y 1,k k ,k y 1 B
²
l r)
3
r
y1 z 12 2r Ž x2 ; x2 .`
s Ž x j z1 .
ž /
xj 2
Ž x2 .`Ž x4 ; x4 .`
2
Qx 2 r Ž x 2Ž cqr .rj . Qx 2 r Ž x 2Ž cqk .j . Qx 4 Ž jy2 .

=
Qx 2 r Ž jy2 .
Qx 4 Ž zy2
1 .
=
Qx 2 r Ž xy2 cy1rz1 . `Qx 2 r Ž x 2Ž cqk .y1rz1 . `
Qx 2 r Ž xz1rj . Ž x 3j z 1 . ` Ž x jy1 zy1
1 .`
= . Ž 4.43 .
Qx 2 Ž x jrz1 . Ž x 3j z 1 ; x 2 . ` Ž x jy1 zy1 2
1 ; x .`
Then the boundary magnetization with a spectral parameter is given as

B
²0 < s 1z <0:B
s lim M Ž0. Ž j ;c,l,u 0 . . Ž 4.44 .
² < :
B 00 B j™1
We used the coherent states to obtain this formula, see Ref. w20x for details.
M Ž0. Ž j ;c,l,u 0 .
´ B ²0 < f´) Ž j . f´ Ž j . <0:B
s Ý
´ s "
² < :
B 00 B
X
l s s
c
< )
B l y 1,l f
² Ž Õ . sX f Ž Õ . s LX Ž u 0 . l < l y 1,l :cB
s Ý l
X
s sl"1
c
< X
B l y 1,l L
² Ž u 0 . l l y 1,l :cB
X
sss "1
X
l s
= Ý ´ t´) Ž u 0 y Õ . sX t´ Ž u 0 y Õ . s
´ s "
s F Ž j ;c,l,u 0 . q F Ž j ;y r y c,2 r y l,u 0 . , Ž 4.45 .

c
² < )
B l y 1,l Fq yÕ Fq yÕŽ . Ž . < l y 1,l :cB
F Ž j ;c,l,u 0 . s g c
:c
B l y 1,ll y 1,l B
²
l ly1
= Ý ´ t´) Ž u 0 y Õ . ly1 t´ Ž u 0 y Õ . l
´s"
c
< )
B l y 1,l Fq
² Ž yÕ . Fy Ž yÕ . LXy2 Ž u 0 . < l y 1,l :cB
qg l
Ž y . Bc ² l y 1,ll y 1,l :cB
l ly1
= Ý ´ t´) Ž u 0 y Õ . ly1 t´ Ž u 0 y Õ . ly2
´ s "
²²Fy Ž yÕ y 1 . Fy Ž yÕ . ::
s yg
wlx EC
dz 1 h 4 Ž u1 q u 0 y 1r2 . h 4 Ž l y Õ y u1 y 1r2 .
=
2p i z 1 w yÕ y u1 q 1r2xw yu 0 y u1 q 1r2x
c
<
B l y 1,l :Fy
² Ž yÕ y 1 . Fy Ž yÕ . xy Ž u1 . : < l,l y 1:cB
= c
:c
B l y 1,ll,l y 1 B
²
=²²Fy Ž yÕ y 1 . xy Ž u1 . ::²²Fy Ž yÕ . xy Ž u1 . ::
dz 1
s E F Ž j , z 1 ;c,l,u 0 . Ž 4.46 .
C 2p i z 1
and
1 Qx 2 r Ž x 2Ž cqr .rj . Qx 2 r Ž x 2Ž cql .j . Qx 4 Ž jy2 .
F Ž j , z 1 ;c,l,u 0 . s y
wlx Qx 2 r Ž jy2 .
h 4 Ž u1 q u 0 y 1r2 . h 4 Ž l y Õ y u1 y 1r2 .
=
w yÕ y u1 q 1r2xw yu 0 y u1 q 1r2x
lrr y1rr
Qx 2 r Ž x z 1rj . Qx 2 r Ž x zy1
1 j
y1
.
= Ž x jz 1 . xy1 Ž jz 1 .
Qx 2 Ž x jrz 1 . Qx 2 Ž x jz 1 .
4
Qx 4 Ž zy2
1 . Ž x2 ; x2 .`
= , Ž 4.47 .
Qx 2 r Ž xy2 cy1rz 1 . Qx 2 r Ž x 2Ž cql .y1rz 1 . Ž x 4 ; x 4 . 2`
where ²² . . . :: are the factors for normal ordering given in Appendix B and
u2
u ip yu
h1 Ž u . s Cq1 ž ; s w u x s fr1 Ž u . Qx 2 r Ž x 2 u . ,
/ fr1 Ž u . s x r Ž 4.48 .
r er
u ip ip p2 ip u
h 4 Ž u . s Cq4 ž ;
r er /
s fr4 Ž u . u y
2e
, fr4 Ž u . s ey 4 r e y r . Ž 4.49 .
The contour C is such that
z 1 s x 2 nq1j , x 2 nq1jy1 , x 2 r ny2 cy1 , x 2 r nq1zy1
0 ,x
2 r nq2Ž cql .y1
Ž n 0 0.
are inside and
z 1 s x 2 ny1j , x 2 nq1jy1 , x 2 r ny2 cy1 , x 2 r nq1zy1
0 ,x
2 r nq2Ž cql .y1
Ž n - 0.
y1
are outside of it. Note that z 1 s x j is not a pole of F Ž j , z 1;c,l,u 0 ..
We used the following identities to make two terms in Ž4.46. together
q4 Ž 2 x . q4 Ž 2 y . q1 Ž 2 z . q1 Ž 2 t .
s q4 Ž x q y q z q t . q4 Ž x q y y z y t . q1 Ž x y y y z q t . q1 Ž x y y q z y t .
y q4 Ž x q y q z y t . q4 Ž x q y y z q t . q1 Ž x y y q z q t . q1 Ž x y y y z y t .
and
n 1 qn 3
n n Ž y. Ž n1 y n 2 . n 4 n3
Ý ´ t´) Ž u . n12 t´ Ž u . n 43 s K
n1 n2
u ,
´s" w n1 x
where
n y nX n q nX
K
nX nX " 1
u s
2 ž
h4 u "
h4
2 / ž / , Ž 4.50 .
n n"1 h1 Ž u .
n q nX n y nX
K
nX nX . 1
u s
h 4 u "
2
h 4 ž2 / ž / . Ž 4.51 .
n n"1 h1 Ž u .
4.3.3. Diagonal K-matrix

From Ž4.20., for the case of the diagonal K-matrix discussed in w14x we have
r r ip
ž
M Ž0. Ž j ;c . s M Ž0. j ;c,ls y c,u 0s1 y y
2 2 2e /
s F Ž j ;c . q F Ž j ;y r y c . , Ž 4.52 .
where
r r ip dz 1
ž
F Ž j ;c . s F j ;c,l s y c,u 0 s 1 y y
2 2 2e
s X
C 2p i z 1
F Ž j , z 1 ;c . , / E
Ž 4.53 .
r r ip
ž
F Ž j , z 1 ;c . s F j , z 1 ;c,ls
2
y c,u 0s1 y
2
y
2e /
1 Qx 2 r Ž x 2Ž cqr .rj . Qx 2 r Ž x rj . Qx 4 Ž jy2 .
s r
yc Qx 2 r Ž jy2 .
2
ry1
=
h4 ž 2
y c y Õ y u1 / Qx 2 r Ž x z 1rj .
1yr ip Qx 2 Ž x jrz 1 . Qx 2 Ž x jz 1 .
yu1 y q
2 2e
1 1qc 3 1qc
Qx 4 Ž zy2
1 .
1 c
2
y
r
y
= x y
2
y
r
yr
j z 12 r
Qx 2 r Ž xy2 cy1rz 1 .
1yr ip 1yr ip
=
ž
fr4 u1 q
2
y
2e / ž fr1 u1 q
2
y
e /Ž 4
x2 ; x2 .`
.
1 Ž x4; x4.`
2
ž
fr1 yÕ y u1 q
2 /
Ž 4.54 .
The contour C X is such that

z 1 s x 2 nq1j , x 2 nq1jy1 , x 2 r ny2 cy1 ,y x 2 r nqry1 Ž n 0 0.
are inside and
z 1 s x 2 ny1j , x 2 nq1jy1 , x 2 r ny2 cy1 ,y x 2 r nqry1 Ž n - 0.
are outside of it. As before, z 1 s xy1j is not a pole of F Ž j , z 1;c,l,u 0 .. We can reproduce
the result of w14x as a special case of these formulae:
ip ip dz 1
ž
M Ž0. j ;c s
2e / ž
s Res z 1s y xy1 ; F j , z 1 ;cs ž /
2 e 2p i z 1 /
Ž x 2jy1 ; x 2 . ` Ž x 2j ; x 2 . ` Ž yx 2 rjy1 . ` Ž yx 2 rj . `
s
Ž yx 2jy1 ; x 2 . ` Ž yx 2j ; x 2 . ` Ž x 2 rjy1 . ` Ž x 2 rj . `
2 2
Ž x 2 ; x 2 . ` Ž yx 2 r . `
= 2 2
, Ž 4.55 .
Ž yx 2 ; x 2 . ` Ž x 2 r . `
Ž z .` s Ž z ; x2r .`, Ž 4.56 .
where we used
ip ip
ž
F j , z 1 ;c s
2e / ž
s yF j , z 1 ;c s yr y
2e / . Ž 4.57 .
The formula for the difference of the boundary magnetizations in w14x can also be
reproduced. We have
M Ž0. Ž j ;c . s M Ž0. Ž jy1 ;c . , Ž 4.58 .
M Ž0. Ž j ;c . s yM
M Ž1. Ž j ;y c . , Ž 4.59 .
which can be seen easily by physical arguments, and are discussed rigorously in Ref.
w13x. Hence
M Ž0. Ž j ;c . y M Ž1. Ž j ;c . s M Ž0. Ž j ;c . q M Ž0. Ž jy1 ;y c . Ž 4.60 .
and the r.h.s. can be written in the simple form
dz 1
ž
M Ž0. Ž j ;c . q M Ž0. Ž jy1 ;y c . s Res z 1 s xy2 cy1 ; F Ž j , z 1 ;c .
2p i z 1 /
dz 1
ž
y Res z 1 s xy1jy1 ; F Ž j , z 1 ;c .
2p i z 1 /
dz 1
ž
y Res z 1 s x 2 cy1 ; F Ž j , z 1 ;y r y c .
2p i z 1 /
dz 1
ž
y Res z 1 s xy1jy1 ; F Ž j , z 1 ;y r y c .
2p i z 1
,
/
Ž 4.61 .
where we used
dz 1 dw1
F Ž jy1 , z 1 ;y c . s F Ž j ,w1 ;c . , Ž 4.62 .
2p i z 1 2p iw1
dz 1 dw1 1
F Ž jy1 , z 1 ;y r q c . s F Ž j ,w 1 ;y r y c . , w1 s . Ž 4.63 .
2p i z 1 2p iw1 x 2z 1
Therefore
dz 1
M Ž1. Ž 1;c . y M Ž0. Ž 1;c . s y2Res z 1s xy2 cy1 ; F Ž 1, z 1 ;c .
ž 2p i z 1 /
2 2 2 2
Ž x r . ` Ž yx r . ` Ž x 2 rq2 c . ` Ž xy2 c . `
s2
Ž x rq 2 c . ` Ž yx rq2 c . ` Ž x ry2 c . ` Ž yx ry2 c . `
2 2
Ž x 4 cq2 ; x 4 . ` Ž xy4 cq2 ; x 4 . ` Ž x 4 ; x 4 . ` Ž x 2 ; x 2 . `
= 2 2 4
.
Ž x 2q2 c ; x 2 . ` Ž xy2 c ; x 2 . ` Ž x2r .`
Ž 4.64 .
Note that we are taking <0:B as the ground state. We can reproduce the result of w13x by
taking the limit x 2 r ™ 0,
lim Ž M Ž 1 . Ž 1;c . y M Ž 0 . Ž 1;c . .

x 2 r™0
Ž x 4 cq2 ; x 4 . ` Ž xy4 cq2 ; x 4 . ` 2 2

s2
2q2 c 2 2 Ž x4 ; x4 .`Ž x2 ; x2 .`, Ž 4.65 .
Žx ; x2 ` . Žx 2y2 c
; x2 `
.
where the parameter r of w13x is identified with xy2 c.
4.3.4. N-point correlation function

The N-point correlation function can be obtained in the same manner as discussed in
w13,15x. But the formula is highly complicated. Here we write down only the essential
part:
B
²0 < f´)X Ž j 1X . . . . f´)X Ž j NX . f´ Ž j N . . . . f´ Ž j 1 . <0:B
1 N N 1
c X X
s s
s ÝB ² l y 1,l < f ) Ž ÕX1 . lsX1 f ) Ž ÕX2 . sX21 . . . f ) Ž ÕNX . sXNNy 1
X X
s1 . . . s N
s1 . . . s N
X
s s s s
=f Ž ÕN . sNN f Ž ÕNy1 . sNNy 1 . . . f Ž Õ 1 . s12 LX Ž u 0 . l 1 < l y 1,l :cB
X X
l s s
=t´)X1 Ž u 0 y ÕX1 . sX1 t´)X2 Ž u 0 y ÕX2 . sX21 . . . t´)XN Ž u 0 y ÕXN . sXNNy 1
X
s s s
=t´ N Ž u 0 y ÕN . sNN t´ Ny 1Ž u 0 y ÕNy1 . sNNy 1 . . . t´ 1Ž u 0 y Õ 1 . s12
c
sg ÝB ² l y 1,l <Fn)X Ž yÕX1 . . . . Fn)X Ž yÕNX .
1 N
n i , n iX
n iy n iX(0
N
X
Ý n iy n i Ž u 0 . < l y 1,l :cB
=Fn N Ž yÕN . . . . Fn 1Ž yÕ1 . LX is 1
X
iy1 n n N n Xm jq1 n n
ly Ý ly Ý q Ý
ns1 ms1 nsN
=Ł t´)Xi Ž u 0 y ÕiX . i
t´ iŽ u 0 y Õi . j
i N
ly Ý ly Ý q Ý nn
ns1 n Xn ms1 n mX nsN i
c
qg Ý ²l y 1,l<Fn)X Ž yÕX1 . . . . Fn)X Ž yÕNX . Fn Ž yÕN . . . . Fn Ž yÕ1 .
1 N N 1
n i , n iXn iy n iX-0
B
X
iy1 n n
N lq Ý
X ns1
=L Ý n y n Ž u 0 . <l y 1,l:cB Ł t´) Ž u 0 y ÕXi .
X
is 1 i
i X
i i
t´ iŽ u 0 y Õi .
i
lq Ý
ns1 n Xn
N n Xm jq1 n n
lq Ý y Ý
ms1 nsN
= N j . Ž 4.66 .
lq Ý y Ý
ms1 n Xm n s N nn
Note that each term in the r.h.s. is an N-fold integral.
5. Discussion
In Ref. w17x, ‘‘m and u 0 ’’-independence is carefully discussed. We explain this point
with Fig. 12. For the bulk problem, the correlation functions of the eight-vertex model
should not depend on the parameters of the surrounding intertwining vectors Ž l and u 0 .
in the thermodynamic limit. ŽWe use l for the m of Ref. w17x.. This is non-trivial
because the essential part of the correlation function has the form
n
trF l , n Ž F Ž u y 1 . n12 . . . L Ž u 0 . x 4 H n . , Ž 5.1 .
2
where Hn is the corner transfer Hamiltonian of the ABF model Žsee Eq. Ž5.14. of Ref.
w17x.. In our case, even in the thermodynamic limit l and u 0 remain as the parameters of
the K-matrix. Therefore the correlation functions should depend on them as they do.
It is argued in w7x that the general solution of the reflection equation of the ABF
model can be constructed from that of the eight-vertex model through the face-vertex
correspondence. Combining this with the argument of Section 4.2, for the ABF model
we can construct the general solution from the diagonal K-matrix. But this method is not
applicable for the eight-vertex model since in our construction of Ž4.11. we can not use a
non-diagonal K-matrix of the ABF model.
Finally we would like to mention the related work of Ref. w9x. In this paper the Bethe
ansatz equation is obtained for the eight-vertex model with two-sided boundaries. The
K-matrix considered is a general solution of the boundary Yang–Baxter equation.
Acknowledgements
The author thanks Hitoshi Konno, Satoru Odake, Yaroslav Pugai, Jun’ichi Shiraishi
and Junji Suzuki for discussions and interest. He is especially indebted to Michio Jimbo
and Atsuo Kuniba for their advice and encouragement.
Appendix A. Notations and definitions of functions
A.1. Elliptic functions
We follow Ch. 15 of Ref. w2x with some modifications. The qi Ž u;t .’s are given as
q1 Ž u;t . s H Ž 2 Iu . , q 2 Ž u;t . s H1 Ž 2 Iu . , Ž A.1 .
q 3 Ž u;t . s Q Ž 2 Iu . , q4 Ž u;t . s Q 1 Ž 2 Iu . , Ž A.2 .
ip t
e s q, Ž A.3 .
X
where the right-hand sides are of Ref. w2x. k,k are elliptic moduli and for half-period
magnitudes we use K, K X instead of I, I X . snŽ u,k ., cnŽ u,k . and dnŽ u,k . are the same as
in w2x. We also use the following:
snh Ž u,k . s yisn Ž iu,k . , cnh Ž u,k . s cn Ž iu,k . , dnh Ž u,k . s dn Ž iu,k . .
Ž A.4 .
For convenience, we gather some notations for elliptic functions
u2
u ip yu
h1 Ž u . s Cq1 ž ; s w u x s fr1 Ž u . Qx 2 r Ž x 2 u . ,
/ fr1 Ž u . s x r
, Ž A.5 .
r er
u ip ip p2 ip u
h 4 Ž u . s Cq4 ž ;
r er /
s fr4 Ž u . u y
2e
, fr4 Ž u . s ey 4 r e y r , Ž A.6 .
Qp Ž z . s Ž z ; p . ` Ž pzy1 ; p . ` Ž p; p . ` , Ž A.7 .
p
Cs ( er
e e r r4 . Ž A.8 .
A.2. Other functions

z4 s Ž z ; x 4 , x 2r . `, Ž A.9 .
Ž z .` s Ž z ; x2r .`, Ž A.10 .
`
Ž z ; p 1 , . . . , pN . ` s Ł Ž 1 y zp1n . . . pNn . .
1 N
Ž A.11 .
n1 , . . . , n N s0
Appendix B. Formulae for normal ordering

We list the formulae for normal ordering. For operators A, B, we write down
²² AB :: such that AB s ²² AB :: = : AB:.
2 z2 Ž x 2 z 2rz1 . `
²² xy Ž u1 . xy Ž u 2 . :: s zy
1 r
q2
1y ž z1 / Ž x 2 ry2 z 2rz1 . `
, Ž B.1 .
1
²²Fy Ž u1 . xy Ž u 2 . :: s z 1r
y1 Ž x 2 ry1 z 2rz1 . `
, Ž B.2 .
Ž xz 2rz1 . `
1
²² xy Ž u 2 . Fy Ž u1 . :: s z 2r
y1 Ž x 2 ry1 z1rz 2 . `
, Ž B.3 .
Ž xz1rz 2 . `
r)
x 2 z 2rz1 4 x 2 rq2 z 2rz1 4
²²Fy Ž u1 . Fy Ž u 2 . :: s z 12 r . Ž B.4 .
x 4 z 2rz1 4 x 2 r z 2rz1 4
As meromorphic functions, the following commutation relations hold:
w u1 y u 2 y 1 x
xy Ž u1 . xy Ž u 2 . s x Žu . x Žu ., Ž B.5 .
w u1 y u 2 q 1 x y 2 y 1
w u1 y u 2 q 1r2x
Fy Ž u1 . xy Ž u 2 . s y x Ž u .F Ž u . , Ž B.6 .
w u1 y u 2 y 1r2x y 2 y 1
Fy Ž u1 . Fy Ž u 2 . s R 0 Ž u1 y u 2 . Fy Ž u 2 . Fy Ž u1 . . Ž B.7 .
Appendix C. Correspondence of the K-matrix with Ref. [12]

As claimed in Section 4.2, we detail how K Ž u;c,l,u 0 . coincides with the K-matrix
given in w12x. The subscript g is for those of w12x which is a general solution of the
boundary Yang–Baxter equation. The relations between parameters are given as
K Xg 2
i sy , Ž C.1 .
Kg t0
ug u
s , Ž C.2 .
2Kg rt 0
hg 1
y1sy , Ž C.3 .
2K rt 0
kskg Ž elliptic modulus. Ž C.4 .
where
ip
t0s . Ž C.5 .
er
In Ref. w12x K g Ž u g ; j g , l g , m g . is given as
" sn Ž j g " u g ;k g .
K g Ž u g ; j g , l g , m g ." s , Ž C.6 .
sn Ž j g ;k g .
"
K g Ž u g ; j g , l g , m g ..
sn Ž 2 u g ;k g . l g Ž 1 y k g sn2 Ž u g ;k g . . . 1 . k g sn2 Ž u g ;k g .
s mg . Ž C.7 .
sn Ž j g ;k g . 1 y k g2 sn2 Ž j g ;k g . sn2 Ž u g ;k g .
We rewrite it as
½
K g Ž u g ; j g , l g , m g . s sn Ž j g ;k g . Ž 1 y k g2 sn2 Ž u g ;k g . sn2 Ž j g ;k g . .
y1
ug 2 jg 2 ug 2
=q 02
ž 2Kg
;y
t0 / ž q 02
2Kg
;y
t0 / ž
q0
Kg
;y
t0 /5
=K g Ž u g ; j g , l g , m g . , Ž C.8 .
"
K g Ž u g ; j g , l g , m g ."
"
u.j 2 u"j 2 2u 2
sc g ," Ž j . q0 ž rt 0
;y
t0 / žq1
rt 0
;y
t0 / ž
q0
rt 0
;y
t0 / , Ž C.9 .
q
K g Ž u g ; j g , l g , m g .y
q
2u 2 uya 2 uqa 2
s mg c g ,y Ž j ,a . q1 ž rt 0
;y
t0 / žq1
rt 0
;y
t0 / ž
q1
rt 0
;y
t0 / ,
Ž C.10 .
y
K g Ž u g ; j g , l g , m g .q
2u 2 uya 2 uqa 2
s m cy
g ,q Ž j ,a . q 1 ž rt 0
;y
t0 / ž
q0
rt 0
;y
t0 / ž q0
rt 0
;y
t0 / , Ž C.11 .
where
j jg
s , Ž C.12 .
rt 0 2Kg
a 2 a 2
lg s
q 02
ž rt 0
;y
t0 / ž q q12
rt 0
;y
t0 / Ž C.13 .
a 2 a 2
q 02
ž rt 0
;y
t0 / ž y q12
rt 0
;y
t0 /
and
4 j 2
c q
g ,q Ž j . s2
k X Ł qi
is1 ž rt 0
;y
t0 / , Ž C.14 .
k 2 2j 2
ž
q 0 0;y
t0 / ž
q1
rt 0
;y
t0 /
j 2 2
c q
g ,y Ž j ,a . s y2 ky1r2
q 02
ž rt 0
;y
/ ž
t0
q 02 0;y
t0 / , Ž C.15 .
a 2 a 2
q 02
ž rt 0
;y
t0 / ž y q12
rt 0
;y
t0 /
c y
g ,y Ž j . s yc gq,q Ž j . , c gy,q Ž j ,a . s yc gq,y Ž j ,a . . Ž C.16 .
On the other hand, K Ž u;c,l,u 0 . of Ž4.12. can be deformed into the same form,
"
K Ž u;c,l,u 0 ."
u.j X 2 u"j X 2 2u 2
"
s c" Ž c,l,u 0 . q0 ž rt 0
;y
t0 / ž
q1
rt 0
;y
t0 / ž q0
rt 0
;y
t0 / ,
Ž C.17 .
q
K Ž u;c,l,u 0 .y
2u 2 u y aX 2 u q aX 2
s cq
y Ž c,l,u 0 . q1 ž rt 0
;y
t0 / ž
q1
rt 0
;y
t0 / žq1
rt 0
;y
t0 / , Ž C.18 .
y
K Ž u;c,l,u 0 .q
2u 2 u y aX 2 u q aX 2
s cy
q Ž c,l,u 0 . q 1 ž rt 0
;y
t0 / ž
q0
rt 0
;y
t0 / žq0
rt 0
;y
t0 / , Ž C.19 .
where
t0 r
j X s j X Ž c,l,u 0 . s i ln R 1 Ž l,c,u 0 . , Ž C.20 .
2p
t0 r
aX s aX Ž l,c,u 0 . s ln R 2 Ž l,c,u 0 . , Ž C.21 .
ip
1 2
C 4 L4 y C 4 L2 q C 2 D 2 L4 y C 2 D 2 q L2 y L
R 1 Ž l,c,u 0 . s
ž C 4 D 2 L4 y C 4 D 2 L2 q C 2 L4 y C 2 q D 2 L2 y D 2 / , Ž C.22 .
1
R 2 Ž l,c,u 0 . s
'2 ½ C 2 q Cy2 q D 2 q Dy2 q C 2 L2 q Cy2 Ly2
yCy2 Dy2 Ly2 y4C 4 D 4 L4 q Ž yC 4 D 2 L4 y C 4 D 2 L2

ž
1 1
2 2 2 2
2 4 2
yC D L y C L y D L y D . 2 2 2 2
/ 5 , Ž C.23 .
ip ip ip
c l d
t0r t0r t0r
Cse , Lse , Dse . Ž C.24 .
The above C has nothing to do with the C of Ž4.6.. Furthermore,
ip X2
q d 2 q2 l 2 q2 c 2 q2 l c .
2 ey t 0 r 2 Ž3 j
cq Ž c,l,u 0 . s
q
t0 2 2j X 2 2j X 2
q 0 0;y
ž t0 / ž
q1
rt 0
;y
/ ž
t0
q0
rt 0
;y
t0 /
j Xqdql 2 j Xydql 2
½ ž
= yq 2
rt 0
;y
t0 / ž
q3
rt 0
;y
t0 /
j Xqc 1 lqcyj X 1
=q1
ž rt 0
X
;y
t0 / ž
q1
rt 0
X
;y
t0 /
j qdyl 2 j ydyl 2
qq2
ž rt 0
X
;y
t0/ ž q3
rt 0
X
;y
t0 /
cyj 1 lqcqj 1
=q1
ž rt 0
;y
t0 / ž
q1
rt 0
;y
t0 /5 , Ž C.25 .
q
cy Ž c,l,u 0 .
ip 3
l 2
q d 2 q2 l 2 q2 c 2 q2 l c .
2 Ž y . ey t 0 r 2 Ž 4 r
s
t0 1 2 yrr2y aX 2 yrr2q aX 2
q1 y
ž t0
;y
t0 / ž q1
rt 0
;y
t0 / ž q1
rt 0
;y
t0 /
yrr2q d q l 2 yrr2y d q l 2
= yq2½ ž rt 0
;y
t0 / ž q2
rt 0
;y
t0 /
yrr2q c 1 l q c q rr2 1
=q1
ž rt 0
;y
t0 / ž q1
rt 0
;y
t0 /
yrr2q d y l 2 yrr2y d y l 2
qq2
ž rt 0
;y
/ ž t0
q2
rt 0
;y
t0 /
c q rr2 1 l q c y rr2 1
y
=q1
ž q
rt 0
;y
t0 / ž
q1
y
rt 0
q
;y
t0 /5 , Ž C.26 .
cy Ž c,l,u 0 . s ycq Ž c,l,u 0 . , cq Ž c,l,u 0 . s ycy Ž c,l,u 0 . . Ž C.27 .

Comparing K g Ž u g ; j g , l g , m g . with K Ž u;c,l,u 0 ., we define

q X
X X
cy Ž c,l,u 0 . cq
g ,q Ž j Ž c,l,u 0 . .
m s m Ž l,c,u 0 . s X X . Ž C.28 .
cq q
q Ž c,l,u 0 . c g ,y Ž j Ž c,l,u 0 . ,a Ž c,l,u 0 . .
Then we have the desired result

q
cq Ž c,l,u 0 .
X K g Ž u; j X ,aX , mX . s K Ž u;c,l,u 0 . , Ž C.29 .
cq
g ,q Ž j Ž c,l,u 0 . .
where we abuse the notation for K g ,

K g Ž u; j ,a, m g . s K g Ž u g ; j g , l g , m g . . Ž C.30 .
References
w1x G.E. Andrews, R.J. Baxter, P.J. Forrester, J. Stat. Phys. 35 Ž1984. 193.
w2x R.J. Baxter, Exactly Solved Models in Statistical Mechanics ŽAcademic, London, 1982..
w3x R.J. Baxter, Ann. Phys. 76, 1.
w6x R.E. Behrend, P.A. Pearce, D.L. O’Brien, J. Stat. Phys. 84 Ž1996. 84.
w7x H. Fan, B. Hou, K. Shi, J. Phys. A 28 Ž1995. 4743.
w8x H. Fan, B. Hou, G. Li, K. Shi, Phys. Lett. A 250 Ž1998. 79.
w9x H. Fan, B. Hou, K. Shi, Z. Yang, Nucl. Phys. B 478 Ž1996. 723.
w10x O. Foda, K. Iohara, M. Jimbo, R. Kedem, T. Miwa, H. Yan, Lett. Math. Phys. 32 Ž1994. 259.
w11x O. Foda, K. Iohara, M. Jimbo, R. Kedem, T. Miwa, H. Yan, Prog. Theor. Phys. Suppl. 118 Ž1995. 1.
w12x T. Inami, H. Konno, J. Phys. A 27 Ž1994. L913.
w13x M. Jimbo, R. Kedem, T. Kojima, H. Konno, T. Miwa, Nucl. Phys. B 441 Ž1995. 437.
w14x M. Jimbo, R. Kedem, H. Konno, T. Miwa, R.A. Weston, Nucl. Phys. B 448 Ž1995. 429.
w15x M. Jimbo, T. Miwa, Algebraic analysis of solvable lattice models, CBMS Regional Conference Series in
Mathematics, Vol. 85 ŽAMS, Providence, RI, 1994..
w16x M. Jimbo, T. Miwa, A. Nakayashiki, J. Phys. A 26 Ž1993. 2199.
w17x M. Lashkevich, L. Pugai, Nucl. Phys. B 516 Ž1998. 623.
w18x M. Lashkevich, L. Pugai, JETP Lett. 68 Ž1998. 257.
w19x S. Lukyanov, Y. Pugai, Nucl. Phys. B 473 Ž1996. 631.
w20x T. Miwa, R.A. Weston, Nucl. Phys. B 448 Ž1995. 429.
w21x E.K. Sklyanin, J. Phys. A 21 Ž1988. 2375.
Thermodynamics of Fateev’s models in the presence of

external fields
Davide Controzzi, Alexei M. Tsvelik
Department of Physics, UniÕersity of Oxford, 1 Keble Road, Oxford, OX1 3NP, UK
Received 29 September 1999; received in revised form 21 December 1999; accepted 21 January 2000
Abstract
We study the Thermodynamic Bethe Ansatz equations for a one-parameter quantum field
theory recently introduced by V.A. Fateev. The presence of chemical potentials produces a kink
condensate that modifies the excitation spectrum. For some combinations of the chemical
potentials an additional gapless mode appears. Various energy scales emerge in the problem. An
effective field theory, describing the low energy excitations, is also introduced. q 2000 Elsevier
Science B.V. All rights reserved.
PACS: 11.10.-z; 71.10.-w; 74.20.De
1. Introduction
Recently V.A. Fateev introduced a series of two-dimensional integrable deformations
of affine Toda theories possessing the property of duality w1,2x. These models exist in
two incarnations – fermionic and bosonic – and the strong coupling limit of one
incarnation corresponds to the weak coupling limit of the other one. Apart from being
amusing examples of integrable field theories, these models may have some interesting
physical applications w3–5x.
In this paper we consider in some details one of the simplest models of that family.
The Lagrangian density of the fermionic version of the theory is given by
pg 2 2 2
L Ž f . s Ý i csgm Em cs q csgm cs q 12 Ž Em f 1 . q Ž Em f 2 .
ž /
ss1,2 2
y M0 c 1 c 1eyb f 1 y M0 c 2 c 2 e bf 2
M02
y w 2e b Žf 2y f 1 . q ey2 bf 1 q e 2 bf 2 x , Ž 1.
2b 2
where g s b 2rŽ4p q b 2 . and we can identify two types of fermions c 1,2 and two
coupled Toda chains described by the bosonic fields f 1,2 .
PII: S 0 5 5 0 - 3 2 1 3 Ž 0 0 . 0 0 0 4 2 - 0
610 D. Controzzi, A.M. TsÕelikr Nuclear Physics B 572 [FS] (2000) 609–625
In the bosonic representation the phonon modes, f 1,2 , disappear and the Lagrangian
density is
2 Em x s Em x s M02
L Ž b. s 12 Ý 2
y x 1 x 1 q x 2 x 2 q g 2r2 Ž x 1 x 1 . Ž x 2 x 2 . ,
ss1 1 q Ž gr2 . x s x s 2
Ž 2.
where the coupling constants of the two models are connected by the duality relation
g s 4prb . Ž 3.
As we have mentioned above, the weak coupling limit of the bosonic theory, g < 1
Ž1 y g < 1., correspond to the strong coupling limit of the fermionic one, b 4 1.
The bosonic form of the model is particularly interesting because the Euclidean
version of the theory can be interpreted as Ginzburg–Landau free energy of two coupled
superconductors. We shall discuss this application in greater details later in the text as
well as in a separate publication.
Models Ž1., Ž2. have UŽ1. = UŽ1. symmetry and the corresponding conserved charges
are given by
i Ž xs E 0 xs y xs E 0 xs .
Q s s d x csg 0 cs s y
H Hd x 2
. Ž 4.
2 1 q Ž gr2 . x s x s
Thus one can introduce two chemical potentials and modify the Hamiltonian,
H ™ H y h1 Q1 y h 2 Q 2 . Ž 5.
The UŽ1. = UŽ1. symmetry of the Hamiltonian reflects in the symmetry of the
particle multiplets. In absence of chemical potentials the spectrum of the model consists
of fundamental particles carrying quantum numbers of UŽ1. groups and their neutral
bound states. The two-body scattering matrix of fundamental particles is given by a
tensor product of two sine-Gordon S-matrices multiplied by a CDD-factor responsible
for cancellation of double poles,
sinh u y i sin Ž prl .
S 12 Ž u . s y SˆŽ l ;u . = SˆŽ l ;u . , Ž 6.
sinh u q i sin Ž prl .
where the SˆŽ l, u . is the soliton scattering matrix of the sine-Gordon model with
g2
ls2ygs1q Ž 7.
4p q g 2
and is given by
sinh Ž lu .
½
Sâab, b Ž l ;u . s e i dlŽ u . d a a d b b yd a b q Ž 1 y d a b .
sinh l Ž u y ip .
i sin Ž pl .
qd a b d b a Ž 1 y d a b .
sinh lŽ u y ip . 5 , Ž 8.
D. Controzzi, A.M. TsÕelikr Nuclear Physics B 572 [FS] (2000) 609–625 611
with
` sin Ž vu . sinh pv Ž ly1 y 1 . r2
dl Ž u . s H0 d v .
v cosh Ž pvr2 . sinh Ž pvr2 l .
To derive thermodynamic equations we start from the discrete Bethe Ansatz equa-
tions which emerge when the system is put in a box of size L with periodic boundary
conditions,
eyi M sinh u j L j Ž a1 ,... a n . s Ł S Ž u j y u k . j Ž a1 ,... a n . , Ž 9.
k/j
n
EsM Ý cosh u j , Ž 10 .
js1
where j Ž a1 , . . . , a n . is the wave function of n fundamental particles in the isotopic

space. The matrix on the right-hand side of Eq. Ž9. is related to the trace of the
monodromy matrix t ,
Ł S Ž u j y u k . s Tr t Ž l s u j ;u 1 , . . . , un . Ž 11 .
k/j
Up to the scalar factor given by the product of CDD-factors the latter matrix is equal to
a tensor product of monodromy matrices of the sine-Gordon models,
tˆ Ž l , u 1 , . . . , un . s tˆ Ž1. Ž l , u 1 , . . . , un . = tˆ Ž2. Ž l , u 1 , . . . , un . . Ž 12 .
The eigenvalues of the sine-Gordon monodromy matrix are known, hence it is
straightforward to write down Ž9. in the diagonal form,
mq
sinh l Ž u j y u a . q ipr2
eyi M sinh u j L s Ł S0 Ž u j y u k . Ł
k/j as1 sinh l Ž u j y u a . y ipr2
my
sinh l Ž u j y Õ b . q ipr2
=Ł , Ž 13 .
bs1 sinh l Ž u j y Õ b . y ipr2
n mq
sinh l Ž u j y u a . q ipr2 sinh l Ž u b y u a . q ip
Ł s Ł , Ž 14 .
js1 sinh l Ž u j y u a . y ipr2 bs1 sinh l Ž u b y u a . y ip
n my
sinh l Ž u j y Õa . q ipr2 sinh l Ž Õ b y Õa . q ip
Ł s Ł , Ž 15 .
js1 sinh l Ž u j y Õa . y ipr2 bs1 sinh l Ž Õ b y Õa . y ip
where
ny1 sinh u y i sin Ž p arl .
S0 Ž u . s e 2 i dlŽ u . Ł . Ž 16 .
as1 sinh u q i sin Ž p arl .
The numbers m " are related to the conserved charges Q1,2 . As we shall demonstrate
later,
m "s 12 n y Ž Q1 " Q 2 . . Ž 17 .
The simplicity of this formula cannot obscure its remarkable meaning: propagating
particles are superpositions of states localized on different edges of the stripe. This result
is valid for other Fateev’s models where the Toda array consists of more than two elastic
chains.
2. Thermodynamic Bethe ansatz (TBA) equations
In order to construct the Thermodynamic Bethe Ansatz ŽTBA. equations from

Ž13. – Ž15. we consider the case 1 y g s 1rn , with integer n G 2, where the complex
solutions of these equations Žthe strings. have the simplest classification. For most of the
results the fact that n is an integer is not important and one can generalize them for
arbitrary g ) 1r2 replacing n by Ž1 y g .y1 .
To save space and make our notations more transparent we shall sometimes denote
the kernels in the integral equations directly in terms of their Fourier transforms. For
example, the convolution of two functions g and f with the function g having Fourier
transform g Ž v . will be written as
q`
X X X
Hy` du g Ž u y u . f Ž u . s g ) f Ž u . s g Ž v . 4 ) f . Ž 18 .
Thermodynamic Bethe Ansatz ŽTBA. equations have been introduced in Refs. w1,4x. The
free energy of the system is expressed in terms of the excitation energies EnŽ u .,
du yE n Ž u .r T
FrL s yT Ý Mn H 2p cosh u ln w1 q e x. Ž 19 .
n
In the specific case Ž1., Ž2. we have only one bound state with energy E1 , and the
fundamental particle with energy E0 . The function E1 satisfies the following equation:
T ln w 1 q e E1Ž u .r T x y G 11 )T ln w 1 q eyE 1Ž u .r T x
s M1cosh u q G 10 )T ln w 1 q eyE 0 Ž u .r T x , Ž 20 .
and is directly coupled only to E0 , which is determined by the equation
T ln Ž 1 q e E 0 Ž u .r T . y Kq )T ln w 1 q eyE 0 Ž u .r T x
hqq hy
s Mcosh u y q G 10 )T ln w 1 q eyE1Ž u .r T x
2
y Ts)ln Ž 1 q e ey1 Ž u .r T . Ž 1 q e e 1Ž u .r T . . Ž 21 .
The two masses are related by
M1 s 2 Msin pr2 Ž 2 y g . Ž 22 .
and the Fourier transform of the kernel is
tanh pvr2 Ž n q 1 . sinh p Ž n y 1 . vr2 Ž n q 1 .
Kqs K sym q 2 a1 ) s s 1 q ,
cosh Ž pvr2 .
Ž 23 .
where K sym is related to the two-particle scattering phase,

1 d
K sym Ž u . s d Ž u . y log Ž S0 Ž u . . , Ž 24 .
2p i du
and its Fourier transform is
sinhpr2 Ž 1 y 1rl . v cosh Ž pvrl .
K sym Ž v . s . Ž 25 .
cosh Ž pvr2 . sinh Ž pvr2 l .
The equations for e n and eyn are
eyn s hy nr2 y s)T ln w 1 q e eyŽ ny 2.r T x , Ž 26 .
ye ny 1 r T eyn q 1 r T
eyn s dn , ny1 hy nr2 q s)T ln w 1 q e xw 1 q e x
q dny2, n s)T ln w 1 q eye yn r T x q dn ,1 s)T ln w 1 q eyE 0 r T x , Ž 27 .
e Ž n y 2.r T
en s hq nr2 y s)T ln w 1 q e x, Ž 28 .
e n s dn , ny1 hq nr2 q s)T ln w 1 q e e ny 1 r T xw 1 q e e nq 1 r T x
q dny2 , n s)T ln w 1 q eye n r T x q dn ,1 s)T ln w 1 q eyE 0 r T x . Ž 29 .
The Fourier transforms of the remaining kernels have the form
y1
s Ž v . s 2cosh pr2 lŽ 1 y g . v ,
4sinh Ž gpvr2 l . sinh p Ž 1 y g . vrl
G 11 Ž v . s
sinh Ž pvr2 l .
cosh Ž l y 1 . pvr2 l sinh w pvr2 l x
= ,
cosh Ž pvr2 .
2sinh Ž gpvr2 l . sinh p Ž 1 y g . vr2 l
G 10 Ž v . s ,
cosh Ž pvr2 .
where l is defined in Ž7. and can be also expressed as l s 1 q ny1 .
The subsequent analysis will show that the fields h " are linear combinations of the
chemical potentials h1,2 ,
h "s h1 " h 2 Ž 30 .
Žrecall Eq. Ž17...
3. Properties of the ground state
In this section we consider the zero temperature limit of the TBA equations. We
assume that the fields h " are strong enough to make E0 negative in some interval
wyB, B x. Then it is obvious from Eq. Ž20. that E1 ) 0,
E1 s M1cosh u y G 01 ) E0Žy. . Ž 31 .
From Eqs. Ž27., Ž29. we deduce that all e " n with n / n are also positive,
e " n s ya1 ) E0Žy. y A n , ny2 s) e "
Žy.
n q Ž n y 1. h " , Ž 32 .
where
sinh Ž n y max Ž n,m . . V sinh min Ž n,m . V
A n m Ž v . s 2coth V ,
sinh nV
sinh p Ž n y n . V pv Ž 1 y g .
an Ž V . s , Vs . Ž 33 .
sinh nV 2Ž 2 y g .
Substituting Eq. Ž32. into Eqs. Ž26., Ž28. we get the equations for e " n ,
sinh Ž n y 2 . V sinh V
e"Žq.
n q 1y½ sinh nV
)e" Žy.
n s h "q 5
sinh nV
) E0Žy. . ½ Ž 34 .5
At last, substituting Eq. Ž32. into Eq. Ž21., we get
sinh V
E0Žq. q K sym ) E0Žy. s Mcosh u y 12 Ž hqq hy . q
sinh nV
) enŽy. q ey
Žy.
½
n . 5
Ž 35 .
Eqs. Ž34., Ž35. determine the structure of the ground state. We cannot solve them in the
general case and therefore we shall consider only two special cases: Ža. hq< hy; Žb.
hqf hy.
In the first case eyn ) 0 and en is negative inside an interval wyQ,Q x, with Q 4 B
and Q ™ ` as hq™ 0. Since enŽq. is small it is convenient to invert the kernel in Eq.
Ž34. to get
sinh nV
½
2sinh V cosh Ž n y 1 . V
) enŽq. q enŽy. 5
1
s n hqr2 q ½ 2cosh Ž n y 1 . V 5 ) E0Žy. . Ž 36 .
For < u < 4 B it is useful to approximate the right-hand side of this equation by its
asymptotics,
1 2z
½ 2cosh Ž n y 1 . V
) E0Žy. ,5p e
a exp Ž yz < u < . , Ž 37 .
where
B X X
ae s HyBdu E0 Ž u X . ey zu Ž 38 .
and
nq1 2yg
zs s . Ž 39 .
ny1 g
Analogously one can rewrite Eq. Ž35. as
1
E0Žq. q K ) E0Žy. s Mcosh u y hyr2 y ½ 2cosh Ž n y 1 . V 5 ) enŽq. , Ž 40 .
where
sinh V cosh Ž 3n y 1 . V
KŽV . s . Ž 41 .
2sinh nV cosh Ž n y 1 . V cosh Ž n q 1 . V
Since enŽq. is very small, to first approximation this reduces to

E0Žq. q K ) E0Žy. s Mcosh u y hyr2 . Ž 42 .
For large B we can approximate Eqs. Ž36. and Ž42. by the Wiener–Hopf ŽWH. ones
w7x. The solutions for ´ 0Žy. Ž u . s E0Žy. Ž u q B . and ńŽq. Ž u . s enŽq. Ž u q B . are
hy 1 1
´ 0Žy. Ž v . s Žy. y , Ž 43 .
Ga Ž v . Ga Ž 0 . i v q 1
Žq.
ivq0
1 hq 1 eyz B 1
ńŽq. Ž v . s Žy .
y ae Žy .
, Ž 44 .
G Žq.
Ž v. G Ž 0 . yi v q 0 G Ž iz . z y i v
where ae is defined in Eq. Ž38., and GaŽq. Ž GaŽy. . and G Žq. Ž G Žy. . are analytic in the
upper Žlower. half plane and satisfy the conditions
GaŽq. Ž v . GaŽy. Ž v . s K Ž v . ,
sinh nu
G Žq. Ž v . G Žy. Ž v . s ; Ž 45 .
z is defined by Eq. Ž39. and B is determined by the following relation:
Me B hy
Žq.
s Žy.
. Ž 46 .
G Ž i. G Ž 0.
To study the low temperature thermodynamic in Section 6 it is useful to introduce also
the ground state densities for E0 and en , s and S , respectively, defined by the
following equations:
M
s Žq. q K ) s Žy. s cosh u , Ž 47 .
2p
sinh nV
S Žy. q ½
) S Žq. 5
1
s ½ 2cosh Ž n y 1 . V 5 ) s Žy. ' S 0 . Ž 48 .
Solutions of the WH equations for s˜ Žy. Ž u . s s Žy. Ž u q B . and S˜ Ž u . s S Ž u q B . are

hy
s˜ Žy. Ž v . s , Ž 49 .
4p G Ž v . G Žq. Ž i . Ž i v q 1 .
Žy.
1 hq eyz B
S̃ Ž v . s y ae . Ž 50 .
G Žq. Ž v . G Ž y . Ž 0 . Ž yi v q 0 . G Žy . Ž i z . Ž z y i v .
In the limit Žb. both e " n are positive and we have for E0 ,
E0Žq. q K sym ) E0Žy. s Mcosh u y h . Ž 51 .
Again using the WH method we find the following solution:
h
´ 0Žy. s Žy. , Ž 52 .
Gb Ž v . Gb Ž 0. Ž i v q 1. Ž i v q d .
Žq.
Fig. 1. Schematic diagram of the ground state energies in presence of chemical potential for hq < hy and
large B. The same situation for small B will be presented in Section 7.
with G Žq.
b
Ž v .G Žy.
b
Ž v . s K sym Ž v .. Substituting this solution in Eq. Ž34. we obtain the
following estimate for the gap in e " n :
Ž2yg .r2
Mn ; h Ž Mrh . . Ž 53 .
In the limit ny1 ™ 0Ž g ™ 1. Mn ; 'Mh .
A schematic diagram of these results is presented in Figs. 1 and 2 for the case Ža. and
Žb., respectively.
Fig. 2. Schematic diagram of the ground state energies in presence of chemical potential for hq f hy and
large B.
4. The limit of free bosons
To check the validity of TBA equations Ž20. – Ž29. and establish the connection
between the fields h " and the chemical potentials h1,2 we consider the free boson limit
of the TBAs. It is realized in the case g ™ 1 or ny1 ™ 0. In this case the kernels in Eqs.
Ž26. – Ž29. become proportional to the delta functions
s Ž u . ™ 12 d Ž u .
and Kq™ I,G11 s G 01 ™ 0, such that all TBA equations become algebraic. The mass of
the bound state in this limit is M1 s 2 M and the Free Energy is equal to
TM 2
FrL s y du cosh u ln w 1 q eyE1 r T x w 1 q eyE 0 r T x .
H ½ Ž 54 . 5
2p
The solution of TBA equations is expressed in terms of j s Mcosh urT,
y1
1 q eyE1 r T s Ž 1 y ey2 j . , Ž 55 .
2
sinh Ž h " n q a " . r2T
1 q ee " n r T s , Ž 56 .
sinh h "r2T
where
sinh a "r2T
s '1 q eyE 0 r T . Ž 57 .
sinh h "r2T
Substituting this into Eq. Ž21. we get
yE 0 r T 1r2
½ coshŽ h r2T . '1 q e q cosh Ž h r2T . q sinh Ž h r2T . e
q
yE 0 r T 2
q
2
q 5
= ½ cosh Ž h r2T . '1 q e
y
yE 0 r T
q cosh Ž h r2T . 2
y
2 yE 0 r T 1r2 j yE 0 r T
qsinh Ž h r2T . e
y 5 se , Ž 58 .
which has the following solution:
1 q eyE 0 r T
sinh2j
s 2
.
Ž cosh j y cosh Ž hqr2T . cosh Ž hyr2T . . y sinh Ž hqr2T . sinh Ž hyr2T .
Ž 59 .
Combining it with Ž55. in the Free Energy, we get
T
FrL s y du Mcosh u ln Ž F . ,
H Ž 60 .
2p
y2
F s Ž 1yey2 j .
sinh2j
=
Ž cosh j ycosh Ž hq r2T . cosh Ž hy r2T . . 2ysinh2 Ž hq r2T . sinh2 Ž hy r2T .
s w 1y eyj y Ž hqq hy . r2T xw 1y ey jq Ž hqq hy . r2T xw 1y ey jy Ž hqy h y . r2T x

y1
=w1y ey jq Ž hqy hy . r2T x 4 . Ž 61 .
The free boson limit of TBAs Ž20. – Ž29. does reproduce the free energy of two
non-interacting complex bosonic fields. This indicates that the TBA equations are
correct. Another important result is the confirmation of Eqs. Ž17., Ž30..
5. Semiclassical analysis of the limit of small g
Let us consider the Euclidean form of the Lagrangian Ž2.. Rescaling the fields,
x s ™ Žgr2. D s , it becomes
2 2 Em D s Em D s
L Ž b. s 2 Ý q M 2 D1 D1 q D2 D2 q 2 Ž D1 D1 .Ž D2 D2 .
½ 5 . Ž 62 .
g ss1 1 q Ds Ds
Then in the limit g ™ 0 the quantum fluctuations are suppressed and one can approxi-
mate the ground state energy as minimum of the functional
2 h s Ds Ds
E Ž D s ,h s . s 2
M 2 D1 D1 q D2 D2 q 2 Ž D1 D1 .Ž D2 D2 . y
½ 5 Ý .
g ss1,2 1 q Ds Ds
Ž 63 .
We will not address this problem, already studied by Fateev w1x, but we will use this
approximation to get more insight in the problem and give an intuitive picture of the low
energy physics. Let us perform the following transformation in the form Ž62.:
D s s sinh r s e i w s . Ž 64 .
Under this transformation the measure transform in Dr Dw sinhŽ2 w . and the Lagrangian
density in the new fields in presence of chemical potentials assume the form
2 2 2
L Ž b. s
g 2 ½Ý
ss1,2
Ž Em rs . q tanh2r s Ž Em ws . y h s tanh2r s E 0 ws
qM 2 sinh2r 1 q sinh2r 2 q 2sinh2r 1 sinh2r 2 5 . Ž 65 .

For h1 s h and h 2 s 0, which corresponds to the situation Žb. in Section 3, we get
2 2 2 2 2
L Ž b. s ½ Ž E r . q tanh r
m 1
2
1 Ž Em w 1 . y h E 0 w 1 q Ž Em r 2 . q tanh2r 2 Ž Em w 2 .
g2
qM 2 Ž sinh2r 1 q sinh2r 2 q 2sinh2r 1 sinh2r 2 . . 5 Ž 66 .

The term hŽ E 0 w 1 . can be absorbed by the shift
w 1 ™ w 1 y Ž hr2 . t , Ž 67 .
2
which generates also an additional term yh r4. Then the Lagrangian density becomes
2 2 2 2
L Ž b. s ½ŽEm r 1 . q Ž Em r 2 . q tanh2r 1 Ž Em w 1 .
g2
2
qtanh2r 2 Ž Em w 2 . q Veffb Ž r 1 , r 2 . , 5 Ž 68 .
where
Veffb Ž r 1 , r 2 . s V1 Ž r 1 . q M 2 sinh2r 2 q 2 M 2 sinh2r 1 sinh2r 2 Ž 69 .
and
V1 Ž r 1 . s M 2 sinh2r 1 y h2r4tanh 2r 1 . Ž 70 .
For hr2 ) M, Veffb Ž r 1 , r 2 . has a minimum at r 1 s r 0 and r 2 s 0 where exp2 r 0
s '8 Ž hrM .. In the vicinity of the minimum we have
Veff Ž r 1 s r 0 q x , r 2 s y . f hM Ž '2 y 2 q '8 x 2 . , Ž 71 .

which gives the masses going like 'hM , in accordance with Ž53..
Thus when the chemical potential exceeds the threshold the kinks condense and
gapless Goldstone mode appear. To see this explicitly we write r 1 s r 0 q x and keeping
only quadratic terms in x obtain the following expression for the Lagrangian:
2 2 2
L Ž b. ; m
2 2
½ Ž E x . q Ž Mhr2. x q tanh r Ž E w . 0 m 1
g2
2 2
q Ž Em r 2 . q tanh2r 2 Ž Em w 2 . q sinh2r 2 hrM . 5 Ž 72 .
Here we can identify a gapless mode w 1 with velocity equal to the bare one, a gapful
field x with the mass m2x s hMr2 and an effective integrable field theory described by
the Lagrangian density
2
1 2 2 2
L2 s
2 ž /½
g
Msinh2r 2 q Ž Em r 2 . q tanh2r 2 Ž Em w 2 . 5. Ž 73 .
In the case h1 ; h 2 corresponding to the situation Ža. of Section 3, the effective

potential acquires the form
Veffa Ž r 1 , r 2 . s V1 Ž r 1 . q V1 Ž r 2 . q 2 M 2 sinh2r 1 sinh2r 2 . Ž 74 .
Repeating the same analysis one finds two independent gapless modes with the same
velocity. Then the semiclassical approximation gives a correct qualitative description of
the low energy behavior of the system. As a byproduct we obtain another confirmation
of Ž30..
6. Low energy physics: massless modes and energy scales
Let us go back to the zero temperature limit described in Section 3. As we have

shown, in the case h1 f h 2 there are two soft modes: E0 and en . At < u < - B the mode en
closely follows E0 ; at < u < 4 B its behavior is determined by the asymptotics Ž37. and
this mode becomes independent. The temperature scale below which the two modes
decouple can be determined by the value of en at u ; B,
Tsep ; enŽy. Ž B . . Ž 75 .
For low temperatures, T - Tsep , the two modes give independent contribution to the
thermodynamics. To calculate the low temperature free energy,
T
FrL s y M du cosh u ln w 1 q eyE 0 Ž u .r T x ,
H Ž 76 .
2p
we use Eqs. Ž36., Ž42. which at finite temperature have the form
T ln Ž 1 q e E 0 r T . y K )T ln Ž 1 q eyE 0 r T .
hy 1
s Mcosh u y
2
y ½ 2cosh Ž n y 1 . V 5 )T ln Ž 1 q e en r T . , Ž 77 .
sinh nV n hq
½ 2sinh V cosh Ž 1 y n . V 5 )T ln Ž 1 q e en r T . y T ln Ž 1 q ey en r T . s
2
y ae ,
Ž 78 .
where ae is defined in Eq. Ž37.. To isolate the contributions that vanish at T s 0 we
rewrite Ž77. like
E0Žq. q K ) E0Žy. q Ž I y K . )T ln Ž 1 q ey< E 0 < r T . s r.h.s. of Eq. Ž 77 . , Ž 79 .
where I is the identity operator. Using Eq. Ž47. we can rewrite the free energy as
T
FrL s y M du cosh u Ž E0Žy. q T ln Ž 1 q ey< E 0 < r T . . ' f 0 q fn ,
H Ž 80 .
2p
where
f 0 s yT du s Ž u . ln Ž 1 q ey< E 0 < r T . f yp 2 T 2r3Vc ,

H Ž 81 .
1
fn s yT duH ½ 2cosh Ž n y 1 . V 5 s Žy. Ž u . T ln Ž 1 q e en r T . f yp 2 T 2r3Vs . Ž 82 .
The velocities of the two modes are determined by

E E0 Ž u . y1
Vc s Ý < < 2ps Ž s. Ž B . u ss B
, Ž 83 .
ss" Eu
Een Ž u . y1
Vs s Ý < < 2pS Ž s. Ž B . u ss B
. Ž 84 .
ss" Eu
For large B all the quantities appearing in these equations can be calculated using the
Wiener–Hopf solutions obtained in Section 3 via the following relations:
2
dE0Ž " .rdu u sB s d ´ 0Ž " .rdu u s0 s lim " Ž i v . e Ž " . Ž v . ,
v™`
s Ž " . Ž B . s s˜ Ž " . Ž 0 . s lim " i vs˜ Žq. Ž v . ,

v™`
and analogous relations for en and S . The results are

Vc s Vs s 1 . Ž 85 .
The fact that the two velocities are the same seems to be peculiar to the model in
consideration and not a general property of the family of Fateev’s models. In general Vs
depends on g and on the number of Toda chains.
Thus we have shown that the presence of chemical potentials introduces additional
energy scales in the problem with nontrivial crossovers. At temperatures T < h, eyn
and e n Ž n / n . modes decouple from the other modes. The same happens for E1 at
energies smaller then their gap and in some physical regions there is also an intermedi-
ate energy scale, as we will see in the next Section. As one lowers the temperature
below Tsep the gapless modes decouple from each other and the picture of two
non-interacting Luttinger liquids emerges.
7. Weak coupling limit: nonperturbative results
Now we wish to focus our attention on the limit g ™ 1 which corresponds to the
weak-coupling limit of the bosonic theory. Naively one can imagine to obtain perturba-
tive results. It turns out however that, in the presence of the chemical potential, this is
not the case, the reason being that the kinks condensate survives and affects the
excitation spectrum. A similar effect is present in the Klein–Gordon limit of the
sine-Gordon model w6x.
To be specific let us consider the situation Ža. of Section 3, all the energies e n
Ž n / n . have gaps of the order of hy and can be omitted at temperatures T < hy. To
obtain information on the excitation spectrum one has to study the fundamental Eqs.
Ž31. and Ž42.. The standard relativistic dispersion law: E1Ž u . s M1cosh u , can be
modified by the last term in Ž31. which emerge in presence of kinks.
In the limit g ™ 1, the kernels simplify K Ž v . s K 0 Ž v .Ž1 q O ŽŽ1 y g . 2 .., and
G 10 Ž v . s G 0 Ž v .Ž1 q O ŽŽ1 y g . 2 .., where
4p 2 cosh Ž pv Ž l q 2 . r2 l .
K0Ž v . s 2
v
bl cosh Ž pvr2 . sinh Ž pvrl .
p cosh Ž pv Ž l q 2 . r2 l .
s Ž1yg . v Ž 86 .
l cosh Ž pvr2 . sinh Ž pvrl .
and
pv sinh Ž pvr2 l .
G0 s Ž 1 y g . . Ž 87 .
l cosh Ž pvr2 .
Again we can study TBA equations with the kernel K 0 Ž v . and G 0 Ž v . in the limits of
large and small B. We focus the attention on this second case where the effects we want
to consider can be seen more clearly. To study Eq. Ž31. for small B we need the kernel
K 0 Ž u . only for small u . This is equivalent to approximating Eq. Ž86. for large v , as
2p
K0Ž v . , Ž1yg . < v < . Ž 88 .
l
Approximating also the r.h.s. of Eq. Ž42. for small u we can rewrite it as
˜ E0 s Mu 2r2 y H ,
K0 ) Ž 89 .
where H s hyr2 y M and ) ˜ denote the convolution of support ŽyB, B . and K 0 is the
Fourier transform of Ž88.. One can absorb the constant part of the kernel in E0
introducing the new function E˜0 s 2lp Ž1 y g . E0 . Eq. Ž42. then becomes
K˜ 0 )
˜ E˜0 s Mu 2r2 y H , Ž 90 .
˜ ˜ Ž . < <
K 0 is the singular integral operator corresponding to the kernel K 0 v s v and
defined
1 B 1
K˜ 0 ) E˜0 Ž u . s y P HyB Ž u y u . X 2
E˜0 Ž u X . du X
p
1 E B 1
s
p Eu
P HyB Ž u y u . E˜ Ž u . d u
X 0
X X
. Ž 91 .
Using this form of the kernel, Eq. Ž90. can be reduced to the canonical form
B 1
P H du X V ŽuX . sg Žu . , Ž 92 .
yB ŽuyuX .
where we have introduced V Ž u . s E E˜0 Ž u .rEu and g Ž u . s p Ž Mu 2r2 y H .. This equa-
tion has the following general solution w8x:
1 B X
gŽuX.
'
V Žu . s 2 B yu P2 2
du H , Ž 93 .
p yB (
B2 yu X2 Ž u X yu .
which gives
Ml 3r2
E0 Ž u . s y 2 Ž B2 yu 2 . , Ž 94 .
3p Ž 1 y g .
where B 2r4 shy r2y M ' D. Analogously the ground state density Ž47. turns out to be
M
Ml
s Žu . s 3
'B 2 y u 2 . Ž 95 .
2p Ž 1 y g .
From here we find the relationship between hy and the charge
hyr2 y M
Qs . Ž 96 .
2p Ž 1 y g .
The Fermi energy, EF s E0 Ž u ., as function of the charge is therefore
3r2
8M 1r2
2p
EF s Ž1yg . Q . Ž 97 .
3p 2 M
We can now use the above results to solve Eq. Ž31.. Rewriting the Fourier transform
of the kernel G 0 as
2 E Ž pr2 Ž 1 y g . . sinh u
G0 ; , Ž 98 .
p Eu sinh2u q Ž pr2 Ž 1 y g . . 2 cosh2u
it is possible to extract the asymptotic behavior of E1 ,
M1cosh u q Mr3cos Ž 2siny1 Ž urB . . , u<B ,
E1 s
½ M1cosh u q Ae yu
, u4B .
Ž 99 .
Fig. 3. Dispersion relation E1 Župper figure. in the presence of kinks for B <1 and g s 0.02, from the
numerical solution of Eq. Ž31.. The dotted line represents the solution in the absence of kink condensate. In the
lower figure we plot the gapless mode E0 .
The numerical solution is shown on Fig. 3. One can clearly see that even in the weak
coupling limit the dispersion law is drastically modified at small u .
This analysis is supposed to clarify the subject of different energy scales present in
the model Žcf. Fig. 4.. At very small doping, Q < Q1 ' M Ž g y 1.y1 r3, EF is smaller
then M and there are two scales besides h: M and EF itself. The mode E1 is
completely decoupled at low temperature. As you increase the doping you reach an
intermediate region, characterized by M Ž1 y g .y1 r3 < Q < Q2 ' M Ž1 y g .y1 , where
EF , M but still much smaller then E1. At very large doping we find a very interesting
Fig. 4. A Schematic diagram describing different energy scales as a function of the charge. We can recognize
three distinct regions as described in Section 7.
region. As we have seen in Section 3 at large Q Žor equivalently at large B . EF , Q,

'
while E1Ž0. , Q . Then there is a regime in which EF is bigger then the gap of the
E1-mode, and there will be temperatures, T < EF , for which the E1-mode becomes soft.
8. Discussion
We have considered in some detail the thermodynamics of one of the integrable field
theories recently introduced by Fateev. Most of the results are quite general and remain
valid also for other models of the family. These models are interesting for various
reasons. First of all they have peculiar mathematical properties mostly related to a dual
representation of the theory. In addition they probably can find application to various
physical systems. Considering the model as a Ž1 q 1.-dimensional Quantum Field
Theory one can interpret it as two one-dimensional conductors coupled via phonon
interaction. The problem of quasi one-dimensional systems with electron–phonon
interaction is a very important one and cannot be approached with perturbative methods,
it is then very important to have exact results for some specific model. This interpreta-
tion of the model and the possible application to physical systems have been discussed
in Ref. w4x, for physical regimes different from the one considered here Ži.e. for
g - 1r2.. On the other side, the Euclidean version of the model, in the bosonic
representation, describes an effective Landau–Ginzburg theory of two coupled supercon-
ductors. In the specific model considered in this work, the coupling between the
superconductors is quite simple. Nevertheless for other models of the Fateev’s family,
where coupling between the fermionic modes is achieved via a higher number of Toda
chains, the resulting coupling between the two superconductors is via elastic modes and
then can be interpreted as two layered superconductors separated by an insulating
stratum. In relation to the high-Tc Žcuprate. superconductors an interesting problem is to
study the dependence of the superconducting properties on the number of insulating
layers between the superconducting planes. We will address this problem in a separate
publication.
Let us give a brief summary of our results. As shown in Section 7 the results are
nonperturbative also in the weak coupling regime and could only be obtained through
exact or nonperturbative methods. This is related to the fact that the chemical potentials
generate a kink condensate that survives in the weak coupling limit. As we have shown,
for some combinations of the chemical potentials two gapless modes emerge and various
energy scales appears, which characterizes crossovers between different regimes. Proba-
bly the most interesting region of the phase diagram is obtained at large doping where
the E1-mode become soft at temperatures for which the E0-one is still frozen. It deserves
further investigation. Another important result is the one contained in Eq. Ž17.. This
equation tells us that charged particles in Fateev’s model are linear combinations of
particles located on the edges, which is important for qualitative understanding of the
physics of the model.
References
w1x V.A. Fateev, Nucl. Phys. B 479 Ž1996. 594.
w2x V.A. Fateev, Nucl. Phys. B 473 Ž1996. 509.
w3x C. Pepin, A.M. Tsvelik, Phys. Rev. Lett. 82 Ž1999. 3859.

w4x C. Pepin, A.M. Tsvelik, cond-mat 9983180.
w5x D. Controzzi, A.M. Tsvelik, unpublished.
w6x J.-S. Caux, A.M. Tsvelik, Nucl. Phys. B 474 Ž1996. 715.
w7x V.A. Fateev, E. Onofri, Al.A. Zamolodchikov, Nucl. Phys. B 406 Ž1993. 521.
w8x S.G. Mikhlin, Integral Equations ŽPergamon press, Oxford, 1964..
Universality and multifractal behaviour of spin–spin

correlation functions in disordered Potts models
Christophe Chatelain a , Bertrand Berche a
a 1
´
Laboratoire de Physique des Materiaux, UniÕersite´ Henri Poincare´ Nancy 1, B.P. 239, F-54506 VandœuÕre
les Nancy Cedex, France
Received 16 November 1999; received in revised form 6 January 2000; accepted 21 January 2000
Abstract
We report a transfer matrix study of the random bond q-state Potts model in the vicinity of the
Ising model q s 2. We draw attention to a precise determination of magnetic scaling dimensions
in order to compare with perturbative results. Universality is checked by the computation of the
spin–spin correlation function decay exponent obtained with different types of probability
distributions of the coupling strengths. Our numerical data, compared to perturbative results for
the second moment of the correlation function, obtained with both replica symmetry and replica
symmetry breaking schemes, are conclusively in favour of the replica symmetric calculations. The
multifractal behaviour of higher moments as well as that of typical correlation functions are also
investigated and a comparison is made with the perturbative expansions. Finally, the shape of the
correlation function probability distribution is analyzed. q 2000 Elsevier Science B.V. All rights
reserved.
PACS: 05.20.-y; 05.50.qq; 64.60.Fr

Keywords: Disordered systems; Universality; Critical exponents; Multiscaling
1. Introduction
Random systems represent the paradigm of spatially inhomogeneous systems where

scale invariance is only preserved on average, but not for specific disorder realizations
w1x. In such systems, not a single exponent but instead an infinite hierarchy of
independent exponents are expected to describe the scaling behaviour of local quantities
like order parameter density profiles or correlation functions. This property, linked to the
non self-averaging behaviour of the corresponding physical quantities w2–5x is usually
referred to as multifractality. The keystone concept which enters the description of
1
Unite´ Mixte de Recherche CNRS No. 7556.
PII: S 0 5 5 0 - 3 2 1 3 Ž 0 0 . 0 0 0 5 0 - X
C. Chatelain, B. Berche r Nuclear Physics B 572 [FS] (2000) 626–650 627
multiscaling properties is that of scaling dimensions associated to the moments of the

local physical property, or equivalently the universal function H Ž a . corresponding to
the Legendre transform of the set of independent scaling indexes Žsee e.g. Refs. w6,7x..
Ten years ago already, in a series of illuminating papers, Ludwig w8,9x and Ludwig
and Cardy w10x reported an extensive analytic study of 2D random bond Potts ferromag-
nets in the regime where bond randomness is slightly relevant, q close to 2, q being the
number of states per spin. Their studies included perturbative calculations of the
conformal anomaly, of the thermal scaling dimension and of the multifractal behaviour
of spin–spin correlation functions.
The essential of the numerical studies dealing with scaling dimensions of average
quantities in random bond Potts models at small values of q were performed by Monte
Carlo ŽMC. simulations combined to standard Finite Size Scaling ŽFSS. w11–14x or
transfer matrix ŽTM. calculations associated to conformal methods w15–19x at the
random fixed point ŽFP. of self-dual disordered models. In particular, an excellent
quantitative agreement for the magnetic scaling dimension in the three-state Potts model
was reported in Refs. w15,18x. In which concerns the multiscaling properties of spin–spin
correlation functions, although some of Ludwig’s predictions have partially been
verified both in cylinder geometry w15,20x and in square geometry w21x, the agreement
with analytical expansions was less conclusive and in particular the shape of the
probability distribution has not been reproduced.
Monte Carlo simulations are not convenient to study numerically the vicinity of the
Ising model, q s 2, where perturbation expansions are supposed to apply. The number
of states per spin, q, is indeed restricted to integer values in such simulations. In this
paper, we therefore use a rather different approach already used by different authors
w15,18,19,22x. This technique is based on the Fortuin–Kasteleyn graph representation
w23x which enables TM calculations w24x where q enters as a parameter that can take non
integer values. We also benefit from previous studies with a bimodal probability
distribution of spin–spin interactions where the disorder amplitude was found to have a
deep influence on the measured critical properties in numerical studies w14x. It should
thus be chosen carefully in order to avoid crossover perturbations due to the unstable
pure model fixed point, r s 1, and the percolation fixed point, r ™ `, r being the ratio
between strong and weak couplings.
In this paper, we are mainly interested in the multiscaling behaviour of the spin–spin
correlation functions. The first section reminds the reader of the essential relevant
theoretical results which have been obtained by several groups using perturbative
techniques around the pure models conformal field theories. Section 3 explains the
methodology and Section 4 gives the numerical results:
1. Universality in quenched disordered ferromagnetic Potts models is checked using

different types of probability distributions of nearest-neighbour couplings.
2. The numerical study of the decay exponent of the second moment of the spin–spin
correlation function is then compared to perturbative results in order to test Replica
Symmetry and Replica Symmetry Breaking scenarios.
3. Finally, we study other moments and deduce the shape of the universal functions
H Ž a . for different q values. The probability distribution of spin–spin correlation
functions is also analyzed.
628 C. Chatelain, B. Berche r Nuclear Physics B 572 [FS] (2000) 626–650
The values of q are chosen in the range 2 to 4, where the pure model exhibits a second
order phase transition, with a special attention paid on the neighbourhood of q s 3.
2. Summary of the perturbation results
2.1. The 2D random Ising model
According to the celebrated Harris criterion w25x, quenched randomness is a marginal

perturbation in the 2D Ising model. This situation has focused a considerable interest on
the critical properties of the random bond Ising model ŽRBIM., and, after partially
conflicting results, disorder was eventually found to be marginally irrelevant, leading to
an unchanged universal behaviour apart from logarithmic corrections for the ensemble
average of some physical quantities w26–31x. These results were then carefully checked
through intensive MC simulations w32–34x and series expansions w35–37x. We mention
that site dilution is still subject to controversial interpretations Žsee e.g. Ref. w37x. in spite
of a conclusive recent work leading to the same conclusions than bond randomness w38x.
The free energy density in a strip of width L with periodic boundary conditions was
obtained by Ludwig and Cardy w10x:
p 1 y3
f Ž L. , f0 y Ž 2 y 128p 3D 3 Ž 1 q 8pD ln L . ., Ž 1.
6 L2
where D, the variance of the Gaussian probability distribution of exchange interactions,
is the strength of disorder related to the ratio r. In this expression, an overbar denotes
the disorder average. The central charge c, defined by the leading size dependence of the
free energy density in the cylinder geometry, p cr6 L2 , thus exhibits logarithmic
corrections which make its exact value 12 difficult to extract numerically w39x.
Using a perturbation expansion, Ludwig later obtained the behaviour of the moments
of the spin–spin correlation function w9x:
² s Ž 0 . s Ž r . : p , ryp r4 Ž D ln r . p Ž py1.r8 , Ž 2.
when D, the strength of disorder, is strong enough. Brackets denote the thermal average.
We can also introduce a reduced correlation function whose leading power-law be-
haviour is
1rp
Ž² s Ž 0. s Ž r . : p . ; ry2 x s . Ž 3.
Ludwig’s results imply that logarithmic corrections are absent in the case of the average
correlation function Ž p s 1., ² s Ž 0 . s Ž r . : ; ry1 r4 , while typical correlation func-
tions Ž p s 0. exhibit such corrections, exp ln² s Ž 0 . s Ž r . : ; ry1 r4 Ž D ln r .y1r8 , ob-
served numerically w40x. Furthermore, a unique scaling dimension xs s 18 describes the
leading power-law decay of all the moments of the spin–spin correlation function,
² s Ž 0 . s Ž r . : p ; ry2 p x s , and no multiscaling behaviour is expected apart from the
logarithmic correction term.
The surface correlation function has also been studied recently and was found to be
self-averaging w41x.
2.2. The random bond Potts model
In the case of the Potts model with q ) 2, disorder is a relevant perturbation which
modifies the universal critical behaviour and leads to new fixed point critical properties.
Studying the effect of a slightly relevant perturbation on the finite-size scaling behaviour
of the free energy density in a strip geometry of width L at the new fixed point, Ludwig
and Cardy w10x, obtained perturbatively the central charge cX Ž q . of the 2D q-state Potts
model with weak quenched bond randomness Žhere and in the following, primes denote
the central charge and the critical exponents at the disordered fixed point, while
unprimed symbols refer to the pure fixed point quantities.. Using an expansion in q y 2
around the Ising model, the random anomaly was deduced from the random free energy
of the strip,
p cX Ž q .
)
f Ž L, g . s A y q O Ž Ly3 . , Ž 4.
6 L2
given in the replica formalism by the quenched free energy E f Ž n.rE n < n ™ 0 :
cX Ž q . s 12 Ž 1 q 74 yH y 169 yH2 y 645 yH3 q O Ž yH4 . . , Ž 5.

where yH is the renormalization group ŽRG. eigenvalue associated to the bond disorder
w25x, yH s arn s 2 y 2 x´ Ž q ., x´ Ž q . being the scaling dimension of the energy density
in the pure model 2 . This latter dimension is obtained for arbitrary q F 4 by the den Nijs
conjecture w42,43x, rigorously proved by Dotsenko and Fateev w44x:
1qm 2
x´ Ž q . s , 0Fms cosy1 Ž 12 q . F 1,
' Ž 6.
2ym p
and to lowest order, the RG eigenvalue is proportional to q y 2: yH s Ž4r3p .Ž q y 2. q
OwŽ q y 2. 2 x. The deviation of the conformal anomaly from its pure fixed point value,
3 m2
cŽ q. s1 y , Ž 7.
2ym
is difficult to measure, since it is only of third order in yH and it requires a very good
accuracy to distinguish between pure and random values.
The thermal exponent was similarly obtained to two-loop order by Ludwig w8x:
x´X Ž q . s x´ Ž q . q 12 yH q 18 yH2 q O Ž yH3 . s 1 q 18 yH2 q O Ž yH3 . , Ž 8.

and the three-loop correction was reported by Jug and Shalaev w45x.
2
In this paper, we use the notations of Refs. w8–10x: The expansion parameter is yH . It is related to the
parameter e , linked to the deviation from the pure Ising model central charge, used by Dotsenko and
co-workers in Refs. w46,47,49,50x by yH s 3 e .
The correction to the magnetic exponent requires a three-loop calculation. It was

obtained by Dotsenko et al. w46x
2 1
G2 y ž / ž / G2
3 6
xsX Ž q . s xs Ž q . q 321 yH3 q O Ž yH4 . , Ž 9.
2
1 2
1
G y ž / ž /3
G y
6
and checked by Picco using MC simulations 12x. In contradistinction with the thermal
w
exponent, the deviation from the pure fixed point value is quite small close to q s 2.
Searching for multiscaling properties, Ludwig obtained, up to linear order, the scaling
3 w x
dimension of the pth moment 9 of the reduced spin–spin correlation function,
p 1r p X
y2 x s p Ž q .
Ž s Ž 0. s Ž r . . ; r
² : and Lewis performed recently the computation up to
the second order w47,48x
xsX p Ž q . s xs Ž q . y 161 Ž p y 1 . yH y 321 Ž p y 1 . A q B Ž p y 2 . yH2 q O Ž yH3 . ,
Ž 10 .
where A s 11 1 '
12 y 4ln2 and B s 24 33 y 29 3 pr3 . Here, the exponent corresponding
Ž .
to the aÕerage critical correlation function at the random fixed point is denoted by
xsX 1 Ž q . ' xsX Ž q ., while the typical behaviour corresponds to p s 0. This result, obtained
in the Replica Symmetry ŽRS. scenario, contains the special case of the second moment
performed by Dotsenko et al. w49x in order to compare between Replica Symmetry:
xsX 2 Ž q . s xs Ž q . y 161 yH q 321 Ž 4ln2 y 12
11
. yH2 q O Ž yH3 . , Ž 11 .
and Replica Symmetry Breaking ŽRSB .4 w50x:
xsXX 2 Ž q . s xs Ž q . y yH q Ž 4ln2 y 125 . yH2 q O Ž yH3 . .
1
16
1
32 Ž 12 .
3. Model and methodology
In this paper, we consider Potts-spin variables, sj g 1,2, . . . ,q on the sites of a square

lattice with independent quenched random nearest-neighbour ferromagnetic interactions
K i j . These exchange couplings are taken from a probability distribution P Ž K i j ., and in
most of our applications, they can take two values, K 1 s K, K 2 s rK, r ) 1, with equal
probabilities:
P Ž K i j . s 12 d Ž K i j y K . q 12 d Ž K i j y rK . , Ž 13 .
where r measures the strength of disorder. Our methodology will be discussed in this
section with the particular case of probability distribution Ž13. and the generalization to
other distributions will be presented in the next section.
3
In the literature, many different notations have been used for the scaling dimensions of the moments of the
X X
correlation functions. Our notation corresponds to that of Lewis w48x: xs p l Ds p. The correspondence with
X X X
w x w x
other works is the following: Ludwig 9 : xs p l X N r N, Dotsenko et al. 49 : xs 2 l Ds 2 r2 and Olson and
X
Young w21x: xs p lhn r2.
4
In Ref. w50x, the thermal and magnetic exponents have been computed with both RS and RSB scenarios.
While Eq. Ž9. for the average behaviour is unchanged up to the third order in the RSB scheme, the thermal
XX
exponent in Eq. Ž8. becomes x´ Ž q . s1q O Ž yH3 ..
The Hamiltonian of the model is thus written

yb H s Ý K i j ds , s .
i j
Ž 14 .
Ži , j.
We only consider self-dual models for which the critical point is exactly known. In
the case of the bimodal distribution Ž13., the self-duality point
exp Ž K c Ž r . . y 1 exp Ž rK c Ž r . . y 1 s q, Ž 15 .
corresponds to the critical point of the model if only one phase transition takes place in
the system as rigorously shown in Ref. w51x.
The degree of dilution in the system can be varied by changing the ratio of the strong
and weak couplings, r. At r s 1, one recovers the perfect q-state Potts model, whereas
for r ™ ` we are in the percolation limit, where Tc s 0. The intermediate regime of
dilution 1 - r - ` is expected to be controlled by the random fixed point located at
some r s r w Ž q . w19x. This optimal disorder amplitude, r w Ž q ., can be obtained numeri-
cally from the maximum condition of the effective central charge of the disordered
system. Dotsenko and co-workers for example considered n q-state Potts models
coupled via energy–energy interactions and obtained perturbatively the central charge
deviation from the decoupling limit Žwhere c is given by the sum of the central charges
of the decoupled models. w52x: D c s y 18 nŽ Žnnyy21. . yH3 q O Ž yH4 .. For n ) 1, D c satisfies the
2
Zamolodchikov’s c-theorem according to which there exists a c-function decreasing

along RG flows and giving the central charge at the fixed point w53x. In the case of
random systems Ž n ™ 0 in the replica approach., the central charge increases and can be
expected to reach a maximum value at an optimal disorder amplitude where the random
FP exponents may be extracted from numerical data. This property, linked to non-unitar-
ity in the presence of disorder, is indeed observed in simulations w15,19,22x.
In the following we used a TM technique, based on the Blote ¨ and Nightingale
connectivity transfer matrix w24x, which enables to compute the physical quantities in
long cylinders. Since transfer operators in the time direction do not commute in
disordered systems, the free energy density is defined by the leading Lyapunov
exponent. For an infinitely long strip of width L with periodic boundary conditions, the
leading Lyapunov exponent is given by the Furstenberg method w54x:
1 m
L0 Ž L . s lim
m™` m
ln ž Ł Tk
ks1
/ < Õ0 : , Ž 16 .
where Tk is the transfer matrix and < Õ 0 : is a unit initial vector. The quenched free
energy density is thus given by
f Ž L . s yLy1L 0 Ž L . . Ž 17 .
The shape of the central charge as a function of the disorder amplitude is shown in
Fig. 1 for several values of q. Each realization of disorder is obtained via 10 6 iterations
of the transfer matrix and the free energy density was averaged over 96 such realiza-
tions.
The central charge follows from a polynomial fit,
pc
f Ž L . s f 0 y 2 q A 4 Ly4 q A 6 Ly6 q A 8 Ly8 , Ž 18 .
6L
Fig. 1. Behaviour of the central charge as a function of the disorder amplitude for different values of the
number of states q Žbinary distribution of Eq. Ž13... The solid lines are parabolic fits. The maximum
corresponds to the optimal value of disorder amplitude.
where the remaining coefficients A i have been included in order to simulate finite size
effects, although the exact dependence of corrections to scaling is not known. The
calculations were performed on strips of widths L s 2 to 8. The central charge is still
sensible to the number of disorder configurations entering the average and to the degree
of the polynomial fit. Nevertheless, it is expected that the position of the maximum
presents small enough deviations for the critical exponents to reflect the disordered fixed
point regime in the neighbourhood of this maximum. We have checked different types
of fits, with less parameters and also including logarithmic terms in the vicinity of q s 2
where Eq. Ž1. is supposed to be valid, but we were not able to improve the results, since,
as it has already been observed w15x, the values of cX Ž q . are systematically below the
perturbative result of Ludwig and Cardy w10x, and even below the pure model central
charge at small values of q. In all the cases, the error can be estimated by the
fluctuations of the results with different fitting procedures and it is of order 10y3 to
10y2 . For example we obtain cX Ž2. s 0.496 with Eq. Ž18., 0.492 with A 6 s A 8 s 0,
0.497 with A 8 s 0 but a Žln L.y3 term added or 0.495 with A 6 s A 8 s 0 and the
log-term present. The error bars given in Table 1 correspond to the fluctuations of the
results obtained with the different fits. The maximal values and optimal disorder
amplitude r ) Ž q . are also given in Table 1.
For a specific disorder realization, the spin–spin correlation function along the long
direction, u, of the strip
q² ds j s jq u : y 1
² s Ž j . s Ž j q u. : s , Ž 19 .
qy1
where ² . . . : denotes the thermal average, is given by a product of non-commuting

transfer matrices. They were computed on strips of widths L s 2 to 8 and then averaged
over 80 000 disorder realizations.
Table 1
Optimal disorder amplitude and corresponding values of the central charges of the disordered Potts model for
different values of q for the binary probability distribution in Eq. Ž13.. The last columns give the numerical
values of the pure model central charge cŽ q . ŽEq. Ž7.. and of the expansion parameter used in the perturbation
results, and shows that the perturbation expansion obviously breaks down when q increases.
X
c Ž q.
q r ) Ž q. TM result Eq. Ž5. cŽ q . yH
1.5 2.25 0.283Ž1. 0.288
2. 2.49 0.496Ž6. 0.5 0.500 0.
2.25 3.18 0.584Ž8. 0.5876 0.588 0.1036
2.5 3.96 0.662Ž8. 0.6661 0.666 0.2036
2.75 4.70 0.732Ž8. 0.7371 0.736 0.3017
3. 5.36 0.797Ž8. 0.8020 0.800 0.4000
3.25 5.94 0.857Ž9. 0.8617 0.858 0.5013
3.5 6.45 0.912Ž11. 0.9174 0.910 0.6101
4. 7.30 1.011Ž12. 1.0312 1.000 1.
We will now assume that conformal covariance can be applied to the order parameter
correlation function and its moments. In the infinite complex plane z s x q i y the
correlation function and its moments exhibit the usual algebraic decay at the critical
point
1rp X
Ž ² s Ž z1 . s Ž z 2 . : p . s const = ry2 x s p Ž q . , Ž 20 .
where r s < z 1 y z 2 <. Multiscaling arises when the exponents xsX p Ž q . are all different,
depending upon the value of the moment order p, and their p-dependence is a convex
Fig. 2. Moments of the spin–spin correlation function ps1,2 and 3 from upper to lower curves Ž q s 3, Ls 7,
r s 5.36.. The solid lines correspond to the average over 96 000 different disorder realizations and the symbols
are deduced from the cumulant expansion up to the fifth order. The fluctuations become extremely large above
the fifth order. Both solid lines and symbols give the same order for the corresponding scaling dimensions
Žrelated to the slopes of these curves., but the direct average leads to more precise results.
function w9x. Under the logarithmic transformation w s Ž Lr2p .ln z s u q i Õ which

maps the infinite plane geometry inside an infinitely long strip of width L, one gets the
exponential decay along the strip
1rp 2p
Ž² s Ž j . s Ž j q u . : p . s const = exp y xsX p Ž q . u . Ž 21 .
L
The scaling dimensions xsX p Ž q . at different strip sizes can thus be deduced from an
exponential fit, and a quadratic extrapolation at L ™ ` is performed to get the
corresponding value in the thermodynamic limit. The calculation of errors follows the
lines explained in Ref. w19x.
This method was used in Ref. w19x for the average correlation function, i.e. for p s 1,
and in the large q regime. In this work, we calculate the higher moments, as well as the
typical behaviour, governed by the derivative of the exponent xsX p with respect to p,
evaluated in the limit p ™ 0. In previous works, it was shown that the results from
exponent extrapolations are sensitive to the value of the disorder strength chosen for the
simulation, since the finite size corrections are very strong unless the calculations are
performed close to the random fixed point w14x as obtained from the maximum condition
of the central charge of the model w19,52x. Our simulations were performed at these
fixed point values of the disorder strength, but for comparison we have also considered
systems with somewhat different values.
Since the correlation functions are not self-averaging, the disorder average must be
performed carefully. We follow the same procedure as in Ref. w19x where we compared
the ensemble average ² s Ž 0 . s Ž u . : to a cumulant expansion in terms of the moments
of ln² s Ž0. s Ž u.:, which are self-averaging w15x:
² s Ž 0 . s Ž u . : p s exp p ln² s Ž 0 . s Ž u . :
2
q 12 p 2 ln² s Ž 0 . s Ž u . :2 y ln² s Ž 0 . s Ž u . :
ž / q... . Ž 22 .
Although the average should in principle be done using the cumulant expansion, we
observe that the direct average, which is compatible with the cumulant expansion, is
more stable than this latter expansion. This is particularly true at high moment orders
and is probably due to the large number of disorder realizations used in the calculations.
This is illustrated for several moments at q s 3 in Fig. 2 where the solid lines represent
the direct average over 96 000 different samples, while the open symbols correspond to
the cumulant expansion up the the fifth order.
From the exponential decay of the average correlation function, the exponent xsX Ž q .
is deduced and presented as a function of q for the case of a binary disorder in Fig. 3.
Although these results are not new, since the same type of curve was reported by Cardy
and Jacobsen in Ref. w18x, the agreement with our results confirms the reliability of the
averaging procedure. The results are compared to the third order expansion of Dotsenko
et al. w46x in Eq. Ž9.. The data themselves are given in Table 2. The agreement is
extremely good especially in the region where the expansion is supposed to be valid
when q is not too far from the Ising model value q s 2.
Even close to the marginally irrelevant case of the Ising model where logarithmic
corrections are known to be present for some quantities, we note that the numerical data
are quite satisfactorily in agreement with the perturbative results. This is due to the
Fig. 3. Scaling dimension of the order parameter Žbinary disorder. compared to the third order expansion of
Dotsenko and co-workers w46x. The scaling dimension corresponding to the pure model is shown for
comparison.
absence of logarithmic corrections for the aÕerage correlation function at q s 2, and we

will see that this observation is no longer true in the following study of other moments.
4. Tests of universality
4.1. UniÕersality of the aÕerage behaÕiour
The question of universality in random systems is not yet solved, especially when
frustration occurs, like in random fields or spin glasses w55x. We address in this section
Table 2
Comparison of the numerical results for the magnetic scaling dimension Žbimodal probability distribution.
X
xs Ž q . with the third order expansion of Dotsenko and co-workers w46x. The error bars systematically contain
the analytical value.
X
q xs
Expansion Ž9. TM result
2 0.12500 0.1252Ž3.
2.25 0.12800 0.1282Ž5.
2.5 0.13051 0.1307Ž7.
2.75 0.13269 0.1328Ž9.
3. 0.13465 0.1347Ž11.
3.25 0.13653 0.1364Ž13.
3.5 0.13845 0.1379Ž14.
the question of the influence of the particular shape of the probability distribution of
exchange couplings on the universality class. There are still subsisting doubts concern-
ing universality since incompatible estimations Žtaking error bars into account. of the
critical exponent xsX were obtained with different probability distributions. For example
at q s 8, the choice of a bimodal probability distribution led to 0.151Ž4. w14x ŽFSS. or
0.1505Ž3. w19x Žconformal invariance., while a continuous distribution gave 0.161Ž3. w21x
ŽFSS..
In this section, we show that the discrepancy is simply due to crossover effects but
does not imply absence of universality. Following the methodology of the previous
sections, the disordered fixed point regime is located at the maximum of the central
charge and the scaling dimension xsX of the average spin–spin correlation functions is
estimated. The results of Fig. 3 are considered as a reference and the same method is
applied to ternary, quaternary and continuous distributions and is shown to lead to
identical critical exponents, within error bars, to those of a binary distribution.
4.1.1. Ternary distribution
The ternary probability distribution is defined by
P Ž K i j . s 13 d Ž K i j y K 0 . q d Ž K i j y K . q d Ž K i j y rK . , Ž 23 .
with the self-duality condition
2
exp Ž K c Ž r . . y 1 exp Ž rK c Ž r . . y 1 s exp Ž K 0 . y 1 s q, Ž 24 .
'
where K 0 s lnŽ1 q q . is the critical coupling of the pure system. The homogeneous
system corresponds to the value r s 1. We tried different kinds of interpolation
procedures for the free energy, as in Eq. Ž18., but always found a monotonic variation of
the corresponding central charge with respect to the disorder amplitude r. The absence
of maximum might be the sign of the presence of strong corrections, possibly due to the
fact that one third of the exchange couplings keeps their pure value K 0 even in the
infinite-disorder limit. Nevertheless, we present in the Table 3 the magnetic exponent as
extracted from the average spin–spin correlation functions for different values of r.
For strong disorder, the scaling dimension xsX , as presented in the Table 3, shows a
plateau with a value compatible within error bars with that of the binary distribution, but
the agreement is not yet conclusive, since the effective central charge was not found to
display a clear maximum.
Table 3
X
Scaling dimension xs of the average order parameter for the q s 3 Potts model with a ternary distribution
compared to the results for a binary distribution at the optimal disorder amplitude and for the pure model.
X
Distribution Disorder amplitude xs Ž3.
Pure r s1 0.1333
Binary r ) , 5.363 0.1347(11)
Ternary r s2 0.1339Ž3.
r s4 0.1343Ž6.
r s 5.363 0.1344(8)
rs8 0.1344(10)
r s12 0.1343Ž13.
r s 20 0.1341Ž15.
4.1.2. Quaternary distribution

The quaternary probability distribution is defined by
P Ž K i j . s 14 d Ž K i j y K . q d Ž K i j y rK . q d Ž K i j y K X . q d Ž K i j y r 2 K X . ,
Ž 25 .
where the four equi-probable exchange couplings K, rK, K X and r 2 K X are related by
exp Ž K c Ž r . . y 1 exp Ž rK c Ž r . . y 1 s exp Ž K cX Ž r . . y 1
exp Ž r 2 K cX Ž r . . y 1 s q Ž 26 .
at the self-dual point of the model. The value r s 1 corresponds to the pure system and
the limit r ™ q` to a percolative regime.
As seen in Fig. 4, the central charge presents both a maximum at r ) , 2.000 and a
‘‘minimum’’ at r) , 3.763. According to Zamolodchikov’s c-theorem in the case of
non-unitary theories, the latter case should correspond to an unstable fixed point while
the former situation is likely to be the disordered fixed point.
In Table 4, we collect the scaling dimensions of the average spin–spin correlation
functions at these two fixed points and, for comparison, those of the pure model and of
the disordered system with a binary distribution.
Inspection of the results in Table 4 reveals that compatible exponents are obtained at
the maximum of the central charge with both probability distributions. These data also
confirm once again that the disorder amplitude r plays an essential role w14x, since the
value at r) is definitely excluded by the perturbative result.
Fig. 4. Behaviour of the central charge as a function of the disorder amplitude r for the quaternary distribution
Ž25. at q s 3. The dotted curve is a guide for the eyes.
Table 4
X
Scaling dimensions xs Ž q s 3. of the average order parameter for the two extrema Ž r ) Žwritten in bold face.
and r) . of the central charge with a quaternary distribution compared to the results for the binary case and for
the pure model.
X
Pure r s1 0.1333
Binary r ) , 5.363 0.1347(11)
Quaternary r ) , 2.000 0.1343(6)
r) s 3.763 0.0097Ž12.
4.1.3. Continuous distribution

Recently, Olson and Young w21x proposed slightly different numerical estimations of
the critical exponents of the average magnetization and of its first moments, compared to
other independent studies w19,20,48x.. They used an interesting continuous probability
distribution of exchange couplings that they claimed to be less sensible to crossover
effects. We show in the following that the randomness amplitude chosen in the
simulations of Olson and Young is not the optimal one, since it is not strong enough to
reach the disordered fixed point. This observation may be at the origin of the slight
discrepancy between the extrapolated values of the exponents.
The continuous probability distribution used by Olson and Young is generalized by
introduction of a parameter r s e l which controls the strength of randomness. Follow-
ing their notation, we have chosen the distribution
1
P Ž yi j . s yi j , Ž 27 .
cosh
l
where
eKijy1
e yi j s . Ž 28 .
' q
Self-duality is ensured by the parity of the distribution P Ž yi j .. The definition of the
disorder amplitude r is such that r s 1 corresponds again to the pure system and r s e
to the Olson–Young distribution. The probability distribution of exchange couplings K i j
is given by
1r l y1
q 1r2 l e K i j Ž e K i j y 1 .
PŽ Ki j. s2 2r l Ž 29 .
pl
q 1r l q Ž e K i j y 1 .
and can be generated by the formula
px
K i j s ln 1 q q tan l
ž ' , / Ž 30 .
2
if x g w0;1w is a uniformly distributed random variable. Examples of probability distribu-
tions at different disorder amplitudes are shown in Fig. 5.
The maximum of the central charge is found in Fig. 6 at the amplitude of disorder
r ) , 3.881. The scaling dimension of the average order parameter, obtained at r ) , is
given in Table 5 and compared to other disorder amplitudes. Again, there is a
convincing agreement with the results obtained with the binary distribution.
Fig. 5. Probability distribution of exchange couplings P Ž K i j . for disorder amplitudes r in the range w1.5;6x.
With the continuous distribution at r s e, Olson and Young measured slightly higher
values for different values of q. We propose here a possible explanation of the small
discrepancy: It was shown that the larger the number of state q of the Potts model the
stronger the disorder should be to reach the disordered fixed point regime w14,19,56x.
Thus, for q G 3 random bond Potts models, the ideal disorder amplitude should become
larger and larger, far from the value r s e , 2.718 implicitly used by Olson and Young.
Fig. 6. Central charge of the q s 3 random bond Potts model for the continuous distribution Eq. Ž29. with
respect to the disorder amplitude r.
Table 5
X
Scaling dimension xs of the average order parameter for the q s 3 random bond Potts model with a
continuous distribution at the optimal disorder amplitude Žwritten in bold face. and other amplitudes as well,
compared to the results obtained in the binary case and for the pure model.
X
Pure r s1 0.1333
Binary r ) , 5.363 0.1347(11)
Continuous r s1.133 0.1338Ž1.
r se 0.1340Ž9.
r ) , 3.881 0.1344(13)
r s 7.389 0.1348Ž17.
On the other hand, since in the weak disorder regime the average exponent is
continuously growing with the disorder amplitude r, a too weak randomness cannot be
the explanation of the slightly too high values for the exponents obtained by Olson and
Young. A possible origin of the small disagreement can be found in the ensemble
average: If the number of disorder realizations is too small, the average behaviour will
give an exponent closer to the typical one, and thus too large.
4.2. Replica symmetry
The question of a possible breaking of replica symmetry w57x in disordered systems is

very controversial and far from being settled, especially in spin glasses Žsee e.g Refs.
w58–61x.. In the context of disordered Potts ferromagnets, the question was first asked
by Dotsenko et al. w49x. The Hamiltonian Ž14. is rewritten,
yb H w s x s K 0 Ý ds , s q Ý Ž K i j y K 0 . ds , s s yb H0 w s x y b H X w s x ,
i j i j
Ži , j. Ži , j.
Ž 31 .
where the deviation from the pure system can be written in the continuum limit
yb H X ; t Ž x . e Ž x . d 2 x ,
H Ž 32 .
with t Ž x . s K Ž x . y K c . The average free energy F s yk B Tln Z can be obtained using

the identity ln Z s lim n ™ 0 Ž Z n y 1 . rn by introduction of n identical copies Žlabelled
by an index a. of the model, coupled by their energy densities. After integration over a
Gaussian probability distribution centered on the value t 0 and with variance D, one is
led to:
Z n s Tr exp yb Ý H0Ž a . q t 0 d 2 x Ý e a Ž x . q D d 2 x
H H Ý ea Ž x . e b Ž x . . Ž 33 .
a a a/b
The first term governed by the average coupling t 0 produces a shift in the critical
temperature, while the second term couples the replicas with each other. In a replica
symmetric scenario, the couplings D between the different copies of the model are
identical, while in a Replica Symmetry Breaking scheme, these couplings are Parisi
matrices and can take different values D a b . Treated as a perturbation this coupling term
leads to different fixed point structures and finally to different scaling dimensions for the
moments of the correlation functions. In order to test between Replica Symmetry and
Replica Symmetry Breaking schemes, Dotsenko et al. performed a second order
expansion of the exponent of the second moment of the spin–spin correlation function
decay in both cases ŽEqs. Ž11. and Ž12... Previous MC simulations have been performed
at q s 3 but were not completely conclusive, although in favour of Replica Symmetry:
The perturbation expansion leads to xsX 2 Ž3. s 0.1176 and xsXX 2 Ž3. s 0.1201 according to
Eqs. Ž11. and Ž12., while previous numerical results lead to 0.113Ž1. w49x, 0.1140Ž5.
w48x, 0.116Ž1. w20x and 0.119Ž2. w21x.
In this section, we report new conclusive results for different values of q. Close to
q s 2, the proximity of the marginally irrelevant Ising FP will surely alter the data, as a
reminiscent effect of the logarithmic corrections present exactly at q s 2 for the second
moment. Too large values of q on the other hand are not very helpful in order to check
perturbation expansions which break down when one explores higher values of the
expansion parameter Žas given for example in Table 1.. One thus has to balance these
two extreme situations and the comparison between numerical data and perturbation
results should be conclusive around q s 3. The TM technique thus appears to be well
adapted, since it is capable to deal with non integer values of q.
The comparison is shown in Fig. 7 for the bimodal probability distribution and the
results are also given in Table 6. In the convenient domain for the test, around q s 3,
results are written in bold face. The agreement with Replica Symmetry is quite
convincing for the bimodal probability distribution. We note that with the continuous
Fig. 7. Exponent of the second moment of the spin–spin correlation function as a function of the number of
states of the disordered Potts model Žbinary disorder.. The comparison is done with Replica Symmetry and
Replica Symmetry Breaking scenarios w49x. The agreement with the RS result is quite good around q s 3.
When q is close to 2, the discrepancy can be attributed to the weak relevance of disorder. We indeed used a
simple exponential fit as can be expected at a stable disordered FP, but at q s 2, one knows from Ludwig’s
results that logarithmic corrections must be added. These corrections can also influence the vicinity of q s 2 in
a numerical approach.
Table 6
Decay exponent of the second moment of the spin–spin correlation function compared to Replica Symmetry
and Replica Symmetry Breaking expressions of Eqs. Ž11. and Ž12.. The results written in bold face correspond
to the range of values of q where the agreement is particularly satisfactory.
Perturbative results
X XX
q xs 2 xs 2 TM result
2.25 0.12213 0.12229 Binary disorder 0.1204Ž5.
2.5 0.12002 0.12067 0.1194Ž8.
2.75 0.11854 0.11997 0.1185(10)
3. 0.11761 0.12011 0.1177(12)
3.25 0.11718 0.12110 0.1172(14)
3.5 0.11723 0.12304 0.1169(16)
3. 0.11761 0.12011 Ternary disorder 0.1182(12)
3. 0.11761 0.12011 Continuous disorder 0.1173(14)
distribution of Eq. Ž29., we obtain also a very good value at q s 3: 0.1173Ž14. and with
the ternary distribution at r s 8 we get 0.1182Ž12..
4.3. Multifractality
The multiscaling behaviour of the spin–spin correlation functions is noticeable in the

1r p
p-dependent set of exponents of the reduced moments Ž ² s Ž 0 . s Ž r . : p . . The
second moment has already been computed when looking for Replica Symmetry, and it
can be generalized in the strip geometry using Eq. Ž21.. We performed an exhaustive
computation of 50 different moments in the range 0 F p F 5 in the strip geometry, and
the associated scaling dimensions followed from a semi-log fit ln² s Ž 0 . s Ž u . : p vs.
ln u, according to Eq. Ž21., followed by an extrapolation to L ™ `. Examples for
q s 2.75, 3 and 3.25 are shown in Fig. 8 Žbimodal probability distribution. where the
numerical results are also compared to the first order expansion of Ludwig and to the
second order expansion in the RS scheme in Eq. Ž10.. The second order result is clearly
very good up to values of p close to 3 and then breaks down as already noticed by
Lewis w48x.
An alternate presentation of the results Žused e.g. by Ludwig w9x. is given by the
scaling dimension of the moment of the correlation function itself, ² s Ž 0 . s Ž r . : p Žnot
1r p
the reduced function Ž ² s Ž 0 . s Ž r . : p . .. The scaling dimension is thus simply
X X
pxs p Ž q ., hereafter denoted by Xs p Ž q . by a simple extension of Ludwig’s notation. An
example, with q s 3., is shown in Fig. 9 where we have also shown the results obtained
with the continuous probability distribution at the optimal disorder amplitude. Once
again, we find a promising agreement between the numerical data and the perturbative
result which confirms universality.
What is interesting in this latter presentation is the link with other fields where
multifractality is observed. One then usually introduces a universal function, the
multiscaling function H Ž a ., which is simply the Legendre transform of the set of
independent scaling indexes XsX p Ž q .. Setting d XsX p Ž q . s a d p, this function is simply
obtained by H Ž a . s XsX p Ž q . y a p. The geometrical interpretation of this Legendre
transform follows from the relation Ž E HrEa . s yp where a is defined by
Ž E XsX p Ž q .rE p . s a . The scaling dimension xsX p Ž q . is obtained on the plot of H Ž a . by
Fig. 8. Comparison of the multifractal exponents Žreduced moment of the correlation function
1r p
Ž ² s Ž0. s Ž u . : p . . with the second order expansion of Lewis in the RS scheme w47x for different values of
q indicated in the figure Žbimodal probability distribution..
the intercept of the tangent of slope yp with the abscissa axis. An example of
multiscaling function H Ž a . deduced from the numerical data with the bimodal probabil-
ity distribution is shown in Fig. 10 for q s 3 and the line of slope y2, leading to the
Fig. 9. Comparison of the multifractal exponents Žmoment of the correlation function ² s Ž 0 . s Ž u . : p. with the
second order expansion of Lewis in the RS scheme w47x for both the bimodal and the continuous probability
distributions.
exponent of the second moment Žsee Table 6. is also shown. For comparison, the
function HŽ k .Ž a . deduced from the k th order Žin yH . perturbative results of Ludwig
Ž k s 1. and Lewis Ž k s 2. Ž10. are also plotted. Other values of q are shown in Fig. 11.
4.4. Correlation function probability distribution
In Ref. w9x, Ludwig presented a remarkable discussion of the spin–spin correlation
function probability distribution. He showed how all the relevant information on the
Fig. 10. Universal function H Ž a . for different q s 3. The functions HŽ1.Ž a . and HŽ2.Ž a . deduced from Eq.
Ž10. at first or second order, respectively, are shown for comparison.
Fig. 11. Universal function H Ž a . for different values of q Žbimodal probability distribution..
large distance behaviour is encoded in the multifractal function H Ž a .. In this section,

we follow Ludwig’s arguments and report a numerical study of the correlation function
probability distribution in the cylinder geometry.
According to the results of the previous section, the moments of the spin–spin
correlation function along the strip asymptotically behaves as follows:
2p u X
G p Ž u . ' ² s Ž 0 . s Ž u . : p ; Bp ey L
Xs p
, Ž 34 .
and are defined in terms of the probability distribution P w GŽ u.x:
1
G p Ž u. s H0 dG Ž u . P G Ž u. G p Ž u. . Ž 35 .
Following Ludwig, we introduce the variable Y Ž u. s ylnGŽ u. and write G p Ž u. s

eyp Y Ž u.. Using the identity P w GŽ u.x dG s P w Y Ž u.x dY and Eqs. Ž34. and Ž35., one
obtains
` 2p u X
H0 dY Ž u . P Y Ž u . eyp Y Ž u. ; Bp ey L
Xs p
, Ž 36 .
which leads to the expression of the probability distribution by inverting the Laplace
transform Ž d ) 0.:
1 dqi` y
2p u X
Xs p y
Y Ž u.
p .
P Y Ž u. s H dpB e p L 2p urL Ž 37 .
2ip dyi`
The amplitude Bp is weakly depending on p. Following Ludwig, it can be rewritten as

Bp s expwyŽ2p urL.Žyln BprŽ2p urL..x, but can be forgotten, since it only introduces
a small correction when 2p urL ™ `. Let us define the function hŽ p . s XsX p y 2pY ŽurL
u.
p.
Fig. 12. Behaviour of the probability distribution P Ž a . as a function of the distance along the strip 2p u r L.
X
The rescaled position along the strip is written u s 2p u r L. Three fixed values of a are shown, the opened
and filled symbols respectively correspond to the strip widths Ls6 and 7. A good collapse of the data at both
sizes is observed, but the behaviour displays a deviation from linearity at small distances Ž q s 3, binary
distribution..
In the large distance limit 2p urL ™ `, the integral can be evaluated by the saddle-point
approximation at the minimum p 0 of hŽ p .:
E X Y Ž u.
žEp
Xs p
/
p0
s
2p urL
. Ž 38 .
Instead of Y Ž u., we define the scaled variable a s Y Ž u.rŽ2p urL., and the saddle point
value at p 0 only depends on this variable hŽ p 0 . s H Ž a ., where H Ž a . is nothing but
Fig. 13. Same as Fig. 12 accounting for the correction in Eq. Ž43. close to the saddle-point approximation.
Fig. 14. Multifractal function H Ž a . as it is deduced from the fit of the probability distribution P Ž a . to Eq.
Ž43. accounting for the correction near the saddle-point approximation. It is compared to the results of Fig. 10
in solid line. The insert shows slightly too large values of H Ž a . when deduced from Eq. Ž41..
the multifractal function defined in the previous section. We thus obtain the probability
distribution
2p u Y Ž u.
P Y Ž u . ; exp y
L
H
ž 2p urL / , Ž 39 .
or, using P w Y Ž u.x dY s P Ž a . d a ,

2p u 2p u
PŽ a . ; exp y HŽ a . . Ž 40 .
L L
The multifractal function contains the essential information on the probability distribu-
tion. In order to check this expression, the value of H Ž a . at fixed a is extracted by
fitting the probability distribution to the expression
2p u 2p u
ln P Ž a . s const q ln HŽ a . .
y Ž 41 .
L L
It is shown in Fig. 12 where the probability distribution of the spin–spin correlation
function was obtained after collecting the results over 96 000 disorder realizations in 50
classes. The values of H Ž a . are slightly too large, compared to the results presented in
the previous section. We can indeed observe in Fig. 12 a deviation from the linear
behaviour which would be expected with these variables, and the shorter the distance u,
the larger the deviation.
This could be due to a correction to the leading behaviour given by the saddle-point
approximation5. If we expand the function hŽ p . close to p 0 , hŽ p . , H Ž a .
5
We mention here that a possible correction to the saddle-point approximation has been suggested by Olson
and Young in their study of higher values of q.
q 12 hXX Ž p 0 .Ž p y p 0 . 2 , with hXX Ž p 0 . ) 0 we obtain, instead of Eq. Ž39., the following

result for the probability distribution P w Y Ž u.x w62x:
y1 r2
2p u 2p u Y Ž u.
P Y Ž u. ; ž / L
exp y
L
H
ž 2p urL / , Ž 42 .
and a correction appears in P Ž a .:

2p u 2p u
ln P Ž a . y 12 ln s const y HŽ a . . Ž 43 .
L L
This is shown in Fig. 13 where a linear behaviour is now obtained in the whole range of
values of urL. A linear fit in the coordinates of Fig. 13 gives the value of the
multifractal function H Ž a . which can be compared to the results of Fig. 10 obtained in
Section 4.3. This fit is performed for all values of 0.034 - a - 0.15 for q s 3 for the
cases of the bimodal probability distribution and of the continuous distribution. The
results are shown in Fig. 14.
This latter figure shows that the correlation function probability distribution is
entirely determined by the universal multifractal function H Ž a ..
5. Conclusions
In this paper, we have investigated the critical behaviour of the moments of the
spin–spin correlation functions of two-dimensional random bond Potts ferromagnets
using Transfer Matrix techniques and conformal methods. New features of our present
work are the following:
1. As far as we know, universality of the critical behaviour of the moments of the

correlation function was checked for the first time in such systems6 . This statement
follows from the numerical evidence that the scaling dimension of the average
spin–spin correlation function, as well as those of higher moments or typical
behaviour do not depend on the details of the probability distribution, provided that
the computations are performed at the disordered fixed point given by the maximum
condition of the central charge. The exponent of the average correlation function is
furthermore in very good agreement with theoretical results using perturbative
conformal field theory available in the literature.
2. The question of a possible breaking of Replica Symmetry is also considered and the
numerical data strongly support the absence of Replica Symmetry Breaking. The
problem is investigated through the comparison of the scaling dimension of the
second moment of the correlation function, which is compared to perturbative results
obtained within both schemes. Although previous numerical results Žusing Monte
Carlo simulations. which led to similar conclusions were already reported at q s 3,
we believe that our study is conclusive, since it extends the work to other non integer
values of q.
6
We mention here that a MC study in the first-order regime of the pure model Ž q s 5. was recently
reported where random bond disorder and dilution were found to belong to the same universality class w63x.
3. The multiscaling behaviour of the spin–spin correlation function is investigated. The

exponential decay of the moments of the correlation functions along very long strips
is used to deduce numerically the corresponding critical exponents. These dimensions
continuously depend on the moment order, as a consequence of multifractality. They
are furthermore very weakly dependent of the probability distribution. At low
moment order, the numerical results are furthermore in agreement with a perturbative
result obtained within the Replica Symmetry scheme. When the moment order
increases, a discrepancy is observed, resulting from the lack of validity of the
perturbation expansion, and possibly of a numerical determination which becomes
less precise. The multifractal function, given by the Legendre transform of the set of
independent scaling dimensions is also computed for different values of q. It is
shown that this universal function completely determines the shape of the correlation
function probability distribution.
The main result of this paper is probably to show that universality in random systems
has to be understood in the sense of a critical behaviour which does not depend on the
choice of the probability distribution Žthis is only true up to some extent, since special
distributions which do not obey the central limit theorem, like Levy flights, would
certainly lead to different results.. By critical behaviour, here we mean the behaviour of
all the moments of a physical quantity, entirely contained in the multifractal function or
the correlation function probability distribution but we want to stress that lack of
self-averaging does not imply absence of universality.
We also note that logarithmic corrections were recently reported in the disconnected
energy–energy correlations function w64x and that disorder was shown to induce
non-vanishing cross-correlations between spin–spin and energy–energy moments w65x.
The multiscaling of energy correlations has been studied very recently by J.L. Jacobsen
w66x.
Acknowledgements
We would like to thank M.A. Lewis and I. Campbell for stimulating discussions. The
computations were performed on the SP2 at the CNUSC in Montpellier under project
No. C990011, and the Power Challenge Array at the CCH in Nancy.
References
w1x V.S. Dotsenko, Phys. Usp. 38 Ž1995. 457.
w2x A. Aharony, A.B. Harris, Phys. Rev. Lett. 77 Ž1996. 3700.
w3x B. Derrida, Phys. Rep. 103 Ž1984. 29.
w4x S. Wiseman, E. Domany, Phys. Rev. Lett. 81 Ž1998. 22.
w5x S. Wiseman, E. Domany, Phys. Rev. E 58 Ž1998. 2938.
w6x H.E. Stanley, P. Meakin, Nature 335 Ž1988. 405.
w7x M. Janssen, Int. J. Mod. Phys. B 8 Ž1994. 943.
w8x A.W.W. Ludwig, Nucl. Phys. B 285 wFS19x Ž1987. 97.
w9x A.W.W. Ludwig, Nucl. Phys. B 330 Ž1990. 639.
w10x A.W.W. Ludwig, J.L. Cardy, Nucl. Phys. B 330 wFS19x Ž1987. 687.
w11x S. Wiseman, E. Domany, Phys. Rev. E 51 Ž1995. 3074.
w12x M. Picco, Phys. Rev. B 54 Ž1996. 14930.

w13x J.K. Kim, Phys. Rev. B 53 Ž1996. 3388.
w14x M. Picco, cond-matr9802092.
w15x J.L. Jacobsen, J.L. Cardy, Nucl. Phys. B 515 Ž1998. 701.
w16x U. Glaus, J. Phys. A 20 Ž1987. L595.
w17x M. Picco, Phys. Rev. Lett. 79 Ž1997. 2998.
w18x J.L. Cardy, J.L. Jacobsen, Phys. Rev. Lett. 79 Ž1997. 4063.
w19x C. Chatelain, B. Berche, Phys. Rev. E 60 Ž1999. 3853.
w20x ´
G. Palagyi, ´ cond-matr9906067.
C. Chatelain, B. Berche, F. Igloi,
w21x T. Olson, A.P. Young, Phys. Rev. B 60 Ž1999. 3428.
w22x C. Chatelain, B. Berche, Phys. Rev. E 58 Ž1998. R6899.
w23x P.W. Kasteleyn, C.M. Fortuin, J. Phys. Soc. Jpn. Suppl. 26 Ž1969. 11.
w24x ¨ M.P. Nightingale, Physica A 112 Ž1982. 405.
H.W.J. Blote,
w25x A.B. Harris, J. Phys. C 7 Ž1974. 1671.
w26x Vik.S. Dotsenko, Vl.S. Dotsenko, Adv. Phys. 32 Ž1983. 129.
w27x B.N. Shalaev, Sov. Phys. Sol. State 26 Ž1984. 1811.
w28x R. Shankar, Phys. Rev. Lett. 58 Ž1987. 2466.
w29x R. Shankar, Phys. Rev. Lett. 61 Ž1988. 2390.
w30x A.W.W. Ludwig, Phys. Rev. Lett. 61 Ž1988. 2388.
w31x B.N. Shalaev, Phys. Rep. 237 Ž1994. 129.
w32x J.S. Wang, W. Selke, Vl.S. Dotsenko, V.B. Andreichenko, Europhys. Lett. 11 Ž1990. 301.
w33x V.B. Andreichenko, Vl.S. Dotsenko, W. Selke, J.S. Wang, Nucl. Phys. B 344 Ž1990. 531.
w34x A.L. Talapov, L.N. Shchur, Europhys. Lett. 27 Ž1994. 193.
w35x A. Roder, J. Adler, W. Janke, Phys. Rev. Lett. 80 Ž1998. 4697.
w36x A. Roder, J. Adler, W. Janke, Physica A 265 Ž1999. 28.
w37x G. Mazzeo, R. Kuhn,¨ Phys. Rev. E 60 Ž1999. 3823.
w38x V.N. Plechko, Phys. Lett. A 239 Ž1998. 289.
w39x S.L.A. de Queiroz, Phys. Rev. E 51 Ž1995. 1030.
w40x S.L.A. de Queiroz, R.B. Stinchcombe, Phys. Rev. E 54 Ž1996. 190.
w41x P. Lajko,´ F. Igloi,
´ cond-matr9908376.
w42x M.P.M. den Nijs, J. Phys. A 12 Ž1979. 1857.
w43x B. Nienhuis, J. Phys. A 15 Ž1982. 199.
w44x Vl.S. Dotsenko, V.A. Fateev, Nucl. Phys. B 240 wFS12x Ž1984. 312.
w45x G. Jug, B.N. Shalaev, Phys. Rev. B 54 Ž1996. 3442.
w46x Vl. Dotsenko, M. Picco, P. Pujol, Nucl. Phys. B 455 wFSx Ž1995. 701.
w47x M.A. Lewis, Europhys. Lett. 43 Ž1998. 189, ŽE. 47 Ž1999. 129.
w48x M.A. Lewis, cond-matr9905401.
w49x Vik. Dotsenko, Vl. Dotsenko, M. Picco, Nucl. Phys. B 250 Ž1998. 633.
w50x Vik. Dotsenko, Vl. Dotsenko, M. Picco, P. Pujol, Europhys. Lett. 32 Ž1995. 425.
w51x L. Chayes, K. Shtengel, cond-matr981103.
w52x Vl. Dotsenko, J.L. Jacobsen, M.A. Lewis, M. Picco, cond-matr9812227.
w53x A.B. Zamolodchikov, JETP Lett. 43 Ž1986. 730.
w54x H. Furstenberg, Trans. Am. Math. Soc. 108 Ž1963. 377.
w55x N. Sourlas, cond-matr9811406.
w56x J.L. Jacobsen, M. Picco, cond-matr9910071.
w57x ´
M. Mezard, G. Parisi, M.A. Virasoro, Spin glass theory and beyond ŽWorld Scientific, Singapore, 1987..
w58x E. Marinari, G. Parisi, J. Ruiz-Lorenzo, F. Ritort, Phys. Rev. Lett. 76 Ž1996. 843.
w59x E. Marinari, C. Naitza, F. Zuliani, G. Parisi, M. Picco, F. Ritort, Phys. Rev. Lett. 81 Ž1999. 1698.
w60x H. Bokil, A.J. Bray, B. Drossel, M.A. Moore, Phys. Rev. Lett. 82 Ž1999. 5174.
w61x E. Marinari, C. Naitza, F. Zuliani, G. Parisi, M. Picco, F. Ritort, Phys. Rev. Lett. 82 Ž1999. 5175.
w62x N.G. De Bruijn, Asymptotic methods in analysis ŽDover, New York, 1981..
w63x R. Paredes, J. Valbuena, Phys. Rev. E 59 Ž1999. 6275.
w64x J.L. Cardy, cond-matr9911024.
w65x T. Davis, J.L. Cardy, cond-matr9911083.
w66x J.L. Jacobsen, cond-matr9912304.
Scalar field theory in the AdSrCFT correspondence revisited

Pablo Minces, Victor O. Rivelles
˜ Paulo, Instituto de Fısica,
UniÕersidade de Sao ´ ˜ Paulo, Brazil
Caixa Postal 66.318, CEP 05315-970, Sao
Received 1 September 1999; received in revised form 17 December 1999; accepted 23 December 1999
Abstract
We consider the role of boundary conditions in the AdS dq1rCFTd correspondence for the
scalar field theory. Also a careful analysis of some limiting cases is presented. We study three
possible types of boundary conditions, Dirichlet, Neumann and mixed. We compute the two-point
functions of the conformal operators on the boundary for each type of boundary condition. We
show how particular choices of the mass require different treatments. In the Dirichlet case we find
that there is no double zero in the two-point function of the operator with conformal dimension
dr2. The Neumann case leads to new normalizations for the boundary two-point functions. In the
massless case we show that the conformal dimension of the boundary conformal operator is
precisely the unitarity bound for scalar operators. We find a one-parameter family of boundary
conditions in the mixed case. There are again new normalizations for the boundary two-point
functions. For a particular choice of the mixed boundary condition and with the mass squared in
the range yd 2r4 - m 2 - yd 2r4 q 1 the boundary operator has conformal dimension comprised
in the interval wŽ d y 2.r2, dr2x. For mass squared m2 ) yd 2r4 q 1 the same choice of mixed
boundary condition leads to a boundary operator whose conformal dimension is the unitarity
bound. q 2000 Elsevier Science B.V. All rights reserved.
PACS: 11.10.Kk; 11.25.Mf

Keywords: AdSrCFT correspondence; Boundary conditions; Holographic principle
1. Introduction
Since the proposal of Maldacena’s conjecture, which gives a correspondence between

a field theory on anti-de Sitter space ŽAdS. and a conformal field theory ŽCFT. on its
boundary w1x, an intensive work has been devoted to get a deeper understanding of its
E-mail addresses: pablo@fma.if.usp.br ŽP. Minces., rivelles@fma.if.usp.br ŽV.O. Rivelles..
PII: S 0 5 5 0 - 3 2 1 3 Ž 9 9 . 0 0 8 3 3 - 0
652 P. Minces, V.O. RiÕellesr Nuclear Physics B 572 [FS] (2000) 651–669
implications. In particular, a precise form to the conjecture has been given in w2,3x. It
reads
ZAdS w f 0 x s Hf Df exp Ž yI w f x . ' Z

0
CFT ¦ žH
w f 0 x s exp
EV
d d x Of 0 /;
, Ž 1.
where f 0 is the boundary value of the bulk field f which couples to the boundary CFT
operator O . This allows us to obtain the correlation functions of the boundary CFT
theory in d dimensions by calculating the partition function on the AdS dq 1 side. The
AdSrCFT correspondence has been studied for the scalar field w3–6x, the vector field
w3,5,7,8x, the spinor field w7,9,10x, the Rarita–Schwinger field w11–13x, the graviton field
w14,15x, the massive symmetric tensor field w16x and the antisymmetric p-form field
w17,18x. In all cases Dirichlet boundary conditions were used. Several subtle points have
been clarified in these papers and all results lend support to the conjecture.
In a broader sense Maldacena’s conjecture is a concrete realization of the holographic
principle w19,20x. We expect that any field theory relationship in AdS space must be
reflected in the border CFT. An example of this is the well known equivalence between
Maxwell–Chern–Simons theory and the self-dual model in three-dimensional Minkowski
space w21x. This equivalence holds also in AdS 3 and using the AdSrCFT correspon-
dence we have shown that the corresponding boundary operators have the same
conformal dimensions w8x. Another situation involves massive scalar fields in AdS
spaces. If the scalar field has mass-squared in the range yd 2r4 - m2 - yd 2r4 q 1
then there are two possible quantum field theories in the bulk w22x. The AdSrCFT
correspondence with Dirichlet boundary condition can easily account for one of the
theories. The other one appears in a very subtle way by identifying a conjugate field
through a Legendre transform as the source of the boundary conformal operator w23x.
The existence of two conjugated boundary operators has been first pointed out in w24x.
Since a field theory is determined not only by its Lagrangian but also by its boundary
terms in the action we expect that the AdSrCFT correspondence must be sensitive to
these boundary terms. This is easily seen to be true by computing the left-hand side of
Eq. Ž1. for a classical field configuration. All that is left is a boundary term. If we start
with different boundary terms in the action then we obtain different correlation functions
on the right-hand side.
The origin of boundary terms in the action is due to the variational principle. In order
to have a stationary action boundary terms, which will depend on the choice of the
boundary conditions, must be introduced. The importance of these boundary terms for
the AdSrCFT correspondence was recognized in the case of spinor fields where the
action is of first order in derivatives and the classical action vanishes on-shell w25x. They
also played an important role in the case of Chern–Simons theory w8x. Therefore it is
crucial to understand the implications of different types of boundary conditions for the
same theory since they in general imply different boundary terms.
In this work we will study the role of different types of boundary condition for the
scalar field theory. We will consider Dirichlet and Neumann boundary conditions, and a
combination of both of them which we will call mixed boundary condition. Each type of
boundary condition requires a different boundary term. We will show that the mixed
boundary conditions are parametrized by a real number so that there is a one-parameter
family of boundary terms consistent with the variational principle.
P. Minces, V.O. RiÕellesr Nuclear Physics B 572 [FS] (2000) 651–669 653
We will also show that different types of boundary condition give rise to different
conformal field theories at the border. For the scalar field this was somehow expected.
The two solutions found in w22x correspond to two different choices of energy–momen-
tum tensor. Both of them are conserved and their difference gives a surface contribution
to the isometry generators. Although these two solutions were found in the Hamiltonian
context by requiring finiteness of the energy they will reappear here by considering
different types of boundary condition which amounts to different boundary terms in the
action. We can also look for the asymptotic behavior of the scalar field near the
boundary according to the chosen type of boundary condition. For the Dirichlet
boundary condition it is well known that the scalar field behaves as x 0d r2y 'd r4qm
2 2
near the border at x 0 s 0. There is no upper restriction on the mass in this case. It
corresponds to one of the solutions found in w22x and gives rise to a boundary conformal
(
operator with conformal dimension dr2 q d 2r4 q m2 . We will show that for a
particular choice of mixed boundary condition and when the mass squared is in the
range yd 2r4 - m 2 - yd 2r4 q 1 the scalar field behaves as x 0d r2q 'd r4qm near the
2 2
border. It corresponds precisely to the second solution of w22x and gives rise to a
(
boundary conformal operator with conformal dimension dr2 y d 2r4 q m2 . Note that
the upper limit for the mass squared yd 2r4 q 1 is consistent with the unitarity bound
Ž d y 2.r2.
Another important point that we will show is the existence of boundary conditions
which give rise to boundary conformal operators for which the unitarity bound Ž d y 2.r2
is reached. They correspond to a massless scalar field with Neumann boundary condition
or to a massive scalar field with m2 ) yd 2r4 q 1 with a particular choice of the mixed
boundary condition Žthe same choice which gives the boundary operator with conformal
(
dimension dr2 y d 2r4 q m2 .. In this way, using different boundary conditions, we
obtain all scalar conformal field theories allowed by the unitarity bound.
We will also analyze carefully two cases where the mass of the scalar field takes
special values. In some cases the usual expansion of the modified Bessel functions in
powers of x 0 breaks down and we must use expansions involving logarithms. When
m2 s yd 2r4 it gives rise to the asymptotic behavior x 0d r2 ln x 0 and the two-point
function is obtained without troubles. This is to be contrasted with the usual limiting
procedure where the mass goes to m2 s yd 2r4 but the two-point function has a double
(
zero in the limit w23x. The other case corresponds to d 2r4 q m2 integer but non-zero.
In this case we just reproduce the known results.
We should stress the fact that the use of different types of boundary condition Žfor
given values of m2 and d . allows us in general to get boundary two-point functions with
different normalizations. This will affect the three-point and higher-point functions.
Maybe this is related to the fact that AdS and field theory calculations agree up to some
dimension dependent normalization factors w26x but we will not discuss this further.
The paper is organized as follows. In Section 2 we find the boundary terms
corresponding to each type of boundary condition. In Section 3 we consider the Dirichlet
case while in Section 4 we treat Neumann boundary conditions. In Section 5 we
consider mixed boundary conditions. Finally Section 6 presents our conclusions. In
Appendix A we list all boundary two-point functions that we computed and Appendix B
contains some useful formulae.
2. The variational principle
We take the usual Euclidean representation of the AdS dq 1 in Poincare´ coordinates

described by the half space x 0 ) 0, x i g R with metric
1 d
ds 2 s Ý dx m dx m . Ž 2.
x 02 ms0
The action for the massive scalar field theory is given by

1
I0 s Hd dq 1
x g Ž g mn Em f En f q m 2f 2 . ,
' Ž 3.
2
and the corresponding equation of motion is
Ž = 2 y m2 . f s 0. Ž 4.
The solution which is regular at x 0 ™ ` reads w5x
d dk
fŽ x. s H Ž 2p . d
eyi kP x x 0d r2 a Ž k . Kn Ž kx 0 . , Ž 5.
where x s Ž x 1, . . . , x d ., k sN k N , Kn is the modified Bessel function, and
d2
ns ( 4
q m2 . Ž 6.
From Eq. Ž5. we also get

d
d dk yi kP x 2
y1 d
E0f Ž x. s H Ž 2p . d
e x0 aŽ k . ž 2 /
q n Kn Ž kx 0 . y kx 0 Knq1 Ž kx 0 . .
Ž 7.
In order to have a stationary action we must supplement the action I0 with a
boundary term IS which cancels its variation. The appropriate action is then
I s I0 q IS . Ž 8.
In order to capture the effect of the Minkowski boundary of the AdS dq 1 , situated at
x 0 s 0, we first consider a boundary value problem on the boundary surface x 0 s e ) 0
and then take the limit e ™ 0 at the very end. Then the variational principle applied to
the action I gives
d I s y d d x eyd q1 E 0 fe dfe q d IS s 0,
H Ž 9.
where fe and E 0 fe are the value of the field and its derivate at x 0 s e respectively.
This equation will be used below to find out the appropriate boundary term IS for each
type of boundary condition.
For Dirichlet boundary condition the variation of the field at the border vanishes so
that the first term in Eq. Ž9. also vanishes and the usual action I0 is already stationary.
Making use of the field equation the action I takes the form
1 1
ID s H d dq 1 x Em Ž g f E mf . s y
' d d x eydq1fe E 0 fe .
H Ž 10 .
2 2
It is to be understood that E 0 fe in Eq. Ž10. is evaluated in terms of the Dirichlet data
fe .
To consider Neumann boundary conditions we first take a unitary vector which is
inward normal to the boundary n m Ž x 0 . s Ž x 0 ,0.. The Neumann boundary condition then
fixes the value of n m Ž e . Em fe ' En fe . The boundary term to be added to the action reads
IS s y d dq 1 x Em Ž g g mnf En f . s d d x eydq1fe E 0 fe ,
H ' H Ž 11 .
so that we find the following expression for the action at the boundary:
1
IN s H d d x eydfe En fe . Ž 12 .
2
Here fe is to be expressed in terms of the Neumann value En fe . Notice that the on-shell
value of the action with Neumann boundary condition Ž12. differs by a sign from the
corresponding action with Dirichlet boundary condition Ž10..
We now consider a boundary condition which fixes the value of a linear combination
of the field and its normal derivative at the border
f Ž x . q a n m Em f Ž x . ' c a Ž x . . Ž 13 .
We will call it as mixed boundary condition. Here a is an arbitrary real but non-zero
coefficient. In this case the surface term to be added to the action is
a a
ISa s H d dq1 x Em g g mn En f n r Er f s y
ž' / d d x eydq2 E 0 fe E 0 fe ,
H Ž 14 .
2 2
and we find the following expression for the action at the boundary:
1
IMa s y Hd d x eydq1cea E 0 fe . Ž 15 .
2
Clearly E 0 fe in the above expression must be written in terms of the boundary data cea.
We then have a one-parameter family of surface terms since the variational principle
does not impose any condition on a . In this way the value of the on-shell action Ž15.
also depends on a .
In the following sections we will consider each boundary condition separately.
3. Dirichlet boundary condition
We begin by recalling the main results for the Dirichlet case w4,5x. Let fe Ž k . be the
Fourier transform of the Dirichlet boundary value of the field f Ž x .. From Eq. Ž5. we
get
eyd r2 fe Ž k .
aŽ k . s , Ž 16 .
Kn Ž k e .
and inserting this into Eq. Ž7. we find
d
d dk d Knq1 Ž k e .
E 0 fe Ž x . s d y fe Ž y .
H H Ž 2p . d
eyi kP Ž xyy . ey1 qnyke .
2 Kn Ž k e .
Ž 17 .
Then the action Ž10. reads
1 d dk
ID s y H d d x d d y fe Ž x . fe Ž y . eyd H Ž 2p . d
eyi kP Ž xyy .
2
d Knq1 Ž k e .
= qnyke . Ž 18 .
2 Kn Ž k e .
The next step is to keep the relevant terms in the series expansions of the Bessel
functions and to integrate in k. We consider first the case n / 0 that is m 2 / yd 2r4.
For completeness we list the relevant modified Bessel functions in Appendix B. For n
not integer we make use of Eqs. ŽB.1., ŽB.5., whereas for n integer but non-zero we use
Eqs. ŽB.2., ŽB.6.. In both cases we get the same result
d d
n/0
ID s y d r2
n G
2
qn ž d d
/H
d x d y fe Ž x . fe Ž y .
e 2Ž ny 2 .
d
q . . . , Ž 19 .
p G Žn . 2Ž q n .
< xyy< 2
where the dots stand for either contact terms or higher order terms in e .
Taking the limit w5x
d
lim e ny 2 fe Ž x . s f 0 Ž x . , Ž 20 .
e™0
to go to the border and making use of the AdSrCFT equivalence in the form
exp Ž yIAdS . ' exp ¦ žH d d x O Ž x . f0 Ž x . /; , Ž 21 .

we find the following two-point function:
d
n/0 n /0
2n
² OD Ž x . OD Ž y . :s d r2
G
2
qn ž / 1
d
. Ž 22 .
p G Žn .
< x y y < 2Ž 2 q n .
Then the conformal operator ODn / 0 on the boundary CFT has conformal dimension
d d r2y n
2 q n . From Eq. 20 we find that near the border f behaves as x 0 f 0 Ž x . as
Ž .
expected. In this way we have extended the results of w4,5x to the case n integer but
non-zero.
For future reference we note that in the particular case m s 0, that is n s dr2, Eq.
Ž22. reads
d G Ž d. 1
² ODnsd r2 Ž x . ODnsd r2 Ž y . :s d r2
, Ž 23 .
p G Ž dr2 . < x y y < 2 d
so that the operator ODnsd r2 has conformal dimension d.
Now we consider the case n s 0, that is m2 s yd 2r4. Since the two-point function
Ž22. has a double zero for n s 0 it was argued w23x that the correct result can be found
by introducing a normalization on the boundary operator. Instead we will make use of
the expansion for the Bessel function K 0 . Using Eqs. ŽB.2., ŽB.4. we get
2
K 1Ž k e . 1 Ž ke . ln k y ln 2 q g
ke sy 1q ln e q O Ž e 2 . 1y qOŽ e 2 .
K0 Ž ke . ln e 2 ln e
ln k
s q..., Ž 24 .
ln2e
where the dots denote all other terms representing either contact terms in the two-point
function or terms of higher order in e . Notice that it is essential to separate the
contributions of k and e in the terms ln k e in order to identify the relevant contribu-
tions. Substituting in Eq. Ž18. we find
1 eyd d dk
IDns0 s H d d x d d y fe Ž x . fe Ž y . H Ž 2p . eyi kPŽ xyy. ln k q . . . . Ž 25 .
2 ln2e d
The integration in k is carried out by making use of Eq. ŽB.6. yielding

G Ž dr2 . eyd 1
IDns0 s y d r2 H d d xd d y fe Ž x . fe Ž y . q.... Ž 26.
4p ln e < x y y < d
2
To go to the border we have to rescale fe using a factor of ln e . This makes the

rescaling somewhat arbitrary since any power of e in ln e would do the job. So
choosing the limit
y1
lim Ž e d r2 ln e . fe Ž x . s f 0 Ž x . , Ž 27 .
e™0
and making use of the AdSrCFT equivalence Ž21. we find the following two-point
function:
G Ž dr2 . 1
² ODns0 Ž x . ODns0 Ž y . :s d r2
. Ž 28 .
2p < xyy< d
Then the conformal operator ODns0 on the boundary CFT has conformal dimension dr2
as expected. As anticipated in w5x the scalar field approaches the boundary as
x 0d r2 ln x 0 f 0 Ž x . due to the logarithm appearing in the expansion of the Bessel function.
4. Neumann boundary condition
Using the Neumann boundary condition we get from Eq. Ž7.

eyd r2 En fe Ž k .
aŽ k . s , Ž 29 .
d
ž 2 /
q n Kn Ž k e . y k e Knq1 Ž k e .
and substituting this in Eq. Ž5. we find
d dk 1
fe Ž x . s d d y En fe Ž y .
H H Ž 2p . d
eyi kP Ž xyy . . Ž 30 .
d Knq1 Ž k e .
qnyke
2 Kn Ž k e .
Then the action Ž12. reads
1 d
d dk
IN s Hd x d d y En fe Ž x . En fe Ž y . eyd H Ž 2p . d
eyi kP Ž xyy .
2
1
= . Ž 31 .
d Knq1 Ž k e .
qnyke
2 Kn Ž k e .
In order to keep the relevant terms in the series expansions of the Bessel functions we
must consider the massive and massless cases separately.
In the massless case we have n s dr2. For d odd we make use of Eq. ŽB.1., whereas
for d even we use Eq. ŽB.2.. In both cases we get for d ) 2
1 y2
s y Ž d y 2. Ž k e . q..., Ž 32 .
d Knq1 Ž k e .
qnyke
2 Kn Ž k e .
up to contact terms and higher order terms in e . Substituting this in Eq. Ž31. we find
dy2
INnsd r2 s y Hd d
x d d y En fe Ž x . En fe Ž y .
2
d dk
=eyd y2 H Ž 2p . d
eyi kPŽ xyy. ky2 q . . . , Ž 33.
and performing the integral in k we get
G Ž dr2 . eyd y2
INnsd r2 s y Hd d
x d d y En fe Ž x . En fe Ž y . q..., Ž 34 .
4p d r2 dy2
< xyy< 2 2
where the dots stand for either contact terms or higher order terms in e .
Taking the limit
d
lim ey 2 y1 En fe Ž x . s En f 0 Ž x . , Ž 35 .
e™0
and making use of the AdSrCFT equivalence of the form
exp Ž yIAdS . ' exp ¦ žH d d x O Ž x . En f 0 Ž x . /; , Ž 36 .

we find the following boundary two-point function:

G Ž dr2 . 1
² ONnsd r2 Ž x . ONnsd r2 Ž y . :s d r2 dy2
. Ž 37 .
2p
< xyy< 2 2
Then for d ) 2, even or odd, the conformal dimension of the operator ONnsd r2 is
precisely the unitarity bound Ž d y 2.r2. From Eq. Ž35. we find that near the border the
scalar field goes as x 0d r2q1 En f 0 Ž x .. Comparing Eqs. Ž23., Ž37. we see that the
conformal dimensions of the boundary operators for the massless Dirichlet and Neu-
mann cases are different and for the later case the unitarity bound is reached.
For the massive scalar field, that is n / dr2, we first consider the case n / 0 i.e.
m 2 / yd 2r4. We have again to consider separately the cases with n not integer and n
integer but non-zero. In both cases we find
d
INn / 0, d r2 s y
n
d r2
1
2
G ž 2
qn /
p d G Žn .
ž 2
yn /
d
e 2Ž ny 2 .
= d d x d d y En fe Ž x . En fe Ž y .
H d
q.... Ž 38.
< x y y < 2Ž 2 q n .
Taking the limit

d
lim e ny 2 En fe Ž x . s En f 0 Ž x . , Ž 39 .
e™0
and making use of the AdSrCFT equivalence Ž36. we find the following boundary
two-point function:
d
² ONn / 0, d r2 Ž x . ONn / 0, d r2 Ž y . :s
2n
d r2
1
2
G ž2
qn / 1
d
.
p d G Žn . 2Ž qn .
< xyy<
ž 2
yn / 2
Ž 40 .
Then the operator ONn / 0, d r2 has conformal dimension 2d q n and the field f goes to the
border as x 0d r2y n En f 0 Ž x .. Comparing Eqs. Ž22., Ž40. we notice that the normalizations
of the boundary two-point functions corresponding to the massive n / 0 Dirichlet and
Neumann cases are in general different.
Now we consider the case n s 0 that is m2 s yd 2r4. Following the now usual steps
we get
G Ž dr2 . eyd 1
INns0 s y 2 d r2 Hd d d
x d y En fe Ž x . En fe Ž y . q.... Ž 41.
dp ln e < x y y < d
2
Taking the limit

y1
lim Ž e d r2 ln e . En fe Ž x . s En f 0 Ž x . , Ž 42 .
e™0
and making use of the AdSrCFT equivalence Ž36. we find the following boundary
two-point function:
2 G Ž dr2 . 1
² ONns0 Ž x . ONns0 Ž y . :s 2 d r2
. Ž 43 .
dp < xyy< d
Then the conformal operator ONns0 on the boundary CFT has conformal dimension
dr2. Near the border the scalar field has a logarithmic behavior x 0d r2 ln x 0 En f 0 Ž x ..
Again we find that the normalizations of the boundary two-point functions correspond-
ing to the n s 0 Dirichlet and Neumann cases are in general different.
5. Mixed boundary condition
Using the mixed boundary condition Ž13. and again Eqs. Ž5., Ž7. we get
eyd r2ce Ž k .
aŽ k . s , Ž 44 .
b Ž a , n . q 2 an Kn Ž k e . y a k e Knq1 Ž k e .
where b Ž a , n . is defined as
d
b Ž a ,n . s 1 q a ž 2
yn . / Ž 45 .
Substituting Eq. Ž44. into Eq. Ž7. we find
d dk
E 0 fe Ž x . s d d y ce Ž y .
H H Ž 2p . d
eyi kP Ž xyy . ey1
d Knq1 Ž k e .
qnyke
2 Kn Ž k e .
= . Ž 46 .
Knq1 Ž k e .
b Ž a , n . q 2 an y a k e
Kn Ž k e .
Using this we can write the action Ž15. as
1 d
IM s y Hd x d d y ce Ž x . ce Ž y . eyd
2
d Knq1 Ž k e .
d qnyke
d k 2 Kn Ž k e .
= H Ž 2p . d
eyi kP Ž xyy . . Ž 47 .
Knq1 Ž k e .
Kn Ž k e .
As we shall see it is important to consider the cases b s 0 and b / 0 separately in order

to find out the relevant terms in the series expansions of the Bessel functions.
Let us start with the case b s 0. For b s 0 we have a s y1rŽ 2d y n . and m / 0.
We first consider the massive case with n / 0,dr2. Again we have to study separately
the cases with n not integer and n integer but non-zero. Let us first consider the case n
not integer. Making use of Eq. ŽB.1. with b s 0 we get
d Knq1 Ž k e . d
qnyke s ynq . . . , Ž 48 .
2 Kn Ž k e . 2
and
1
Knq1 Ž k e .
Kn Ž k e .
d
yn
2
sy . Ž 49 .
1 2 1y2 n
G Ž1yn . 2n
Ž ke . y2 Ž ke . q . . .
2Ž 1 y n . G Žn .
Notice that for 0 - n - 1 the dominating term in the denominator of the r.h.s of Eq. Ž49.
is Ž k e . 2 n . Substituting in Eq. Ž47. we get
2
d G Žn .
IMbs0,0 - n - 1 s y2 2 ny2 ž yn / Hd d
x d d y ce Ž x . ce Ž y .
2 G Ž1yn .
d dk
=ey2 nyd H Ž 2p . d
eyi kPŽ xyy. ky2 n q . . . Ž 50.
Integration over k thus yields

d
/ žŽ ./
1 d 2 G yn
2
IMbs0,0 - n - 1 s y ž yn
4p d r2 2 G 1yn
d
d d
ey2Ž nq 2 .
= d x d y ce Ž x . ce Ž y .
H d
q.... Ž 51.
2Ž yn .
< xyy< 2
For n ) 1 the dominating term in the denominator of the r.h.s of Eq. Ž49. is Ž k e . 2 and
Eq. Ž47. reads
2
d
IMbs0, n ) 1 s y Ž n y 1 . ž yn /H d d x d d y ce Ž x . ce Ž y .
2
d dk
=eyd y2 H Ž 2p . d
eyi kPŽ xyy. ky2 q . . . . Ž 52 .
Integration over k is carried out for d ) 2 thus giving

dy2
/ ž /H
d 2 G
2
IMbs0, n ) 1 s y Ž n y 1 . ž yn d r2
d d x d d y ce Ž x . ce Ž y .
2 4p
eyd y2
= dy 2
q.... Ž 53 .
2
< xyy< 2
For the case n integer and non-zero we make use of Eq. ŽB.2.. The logarithmic terms
vanish in the limit e ™ 0 and we find that the same result Ž53. holds for n integer and n
not integer.
Now in the action Ž51. we take the limit
d
lim ey ny 2 ce Ž x . s c 0 Ž x . , Ž 54 .
e™0
whereas in the action Ž53. the limit to be taken is

d
lim ey 2 y1 ce Ž x . s c 0 Ž x . . Ž 55 .
e™0
Using the AdSrCFT equivalence
exp Ž yIAdS . ' exp ¦ žH d d x O Ž x . c0 Ž x . /; , Ž 56 .
we get the following boundary two-point functions:
² OMbs0,0 - n - 1 Ž x . OMbs0,0 - n - 1 Ž y . :
d
/ žŽ ./
1 d 2 G yn
1
2
s
2p d r2 ž 2
yn
G 1yn
< xyy< 2Ž
d
2
yn .
, Ž 57 .
dy2
² OMbs0,n ) 1 Ž x . OMbs0,n ) 1 Ž y . :s Ž n y 1. ž d
yn / ž 2 G
2
d r2
/ 1
dy2
.
2 2p
< xyy< 2 2
Ž 58 .
Then the operators OMbs0,0 - n - 1 and OMbs0, n ) 1 have conformal dimensions y n and d
2
Ž d y 2.r2 respectively. For 0 - n - 1 the field f approaches the boundary as
x 0d r2q nc 0 Ž x .. The derivation of the conformal dimension 2d y n for its associated
boundary operator OMbs0,0 - n - 1 is a rather striking feature. It is worth noting that the
upper constraint n - 1 in Eq. Ž57. is consistent with the unitarity bound.
For n ) 1 we found a boundary operator whose conformal dimension is the unitarity
bound Ž d y 2.r2. Whereas we have already found such a conformal dimension in the
massless Neumann case Ž37. we have here a different normalization for the boundary
two-point function. We note that the behavior of the scalar field for small x 0 is as it
should be.
Now we consider the case n s 0, that is m2 s yd 2r4, keeping still a s y2rd. We
then find
d 2G Ž dr2 . eyd
IMbs0, ns0 s y H d d x d d y ce Ž x . ce Ž y . q.... Ž 59 .
16p d r2 < xyy< d
Taking the limit
lim eyd r2ce Ž x . s c 0 Ž x . , Ž 60 .

e™0
and making use of the AdSrCFT equivalence Ž56. we get the following boundary
two-point function:
d 2G Ž dr2 . 1
² OMbs0,ns0 Ž x . OMbs0,ns0 Ž y . :s d r2
. Ž 61 .
8p < xyy< d
Then the conformal operator OMbs0, ns0 on the boundary CFT has conformal dimension
dr2. Now the field f goes to the border as x 0d r2c 0 Ž x . and no logarithm is present. We
find again that the normalization of the two-point function is different from the
corresponding ones of the Dirichlet and Neumann cases.
Let us now consider the case when b / 0. We study first the case n / 0. Again the
cases n not integer and n integer but non-zero must be considered separately. We first
study the case n not integer. Up to contact terms or higher order terms in e we find
d Knq1 Ž k e . d 2 1y 2 n G Ž 1 y n . 2n
2
qnyke
Kn Ž k e .
s ž 2
yn / 1y
d
yn
G Žn .
Ž ke . q . . . , Ž 62.
2
and
1
Knq1 Ž k e .
Kn Ž k e .
1 2 1y 2 na G Ž 1 y n . 2n
s 1q Ž ke . q . . . . Ž 63.
b Ž a ,n . b Ž a ,n . G Žn .
Substituting in Eq. Ž47. we get

1 1 G Ž1yn .
IMb / 0, n / 0, d r2 s 2n 2 Hd d
2 b Ž a ,n . G Žn .
d dk
=e 2 nyd H Ž 2p . d
eyi kPŽ xyy. k 2 n q . . . . Ž 64 .
Integration over k yields

d
IMb / 0, n / 0, d r2 s y
n
d r2 2
1
G ž2
qn /
p b Ž a ,n . G Žn .
d
e 2Ž ny 2 .
= d d x d d y ce Ž x . ce Ž y .
H d
q.... Ž 65.
< x y y < 2Ž 2 q n .
Consider now the case n integer and non-zero. We find
d Knq1 Ž k e .
qnyke
2 Kn Ž k e .
d n 2 2y 2 n 1 2n
s ž 2
yn / 1 y Ž y1 .
d
yn
2
G Žn .
Ž k e . ln k q . . . , Ž 66.
2
and
1
Knq1 Ž k e .
Kn Ž k e .
1 n 2 2y 2 na 1 2n
s 1 q Ž y1 . 2 Ž k e . ln k q . . . . Ž 67.
b Ž a ,n . b Ž a ,n . G Ž n .
Substituting in Eq. Ž47. we get
n 1 1
IMb / 0, n / 0, d r2 s Ž y1 . 2 1y2 n 2 2 Hd d
b Ž a ,n . G Ž n .
d dk
=e 2 nyd H Ž 2p . d
eyi kPŽ xyy. k 2 n ln k q . . . Ž 68.
Making use of Eq. ŽB.6. we get Eq. Ž65. again. So both cases n integer and n not
integer yield the same result.
Now taking the limit
d
lim e ny 2 ce Ž x . s c 0 Ž x . , Ž 69 .
e™0
and making use of the AdSrCFT correspondence Ž56. we find the following boundary
two-point function:
² OMb / 0,n / 0, d r2 Ž x . OMb / 0 ,n / 0, d r2 Ž y . :

d
s
2n
d r2 2
1
G ž2
qn / 1
d
, Ž 70 .
p b Ž a ,n . G Žn . 2Ž qn .
< xyy< 2
so that the operator OMb / 0, n / 0, d r2 has conformal dimension 2d q n . From Eq. Ž69. we
find that the behavior of f for small x 0 is as expected. Comparing Eqs. Ž22., Ž40., Ž70.
we conclude that the normalizations of the boundary two-point functions corresponding
to the massive n / 0 Dirichlet, Neumann and b / 0 mixed cases are different.
We now consider the case n s 0 that is m 2 s yd 2r4. We find
1 G Ž dr2 . eyd 1
IMb / 0, ns0 s y 2 d r2 Hd d
x d d y ce Ž x . ce Ž y . q....
b Ž a ,0 . 4p ln e < x y y < d
2
Ž 71 .
Taking the limit
y1
lim Ž e d r2 ln e . ce Ž x . s c 0 Ž x . , Ž 72 .
e™0
and making use of the AdSrCFT correspondence Ž56. we find the following boundary
two-point function:
1 G Ž dr2 . 1
² OMb / 0,ns0 Ž x . OMb / 0,ns0 Ž y . :s 2 d r2
, Ž 73 .
b Ž a ,0 . 2p < xyy< d
so that the conformal operator OMb / 0, ns0 on the boundary CFT has conformal dimen-
sion dr2. For small x 0 we find a logarithmic behavior x 0d r2 ln x 0 c 0 Ž x .. Again the
normalization of the two-point function is different when compared to the corresponding
ones of the Dirichlet, Neumann and b s 0 mixed cases.
In the massless case we have n s dr2. For d odd we make use of Eqs. ŽB.1., ŽB.5.,
whereas for d even we use Eqs. ŽB.2., ŽB.6.. In both cases we get
d G Ž d. 1
IMnsd r2 s y d r2 Hd d
x d d y ce Ž x . ce Ž y . q.... Ž 74.
2p G Ž dr2 . < xyy< 2 d
Taking the limit
lim ce Ž x . s c 0 Ž x . , Ž 75 .
e™0
and making use of the AdSrCFT equivalence Ž56. we get the following boundary
two-point function:
d G Ž d. 1
² OMnsd r2 Ž x . OMnsd r2 Ž y . :s d r2
. Ž 76 .
p G Ž dr2 . < x y y < 2 d
Then the operator OMnsd r2 has conformal dimension d. The scalar field goes to the
border as c 0 Ž x . as expected. Comparing Eqs. Ž23., Ž76. we conclude that the boundary
CFT’s corresponding to the massless Dirichlet and mixed cases are equal.
6. Conclusions
We have shown how different boundary conditions in the AdSrCFT correspondence

allow us to derive boundary two-point functions which are consistent with the unitarity
bound. We have also done a careful analysis of the particular cases when n is an integer
and we have shown that when n s 0 there are no double zeroes in the two-point
functions.
In general the use of different types of boundary conditions lead to boundary
two-point functions with different normalizations. It is not clear at this point whether
they are important or not. It is necessary to compute the three-point functions in order to
clarify if the different normalizations are leading to different boundary CFT’s.
In the Neumann case the unitarity bound is obtained for m s 0 while with mixed
boundary conditions it is reached when b s 0 and m 2 ) yd 2r4 q 1. The corresponding
two-point functions have different normalizations. The conformal dimension dr2 y n is
obtained in the case of mixed boundary condition with b s 0 and yd 2r4 - m2 -
yd 2r4 q 1, and the normalization of the corresponding boundary two-point function
differs from the one found in w23x.
We have also tried to relate our formalism to the Legendre transform approach w23x.
We could think that both formalisms are related through some field redefinition but this
is not the case. It is not possible to redefine the scalar field in order to turn a Dirichlet
boundary condition into a Neumann or mixed one. If there is any relation between the
two approaches it must be a very subtle one.
Another important point is the interpretation of the new boundary conditions in the
string theory context. Dirichlet boundary conditions are natural when thinking of the
asymptotic behavior of the supergravity fields reaching the border of the AdS space.
Possibly Neumann and mixed boundary conditions are related to more complex solu-
tions involving strings and membranes reaching the border in more subtle ways. This of
course needs a more detailed study.
Acknowledgements
P.M. acknowledges financial support by CAPES. V.O.R. is partially supported by

CNPq and acknowledges a grant by FAPESP.
Appendix A. Boundary two-point functions
The coefficients n , a and b Ž a , n . are defined in Eqs. Ž6., Ž13., Ž45. respectively.
Let us also define
d
s Žn . s yn . Ž A.1 .
2
A.1. Dirichlet boundary condition
d
2n
G
² ODn / 0 Ž x . ODn / 0 Ž y . :s d r2 2
ž qn / 1
d Ž A.2 .
p G Žn .
< x y y < 2Ž 2 q n .
d G Ž d. 1
² ODnsd r2 Ž x . ODnsd r2 Ž y . :s d r2 Ž A.3 .
p G Ž dr2 . < x y y < 2 d
G Ž dr2 . 1
² ODns0 Ž x . ODns0 Ž y . :s d r2 Ž A.4 .
2p < xyy< d
A.2. Neumann boundary condition
1
² ONn / 0, d r2 Ž x . ONn / 0, d r2 Ž y . :s 2
² ODn / 0 Ž x . ODn / 0 Ž y . : Ž A.5 .
s Žn .
G Ž dr2 . 1
² ONnsd r2 Ž x . ONnsd r2 Ž y . :s d r2 dy2
, Ž A.6 .
2p
< xyy< 2 2
1
² ONns0 Ž x . ONns0 Ž y . :s 2
² ODns0 Ž x . ODns0 Ž y . : Ž A.7 .
s Ž 0.
A.3. Mixed boundary condition
² OMbs0,0 - n - 1 Ž x . OMbs0,0 - n - 1 Ž y . :
d
ss Žn .2
1
d r2
G
2žyn / 1
d
, Ž A.8 .
2p G Ž1yn .
< x y y < 2Ž 2 y n .
dy2
² OMbs0,n ) 1 Ž x . OMbs0,n ) 1 Ž y . :s s 2 Ž n . Ž n y 1.
G ž 2
d r2
/ 1
dy2
,
2p
< xyy< 2 2
Ž A.9 .
1
² OMb / 0,n / 0, d r2 Ž x . OMb / 0 ,n / 0, d r2 Ž y . :s 2
² ODn / 0 Ž x . ODn / 0 Ž y . :,
b Ž a ,n .
Ž A.10 .
² OMnsd r2 Ž x . OMnsd r2 Ž y . :s² ODnsd r2 Ž x . ODnsd r2 Ž y . :, Ž A.11 .
1
² OMb / 0,ns0 Ž x . OMb / 0,ns0 Ž y . :s 2
² ODns0 Ž x . ODns0 Ž y . :, Ž A.12 .
b Ž a ,0 .
² OMbs0,ns0 Ž x . OMbs0,ns0 Ž y . :s s 2 Ž 0. ² ODns0 Ž x . ODns0 Ž y . : Ž A.13 .

Appendix B. Some useful formulae
B.1. Series expansions for the modified Bessel functions Kn
For n not integer

2n
1 yn Ž zr2.
Kn Ž z . s G Ž n . G Ž 1 y n . Ž zr2 . Ý
2 n00 n! G Ž n q 1 y n .
2n
2n Ž zr2.
y Ž zr2 . Ý . Ž B.1 .
n00 n! G Ž n q 1 q n .
For n integer and non-zero

1 ny1 G Ž n y n.
yn n 2n n n
Kn Ž z . s Ž zr2. Ý Ž y1. Ž zr2. y Ž y1. Ž zr2.
2 ns0 n!
2n
lŽ n q 1. q lŽ n q n q 1. Ž zr2.
= Ý ln Ž zr2 . y ,
n00 2 n! G Ž n q 1 q n .
Ž B.2 .
where
ny1 1
lŽ 1 . s yg , lŽ n . s yg q Ý , Ž n 0 2. , Ž B.3 .
ms1 m
and g is the Euler constant.
For n s 0
2n
Ž zr2.
K0Ž z . sy Ý ln Ž zr2 . y l Ž n q 1 . . Ž B.4 .
n00 n! G Ž n q 1 .
B.2. Integration oÕer the momenta
d dk 1
H Ž 2p . eyi kP x k r s Cr , r / yd,y d y 2, . . . , Ž B.5 .
d
< x < dq r
d dk dCr 1 ln < x <
H Ž 2p . eyi kP x k r ln k s dq r
q Cr , r / yd,y d y 2, . . . ,
d
dr < x< < x < dq r
Ž B.6 .
where
dqr
Cr s
2 r
d r2
G ž 2
r
/ . Ž B.7 .
p G y ž /
2
References

w2x S. Gubser, I. Klebanov, A. Polyakov, Phys. Lett. B 428 Ž1998. 105. hep-thr9802109.
w4x ¨ K.S. Viswanathan, Phys. Rev. D 58 Ž1998. 41901. hep-thr9804035.
W. Muck,
w5x D.Z. Freedman, S.D. Mathur, A. Matusis, L. Rastelli, Nucl. Phys. B 546 Ž1999. 96. hep-thr9804058.
w6x V. Balasubramanian, P. Kraus, A. Lawrence, Phys. Rev. D 59 Ž1999. 046003. hep-thr9805171.
w7x ¨ K.S. Viswanathan, Phys. Rev. D 58 Ž1998. 106006. hep-thr9805145.
W. Muck,
w8x P. Minces, V.O. Rivelles, Phys. Lett. B 455 Ž1999. 147. hep-thr9902123.
w9x M. Henningson, K. Sfetsos, Phys. Lett. B 431 Ž1998. 63. hep-thr9803251.
w10x A. Ghezelbash, K. Kaviani, S. Parvizi, A. Fatollahi, Phys. Lett. B 435 Ž1998. 291. hep-thr9805162.
w11x A. Volovich, J. High En. Phys. 9809 Ž1998. 22. hep-thr9809009.
w12x A. Koshelev, O. Rytchkov, Phys. Lett. B 450 Ž1999. 368. hep-thr9812238.
w13x P. Matlock, K.S. Viswanathan, The AdSrCFT correspondence for the massive Rarita–Schwinger field,
hep-thr9906077.
w14x H. Liu, A. Tseytlin, Nucl. Phys. B 533 Ž1998. 88. hep-thr9804083.
w15x ¨ K.S. Viswanathan, The graviton in the AdS–CFT correspondence: solution via the Dirichlet
W. Muck,
boundary value problem, hep-thr9810151.
w16x A. Polishchuk, J. High Energy Phys. 9907 Ž1999. 007. hep-thr9905048.
w17x G. Arutyunov, S. Frolov, Phys. Lett. B 441 Ž1998. 173. hep-thr9807046.
w18x W. l’Yi, Phys. Lett. B 448 Ž1999. 218. hep-thr9809132.
w19x L. Susskind, J. Math. Phys. 36 Ž1995. 6377. hep-thr9409089.
w20x G.’tHooft, Dimensional reduction in quantum gravity, gr-qcr9310026.
w21x S. Deser, R. Jackiw, Phys. Lett. B 139 Ž1984. 371.
w22x P. Breitenlohner, D. Freedman, Ann. Phys. 144 Ž1982. 249.
w23x I. Klebanov, E. Witten, AdSrCFT correspondence and symmetry breaking, hep-thr9905104.
w24x V.K. Dobrev, Intertwining operator realization of the AdSrCFT correspondence, hep-thr9812194, to
appear in Nucl. Phys. B.
w25x M. Henneaux, Boundary terms in the AdSrCFT correspondence for spinor fields, hep-thr9902137, to
appear in the Proc. of the Int. Workshop ISMP ŽTbilissi, September 1998..
w26x S. Lee, S. Minwalla, M. Rangamani, N. Seiberg, Adv. Theor. Math. Phys. 2 Ž1998. 697. hep-thr9806074.
Cumulative Author Index B571–B572
Ahn, C. B572 Ž2000. 188 Frixione, S. B571 Ž2000. 169

Alexandrou, C. B571 Ž2000. 257 Fukae, M. B572 Ž2000. 71
Alford, M. B571 Ž2000. 269
ALPHA Collaborations B571 Ž2000. 237 Gaillard, M.K. B571 Ž2000. 3
Anastasiou, C. B572 Ž2000. 307 Ganchev, Ch.A. B571 Ž2000. 457
Angelantonj, C. B572 Ž2000. 36 Garden, J. B571 Ž2000. 237
Antoniadis, I. B572 Ž2000. 36 Glover, E.W.N. B572 Ž2000. 307
Gorsky, A. B571 Ž2000. 120
Bak, D. B572 Ž2000. 151 Greco, M. B571 Ž2000. 137
Ball, P. B572 Ž2000. 3 Guillet, J.Ph. B572 Ž2000. 361
Berche, B. B572 Ž2000. 626 ¨
Gunaydin, M. B572 Ž2000. 131
Berges, J. B571 Ž2000. 269 Gyulassy, M. B571 Ž2000. 197
Binoth, T. B572 Ž2000. 361
Blas Achic, H.S. B571 Ž2000. 607 Hara, Y. B572 Ž2000. 574
Bouwknegt, P. B572 Ž2000. 547 Hart, A. B572 Ž2000. 243
Buchbinder, I.L. B571 Ž2000. 358 Hastings, M.B. B572 Ž2000. 535
Hebecker, A. B571 Ž2000. 26
Cacciari, M. B571 Ž2000. 185 Heinrich, G. B572 Ž2000. 361
Chatelain, C. B572 Ž2000. 626 Heitger, J. B571 Ž2000. 237
Chim, L. B572 Ž2000. 547 Hoppe, J. B571 Ž2000. 479
Controzzi, D. B572 Ž2000. 609 Howe, P.S. B571 Ž2000. 71
ˇ M.
Cvetic, B571 Ž2000. 358 Hsu, S.D.H. B572 Ž2000. 211
Damgaard, P.H. B572 Ž2000. 478 Jegerlehner, F. B571 Ž2000. 511

D’Appollonio, G. B572 Ž2000. 36
Dasgupta, T. B572 Ž2000. 95 Kabat, D. B571 Ž2000. 419
Del Duca, V. B571 Ž2000. 51 Kac, V.G. B571 Ž2000. 515
Diakonov, D. B571 Ž2000. 91 Kazakov, V. B571 Ž2000. 479
Dixon, L. B571 Ž2000. 51 Komori, Y. B571 Ž2000. 632
Dorey, P. B571 Ž2000. 583 Kostov, I.K. B571 Ž2000. 479
Dudas, E. B572 Ž2000. 36 ¨
Kramer, M. B571 Ž2000. 169
Engels, J. B572 Ž2000. 289 Laenen, E. B571 Ž2000. 169

Ermolaev, B.I. B571 Ž2000. 137 Langfeld, K. B572 Ž2000. 266
´
Levai, P. B571 Ž2000. 197
Ferreira, L.A. B571 Ž2000. 607 Lifschytz, G. B571 Ž2000. 419
Fleischer, J. B571 Ž2000. 511 Lima, E. B572 Ž2000. 112
` J.-M.
Frere, B572 Ž2000. 3 ¨ H.
Lu, B572 Ž2000. 112
672 CumulatiÕe Author Index B571–B572
Maltoni, F. B571 Ž2000. 51 Selivanov, K. B571 Ž2000. 120

´ C.P.
Martın, B572 Ž2000. 387 Shin, G. B572 Ž2000. 266
Matias, J. B572 Ž2000. 3 Shore, G.M. B571 Ž2000. 287
Maul, M. B571 Ž2000. 91 Sierra, G. B572 Ž2000. 517
Meggiolaro, E. B571 Ž2000. 26 Smilga, A.V. B571 Ž2000. 515
Mendes, T. B572 Ž2000. 289 Sokatchev, E. B571 Ž2000. 71
Minasian, R. B572 Ž2000. 499 Sommer, R. B571 Ž2000. 237
Minces, P. B572 Ž2000. 651 Splittorff, K. B572 Ž2000. 478
Mueller, A.H. B572 Ž2000. 227 ´
Stefanski, Jr. B. B572 Ž2000. 95
Nachtmann, O. B571 Ž2000. 26 Tarasov, O.V. B571 Ž2000. 511

Nauta, B.J. B571 Ž2000. 151 Tateo, R. B571 Ž2000. 583
Nelson, B.D. B571 Ž2000. 3 Teschner, J. B571 Ž2000. 555
Nishino, A. B571 Ž2000. 632 Troyan, S.I. B571 Ž2000. 137
Tsimpis, D. B572 Ž2000. 499
Oleari, C. B572 Ž2000. 307 Tsvelik, A.M. B572 Ž2000. 609
Osborn, H. B571 Ž2000. 287
Ovrut, B.A. B572 Ž2000. 112 Ujino, H. B571 Ž2000. 632
UKQCD Collaborations B571 Ž2000. 237
Panagopoulos, H. B571 Ž2000. 257
Petkova, V.B. B571 Ž2000. 457
Van Weert, Ch.G. B571 Ž2000. 151
Petrov, A.Yu. B571 Ž2000. 358
Veretin, O.L. B571 Ž2000. 511
Philipsen, O. B572 Ž2000. 243
Vicari, E. B571 Ž2000. 257
Pope, C.N. B572 Ž2000. 112
Vitev, I. B571 Ž2000. 197
Rajagopal, K. B571 Ž2000. 269
Rey, S.-J. B572 Ž2000. 151 Wadati, M. B571 Ž2000. 632
Rey, S.-J. B572 Ž2000. 188 Watts, G.M.T. B571 Ž2000. 457
Ridout, D. B572 Ž2000. 547 West, P.C. B571 Ž2000. 71
Rivelles, V.O. B572 Ž2000. 651 Wittig, H. B571 Ž2000. 237
Sagnotti, A. B572 Ž2000. 36 Yamada, Y. B572 Ž2000. 71

´
Sanchez-Ruiz, D. B572 Ž2000. 387 Yang, S.-K. B572 Ž2000. 71
Schubert, C. B571 Ž2000. 71
Schwetz, M. B572 Ž2000. 211 Zagermann, M. B572 Ž2000. 131

Nucl - Phys.B v.572 PDF

Uploaded by

Copyright:

Available Formats

You might also like

Nucl - Phys.B v.572 PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Nucl - Phys.B v.572 PDF

Uploaded by

Copyright:

Available Formats

Nuclear Physics B 572 Ž2000.

Anatomy of mixing-induced CP asymmetries in left–right

In this paper we investigate in numerical detail an attractive extension of the standard

2. Definition of the model

The quark mass matrices read

and acquire the VEVs

The spontaneous breakdown of SUŽ2.L = SUŽ2.R = UŽ1. to UŽ1.em generates the

The Yukawa interaction part of the Lagrangian reads

Ž VLU . 11 e 2 i a 1 Ž VLU . 12 e i Ž a 1q a 2q e 1 . Ž VLU . 13 e i Ž a 1q a 3q e 1q e 2 .

I and class II solutions, respectively. This induces another twofold multiplicity in

c12 c13 s12 c13 s13 eyi d

with, from the experimental results, Eq. Ž14.,

In Fig. 1 we plot the independent phases as functions of b in the allowed range

4. The SB–LR parameter space and strategies for constraints

– M2 ; O Ž1 TeV., the mass of the predominantly right-handed weak gauge

limit of the SB–LR model experimentally by measuring d from D m B d or e . To date, the

– the suppression factor Ž mW rM2 . 2 is partially compensated by large matching

– the ‘‘natural’’ region with r ; O Ž m brm t . ; 0.02, which implies z ; O Ž10y4 .;

As for CP conserving quantities, constraints can be derived from D m, the mass

5. B 0 –B 0 mixing in the SB–LR

5.1. Constraints from D m B d and Predictions for D m B s

=² B 0 < Ž db . Vy A Ž db . VyA Ž m . < B 0 : . Ž 24 .

mtŽ mt . h 2B,LO Ž m b . hˆ 2B,LO hˆ 2B,NLO BB Ž m b . BˆBNLO (

and the Inami–Lim function S is defined as

On the other hand, the matrix element of OS can be written as9

h1L R Ž m b . f 1.8, h 2L R Ž m b . f 1.7. Ž 29 .

Numerically, <4 F1 < < < F2 <. We thus approximate

function of only two variables, k and b , instead of expressing k in terms of M2 and

which roughly translates into

s Ž 1.31 " 0.19 . 2

which translates into the bound

the SB–LR predicts values close to 0 or p ; this result depends, however, on

5.2. Constraints from Bd0 ™ J r c K S0

Other interesting constraints can be obtained from the measurement of the CP

the SM model expectation is sin2 bCK M s 0.73y0.06 5 .

6. Constraints from the K system

generic CKM factor Ž l iA B . 2 , A, B g  L, R4 , i g  u,c,t 4 , whereas a0 contains the factor

Fig. 8. Allowed values for M2 and MH from the K-physics constraints.

7. Combining constraints from K and B system

Combining all the constraints from D m K , D m B d , D m B s, e , e X and sin2 bCKM eff

– a lower bound from D mLR

Fig. 10. Allowed region in Ž M2 , MH ., taking into account all constraints.

As for b , we find that

8. Summary and discussion

P.B. is supported by DFG through a Heisenberg fellowship. J.M.F. gratefully

Appendix A. Non-factorizable contributions to BSB r B B

As factorization is exact in the large-Nc limit, BB becomes scale-independent and it is

Unfortunately, both D B in the numerator and denominator are scale-dependent so that

with b 0 s 11 y 2r3 P 4, g B S s y8 and g B s 2; s0 f 34 GeV 2 is the continuum thresh-

w1x ` et al., Phys. Lett. B 314 Ž1993. 289.

Type I vacua with brane supersymmetry breaking

Received 12 November 1999; accepted 21 January 2000

2. Z 2 = Z 2 orientifolds and discrete torsion

g : Ž q,y,y . , f : Ž y,q,y . , h: Ž y,y,q . . Ž 2.1 .

where the 16 Z 2 = Z 2 characters t k l w16,17x, combinations of products of level-one

where we have resorted to a compact notation, used extensively in the following:

for the NS portion, and

2.1. Massless spectra

generic CKM factor Ž l iA B . 2 , A, B g L, R4 , i g u,c,t 4 , whereas a0 contains the factor