Nucl - Phys.B v.791

Nuclear Physics B 791 (2008) 119
Higgs boson production at the LHC:

Transverse-momentum resummation and rapidity
dependence
Giuseppe Bozzi a , Stefano Catani b , Daniel de Florian c ,
Massimiliano Grazzini b,
a Institut fr Theoretische Physik, Universitt Karlsruhe, PO Box 6980, D-76128 Karlsruhe, Germany
b INFN, Sezione di Firenze and Dipartimento di Fisica, Universit di Firenze, I-50019 Sesto Fiorentino, Florence, Italy
c Departamento de Fsica, Facultad de Ciencias Exactas y Naturales, Universidad de Buenos Aires,
(1428) Pabelln 1 Ciudad Universitaria, Capital Federal, Argentina

Received 30 May 2007; accepted 24 September 2007
Available online 10 October 2007
This paper is dedicated to the memory of Jiro Kodaira, great friend and distinguished colleague
Abstract
We consider Higgs boson production by gluon fusion in hadron collisions. We study the doublydifferential transverse-momentum (qT ) and rapidity (y) distribution of the Higgs boson in perturbative
QCD. In the region of small qT (qT MH , MH being the mass of the Higgs boson), we include the effect of logarithmically-enhanced contributions due to multiparton radiation to all perturbative orders. We
use the impact parameter and double Mellin moments to implement and factorize the multiparton kinematics constraint of transverse- and longitudinal-momentum conservation. The logarithmic terms are then
systematically resummed in exponential form. At small qT , we perform the all-order resummation of large
logarithms up to next-to-next-to-leading logarithmic accuracy, while at large qT (qT MH ), we apply
a matching procedure that recovers the fixed-order perturbation theory up to next-to-leading order. We
present quantitative results for the differential cross section in qT and y at the LHC, and we comment on
the comparison with the qT cross section integrated over y.
2007 Elsevier B.V. All rights reserved.
PACS: 12.38.Cy; 14.80.Bn
Keywords: Higgs; QCD; Hadronic colliders; LHC
* Corresponding author.
E-mail address: massimiliano.grazzini@cern.ch (M. Grazzini).

0550-3213/$ see front matter 2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.nuclphysb.2007.09.034
G. Bozzi et al. / Nuclear Physics B 791 (2008) 119
1. Introduction
The search for the Higgs boson [1] and the study of its properties (mass, couplings, decay
widths) at hadron colliders require a detailed understanding of its production mechanisms. This
demands reliable computations of related quantities, such as production cross sections and the
associated distributions in rapidity and transverse momentum. In this paper we consider the production of the Standard Model (SM) Higgs boson by the gluon fusion mechanism.
The gluon fusion process gg H , through a heavy-quark (mainly, top-quark) loop, is the
main production mechanism of the SM Higgs boson H at hadron colliders. When combined
with the decay channels H and H ZZ, this production mechanism is one of the most
important for Higgs boson searches and studies over the entire range, 100 GeV MH 1 TeV,
of Higgs boson mass MH to be investigated at the LHC [2]. In the mass range 140 GeV
MH 180 GeV, the gluon fusion process, followed by the decay H W W + , can
be exploited as main discovery channel at the LHC and also at the Tevatron [3], provided the
background from t t production is suppressed by applying a veto cut on the transverse momenta
of the jets accompanying the final-state leptons.
The dynamics of the gluon fusion mechanism is controlled by strong interactions. Detailed
studies of the effect of QCD radiative corrections are thus necessary to obtain accurate theoretical
predictions.
In QCD perturbation theory, the leading order (LO) contribution to the total cross section
for Higgs boson production by gluon fusion is proportional to S2 , S being the QCD coupling.
The QCD radiative corrections to the total cross section are known at the next-to-leading order
(NLO) [47] and at the next-to-next-to-leading order (NNLO) [812]. The Higgs boson rapidity
distribution is also known at the NLO [13] and at the NNLO [14,15]. The effects of a jet veto
have been studied up to the NNLO [11,14,15]. We recall that all the results at NNLO have been
obtained by using the large-Mt approximation, Mt being the mass of the top quark. This approximation is justified by the fact that the bulk of the QCD radiative corrections to the total cross
section is due to virtual and soft-gluon contributions [911,16,17]. The soft-gluon dominance
also implies that higher-order perturbative contributions can reliably be estimated by applying
resummation methods [9] of threshold logarithms, a type of logarithmically-enhanced terms due
to multiple soft-gluon emission. In Ref. [17], the NNLO calculation of the total cross section
is supplemented with threshold resummation at the next-to-next-to-leading logarithmic (NNLL)
level; the residual perturbative uncertainty at the LHC is estimated to be at the level of better
than 10%. The NNLL + NNLO results [17] are nicely confirmed by the more recent computation [1820] of the soft-gluon terms at N3 LO; the quantitative effect [18] of the additional
(i.e., beyond the NNLL order) single-logarithmic term at N3 LO is consistent with the estimated
uncertainty at NNLL + NNLO. The effect of threshold logarithms on the rapidity distribution of
the Higgs boson has been considered in Ref. [21].
The gluon fusion mechanism at O(S2 ) produces a Higgs boson with a vanishing transverse
momentum qT . A large (or, however, non-vanishing) value of qT can be obtained only starting
from O(S3 ), when the Higgs boson is accompanied by at least one recoiling parton in the final
state. This mismatch by a power of S is a preliminary indication of the fact that the small-qT
and large-qT regions are controlled by different dynamics regimes.
The large-qT region is identified by the condition qT MH . In this region, the perturba2 ), and calculations based on the
tive series is controlled by a small expansion parameter, S (MH
truncation of the series at a fixed order in S are theoretically justified. The LO, i.e., O(S3 ), calculation is reported in Ref. [22]. The results of Ref. [22] and the higher-order studies of Refs. [23,
24] show that the large-Mt approximation is sufficiently accurate also in the case of the qT distribution when qT MH , provided qT Mt . Using the large-Mt approximation, the NLO QCD
computation of the qT distribution of the SM Higgs boson is presented in Refs. [14,15,2527].
QCD corrections beyond the NLO are evaluated in Ref. [28], by implementing threshold resummation at the next-to-leading logarithmic (NLL) level. The results of the numerical programs of
Refs. [14,15] can also be safely (i.e., without encountering infrared divergences) extended from
large values of qT to qT = 0: in the small-qT region these programs evaluate the qT distribution
up to NNLO.
In the small-qT region (qT MH ), where the bulk of events is produced, the convergence
of the fixed-order expansion is definitely spoiled, since the coefficients of the perturbative series
2 ) are enhanced by powers of large logarithmic terms, lnm (M 2 /q 2 ). The logarithmic
in S (MH
H T
terms are produced by multiple emission of soft and collinear partons (i.e., partons with low
transverse momentum). To obtain reliable perturbative predictions, these terms have to be resummed to all orders in S . The method to systematically perform all-order resummation of
classes of logarithmically-enhanced terms at small qT is known [2936]. In the case of the SM
Higgs boson, resummation has been explicitly worked out at leading logarithmic (LL), NLL [35,
37] and NNLL [38] level.
The fixed-order and resummed approaches at small and large values of qT can then be matched
at intermediate values of qT , to obtain QCD predictions for the entire range of transverse momenta. Phenomenological studies of the SM Higgs boson qT distribution at the LHC have been
performed in Refs. [3945], by combining resummed and fixed-order perturbation theory at different levels of theoretical accuracy. Other recent studies of various kinematical distributions of
the SM Higgs boson at the LHC are presented in Refs. [4650].
In Refs. [41,44] we studied the Higgs boson qT distribution integrated over the rapidity. In
the small-qT region, the logarithmic terms were systematically resummed in exponential form
by working in impact-parameter and Mellin-moment space. A constraint of perturbative unitarity was imposed on the resummed terms, to the purpose of reducing the effect of unjustified
higher-order contributions at large values of qT and, especially, at intermediate values of qT .
This constraint thus decreases the uncertainty in the matching procedure of the resummed and
fixed-order contributions. Our best theoretical predictions were obtained by matching NNLL resummation at small qT and NLO perturbation theory at large qT . NNLL resummation includes
the complete NNLO result at small qT , and the unitarity constraint assures that the total cross
section at NNLO is recovered upon integration over qT of the transverse-momentum spectrum.
Considering SM Higgs boson production at the LHC, we concluded [44] that the residual perturbative QCD uncertainty of the NNLL + NLO result is uniformly of about 10% from small to
intermediate values of transverse momenta.
In this paper we extend our study to include the dependence on the rapidity of the Higgs
boson. Using the impact parameter and double Mellin moments, we can perform the extension
by maintaining all the main features of the resummation formalism of Refs. [36,44]. We are then
able to present results up to NNLL + NLO accuracy for the doubly-differential cross section
in qT and rapidity at the LHC.
The paper is organized as follows. In Section 2 we recall the main aspects of the resummation
formalism, and we illustrate the steps that are necessary to include the dependence on the rapidity
in the qT resummed formulae. In Section 3 we apply the formalism to the production of the SM
Higgs boson at the LHC, and we perform quantitative studies on the qT and rapidity dependence
of the doubly-differential cross section. Some concluding remarks are presented in Section 4.
Additional technical details on the double Mellin moments of the resummation formulae are
given in Appendix A.
2. Rapidity dependence in qT resummation
We consider the inclusive hard-scattering process
h1 (p1 ) + h2 (p2 ) H (y, qT , MH ) + X,
(1)
where the collision of the two hadrons h1 and h2 with momenta p1 and p2 produces the Higgs
boson H , accompanied by an arbitrary
and undetected final state X. The centre-of-mass energy
of the colliding hadrons is denoted by s. The rapidity, y, of the Higgs boson is defined in the
centre-of-mass frame of the colliding hadrons, and the forward direction (y > 0) is identified by
the direction of the momentum p1 .
According to the QCD factorization theorem, the doubly-differential cross section for this
process is

d
(y,
q
,
M
,
s)
=
dx
dx2 fa1 / h1 x1 , 2F fa2 / h2 x2 , 2F
T
H
1
2
dy dqT
a1 ,a2
1

d a1 a2
y,
qT , MH , s ; S 2R , 2R , 2F ,
2
d y dqT
(2)
where fa/ h (x, 2F ) (a = qf , qf , g) are the parton densities of the colliding hadrons at the factorization scale F , d ab are the partonic cross sections, and R is the renormalization scale.
Throughout the paper we use parton densities as defined in the MS factorization scheme, and
S (q 2 ) is the QCD running coupling in the MS renormalization scheme. The rapidity, y,
and the
centre-of-mass energy, s , of the partonic cross section (subprocess) are related to the corresponding hadronic variables y and s:
y = y
1 x1
ln ,
2 x2
s = x1 x2 s,
(3)

with the kinematical boundary |y|
< ln s /M 2 (|y| < ln s/M 2 ) and s > M 2 (s > M 2 ).
The partonic cross section d ab is computable in QCD perturbation theory. Its power series
2 /q 2 ), that we
expansion in S contains the logarithmically-enhanced terms, (Sn /qT2 ) lnm (MH
T
want to resum. To this purpose, we use the general (process-independent) strategy and the formalism described in detail in Ref. [44]. The only difference with respect to Ref. [44] is that the
resummation is now performed at fixed values of the rapidity y, rather than after integration over
the rapidity phase space. In the following we briefly recall the main steps of the resummation
formalism, and we point out explicitly the differences with respect to Ref. [44].
We first rewrite (see Section 2.1 in Ref. [44]) the partonic cross section as the sum of two
terms,
(res.)
(fin.)
d a1 a2
d a1 a2
d a1 a2
=
+
.
d y dqT2
d y dqT2
d y dqT2
(4)
(res.)
The logarithmically-enhanced contributions are embodied in the resummed component d a1 a2 .

The finite component d a(fin.)
1 a2 is free of such contributions, and it can be computed by truncation of the perturbative series at a given fixed order (LO, NLO and so forth). In practice, after
having evaluated d a1 a2 and its resummed component at a given perturbative order, the finite
(fin.)
component d a1 a2 is obtained by the matching procedure described in Sections 2.1 and 2.4 of
Ref. [44].
The resummation procedure of the logarithmic terms has to be carried out [3034] in the
impact-parameter space, to correctly take into account the kinematics constraint of transversemomentum conservation. The resummed component of the partonic cross section is then obtained
by performing the inverse Fourier (Bessel) transformation with respect to the impact parameter b.
We write1
(res.)
2
MH
d a1 a2
b
(
y,
q
,
M
,
s
)
=
db J0 (bqT )Wa1 a2 (y,
b, MH , s ; S ),
(5)
T
H
S
s
2
d y dqT2
0
where J0 (x) is the 0th-order Bessel function, and the factor W embodies the all-order dependence on the large logarithms ln(MH b)2 at large b, which correspond to the qT -space terms
2 /q 2 ) (the limit q M corresponds to M b 1, since b is the variable conjugate
ln(MH
T
H
H
T
to qT ).
In the case of the qT cross section integrated over the rapidity, the resummation of the large
logarithms is better expressed [36,44] by defining the N -moments WN of W with respect to
2 /
z = MH
s at fixed MH . In the present case, where the rapidity is fixed, it is convenient (see,
e.g., Refs. [51,52])
to consider double
(N1 , N2 )-moments with respect to the two variables

z1 = e+y MH / s and z2 = ey MH / s at fixed MH (note that 0 < zi < 1). We thus introduce
W (N1 ,N2 ) as follows:
1
1 ,N2 ) (b, M ; ) =
Wa(N
H
S
1 a2
dz1 z1N1 1
1
dz2 z2N2 1 Wa1 a2 (y,

b, MH , s ; S ).
(6)
More generally, any function h(y; z) of the variables y (|y| < ln z ) and z (0 < z < 1) can
be considered as a function of the two variables z1 = e+y z and z2 = ey z. Thus, throughout

the paper, the (N1 , N2 )-moments h(N1 ,N2 ) of the function h(y; z) are defined as
1
(N1 ,N2 )
dz1 z1N1 1
1
dz2 z2N2 1 h(y; z),
where y =
1 z1
ln , z = z1 z2 .
2 z2
(7)
Note that the double Mellin moments can also be obtained (see, e.g., Ref. [53]) by introducing
a Fourier transformation with respect to y (with conjugate variable = i(N2 N1 )) and then
performing a Mellin transformation with respect to z (with conjugate variable N = (N1 +N2 )/2):
1
(N1 ,N2 )
dz z
N1
+
dy eiy h(y; z),
where N1 = N + i/2, N2 = N i/2.
(8)
The convolution structure of the QCD factorization formula (2) is readily diagonalized by
considering (N1 , N2 )-moments:

1 ,N2 ) ,
fa1 / h1 ,N1 +1 fa2 / h2 ,N2 +1 d a(N
d (N1 ,N2 ) =
(9)
1 a2
a1 ,a2
1 In the following equations, the functional dependence on the scales and is understood.
R
F
1
where fa/ h,N = 0 dx x N1 fa/ h (x) are the customary N -moments of the parton distributions.
The use of Mellin moments also simplifies the resummation structure of the logarithmic terms
(res.)(N ,N )
(N ,N )
in d a1 a2 1 2 . The perturbative factor Wa1 a12 2 can indeed be organized in exponential form
as follows:

,
W (N1 ,N2 ) (b, MH ; S ) = H(N1 ,N2 ) (MH , S ) exp G (N1 ,N2 ) (S , L)
(10)
where

2 2
MH b
+
1
,
L = ln
b02
(11)
b0 = 2eE (E = 0.5772 . . . is the Euler number) and, to simplify the notation, the dependence
on the flavour indices has been understood.
The structure of Eq. (10) is in close analogy to the cases of soft-gluon resummed calculations
for hadronic event shapes in hard-scattering processes [54] and for threshold contributions to
hadronic cross sections [51,55,56]. The function H(N1 ,N2 ) (which is process dependent) does not
depend on the impact parameter b and, therefore, its evaluation does not require resummation of
large logarithmic terms. It can be expanded in powers of S as

2

S
S
H(N1 ,N2 ) (MH , S ) = 0 (S , MH ) 1 + H(N1 ,N2 )(1) +
H(N1 ,N2 )(2) + , (12)
where 0 (S , MH ) is the lowest-order partonic cross section for Higgs boson production. The
form factor exp{G} is process independent2 ; it includes the complete dependence on b and, in
particular, it contains all the terms that order-by-order in S are logarithmically divergent when
b . The functional dependence on b is expressed through the large logarithmic terms Sn L m
with 1 m 2n. More importantly, all the logarithmic contributions to G with n + 2 m 2n
are vanishing. Thus, the exponent G can systematically be expanded in powers of S , at fixed
as follows:
value of = S L,
= Lg
(1) (S L)
+ g (2)(N1 ,N2 ) (S L)
+ S g (3)(N1 ,N2 ) (S L)
+ .
G (N1 ,N2 ) (S , L)
(13)
(1) collects the leading logarithmic (LL) contributions n L n+1 ; the function g (2)
The term Lg
S
resums the next-to-leading logarithmic (NLL) contributions Sn L n ; g (3) controls the next-to-nextto-leading logarithmic (NNLL) terms Sn L n1 , and so forth.
Note that we use the logarithmic variable L (see Eq. (11)) to parametrize and organize the
resummation of the large logarithms ln(MH b)2 . We recall the main motivations [44] for this
choice. In the resummation region MH b 1, we have L ln(MH b)2 and the use of the variable L is fully legitimate to arbitrary logarithmic accuracy. When MH b 1, we have L 0
1. Therefore, the use of L reduces the effect
(whereas3 ln(MH b)2 !) and exp{G(S , L)}
2 More precisely, it depends only on the flavour of the colliding partons (see Appendix A).
3 As shown in Appendix B of Ref. [44] (see Eqs. (131) and (132) therein), after inverse Fourier transformation to
qT space, the b-dependent functions lnn (MH b)2 and L n lead to quite different behaviours at large qT . When qT MH ,
the behaviour (1/qT2 ) lnn1 (qT /MH ) (which is not integrable when qT ) produced by lnn (MH b)2 is damped (and
made integrable) by the extra factor qT /MH exp(b0 qT /MH ) produced in the case of L n .
produced by the resummed contributions in the small-b region (i.e., at large and intermediate values of qT ), where the large-b resummation approach is not justified. In particular, setting b = 0
= 1: this prop(which corresponds to integrate over the entire qT range) we have exp{G(S , L)}
erty can be interpreted [44] as a constraint of perturbative unitarity on the total cross section; the
dynamics of the all-order recoil effects, which are resummed in the form factor exp{G(S , L)},
produces a smearing of the fixed-order qT distribution of the Higgs boson without affecting its
total production rate.
The resummation formulae (10), (12) and (13) can be worked out at any given (and arbitrary)
logarithmic accuracy since the functions H and G can explicitly be expressed (see Ref. [44]) in
(n)
(n)
terms of few perturbatively-computable coefficients denoted by A(n) , B (n) , H (n) , CN , N . The
key role of these coefficients to fully determine the structure of transverse-momentum resummation was first formalized by Collins, Soper and Sterman [32,34,36]. The present status of the
calculation of these coefficients for Higgs boson production is recalled in Section 3.
In the case of the qT cross section integrated over the rapidity, Eq. (10) is still valid,
provided the double (N1 , N2 )-moments are replaced by the corresponding single N -moments
WN , HN , GN (see Section 2.2 in Ref. [44]). The relation between double and single moments
can easily be understood by inspection of Eqs. (6)(8). We see that setting = 0 in Eq. (8) is
exactly equivalent to integrate the cross section over the rapidity. Therefore, the functions WN ,
HN , GN in Ref. [44] are obtained by simply setting N1 = N2 = N in the corresponding functions
W (N1 ,N2 ) , H(N1 ,N2 ) , G (N1 ,N2 ) of Eq. (10).
Moreover, from the results presented in Ref. [44], we can straightforwardly obtain the functions H(N1 ,N2 ) and G (N1 ,N2 ) from the functions HN and GN . Roughly speaking, we simply have
1
G (N1 ,N2 ) = (GN1 + GN2 ),
2
H(N1 ,N2 ) = [HN1 HN2 ]1/2 .
(14)
More precisely, these equalities are valid in the simplified case where there is a single species
of partons (e.g., only gluons). In the following we comment on the physical picture that leads to
Eq. (14). The generalization to considering more species of partons does not require any further
conceptual steps: it just involves algebraic complications related to the treatment of the flavour
indices. The multiflavour case is briefly illustrated in Appendix A.
In the small-qT (large-b) region that we are considering, the kinematics of the Higgs boson is fully determined by the radiation of soft and collinear partons from the colliding partons (hadrons) in the initial state. The radiation of soft partons cannot affect the rapidity of
the Higgs bosons. On the contrary, the radiation of partons that are collinear to p1 (p2 ), i.e.,
in the forward (backward) region, decreases (increases) the rapidity of the Higgs boson as
a consequence of longitudinal-momentum conservation (see Eq. (3)). Since the emissions of
collinear partons from p1 and p2 are dynamically uncorrelated (factorized from each other),
correlations arise only from kinematics. The use of the (N1 , N2 )-moments exactly factorizes
(see Eqs. (2) and (9)) the kinematical constraint of longitudinal-momentum conservation. It
follows that the (N1 , N2 )-dependence of W (N1 ,N2 ) is given by the product of two functions
(N )
(N )
(say, W (N1 ,N2 ) = M1 1 M2 2 ) that depends only on N1 or N2 , respectively. If all the partons have the same flavour, the two functions should be equal, and Eq. (14) directly follows from
[W (N1 ,N2 ) ]N1 =N2 =N = WN .
The formalism illustrated in this section defines a systematic order-by-order (in extended
sense) expansion [44] of Eq. (4): it can be used to obtain predictions with uniform perturbative
accuracy from the small-qT region to the large-qT region. The various orders of this expansion
are denoted4 as LL, NLL + LO, NNLL + NLO, etc., where the first label (LL, NLL, NNLL,
. . .) refers to the logarithmic accuracy at small qT and the second label (LO, NLO, . . .) refers to
the customary perturbative order5 at large qT . To be precise, the NLL + LO term of Eq. (4) is
obtained by including the functions g (1) , g (2) and the coefficient H(1) (see Eqs. (13) and (12))
in the resummed component, and by expanding the finite (i.e., large-qT ) component up to its LO
(3)
term. At NNLL + NLO accuracy, the resummed component includes also the function gN and
the coefficient H(2) (see Eqs. (13) and (12)), while the finite component is expanded up to NLO.
It is worthwhile noticing that the NNLL + NLO (NLL + LO) result includes the full NNLO
(NLO) perturbative contribution in the small-qT region.
We recall [44] that, due to our actual definition of the logarithmic parameter L in Eq. (10)
and to our matching procedure with the perturbative expansion at large qT , the integral over
qT of the qT cross section exactly reproduces the customary fixed-order calculation of the total
cross section. This feature is not affected by keeping the rapidity fixed. Therefore, the NNLO
(NLO) result for total cross section at fixed y is exactly recovered upon integration over qT of
the NNLL + NLO (NLL + LO) qT spectrum at fixed y.
Within our formalism, resummation is directly implemented, at fixed MH , in the space of
the conjugate variables N1 , N2 and b. To obtain the cross section in Eq. (2), as function of the
kinematical variables s, y and qT , we have to perform inverse integral transformations. These
integrals are carried out numerically. We recall [44] that the resummed form factor (i.e., each of
in Eq. (13)) is singular at the value of b where S (2 )L = /0 (0 is
the functions g (k) (S L)
R
the first-order coefficient of the QCD function). This singularity has its origin from the presence
of the Landau pole in the running of the QCD coupling S (q 2 ) at low scales. When performing
the inverse Fourier (Bessel) transformation with respect to the impact parameter b (see Eq. (5)),
we deal with this singularity by using a minimal prescription [56,57]: the singularity is avoided
by deforming the integration contour in the complex b space (see Ref. [57]). We note that the
position of the singularity is completely independent of the values of N1 and N2 . Thus, the
inversion of the Mellin moments is performed in the customary way (in Mellin space there are no
singularities for sufficiently-large values of Re N1 and Re N2 ). In this respect, going from single
N -moments (as in Ref. [44]) to double (N1 , N2 )-moments (as in the present case, where the
rapidity is kept fixed) is completely straightforward, with no additional (practical or conceptual)
complications.
3. Higgs boson production at the LHC
In this section we apply the resummation formalism of Section 2 to the production of the
Standard Model Higgs boson at the LHC. We closely follow our previous study of the single
differential (with respect to qT ) cross section, with the same choice of parameters as stated in
Section 3 of Ref. [44]. Therefore, the integration over y of the double differential (with respect
to y and qT ) cross sections presented in this section returns the qT cross sections of Ref. [44].
As a cross-check of the actual implementation of the calculation, we have verified that after integration over the rapidity the numerical results in Ref. [44] are reobtained within a high accuracy.
4 In the literature on q resummation, other authors sometime use the same labels (NLL, NLO and so forth) with a
T
meaning that is different from ours.

5 We recall that the LO term at small q (i.e., including the region where q = 0) is proportional to 2 , whereas the
T
T
S
LO term at large qT is proportional to S3 . This mismatch of one power of S (and the ensuing mismatch of notation)
persists at each higher order (NLO, NNLO, . . . ).
As in Refs. [17,44], we use an improved version [16] of the large-Mt approximation. The
cross section is first computed by using the large-Mt approximation. Then, it is rescaled by a Born
level factor, such as to include the exact lowest-order dependence on the masses, Mt and Mb ,
of the top and bottom6 quarks, which circulates in the heavy-quark loop that couples to the
Higgs boson. We use the values Mt = 175 GeV and Mb = 4.75 GeV. As discussed in Ref. [17]
and recalled in Section 1, this version of the large-Mt approximation is expected to produce an
uncertainty that is smaller than the uncertainties from yet uncalculated perturbative terms from
higher orders.
For the sake of brevity, we present quantitative results only at NNLL + NLO accuracy, which
is the highest accuracy that can be achieved by using the present knowledge of exact perturbative
QCD contributions (resummation coefficients and fixed-order calculations [2527]). We use the
MRST2004 set [58] of parton distribution functions at NNLO. The use of NNLO parton densities
consistently matches the NNLL (NNLO) accuracy of our partonic cross section in the region of
small and intermediate values of qT .
Resummation up to the NLL level is under control from the knowledge of the perturbative
coefficients A(1) , B (1) , A(2) [35] and H(1) [37]. To reach the NNLL + NLO accuracy, the form
factor function G (N1 ,N2 ) in Eq. (13) must include the contribution from g (3)(N1 ,N2 ) (which is
controlled by the coefficients B (2) [38] and A(3) [59]), and the coefficient function H(N1 ,N2 ) in
Eq. (12) has to be evaluated up to its second-order term H(2)(N1 ,N2 ) . In Ref. [44] we exploited
b=0 = 0 to numerically derive an approximated form of the
the unitarity constraint G(S , L)|
coefficient H(2) from the NNLO calculation [12] of the total cross section. The recent calculation
of Ref. [15], which is based on the complete evaluation of H(2)(N1 ,N2 ) in analytic form, allows us
to gauge the quality of the approximated form. We find that the use of the H(2) of Ref. [44] leads
to differences of about 1% with respect to the exact computation of the rapidity cross section at
NNLO.
All the numerical results in this section are obtained by fixing the renormalization and factorization scales at the value R = F = MH . The resummation scale Q (the auxiliary scale
introduced in Ref. [44] to gauge the effect of yet uncalculated logarithmic terms at higher orders) is also fixed at the value Q = MH . The mass of the Higgs boson is set at the value
MH = 125 GeV.
We start our presentation of the predictions for Higgs boson production at the LHC by considering the qT dependence of the cross section at fixed values of the rapidity. In Fig. 1, we set
y = 0 and we compare the customary (when qT > 0) NLO calculation (dashed line) with the
resummed NNLL + NLO calculation (solid line).
As expected, the NLO result diverges to as qT 0 and, at small values of qT , it has an
unphysical peak that is produced by the numerical compensation of negative leading logarithmic
and positive subleading logarithmic contributions. The presence of this peak is not accidental. At
large qT , the perturbative expansion at any fixed order has no pathological behaviour: it leads to a
positive cross section, whose value decreases as qT increases. When qT 0, instead, any fixedorder calculation diverges alternatively to depending on the perturbative order. Therefore,
6 We note that the Born level cross section is not insensitive to the contribution of the bottom quark. Adding the
bottom-quark loop to the top-quark loop in the scattering amplitude produces a non-negligible interference effect in
the squared amplitude. The relative effect of the bottom quark decreases the Born level cross section by about 11% if
MH = 125 GeV, and by about 3% if MH = 300 GeV. If MH 500 GeV, the relative effect of the bottom quark is
always smaller than 1%.
10
Fig. 1. The qT spectrum at the LHC with MH = 125 GeV and y = 0: results at NNLL + NLO (solid line) and NLO
(dashed line) accuracy. The inset plot shows the ratio K (see Eq. (15)) of the corresponding qT cross sections, fixing
y = 0 (solid line) and integrating them over the full rapidity range (dashed line).
to go smoothly from the large-qT behaviour to the small-qT limit, the NLO (or N3 LO, and so
forth) calculation of the cross section has to show at least one peak in the intermediate-qT region.
We recall once more that the label NLO in Fig. 1 refers to (and originates from) the perturbative expansion at large qT . To avoid possible misunderstandings (coming from such a label)
when interpreting the dashed (NLO) curve in the small-qT region, we point out that, the only
difference produced in Fig. 1 by the NNLO calculation at small qT (this calculation can be carried out, for example, by using the NNLO codes of Refs. [14,15]) is a spike around the point
qT = 0. More precisely, as long as qT
= 0, the dashed curve is exactly the result of the NNLO
calculation of the qT cross section at small qT . The only difference introduced in the plot by this
NNLO calculation would occur in the first bin (with arbitrarily small size) that includes the point
qT = 0. The NNLO value of the qT cross section in this first bin is positive and fixed by the value
of the NNLO total cross section.7 Of course, owing to the increasingly negative behaviour of the
qT distribution when qT 0, the NNLO value of the qT cross section in the first bin increases
by decreasing the size of that bin.
The resummed NNLL + NLO result in Fig. 1 is physically well behaved at small qT (it
vanishes as qT 0 and has a kinematical peak at qT 12 GeV), and it converges to the expected
NLO result only when qT is definitely large (qT MH ).
7 By definition, the integral over q of d 2 /(dq dy) at NNLO is equal to d/dy at NNLO.
T
T
11
Fig. 2. The qT spectrum at the LHC with MH = 125 GeV and y = 2: results at NNLL + NLO (solid line) and NLO
(dashed line) accuracy. The inset plot shows the ratio K (see Eq. (15)) of the corresponding qT cross sections, fixing
y = 2 (solid line) and integrating them over the full rapidity range (dashed line).
To quantify more clearly the effect of the resummation on the NLO result, the value at y = 0
of the qT dependent K-factor,
K(qT , y) =
dNNLL+NLO /(dqT dy)

,
dNLO /(dqT dy)
(15)
is shown in the inset plot of Fig. 1. The dashed line shows the analogous K-factor as computed
from the ratio of the rapidity integrated cross sections. The similarity between these two K-factors
is a first indication of the mild rapidity dependence of the resummation effects. By inspection of
the inset plot, we note that NNLL resummation is relevant not only at small qT , but also in the
intermediate-qT region: as soon as qT 80 GeV, the resummation effects are larger than 20%.
Of course, the fact that K 1 at qT 24 GeV is purely accidental: it simply follows from the
unphysical behaviour of the fixed-order perturbative expansion at small qT .
Considering other values of the rapidity, from the central to the off-central rapidity region, we
find the same features as observed at y = 0. Our results of the qT spectrum at y = 2 are presented
in Fig. 2. The NNLL + NLO spectrum has a peak at qT 11 GeV. As happens in the case of the
qT distribution integrated over y, the effect of NNLL resummation is definitely non-negligible
starting from relatively-high values of qT . For example, at qT = 50 GeV the NNLL + NLO
result is about 30% higher than the NLO result.
To analyze the rapidity dependence in more detail, we study the doubly-differential cross
section at fixed values of qT . In Figs. 3 and 4, we show quantitative results at two typical values
12
Fig. 3. The rapidity spectrum at the LHC with MH = 125 GeV and qT = 15 GeV: results at NNLL + NLO (solid line)
and NLO (dashed line) accuracy. The inset plot shows the K-factor as defined in Eq. (15).
of the transverse momentum, qT = 15 GeV and qT = 40 GeV, in the small-qT and intermediateqT region, respectively.
Fig. 3 shows the rapidity distribution at NNLL + NLO (solid line) and NLO (dashes) accuracy
when qT = 15 GeV. At this value of qT , the effect of NNLL resummation reduces the cross
section. For example, when y = 0 the reduction effect is about 25%. As can be observed in the
inset plot, the relative contribution from the resummed logarithmic terms is rather constant in the
central rapidity region, and its dependence on y only appears in forward (and backward) region,
where the cross section is quite small.
When qT = 40 GeV (see Fig. 4), instead, the effect of NNLL resummation increases the
absolute value of the cross section. For example, when y = 0 the NLO cross section is increased
by about 22%. Nonetheless, as for the relative effect of resummation and the rapidity dependence
of the K-factor, we observe features that are very similar to those in Fig. 3. The resummation
effects have a very mild dependence on y in the central and (moderately) off-central regions,
and this explains the remarkable similarity between the solid and dashed lines in the inset plot
of Figs. 1 and 2. Since the kinematical region where |y| 2 accounts for most of the total cross
section, when comparing the ratio K(qT , y) to the analogous ratio of the y-integrated cross
sections, hardly any differences are expected, unless the large-rapidity region is explored.
The mild rapidity dependence of the qT shape of the resummed results can be studied with a
finer resolution by defining the following ratio:
R(qT ; y) =
d 2 /(dqT dy)
.
d/dqT
(16)
13
Fig. 4. The rapidity spectrum at the LHC with MH = 125 GeV and qT = 40 GeV: results at NNLL + NLO (solid line)
and NLO (dashed line) accuracy. The inset plot shows the K-factor as defined in Eq. (15).
This ratio gives the doubly-differential cross section normalized to the qT cross section integrated
over the full rapidity range. For comparison, we consider also the qT -integrated version of the
cross section ratio in Eq. (16), and we define the ratio
Ry =
d/dy
(17)
of the rapidity cross section d/dy over the total cross section .
We have computed the ratio in Eq. (16) by using the resummed qT cross sections at NNLL +
NLO accuracy. The results, as a function of qT , are presented in Fig. 5 (solid lines) at two different values, y = 0 and y = 2, of the rapidity. The results of the analogous (qT -independent)
ratio Ry (computed8 at NNLO with the numerical programs of Refs. [14,15]) at the corresponding values of rapidity are also reported (dotted lines) in Fig. 5. The dashed lines in Fig. 5
correspond to the computation of Eq. (16) by using the qT cross sections at NLO: we see that the
dashed and solid lines are very similar (as expected from the similarity of the dashed and solid
lines in the inset plot of Figs. 1 and 2). As discussed below, the results in Fig. 5 show that the
cross section decreases and the qT spectrum softens when the rapidity increases.
8 The numerical accuracy of this computation is better than about 2%3%. Owing to the unitarity constraint in our
resummation formalism, the same result (with a similar numerical accuracy) can be obtained by integration over qT of
the resummed qT cross sections.
14
Fig. 5. The rescaled qT spectrum (as defined by the ratio R(qT ; y) in Eq. (16)) at the LHC with MH = 125 GeV. The
solid (dashed) lines correspond to the NNLL + NLO (NLO) results at two different values of the rapidity: y = 0 (upper)
and y = 2 (lower). The dotted lines refer to the corresponding values of the ratio Ry (see Eq. (17)).
We observe that the lines at y = 0 lie above the lines at y = 2; this is just a consequence of
the fact that the cross sections (both at fixed qT and after integration over qT ) decrease when
y increases.
At fixed y, R(qT ; y) is not constant: it depends (though very slightly) on qT . We note that the
corresponding upper and lower lines in Fig. 5 have different slopes with respect to qT : fixing qT ,
the qT slope of R(qT ; y) decreases from positive to negative values as y increases from y = 0 to
y = 2, thus showing that the qT spectrum becomes slightly softer at larger rapidity. In general, as
|y| increases, the hardness of the qT shape of d 2 /(dqT dy) decreases. Since the cross section
decreases by increasing the rapidity, the hardness of d/dqT (the denominator in Eq. (16)) is
intermediate between the values of the hardness of d 2 /(dqT dy) (the numerator in Eq. (16)) at
y = 0 and at large |y|. As a consequence, the qT slope of R(qT ; y) is necessarily positive when
y = 0. Note that the qT slope is already negative when y = 2 (Fig. 5): this is a consequence of
the fact that the bulk of the cross section is in the rapidity region |y| 2.
Our qualitative illustration of the results in Fig. 5 can be accompanied by some quantitative
observations. We note that the rapidity dependence of the cross sections is sizeable: going from
y = 0 to y = 2, the ratio Ry decreases by about 43%; comparable variations affect the ratio
R(qT ; y), which is not very different from Ry and it is slowly dependent on qT . Indeed, at
fixed y, the ratio R(qT ; y) at NNLL + NLO accuracy has a small and nearly constant slope
from low values of qT around the peak (say, qT 10 GeV) to qT = 100 GeV; varying qT in
this region, R(qT ; y) increases by about 11% when y = 0, and it decreases by about 16% when
y = 2. In the same range of qT and y, the values of R(qT ; y) at NNLL + NLO (solid lines)
15
and at NLO (dashed lines) are very similar: although this is expected at large qT , the differences
never exceed the level of about 4% even at values of qT as low as qT 10 GeV.
In summary, the results in Fig. 5 show that, when |y| increases from the central to the (moderately) off-central region, the cross sections vary more in absolute value than in qT shape. These
features deserve some words of discussion.
We first consider the total cross section and the rapidity cross section d/dy. We recall (see
Section 1) that the value of these cross sections is sizeably affected by QCD radiative corrections.
The bulk of the effect is due to the radiation of virtual and soft gluons, and they cannot affect the
rapidity of the Higgs boson. As a consequence, the ratio Ry has little sensitivity to perturbative
QCD corrections. The decreases of Ry as |y| increases is mainly driven by the decrease of the
2 ) as x increases. Considering the large-q region, similar arguments apgluon density fg (x, MH
T
ply to the qT cross sections d/dqT and d/(dqT dy), and similar conclusions apply to the ratio
R(qT , y). In the small-qT region, we have to consider the additional and large effect produced
2 /q 2 ). These terms are
on the qT cross sections by the logarithmically-enhanced terms lnm (MH
T
due to the radiation of soft and collinear partons. As already discussed in Section 2, the rapidity
of the Higgs boson can be varied only by collinear radiation, while soft radiation can only lead
to on overall (independent of y) rescaling of the qT cross sections. At the LL level, only soft
radiation contributes (the LL function g (1) in Eq. (13) does not depend on N1 and N2 ) and all the
logarithmic terms cancel in the ratio R(qT , y). The y sensitivity of R(qT , y) starts at the NLL
level. The corrections produced on the dominant soft-gluon effects by the collinear radiation are
physically [29] well approximated by varying the scale of the gluon density from MH
to qT . As a consequence, the variations of the hardness of the qT cross sections are mainly
driven by d ln fg (x, qT2 )/d ln qT2 , the amount of scaling violation of the gluon density. Since the
scaling violation decreases as x increases, the hardness of d/(dqT dy) decreases and the qT
spectrum softens as |y| increases. Note that, by increasing x, the gluon density decreases faster
than its scaling violation: this explains why d/(dqT dy) varies more in absolute value than in
qT shape when |y| increases.
We conclude this section with some comments about the theoretical uncertainties on the
doubly-differential cross section d/(dqT dy) at NNLL + NLO accuracy. In Ref. [44] the
perturbative QCD uncertainties on d/dqT were investigated by comparing the results at
NNLL + NLO and NLL + LO accuracies and by performing scale variations at NNLL + NLO
level. We also considered the inclusion of non-perturbative contributions, and we found that they
lead to small corrections provided qT is not very small. From these studies we concluded that the
NNLL + NLO result has a QCD uncertainty of about 10% in the region from small (around
the peak of the qT distribution) to intermediate (say, roughly, qT MH /3) values of transverse
momenta. Similar studies can be carried out in the case of the doubly-differential cross section
d/(dqT dy). These studies are not reported here since their results and the ensuing conclusions
are very similar to those in Ref. [44]. The reason for this similarity is a feature that we have
pointed out throughout this section: the qT resummation effects have a very mild dependence on
the rapidity and, thus, they are almost unchanged when comparing d/(dqT dy) with d/dqT
(equivalently, they largely cancel in the ratio in the cross section ratio of Eq. (16)).
4. Summary
We have considered the resummation of the logarithmically-enhanced QCD contributions that
appear at small transverse momenta when computing the qT spectrum of a Higgs boson produced
in hadron collisions. In our previous work on the subject [41,44], the rapidity of the Higgs boson
16
was integrated over: resummation was implemented by using a formalism based on a transform
to impact parameter and Mellin moment space. In this paper we have extended the resummation
formalism to the case in which the rapidity is kept fixed, and we have considered the doublydifferential cross section with respect to the transverse momentum and the rapidity. We have
shown that this extension can be carried out without substantial complications: it is sufficient to
enlarge the conjugate space by introducing a suitably-defined double Mellin transformation.
The main aspects of our method [36,44], which are recalled here, are unchanged by the inclusion of the rapidity dependence. The resummation is performed at the level of the partonic
cross section, and the parton densities are factorized as in the customary fixed-order calculations. The formalism is completely general and it can be applied to other processes: the large
logarithmic contributions are universal and, thus, they are systematically exponentiated in a
process-independent form (see Eqs. (10) and (13)); the process-dependent part is factorized in
the hard-scattering coefficient H. A constraint of perturbative unitarity is imposed on the resummed terms (see Eq. (11)), so that the qT smearing produced by the resummation does not
change the total production rate. This constraint reduces the effect of unjustified higher-order
contributions at intermediate qT and facilitates the matching procedure with the complete fixedorder calculations at large qT . In particular, when the rapidity is kept fixed, the integration over
qT of d/(dqT dy) at NNLL + NLO accuracy returns d/dy at NNLO.
We have presented numerical results for Higgs boson production at the LHC. Comparing
fixed-order and resummed calculations, we find that the resummation effects are large at small qT
(as expected) and still sizeable at intermediate qT . The inclusion of the rapidity dependence has
little quantitative impact on this picture since, as we have shown, the qT resummation effects are
mildly dependent on the rapidity. Going from the central to the (moderately) off-central rapidity
region, the qT shape of the spectrum slightly softens. In the range from small to intermediate values of qT , the residual perturbative uncertainty of the NNLL + NLO predictions for d/(dqT dy)
is comparable to that of advanced (NNLO or NNLL + NNLO) calculations of the qT inclusive
cross sections d/dy and .
Acknowledgements
The work of D.dF. was supported in part by CONICET. D.dF. wishes to thank the Physics
Department of the University of Florence and INFN for support and hospitality while this work
was completed.
Appendix A
In this appendix we present the structure of the resummation formula (10) by explicitly including the dependence on the flavour indices of the colliding partons.
In the context of our resummation formalism, a detailed derivation of exponentiation in the
multiflavour case is illustrated in Appendix A of Ref. [44]. Considering a generic LO partonic
subprocess c + c F (F = H and c = c = g in the specific case of Higgs boson production by
gluon fusion), and performing qT resummation after integration over the rapidity, the resummed
(res.)
component d a1 a2 /dqT2 of the partonic cross section is controlled by the N -moments WaF1 a2 ,N .
The final exponentiated result for these N -moments is given by the master formulae (106)(108)
of Ref. [44]. We recall the master formula (106) in the following form:
{I },F

,
Ha1 a2 ,N (M, S ) exp G{I },N (S , L)
WaF1 a2 ,N (b, M; S ) =
(A.1)
{I }
17
where the sum extends over the following set of flavour indices:
{I } = c, c,
ii , i2 , b1 , b2 ,
(A.2)
and, for simplicity, the functional dependence on various scales (such as the renormalization and
{I },F
factorization scales) is understood. The functions G{I },N and Ha1 a2 ,N are given in the master
formulae (107) and (108), respectively. In the present paper, qT resummation is performed at
(N ,N )F
fixed values of the rapidity, and the double (N1 , N2 )-moments Wa1 a12 2 in Eq. (6) replace the
F
N -moments Wa1 a2 ,N of Ref. [44]. The generalization of Eq. (10) to the multiflavour case is
straightforwardly obtained from Eq. (A.1) by the simple replacement N (N1 , N2 ):

1 ,N2 )F (b, M; ) =
1 ,N2 )F (M, ) exp G (N1 ,N2 ) ( , L)
Wa(N
(A.3)
Ha{I1},(N
S
S
S .
a
a2
{I }
1 2
{I }
(N ,N )
The exponent G{I }1 2 of
{I },(N ,N )F
factor Ha1 a2 1 2 are
the process-independent form factor and the process-dependent hard
(N ,N2 )
= Gc + Gi1 ,N1 + Gcb1 ,N1 + Gi2 ,N2 + Gcb

2 ,N2 ,

(i1 ) 1
1 ,N2 )F = (0) H F S C
2 ,N2 E (i2 ) V 1 U N2
Ha{I1},(N
a2
c c cb1 ,N1 E N1 V N1 U N1 b a Ccb
cc,F
N2
N2
b
G{I }1
1 1
2 a2
(A.4)
.
(A.5)
The expressions in Eqs. (A.4) and (A.5) are completely analogous to the master formulae (107)
and (108) in Ref. [44] (the functional dependence on the scales M, R , F and Q is explicitly
denoted in those formulae). In particular, we note that the dependence of G (N1 ,N2 ) and H(N1 ,N2 )
on the Mellin variables N1 and N2 is completely factorized: each of terms on the right-hand side
of Eqs. (A.4) and (A.5) depends only on one Mellin variable (either N1 or N2 ). This factorized
structure is completely consistent with Eq. (14) and with the physical picture discussed below
Eq. (14); the dependence on N1 (N2 ) follows the longitudinal-momentum flow and the flavour
flow a1 b1 i1 c (a2 b2 i2 c)
that are produced by collinear radiation from the
(i)
initial-state parton with momentum p1 (p2 ). The various Mellin functions (Gi,N , E N , U N and
so forth) in Eqs. (A.4) and (A.5) can be found in Ref. [44].
References
[1] For a review on Higgs physics in and beyond the Standard Model, see J.F. Gunion, H.E. Haber, G.L. Kane, S. Dawson, The Higgs Hunters Guide, AddisonWesley, Reading, MA, 1990;
M. Carena, H.E. Haber, Prog. Part. Nucl. Phys. 50 (2003) 63;
A. Djouadi, report LPT-ORSAY-05-17, hep-ph/0503172;
A. Djouadi, report LPT-ORSAY-05-18, hep-ph/0503173.
[2] ATLAS Collaboration, ATLAS Detector and Physics Performance: Technical Design Report, vol. 2, report CERN/
LHCC/99-15, 1999;
S. Asai, et al., Eur. Phys. J. C 32 (Suppl. 2) (2004) 19;
CMS Collaboration, CMS Physics Technical Design Report: Physics Performance, vol. 2, report CERN/
LHCC/2006-021, 2006.
[3] M. Carena, et al., Report of the Tevatron Higgs working group, hep-ph/0010338;
CDF Collaboration, D Collaboration, Results of the Tevatron Higgs Sensitivity Study, report FERMILAB-PUB03/320-E;
V.M. Abazov, et al., D Collaboration, Phys. Rev. Lett. 96 (2006) 011801;
A. Abulencia, et al., CDF Collaboration, Phys. Rev. Lett. 97 (2006) 081802;
The TEVNPH working group, for the CDF and D Collaborations, hep-ex/0612044.
18
[4] S. Dawson, Nucl. Phys. B 359 (1991) 283;

A. Djouadi, M. Spira, P.M. Zerwas, Phys. Lett. B 264 (1991) 440.
[5] M. Spira, A. Djouadi, D. Graudenz, P.M. Zerwas, Nucl. Phys. B 453 (1995) 17.
[6] U. Aglietti, R. Bonciani, G. Degrassi, A. Vicini, JHEP 0701 (2007) 021;
C. Anastasiou, S. Beerli, S. Bucherer, A. Daleo, Z. Kunszt, JHEP 0701 (2007) 082.
[7] M. Muhlleitner, M. Spira, report PSI-PR-06-15, hep-ph/0612254.
[8] R.V. Harlander, Phys. Lett. B 492 (2000) 74;
V. Ravindran, J. Smith, W.L. van Neerven, Nucl. Phys. B 704 (2005) 332;
T. Gehrmann, T. Huber, D. Maitre, Phys. Lett. B 622 (2005) 295.
[9] S. Catani, D. de Florian, M. Grazzini, JHEP 0105 (2001) 025.
[10] R.V. Harlander, W.B. Kilgore, Phys. Rev. D 64 (2001) 013015.
[11] S. Catani, D. de Florian, M. Grazzini, JHEP 0201 (2002) 015.
[12] R.V. Harlander, W.B. Kilgore, Phys. Rev. Lett. 88 (2002) 201801;
C. Anastasiou, K. Melnikov, Nucl. Phys. B 646 (2002) 220;
V. Ravindran, J. Smith, W.L. van Neerven, Nucl. Phys. B 665 (2003) 325.
[13] C. Anastasiou, L.J. Dixon, K. Melnikov, Nucl. Phys. B (Proc. Suppl.) 116 (2003) 193.
[14] C. Anastasiou, K. Melnikov, F. Petriello, Phys. Rev. Lett. 93 (2004) 262002;
C. Anastasiou, K. Melnikov, F. Petriello, Nucl. Phys. B 724 (2005) 197.
[15] S. Catani, M. Grazzini, Phys. Rev. Lett. 98 (2007) 222002.
[16] M. Kramer, E. Laenen, M. Spira, Nucl. Phys. B 511 (1998) 523.
[17] S. Catani, D. de Florian, M. Grazzini, P. Nason, JHEP 0307 (2003) 028.
[18] S. Moch, A. Vogt, Phys. Lett. B 631 (2005) 48.
[19] E. Laenen, L. Magnea, Phys. Lett. B 632 (2006) 270.
[20] A. Idilbi, X.d. Ji, J.P. Ma, F. Yuan, Phys. Rev. D 73 (2006) 077501.
[21] V. Ravindran, Nucl. Phys. B 746 (2006) 58;
V. Ravindran, Nucl. Phys. B 752 (2006) 173;
V. Ravindran, J. Smith, W.L. van Neerven, report HRI-04-2006, hep-ph/0608308.
[22] R.K. Ellis, I. Hinchliffe, M. Soldate, J.J. van der Bij, Nucl. Phys. B 297 (1988) 221;
U. Baur, E.W. Glover, Nucl. Phys. B 339 (1990) 38.
[23] V. Del Duca, W. Kilgore, C. Oleari, C. Schmidt, D. Zeppenfeld, Nucl. Phys. B 616 (2001) 367;
V. Del Duca, W. Kilgore, C. Oleari, C.R. Schmidt, D. Zeppenfeld, Phys. Rev. D 67 (2003) 073003.
[24] J. Smith, W.L. van Neerven, Nucl. Phys. B 720 (2005) 182.
[25] D. de Florian, M. Grazzini, Z. Kunszt, Phys. Rev. Lett. 82 (1999) 5209.
[26] V. Ravindran, J. Smith, W.L. Van Neerven, Nucl. Phys. B 634 (2002) 247.
[27] C.J. Glosser, C.R. Schmidt, JHEP 0212 (2002) 016.
[28] D. de Florian, A. Kulesza, W. Vogelsang, JHEP 0602 (2006) 047.
[29] Y.L. Dokshitzer, D. Diakonov, S.I. Troian, Phys. Lett. B 79 (1978) 269;
Y.L. Dokshitzer, D. Diakonov, S.I. Troian, Phys. Rep. 58 (1980) 269.
[30] G. Parisi, R. Petronzio, Nucl. Phys. B 154 (1979) 427.
[31] G. Curci, M. Greco, Y. Srivastava, Nucl. Phys. B 159 (1979) 451.
[32] J.C. Collins, D.E. Soper, Nucl. Phys. B 193 (1981) 381;
J.C. Collins, D.E. Soper, Nucl. Phys. B 213 (1983) 545, Erratum;
J.C. Collins, D.E. Soper, Nucl. Phys. B 197 (1982) 446.
[33] J. Kodaira, L. Trentadue, Phys. Lett. B 112 (1982) 66;
J. Kodaira, L. Trentadue, report SLAC-PUB-2934, 1982;
J. Kodaira, L. Trentadue, Phys. Lett. B 123 (1983) 335.
[34] J.C. Collins, D.E. Soper, G. Sterman, Nucl. Phys. B 250 (1985) 199.
[35] S. Catani, E. DEmilio, L. Trentadue, Phys. Lett. B 211 (1988) 335.
[36] S. Catani, D. de Florian, M. Grazzini, Nucl. Phys. B 596 (2001) 299.
[37] R.P. Kauffman, Phys. Rev. D 45 (1992) 1512;
C.P. Yuan, Phys. Lett. B 283 (1992) 395.
[38] D. de Florian, M. Grazzini, Phys. Rev. Lett. 85 (2000) 4678;
D. de Florian, M. Grazzini, Nucl. Phys. B 616 (2001) 247.
[39] C. Balazs, C.P. Yuan, Phys. Lett. B 478 (2000) 192;
C. Balazs, J. Huston, I. Puljak, Phys. Rev. D 63 (2001) 014021.
[40] E.L. Berger, J.w. Qiu, Phys. Rev. D 67 (2003) 034026;
E.L. Berger, J.w. Qiu, Phys. Rev. Lett. 91 (2003) 222003.
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
[50]
[51]
[52]
[53]
[54]
[55]
[56]
[57]
[58]
[59]
19
G. Bozzi, S. Catani, D. de Florian, M. Grazzini, Phys. Lett. B 564 (2003) 65.

A. Kulesza, W.J. Stirling, JHEP 0312 (2003) 056.
A. Kulesza, G. Sterman, W. Vogelsang, Phys. Rev. D 69 (2004) 014012.
G. Bozzi, S. Catani, D. de Florian, M. Grazzini, Nucl. Phys. B 737 (2006) 73.
Q.H. Cao, C.R. Chen, report UCRHEP-T428, arXiv: 0704.1344.
A. Gawron, J. Kwiecinski, Phys. Rev. D 70 (2004) 014003;
G. Watt, A.D. Martin, M.G. Ryskin, Phys. Rev. D 70 (2004) 014012;
G. Watt, A.D. Martin, M.G. Ryskin, Phys. Rev. D 70 (2004) 079902, Erratum;
M. Luszczak, A. Szczurek, Eur. Phys. J. C 46 (2006) 123.
G. Davatz, G. Dissertori, M. Dittmar, M. Grazzini, F. Pauss, JHEP 0405 (2004) 009.
A.V. Lipatov, N.P. Zotov, Eur. Phys. J. C 44 (2005) 559.
F. Stockli, A.G. Holzner, G. Dissertori, JHEP 0510 (2005) 079;
G. Davatz, F. Stockli, C. Anastasiou, G. Dissertori, M. Dittmar, K. Melnikov, F. Petriello, JHEP 0607 (2006) 037.
C. Anastasiou, G. Dissertori, F. Stockli, report CERN-PH-TH/2007-118, arXiv: 0707.2373.
S. Catani, L. Trentadue, Nucl. Phys. B 327 (1989) 323;
S. Catani, L. Trentadue, Nucl. Phys. B 353 (1991) 183.
H. Kawamura, J. Kodaira, K. Tanaka, hep-ph/0703079.
G. Sterman, W. Vogelsang, JHEP 0102 (2001) 016.
S. Catani, L. Trentadue, G. Turnock, B.R. Webber, Nucl. Phys. B 407 (1993) 3.
G. Sterman, Nucl. Phys. B 281 (1987) 310.
S. Catani, M.L. Mangano, P. Nason, L. Trentadue, Nucl. Phys. B 478 (1996) 273.
E. Laenen, G. Sterman, W. Vogelsang, Phys. Rev. Lett. 84 (2000) 4296;
A. Kulesza, G. Sterman, W. Vogelsang, Phys. Rev. D 66 (2002) 014011.
A.D. Martin, R.G. Roberts, W.J. Stirling, R.S. Thorne, Phys. Lett. B 604 (2004) 61.
A. Vogt, S. Moch, J.A.M. Vermaseren, Nucl. Phys. B 691 (2004) 129.
Long wavelength limit of evolution of cosmological

perturbations in the universe where scalar fields
and fluids coexist
Takashi Hamazaki
Kamiyugi 3-3-4-606 Hachioji-city, Tokyo 192-0373, Japan
Received 7 July 2007; accepted 24 September 2007
Available online 29 September 2007
Abstract
We present the LWL formula which represents the long wavelength limit of the solutions of evolution
equations of cosmological perturbations in terms of the exactly homogeneous solutions in the most general
case where multiple scalar fields and multiple perfect fluids coexist. We find the conserved quantity which
has origin in the adiabatic decaying mode, and by regarding this quantity as the source term we determine
the correction term which corrects the discrepancy between the exactly homogeneous perturbations and the
k 0 limit of the evolutions of cosmological perturbations. This LWL formula is useful for investigating the evolutions of cosmological perturbations in the early stage of our universe such as reheating after
inflation and the curvaton decay in the curvaton scenario. When we extract the long wavelength limits of
evolutions of cosmological perturbations from the exactly homogeneous perturbations by the LWL formula,
it is more convenient to describe the corresponding exactly homogeneous system with not the cosmological
time but the scale factor as the evolution parameter. By applying the LWL formula to the reheating model
and the curvaton model with multiple scalar fields and multiple radiation fluids, we obtain the S formula
representing the final amplitude of the Bardeen parameter in terms of the initial adiabatic and isocurvature
perturbations.
PACS: 98.80.Cq
Keywords: Cosmological perturbations; Long wavelength limit; Reheating; Curvaton
E-mail address: yj4t-hmzk@asahi-net.or.jp.

doi:10.1016/j.nuclphysb.2007.09.028
T. Hamazaki / Nuclear Physics B 791 (2008) 2059
21
1. Introduction and summary

Recently we come to be required to investigate the evolution of cosmological perturbations in
the very early universe [1,2]. According to the inflationary scenarios and the curvaton scenario, in
the early universe the wavelength of cosmological perturbations responsible for the present cosmic structures such as galaxies and clusters of galaxies is much larger than the horizon scales.
Therefore the methods for researching the cosmological perturbations on superhorizon scales
have been sought. In this context, Nambu and Taruya pointed out that there exists an LWL formula representing the k 0 limit of the cosmological perturbations in terms of the exactly
homogeneous perturbations [3]. Soon later in the multiple scalar fields system [4,5], in the multiple fluids system [4] the complete LWL formulae were constructed. Since the evolution equations
of the corresponding exactly homogeneous universe look simpler than the evolution equations
of cosmological perturbations, the LWL formula brought about great simplification. For this reason, the LWL formula was used for investigating the evolution of cosmological perturbation on
superhorizon scales by several authors [3,58]. Especially the LWL formula was used effectively
in the analysis of the system containing oscillatory scalar fields. By using such LWL formula,
in single oscillatory scalar field system [6] and in nonresonant multiple oscillatory scalar fields
system [7] it was shown that the Bardeen parameter is conserved, and in resonant multiple oscillatory scalar fields system [8] it was shown that the cosmological perturbation including the
Bardeen parameter can grow. In the papers [7,8], the averaging method representing the averaging over the fast changing angle variables was used for investigating the corresponding exactly
homogeneous system, and by the LWL formula the evolution of the cosmological perturbation
in the long wavelength limit was constructed from the corresponding exactly homogeneous perturbation. In addition, the viewpoint that the evolutions of the cosmological perturbations on
superhorizon scales are governed by the stability and/or instability of the corresponding exactly
homogeneous universe [7,8] is physically important. In this context, the phase space of the corresponding exactly homogeneous system was investigated in detail and the role of the fixed points
in the phase space in the stability and instability of cosmological perturbation was discussed [8].
By replacing the oscillating scalar fields with the dust fluids, the evolution of the cosmological
perturbations during reheating after the inflation [9] and in the curvaton scenario [10] were investigated. These authors treated the system dominated by dust-like scalar field fluid and radiation
and investigated the influence of the entropy perturbation originating from the multicomponent
property to the evolution of the total curvature perturbation variables such as the Bardeen parameter. The purpose of these analyses was to determine the initial perturbation of the present
Friedmann universe in terms of the early stage seed perturbation. Although it was shown partially
that this replacement is physically reasonable [6,9], it is dangerous that there can be possibility to
miss the instabilities characteristic to the rapidly oscillating scalar fields [8,1114]. The instability of the oscillatory scalar fields has been investigated [1518]. Therefore the direct investigation
of the system where the scalar fields and fluids coexist becomes more and more necessary and
the LWL formula of the scalar-fluid composite system is required eagerly. If we have such LWL
formula, by investigating the deformation of the phase space of the corresponding exactly homogeneous system due to the dissipation of the oscillatory scalar fields into fluids we will be able
to interpret the evolution of cosmological perturbation more physically.
This paper is organized as follows. In Section 2, we construct the LWL formula as for such
scalar-fluid composite system based on the philosophy of the paper [4]. As shown in the paper
[4], the discrepancy exists between the evolution equations of the cosmological perturbations
in the k 0 limit and the evolution equations of the exactly homogeneous perturbations be-
22
cause the former contains k 2 = O(1) terms and the latter does not, therefore this discrepancy
should be corrected by the correction term which contributes the adiabatic decaying mode, but
any general methods for determining such correction term have not been presented yet, and only
in the multiple scalar fields system such correction term was determined. We show in the k 0
limit the existence of the conserved quantity which has origin in the adiabatic decaying mode
and which is related with k 2 . By regarding this conserved quantity as the source term, and
obtaining the special solution A , we correct the exactly homogeneous perturbation A and we
obtain the complete LWL formula A = A + A in the most general scalar-fluid composite system. In Section 3, we point out that it is more appropriate to use the scale factor a rather than
the cosmological time t as the evolution parameter when we use the LWL formula. As for the
scalar quantity T , we use the perturbation variable DT , D is the operator which maps the exactly
homogeneous scalar quantity T to the gauge invariant perturbation variable representing the T
fluctuation in the flat slice. D defined in this way can be interpreted as a kind of derivative operator. In fact, the exactly homogeneous part (DT ) can be expressed as the derivative of T with
respect to the solution constant with the scale factor a fixed. In order to investigate the exactly
homogeneous system containing oscillatory scalar fields, we use the action angle variables Ia , a .
By using D defined in this way, we can define the action angle perturbation variables DIa , Da
whose exactly homogeneous parts are given as the derivatives of Ia , a with solution constant C
with the scale factor a fixed. When we use the derivative operator D and the LWL formulae,
it is essential to use the scale factor a as the evolution parameter. In Section 4, we apply the
LWL formulae to the non-interacting multicomponents system and discuss the long wavelength
limit of the evolution of the Bardeen parameter. In Section 5, we apply the averaging method by
which the system is averaged over the fast changing angle variables to the decaying scalar fields
which have been discussed in the reheating model and the curvaton model. By evaluating the
corrections produced by the averaging process and the errors produced by the truncation of the
sufficient reduced angle variables dependent part, the validity of the averaging method is established. In Sections 6, 7, we apply the LWL formula and the averaging method to the interacting
multicomponents model such as the reheating model, the curvaton model, respectively. We assume that the multiple scalar fields and the multiple radiation fluid components exist. In these
models, we construct the S formulae representing the final amplitude of the Bardeen parameter in
terms of the initial adiabatic and isocurvature perturbations. In our previous paper [8], the evolutionary behaviors of cosmological perturbations in the early universe where multiple oscillatory
scalar fields interact with each other have been investigated. This S formula gives the information about how the cosmological perturbations which grew in such early era are transmitted into
the radiation energy density perturbations through the energy transfer from the scalar fields into
the radiation fluids. We present the necessary condition for the initial entropic perturbations produced in the early era to survive until the late radiation dominant universe. Section 8 is devoted
to discussions containing consideration of the case where the decay rate depends on the physical
quantities. In Appendices A and B, the proofs of the propositions presented in Section 5 and the
evaluations of the useful mathematical formulae used in Section 6 are contained.
In this paper, we consider the case where the homogeneous scalar fields obey the phenomenological evolution equations as
U
+ 3H +
+ S = 0.
(1.1)
The interactions between scalar fields are described by the interaction potential U , while the
interactions between scalar fields and fluids are described by S. This analysis includes the well-
23
known case S = [19,20] and the general case where S is an arbitrary analytic function of ,
which was discussed in the paper [21] but whose perturbations have not been investigated yet.
Another supplemental purpose of our paper is to present the evolution equations of cosmological
perturbations corresponding to the homogeneous system with various source term S especially
dependent on .
The notations used in this paper are based on the review [22] and the paper [4].
2. Derivation of the LWL formula
We give the definitions and the evolution equations as for the background and the perturbation
variables. Based on these notations, in the most general model where the multiple scalar fields
and the multiple perfect fluid components interact, we give the LWL formula representing the
evolutions of the perturbations variables in terms of the exactly homogeneous solutions.
We consider perturbations on a spatially flat RobertsonWalker universe given by
2 = (1 + 2AY ) dt 2 2aBYj dt dx j
ds

+ a 2 (1 + 2HL Y )j k + 2HT Yj k dx j dx k ,
(2.1)
where Y , Yj and Yj k are harmonic scalar, vector and tensor for a scalar perturbation with wave
vector k on flat three-space:

kj
kj k k
1
Yj := i Y,
Yj k :=
j k 2 Y.
Y := eikx ,
(2.2)
k
3
k
By using the gauge dependent variables R and g representing the spatial curvature perturbation
and the shear, respectively:
1
a
g := H T B,
R := HL + HT ,
3
k
we can define two independent gauge invariant variables:

R
aH
A := A
,
:= R
g .
H
k
(2.3)
(2.4)
In order to define the matter perturbation variables, we will consider the scalar quantity perturbation variables generally. As for covariant scalar quantity T = T + T Y , we define the gauge
invariant perturbation variable representing the T fluctuation in the flat slice:
DT := T
T
R.
H
(2.5)
Next we consider the covariant scalar quantity T2 whose background quantity is the time derivative of T : T . The extension of T into the covariant scalar quantity T2 is not unique. For example,

1/2
T2 = sgn(0 T ) g T T
(2.6)
,
and
T2 = n T ,
(2.7)
where n is an arbitrary vector field satisfying

n n = 1,
(2.8)
24
have the same T as the background part. But these different T2 s give the unique perturbation
part: DT2 = (DT ) T A. Therefore we can define D T by
D T := (DT ) T A.
(2.9)
We consider the universe where the scalar fields a (1 a NS ) and the fluids , P (1
Nf ) coexist, whose energymomentum tensor is divided into A = (S, f ) parts where S
represents the multiple scalar fields, f represents the multiple fluids. The energymomentum
tensor of f parts are further divided into individual fluids parts . On the other hand, the energy
momentum tensor of S part cannot be divided into individual scalar fields parts a, since the
interaction potential U contains the terms consisting of plural scalar fields a :

T = T S + T f = T S +
(2.10)
,
T
)S +
0 = (Q )S + (Q )f = (Q
Q ,
(2.11)
where the energymomentum transfer vector Q A is defined by
= Q A = Q A u + fA ,
TA
(2.12)
where u is the four velocity of the whole matter system and the momentum transfer fA
satisfies u fA = 0. For the scalar perturbation, the energymomentum tensor and the energy
momentum transfer vector of each individual component are expressed as
0
= (A + A Y ),
TA0
0
TAj = a(A + PA )(vA B)Yj ,

j
j
j
j
TAk = PA k + PA Y k + T A Yk ,
(2.13)
(2.14)
(2.15)
and

Q A0 = QA + (QA A + QA )Y ,

Q Aj = a QA (v B) + FcA Yj ,
(2.16)
(2.17)
where A , PA and QA are the background quantities of the energy density, the pressure and the
energy transfer of the individual component A, respectively. The anisotropic pressure perturbation T A and the momentum transfer perturbation FcA are already gauge invariant. As for the
scalar quantities T = (A , PA , QA ), we use DT as the gauge invariant perturbation variables.
As for the gauge invariant velocity perturbation variable, we use
aH
(vA B).
k
The energymomentum tensor of scalar fields part is given by
ZA := R

1
T S = + 2U .
2
Since divergence of the energymomentum tensor is given by

U
a
T S = 2
a ,
a
(2.18)
(2.19)
(2.20)
25
in order that the phenomenological equations of motion of the scalar fields become
a
2
U
= Sa ,
a
(2.21)
we assume that
(Q )S = Sa a .
(2.22)
By using the scalar fields background variables a , a , Sa and the corresponding perturbation
variables Da , D a , DSa , the background part of the fluid variables is given by
1 2
+ U,
S = ()
2
1 2
U,
PS = ()
2
2,
hS = ()
QS = S ,
(2.23)
(2.24)
(2.25)
(2.26)
and the perturbation part of fluid variables is given by

S
S
D ,
D +

PS
PS
(DP )S =
D ,
D +

(hZ)S = H D,
(D)S =
(2.27)
(2.28)
(2.29)
(T )S = 0,
(2.30)
(DQ)S = S D DS,

k
(aFc )S = Sa kDa a Z .
H
(2.31)
(2.32)
When the source of the scalar field a , Sa is given as functions of the covariant scalar quantities
T and T2 whose background part is T , that is Sa = Sa (T , T2 ), DSa is given by
Sa
Sa
D T .
DT +
T
T
In such case, (DQ)S can be written as
DSa =
(DQ)S =
QS
QS
D T ,
DT +
T
T
(2.33)
(2.34)
which is assumed from now on. In the same way as the individual components TA , as for the

total energymomentum tensor T = A TA , we can define the gauge invariant perturbation

variables such as D, DP , hZ and T . From (2.10), (2.11), we obtain the background equations
as

,
= S +
(2.35)
P = PS +
P ,
(2.36)
26
h = hS +
(2.37)
h ,
0 = QS +
(2.38)
Q ,
and perturbation equations as

D ,
D = DS +
(2.39)
DP = DPS +
(2.40)
DP ,
hZ = (hZ)S +
(2.41)
h Z ,
T = (T )S +
(2.42)
T ,
0 = (DQ)S +
(2.43)
DQ ,
0 = (Fc )S +
(2.44)
Fc .
This Z is known as the Bardeen parameter [2224]. In the long wavelength limit, the Bardeen
parameter is conserved in the case where the entropy perturbations are negligible. But in various
systems it was reported that the entropy perturbations cannot be neglected [2,8,11,12], so in
the present paper we will investigate the evolutionary behavior of the Bardeen parameter more
carefully. Until now, as for the gauge invariant scalar quantity perturbation variables, we use D.
But traditionally most scalar quantity perturbation variables have been written without using D:
Ya := Da ,
g := D ,
P L := DP ,
Q Eg := DQ .
(2.45)
This Ya has been called the SasakiMukhanov variable [25,26].

In terms of the gauge independent variables defined above, we give the evolution equations
of cosmological perturbations. From (2.21), the background and the perturbation parts can be
written as
U
+ S = 0,
+ 3H +
(2.46)
k2
k 2
L1 (DT , A) = 2 D 2 ,
(2.47)
a
a H
where

2U
U
L1 (DT , A) = (D) + 3H (D) +

(2.48)
D + DS A + 2
+ S A.
As for the fluid components, T = Q gives the background equations as
= 3H h + Q ,
(2.49)
and the perturbation equations as

L2 (DT , A) =
k2
h ( Z ),
a2H
(2.50)
h Z
H
2
a
Q
Z,
+ 3h Z + h A + DP T = Fc +
3
k
H
27
(2.51)
where
L2 (DT , A) = (D ) + 3H D + 3H DP Q A DQ .
(2.52)
G
= T gives the background equations as
2
,
3
= 3H h,
3
H = (1 + w)H 2 ,
2
and the perturbation equations as
H2 =
(2.53)
(2.54)
(2.55)
k2
,
3a 2 H 2
2
k2
L4 (DT , A) = T 2 ,
3
a
3
A + (1 + w)Z = 0,
2

2
1 a
= 2 a 2 T ,
A+
a H
k
L3 (DT , A) = 2
(2.56)
(2.57)
(2.58)
(2.59)
where
L3 (DT , A) = 2A + D,
L4 (DT , A) = H A + 2H A
(2.60)
2
(D + DP ).
(2.61)
2
The dynamical perturbation variables are classified into two groups, that is, what has analogy
with the exactly homogeneous perturbations and what is not related with the exactly homogeneous perturbations at all. The dynamical perturbation variables of the former type are DT
representing the scalar quantity T = (, P , , Q, S) perturbation in the flat slice, D T and the
metric perturbation variable A. The dynamical perturbation variables of the latter type are the
Newtonian gravitational potential and ZA , FcA , T A which have vector or tensor origin.
In the above Li (i = 1, . . . , 4) equations, the former type dynamical perturbation variables are
contained in the left-hand side while the latter type perturbation variables are collected in the
right-hand side. The exactly homogeneous perturbations DT and A corresponding to DT
and A, respectively are constructed as

T
T

(DT ) :=
(2.62)
R ,
C t H

R
A :=
(2.63)
,
H

1 a
R :=
(2.64)
,
a C t
28
where C is the solution constant of the background solution and the subscript t implies that
the derivative with respect to C is performed with the cosmological time t fixed. On the other
hand, the dynamical perturbation variables of the latter type such as , ZA , FcA and T A do not
have exactly homogeneous counterparts. The evolution equations of cosmological perturbations
containing Li (i = 1, . . . , 4) have analogy in the exactly homogeneous perturbation equations. In
fact, the variations of the exactly homogeneous equations (2.46), (2.49), (2.53) and (2.55) give

Li DT , A = 0 (i = 1, . . . , 4),
(2.65)
respectively. The only difference between the exactly homogeneous perturbation Li (i =
1, . . . , 4) equations and the actual k = 0 cosmological perturbation Li (i = 1, . . . , 4) equations
is that k 2 terms exist in the latter but k 2 terms do not exist in the former. Then the effect of
the source term k 2 is corrected in the following way. In performing the correction process, it is
important to notice that the source terms k 2 can be represented in terms of conserved quantity
which has origin in the universal adiabatic decaying mode. In fact, as for f defined by

1
k2
3
a,
f = a H A + g =
(2.66)
2
3H
using (2.50), (2.56), (2.57) yields
1
df
= a 3 H 2 wT + ak 2 (1 + w)Z.
dt
2
When we assume that for k 0 limit
T 0,
kZ 0,
(2.67)
(2.68)
are satisfied, the quantity f is conserved, whose value is written as c. Therefore for k 0 limit,
3H
c = O(1).
(2.69)
a
This expression of is well known as that of the universal adiabatic decaying mode [4]. In
the Li (i = 1, . . . , 4) equations containing DT , A, the Newtonian potential appears only
in the form k 2 , that is, accompanied by k 2 . When we assume that DT = O(1), A = O(1),
k 2 behaves as O(1). Since in the linear perturbation, the scale of the perturbation variables is
arbitrary, the fact that = O(1/k 2 ) does not imply the breakdown of the linear perturbation. If
one wants to get = O(1), one simply assumes that DT = O(k 2 ), A = O(k 2 ). But as explained
later, we cannot assume that is vanishing, since c defined by (2.69) must satisfy the constraint
(2.80). Therefore in the k 0 limit where (2.68), (2.69) are satisfied, (2.47), (2.50), (2.56),
(2.57) can be written as
k2
3 a
c,
a3
3h
L2 (DT , A) = 3 c,
a
2
L3 (DT , A) = 3 c,
a H
3H
L4 (DT , A) = 3 c.
a
L1a (DT , A) =
(2.70)
(2.71)
(2.72)
(2.73)
29
It can be verified that above four sets of Eqs. (2.70), (2.71), (2.72) and (2.73) are satisfied by
g
3
A = (1 + w)g + ,
2
H
T
DT = g,
H
where
1
g = c dt 3 .
a
(2.74)
(2.75)
(2.76)
t0
This special solution for A = (DT , A) is written as A . Since the variation of the exactly homogeneous solution A satisfies (2.65), the general solutions of (2.70), (2.71), (2.72), (2.73)
A = (DT , A) can be expressed as
A = A + A .
(2.77)
The perturbation equations except Li (i = 1, . . . , 4) equations have vector origin, that is, they
are derived from the space component of the Einstein equations. Therefore these perturbation
equations do not have any analogy with the exactly homogeneous perturbation equations. As
explained in the paper [4], these perturbation equations determine the evolutions of the dynamical
perturbation variables which have vector or tensor origin, that is, which have no correspondence
with the exactly homogeneous perturbations, or give the constraint which should be satisfied
in order that the exactly homogeneous perturbations become the k 0 limit of evolutions of
cosmological perturbations. Therefore (2.51), (2.58) can be interpreted as the decision of the
evolution of the variables Z which is not related to the exactly homogeneous solution at all in
terms of A = (DT , A), the constraint to the exactly homogeneous perturbations, respectively.
Integrating (2.51) yields

H
a
Q
Z .
h Z 3 C + dt a 3 h A DP Fc +
(2.78)
k
H
a
t0
By summing (2.78) with respect to all the fluid components, we obtain

H
3
(hZ)f = 3
C + dt a hf A (DP )f S D
a
t0

H
2 a3
2 a3
= 3
C +
A + a 3 D
A + a 3 D
. (2.79)
3H
3H
a
0
Therefore (2.58) gives the constraint between C , c defined by (2.69), and 2NS + Nf solution
constants of the exactly homogeneous perturbation as

2
2 a3
C + 2 c
A + a 3 D = 0.
(2.80)
3H
Integrating (2.59) gives

H
Ct dt aA ,
=
a
t0
(2.81)
30
where the first term containing Ct is well-known universal adiabatic decaying mode [4] and by
comparing with (2.69) we obtain
3
c.
(2.82)
k2
If we assume c = 0, since (2.80) gives one constraint relation, we obtain 2NS + 2Nf 1 solutions
and the Newtonian potential is obtained by (2.81) with Ct = 0:
H
dt aA.
=
(2.83)
a
Ct =
t0
If we assume that c is nonvanishing, since c = O(1), A O(1), therefore

A
O k2 ,
Ct
we obtain the k 0 limit of the universal adiabatic decaying mode [4]:
(2.84)
H
1 2
k c,
(2.85)
3
a
which is consistent with (2.69). Then we have obtained the long wavelength limit of all the
solutions to the evolution equations of cosmological perturbations.
3. Use of the scale factor as the evolution parameter
As the gauge invariant variable representing the fluctuation of the scalar quantity T , we adopt
DT defined by (2.5) which represents the T fluctuation in the flat slice, since it is the easiest to
see the correspondence with the exactly homogeneous perturbation of T . While until now we
described the exactly homogeneous variables as functions of t , C where t is the cosmological
time and Cs are solution constants, we can describe the exactly homogeneous variables as functions of a, C where a is the scale factor. For an arbitrary scalar quantity such as , P , S, Q, ,
from (2.62), (2.64), (DT ) can be written as the partial derivative of the corresponding exactly
homogeneous scalar quantity T with respect to solution constant C with the scale factor a fixed:

T

(DT ) =
(3.1)
.
C a
Since

1
1
1
=
a =
(D) = A ,
a C a 2 C a 2
this property of D also holds as for the time derivative of the scalar quantity T :

(D T ) =
T .
C
a
(3.2)
(3.3)
Therefore the operator D defined by (2.5) can be interpreted as a kind of derivative operator, that
is, the derivative with respect to the solution constant C with a fixed. Because of this derivative
we can
property of D, as for the scalar quantities T = (, P , S, Q) which are functions of , ,
understand
T
T
DT =
(3.4)
D ,
D +
31
easily. The contribution to D T from the adiabatic decaying mode is given by the same form as
that of DT :
T
g,
(3.5)
H
where g is defined by (2.76).
As seen from the above discussion, in order to derive the long wavelength limit of cosmological perturbations from the corresponding exactly homogeneous system by using the LWL
formula, it is more appropriate to use as the evolution parameter the scale factor a than the
cosmological time t . For example, as for the scalar-fluid composite system, the corresponding
exactly homogeneous expressions are obtained by solving the first order differential equations
setting a , pa := a 3 a , as independent variables and the scale factor a as the evolution parameter:
(D T ) =
d
1 pa
,
a =
da
H a3
a 3 U
a3
d
Sa ,
a pa =
da
H a
H
Q
d
,
a = 3h +
da
H
replacing H with the right-hand side of the Hubble law:

2 1 2
p
+
U
()
+
.
H2 =
3 2a 6 a a
(3.6)
(3.7)
While under use t as the evolution parameter our system is the constrained system with the
Hamiltonian constraint (3.7), under use a as the evolution parameter our system becomes the
unconstrained system with respect to independent variables a , pa := a 3 a , .
For some time, we consider the system consisting of multiple scalar fields a only. Since the
evolutions of a , pa := a 3 a can be described in terms of the Hamilton equations of motion,
the evolutions of the corresponding perturbation variables Ya = Da , Pa := a 3 D a can also be
written in terms of the Hamilton equations of motion:
H
dYa
,
=
dt
Pa
dPa
H
,
=
dt
Ya
(3.8)
whose Hamiltonian is given by

a3
2
1
a b Pa Yb ,
H = 3 Pa Pa + Vab Ya Yb +
2
2H
2a

U
2U
3 2
2 U
k2
+
a b +
+ 2 ab .
Vab =
b + a
a b
2
2H a
b
a
(3.9)
(3.10)
Although the evolutions of Ya , a 3 Ya can be formularized as the Hamiltonian dynamical system,

Ya , Pa := a 3 D a are more convenient to see the correspondence with the long wavelength limit
expressions of cosmological perturbations given in terms of the derivative of a , pa := a 3 a with
respect to the solution constant with a fixed than Ya , a 3 Ya . Therefore this set of variables Ya ,
Pa := a 3 D a is useful as the starting point from which to derive the finite wavelength correction
around the long wavelength limit calculated in the previous papers [7,8]. The connection between
32
the old canonical variables Ya := Ya , Pa := a 3 Ya and the new canonical variables Ya , Pa :=

a 3 D a are given by the canonical transformation defined by the generating function:
3 a3H
a b Ya Yb .
W = Ya Pa +
4
(3.11)
When we treat the oscillatory scalar fields, the action angle variables Ia , a are useful [7,8]:

2Ia
1
a = 3/2
cos a ,
ma
a

pa = a 3/2 2ma Ia sin a ,
(3.12)
where ma is the mass of the scalar field a . The action angle variables obey the evolution equation
as

d
a 3 Uint a 3/2 2Ia
a Ia =
(3.13)
+
sin a Sa + 3Ia cos 2a ,
da
H a
H
ma
a
d
ma a 3 Uint a 3/2
3
1
+
cos a Sa sin 2a .
a =
+
da
H
H Ia
H
2
2ma Ia
(3.14)
In order to investigate the cosmological perturbations in the universe containing oscillatory scalar
fields, by using D defined in the above we define the action angle perturbation variables DIa ,
Da starting from Ya := Da , Pa := a 3 D a . In the universe dominated by scalar fields only, both
Ya , Pa and DIa , Da are pairs of canonical conjugate variables. These perturbation variables
pairs are appropriate in order to interpret the effect of the dissipation from the scalar fields to the
perfect fluids as the deviation from the Hamiltonian dynamical system. DIa , Da are defined by
the following expressions:

2Ia
1
Ya = D 3/2
cos a ,
ma
a

3/2
2ma Ia sin a ,
Pa = D a
(3.15)
where D in the right-hand side is interpreted as

D=
DIa
+
Da
.
I
a
a
a
a
(3.16)
The expressions obtained from variations of Ia , a with a fixed in the previous papers [7,8] are
the long wavelength limits of DIa , Da defined by (3.15). In fact, the LWL formulae

Ia
Ia
+
DIa =
(3.17)
3Ia g,
C a
H

a
a
+ g,
Da =
(3.18)
C a H
where g is defined in (2.76) hold. While the third term in the right-hand side of (3.17) appears
because of the scale factor a dependence of the transformation law from a , a to Ia , a , the
parts of (3.17) and (3.18) reflect the fact that D is the derivative operator with respect to the
solution constant with the scale factor a fixed.
33
In order to solve the dynamics of the system containing the oscillatory scalar fields, we are
required to perform the averaging over the fast changing angle variables a [7,8]. If we use the
cosmological time t as the evolution parameter, our system is a constrained system. Therefore we
must check that our averaging procedure is consistent with the constraint and this process is rather
cumbersome. But if we use the scale factor a as the evolution parameter, our system becomes
unconstrained system, so the definition of the averaging procedure becomes rather simple.
We can conclude that use of the scale factor a as the evolution parameter brings about the two
merits. One is that it becomes easier to use the LWL formula and the other is the more simple
definition of the averaging process.
4. Application of the LWL formula to the non-interacting multicomponent system
Based on the results obtained in Sections 2, 3, we consider the long wavelength limit of the
evolutions of cosmological perturbations in the universe consisting of multiple cosmic components. For simplicity, we consider the case where each component does not interact with each
other.
We consider the w fluid where w is constant and which does not interact with other components Q = 0. is solved as
=
A
,
3(1+w
)
a
(4.1)
where A is a solution constant and by differentiating with respect to A with the scale factor a
fixed, we obtain
A
(D ) = 3(1+w ) ,
(4.2)
a
where A is a perturbation solution constant corresponding with A . Therefore we obtain
A
(4.3)
= const.
A
Next we consider the case that the oscillatory scalar field a does not interact with other cosmic
components, that is Uint = 0, Sa = 0. Since in such case the right-hand side of (3.13) is oscillatory
function depending on the angle variable a with vanishing mean value, by taking the averaging
over a , we obtain
g =
Ia
= Aa ,
(4.4)
where Aa is constant. The estimate about the effects of the oscillations due to the fast changing
angle variables a was discussed in the previous papers [7,8]. Therefore by taking the derivative
with respect to solution constant Aa with the scale factor a fixed, we obtain
DIa
= Aa ,
(4.5)
where Aa is a perturbation constant corresponding with Aa . Since

ma I a
ma DIa
(4.6)
,
Da =
,
a3
a3
where the above second expression is given by the D operation to the above first expression, we
obtain
a =
ga =
DIa Aa
= const.
=
Ia
Aa
(4.7)
34
As seen from the above expressions, as for the oscillatory scalar field a the energy density a

and the energy density perturbation a ga behave like those of the dust fluid. We can summarize

that for non-interacting cosmic components, g , ga are conserved.
We consider the evolution of the Bardeen parameter Z in the universe dominated by noninteracting cosmic components. Following the paper [27], we define as the gauge invariant
variable representing the curvature perturbation in the uniform density slice:
:=
H
H
D = R .
(4.8)
From (2.56), (2.58) we can see that the Baredeen parameter Z and are closely connected as
k2
2 1
,
9 1 + w a2H 2
whose k 0 limit is
Z=
(4.9)
2 1
c
(4.10)
,
3 1 + w a3H
where c is constant related with the adiabatic decaying mode defined by (2.69). While the
Bardeen parameter Z is expressed as the weighted sum of ZS , Z which do not related with
the exactly homogeneous quantity at all and whose evolution is written in the rather cumbersome integral form (2.78), is connected with the corresponding exactly homogeneous quantity
and evolution can be written in terms of the derivative of the total energy density with
respect to solution constant. Then we consider much easier evolution.
As for the multicomponent non-interacting fluids system, we obtain

A
A
(1
+
w
)
3
=
(4.11)
3(1+w ) .
a 3(1+w )
a
Z=
We can see that is exactly conserved for the adiabatic growing mode defined by
A
1
=
A 1 + w
independent.
(4.12)
For some time, we consider the two components system consisting of dust and radiation. In
this case, (4.11) is reduced to
=
aAd + Ar
,
3aAd + 4Ar
(4.13)
where the suffixes d, r imply dust, radiation, respectively. We obtain

init =
1 Ar
,
4 Ar
fin =
1 Ad
,
3 Ad
(4.14)
in the limit a 0, a , respectively. We adopt different, more physical parametrization:

Ad = 3 Ad Ad ,
Ar = 4 Ar .
(4.15)
represents the adiabatic growing mode and represents the isocurvature mode defined by
init = 0,
3

Srd = gr gd = const =: .
4
(4.16)
35
Then we obtain
1

fin = init ,
(4.17)
3
which is the famous formula [28].
Although in the paper [27], (4.13) has already been derived essentially without using the LWL
formula, our result is more rigorous in the point that we treat the contribution from the adiabatic
decaying mode characterized by c defined by (2.69) more appropriately, while the paper [27]
simply assumes that k 2 is vanishing. In this section, the consideration of the oscillatory scalar
fields is also contained unlike the paper [27].
5. Application of the averaging method to the decaying scalar fields
We derive the evolution equations of the multiple scalar fields decaying into the multiple radiation fluids. By solving these evolution equations and taking the exactly homogeneous perturbations, we can obtain the information of the evolutionary behaviors of cosmological perturbations
during reheating and in the curvaton model. We assume that the source Sa is given by
Sa = a a .
(5.1)
We nondimensionalize the dynamical quantities as

Ia
Ia ,
I0
a
a,
a0
,
0
Uint
Uint = O(),
0
(5.2)
and the parameters as

ma
ma ,
m0
a
a ,
0
(5.3)
where is the small parameter implying the ratio of the interaction energy to the free part energy
defined by
0 :=
m0 I0
a03
(5.4)
Then we obtain the dimensionless parameters
1/2
30
0
,
:= 1/2 ,
:=
m
3 0
(5.5)
which imply the ratio of the Hubble parameter to the mass of the scalar fields H /m, the ratio of
the decay rate to the Hubble parameter /H at the initial time a = a0 , respectively. By using the
above dynamical variables and parameters, the evolution equations can be expressed as
d
Ia
1 a 3 Uint
a 1/2 (1 cos 2a ) + 3Ia cos 2a ,
Ia = 1/2
da

a
1 ma
1 a 3 Uint 1 a
3
d
+
1/2 sin 2a sin 2a ,
a a =
1/2
1/2
da

Ia
2
2
a
d
a ma Ia (1 cos 2a ),
a = 1/2
da
a
a
(5.6)
(5.7)
(5.8)
36
where is defined by
(5.9)
,
a4
and a is the decay rate from the scalar field Ia to the radiation component and therefore a
is given by

a .
a =
(5.10)
=
In this paper, we investigate the evolutionary behavior of cosmological perturbations during the
period when the decay rates from the scalar fields to the radiation fluids are large compared to
the interaction between the scalar fields, that is /, while in the paper [8] the evolutions
of cosmological perturbations during the period when the interaction between the scalar fields is
dominant / were discussed, which is thought to give the initial conditions for the present
studies.
Next we show that there exists a transformation such that in a system obtained by that transformation, the dynamics of the action variables Ia and the radiation energy densities can
be determined independently of the angle variables a . In order to show this statement, we put
several assumptions.
(i) The interaction energy of the scalar fields Uint is analytic with respect to the dynamical
variables Ia , a and is 2 periodic with respect to a . Uint is bounded as
|I |,
a 9/2
which implies that there exists a positive constant M such that

9/2
a Uint M|I |,
Uint
where
|I | :=
|Ia |.
(5.11)
(5.12)
(5.13)
(ii) As for the total energy density

ma Ia
=
+
+ Uint ,
a3
a4
a
the action variables Ia and the radiation energy densities satisfy

ma Ia +
c2 ,
c1 a
a
(5.14)
(5.15)
for some positive constants c1 , c2 .

Note that the evolution equations of the system can be written as
d
Ia = Fa (I, , , a),
da
d
a = F (I, , , a),
da
(5.16)
(5.17)
d
1
a = a (I, , a) + Ga (I, , , a),
da

37
(5.18)
where Fa , F , Ga are analytic with respect to the dynamical variables Ia , , a and 2 periodic
with respect to the angle variables . We say that the evolution equations is of the type Ck :
(i) The averaged parts of Fa , F , Ga are bounded as
Fa
a 2 |I |,
(5.19)
F
a |I |,
(5.20)
Ga
a ,
(5.21)
where A
implies the averaging over the angle variables a :
1
A
:=
(2)NS
2
d NS A.
(5.22)
In case of the resonant case, it is prescribed that the averaging is performed with respect to
the fast angle variables only [8].
(ii) The oscillatory parts of Fa , F , Ga are bounded as
Fa k a 2 |I |,
F k a 3 |I |,
(5.23)
a k a2,
G
(5.25)
(5.24)
where A implies the residual part after the averaging over the angle variables a :
A := A A
.
(5.26)
Under this notation, the following proposition holds.

Proposition 1. Let k be some non-negative integer and consider the evolution equations as

d (k)
Ia = Fa(k) I (k) , (k) , (k) , a ,
da

d
a (k) = F(k) I (k) , (k) , (k) , a ,
da

(k) (k) (k)

d
1
a a(k) = a(k) I (k) , (k) , a + G(k)
a I , , ,a .
da

a
(5.27)
(5.28)
(5.29)
Suppose that this set of evolution equations is of the type Ck , there exists a transformation
(k+1) (k+1) (k+1)
Ia(k) = Ia(k+1) + u(k)
(5.30)
,
,
,a ,
a I

(k)
(k+1)
(k) (k+1)
(k+1) (k+1)
=
(5.31)
+ u I
,
,
,a ,

(k)
(k+1)
(k) (k+1)
(k+1) (k+1)
a = a
(5.32)
+ va I
,
,
,a ,
satisfying the following conditions:
38

(k)
(k)
(k)
(k+1)
(k+1)
(i) ua , u , va are analytic with respect to the dynamical variables Ia

,
(k+1)
are 2 periodic with respect to the angle variables a
, and are bounded as

(k)
k+1 (k+1)
ua
I
,

(k)
k+1 (k+1)
,
u a I
(k)
va
k+1 .
(5.33)
(5.34)
(5.35)
(k+1)
,
Ia
(ii) The evolution equations of the transformed variables

Ck+1 and the changes of Fa
, F
, Ga
are bounded as

Fa
k+1 a 2 I (k+1) ,

F
k+1 a 3 I (k+1) ,
Ga

(k+1)
, a
k+1 2
a ,
where A
is defined by

A
:= A(k+1) I (k+1) , (k+1) , (k+1) , a

A(k) I (k+1) , (k+1) , (k+1) , a .
(k+1)
(k+1)
, a
are of the type

(5.36)
(5.37)
(5.38)
(5.39)
(For the proof, see Appendix A.)

This proposition implies that we can make the part depending on the angle variables arbitrarily small by taking the original set of the evolution equations as the starting point k = 0 and
applying transformations given in the proposition iteratively. Therefore it can be expected that
the evolution of the dynamical variables can be described by the truncated system obtained by
discarding the angle variable dependent part with sufficiently good accuracy, if we take sufficiently large k. By estimating the errors produced by the truncation, we show that the above
expectation is correct. We use the symbol to represent the difference of a quantity for the exact
system and a corresponding quantity for the truncated system. For A = (Ia , , a ), the errors of
the background variables are written as
A := A Atr ,
(5.40)
and the errors of the perturbation variables are written as

A := A Atr ,
(5.41)
where A, A represent quantities of the exact system and Atr , Atr represent quantities of the
corresponding truncated system. For a function f (a), let us write

f (a) = E a 2 ,
(5.42)
when f (a) is bounded as

f (a) p(a) exp a 2
(5.43)
for a polynomial of a:
f (a) by

f (a) := sup
(5.44)
p(a), and for a positive number . For a function f (a), let us define
1a a

f (a ).
When we write all the inequalities, it is prescribed that all the coefficients of order unity are
omitted. The truncation error for the mth order system can be estimated as follows.
39
Proposition 2A. Let m be an integer larger than or equal to 2. For the mth order system, the
truncation errors of the background variables are given by

|I | E a 2 m ,
(5.45)
(a) m ,
(5.46)
| | a
(5.47)
2 m1
and the truncation errors of the perturbation variables are given by

|I | E a 2 m1 A1 (1),
| |
m1
|| a
A1 (1),
2 m2
A1 (1) +
m1

exp a 2 m a 2 (1) + a 4 Am (1) ,
(5.48)
(5.49)
(5.50)
where

Am (1) := I (1) + (1) + m (1),
(5.51)
under the initial conditions

I (1) = (1) = (1) = 0,
(5.52)
I (1) = (1) = (1) = 0.
(5.53)

By the transformation laws, the errors of the mth order variables affect the original variables
as shown by the next proposition.
Proposition 2B. The difference between A(0) obtained from A(m) by the transformation laws and
(0)
(m)
Atr obtained from Atr by the same transformation laws has upper bound
(0)

I E a 2 m ,
(5.54)
(0)
m ,
(5.55)
(0)
2
m1
a
,
(5.56)
and the corresponding difference as for the perturbation variables has upper bound

I (0) E a 2 m1 A(m) (1),
1

(0) m1 A(m) (1),
1

(0) a 2 m2 A(m) (1) + m1 exp a 2 m a 2 (m) (1) + a 4 A(m) (1) ,
m
1
(5.57)
(5.58)
(5.59)
under the same initial condition as in the previous proposition.

According to the above proposition, we can conclude that we can make the truncation errors
as for the original variables arbitrarily small if we truncate the system at the arbitrarily large mth
order system.
From Proposition 1, we can see that the part independent of the angle variables of the evolution equations are shifted after the transformations reducing the part dependent on the angle
variables. By the truncation, our system becomes much simpler than the original system. But
40
it is still difficult to solve the truncated evolution equations with such correction terms because
the evolution equations are complicatedly entangled with each other. In order to solve the evolution equations analytically, we want to discard such correction terms. The errors produced by
discarding such corrections are evaluated in the following proposition.
Proposition 3. The difference between the system with the correction terms produced by the
transformation and the system obtained by discarding such correction terms is evaluated in the
following way. As for the background variables, the discard errors are evaluated as

|I | E a 2 ,
(5.60)
| | ,
(5.61)
| | a ,
(5.62)
and as for the perturbation variables, the discard errors are evaluated by

|I | E a 2 B(1),
(5.63)
| | B(1),
(5.64)
|| a B(1),
(5.65)
where

B(1) := I (1) + (1),
(5.66)
under the initial conditions

I (1) = (1) = (1) = 0,
(5.67)
I (1) = (1) = (1) = 0.
(5.68)

By the above propositions, it can be understood that we can obtain the information of the original system with sufficiently good accuracy by investigating the evolution equations by simply
dropping the part dependent on the angle variables, because the errors produced by dropping are
sufficiently small and in particular , are bounded. The reason why these errors are mild
is that the final state in which all the energy of the scalar fields is completely transferred into that
of radiation fluids is the attracting equilibrium around which the perturbations do not grow.
6. Application of the LWL formula to the multicomponent reheating model
In this section, we apply the LWL formula to the reheating where the energy of the multiple
scalar fields is transferred into that of the multiple radiation fluids. The decay rate from the scalar
field a to the radiation fluid is given by a . When the interactions between the scalar fields
a : Uint are neglected, the background quantities are solved as

1
ma Ia = Aa exp a da 1/2 ,
(6.1)
a
1

a ma Ia ,
= B + da 1/2
(6.2)
a
1
41
where Aa , B are integration constants. As long as we do not give the expression of the total
energy density in the integrals, the above solutions do not give any physical information of reheating. But it is difficult to solve the evolution equations in the form where the exact expression
of is explicitly described, because the evolution equations of Ia , are complicated and highly
nonlinear. Then we expect that the contribution to the integrations owes mainly to the period
when the energy of the scalar fields is dominant, that is can be approximated as
=
B
A
+ 4,
3
a
a
where
A :=
Aa ,
(6.3)
B :=
B .
(6.4)
We assume that the initial radiation energy density B is negligibly small B A. For simplicity,
in all the following discussions in Sections 6, 7, we assume that all the decay rates a defined
by (6.6) are of the same orders of magnitude. By substituting the above expression to the
solutions (6.1), (6.2), by expanding the solutions with respect to B around B = 0, we obtain

Aa B 1/2
2 a 3/2
2 a 3/2
a
a
exp
a
+
+ O B2 ,
ma Ia = Aa exp
a 3/2
1/2
1/2
3A
3A
A

4
4
a r = a
2/3

3
A1/3 Aa
1
3
= B 1 + G(2) G(1) +
G(5/3) 2/3
+ O B2 ,
2/3
2
2
2
a a
where
a :=
a ,
(6.5)
(6.6)
and G(t) is the Gamma function defined by
G(t) :=
dx x t1 ex ,
(6.7)
which is convergent for t > 0. By taking the exactly homogeneous perturbation of (6.5) defined
by

,
+ B
D := A
(6.8)
A
B a,B=0
we can obtain the Bardeen parameter in the final state a :
1 Dr
4 r

Aa
1 2 2/3 2/3
3
1
=
B
1
+
G(5/3)
G(2)
G(1)
2/3
4 3
2
2
A1/3
a a

Aa Aa
1 1 A Aa
+
+
.
2/3
2/3
4 3 A a a2/3
a a
a a
fin
=
(6.9)
42
From now on, we name the expressions representing the final amplitude of the Bardeen parameter

fin in terms of the initial perturbation amplitudes such as Aa , B , S formula after S matrix in
the quantum mechanics. As for the initial energy density perturbations of the scalar fields Aa ,
by adopting more physical parametrization introduced by
1 A
=: ,
3 A
Sab =
Aa Ab
=: ab ,
Aa
Ab
Aa can be written as
Aa Ab
Aa = 3Aa +
ab .
A
(6.10)
(6.11)
represents the adiabatic growing mode and ab represent the isocurvature modes. Since ab
satisfy
ab = ba ,
ab + bc = ac ,
(6.12)
the independent quantities are given by , 12 23 NS 1NS . By using this parametrization, Aa

dependent part of fin is written as

A
1
1
1
Aa Ab
a
fin +
(6.13)
ab
2/3
2/3
2/3
8
A
a
a
a
ab
b
where A B implies that B is contained by A, that is A = B + . From this S formula, we
can conclude that for the adiabatic growing mode , the Bardeen parameter is conserved, and
that the initial entropy perturbations survive in the case that the decay rates are dependent on the
scalar field a from which the radiation energy comes, that is a = b (a = b). In the case where
multiple scalar fields exist, there is no reason why the perturbation has only the adiabatic component, and it is natural to think that in the perturbation the adiabatic components and the entropic
components coexist. In such mixed cases, so-called conservation of the Bardeen parameter does
not hold and the above S formula gives useful tool for calculating the final Bardeen parameter.
We can consider the case where the energy transfer rates a fluctuates [29], which is called
as the modulated reheating scenario [30]. For simplicity, we consider the one scalar field case.
By taking the derivative of (6.5) with respect to with B vanishing, we obtain
1
(6.14)
,
6
which is well-known formula derived in the paper [30], and where the coefficient is successfully
determined in this paper.
We consider the influence of the resonant interaction between scalar fields on the final amplitude of the Bardeen parameter. We take the interaction Uint into account by iteration. As the first
order correction from the interaction term is
fin
1 a 3 Uint
d
a Ia ,
1/2 a
da
we obtain
da
1

1/2
a (ma Ia ),
(6.15)
(6.16)
where
(ma Ia ) := exp a da
43
1
1/2 a

ma
1
1 a 3 Uint
da
.
exp a da 1/2
1/2
a

a
a
(6.17)
Since as the zeroth order approximation Ia obeys (6.1), we can write

Uint Uint
1
1
=
exp
da
,
a
a a=1 a 3n/2
1/2 a
(6.18)
where we assumed that Uint contains an nth order interaction term and that is the appropriate
sum of a /2. By substituting the above expression and by expanding the correction term with
respect to B around B = 0, we obtain

n8/3

Uint
1
2
1
a ma
G(5/3, n + 3, a , )
a

1/2
a a=1

3A
a

2/3 1/3
B
1
+
( a )G(5/3, n + 10/3, a , )
A
2
A1/2
1/3

3
1 2 2/3
G(5/3, n + 7/3, a , ) +
a G(2, n + 3, a , )
2 3
2

1 2 2/3
G(1, n + 3, a , ) ,
(6.19)
2 3
where
G(n1 , n2 , 1 , 2 ) :=
dx x
n1 1
x0
x
exp(1 x)

dy y n2 1 exp (1 2 )y ,
(6.20)
x0
where
2
2 3/2
x0 := 1/2 .
a ,
(6.21)
A1/2 3
A 3
The evaluation of the double Gamma function defined by (6.20) is treated in Appendix B. We
consider the concrete example defined by
x :=
Uint = 12 22 ,
m1 = m2 .
(6.22)
In this case, we use the independent variables defined by

1 = q0 ,
2 = q 0 + q1 ,
I1 = p0 p1 ,
I 2 = p1 ,
(6.23)
where (q0 , p0 ) and (q1 , p1 ) are called fast and slow action-angle variables, respectively [8].
Because of the resonant relation satisfied by the masses of the scalar fields, the slow angle
variable q1 moves much more slowly than the fast angle variable q0 . The averaging over the
slow angle variable q1 cannot be justified, and therefore the slow action-angle variables (q1 , p1 )
can have evolutions. In the previous paper [8], we investigated the influences of the resonant
interaction on the evolution of the cosmological perturbations before the energy transfer from
44
the scalar fields to the radiation fluids begins. According to this study, the slow action-angle
variables can suffer from the instability near the hyperbolic fixed point in the phase space of the
slow action-angle variables. Since the initial adiabatic perturbation and the initial isocurvature
perturbation 12 are given by
=
1 p0
3 p0
(6.24)
and
12 =
p1 p0 p0 p1
,
(p0 p1 )p1
(6.25)
respectively, the instability of the slow action variable p1 has influence on the isocurvature mode.
Therefore from (6.13) in the case 1 = 2 the instability of the action-angle variables survives
in the final amplitude of the Bardeen parameter. Next we calculate the first order correction term
(6.19) in the present model (6.22). The present model has the hyperbolic fixed point at
2p1 = p0 = c,
q1 = (2k + 1),
(6.26)
2
where k is an integer. At this hyperbolic fixed point, the first order correction to the final amplitude of the Bardeen parameter (6.19) is calculated as

Aa
1 1 2
1
1

q
(1)
,
c
fin
(6.27)
1
2/3
2/3
2/3
12 A1/2
1
2
a
where
:=
I0
(6.28)
m30 a03
and the non-dimensional masses are scaled as m1 = m2 = 1. In the present model, as for the first
order correction term also, in order that the slow action-angle variables instability has influence
on the final Bardeen parameter, 1 = 2 is necessary.
Until now, we evaluate Ia , by assuming that is given by (6.3). Now we evaluate the
contribution to Ia , from the late stage of reheating when is given by
2/3
3
1 A1/3 Aa
G(5/3) 4 2/3
.
late =
(6.29)
2/3
2
a
a a
Such late stage of reheating begins at
2/3
Aa
3
1
G(5/3) 2/3 2/3
,
a1 = d
2/3
2
A
a a
(6.30)
because at this a1 , A/a 3 is almost equal to late . d is a numerical factor which can be assumed to
be larger than unity. By using late , a1 , we can evaluate the contribution to r from the late stage
of reheating as

a Ab 3/2
1 3 2/3
G(5/3)
d exp d 3/2 G(5/3)3/2 3/2
r 4
2/3
a 2
A
a
b b
Ab
Aa
2/3 2/3
(6.31)
,
2/3
A
whose size is characterized by

rl/e := d exp d 3/2 G(5/3)3/2 ,
45
(6.32)
which is the ratio of the late contribution to the main early contribution to r . The value of rl/e
is 0.43, 0.18 and 0.035 for d = 1, 2 and 3, respectively. Therefore we can conclude that the
S formula which is derived by using (6.3) is rather good approximation to the real S formula.
Finally we consider the case where the decay rates a depend on the radiation temperature T .
In the high temperature limit T m where m is the mass scale of the oscillatory scalar fields, the
decay rate a depends on the radiation temperature [21]. When the decay product is the fermion,
a is given by
1
.
T
When the decay product is the boson, a is given by
a = a
(6.33)
a = a T .
(6.34)
According to the paper [21] the reason is following. We consider the case where r T 4 is
sufficiently high. In the fermion case, the Pauli exclusion principle inhibits the decay of a into
fermions since the fermions have already occupied the energy levels into which a would decay.
In the boson case, the induced effect promotes the decay of a into bosons since the bosons
occupy the energy levels into which a decay. For simplicity, we consider the case where the
radiation consists of one component. We interpret that the radiation temperature T appearing
in (6.33), (6.34) is the temperature Ta := [r (a(a ))]1/4 at the time when the decay process
proceeds given by
a(a ) :=
A1/3
2/3
2/3 a
In the fermion case, by substituting Ta defined above to (6.33) it can be verified that

Aa
2/9 Aa 10/9
=
.
2/3
A1/9 a a2/5
a a
Therefore we obtain

A2/9 Aa 10/9
,
a 4 r 4/9
2/5
a a

A
5 1
1
Aa Ab
a

fin +
ab
2/5
2/5
2/5
36
A
b
a a
ab a
In the same way as in the fermion case, in the boson case we can obtain

A2/3 Aa 2/3
,
a 4 r 4/3
a2
a

A
1 1
1 Aa Ab
a

fin +
ab
2
12
A
a2 b2
a
a
(6.35)
(6.36)
(6.37)
(6.38)
(6.39)
(6.40)
ab
The radiation temperature dependence of the decay rate a affects how the isocurvature modes
are transmitted into the final amplitude of the Bardeen parameter.
46
7. Application of the LWL formula to the multicomponent curvaton model

In this section, we apply the LWL formula to the curvaton scenario where multiple weakly
coupled massive scalar fields called curvatons decay into multiple radiation fluids some time later
after the inflation has ended. In this curvaton scenario, the curvaton fields other than the inflaton
fields driving the inflation are responsible for the origin of the cosmic structures.
First we consider the limit where in the initial time the curvaton fields energy A is small
compared to the radiation fluids energy B: A B. By substituting the expression (6.3) to
(6.1), (6.2) and by expanding it with respect to A around A = 0, we obtain

1 a 2
ma Ia = Aa exp
a + O A2 ,
1/2
2B

B 1/4 Aa
= B + 2G(3/2) 1/2
+ O A2 .
a 4 r = a 4
(7.1)
1/2
a a
By taking the exactly homogeneous perturbation, that is D operation around A = 0, we obtain

Aa
2
1 B
1
1 D
fin =
(7.2)
.
=
+
G(3/2) 1/2 3/4
1/2
4
4 B
4
B
a a
Next we consider the case where the energy densities of the curvatons are large compared with
those of radiation fluids when the energy transfer from the curvatons to the radiation fluids proceeds. In this case, the exponent of (6.1) is written as

a 2 3/2
1
4
a da 1/2 = 2
(7.3)
x 2Bx 1/2 + B 3/2 ,
3
3
a
A
1
where
x := Aa + B.
(7.4)
As for x(a ) defined by

{x(a
)B}/A
da
1
= 1,
1/2 a
(7.5)
we obtain

x(a ) = x0 (a ) 1 + 2

3/2

B
B2
4
B
+O
,
x0 (a ) 3 x0 (a )
x0 (a )2
(7.6)
where
2/3 2 2/3
3
A
x0 (a ) :=
.
2
a
(7.7)
The expansion parameter B/x0 (a ) implies the ratio of the energy of radiations to the energy
of the curvaton a when the energy transfer proceeds. By using x(a ), we approximate the
exponential function by the step function:

1
exp a da 1/2
(7.8)
x(a ) x .
a
1
By using this approximation, we obtain the radiation energy density in the a limit:

1
2 3 5/3 A1/3 Aa
= 4 B +
+
B
+
.
r =
5 2
a
2/3 a a2/3
47
(7.9)
By taking the exactly homogeneous perturbation, we obtain the S formula:

fin =
1 A 1 B
B
+
A + ,
4 A
2 A 2A2
(7.10)
where

2 3 5/3 A1/3 Aa
A :=
.
5 2
2/3 a a2/3
We define more physical parametrization by

Aa Ab
ab ,
Aa = 3Aa + 3Aa +
A
b
B B
B = 4B +
.
B
(7.11)
(7.12)
(7.13)
In particular, since
A = 3A + 3A,
B = 4B,
(7.14)
implies the adiabatic growing mode and means the isocurvature mode between the total
curvatons and the total radiations. By using this parametrization, the S formula can be rewritten
as

B
fin = + 1 2
A

A
1
B
1
Aa Ab
1
a
+ .
+
(7.15)
ab
2/3
2/3
2/3
8 4A
A
b
a a
ab a
In the most simple curvaton scenario of one curvaton field and one radiation fluid, the empirical
S formula was obtained from the numerical calculation [31]:
fin = r(p),

0.924 1.24
r(p) = 1 1 +
,
p
1.24
(7.16)
(7.17)
where
p :=
A
.
1/2 1/2 B 3/4
(7.18)
In the limits p 1, p 1, this empirical S formula gives

fin = 0.924p,

1.44
fin = 1 1.24 ,
p
p 1,
(7.19)
p 1,
(7.20)
48
respectively. Our analytical results (7.2), (7.15) give

fin 0.940p,

2.54

fin 1 1.33 ,
p
p 1,
(7.21)
p 1,
(7.22)
respectively. In the case p 1, the empirical formula and our analytic result agree with good
accuracy. In the case p 1, our analytic result is obtained by rather rough treatment approximating the exponential function by the step function. But our analytic S formula agree well with
the empirical formula at least qualitatively.
8. Discussion
In this paper, we constructed the LWL formula expressing the long wavelength limit of
evolution of cosmological perturbations in terms of the corresponding exactly homogeneous
perturbations in the most general scalar-fluid composite system. We determined the correction
term which corrects the difference between the long wavelength limit of cosmological perturbations and the exactly homogeneous perturbations, and we showed that the correction term
contributes the well-known adiabatic decaying mode. It was pointed out that when we extract
the long wavelength limits of evolutions of cosmological perturbations from the exactly homogeneous variables, the use of the scale factor a as the evolution parameter is more useful. The
scalar-fluid composite system whose LWL formula is constructed in this paper can be used to
describe the early stage of the universe such as reheating after inflation and the curvaton decay in
the curvaton scenario, when the fluid is assumed to be radiation. In this paper, the LWL formula
is applied to the most general case of reheating and of the curvaton decay containing the multiple
scalar fields and the multiple radiation fluids, and the S formulae representing the final amplitude
of the Bardeen parameter in terms of the initial adiabatic and entropic perturbations are constructed. In case where the decay rates a are dependent on the label of the scalar fields a; that
is a = b for a = b, the initial isocurvature modes survive in the final amplitude of the Bardeen
parameter.
Our evolution equations have arbitrary functions S which describe the energy transfer between
scalar fields and perfect fluids. The source functions S can be determined from the microscopic
dynamics between the coherently oscillating scalar fields and radiation, concretely speaking, by
path integrating out the fields constituting the radiation and interacting with the coherently oscillating scalar fields in the effective inin action [21]. When the scalar fields oscillate coherently,
the source term such as S = is important in order that the energy is transferred from the
scalar fields into radiation effectively. As shown in the paper [21], is given as the function
r ) where r is the energy density of radiation.
= (, ,
We consider the case where is a function of the scalar quantities which are not related
to reheating process directly [29]. The scenario where the energy transfer is controlled by the
scalar quantities not related to reheating is called as modulated reheating scenario [30]. As such
scalar quantities, we can choose flat direction scalar field which does not govern the energy of the
universe but fluctuates of the order of the Hubble parameter during the inflationary expansion,
or the scalar field written in terms of the slow action variable which suffers from the hyperbolic
instability due to the resonance of the masses of the scalar fields during the oscillatory stage [8].
As seen from the exactly homogeneous evolution equations, it can be seen that r changes as
changes. Therefore the fluctuation of such modulating scalar quantities is imprinted on the r
fluctuation. Please see the formula (6.14).
49
In order to generate the evolution of cosmological perturbations in the long wavelength limit
from the corresponding exactly homogeneous perturbations by the LWL formula, under the various source term S we need to solve the exactly homogeneous system precisely. These analyses
will be in future publication.
We consider the system where multiple oscillatory scalar fields and multiple radiation fluids
interact. As for the system without radiation, its evolution of cosmological perturbations have
been investigated in detail [7,8]. The system of multiple scalar fields only can be written in terms
of Hamiltonian form. Any Hamiltonian system obeys the Liouville theorem, that is, volume occupied by group of orbits are invariant. Therefore according to the LWL formula, in the stable
case the perturbations do not grow and in the unstable case the same numbers of growing modes
and decaying modes appear because of the squeezing of the phase space volume. The former case
occurs in the case where masses of scalar fields are incommensurable and near the elliptic fixed
points in the case where masses of scalar fields are commensurable. The latter case occurs near
the hyperbolic fixed points in the case where masses of scalar fields are commensurable. Then we
include dissipative interaction with radiation. In this case our system is not a Hamiltonian system
and it does not obey the Liouville theorem. In this dissipative system, in addition to two possibilities mentioned above we can expect the third possibility where group of orbits are attracted into
the attracting set. The final state where all the energy of the scalar fields is transferred into that of
radiation fluids is the attracting equilibrium. Around the attracting set, the adjacent orbits come
nearer and nearer, therefore the LWL formula tells us that all the perturbation modes are stable,
that is, converge into some constants or decay. It is useful to investigate how the behavior of
cosmological perturbations around the hyperbolic fixed points is changed due to the dissipative
interaction with radiation under the spirit of the LWL formula. In this line of researches, we will
have new explanation of the backreaction which suppresses the instability due to the resonance.
In the future publication, we will return to this problem.
Acknowledgements
The author would like to thank Professors H. Kodama, S. Mukohyama, T. Nakamura,
M. Sasaki, K. Sato, N. Sugiyama, T. Tanaka, A. Taruya, J. Yokoyama for continuous encouragements. He would like to thank Professor V.I. Arnold for writing his excellent textbook and/or
review [32], from which he learned a lot about the dynamical system.
Appendix A. Proofs of the propositions in Section 5
A.1. Technical lemmas
Technical Lemma 1. Under the assumption (5.15), for |f | 1,

f
f 1
a,
,
I
a a
and for |f | |I |,

f
1,
I

f |I |

.
a
a
(A.1)
(A.2)
50
Technical Lemma 2. For the general physical quantity A(I, , ) with Fourier decomposition
as

Ak (I, ) exp(ik ),
A = A0 (I, ) +
(A.3)
k=0
the solution to the first order partial differential equation

a
S = A A
,
a
(A.4)
is given by
S = {A},
(A.5)
where
{A} :=

k=0
Ak
exp(ik ).
i(k )
(A.6)
A.2. Proof of Proposition 1

In order to make the notation simple, we omit the superscript (k) and replace (k + 1) with (1).
By substituting the transformation laws of I , , to the evolution equations of Ia , , a , we
obtain

ua
Fa
+ Fa (I, , , a) Fa I (1) , (1) , (1) , a a
a
ua (1)
ua (1)
ua (1)
(1)
F + (1) Gb ,
= Fa + (1) Fb +
(A.7)
(1)
Ib
b

u
F
+ F (I, , , a) F I (1) , (1) , (1) , a a
a
u (1)
u (1)
u (1)
(1)
= F + (1) Fb +
(A.8)
F + (1) Gb ,
(1)
Ib
b
and
1 a
1
1 a
1
ub
u
a (I, , a) a I (1) , (1) , a

I (1)
(1)
b

va
+ Ga
+ Ga (I, , , a) Ga I (1) , (1) , (1) , a a
a
v
v
v
a
a
a
(1)
(1)
(1)
F +
F + (1) Gb ,
= G(1)
a +
(1) b
(1)
Ib
(A.9)
when we choose ua , u , va as
ua = {Fa } k+1 |I |,
(A.10)
u = {F } a|I |,

a
a
u +
u + Ga k+1 .
va =
(1) b
(1)
Ib
(A.11)
k+1
(A.12)
51
As for Fa , F , Ga defined by

Fa := Fa(1) Fa
,
(A.13)
F := F(1) F
,
Ga := G(1)
a Ga
,
(A.14)
(A.15)
applying the mean value theorem to (A.7), (A.8), (A.9) gives

Fa + k+1 Fb + k+1 |I |F + k+1 |I |Gb = k+1 a 2 |I |,
(A.16)
F +
(A.17)
k+1
Ga +
aFb +
k+1
k+1
aFb +
a|I |F +
k+1
F +
k+1
k+1
a|I |Gb =
Gb =
a |I |,
k+1 3
k+1 2
a ,
(A.18)
where all coefficients of order unity are omitted. By solving the above three equations, we obtain
Fa = k+1 a 2 |I |,
(A.19)
F =
(A.20)
a |I |,
k+1 3
Ga =
k+1 2
(A.21)
a ,
Fa , F , Ga are decomposed into the angle variables independent parts Fa
, F
,
(1)
Ga
and the angle variables dependent parts Fa(1) , F(1) , G
a .
A.3. Lemmas and the preparatory propositions
Lemma 1. The solution to the differential equation

d
A = a 2 A + E a 2 B(a),
da
where is a positive constant, is bounded as

A(a) exp a 2 A(1) + E a 2 B(a).
2
a
Lemma 2. When B satisfies

d

B E a 2 |B| + E a 2 C + B(a) ,
da
where C is a positive constant, for an arbitrary a 1

a1
B(a)
C + B(a1 ) ,
a1 1
(A.22)
(A.23)
(A.24)
(A.25)
where a1 is a constant satisfying a1 > 1. For example, by putting a1 = 2 we obtain

B(a) C + B(2).
Proof. By solving the differential equation, we obtain

B(a) B(1) + daE a 2 C + B(a) .
1
(A.26)
(A.27)
52
Since for a satisfying 1 a a1 ,

B(a) C + B(a1 ),
(A.28)
and for a a1

B(a) C + B(a1 ) + E a 2 B(a),
1
(A.29)
then for an arbitrary a 1

B(a) C + B(a1 ) + E a 2 B(a),
1
(A.30)
whose right-hand side is an increasing function of a. Since E(a12 ) 1/a1 , we obtain

(A.25). 2
Lemma 3. When I , satisfy

d
I a|I | + | | + |A|,
da

d
a 2 |I | + a| | + |B|,
da
the following inequality hold:

(2) I (1) + (1) + A(2) + B(2).
(A.31)
(A.32)
(A.33)
Proof. We consider the differential equation as for a|I | + | |:

d
a|I | + | | a a|I | + | | + a|A| + |B|.
da
(A.34)
Then we obtain

1 2
I (1) + (1) + a 2 A(a) + aB(a) .
(a) exp a
2
We put a = 2.
(A.35)
Proposition Ap1. For the mth order system, for the background quantities, the following inequalities hold:

| | 1.
|I | exp a 2 ,
(A.36)
Proof. We solve the evolution equations given by

d
a I a 2 |I | + a 2 m |I |,
da

d
a a 3 |I | + a 3 m |I |.
da
2
(A.37)
(A.38)
53
Proposition Ap2. Let m be an integer larger than or equal to 2. For the mth order system, for
the perturbation quantities, the following inequalities hold:

I (a) E a 2 Am (1),
(A.39)
(a) Am (1),

2

(a) exp a 2 m (1) + a Am (1) ,

(A.40)
(A.41)
where Am (1) is defined by (5.51).

Proof. The perturbation variables satisfy the evolution equations as
d
(A.42)
I = aI + a|I | + a m |I |,
da
d
(A.43)
= a 2 I + a 2 |I | + a 2 m |I |,
da
a2
a
d
= I + + a m ,
(A.44)
da

where |I | means a function bounded by M|I | for a positive constant M. It is important to notice the coefficient of I in (A.42) is negative. Although in (A.42), terms such as a 2 |I |I are
contained, such terms can be neglected because
(A.45)
da a 2 |I | 1.
1
By substituting the estimation obtained from (A.44)

a3

a2
| | exp a 2 m (1) + I (a) + (a) ,

(A.46)
into (A.42), by applying Lemma 1, we obtain

I (a) E a 2 I (1) + m (1) + (a) .
(A.47)
By using the above inequalities in (A.43) and applying Lemma 2, we obtain

(a) I (1) + m (1) + (2).
(A.48)
On the other hand, by applying Lemma 3, we obtain

(2) Am (1).
Then we can prove the results.
(A.49)
2
A.4. Proof of Proposition 2A

The evolution equations of A where A = (I, , ) are given by
d
I = aI + a|I | + a m |I |,
da
(A.50)
54
d
= a 2 I + a 2 |I | + a 2 m |I |,
da
a2
a
d
= I + + a m .
da

In the same way as the proof of Proposition Ap2, we obtain

|I | E a 2 I (1) + (1) + m ,

(a) I (1) + (1) + m ,

|| (1) +
(A.51)
(A.52)
(A.53)
(A.54)

I (1) + (1) + m .
a 2
(A.55)

In the above estimations, we put A(1) = 0.
Next we consider the evolutions of A where A = (I, , ). We take differences between
the upper equations and the lower equations. As for I

d
I = aI + a|I | + a m I + |I | + |I | ,
da
d
Itr = aItr + a|Itr |tr ,
da
as for

d
= a 2 I + a 2 |I | + a 2 m I + |I | + |I | ,
da
d
tr = a 2 Itr + a 2 |Itr |tr ,
da
and as for
(A.56)
(A.57)
(A.58)
(A.59)
d
a2
a
(A.60)
= I + + a m (aI + + ),
da

a2
a
d
tr = Itr + tr .
(A.61)
da

By taking into account the fact that in the first two terms on the right-hand sides the coefficients
depending on I , , not on are multiplied, and using the estimations of A, A where A =
(I, ), we obtain
I Itr = I + I (aI + )

= I + E a 2 Am (1) m ,

I Itr tr = Itr + I + |Itr |

= E a 2 + E a 2 Am (1) m ,
(A.62)
(A.63)
tr = + (aI + )
= + Am (1) m .
(A.64)
By using estimations of A where A = (I, , ), we obtain

1
I + |I | + |I | = E a 2 A1 (1),

a2
2 m

(1) + Am (1) .
aI + + = exp a

(A.65)
(A.66)
55
Therefore we get

d
I = aI + E a 2 + m1 E a 2 A1 (1),
da

d
= a 2 I + E a 2 + m1 E a 2 A1 (1),
da

d
a2
a
= I + + a m1 exp a 2 m (1) + a 2 Am (1) .
da


|I | E a 2 I (1) + (1) + m1 A1 (1) ,

(a) I (1) + (1) + m1 A1 (1),

a 2

I (1) + (1) + m1 A1 (1)
| | (1) +

+ m1 exp a 2 m a 2 (1) + a 4 Am (1) .
(A.67)
(A.68)
(A.69)
(A.70)
(A.71)
(A.72)
In the above estimations, we put A(1) = 0 where A = (I, , ). Then we complete the proof.
A.5. Proof of Proposition 2B
From Proposition 1, we obtain

I (0) = I (m) + I (m) ,

(0) = (m) + a I (m) ,
(0)
(m)
+ ,
(A.73)
(A.74)
(A.75)
where |I (m) | means the function of A(m) where A = (I, , ) bounded by M|I (m) | for a positive
constant M. As for A where A = (I, , ), we obtain
(0)

I I (m) I (m) + I (m) (m) + I (m) (m)

m E a 2 ,
(A.76)
(0)

(m) a I (m) + I (m) (m) + I (m) (m)

m E a 2 ,
(A.77)
(0)

(m) (m) (m)
(m)
a I + +
ma2.
(A.78)
By using the estimations of Proposition 2A and the above evaluations, we obtain the results of
the former part.
Next we consider A where A = (I, , ). By taking the variations of the transformation
laws, we obtain

I (0) I (m) = I (m) + I (m) (m) + I (m) (m) ,
(A.79)

(0) (m) = a I (m) + I (m) (m) + I (m) (m) ,
(A.80)

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

(A.81)
,
= aI
+
+
56
where the coefficients are the functions of A(m) where A = (I, , ). We take the differences of
the transformation laws of the exact variables A and those of the truncated variables Atr . By
using

(m)
I (m) Itr = I (m) + I (m) aI (m) + (m) + (m) ,
(A.82)

(m)
I (m) Itr tr = Itr (m) + (m) I (m) + Itr (m) + Itr (m) ,
(A.83)

(m)
(m)
(m)
(m)
(m)
(m)
(m)
aI
,
tr =
+
+
+
(A.84)

(m) tr(m) = (m) + (m) aI (m) + (m) + (m) ,
(A.85)
(m)

(m)
(m)
(m)
(m)
(m)
(m)
I
,
Itr tr = Itr
+
+ Itr
+ Itr
I
(A.86)
and by using the estimations of A(m) and A(m) where A = (I, , ), we obtain

I (0) I (m) = m1 E a 2 A1 (1),

(0) (m) = m1 E a 2 A1 (1),

(0) (m) = exp a 2 m m1 a 2 (1) + a 4 Am (1) .
(A.87)
(A.88)
(A.89)
By using the estimations of Proposition 2A and the above evaluations, we obtain the results of
the latter part. We complete the proof.
A.6. Proof of Proposition 3
As for A where A = (I, , ), we can obtain the results by putting m = 1 in the proof of
Proposition 2A.
Next we consider the evolutions of A where A = (I, , ). We take differences between
the upper equations and the lower equations. As for I

d
Itr = aItr + a|Itr |tr + a Itr + |Itr |tr ,
da
d
In = aIn + a|In |n ,
da
(A.90)
(A.91)
as for

d
tr = a 2 Itr + a 2 |Itr |tr + a 2 Itr + |Itr |tr ,
da
d
n = a 2 In + a 2 |In |n ,
da
(A.92)
(A.93)
and as for
d
a2
a
tr = Itr + tr + a(aItr + tr ),
da

d
a2
a
n = In + n .
da

(A.94)
(A.95)
In the above equations, the subscript tr implies that in the present system, the angle variables
dependent parts which have been made sufficiently small by the transformations defined in the
57
proof of Proposition 1 have already been truncated, and the subscript n means the further neglection of -order corrections produced by such transformations. We obtain the evolution equations:

d
I = aI + E a 2 + B(1) ,
da

d
= a 2 I + E a 2 + B(1) ,
da
d
a2
a
= I + + aB(1).
da


|I | E a 2 I (1) + (1) + B(1) ,

(a) I (1) + (1) + B(1),

| | (1) +
(A.96)
(A.97)
(A.98)
(A.99)
(A.100)

I (1) + (1) + B(1) .
a 2
(A.101)

By putting A(1) = 0 where A = (I, , ) in the above inequalities, we obtain the results of
the latter part. We complete the proof.
Appendix B. Evaluation of the gamma-like function
In the present paper, we often have to evaluate the integrals defined by
G(t, ) :=
dx x t1 e x ,
(B.1)
x0
where x0 is defined by
2
,
(B.2)
3 A1/2
where is assumed to be sufficiently small. By expanding with respect to the small parameter x0
by the partial integration, we obtain the evaluations as follows. For t > 0,
x0 :=
G(t, ) =
1
G(t) = O(1),
t
(B.3)
for t = 0,
G(0, ) = ln x0 + O(1),
(B.4)
for t < 0 and t is an integer,

1
G(t, ) = x0t + O x0t+1 , ln x0 ,
t
and for t < 1 and t is not an integer,

1
G(t, ) = x0t + O x0t+1 , 1 .
t
G(t) is the well-known Gamma function.
(B.5)
(B.6)
58
Next by using the above evaluations, we evaluate the integrals defined by (6.20) which appear when we evaluate the effects of the interactions between scalar fields on the final radiation
energy density = /a 4 . By expanding with respect to the small parameter x0 by the partial
integration, we obtain the evaluations as follows. We assume that n1 > 0. For n2 > 0,
G(n1 , n2 , 1 , 2 ) = O(1),
(B.7)
for n2 = 0,
G(n1 , 0, 1 , 2 ) =
1
ln x0 G(n1 ) + O(1),
1n1
(B.8)
and for n2 < 0,

G(n1 , n2 , 1 , 2 ) =

n1 +n2 n2 +1
1 1 n2
, x0 , ln x0 .
n1 x0 G(n1 ) + O x0
n2 1
(B.9)
Finally we evaluate the incomplete Gamma function defined by
G(t; x1 ) :=
dx x t1 ex ,
(B.10)
x1
for large x1 . By partial integration, we obtain

G(t; x1 ) = x1t1 ex1 + O x1t2 ex1 ,
for sufficiently large x1 .
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
D. Polarski, A.A. Starobinsky, Nucl. Phys. B 385 (1992) 623.

C. Gordon, D. Wands, B.A. Bassett, R. Maartens, Phys. Rev. D 63 (2001) 123506.
A. Taruya, Y. Nambu, Phys. Lett. B 428 (1998) 3743.
H. Kodama, T. Hamazaki, Phys. Rev. D 57 (1998) 71777185.
M. Sasaki, T. Tanaka, Prog. Theor. Phys. 99 (1998) 763782.
H. Kodama, T. Hamazaki, Prog. Theor. Phys. 96 (1996) 949970.
T. Hamazaki, Phys. Rev. D 66 (2002) 023529.
T. Hamazaki, Nucl. Phys. B 698 (2004) 335385.
T. Hamazaki, H. Kodama, Prog. Theor. Phys. 96 (1996) 11231146.
K.A. Malik, D. Wands, C. Ungarelli, Phys. Rev. D 67 (2003) 063516.
B.A. Bassett, F. Viniegra, Phys. Rev. D 62 (2000) 043507.
F. Finelli, R. Brandenberger, Phys. Rev. D 62 (2000) 083502.
J.P. Zibin, R. Brandenberger, D. Scott, Phys. Rev. D 63 (2001) 043511.
J. Yoshida, S. Tsujikawa, Class. Quantum Grav. 23 (2006) 353.
J. Traschen, R.H. Brandenberger, Phys. Rev. D 42 (1990) 2491.
K.A. Kofman, A.D. Linde, A.A. Starobinsky, Phys. Rev. Lett. 73 (1994) 3195.
Y. Shtanov, J. Traschen, R.H. Brandenberger, Phys. Rev. D 51 (1995) 5438.
K.A. Kofman, A.D. Linde, A.A. Starobinsky, Phys. Rev. D 56 (1997) 3258.
A. Hosoya, M. Sakagami, Phys. Rev. D 29 (1984) 2228.
M. Morikawa, Phys. Rev. D 33 (1986) 3607.
J. Yokoyama, Phys. Rev. D 70 (2004) 103511.
H. Kodama, M. Sasaki, Prog. Theor. Phys. Suppl. 78 (1984) 1166.
J.M. Bardeen, Phys. Rev. D 22 (1980) 1882.
V.F. Mukhanov, H.A. Feldman, R.H. Brandenberger, Phys. Rep. 215 (1992) 203.
M. Sasaki, Prog. Theor. Phys. 76 (1986) 1036.
(B.11)
[26]
[27]
[28]
[29]
[30]
[31]
[32]
V.F. Mukhanov, Sov. Phys. JETP 67 (1988) 12971302.

D. Wands, K.A. Malik, D.H. Lyth, A.R. Liddle, Phys. Rev. D 62 (2000) 043527.
H. Kodama, M. Sasaki, Int. J. Mod. Phys. A 2 (1987) 491.
Y. Sasaki, J. Yokoyama, Phys. Rev. D 44 (1991) 970.
G. Dvali, A. Gruzinov, M. Zaldarriga, Phys. Rev. D 69 (2004) 023505.
S. Gupta, K.A. Malik, D. Wands, Phys. Rev. D 69 (2004) 063513.
V.I. Arnold, Mathematical Methods of Classical Mechanics, Springer, New York, 1978;
V.I. Arnold, A. Avez, Problmes ergodiques de la mcanique classique, GauthierVillars, Paris, 1967.
59
Systematic parameter space search of extended

quarklepton complementarity
Florian Plentinger, Gerhart Seidl , Walter Winter
Institut fr Theoretische Physik und Astrophysik, Universitt Wrzburg, D-97074 Wrzburg, Germany
Received 4 January 2007; received in revised form 3 August 2007; accepted 3 September 2007
Abstract
We systematically investigate the parameter space of neutrino and charged lepton mass matrices for
textures motivated by an extended quarklepton complementarity. As the basic hypothesis, we postulate
that all mixing angles in U and U be either maximal or described by powers of a single small quantity
C . All mass hierarchies are described by this as well. In this study, we do not assume specific forms
for U and U , such as large mixing coming from the neutrino sector only. We perform a systematic scan
of the 262 144 generated mixing matrices for being compatible with current experimental data, and find a
sample of 2468 possibilities. We then analyze and classify the effective charged lepton and neutrino mass
textures, where we especially focus on a subset of models getting under pressure for small 13 . In addition,
we predict the mixing angle distributions from our sample of all valid textures, and study the robustness
of this prediction. We also demonstrate how our procedure can be extended to predictions of the Dirac and
Majorana phases in UPMNS . For instance, we find that CP conservation in neutrino oscillations is preferred,
and we can impose a lower bound on the mixing matrix element for 0 decay.
PACS: 12.15.Ff; 14.60.Pq; 14.60.St
Keywords: Charged lepton and neutrino masses and mixing; Cabibbo angle; Textures; Quarklepton complementarity
E-mail addresses: florian.plentinger@physik.uni-wuerzburg.de (F. Plentinger), seidl@physik.uni-wuerzburg.de

(G. Seidl), winter@physik.uni-wuerzburg.de (W. Winter).
doi:10.1016/j.nuclphysb.2007.09.016
F. Plentinger et al. / Nuclear Physics B 791 (2008) 6092
61
1. Introduction
The Standard Model (SM) of elementary particle physics with three light Majorana neutrinos
contains 28 free parameters. Most of them, in total 22, describe the masses and mixings of the
fermions (the remaining six parameters are the three SM gauge couplings, the vacuum expectation value and quartic coupling of the Higgs, and the QCD parameter). This large number of
parameters, especially in the fermion sector, is generally considered as an unsatisfactory feature
of the SM and one therefore seeks for models in which the number of parameters can be minimized. One possibility to reduce the number of parameters is to embed the SM into a Grand
Unified Theory (GUT).
By putting quarks and leptons into GUT multiplets, the masses and mixing angles in the quark
and lepton sectors become related. From this point of view, it is thus reasonable to describe the
observed hierarchical pattern of the masses and mixing angles of quarks and charged leptons [1]
in terms of powers of a single small expansion parameter . The expansion parameter might,
for example, represent a low-energy remnant of a flavor symmetry that has been broken at some
high scale. In fact, the CKM mixing matrix VCKM [2,3] exhibits quark mixing angles of the
orders
|Vus | ,
|Vcb | 2 ,
|Vub | 3 ,
(1)
where the quantity is of the order of the Cabibbo angle C 0.2. Similarly, for the same value
C , the mass ratios of the up-type quarks, down-type quarks, and the charged leptons can be
approximated, e.g., by1
mu : mc : mt = 6 : 4 : 1,
md : ms : mb = 4 : 2 : 1,
me : m : m = 4 : 2 : 1,
(2)
where mb /mt 2 , m /mb 1, and mt 175 GeV. While the CKM angles and charged
fermion masses are thus strongly hierarchical, there are striking differences in the neutrino
sector. In the past few years, solar [4,5], atmospheric [6], reactor [7,8], and accelerator [9]
neutrino oscillation experiments have established with increasing precision that among the
leptonic mixing angles only the reactor angle 13 is small whereas the solar angle 12 and
the atmospheric angle 23 are both large. Moreover, neutrino oscillation data tells us that
the neutrinos have only a mild hierarchy. To be specific, expressing the neutrino mass ratios as in Eq. (2) in terms of powers of , the neutrino mass spectrum can, e.g., be written
as
m1 : m2 : m3 = 2 : : 1,
m1 : m2 : m3 = 1 : 1 : ,
m1 : m2 : m3 = 1 : 1 : 1,
(3)
where m1 , m2 , and m3 denote the 1st, 2nd, and 3rd neutrino mass eigenvalue. In Eq. (3), the first
equation corresponds to a normal hierarchical, the second to an inverted hierarchical, and the
third to a degenerate neutrino mass spectrum.2 In addition, we know from cosmological observations that the absolute neutrino mass scale is of the order 102 . . . 101 eV [10]. An attractive
origin of the smallness of neutrino masses is provided by the seesaw mechanism [11,12]. For
a summary of current values and errors for the neutrino oscillation parameters from a global
analysis, see Table 1.
1 We take here a fit compatible with an SU(5) GUT.
2 More precisely, we have in Eq. (3) for the inverse hierarchical case m > m and (m m )/m 2 .
2
1
2
1
2
62
Table 1
Current best-fit values with 1 errors, relative accuracies at 1 , as well as 2 and 3 allowed ranges of three-flavor neutrino oscillation parameters (from a combined analysis of all available data [13]). In addition, we show as the last column
the projected improvement of the mixing angle errors on a time scale of about ten years from now. For this projection, we
assume that the current best-fit values remain unchanged, in particular, that sin2 213 will not be discovered. We use the
experimental bounds sin2 213 0.03 (3 , year 2016 from Ref. [14] based on NOA), 10% precision of sin2 23 (1 ,
from Ref. [15], based on beam experiments), 4.6% precision of sin2 12 (1 , from Ref. [16] based on SPMIN/SADO +
others with an exposure of 10 GW kt yr, which seems to be reasonable on that time scale for a KamLAND-scale detector).
These values should be interpreted with care, because they depend on the experimental strategy
Parameter
Best-fit 1
1 acc.
2 range
3 range
Projection 2016
m221 [105 eV2 ]
7.9 0.3
4%
7.38.5
7.18.9
| m231 | [103 eV2 ]
+0.20
2.50.25
10%
2.13.0
1.93.2
sin2 12
+0.02
0.300.03
+0.08
0.500.07
9%
0.260.36
0.240.40
4.6% (1 )
16%
0.380.64
0.340.68
10% (1 )
0.025
0.041
0.0076 (3 )
sin2
23
sin2 13
In spite of the qualitative differences between the quark and the lepton sector, there have
recently been proposed interesting quarklepton complementarity (QLC) relations [1719]
(for an early approach see Ref. [20]) that might point to quarklepton unification. The QLC
relations express that the solar angle 12 and the atmospheric angle 23 seem to be connected to
the quark mixing angles by
12 + C /4,
23 + cb /4,
(4)
where cb = arcsin Vcb . A simple interpretation of how the QLC relations in Eq. (4) might arise
is to assume that the small and the maximal angles in this relation, C , cb , and /4, describe
the mixing of the left-handed charged leptons and the neutrinos in flavor basis. Consequently,
the observed large leptonic mixing angles 12 and 23 can only arise as a result of taking the
product of the charged lepton mixing matrix U and the neutrino mixing matrix U in the PMNS
[21,22] mixing matrix UPMNS = U U . This point of view is supported by the fact that in explicit
models, C and cb are usually given from the outset and also maximal leptonic mixing has meanwhile been obtained in many models (for a review on recent developments see, e.g., Ref. [23]).
Therefore, if the QLC relations really arise from taking the product of U and U in UPMNS , it
is interesting to study models with mass matrices where, e.g., the value of 12 can be understood
in terms of C and 23 as a result of combining U and U into UPMNS . There has been a considerable amount of work on QLC. For example, deviations from bimaximal neutrino mixing [24]
that are introduced by C have been studied in Refs. [2527]. Sum rules in the context of QLC
using some assumptions on the mixing angles were presented in Refs. [28,29]. Parameterizations
of UPMNS in terms of C as an expansion parameter were given in Refs. [3034]. For renormalization group effects and QLC see Ref. [35] and references therein. Model building realizations
of QLC were discussed in Refs. [36,37] and for the special case C 13 in Ref. [38].
In this paper, we study systematically 262 144 pairs of charged lepton and neutrino mass matrices that lead to the observed leptonic masses and mixing angles. Motivated by QLC, our only
assumption on the mixing angles in U and U is that they can take any of the values /4, , 2 , 0,
where is of the order of the Cabibbo angle C and represents the only small expansion parameter in our approach. Different from earlier studies of QLC, which assume very specific forms
of U and U [2529], we allow maximal ( /4) and small ( , 2 , . . .) mixing angles to be
63
generated in the charged lepton as well as in the neutrino sector. This means, in particular, that we
do not assume that U or U are necessarily of a bimaximal mixing form but obtain this, instead,
as a special case. We consider all possible combinations of the mixing angles /4, , 2 , . . . in
U and U , and select those pairs U and U that give PMNS mixing angles 12 , 13 , and 23
in agreement with observations. For the detailed description of our method, see Section 2. We
then unambiguously reconstruct directly the charged lepton and the neutrino mass matrices from
the known charged lepton and neutrino mass spectra in Eqs. (2) and (3). The matrix pairs that
we obtain in this way are then presented in the flavor basis with order of magnitude entries that
are expressed as powers of C , which takes the role of a small expansion parameter for the
matrices. Such approximate forms or representations of mass matrices expanded in terms of
will be called in the following mass matrix textures or simply textures. They are presented
and discussed in Section 3. We analyze the predictions for the mixing angles from the set of
mixing matrices consistent with current observations in Section 4, and we study the robustness
of these predictions. Finally, we show how our framework can be applied to phase predictions
in Section 5, such as for CP violation in neutrino oscillations or the mixing matrix element in
neutrinoless double beta decay.
2. Method
In this section, we will introduce our method for a systematic scan of the textures that exhibit
QLC. First, we set up the notation for the mass and mixing matrices. Next, we motivate and
present our hypothesis for the charged lepton and neutrino mixing angles. We show how the
resulting PMNS matrices are compared with observation using a selector and discuss how to
construct from the selected mixing matrices the textures for the charged leptons and neutrinos.
2.1. Mixing formalism and notation
Let us now describe the formalism and notation for the leptonic mixing parameters. Here, we
follow closely Ref. [27]. A general unitary 3 3 matrix Uunitary can always be written as

Uunitary = diag ei1 , ei2 , ei3 U diag ei1 , ei2 , 1 ,
(5a)
where the phases 1 , 2 , 3 , 1 , and 2 , take their values in the interval [0, 2] and
c12 c13
s12 c13
s13 ei
U = s12 c23 c12 s23 s13 ei c12 c23 s12 s23 s13 ei
s23 c13
s12 s23 c12 c23 s13
e i
c12 s23 s12 c23 s13
e i
(5b)
c23 c13
is a CKM-like matrix in the standard parameterization with sij = sin ij , cij = cos ij , where
ij {12 , 13 , 23 } lie all in the first quadrant, i.e., ij [0, 2 ], and [0, 2]. The matrix U
is thus described by 3 mixing angles ij and one phase , i.e., it has 4 parameters. The matrix
Uunitary has five additional phases and contains therefore in total 9 parameters.
Depending on whether neutrinos are Majorana or Dirac particles, the low-energy effective
Lagrangian for lepton masses takes one of the forms
1 Maj
M ij i j + h.c.,
2

LD = (M )ij ei ejc MDirac ij i jc + h.c.,
LM = (M )ij ei ejc
(6)
64
where ei and i are the left-handed charged leptons and neutrinos that are part of the SU(2)L lepton doublets i = (i , ei )T , while eic are the right-handed (SU(2)L singlet) charged leptons, and
i = 1, 2, 3 is the generation index. We have also extended the SM by adding to each generation i
Maj
one right-handed SM singlet neutrino ic . In Eq. (6), M , M , and MDirac , denote the charged
Maj
lepton (M ), the Majorana neutrino (M ), and the Dirac neutrino mass matrix (MDirac ), respecMaj
tively. The Dirac matrices M and MDirac are general complex 3 3 matrices, whereas M is a
complex symmetric 3 3 matrix. The charged lepton mass matrix is diagonalized by a biunitary
transformation
diag
M = U M
U ,
(7a)
where U and U are unitary matrices acting on the left-handed (U ) and right-handed (U )
charged leptons ei and eic , which span the rows and columns of M , respectively. In Eq. (7a),
diag
diag
the matrix M is on the diagonal form M = diag(me , m , m ). Using the freedom of rephasing the charged lepton fields, we can, in what follows, assume that U is on a CKM-like form
that is parameterized as in Eq. (5b). We then define the 4 mixing parameters of the left-handed
, , , and the phase , by identifying in Eq. (5b)
charged leptons, the three mixing angles 12
13 23

ij ij and . The Majorana and Dirac neutrino mass matrices are given by
Maj
diag
= U M
UT ,
diag
MDirac = U M
U ,
(7b)
where the unitary neutrino mixing matrix U acts on the left-handed neutrinos i , while
diag
diag
is a diagonal matrix M =
U acts on the right-handed neutrinos ic . In Eq. (7b), M
diag(m1 , m2 , m3 ). Since U has already been brought to a CKM-like form, we have no longer
the same freedom to remove phases in U and, thus, we find that the PMNS matrix is in general
written as
UPMNS = U U = U D U K,
(8a)
where U is a CKM-like matrix that is on the form as in Eq. (5b) while D = diag(1, ei1 , ei2 ) and
K = diag(ei1 , ei2 , 1) are diagonal matrices with phases in the range 1 , 2 , 1 , 2 [0, 2].
Note that we have already removed in UPMNS an unphysical overall phase. The CKM-like matrix U in Eq. (8a) contains four neutrino mixing parameters, the three neutrino mixing angles
, , , and a neutrino phase , which we define in the standard parameterization by iden12
13 23
tifying in Eq. (5b) the neutrino mixing angles as ij ij and the neutrino phase as . The
matrix in Eq. (8a) is written in terms of the six phases , , 1 , 2 , 1 , and 2 , which lead to
three physical phases. The PMNS matrix in Eq. (8a) can, equivalently, also be directly written as

UPMNS = U U = U diag ei1 , ei2 , 1 ,
(8b)
where U is a CKM-like matrix that is on the form as in Eq. (5b) and the phases 1 and 2 are
Majorana phases. The CKM-like matrix U in Eq. (8b) is described by the solar angle 12 , the
reactor angle 13 , the atmospheric angle 23 , and one Dirac CP-phase , which we identify in
the standard parameterization of Eq. (5b) as ij ij and . The PMNS matrix has thus
3 mixing angles and 3 phases and contains therefore 6 physical parameters. Writing the matrix
elements of UPMNS as Uij (UPMNS )ij , we read off in the standard parameterization of Eq. (5b),
the 6 leptonic mixing parameters as follows
sin 13 = |U13 |,
(9a)
65
U12
, 12 = if U11 = 0,
tan 12 =
(9b)
U
11
U23
, 23 = if U33 = 0,
tan 23 =
(9c)
U33
2

U U U U
arg( 12 13 22 2 23 + s12c12s13c23s23 ), for ij
= 0, 2 ,
s
=
(9d)
12 s13 s23 c12 c13 c23
0
else,
arg(ei U13 U11 ), for ij

= 2 and 13
= 0,
),
1 = arg(U31 U33
(9e)
for ij
= 2 and only 13 = 0 allowed,
0
else,
i
arg(e U13 U12 ), for ij
= 2 and 12 , 13
= 0,
i1
2 = arg(e U11 U12 ), for 13 , 23 = 0 and 12
= 0, 2 ,
(9f)
for 12 or 13 = 0 but
= 2 , and 23
= 0, 2 ,
arg(U22 U23 ),
0
else.
Note that we set the phases to 0 if they are undefined. Since we can easily identify these cases
if needed, there is no bias or constraint introduced by this choice. For example, for 13 = 0, is
undefined. In this case, does not affect the mixing matrix, i.e., an arbitrary choice of 0 does
not change the physics. However, for the phase prediction of , the irrelevant cases can be easily
eliminated by the identification of the corresponding 13 = 0. Note that the above relations are
valid for a general unitary matrix Uunitary as in Eq. (5a) as well. In addition, if UPMNS is already a
).
CKM-like matrix, i.e., all unphysical phases are already removed, we simply have = arg(U13
2.2. Generating the PMNS matrices and textures
Let us now introduce our method for a systematic scan of the textures that exhibit QLC.
Instead of starting out with various forms for the lepton mass matrix textures and calculate from
these the lepton masses and mixing angles, we follow in our procedure a reverse process: we
reconstruct the mass matrix textures in flavor basis from the masses and mixing parameters of
charged leptons and neutrinos.
Although neutrino oscillation experiments will in future allow to pin down the leptonic mixing
angles with increasing precision, the PMNS matrix does not uniquely reveal the individual mixing parameters in U and U . Existing models and studies mostly suppose that the observed large
solar and atmospheric mixing angles arise mainly in the neutrino sector. It is, however, important to emphasize that maximal atmospheric mixing can also equally well arise from the charged
lepton sector as proposed, e.g., in lopsided GUT models [39] (for realizations of bilarge neutrino mixing in lopsided models see also, e.g., Refs. [4042]). Moreover, the QLC relations in
Eq. (4) suggest, in particular, that both maximal ( /4) and small ( C , cb ) mixing angles
might be expected in the charged lepton and in the neutrino sector. Motivated by QLC, we will
thus assume in our approach that the mixing angles ij in U and ij in U can a priori take any
of the values in the sequence /4, , 2 , . . . , where C 0.2, and then compare the resulting
PMNS mixing angles ij with current data. The choice of the angles in this sequence is a simple
and straightforward interpretation of the QLC relations in Eq. (4) in the sense that the solar angle 12 33 can only arise as a result of taking the product of U and U in UPMNS = U U .
The assumption that any of the angles ij and ij can assume any of the values /4, , 2 , . . . ,
generalizes the definition of QLC to what we call an extended QLC.
66
Let us next specify the values of the mass parameters in our approach. Motivated by the mass
ratios of quarks and leptons in Eqs. (2) and (3), we will approximate the diagonalized charged
diag
lepton mass matrix M in Eq. (7a) by

diag
M = m diag 4 , 2 , 1 ,
(10a)
diag
and choose for the diagonalized neutrino mass matrix M in Eq. (7b) any of the cases

diag
diag
diag
M = m3 diag 2 , , 1 ,
M = m2 diag(1, 1, ),
M = m3 diag(1, 1, 1),
(10b)
where the first, second, and third equations correspond to a normal, inverted, and degenerate
neutrino mass spectrum, respectively. We thus see that the quantity C will appear in our
approach as a small expansion parameter, which is in charge of all the hierarchies and small
effects that are exhibited by the mass spectra and mixing angles of the fermions.
Our systematic search for textures satisfying QLC hence consists in the following simple
three-step procedure:
First step: In the first step of our procedure, we consider the PMNS matrix in the parameterization UPMNS = U U = U D U K of Eq. (8a) and assume here that the entries sin(ij ) and
sin( ) in the CKM-like matrices U and U can take all the values
ij
sij , sij

,
1 , , 2 , 0
2
(11)
where we have defined sij = sin(ij ) and sij = sin(ij ). For definiteness, we choose in Eq. (11)
= 0.2 as our standard value. Moreover, we will first restrict ourselves to the case of real mixing
matrices U and U , and assume in Eq. (8a) that the phases , , 1 , 2 , 1 , and 2 , are only 0
or . In total, this gives
43 2 43 25 = 262 144
(12)
possible distinct pairs of mixing matrices {U , U } (cf. Eq. (8a)). Later, in Sections 4.2 and
4.3, we will vary in the range 0.15 0.25 and the phases and in the whole range
, [0, 2]. In the following, we will call a model the set of twelve mixing parameters
, , , , , , , , , , , } that describes a pair of charged lepton and neu{12
1 2 1 2
13 23 12 13 23
trino mixing matrices {U , U }.
Second step: In the second step of our method, we obtain for each of the 262 144 matrix pairs
the corresponding PMNS matrix UPMNS = U U , read off the PMNS mixing parameters 12 ,
13 , 23 , , 1 , and 2 using Eq. (9), and extract those pairs {U , U }, which are in agreement
with current neutrino oscillation data. In order to have an automatic selection criterion for our
candidate pairs {U , U }, we create our best sample by defining a selector
2
2

sin 12 0.3 2
sin 23 0.5 2
S
(13)
+
.
0.3 12
0.5 23
This selector corresponds to a Gaussian 2 with the current best-fit values and the given relative
1 errors 12 9% (for sin2 12 ) and 23 16% (for sin2 23 ) from Table 1, where we assume
a Gaussian distribution in sin2 12 and sin2 23 as an approximation. For the best sample, we
choose all models that satisfy the selection criterion
S 11.83 and
sin2 13 13 0.04,
(14)
67
which corresponds to a 2 for the 3 confidence level with two degrees of freedom, and the
hard cut on 13 represents the current 3 bound according to Table 1. We do not include the
numerical value of sin2 13 in the selector in Eq. (13), because its current best-fit value would
introduce a bias towards small 13 . In some cases, we will use different values for the selection,
such as to test the experimental pressure on the model space. For example, on a time scale of
about ten years from now, we choose the values from the last column in Table 1. This selection
process is a conservative guess/crude estimator for the sample of models which can be still accommodated with data within the 3 confidence level. Note that we have, so far, neither used the
mass squared differences (and mass hierarchy), nor extracted/predicted them. At this stage, in
steps 1 and 2, our procedure is independent from the mass spectra of charged leptons and neutrinos. They become, however, important for constructing the mass matrix textures in the next step.
Third step: The third step of our approach consists of considering all the pairs {U , U } that
have been selected in the previous step 2 and rotate for each such pair the diagonal matrices
diag
diag
M and M given in Eqs. (10) back to flavor basis according to
diag
M = U M
Maj
diag
= U M
diag
MDirac = U M
UT ,
(15)
where we have chosen, for simplicity, U = U = 13 . The Dirac mass matrices M and MDirac
can thus be viewed as representatives of a class of mass matrices that is obtained by introducing
arbitrary rotation matrices U and U acting on the right-handed leptons. Moreover, note that
Eq. (15) actually describes six different cases, depending on whether one chooses from Eq. (10b)
a normal hierarchical, inverted hierarchical, or a quasi-degenerate neutrino mass spectrum. In the
Maj
following, we will, as already mentioned in the introduction, denote the mass matrices M , M ,
and MDirac , in Eq. (15) expanded in terms of as mass matrix textures or simply textures.
Let us briefly summarize again our three-step procedure for a systematic scan of real PMNS
matrices and textures with QLC:
(1) Generate 226 144 PMNS matrices by inserting all distinct combinations of sij , sij , , ,
1 , 2 , 1 , and 2 , into Eq. (8a). Here, s and s are taken from { 1 , , 2 , 0}, whereas the
ij
ij
phases can assume values and 0.

(2) Read off the mixing parameters 12 , 13 , 23 , , 1 , and 2 , using Eq. (9) and select the pairs
{U , U } where UPMNS = U U satisfies the selection criterion in Eq. (14).
(3) For each selected pair {U , U } rotate the mass matrices in Eq. (10) back to flavor basis to
Maj
obtain the textures M , M , and MDirac (cf. Eq. (15)).
We wish to emphasize that the above procedure leaves complete freedom to let the observed leptonic mixing angles be generated by contributions from both the charged lepton and/or neutrino
sector in the most general way. Another advantage of our method, that we will exploit later, is
that it allows to scan very quickly the parameter space of complex phases by varying the parameters directly in the PMNS matrix instead of varying the highly redundant phases in the mass
matrices. Our procedure to construct mass matrix textures in this way is extremely simple and
efficient. In particular, to derive the textures, there is no need to perform any diagonalization in
the whole process.
68
3. Scanning the parameter space for individual models

In this section, we apply our procedure outlined in Section 2.2 for extracting the mass matrix
textures from the best sample obtained with the selector in Eq. (13), and make it explicit for
several examples. Thereby, as explained above in step 3 of our procedure, we start with the
diagonal forms of the charged lepton and normal hierarchical neutrino mass matrices in Eqs. (10)
and obtain the mass matrices by rotating back to flavor basis following Eq. (15), where we set
for convenience U and U equal to the unit matrix. In the following, we make first a subjective
choice and focus in Section 3.1 on several textures which are of special interest like (i) the
combination of a CKM-like matrix for U with a bimaximal mixing matrix U , (ii) a texture
with very small 13 , and (iii) the sets which represent the best approximation to the current
best-fit values, i.e., those for which the selector S in Eq. (13) takes the lowest value. Later, in
Section 3.2, we present a complete list of the textures that exhibit a small 13 , as well as survive
increased experimental pressure from the other parameters.
3.1. Several examples of models compatible with current bounds
Here, we present a subjective selection of models compatible with current bounds. In addition,
we describe how we identify the leading order terms in the textures.
3.1.1. Matrices with CKM-like plus bimaximal mixing
Let us first consider the case where U is CKM-like and U is on a bimaximal mixing form. In
the parameterization of Eq. (8a), this can be realized in our procedure by taking for the charged
lepton and neutrino mixing parameters

s12 , s13 , s23 , = , 0, 2 , ,
(16a)

1

1
s12 , s13 , s23 , , 1 , 2 , 1 , 2 = , , , 0, 0, 0, , ,
(16b)
2
which lead to the PMNS mixing angles

(12 , 13 , 23 ) = (36.5 , 3.6 , 43.8 ).
(16c)
The corresponding charged lepton, neutrino, and PMNS mixing matrices are
2
2
1
2
1 +
+

0
1 2
2
2 2
2
2 2
1
2
12 + 2
U =
U =
1 2 2 ,
2+2
1
0
2
1
1 +
+
2
0.7 + 0.5 + 1.2 2

UPMNS = 0.5 + 1.2 0.8 2
0.5 + 0.5 0.5 2
0.7 0.5 + 1.2 2

0.5 + 1.2 + 0.8 2
0.5 + 0.5 + 0.5 2

1
2
1
2
1.7
0.7 2 .
0.7 1.1 2
The CKM matrix and the bimaximal mixing matrix are, on the other hand, given by
1
0
0
1 2
2
2
1
2
1
,
VCKM =
1 2 A2 and Ubimax =
2
2
2
1
1
1
2
0
A
1
2
2
2

2 2
2
2 2
(17a)
(17b)
(17c)
69
where we have used for VCKM the Wolfenstein parameterization to second order in = 0.22. The
Ubimax are (12 , 13 , 23 )CKM + bimax = (36 , 9 , 45 ).

PMNS mixing angles in UPMNS = VCKM
Comparison of Eqs. (17a) and (17c) yields that U reproduces exactly the CKM matrix in the
Wolfenstein parameterization, whereas the entries of U differ from those in Ubimax by terms
of the orders or 2 . These terms lead to the deviations between the numerical values of
(12 , 13 , 23 )CKM + bimax and (12 , 13 , 23 ) in Eq. (16c). Note that choosing the phases of U
such that all signs of the entries in Ubimax are reproduced but without changing U , changes the
PMNS mixing angles in Eq. (16c) to (12 , 13 , 23 ) (33.8 , 10.0 , 43.4 ).
diag
diag
Now, we rotate the mass matrices M = m diag( 4 , 2 , 1) and M = m3 diag( 2 , , 1)
back to flavor basis following Eq. (15), where U and U are parameterized in terms of powers
of as in Eq. (17a). In flavor basis, the resulting mass matrix textures of charged leptons and
neutrinos are then given by

0 0 0
0 0 0
2
2
2
2
M = m 0
(18a)
0 ,
0 0 1
0 0 1
3 2
2

3

2 + 2
2 2
2
2 2

3
2
2
2
Maj
1
1

3

3
M = m3
(18b)
2 4 4 1 1 ,
2 2 2 2+4 4
2
2

1
1

1
1

3

2 4 4
2 + 4 + 4
2 2
and
MDirac = m3
2
2
2
2 +

2
2
2
2
2
2
1
2
1
2
2 2
2
2
2 2
2
2
2

1 ,
1
(18c)
where we have, for simplicity, assumed the special case of a normal hierarchical neutrino mass
diag
diag
spectrum. Starting, instead with M = m3 diag(1, 1, ) or M = m3 diag(1, 1, 1), we obtain
the corresponding textures of the neutrinos for the case of inverse hierarchical or degenerate
neutrino masses. In Eqs. (18), symbolizes, up to an overall mass scale, the identification of
the leading order terms in the expansion in that contribute to the mass matrix elements in M ,
Maj
M , and MDirac . This identification is possible for all the mass matrices which are currently
valid at the 3 level, since it turns out that the contributions to a mass matrix element belonging to
different orders in can always be clearly separated from each other. In other words, for the given
expansions of the mass matrix elements in powers of , the zeroth order term in (if non-zero),
is always larger than the higher order terms, and the first order term in (if non-zero) is always
larger than the second order term. In the thus identified leading order contributions to the mass
matrices, we can then further approximate and set the order unity coefficients equal to one. This
leads us finally to a rough texture with matrix elements 1, , 2 , and 0. These entries have thus
to be understood as order of magnitude entries. Note that such textures become t Hooft natural
[43] when they arise from a spontaneously broken flavor symmetry [44] (see also Ref. [45]). An
origin of the small number is provided in terms of an anomalous U (1) symmetry [46], which
has been employed for generating textures in various models [47] (for an anomaly-free approach
see, e.g., Ref. [48]).
In the rest of the text, we will always proceed exactly the same way and determine the textures
for charged leptons and neutrinos as it was done above in arriving at Eqs. (18).
70
3.1.2. Matrices with very small 13 Near-CKM plus bimaximal mixing

We find a class of 64 models with very small 13 0.2 . All these models have in common that
12 is slightly off the current best-fit value, which means that they will be excluded in ten years
by our assumptions. Note, however, that experimentally 13 and 12 are two different degrees
of freedom, which means that there could well be much stronger pressure coming from 13
than from 12 . In this case, this class of models may be surviving very long. For one typical
representative combining near-CKM-type mixing in U with the near-bimaximal mixing in U ,
the charged lepton and neutrino mixing parameters are

s12 , s13 , s23 , = (, , 0, ),

s12 , s13 , s23 , , 1 , 2 , 1 , 2 = 1 , 0, 1 , 0, 0, 0, , ,
2
(19a)
(19b)

(12 , 13 , 23 ) = (28.7 , 0.2 , 46.1 ).
(20)
The charged lepton, neutrino, and PMNS mixing matrices are

1
1
1 2

2
2
1
2

1
0
2
U = 2
U =
,
2
2
2
1
1

1 2
2
2
1
2
2
1 + +
0
+
2
2
2
2
2
2
1

3 2
1 + .
UPMNS =
12 + 34
2 4
12 +
2

2
2
4
1
2

2
2
1
2
2
4
2 2
2

2 2
0
1
2
1
2
(21a)
(21b)
Here, U is on a bimaximal mixing form but U is not exactly the CKM-matrix (mainly due to
the entry in the 13 and 31 element). For the textures of charged leptons and neutrinos we
thus find

0 0

0 0
2
2
M = m 0
(21c)
0 0 ,
0
2
0 0 1 2
0 0 1
2
2

2

+

2 + 2
2 2
2 2
2 2
2 2

2
2
2
Maj
1

1
M = m3
(21d)
2 + 4 + 4
2 4 4 1 1 ,
2 2
2 2

1
1
2
2
2
1

1

+
2 4 4
2 + 4 + 4
2 2
2 2
and
MDirac
= m3
2
2
2
2
2

2

2
0
1
2
1
which is similar to the example given before.
2
2
2

0
1 ,
1
71
3.1.3. Best-fit matrices

Let us now consider the two best examples of models, which provide perfect fits to current
data. These two examples minimize the selector in Eq. (13), which takes for these models the
smallest possible value S = 0.12. We denote these textures as best-fit matrices since they provide a perfect fit to current data. Both examples exhibit similar PMNS mixing angles with a 13
that is in the reach of next-generation neutrino oscillation experiments. For example, 13 is large
enough to be discovered by Double Chooz, which means that for large 13 , these representatives
would be the perfect candidates.
Example (a). Our first example for the best-fit matrices has the charged lepton and neutrino
mixing parameters

1
s12 , s13 , s23 , = , , 1 , ,
(22a)
2
2

s12 , s13 , s23 , , 1 , 2 , 1 , 2 = 2 , 1 , , 0, , , , ,
(22b)
2

(12 , 13 , 23 ) = (33.4 , 7.5 , 43.5 ).
(23)
The charged lepton, neutrino, and PMNS mixing matrices are

1
U =
2
2 2
12 + 2
2
2
1
2
2 2

1
2 + 2
1
2

2
1
2

2
U =
2
2
1 +
2
2 2
12 +

2
2
2
2
2
2
,
2 2
2

2 2
1
2
,

2
2
1

+
2
2 2
1
2
1
2
0.9 + 0.6 2
UPMNS = 0.1 0.7 0.8 2
0.5 + 0.2 0.2 2
0.5 1.1 2
0.5 + 0.9 2
0.7 0.7 1.2 2
(24a)
0.1 0.4 2
0.9 0.7 0.8 2 .
0.5 1.2 + 0.5 2
(24b)
Note that, in this case, U is on a bimaximal mixing form but U is neither the CKM-matrix nor
CKM-like. For the textures of charged leptons and neutrinos we find
M = m
0
2
2
2
2
2
0 2
1 2
2 + 2
Maj
M = m3 2
12

1
2
1
3 2
4
0 2

0 2 1 ,
2 2
2
0 2 1

2 2
2

2
12 + 34
1
2

2

+ 2
2
1
1

2
2
2
2
(24c)

1
,
1
(24d)
72
and
MDirac
0
= m3
2

0

2
1
2

2
2
1
+
2
2 2
2
0
2
0

2

1
.
1
(24e)
We thus see that M is on a typical lopsided form with a strictly hierarchical structure within
Maj
each row. On the other hand, M and MDirac have nearly degenerate large entries in the first
and third row/column with small entries in the second row/column.
Example (b). Our second example for the best-fit matrices has the charged lepton and neutrino
mixing parameters

1
s12 , s13 , s23 , = , , , 0 ,
(25a)
2

2 1 1

s12 , s13 , s23 , , 1 , 2 , 1 , 2 = , , , , , , , ,
(25b)
2

(12 , 13 , 23 ) = (33.4 , 7.5 , 43.5 ).
(26)
Thus, the charged lepton, neutrino, and PMNS mixing matrices are
2
2
1

1

2
2
2 2
2
2 2
1
2
2
2
1

3
1
U =
U =

,
2 22 2 22
2 2
1
2
0
2
1 2
2 +
2
0.9 + 0.6 2
UPMNS = 0.1 0.7 0.8 2

0.5 0.2 + 0.2 2
0.5 1.1 2
0.5 + 0.9 2
0.7 + 0.7 + 1.2 2
2
2
1
+
2
2
2
1 + 2
2
and
MDirac
= m3
2
2
12
(27a)
(28a)
(28b)
2 2

2
2
2
1
2
0.1 + 0.4 2
0.9 + 0.7 + 0.8 2 . (27b)
0.5 1.2 + 0.5 2
As a consequence, we find for the textures of charged leptons and neutrinos

0
0 2
2
2
2
M = m

0 2
0 ,
0 0 1
0 0 1 2
2
2
2
1
1
1

2 + 2
2 2
2 2
2 2
2 2
1 1 1
1
2
2
2
Maj
1

1
M = m3
2 2 2 2 4 + 2 + 4 4 2 + 4 1 1 1 ,
1 1 1
2
2
2
1
1

1

4 2 + 4
4 + 2 + 4
2 2
0

2

2
2
1
2
12
2
2
2
0

1
1 .
1
(28c)
73
In this case, M is no longer of the usual lopsided form but has only entries of similar (small)
orders in the first and second row. The neutrino mass matrices, however, have entries in the last
column that differ only by order unity factors, which leads, after neglecting these factors, for
Maj
M to a democratic or anarchic texture.
3.2. Textures with small 13 surviving increased experimental pressure
Let us now concentrate on models with a small 13 that would survive a projected increased experimental pressure in ten years, i.e., pressure from a stronger 13 bound as well
as from the other parameters as estimated in the last column of Table 1. The reason for considering textures with small 13 is two-fold. First, from a theoretical point of view, an understanding of textures with small 13 seems to require interesting model-building assumptions,
e.g., discrete non-Abelian flavor symmetries [49]. Second, from an experimental perspective,
a general survey shows that most of the existing GUT model-building approaches predict a
comparatively large 13 , which is close to the current upper bound [50]. These predictions
from model-building can soon be tested in next-generation experiments such as Double Chooz,
i.e., the measurement of 13 exerts a strong experimental pressure on currently allowed models.
In Table 2, we have summarized all possible real textures (only leading order entries) that
would survive a projection of the experimental bounds on the leptonic mixing angles in ten years.
In order to extract these, we have used the extrapolated errors from the last column in Table 1
in Eqs. (13) and (14) by assuming that the current best-fit values hold. Of course, we could as
well extract the textures for any other set of best-fit values with the same procedure, but this
Maj
would exceed the scope of this paper (see also Ref. [60]). We have listed the textures M , M ,
and MDirac , for normal hierarchical, inverted hierarchical, and degenerate neutrino masses. The
textures in Table 2 have been selected from the 264 144 initial models following the procedure
in Section 2.2, which assumes that the charged lepton and neutrino mixing matrices U and U
are real. The textures in Table 2 therefore represent all the models which survive the projected
experimental bounds in ten years for the CP conserving case. The number of such models is less
than 0.01% of the initial sample.
In the notation and parameterization of Section 2.2, Table 2 shows the individual mixing
, s , s ), (s , s , s ), and ( , , , ), of U and U , as well as the PMNS
parameters (s12
1 2
12 13 23
13 23
mixing angles (12 , 13 , 23 ). Notice that we have dropped the phases 1 and 2 , since they can
be absorbed into the definition of the Majorana phases 1 and 2 . In the CP conserving case
considered here, however, the Majorana phases are either 0 or , and thus play no role for the
structure of the textures (or the PMNS mixing angles). Moreover, Table 2 contains for degenerate
neutrino masses only the Dirac neutrino mass matrix MDirac , since in the CP conserving case
Maj
M would only be proportional to the unit matrix.
For the textures #6, #11, and #12, we have given two sets of PMNS mixing angles, which
correspond to different choices of the phases 0 and in U and U , resulting in a variation of the
mixing parameters. This is not the case for all other textures in the list. Moreover, even though
different phase combinations change the PMNS mixing angles, the textures of the mass matrices
are, for normal hierarchical neutrino masses, invariant under such permutations of the phases.
For an inverted hierarchical or degenerate neutrino mass spectrum, however, a different choice
of phase combinations can manifest itself in a slightly different mass matrix texture, even in the
CP conserving case.
Normal hierarchy
M
Maj
M
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
2
0
0
2
0
0
2
0
0
2
0
0
2
0

0
1

2
1
2
0
1
2
2
1

0
1

0
2
0

2

1
0
2
0
2
0
1

Maj
M

1 1
1 1

1 1
1 1

1 1
1 1

1 1
1 1

1 1
1 1
2

2
2
2

2
2
2

2
2
2

2
2
2

2
2

1
1

1
1

1
1

1
1
2
1
1
1

1

1

1

1
2
2

1 1
1 1
2

2
2

2
1
1

1 1
1 1
2

2
2

2
1
1
1
2
2
1
2
2
MDirac

1
1

1
1

1
1

1
1
2
1
1

1
1

1
1

1
1

1
1
2
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
2
1
1
2
1
1
2
1
1
2
1
1
1
1
1
1
1
1
1 2
1
1
1 2
1
1
1 2
1
1
1 2
1
1
1 0
1
1
Degenerate
MDirac
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1 0
1
1
1 0
1
1
1
1 1
1 1
1
1 1
1 1
1
1 1
1 1
1
1 1
1 1
1 2
1 1
1 1
1 1 2
1 1 1
1 1 1
1 1 2
1 1 1
1 1 1
, s , s )
(s12
13 23
(12 , 13 , 23 )
( 1 , , 1 )
(35.2 , 4.9 , 43.8 )
, s , s )
(s12
13 23
( , , 1 , 2 )
( 2 , , 0)
2
(, , 0, + )
( 2 , , 2 )
( 1 , , 1 )
2
(35.5 , 4.5 , 41.6 )
(, , 0, 0)
(, 2 , 0)
( 1 , , 1 )
2
(35.2 , 4.8 , 46.6 )
(, 0, 0, + )
(, 2 , 2 )
( 1 , , 1 )
2
(35.5 , 4.5 , 48.9 )
(0, 0, 0, )
( 2 , , 0)
( 1 , 2 , 1 )
2
(35.2 , 3.7 , 45.1 )
(, , 0, + )
( 2 , , 2 )
( 1 , 2 , 1 )
2
(0, , 0, )
(, , 0, 0)
(, 2 , 0)
( 1 , 2 , 1 )
2
(, 0, 0, + )
(35.0 , 3.8 , 47.7 )

(35.5 , 4.6 , 43.1 )
(35.2 , 3.7 , 45.1 )
Inverted hierarchy
MDirac
74
Table 2
Mass matrix textures surviving the projected increased experimental pressure in ten years (see Table 1). For the mass matrices of charged leptons and Dirac neutrinos (for normal
hierarchical, inverted hierarchical, and degenerate neutrinos respectively), we use the basis in which the mixing angles of the right-handed fields are zero. Note that in the case
of degenerate neutrinos, the Majorana mass matrix textures are simply the unit matrix and they are therefore neglected. The corresponding mixing angles and phases of U , U
(with {0, }) as well as the mixing angles of UPMNS are also shown. Note that in some cases, a phase variation in U and U , without changing their mixing angles, can lead
to different forms of UPMNS . In these cases, we relate the causal phases of U and U to their corresponding PMNS mixing angles
Table 2 (continued)
#
Normal hierarchy
M
Maj
M
0 0
0 2
0 2
10
0 0
0 2
0 2
11a
0 0
0 2
0 2
11b
12a

1
1

1
1

1
1
0 0
0 2
0 2

1
1
2
0 0
0 2
0 2
12b
2
2

1
0 0
0 2
0 2

2

2
1
1

2

1
1

2

1
1

2

2

2

2

2

2

1
1
2

2
2

2

1
2

2
0

2
2

2
0

2
1
2

2
1
2

2
1
2
2
1
2

2
1
2

2
0
2

2
0
2

2
0
2

2
0
MDirac
1 1 0
1 1
1 1

2

1 1 2
1 1 0

1 1
2
1 1
1
( 1 , , 2 )
(, 2 , 1 )
Maj
M
MDirac

2
2
1
1
1
2
2

2

1

2
1
2

2
0 1
2

2
0 1

0
2
2
1 0
0 1
2
1 0
0 1
2
1
0
2
1
0
2
1
0
2

2
0 1
2
1
1
1
0
2
0
1
2
0
1
2
0
1
2
0
1
2
, s , s )
(s12
13 23
Degenerate
MDirac
2
1
1

2

2

2

2

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

1 1
1 1
0 2
1
1
2
0
0
1 0
1 0
0

2
1
1
2
1 1
1 1
0 2

2
(12 , 13 , 23 )
, s , s )
(s12
13 23
0
0
1 0
1 0
0
1 1 2
1 1 1
1 1 1
(, 2 , 2 )
( 1 , 2 , 1 )
2
(, 0, 0, 0)
(0, 0, 0, )
( 2 , , 1 )
2
( 1 , , 2 )
( 2 , , 1 )
1 1
1 1
0 2
1
1
2
1
1 2
0 1
1 1
1 1
0 2
1
1
2

2
1
( 1 , 2 , 2 )
2
(35.5 , 4.7 , 47.3 )
(0, , 0, )
( 2 , , 1 )
2
( 1 , 2 , 2 )
2
(35.0 , 3.9 , 42.8 )
(0, 0, 0, 0)
( 1 , 2 , 2 )
(0, , 0, 0)
1
1 2
0 1
(35.5 , 4.5 , 48.9 )
(0, 0, 0, )
2
2
2
(35.5 , 4.5 , 41.6 )
(0, 0, 0, 0)
1 1
1 1 2
1
2
(35.0 , 3.9 , 42.8 )

(35.5 , 4.7 , 47.3 )
(, 2 , 1 )
2
0 0
0 2
0 0
Inverted hierarchy
(35.5 , 4.6 , 43.1 )
(, 2 , 1 )
2
( 1 , 2 , 2 )
2
(35.0 , 3.8 , 47.7 )
(0, 0, 0, )
(continued on next page)
75
76
Table 2 (continued)
#
Normal hierarchy
M
Maj
M
13
0
0
0
14
0
0
0
15
0
0
0
16
0
0
0
17
0
0
0
18
0
0
0
19
0
0
0
20
0
0
0
0
2
2
0
2
2
0
2
2
2
2
2
2
2
2
2
2
0
2
2
0

1
1
2

1
1

1
1
2

1
1

1
1

1
1

0
1

2
1

2

2
1
2
1
1
2
1
1
1
1
1
1
1
Maj
M
MDirac

0
1
1 0
0 1 0
0
MDirac

2

2

0

2

1
2

2
0
2

2
0

2
1
2
0
1
2

0
0 1
2

2
2

2
1
1
1
1
1
1
1
2
1
1
2
1
1
1
1
1
1
1
2

2
0
2

2
0
2

0
2
2

0
2
2

2
2
2

2
2

2

2

0

0
1
2
2

2
2

2
2

2

1 0
0 1 0
0
0
1
1
0
2

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

1
1 0
1
1
1 1
1 1
(12 , 13 , 23 )
( , , 1 , 2 )
1 1
1 1 0

2
, s , s )
(s12
13 23
, s , s )
(s12
13 23
Degenerate
MDirac
1
1 1
1 1
( 2 , , 1 )
2
( 1 , , 0)
2
1 1
1 1 0
1
1
1
2
1
1
2
1
1
2
1
1
2
( 1 , , 0)
2
0
1
2

0
1
( 2 , , 1 )
2
( 1 , 2 , 0)
2
(, 2 , 1 )
2
( 1 , 2 , 0)
2
( 1 , , 1 )
2
(, 1 , 0)
2
( 1 , , 1 )
( 1 , , 0)
1 1
1 1 1
1 1 1
(35.2 , 3.8 , 50.8 )
(, , , + )
(35.2 , 3.7 , 45.1 )
(0, , 0, + )
(, 1 , 2 )
1 1
1 1 1
1 1 1
(35.2 , 4.2 , 45.4 )
(0, , 0, )
1 1
1 2
1 1
(35.2 , 4.8 , 46.6 )
(0, , 0, + )
2
1 1
1 0
1 1
(35.2 , 4.9 , 43.8 )
(33.6 , 3.1 , 52.2 )
(, , , 0)
2
(, 1 , 1 )
2
(35.2 , 3.8 , 50.8 )
(, , , + )
( 1 , , 2 )
2
(, 1 , 1 )
2
(, , , 0)
(33.6 , 3.1 , 52.2 )
0
2
2
Inverted hierarchy
77
The textures in Table 2 can be roughly divided into four classes:

and are maximal, all other mixing angles are small.
(1) Semi-anarchic textures (#18): 12
23
Maj
Here, M has a strongly hierarchical structure, whereas M and MDirac have several leading order entries.
and , are maximal, all other mixing angles are small. M
(2) Lopsided textures (#916): 23

12
is on a lopsided (highly asymmetric) form.
, , and , are maximal, all other mixing angles are
(3) Diamond textures (#1718): 12
23
13
small. M is on a lopsided form. In the neutrino mass matrix, leading order entries are situated in the corners.
, , and , are maximal, all other mixing angles are small.
(4) Anarchic textures (#1920): 12
13
23
Maj
M has a strongly hierarchical structure whereas M and MDirac have many leading order
entries.
We have called the matrices #17 and #18 diamond textures, because of the diamond-like strucMaj
ture of the small entries in M . Most of the textures in Table 2 fall either into the first or the
second class: 8 textures are semi-anarchic and 8 are lopsided. The remaining four textures are
equally distributed between the diamond and the anarchic forms. For the semi-anarchic and lopsided textures, the charged lepton and neutrino mixing matrices exhibit each a single maximal
mixing angle. The diamond and anarchic textures, on the other hand, have in U (diamond) or
U (anarchic) an additional maximal mixing angle such that the total number of maximal mixing
angles in the charged lepton and neutrino sectors is three.
The diamond textures (#17 and #18) are interesting, since they can be viewed as representing
a generalization of the lopsided models with two maximal mixing angles in the charged lepton
is maximal. Let
sector. They also have (like the anarchic examples) the unusual feature that 13
us therefore briefly consider these mass matrices in some more detail. For the diamond textures,
the charged lepton, neutrino, and PMNS mixing matrices are
1
U =
2

2
2
2 2
12
2
1
2
2 2
1

2 + 2

2
1
2

2
1
2

1
2
1
2
2 2
2
2 2
U =
1
2
2
2 2
2
2

1
2

2
2
2 2
1

2
1
2
0
,
1
2
(29)
0.9 0.1 0.2 2
UPMNS = 0.1 + 0.9 + 0.2 2

0.5 0.5 2
0.5 + 0.4 + 0.1 2

0.5 0.4 + 0.6 2
0.7 + 0.5
0.1 0.4 0.3 2

0.9 0.4 0.3 2 .
0.5 0.7 + 0.3 2
(30)
We thus obtain for the expansion of the diamond textures in powers of the forms
M = m
0
2
2
2
2
2

1
2
1
2
2 2
2
2 2
(31a)
78
and
Maj
M
1
2
2
2
2
= m3
2
12

2
2
2

2
12 +
2
2
2
1
2
2
2
2

2
MDirac = m3 0
2

2

2

2

2
1
2
0 .
1
2
(31b)
Keeping in the expansion in only the leading order coefficients and approximating these by
one, we arrive from Eqs. (31) at the textures #17 and #18 in Table 2. The best-fit matrices of
Section 3.1 that provide a perfect fit to current data are not contained in Table 2, since they exhibit
a reactor angle 13 = 7.5 that would be ruled out by the projected confidence levels in 10 years.
The diamond textures #17 and #18 can, however, be considered as a crude approximation of the
best-fit matrices.
It is interesting to compare the results in Table 2 also with tribimaximal mixing [51], which
has the PMNS angles (12 , 13 , 23 )tribi = (35 , 0 , 45 ). In our procedure, neitherU or U are
tribi = 1/ 3. Nevertheon a tribimaximal form, since our parameter space does not contain s12
less, we can associate some textures in Table 2 with perturbations to tribimaximal mixing. For
instance, the textures #5, #7, #15, and #16, could be viewed as describing (small) deviations
from tribimaximal mixing (deviations from tribimaximal mixing have been considered previously e.g., in Ref. [52] and references therein). Different from the literature, however, none of the
examples for tribimaximal-like mixing in Table 2 is associated with a charged lepton mixing
matrix U that is close to unity. The tribimaximal-like mixing emerges in these cases, instead,
from large mixings in both the charged lepton and the neutrino sector.
Up to corrections of the order 3 , all the data models (in the sense of Section 2.2) for the
textures #116 in Table 2 exhibit values for 12 that follow the relation

2
12 + + = .
4
2
2
(32)
This corresponds to the usual QLC relationship in Eq. (4). The models for the diamond and
anarchic textures #1720 satisfy, instead, the new sum rule
12 +
= arctan(2
5+2 2
2 ).
(33)
This new relation is an outcome of our extended QLC approach. The models for the textures
#116 exhibit

13 = O() and 23 = + O 2 ,
(34)
4
whereas the models for the diamond and anarchic textures #1720 reveal

1
13 = arcsin 14 (2 2 )
(35a)
,
5+2 2

1
23 = arctan 1 + 1 + (2 11 2 ).
(35b)
2
17
In Eq. (34), 13 is proportional to , but it is nevertheless small due to a small coefficient
multiplying . In contrast to this, in Eq. (35a), a small 13 arises from corrections to a
small O(1) 8 term. A close to maximal 23 results in Eq. (35b) from a correction to
79
arctan(1 + 1/ 2 ) 60 . We would like to emphasize again that the new sum rules in Eqs. (33)
and (35) are, together with the known relations of Eqs. (32) and (34), consequences of the extended QLC approach.
4. Distribution of observables: Mixing angles
In this section, we investigate the ensemble of models found in our best sample. This includes
all models as defined by our selector (cf. Eq. (13)) for the currently allowed parameter ranges.
We will be mainly interested in distributions of observables from this ensemble of models on
a statistical basis rather than individual models. Note that there will be no predictions for the
mass hierarchy, as well as there will be no dependence of the predictions on the hierarchy. The
reason for this is the independence of the selector and the model building process on the mass
hierarchy (only the neutrino mass textures depend on the hierarchy, which we do not consider in
this section; this means that step 3 in Section 2.2 will not be necessary). For most of this section,
we assume real mass matrices, but we will demonstrate the dependence on the Dirac-like phases
and . In all of the figures shown in this section, a valid model corresponds to a model
generated by our procedure which is consistent with current bounds. Therefore, the interpretation
of the figures is different from so-called scatter plots showing the parameter space density for
certain assumptions on the input variable distributions. Our figures represent discrete predictions
for specific models, where each point/model can be connected with a specific texture.
Note that the interpretation of the distributions of observables as predictions has to be done
with great care, since in nature only one model is actually implemented. In addition, one has to
use a measure in theory space, which we have done by our discrete choices of mixing angles
in U and U . However, one can use excluded regions in the parameter space (for the given assumptions) for strong conclusions. In addition, it is useful to study the model parameter space as
a whole in order to obtain hints on how experiments can test this parameter space most efficiently
(cf. Section 4.4).
4.1. Distribution of mixing angles for fixed
Let us first of all fix to 0.2, as we have done before. In this case, our best sample contains
all allowed models for this fixed , which are 2468 out of the initial 262 144 models. We show
the distributions of mixing angles for these models in Fig. 1 in terms of histograms, where the
gray-shaded regions mark the current 3 -excluded regions. The vertical lines/arrows mark the
3 exclusion potential of selected future experiments.3 For sin2 213 , the distribution is peaking
at rather large values, i.e., most models are within the range of Double Chooz or the T2K and
NOA superbeams. T2HK could then exclude most of the rest, and a neutrino factory almost
all of the rest. We will discuss the number of allowed models as a function of the expected
precisions in greater detail later. In fact, the smallest sin2 213 in all of the best sample models
is 3.3 105 , i.e., there is not a single model with sin2 213 0. This value is not far below the
reach of a neutrino factory.4
3 The sensitivities are, for Double Chooz, taken from Ref. [14], for T2HK, taken from Ref. [53], for the neutrino
factory, taken from Ref. [54] for two baselines, and for SPMIN/SADO (including solar data), taken from Refs. [16,55]
for a luminosity of 20 GW kt yr.
4 Extending the allowed values for s and s by arctan 1/ 2 would produce models with sin2 2 0. However,
13
ij
ij
this class of models is not included in our initial hypothesis.
80
For sin2 23 , the selected models peak around maximal mixing and at sin2 23 = 0.5 , as one
may expect by considering as a perturbation. This means that the value of may determine the
deviation from maximal mixing for a large class of models. Interestingly, the peaks around 0.3
and 0.7 are already under strong experimental pressure right now, which means that an exclusion
may come shortly. However, choosing a somewhat smaller may change this argument, as we
will discuss in the next subsection. Note that there are gaps between the outer peaks and the
maximal mixing peak with no models at all, which makes these two classes very distinguishable.
We have also tested the predictions for 23 > /4 versus 23 < /4, but we have not found a
substantial deviation from a 50:50 distribution.
As far as sin2 12 is concerned, the distribution is rather flat in the currently allowed region.
In fact, the main peak around sin2 12 0.36 is off the current best-fit value (but well within the
currently allowed range), which means that future precision measurements (such as by a large
reactor experiment SADO/SPMIN) could exert strong pressure on the models. Except from these
parameters, we have also tested the predictions for . Since we assume real matrices in this part,
only 0 and can be generated (CP conservation). We have not found any substantial deviation
from an equal distribution.
It is now interesting to compare our distributions with the literature, where we have chosen
two specific examples for sin2 213 predictions. In Ref. [56], random mixing matrices and their
predictions for sin2 13 were investigated (anarchy). Our distribution for sin2 213 follows the
general anarchy trend, which is not surprising for a statistical ensemble of models. Note, however,
that our initial assumptions are qualitatively very different and we have a class of discrete models
here. For example, we observe a number of discrete possible models for very small sin2 213 . One
can also see this excess on a linear scale in sin2 213 : The more or less uniform distribution in
sin2 213 (creating a distribution in log(sin2 213 ) peaking at large values) has a tendency to small
values of sin2 213 . Another study, which reviews existing models in the literature by the model
class and their predictions for sin2 213 , is Ref. [50]. Although our model is motivated by GUTs,
it does not at all show the sin2 213 distribution obtained for GUT models in the literature (which
peak around sin2 213 0.04). In particular, we obtain an excess of very large and very small
sin2 213 models. We find this result very interesting, because our approach does not imply a
particular matter of taste for the result, whereas in the literature, many models might have been
biased with respect to the outcome.
4.2. Tuning
So far, we have assumed C 0.2 fixed. However, it may well be that a slightly different
choice for will shift specific models in our selection range, i.e., that there will be more models
allowed. From neutrino physics only, one can derive using specific assumptions. For example,
for a normal hierarchy of neutrino masses using m1 : m2 : m3 = 2 : : 1, we obtain from the
best-fit values and allowed ranges in Table 1 the range 0.15 0.22 (3 ) with a best-fit value
0.18. Of course, these values and ranges are based on more assumptions that we have used
before, and one may find counter-arguments from the quark sector (such as using the Cabibbo
angle instead). Therefore, in order to test the effect of different s, we vary in a symmetric
range around 0.2 between 0.15 and 0.25. Then we choose, for each model, the which minimizes
our selector value.
First of all, we observe that the number of valid models according to our selector increases
from 2468 to 3316. We show the distribution of these models in Fig. 2, which is similar to Fig. 1
and should be compared with that. We find that for sin2 23 (middle panel), the gaps between
81
Fig. 1. Distributions of sin2 213 (left), sin2 23 (middle), and sin2 12 (right) of the models of our best sample for = 0.2
fixed. The bars show the number of selected models per bin, i.e., per specific parameter range. The gray-shaded regions
mark the current 3 -excluded regions (cf. Table 1).
Fig. 2. Distributions of sin2 213 (left), sin2 23 (middle), and sin2 12 (right), of the models of our best sample for the
optimal (see main text).
sin2 23 0.5 0.2 and sin2 23 0.5 (maximal mixing) are getting filled by some models. This
observation supports the hypothesis that the positions of the outer peaks are indeed determined
by the value of . For sin2 12 (right panel), the main peak is now at the current best-fit value
sin2 12 0.3. This means that the relatively small error on sin2 12 exerts pressure on the choice
of . However, this also implies that an adjustment of can, for most models, circumvent the fact
that for fixed most models do not hit the currently allowed value of sin2 12 for future precisions
of sin2 12 (cf. Fig. 1, right panel). As the last indicator, we observe that the sin2 213 distribution
is now even more peaking at larger values of sin2 213 (cf. Fig. 2, left panel), with most models
being very close to the current bound. However, a small number of models with small sin2 213
still survives. In summary, we find that adjusting does not change the qualitative conclusions
from the previous section, but it allows for even more allowed models with a tendency of better
sin2 12 fits and larger predicted values for sin2 213 .
4.3. Impact of non-vanishing Dirac-like phases
We now discuss the impact of Dirac-like CP phases and in U and U , respectively.
So far, we have assumed that all phases be 0 or in Eq. (8a), and we have obtained the CP
conserving values 0 and for in UPMNS without significant preference for either one. Let us
now still fix 1 , 2 , 1 , and 2 , in (8a) to their CP conserving values 0 or , and average over
82
Fig. 3. Distributions of sin2 213 (left), sin2 23 (middle), and sin2 12 (right), of the models of our best sample averaged
over the Dirac phases and using uniform distributions (see main text).
all possible Dirac-like phase values of and . Note that this implies that there will be nontrivial values for both the Dirac and Majorana phases in UPMNS , which can be interpreted for
predictions as well. Since these predictions are a bit more model-dependent than most of this
section, we discuss them in the next section.
As far as the procedure is concerned, we now vary and from 0 to 2 in 32 equi-distant
steps (excluding 2 ), i.e., we generate 32 32 = 1024 models instead of four before (two values 0 and times two phases). This implies that we simulate a uniform distribution in these
phases. Then we choose the allowed models according to our selector, calculate the parameter
predictions, and, in order to compare with previous results, divide the number of valid models in
each bin by 1024/4 = 256 (averaging). Indeed, using this averaging process, we find a number of
1570 valid models (instead of 2468 for the Dirac-like phases fixed to the CP conserving values).
Though the order of magnitude is the same, this implies that about 64% of all phases are as good
as 0 and (on average), whereas one may naively expect about 50%.
We show the effects of the Dirac phase averaging in Fig. 3, which should be compared with
Fig. 1. The qualitative result is, again, very similar to Fig. 1. For sin2 213 (left panel), the distribution becomes smoother because of the averaging process. Similar to the adjustment, larger
values of sin2 213 become more pronounced. For sin2 23 (middle panel), we find some models
in the gaps between 0.5 and 0.5 0.2, as one may expect from a smearing of a subset of models.
In addition, some peaks at around 0.5 0.2 become relatively more pronounced, because there
= 0 and/or = 0). For sin2 ,
is a subset of models which is independent of and (for 13
12
13
we observe that this smearing fills the gaps which are present for fixed , but the qualitative
picture remains unchanged. Therefore, we do not find significant deviations from our earlier distributions. However, there is a slight excess of models favoring 23 < /4 versus 23 > /4, as
it is already obvious from Fig. 3, middle panel. As the most interesting part, we now also obtain
non-trivial predictions for and the Majorana phases in UPMNS . We discuss these in the next
section.
4.4. Increase of the experimental pressure
Here, we analyze future improving experimental constraints and how they affect the selection
of models, where we restrict this discussion to = 0.2 fixed. First of all, we show in Fig. 4 the
distribution of models in the sin2 213 sin2 23 (upper left), sin2 12 sin2 23 (upper right), and
sin2 213 sin2 12 (lower left) planes, where each point corresponds to one or more models with
83
Fig. 4. Distribution of models for fixed = 0.2 for the best sample based on current experimental bounds (all marks), as
well as remaining models for the bounds expected in ten years from now (dark/blue stars only); cf. last column of Table 1
for this projection. The distributions are given in the sin2 213 sin2 23 (upper left), sin2 12 sin2 23 (upper right), and
sin2 213 sin2 12 (lower left) planes, where the best-fit values are marked by lines. Each point corresponds to one or
more models predicting these parameter values. (For an interpretation of the references to colour in this figure legend,
the reader is referred to the web version of this article.)
a specific parameter combination. The figure shows all the models from our best sample based on
current experimental bounds, whereas the dark stars mark the models remaining after increasing
the experimental pressure on a time scale of ten years. The corresponding scenario in ten years
is defined in Table 1, where we assume that 13 will not be discovered. Obviously, the improving
sin2 213 bound restricts the model space very strongly (left panels), where many models close
to maximal mixing will survive (upper left panel). However, only very few models close to the
sin2 12 best-fit value survive (lower left panel). In particular, none of the remaining models can
reproduce both the solar and atmospheric best-fit values closely (upper right panel). Note that the
models with extremely small sin2 213 will be eliminated as well, mainly because of the pressure
from sin2 12 (cf. lower left panel). Therefore, we expect that the strongest experimental pressure
on this model space will come from sin2 213 and sin2 12 , which are two different degrees of
freedom since they will be driven by different classes of experiments (such as beams and shortbaseline reactor experiments versus solar and long-baseline reactor experiments). Note that this
discussion assumes that sin2 213 will not be discovered, whereas a sin2 213 discovery would,
depending on the new best-fit value, select a different class of models.
84
Fig. 5. Number of valid models as function of the sin2 213 bound (left), error on sin2 23 (middle), and error on sin2 12
(right). The right edges of the plots correspond to the current errors, the left edges to exactly known parameters. The light
(yellow) curves refer to the current errors for the other two degrees of freedom. The dark (red) curves assume the other
errors to be as in the scenario in ten years (cf. Table 1). The other indicated experiments marked by the vertical lines
are referred to in the caption of Fig. 1. (For an interpretation of the references to colour in this figure legend, the reader
is referred to the web version of this article.)
In order to demonstrate the continuous dependence on experimental constraints, we show in

Fig. 5 the number of valid models as a function of the sin2 213 bound (left), error on sin2 23
(middle), and error on sin2 12 (right). The right edges of the plots correspond to the current
errors, the left edges to exactly known parameters. The light (yellow) curves fix the errors for
the other two degrees of freedom to their current values, whereas the dark (red) curves assume
the other errors from the scenario in ten years defined in the caption of Fig. 4. Comparing the
light/yellow curves only, the strongest experimental pressure will be exerted by sin2 213 and
sin2 12 , which may each independently reduce the number of models to one third on a timescale
of ten years. This means that sin2 213 experiments affect the valid models as much as solar and
potential long-baseline reactor experiments (SPMIN/SADO), and already one of these experiment classes already acts as a strong model discriminator. The reason is the generic prediction of
the value of sin2 12 from the concepts of maximal mixing combined with deviations, whereas
maximal mixing is used as an initial hypothesis for our models. In order to estimate the combined potential of different experimental degrees of freedom, we show as the dark/red curves
the number of valid models for smaller errors on the parameters not shown (approximately on a
time scale of ten years). Obviously, sin2 213 is a necessary discriminator to exclude all possible
models very quickly, whereas no further increase in the sin2 213 bound would make a further
model discrimination very hard.
5. Distribution of observables: Dirac and Majorana PMNS phases
We now demonstrate that the results in Section 4.3 can be used for non-trivial predictions of
the Dirac and Majorana phases in UPMNS . We have decoupled this discussion from the previous
sections, because the results in this section will be more dependent on the assumptions used.
However, we believe that already the procedure used in this section warrants some attention,
because it opens a new way of model building predictions. The actual dependence on the assumptions, which need to be chosen according to the underlying theory, deserves further studies.
Note that, as in the last section, the predictions can be used to study the impact of experiments
on the parameter space of models, but one has to be careful to interpret them as predictions for
the actual theory.
85
Fig. 6. Distribution of the Dirac phase for our best sample for all models of that sample (left), only models with at least
> 0 or > 0 (middle), and models with both > 0 and > 0. There are no particular assumptions for
one 13
13
13
13
13
(in UPMNS ).
In Section 4.3, we have assumed that 1 , 2 , 1 , and 2 , in Eq. (8a) be 0 or , but the
Dirac-like phases and be uniformly distributed. We have then generated all possible models for different phases and selected our best sample which has been compatible with current
data. These phases generate non-trivial Dirac () and Majorana (1 , 2 ) phases in the product
UPMNS = U U , which can be predicted from our best sample. One can now argue if our set
of assumptions is plausible (such as the set of values used for the mixing angles), if one should
include variations of the other phases, if the uniform distributions for the Dirac-like phases make
sense, etc. Therefore, one has to be careful with the interpretation of the results in this section. It
is, however, the purpose of this section to demonstrate how our procedure can provide non-trivial
predictions for phases under certain assumptions.
First of all, we show in Fig. 6 the distribution for (in UPMNS ) for our best sample. As before,
we have generated 32 32 models for all possible pairs of Dirac-like phases and , have
chosen our best sample, and then have normalized the histograms by 4/1024 = 1/256 (because
we replace four choices of the phases by 1024). In the left plot, we show the distribution for
all models in the best sample. Obviously, there is a strong peak at , which predominantly
= = 0. In this case, U and U are real matrices, and the product
comes from models with 13
13
has to be real as well, which only allows the phases 0 and for . There is no smearing of these
models coming from the phase averaging, because and are undefined. This means that
the averaging over these phases leaves the peak untouched and smears out many of the other
models, leading to a relative enhancement at the CP conserving choice. The strong preference
of compared to 0 comes from the selector from a non-trivial restriction of the parameter space
in all three mixing angles, it is not present before the selection process.
In some sense, the overall preference of CP conservation is connected with our choice of 0
and , which means that it is model-dependent.
as one of the values for the mixing angles 13
13
= 0 or = 0
One can now argue that it does not make sense to consider the cases with 13
13
because in this case the Dirac-like phases are not physical and therefore introduce a bias (these
models have a stronger relative weight because the phases are not varied). Therefore, we show
in Fig. 6, middle and right, only the models with at least one physical (= defined) Dirac-like
phase (middle) and two physical Dirac-like phases (right) in U and U . Obviously, there is a
shift in the preference of . However, all of the plots in Fig. 6 have in common that maximal CP
violation is disfavored. The reason for the peaks close to CP conservation is the generation of
the PMNS Dirac phase in (UPMNS )e3 as a combination of the Dirac-like phases from U and U
86
Fig. 7. Distribution of the Majorana phases 1 and 2 from our best sample. Each point corresponds to one or more
models predicting the corresponding phases.
(times some factors) plus a constant term acting CP conserving. If this constant term is larger
than the phase-dependent factors, there will be CP conservation disturbed by some -size (or
2 -size) contribution coming from the mixing angles. If this constant term is smaller than the
phase-dependent factors, the uniform phase distributions will translate into uniform distributions
for the PMNS Dirac phase. Note that we have still kept all the other phases 1 , 2 , 1 and 2 ,
fixed at their CP conserving valueswhich is a model dependent assumption (and a constraint
from computation power). However, note that 1 and 2 only affect the Majorana phases in
UPMNS .
Our approach does not only predict the PMNS Dirac phase, but also a non-trivial set of Majorana phases. We show in Fig. 7 this distribution for our best sample. Obviously, there is some
clustering close to the CP conserving values, but, in principle, the whole parameter space is covered. In addition, note that the parameter space [, 2] is redundant for 0 decay, because
only the square of the phase enters.
The predictions for these phases for one specific model become very interesting in combination with the mixing angles for that model, because this set of parameters can be used for a direct
prediction of the mass matrix element for 0 decay for Majorana neutrino masses. We follow
the calculations in Ref. [57] (for a recent review, see also Ref. [58] and references therein), where
the absolute value of the mass matrix element mee , which is proportional to the rate of the 0
decay, is given by5

(2)
2i( )
(3)
2i( +)
2
1 +
m
1
|mee |
Uei2 mi
with mee =
m(1)
ee + mee e
ee e
(36)
and
5 Note the change in notation for the phases: We read off and in a way which is related to and in Ref. [57]
1
2
by = 2 1 , = 1 .
87
Fig. 8. Distribution of |mee | from our best sample for a normal mass hierarchy and m1 as chosen in the plot labels.
(1)
m
= m1 |Ue1 |2 = m1 c2 c2 ,
ee
12 13
(2)
m
= m2 |Ue2 |2 = m2 s 2 c2 ,
ee
12 13
(3)
m
= m3 |Ue3 |2 = m3 s 2 .
ee
13
(37)
Given the mass hierarchy and using the mass squared differences from Table 1, one can use
these equations to compute |mee | as function of the lightest neutrino mass m (m1 for the normal hierarchy, and m3 for the inverted hierarchy). Each of our models predicts the combination
(1 , 2 , 13 , 12 , ) relevant for 0 decay, which results in a particular prediction of |mee | as a
function of m. We show this prediction for the normal mass hierarchy for two different values of
m = m1 in Fig. 8. The left plot corresponds to a vanishing m1 , the right plot to the chimney,
where |mee | may even vanish. From these distributions one can read off what may be obvious:
Making the 0 decay rate vanish could mean some sort of fine-tuning because it requires a specific combination of phases and mixing angles (cf. Fig. 1 in Ref. [57] for illustration). Therefore,
none of our model predicts a vanishing 0 decay rate.
We show in Fig. 9 the theory allowed regions (left) and our model predictions (right) as
function of the lightest neutrino mass. The general theory prediction is obtained by varying the
mixing angles (and phases) in their current 3 allowed ranges. Because of the relatively weak
sin2 213 bound the chimney is not explicitly visible in the left plot, and a vanishing 0 decay
rate is possible for the normal hierarchy. In the right panel of Fig. 9, we compute for each m the
minimum and maximum of |mee | from all models in the best sample. In addition, we show the
curves where 99% and 95% of all models are above. As the most interesting result, we find that
99% of all normal mass hierarchy models are above 0.001 eV independent of m1 . In addition,
comparing the two panels of Fig. 9, we find that the actually predicted ranges will be much more
narrow than the corresponding theoretically allowed ranges. Again, this result depends on the
assumptions used for the phase generation, in particular, the selection of values for the mixing
angles, the uniform distributions of the Dirac-like phases and , and the selection of CP
conserving values for the other phases. However, the obvious observation that one needs to finetune the parameters in order to make 0 decay vanish, seems to be quite general. Therefore, we
88
Fig. 9. General theory allowed regions (left) and distribution of our best sample prediction (right) for |mee | as a function
of the lightest neutrino mass m. In the left plot, we show the currently allowed regions for |mee |, where the mixing angles
are varied in the current 3 ranges from Table 1 (computed with the formulas given in Ref. [57]). Note that fixing 13
would result in the appearance of the chimney. In the right plot, the bands correspond to all possible models from our
best sample. The different intermediate curves limit the regions where 99% and 95% of the models can be found above.
The vertical line corresponds to the choice of Fig. 8, right panel (for the normal hierarchy). Note that in the right panel,
we use both the mixing angle and Majorana phase predictions for each model, i.e., we use the combination of all available
predictions for that model (and not only the Majorana phases). In both panels, the limit from cosmology is obtained from
a combined SDSS and WMAP analysis [10], and the limit from 0-decay is obtained by the HeidelbergMoscow
Collaboration [59] (with an uncertainty coming from the calculations of the nuclear matrix elements). We fix the mass
squared differences in both panels to their best-fit values in Table 1.
expect that in most concrete realizations, there will be non-vanishing 0 decay for Majorana
neutrino masses.
6. Summary and conclusions
As a minimal unified approach to the fermion mass and mixing parameters in the Standard
Model, it is attractive to assume that all deviations from symmetries or zeros, such as deviations
from maximal atmospheric neutrino mixing, the small mixing angle 13 in UPMNS , the Cabibbo
angle C in VCKM , and the mass hierarchies, may be described by powers of a single small
quantity C . This small quantity can be motivated by Grand Unified Theories connecting
quarks and leptons. Quarklepton complementarity can be interpreted as a phenomenological
implementation of this approach, obtaining the solar mixing angle as 12 + C /4. We have
introduced an extended quarklepton complementary approach as an extension of this relationship. Since UPMNS arises as a product of the charged lepton and neutrino mixing matrices, i.e.,
UPMNS = U U , we have postulated that all mixing angles in U and U be given by either maximal mixing or powers of . In this way, the observed solar mixing angle can only result from
taking the product in UPMNS = U U . In addition, deviations from maximal atmospheric mixing, 13 , etc., emerge as predictions in this minimal unified approach as being given by powers
of . Note that we have not assumed specific forms for U and U , such as bimaximal mixing.
However, we have obtained configurations involving bimaximal mixing as special cases.
89
Using this assumption for the mixing angles in U and U up to the order 2 , i.e., the mixing
angles be /4, , 2 , or 0, we have systematically tested all possible CP conserving configurations, a total of 262 144 generated models. Naturally, we had to introduce an automated selector
for the search for models compatible with current experimental data, leading to a best sample of
2468 models. As a first analysis, we have scanned for particularly interesting textures, such as
with U VCKM and U Ubimax , or models being perfectly consistent with current data. We
have found that these models imply that 13 be large, which means that many of these models
could be ruled out very soon due to stronger bounds on 13 . However, if 13 turns out to be large,
we have identified examples with (12 , 13 , 23 ) = (33.4 , 7.5 , 43.5 ), providing a perfect fit.
As the next step, we have particularly focused on models with small 13 and increased pressure
on the other oscillation parameters in order to further reduce our sample. These models are interesting because they could survive the next ten years even if 13 was not discovered. We have
classified the corresponding textures systematically by their leading order entries, and we have
found 20 distinctive sets of textures. These textures can be divided into different classes, such as
lopsided models, anarchic models, etc. As a very interesting class, we have identified a number
, , and , which we have called diamond models (beof models with maximal mixings 12
13
23
Maj
cause of the diamond-like structure of the small entries in M ). We have shown that the sum
rules for this and the anarchic-like classes of models are qualitatively different from the ones in
standard quarklepton complementarity.
In an independent approach, we have investigated the mixing angle distributions for our best
sample compatible with current data. These distributions can be used to study the impact of future
experiments on the model parameter space, and should be interpreted as predictions in that way.
As the main result, we have found rather large values of sin2 213 preferred, and sin2 213 >
3.3 105 for all of our models. In addition, sin2 23 peaks around 0.5 and at 0.5 , whereas
sin2 12 peaks above the current best-fit value. Compared to the GUT model literature (surveyed
in Ref. [50]), we have not found a characteristic peak of models at around sin2 213 0.04. Since
our matrix generation and selection process has been quite un-biased, this could point towards
a bias in the building of specific GUT models. Compared to anarchic models, we find an excess
for small sin2 213 . In the future, especially sin2 213 experiments (such as superbeams or shortbaseline reactor experiments) and potential sin2 12 experiments (such as a long-baseline reactor
experiments) will put the model parameter space under pressure. Because we have used maximal
mixing as an input and have generated deviations from that by powers of C , sin2 12 is an
important discriminator for this class of models. However, in order to exclude all models, very
strong bounds for sin2 213 are needed, such as those coming from a neutrino factory.
In a more specific predictability part, we have generated (in U ) and (in U ) with uniform distributions in order to obtain predictions for and the Majorana phases in UPMNS . This
set of assumptions is more model-dependent than the rest of this work, but it allows a very powerful handle on phase predictions. For example, we have found that for (in UPMNS ), maximal
CP violation is significantly disfavored, because it requires large imaginary parts that are hard
to obtain from the construction UPMNS = U U . Furthermore, we have combined the non-trivial
predictions of the Majorana phases with the mixing angle predictions for each model, and we
have predicted the 0 decay rates. Not a single of our models predicts a vanishing 0 decay
rate, because the necessary phase cancellation requires fine-tuning not present in the discussed
model parameter space. We have found that 99% of all models predict |mee | > 0.001 eV independent of the mass hierarchy and lightest neutrino mass, i.e., the chimney is, in practice, not
present.
90
In conclusion, we have used a novel approach for studying neutrino mass matrices, which is a
mixture between basic fundamental assumptions and the systematic machinized parameter space
scan of a very high dimensional parameter space. This approach has turned out to be extremely
powerful, because it does not require the diagonalization of matrices. The primary objective of
this work has been to stay as far away from specific assumptionswhich may introduce a bias
as currently possible from the computational point of view. With this approach, we have not only
been able to scan a large parameter space systematically, but also been able to make predictions
based on a large sample of models compatible with current data. Each of these predictions for the
mixing angles can be connected to a particular texture, which is different from parameter space
scans using particular assumptions for the input variable distributions and investigating the parameter space density. Only in the last part we have combined these two methods because we have
used uniform distributions for the Dirac-like phases as an assumption. Naturally, we have not
been able to present all of our results here, but have focused on the most interesting partsin our
opinion, of course. The interested reader can find a tool to view all models from our best sample
compatible with current data in Ref. [60]. Finally, we believe that the connection between quarks
and leptons could be the key element in the motivation of future neutrino facilities, and we have
demonstrated how this element can be implemented phenomenologically in a straightforward
scheme.
Acknowledgements
The research of F.P. is supported by Research Training Group 1147 Theoretical Astrophysics
and Particle Physics of Deutsche Forschungsgemeinschaft. G.S. was supported by the Federal
Ministry of Education and Research (BMBF) under contract number 05HT1WWA2. W.W. would
like to acknowledge support from the Emmy Noether program of Deutsche Forschungsgemeinschaft.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
W.M. Yao, et al., Particle Data Group, J. Phys. G 33 (2006) 1.

N. Cabibbo, Phys. Rev. Lett. 10 (1963) 531.
M. Kobayashi, T. Maskawa, Prog. Theor. Phys. 49 (1973) 652.
S. Fukuda, et al., Super-Kamiokande Collaboration, Phys. Lett. B 539 (2002) 179, hep-ex/0205075.
Q.R. Ahmad, et al., SNO Collaboration, Phys. Rev. Lett. 89 (2002) 011302, nucl-ex/0204009.
Y. Fukuda, et al., Super-Kamiokande Collaboration, Phys. Rev. Lett. 81 (1998) 1562, hep-ex/9807003.
T. Araki, et al., KamLAND Collaboration, Phys. Rev. Lett. 94 (2005) 081801, hep-ex/0406035.
M. Apollonio, et al., CHOOZ Collaboration, Eur. Phys. J. C 27 (2003) 331, hep-ex/0301017.
E. Aliu, et al., K2K Collaboration, Phys. Rev. Lett. 94 (2005) 081802, hep-ex/0411038.
M. Tegmark, et al., SDSS Collaboration, Phys. Rev. D 69 (2004) 103501, astro-ph/0310723.
P. Minkowski, Phys. Lett. B 67 (1977) 421;
T. Yanagida, in: Proceedings of the Workshop on the Unified Theory and Baryon Number in the Universe, KEK,
Tsukuba, 1979;
M. Gell-Mann, P. Ramond, R. Slansky, in: Proceedings of the Workshop on Supergravity, Stony Brook, New York,
1979.
R.N. Mohapatra, G. Senjanovic, Phys. Rev. Lett. 44 (1980) 912;
R.N. Mohapatra, G. Senjanovic, Phys. Rev. D 23 (1981) 165;
J. Schechter, J.W.F. Valle, Phys. Rev. D 22 (1980) 2227;
G. Lazarides, Q. Shafi, C. Wetterich, Nucl. Phys. B 181 (1981) 287.
T. Schwetz, Phys. Scripta T 127 (2006) 1, hep-ph/0606060.
P. Huber, J. Kopp, M. Lindner, M. Rolinec, W. Winter, JHEP 0605 (2006) 072, hep-ph/0601266.
S. Antusch, P. Huber, J. Kersten, T. Schwetz, W. Winter, Phys. Rev. D 70 (2004) 097302, hep-ph/0404268.
91
[16] H. Minakata, H. Nunokawa, W.J.C. Teves, R. Zukanovich Funchal, Phys. Rev. D 71 (2005) 013005, hep-ph/
0407326.
[17] A.Y. Smirnov, hep-ph/0402264.
[18] M. Raidal, Phys. Rev. Lett. 93 (2004) 161801, hep-ph/0404046.
[19] H. Minakata, A.Y. Smirnov, Phys. Rev. D 70 (2004) 073009, hep-ph/0405088.
[20] S.T. Petcov, A.Y. Smirnov, Phys. Lett. B 322 (1994) 109, hep-ph/9311204.
[21] B. Pontecorvo, Sov. Phys. JETP 6 (1957) 429.
[22] Z. Maki, M. Nakagawa, S. Sakata, Prog. Theor. Phys. 28 (1962) 870.
[23] G. Altarelli, hep-ph/0611117.
[24] F. Vissani, hep-ph/9708483;
V.D. Barger, S. Pakvasa, T.J. Weiler, K. Whisnant, Phys. Lett. B 437 (1998) 107, hep-ph/9806387;
A.J. Baltz, A.S. Goldhaber, M. Goldhaber, Phys. Rev. Lett. 81 (1998) 5730, hep-ph/9806540;
G. Altarelli, F. Feruglio, Phys. Lett. B 439 (1998) 112, hep-ph/9807353;
M. Jezabek, Y. Sumino, Phys. Lett. B 440 (1998) 327, hep-ph/9807310;
D.V. Ahluwalia, Mod. Phys. Lett. A 13 (1998) 2249, hep-ph/9807267.
[25] M. Jezabek, Y. Sumino, Phys. Lett. B 457 (1999) 139, hep-ph/9904382.
[26] C. Giunti, M. Tanimoto, Phys. Rev. D 66 (2002) 113006, hep-ph/0209169.
[27] P.H. Frampton, S.T. Petcov, W. Rodejohann, Nucl. Phys. B 687 (2004) 31, hep-ph/0401206.
[28] T. Ohlsson, Phys. Lett. B 622 (2005) 159, hep-ph/0506094.
[29] S. Antusch, S.F. King, Phys. Lett. B 631 (2005) 42, hep-ph/0508044.
[30] W. Rodejohann, Phys. Rev. D 69 (2004) 033005, hep-ph/0309249.
[31] N. Li, B.-Q. Ma, Phys. Rev. D 71 (2005) 097301, hep-ph/0501226.
[32] Z.-z. Xing, Phys. Lett. B 618 (2005) 141, hep-ph/0503200.
[33] A. Datta, L. Everett, P. Ramond, Phys. Lett. B 620 (2005) 42, hep-ph/0503222.
[34] L.L. Everett, Phys. Rev. D 73 (2006) 013011, hep-ph/0510256.
[35] M.A. Schmidt, A.Y. Smirnov, hep-ph/0607232;
A. Dighe, S. Goswami, P. Roy, Phys. Rev. D 73 (2006) 071301, hep-ph/0602062.
[36] P.H. Frampton, R.N. Mohapatra, JHEP 0501 (2005) 025, hep-ph/0407139.
[37] S. Antusch, S.F. King, R.N. Mohapatra, Phys. Lett. B 618 (2005) 150, hep-ph/0504007.
[38] T. Ohlsson, G. Seidl, Nucl. Phys. B 643 (2002) 247, hep-ph/0206087.
[39] C.H. Albright, K.S. Babu, S.M. Barr, Phys. Rev. Lett. 81 (1998) 1167, hep-ph/9802314.
[40] Y. Nir, Y. Shadmi, JHEP 9905 (1999) 023, hep-ph/9902293.
[41] Y. Nomura, T. Sugimoto, Phys. Rev. D 61 (2000) 093003, hep-ph/9903334.
[42] C.H. Albright, S.M. Barr, Phys. Rev. D 64 (2001) 073010, hep-ph/0104294.
[43] G. t Hooft, Lecture at the Cargese Summer Institute, 1979.
[44] C.D. Froggatt, H.B. Nielsen, Nucl. Phys. B 147 (1979) 277.
[45] M. Leurer, Y. Nir, N. Seiberg, Nucl. Phys. B 398 (1993) 319, hep-ph/9212278;
M. Leurer, Y. Nir, N. Seiberg, Nucl. Phys. B 420 (1994) 468, hep-ph/9310320.
[46] M.B. Green, J.H. Schwarz, Phys. Lett. B 149 (1984) 117;
M. Dine, N. Seiberg, E. Witten, Nucl. Phys. B 289 (1987) 589;
J.J. Atick, L.J. Dixon, A. Sen, Nucl. Phys. B 292 (1987) 109.
[47] L.E. Ibanez, G.G. Ross, Phys. Lett. B 332 (1994) 100, hep-ph/9403338;
P. Binetruy, P. Ramond, Phys. Lett. B 350 (1995) 49, hep-ph/9412385;
P. Binetruy, S. Lavignac, P. Ramond, Nucl. Phys. B 477 (1996) 353, hep-ph/9601243;
K.S. Babu, T. Enkhbat, I. Gogoladze, Nucl. Phys. B 678 (2004) 233, hep-ph/0308093;
H.K. Dreiner, H. Murayama, M. Thormeier, Nucl. Phys. B 729 (2005) 278, hep-ph/0312012.
[48] T. Enkhbat, G. Seidl, Nucl. Phys. B 730 (2005) 223, hep-ph/0504104.
[49] See, e.g., K.S. Babu, E. Ma, J.W.F. Valle, Phys. Lett. B 552 (2003) 207, hep-ph/0206292;
G. Seidl, hep-ph/0301044;
G. Altarelli, F. Feruglio, Nucl. Phys. B 720 (2005) 64, hep-ph/0504165;
J. Kubo, Phys. Lett. B 622 (2005) 303, hep-ph/0506043;
K.S. Babu, X.G. He, hep-ph/0507217;
C. Hagedorn, M. Lindner, R.N. Mohapatra, JHEP 0606 (2006) 042, hep-ph/0602244;
C. Hagedorn, M. Lindner, F. Plentinger, Phys. Rev. D 74 (2006) 025007, hep-ph/0604265;
I. de Medeiros Varzielas, S.F. King, G.G. Ross, hep-ph/0607045;
Y. Cai, H.B. Yu, hep-ph/0608022;
92
[50]
[51]
[52]
[53]
[54]
[55]
[56]
[57]
[58]
[59]
[60]
Y. Kajiyama, J. Kubo, H. Okada, hep-ph/0610072;

S.F. King, M. Malinsky, hep-ph/0610250;
E. Ma, hep-ph/0612013.
C.H. Albright, M.-C. Chen, hep-ph/0608137.
P.F. Harrison, D.H. Perkins, W.G. Scott, Phys. Lett. B 458 (1999) 79, hep-ph/9904297;
P.F. Harrison, D.H. Perkins, W.G. Scott, Phys. Lett. B 530 (2002) 167, hep-ph/0202074.
F. Plentinger, W. Rodejohann, Phys. Lett. B 625 (2005) 264, hep-ph/0507143.
P. Huber, M. Lindner, M. Rolinec, W. Winter, Phys. Rev. D 73 (2006) 053002, hep-ph/0506237.
P. Huber, W. Winter, Phys. Rev. D 68 (2003) 037301, hep-ph/0301257.
A. Bandyopadhyay, S. Choubey, S. Goswami, S.T. Petcov, Phys. Rev. D 72 (2005) 033013, hep-ph/0410283.
A. de Gouvea, H. Murayama, Phys. Lett. B 573 (2003) 94, hep-ph/0301050.
M. Lindner, A. Merle, W. Rodejohann, Phys. Rev. D 73 (2006) 053005, hep-ph/0512143.
S.T. Petcov, New J. Phys. 6 (2004) 109.
H.V. Klapdor-Kleingrothaus, et al., Eur. Phys. J. A 12 (2001) 147, hep-ph/0103062.
F. Plentinger, G. Seidl, W. Winter, http://theorie.physik.uni-wuerzburg.de/~winter/Resources/Textures/index.html.
Strong-coupling expansion of cusp anomaly

and gluon amplitudes from quantum open strings
in AdS5 S 5
M. Kruczenski a , R. Roiban b, , A. Tirziu c , A.A. Tseytlin d,1
a Department of Physics, Purdue University, W. Lafayette, IN 47907-2036, USA
b Department of Physics, The Pennsylvania State University, University Park, PA 16802, USA
c Department of Physics, The Ohio State University, Columbus, OH 43210, USA
d Blackett Laboratory, Imperial College, London SW7 2AZ, UK
Received 14 August 2007; accepted 10 September 2007

Abstract
An important observable of planar N = 4 SYM theory is the scaling function f () that appears in the
anomalous dimension of large spin twist 2 operators and also in the cusp anomaly of light-like Wilson loop.
The non-trivial relation between the anomalous dimension and the Wilson loop interpretations of f () is
well-understood on the perturbative gauge theory side of the AdS/CFT duality. In the first part of this paper
we present the dual string-theory counterpart of this relation, i.e., the equivalence between the closed-string
and the open-string origins of f (). We argue that the coefficient of the log S term in the energy of the
closed string with large spin S in AdS5 should be equal to the coefficient in the logarithm of expectation
value of the null cusp Wilson loop, to all orders in 1/2 expansion. The reason is that the corresponding
minimal surfaces happen to be related by a conformal transformation (and an analytic continuation). As a
check, we explicitly compute the leading 1-loop string sigma model correction to the cusp Wilson loop,
reproducing the same subleading coefficient in f () as found earlier in the spinning closed string case.
The same function f () appears also in the resummed form of the 4-gluon amplitude as discussed at weak
coupling by Bern, Dixon and Smirnov and recently found at the leading order at strong coupling by Alday
and Maldacena (AM). Here we attempt to extend the latter approach to a subleading order in 1/2 by
computing the IR singular part of the 1-loop string correction to the corresponding T -dual Wilson loop. We
E-mail addresses: markru@purdue.edu (M. Kruczenski), radu@phys.psu.edu (R. Roiban),

tirziu@mps.ohio-state.edu (A. Tirziu), tseytlin@imperial.ac.uk (A.A. Tseytlin).
1 Also at Lebedev Institute, Moscow.
doi:10.1016/j.nuclphysb.2007.09.005
94
M. Kruczenski et al. / Nuclear Physics B 791 (2008) 93124
discuss explicitly the 1-cusp case and comment on apparent problems with the dimensional regularization
proposal of AM when directly applied order by order in strong coupling (string inverse tension) expansion.
PACS: 11.25.Tq; 11.15.Me; 11.25.Mj
Keywords: AdS/CFT duality; Strong coupling expansion; Supersymmetric gauge theory
1. Introduction
As is well known, in perturbative (planar N = 4) gauge theory there are two alternative routes
that lead to the same scaling function f (): it can be found as a coefficient in the anomalous
dimension of gauge-invariant large spin twist two operator [1,2] or as a cusp anomaly of a lightlike Wilson line [36]. One can give a general proof of the equivalence between the two pictures
in perturbative gauge theory [5].
On the dual perturbative AdS5 S 5 string theory side the anomalous dimension of minimal
twist operator is represented by the energy
of a closed string with large spin S 1 in AdS5 [7],
E = S + f () ln S + , f ()1 = + . The same result for the strong-coupling limit of

f () was shown in [8] (see also [9]) to follow from the open string picture, i.e., from the area of
a surface ending on a cusp formed by two light-like Wilson lines on the boundary of AdS5 .
The definitions of f () in [7] and [8] seem very different and a priori unrelated, in contrast
to the known perturbative gauge theory equivalence of the anomalous dimension of large spin
twist 2 operator and the Wilson loop cusp anomaly function. In particular, while it was possible
to compute the two subleading quantum corrections to f () in the closed spinning string picture
[1012]2
a2
f ()1 = a0 + a1 + + ,
a0 =
1
3
, a1 = ln 2,
(1.1)
the direct computation of the quantum string corrections in the Wilson loop approach [8,13,14]
appeared to be harder (for previous attempts in that direction see [1517]).
One of our aims below will be to explain the relation between these two approaches, making
their equivalence manifest (to all orders in the strong-coupling, i.e., 1 expansion). In particular,
we shall demonstrate that computing the quantum open AdS5 S 5 string fluctuations near the
cusp surface of [8] leads to the same 1-loop coefficient a1 in (1.1) as found in the closed-string
picture in [10].
The key observation [10,11] that simplifies dramatically the computation of quantum string
corrections to the closed string energy in the large spin limit is that in order to compute the coefficient of the leading ln S term in E it is enough to consider a scaling limit of the full (elliptic
function) solution of [7]. In this limit the string is stretched homogeneously along the radial direction of AdS5 all the way to the boundary, i.e., = , with 1 ln S 1. One may also
ignore the boundary (turning-point) contributions since they are subleading in the large S limit.
This scaling-limit solution is formally related [11,12] via an analytic continuation to the circular
rotating string with two equal S 5 angular momenta J1 = J2 [18,19] and an imaginary value of
2 The expression for a can be found in [12].
2
95
the winding parameter. This reveals a simple homogeneous nature of this string background:
the only non-trivial fields are isometric angles which are linear in the world-sheet coordinates
(, ), i.e., their derivatives are constant and so are the coefficients in the fluctuation Lagrangian.
This makes the computation of the quantum corrections rather straightforward.
Remarkably, the cusp Wilson loop solution also admits [8] a similar simple scaling limit in
which the Euclidean open-string world surface ends on two light-like lines on the boundary. This
limit (combined with a regularization of the world-sheet area) was enough to reproduce [8] the
leading term a0 in the strong-coupling expansion (1.1) of the cusp anomaly function.
As we shall explain below in Section 2, the two limiting (scaling) solutions are actually
closely related: they become equivalent upon certain analytic continuation (that is needed in particular to convert the Minkowski world sheet coordinates in the closed spinning string case into
the Euclidean one in the open string Wilson loop case)3 and an AdS5 isometry, i.e., a conformal
SO(2, 4) transformation.
That relation implies that all quantum string corrections to the two partition functions which
are computed by expanding the string action near the equivalent classical solutions should also
agree. The logarithm of the string partition function in the conformal gauge determines (after
dividing by the time interval) the quantum string corrections to the closed string energy in the
scaling limit [11,12]. The same open-string partition function on the disc leads (after an appropriate regularization of the world-sheet area [8]) to the cusp anomaly function. Given the general
nature of the relation between the two scaling string solutions (and their homogeneous nature
allowing one to ignore boundary effects) the equivalence of the quantum corrections to the corresponding string partition functions and thus to the f () function should thus extend to all
orders in strong coupling expansion. This provides a proof of the equivalence between the two
definitions of the scaling function also on the dual string-theory side, which is another remarkable
manifestation of the AdS/CFT duality.
We will explicitly check this general statement in Section 2.2 by first showing that the null
cusp solution of [8] has a hidden homogeneous structure: the Lagrangian for string fluctuations
near it has indeed (after an appropriate field redefinition) constant coefficients and produces
the same characteristic spectrum as in the scaling limit of the spinning closed string solution
in [10,11]. This implies that the 1-loop correction to the null cusp Wilson loop leads to the same
a1 coefficient in f () in (1.1) as found in the spinning closed string picture.
The cusp anomalous dimension of the Wilson loop (which in gauge theory is determined
by a UV singularity) governs also [20] the IR asymptotics of the gluon amplitudes in QCD
and in N = 4 SYM theory. The on-shell gluon amplitudes are IR divergent; in dimensional
regularization D = 4 2, < 0 the IR poles in exponentiate and the double pole is determined
in terms of the same f () function [22]. Moreover, f () controls also the finite ln2 st part of
the exponentiated form of the 4-gluon amplitude [23,24] (which is apparently determined by
conformal invariance considerations4 relating it the IR singular part [25]).
The same exponential expression for the 4-gluon amplitude was found also at the leading
order in strong coupling expansion using the AdS/CFT correspondence in [26]. Symbolically,
3 A similar analytic continuation (in spin) was used to relate the anomalous dimension of twist 2 operators to the Wilson
loop cusp anomaly picture on the gauge theory side [5] (which involved also separating the fields in the twist 2 operator
and inserting a Wilson line).
4 This is apparently no longer so for n-point amplitudes [40].
96
assuming some IR cutoff 0, the IR singular part of the 4-gluon amplitude may be written as

2
A4 Adiv (, s)Adiv (, t) ,

2 1
1
2
g() ln
,
Adiv (, s) = exp f () ln2
(1.2)
8
|s| 4
|s|
where s = (k1 + k2 )2 and g() is a non-universal function depending on a choice of the IR
cutoff. As was found in [26] at the classical string order using a special dual version of the
dimensional regularization prescription
f () =
(1.3)
+ O(1),
g() =
(1 ln 2) + O(1).
2
Our aim in the second part of this paper (Section 3) will be to attempt to extend the proposal
of [26] to the quantum (1-loop) string level. One motivation is to explicitly check that it is the
same universal scaling function f () that indeed appears in the exponential form of the 4-gluon
amplitude when computed in strong coupling expansion. Another is to clarify the structure of
the large expansion of the second scaling function g() with a hope of understanding the
interpolation to the 2-loop [23] and 3-loop weak-coupling result for it in [24].
The proposal of [26] was based on starting with a dimensionally extended analog of the
near-horizon D3-brane background to which one should apply T -duality transformation along
the D = 4 2 longitudinal directions.
The use of the 2d duality transformation or, in the target space language, of the T -duality
transformation was one of the key observations of [26] that suggested a relation between the
strong-coupling (semiclassical) computation of the 4-gluon amplitude and the computation of the
null cusp Wilson loop. Thinking of gluons as represented by open strings (attached to D3-branes)
and considering the leading strong-coupling approximation one may expect (by analogy with
high-energy asymptotics of string scattering amplitudes in flat space [27]) that their scattering
amplitude should be dominated by a semiclassical string solution depending on some fixed lightlike 4-momenta kim (i.e., on the s and t kinematic variables in the 4-gluon case). At a qualitative
level, since for open strings in flat space the T -duality exchanges the Neumann and Dirichlet
boundary conditions [28], it should transform a configuration of strings with free ends (thus having specified conserved target-space momenta) to the one with the ends fixed at certain positions.
As discussed in [26], in the T -dual picture the open string world surface should then end on a
closed contour with straight sides determined by the light-like momenta ki .5 The resulting world
surface is then related (before an IR regularization) by an SO(2, 4) transformation [26] to the
cusp Wilson loop surface found in [8] and discussed below in Section 2.
This T -duality transformation appears thus to relate the scattering amplitudes to the momentum space Wilson loops [20,25,29]. The latter were first discussed in a closely related
gaugestring duality context in [30].
In general, given a string sigma model with the metric ds 2 = G(z) dx m dx m + dz2 + the
corresponding classical solution can be related to a classical solutions in the T -dual metric d s 2 =
5 Consider a flat target space and the open-string world sheet as a half-plane ( > 0, < < ). Then a string
solution x m k m ln( 2 + 2 ) that is sourced by an external momentum term (originating from the vertex operator
insertions as in [27]) is a semi-infinite line representing a point-like string coming from infinity with momentum k m . The
formal dual solution (a x m = ab b x m ) is x m k m arctan and near the origin it is a segment of length k m instead
of semi-infinite line. Formally, the 2d duality is an equivalence in the absence of source terms, but the argument should
still go through for sources that are localized, i.e., are of delta-function type.
97
G1 (z) dy m dy m +dz2 + via y m = x m , G(z)a x m = ab b x m .6 The case of the standard AdS5
in Poincar coordinates is special7 since here the original ds 2 = z2 (dx m dx m + dz2 ) and the
dual d s 2 = z2 dy m dy m + z2 dz2 metrics are, in fact, related by the coordinate transformation
(interchanging the boundary and the horizon): z z1 . Here one may say that the 2d duality
acts on the same space of all classical solutions, provided we combine it with the coordinate
transformation z z1 . In particular, both the original scattering solution and its Wilson loop
counterpart may be viewed, at the classical level, as solutions of the same AdS5 sigma model,
albeit with different boundary conditions. That does not mean that solutions related by such
transformation are equivalent (given, in particular, that they satisfy different boundary conditions
in (, ) and thus describe different physical situations), but some of their properties are indeed
closely related.8
In Section 3.1 we shall first ignore the dimensional regularization aspect (imposing a formal
regularization on the world sheet area only at the very end) and assume that the T -duality relation between the semiclassical world-sheet describing gluon scattering and the light-like cusped
Wilson loop suggested in [26] extends beyond the classical level to the full quantum world-sheet
theory as defined by the AdS5 S 5 superstring action [31]. The open string scattering solution in
the original near-horizon D3-brane background, i.e., in AdS5 S 5 supported by the 5-form flux
should thus be related to the light-like cusp Wilson loop surface in the T -dual background, i.e.,
in the near-core region of the smeared D-instanton solution. The latter has the same AdS5 S 5
metric but is supported by a dilaton and a RR 1-form background. As a result, while the bosonic
sigma model part of the associated superstring action is the same, the fermionic part is formally
different. A simple form of the corresponding superstring action was found in [15] by applying
the 2d duality to the Poincar patch x m coordinates in the action of [31] (written in a particular
Killing-spinor -symmetry gauge). The transformation x m x m y m may be viewed as a
quantum change of variables in the string partition function, so the two partition functions (when
properly defined to account for the boundary conditions) should actually agree. Computing the
1-loop partition function in the T -dual geometry we shall indeed find the same result for the a1
coefficient in f ().
In Section 3.2 we shall address the issue of how to extend the dimensional regularization
prescription of [26] to the quantum string level. We will show how one can redo the 1-loop
computation for the 1-cusp Wilson loop using the D = 4 2 prescription of [26]. We will find
that, contrary to what happened at the classical string level [26], keeping finite does not provide
the required regularization of the result.
In Section 3.3 we shall discuss several problems with the quantum string version of the
dimensional regularization proposal of [26], suggesting that it may not apply order by order in
the inverse string tension ( 1 ) expansion.
The Wilson loop directly related to the 4-gluon scattering amplitude is the 4-cusp Wilson loop
which, in the absence of the regularization, i.e., in D = 4, can be found [26] by applying a conformal transformation to the 1-cusp [8] solution. In the regularized D = 4 2 case finding this
solution appears to be complicated (one can no longer use the conformal transformation trick).
6 There is an extra factor of i in front of in the case of the euclidean 2d signature.
ab
7 Here we are ignoring a possible modification of AdS S 5 due to an IR regularization [26].
5
8 Let us mention also that the above T -duality effectively inverting the z-coordinate inverts also the notion of the
UV and IR singularities: the IR divergences of the amplitudes are mapped to UV divergences of the momentum-space
Wilson loops (similar relation was already observed at weak coupling, see [25] and references therein). Note also that in
the dimensional regularization framework of [26] inverting z effectively corresponds to changing sign of = 4D
2 .
98
In Appendix A we describe the construction of the leading order term in this regularized
solution and comment on the structure of the corresponding small fluctuation Lagrangian.
In Appendix B we attempt to resolve the problems discussed in Section 3.3 by suggesting a
modification of the dimensional regularization prescription of [26] that may apply order by order
in string perturbative expansion.
2. Strong coupling expansion of the scaling function: Equivalence of spinning closed
string and null cusp Wilson loop pictures
Below we shall use the following notation for coordinates in AdS5 (we shall often set its radius
to 1). The global coordinates (, t, , 1 , 2 )

ds 2 = d 2 cosh2 dt 2 + sinh2 d 2 + cos2 d12 + sin2 d22
(2.1)
are related to the embedding coordinates XM (M = 0, . . . , 5) on which SO(2, 4) is acting linearly
by
X0 + iX5 = cosh eit ,
X1 + iX2 = sinh cos ei1 ,
X3 + iX4 = sinh sin ei2 ,

ds = dX dXM ,
2
(2.2)
XM X02
X52
+ X12
+ X22
+ X32
+ X42
= 1.
(2.3)
In the Poincar coordinates with the boundary at z = 0 one has

1 m
dx dxm + dz2 ,
x m xm x02 + xi2 ,
2
z
The relation to the embedding coordinates is
ds 2 =
x0
xi
,
Xi = ,
z
z

1
2
m
X5 =
1 + z + x xm .
2z
X0 =
X4 =
i = 1, 2, 3.
(2.4)

1
1 + z2 + x m xm ,
2z
(2.5)
2.1. Scaling limits of the spinning closed string and cusp Wilson loop minimal surfaces and
their equivalence
Let us start with recalling the scaling limit [10,11] of the spinning closed string solution
of [7]. Here we shall use the conformal gauge with Minkowski 2d signature, ds 2 = d 2 + d 2 .
In the limit of large spin the spinning closed string [7] written in global coordinates (2.1) can be
approximated by
t = ,
= ,
1 = ,
= 2 = 0,
1
S
ln 1.
(2.6)
Here 1 is the rotation angle and the string is folded and stretched along with = as
turning points. Rescaling the world-sheet coordinates by one effectively decompactifies , i.e., one may consider the world sheet as a plane. The world-sheet area then scales as
2 and the classical spacetime energy scales as , i.e., as ln S. The same is true to all orders
in quantum
1 expansion since the solution happens to be homogeneous: the fluctuation
Lagrangian has constant coefficients. Thus the quantum fluctuation problem is translationally
99
invariant and the quantum effective action or ln Z is proportional to the area 2 to all orders in
1 [12].
Writing (2.6) in the embedding coordinates we get (after rescaling , by and thus assuming that they take values in the infinite interval in the limit )
X0 = cosh cos ,
X5 = cosh sin ,
X1 = sinh cos ,
X2 = sinh sin ,
X3 = X4 = 0,
(2.7)
X 0 X2 = X 5 X1 .
(2.8)
The solution (2.7) thus belongs to the same class of homogeneous string solutions as the rigid
circular string found in [18,19]. In fact, it is formally equivalent, upon an analytic continuation
and re-interpretation of the parameters, to the background representing the circular rotating string
in S 5 with two equal angular momenta [11].
Let us now find a counterpart of this solution with euclidean 2d world sheet. If we set
i then X5 and X2 in (2.7) become imaginary, effectively exchanging places in (2.3) while still
preserving the ( + + ++) signature of the 6d metric of the embedding space. Then we can
get a real AdS5 string solution with euclidean world sheet by simply renaming the coordinates:
X0
= X0 ,
X1
= X1 ,
X2
= iX5 ,
X5
= iX2 ,
X3
= X3 ,
X4
= X4 ,
(2.9)
where X0
2 X5
2 + X1
2 + X2
2 + X3
2 + X4
2 = 1, i.e.,
X0
= cosh cosh ,
X5
= sinh sinh ,
X1
= sinh cosh ,
X2
= cosh sinh ,
X0
X5
X3
= X4
= 0,
(2.10)
= X1
X2
.
(2.11)
As we shall see below, this euclidean world sheet counterpart of the scaling limit of the spinning
closed string solution is directly related to the null cusp Wilson line solution of [8]. Note that
written in the Poincar coordinates (2.5) the solution (2.10) becomes
z
= (sinh sinh )1 ,
x1
= coth ,
x0
= coth coth ,
x2
= coth ,
x3 = 0.
(2.12)
Next, let us review the cusp solution found in [8] in the Poincar coordinates (2.4), i.e.,

1 2
(2.13)
dz du2 + u2 d 2 + dx22 + dx32 , x0 = u cosh , x1 = u sinh .
2
z
The limiting minimal surface ending on two light-like lines at the boundary z = 0 found in [8] in
the static gauge (i.e., with (u, ) as world-sheet directions) is

z = 2u = 2 x02 x12 ,
(2.14)
x2 = x3 = 0.
ds 2 =
The corresponding induced 2d metric has euclidean signature: ds 2 = 12 ( du

+ d 2 ). To regularu2
ize the area of this surface one may assume that < u < L, 2 < < 2 where L, ,
0 [8]. Then the coefficient in the area representing the strong-coupling limit of the cusp
anomaly reproduces [8] the same value of the leading strong-coupling coefficient a0 in the scaling function (2.6) (see also [9]).
100
The same solution in the conformal gauge (ds 2 = d 2 + d 2 ) is
z = 2u,
u = e 2 ,
= 2.
(2.15)
Looking for a more general solution in the conformal gauge that satisfies the limiting z = 2u
condition one finds that it is the same as (2.15) up to an SO(2) rotation in the (, ) plane9 :
u = e ,
= + ,
2 + 2 = 2,
z = 2u,
(2.16)
x0 = e cosh( + ),
x1 = e sinh( + ),
x2 = x3 = 0. (2.17)
The two natural simple choices are = 2, = 0 in (2.15) and = = 1 that we will use
below.
Using (2.5) the conformal-gauge solution can be written in embedding coordinates as
1
X0 = cosh( + ),
2
1
X1 = sinh( + ),
2
1
X5 = cosh( ),
2
1
X4 = sinh( ),
2
X2 = X3 = 0.
(2.18)
X3 = X4 = 0
(2.19)
We can get an equivalent form of this solution

1
X0 = cosh( + ),
2
1
X1 = sinh( + ),
2
1
X5 = cosh( ),
2
1
X2 = sinh( ),
2
by performing a discrete SO(2, 4) transformation that interchanges X2 and X4 ,

X2 X 4 ,
X4 X2 .
(2.20)
Note that for the solution (2.19) [8]

1
X02 X12 = X52 X22 = .
2
(2.21)
The transformation (2.20) effectively produces a non-zero value of x2 : if we project (2.19)

back to the Poincar patch using (2.5) we get instead of (2.17)
x0 = e cosh( + ),
z = 2e ,
x1 = e sinh( + ),
x2 = e sinh( ),
x3 = 0.
(2.22)
It is easy to check that (2.19) (or (2.18)) satisfies the string equations and the conformal gauge
constraints (in euclidean 2d metric) written in the embedding coordinates:
a a XM XM = 0,
X M XM = 1,
X M XM X M XM = 0,
= a X M a XM = 2,
X M XM = 0.
(2.23)
(2.24)
9 One may also consider a general 2d conformal transformation on (, ); the resulting value of the classical string
action (area) will be formally the same before one introduces a regularization.
101
This is thus a special case of the constant Lagrange multiplier ( = const) solutions for which
XM satisfies a linear constant-mass 2d equation discussed in [19]. Such string solutions include
rigid circular rotating strings and are effectively homogeneous.10
Applying another discrete SO(2, 4) transformation one can show explicitly that (2.19) is a
homogeneous solution, i.e., it can be put into the form where only the isometric angles of the
AdS5 metric (i.e., the analogs of t, 1 , 2 in the parametrization (2.1)) are non-zero and linear in
(, ). As a result, the fluctuation Lagrangian will have constant coefficients (after an appropriate
choice of the basis of fluctuation fields). To see this explicitly let us group XM as (X02 X12 ) +
(X52 X22 ) (X32 + X42 ) = 1 and introduce new global AdS5 coordinates as
X0 X1 = r1 ep ,
where ri satisfy
r12
+ r22
X5 X2 = r2 eq ,
r32
r1 = cosh r cos f,
X3 iX4 = r3 eih ,
(2.25)
= 1, i.e.,
r2 = cosh r sin f,
r3 = sinh r.
(2.26)
The AdS5 metric written in terms of the independent coordinates p, q, , r, f becomes:

ds 2 = cosh2 r df 2 + dr 2 + cosh2 r cos2 f dp 2 + sin2 f dq 2 + sinh2 r dh2 .
(2.27)
Our solution is then

r = 0,
f=
,
4
p = + ,
q = ,
h = 0,
(2.28)
i.e., is homogeneous.11
Finally, let us now demonstrate that the solution (2.19) is, in fact, SO(2, 4)-equivalent to the
euclidean world sheet version of the scaling limit of the spinning closed string solution in (2.10).
Let us choose = = 1 in (2.19) and apply two discrete SO(2) rotations in the (X0 , X5 ), and
(X1 , X2 ) planes (which of course preserve the ( + + ++) metric)
1
1
X0
= (X0 + X5 ),
X5
= (X0 X5 ),
2
2
1
1
X1
= (X1 + X2 ),
(2.29)
X2
= (X1 X2 ).
2
2
The resulting background is then exactly the same as in (2.10), (2.11).
As was implicit in [8] and recently discussed in detail in [26] the Poincar-patch solution (2.15) interpreted in the global coordinates (2.19), (2.21) describes a surface ending on
a closed line with four and not just one null cusp. The presence of the four cusps was made
clear in [26] by applying the transformation (2.20) and the same SO(2, 4) transformation as
in (2.29). The resulting Poincar patch solution is similar to the one in (2.12) (with X0 X5 ,
i.e., sinh cosh, coth tanh) and it ends on a rectangular contour with 4 null cusps.12 Taking
10 Note that (2.18) cannot describe a regular closed string since it is not periodic in (unless one considers a scaling
limit as discussed above in which is rescaled by a large parameter) but it may be interpreted as an open-string solution
with an infinite range of .
11 The metric (2.27) is of course related to S 5 metric by an analytic continuation so there is formally a similar homogeneous S 5 solution.
12 The corresponding open-string solution ending on a rectangular contour at the boundary z = 0 was interpreted in [26]
as being 2d-dual (T -dual) to the semiclassical solution describing the massless open string (gluon) 4-point scattering
amplitude with the Mandelstam variables s = t .
102
into account
the euclidean continuation and the subsequent rotation in the (, ) plane (going
from = 2, = 0 case in (2.15) to the = = 1 in (2.19), (2.29)) one may relate the origin
of the 4 cusps to the presence of the 4 special points in the spinning closed (and folded) string
surface: the two ends where the -derivatives vanish and the center where the -derivatives vanish.13
To conclude, we have seen that the scaling (large spin) limit [10,11] of the spinning closed
string solution of [7] is formally equivalent, upon an analytic continuation to the euclidean world
sheet combined with a discrete SO(2, 4) rotation in AdS5 , to the global AdS5 version of the null
cusp solution of [8]. This explains the agreement between the leading strong-coupling expressions for the scaling (cusp anomaly) function found respectively in [7] and in [8].
This equivalence, combined with the homogeneous nature of the string background, implies
that the correspondence between the two picturesthe spinning closed string energy (or the
minimal twist anomalous dimension) and the cusp Wilson loopextends also to the quantum
string level. The homogeneity of the scaling-limit string solution implying the (, ) translational invariance of the quantum fluctuation Lagrangian means that one can essentially ignore
the difference in the boundary conditions in the closed and open string cases. The contributions
of the quantum string fluctuations to the energy of the closed string or to the open string partition
function will be equivalent since in this scaling limit they are the same at each interior point of
the string world sheet, i.e., they are proportional to the (regularized) area of the world sheet.
To check this argument, below in Section 2.2 we shall explicitly find the spectrum of quadratic
fluctuations near the cusp open string surface (2.19) and verify that they indeed lead to the same
1-loop cusp anomaly coefficient a1 in (1.1) as found [10] in the spinning closed string case.
Incidentally, that apparently will be the first explicit computation of the 1-loop string correction
to a (non-BPS) Wilson loop surface done so far in the literature.
Before turning to the discussion of the quantum fluctuations let us mention several possible
generalizations of the above discussion. One immediate extension is to consider a more general
cusp Wilson loop to include an angular momentum J in S 5 , in direct analogy with what was done
in [10,11] for the rotating string (large finite J corresponds to operators with large finite twist).14
It is of interest also to consider the generalization of the spinning closed string to spinning string
with n > 2 spikes [33] to find its scaling (large spin) limit and to try to identify then a related
euclidean open string world surface.15
13 One may also think of the closed string as a combination of two coinciding open strings with the ends at the two
folds; this also suggests the presence of 4 cusps in the joint (euclidean) world sheet surface.
14 The scaling limit [11] of the (S, J ) folded closed string solution of [10] is the following generalization of (2.6):

t = , = m , 1 = , = where m = 2 2 , and is from S 5 and J = . As one can verify, the

corresponding generalization of the cusp Wilson line solution in the conformal gauge (2.15) corresponding
to the metric
(2.13) with an extra d 2 term and having the euclidean-signature world-sheet metric is: z = 2u, u = e +m , =
+ m, =
, where 2 = m2
2 , and
= i (in going from Minkowski world sheet solution for a closed string
to euclidean world sheet solution for the open string in Section 2.1 we rotated i ). The induced metric is then
ds 2 = m2 (d 2 + d 2 ) so one may set m = 1 to put it into the canonical form as we did in (2.15). Other Wilson loop
solutions with rotation in S 5 were considered in [32].
15 Below we shall also discuss a generalization of the above cusp solution to the dimensionally regularized case, following the suggestion of [26]. It would be interesting to find explicitly the corresponding form of the spinning closed
string solution (dimensional continuation to D = 4 2 breaks SO(2, 4) symmetry so this is non-trivial). In contrast to
the area of the cusp the energy of the closed string should be regular in the 0 limit.
103
2.2. 1-loop string correction to null cusp Wilson loop expectation value
Let us now go back to the conformal-gauge solution (2.15) or (2.19) with = 2, = 0 and
compute the spectrum of small fluctuations near it, demonstrating explicitly that it is the same
as in the case of the scaling limit of the spinning closed string in [10,11]. That (together with
homogeneity of the solution) will imply that the 1-loop correction to the cusp anomaly is the
same as the 1-loop correction to the coefficient of ln S in the fast-spinning string energy.
Let us start with bosonic fluctuations. It is easiest to consider the string action in terms
of the embedding coordinates, though one can get the same fluctuation spectrum using the
Poincar coordinates (that will require a non-trivial choice of the quantum fluctuation fieldssee
Appendix A). The euclidean world-sheet Lagrangian in the conformal gauge is

1
1
L = a X M a XM + X M XM + 1 ,
2
2
with the solution (2.19) satisfying (2.23). Introducing the fluctuations
XM XM + X M ,
+ ,
(2.30)
(2.31)
we obtain the quadratic part of the fluctuation Lagrangian with X M subject to a linear constraint
( = 2)
1
L 2 = a X M a X M + X M X M , X M X M = 0.
2
The explicit form of the latter constraint is
X 0 cosh 2 X 1 sinh 2 + X 5 cosh 2 X 2 sinh 2 = 0.
(2.33)
Performing the field redefinition (X 0 , X 1 ) (Z0 , Z1 )
Z1 = X 0 sinh 2 + X 1 cosh 2
Z0 = X 0 cosh 2 X 1 sinh 2,
(2.34)
(2.32)
and similarly (X 5 , X 2 ) (Z5 , Z2 ), the constraint (2.33) takes the form

Z0 + Z5 = 0.
(2.35)
This allows us to eliminate Z5 from the fluctuation Lagrangian which then becomes

1
L 2 = a Z0 a Z0 + a Z1 a Z1 + a Z2 a Z2 2 2(Z1 Z0 Z2 Z0 )
2

1 a
+ X3 a X 3 + a X 4 a X 4 + 2X 32 + 2X 42 .
(2.36)
2
This fluctuation Lagrangian is essentially equivalent to the euclidean version of the one found
in [10,11] for the scaling limit of the spinning closed string solution. Diagonalizing it we get
two massless modes (whose contribution is compensated by that of
the two massless conformal
gauge ghosts), one mode with mass 2 and two modes with mass 2. In addition, there are five
massless modes from the fluctuations in S 5 directions.
The quadratic fermionic action [31] is given by16

ab I J
i JK
ab I J I
JK
LF 2 = i s /ea Db /eb K ,
(2.37)
2
16 For details and notation see, e.g., [12]. In particular, here M = 0, 1, . . . , 9, sI J = (1, 1), = i
01234 , etc.
104

A
/ea = eM
a x M A ,
1
Da = a + M AB a x M AB .
4
(2.38)
It simplifies in the 1 = 2 -symmetry gauge used also in [10,12]. It is most straightforward

to find the fermionic spectrum (after a continuation to euclidean world sheet) in the coordinates (2.27) in which the solution takes the explicitly homogeneous form (2.28) and thus all the
coefficients in (2.37) are constant. As in [10], we find eight fermionic modes with mass 1.
As a result, the logarithm of the 1-loop euclidean partition function is (here we assume that
the rescaled and coordinates change in the infinite limits)

d 2p
1 = ln Z1 = V2
(2.39)
Z1 p 2 ,
2
(2)
1

Z1 p 2 = ln p 2 + 4 + 2 ln p 2 + 2 + 5 ln p 2 8 ln p 2 + 1 .
(2.40)
2
Here
the volume V2 factorizes since our background is translationally invariant. Note that V2 =
d d = 2A2 , where A2 corresponds to an area of a single cusp (i.e., V2 is twice the area of
a half-plane world sheet as appropriate for the open string case).
Computing the integral one finds as in [1012]
V2
1 =
4
dv Z1 (v) = a1 A2 ,
a1 =
3 ln 2
.
(2.41)
For comparison, the value of the classical action is ( =
R2
1 a M
d d X a XM =
V2 = a0 A2 ,
I=
2
2
2
a0 =
1
.
(2.42)
The area of the cusp open-string world sheet regularized as in [8] is A2 = 4 ln L , where
is a boost parameter for the two lines forming a cusp and L and are the IR and the UV cutoffs.17
3. Strong-coupling corrections to IR singular part of gluon scattering amplitude: Dual
Wilson loop expectation value
In Section 2.2 we have computed the 1-loop string correction to the 1-cusp Wilson loop and
found that it contains the same 1-loop corrected f () function as appearing in the energy of
the spinning closed string (and we argued that this agreement should hold also to all orders in
strong coupling expansion). We used the standard AdS5 S 5 superstring action with the 5-form
coupling in the fermionic part.
Below we shall repeat this computation starting with a T -dual superstring action which appears in the context of the proposed relation between gluon scattering amplitudes and Wilson
loops in [26]. It was suggested in [26] that using T -duality one can relate the expression for the
4-gluon scattering amplitude to the expectation value of a certain 4-cusp Wilson loop. It was
argued that the world surface relevant for the 4-gluon high energy scattering can be found via
L
L 2
17 Imposing a cutoff at z = 0 in (2.13) implies that
max ln where L is a cutoff on u. Then A2 (ln ) (see
also [9]).
105
2d duality by starting with the null cusp solution of [8], lifting it to global AdS5 coordinates,
applying a certain SO(2, 4) conformal transformation (a discrete one plus a conformal boost
depending on s/t), and then reinterpreting the result back in the Poincar coordinates.
If we first formally ignore the issue of IR regularization,18 we can argue that since the superstring partition function is supposed to be invariant under the global symmetry of AdS5 S 5
and, moreover, since the quantum fluctuations should not distinguish between the global and the
Poincar coordinates (a change of coordinates in AdS5 S 5 is a local field redefinition of the
quantum 2d fields) it should not matter which form of the 1-cusp Wilson loop solution we use
for the 1-loop computation.
Then assuming an a posteriori cutoff on the area of world surface we shall find in Section 3.1
the same expression for the coefficient of the IR singular part of the partition function, confirming
at strong coupling that f () that appears in (1.2) is the same universal cusp anomaly as found in
the spinning closed string and the 1-cusp Wilson loop cases.
The second part of the proposal of [26] states that the counterpart of the IR dimensional regularization on the gauge theory side should be to consider the dual string theory defined in a
dimensionally regularized version of the AdS5 S 5 background with the analog of the boundary having D = 4 2 dimensions. This proposal worked fine at the level of the classical string
theory, i.e., the leading order in strong coupling expansion, reproducing [26] the same structure of the 4-gluon amplitude as conjectured [24] from the resummation of the weak coupling
perturbation theory.
In Section 3.2 we shall explore how this dual version of dimensional regularization works
at the quantum string level. This built-in IR cutoff breaks SO(2, 4) invariance and thus distinguishes between the 1-cusp and 4-cusp solutions, with the latter being relevant for the physical
scattering amplitude case. Since it appears to be hard to find the explicit dimensionally regularized version of the 4-cusp solution19 we shall limit ourselves to the case of the D = 4 2
regularized version of the 1-cusp solution (which happens to be a simple generalization [26] of
the solution of [8]).
Our motivation will be based on the expectation that the 1-cusp contribution should be closely
related to the leading singularity of the IR divergent part (the Sudakov form factor) of the scattering amplitude, but we should admit that there is no obvious argument supporting that beyond
the classical string level.20
We shall then redo the same 1-loop computation as in Sections 2.2 and 3.1 for the null 1-cusp
solution now in dimensionally regularized T -dual geometry. We shall find that, contrary to
18 By IR regularization we mean the one that regularizes the massless gluon amplitudes. In the T -dual Wilson loop
picture it appears as a UV regularization at small values of the coordinate z. In addition, the area of the 1-cusp solution
has also an IR divergence at large values of coordinates yi . This IR divergence is automatically absent in the 4-cusp
case, being effectively cut off by the lengths of the sides of the contour related to momentum invariants s and t [26].
19 One could attempt to use perturbation theory in (see Appendix A), and even the leading-order form of the solution
was apparently enough at the classical level [26]. However, it is not clear why the same should apply at the 1-loop
level. There is also a technical problem that expanding the exact-in- action near a perturbative in solution will lead to
spurious (off-shell) 2d divergences and -symmetry gauge dependence.
20 At the tree level it was argued in [26] that the single-cusp contributions determine the 1 part of the area of the 4-cusp
2
surface. Moreover,it was noticed in [38] that with a special choice of the IR cutoff on the length of the two sides of the
cusp (equal to 2 s where s is the Mandelstam variable) which complements the UV dimensional regularization at
small z one gets exactly the same answer for the two leading singular terms 12 and 1 in the 1-cusp area as found in the

area for the full 4-cusp surface in [26]. It is not clear, however, why this observation should extend to the quantum string
case.
106
what happened at the string tree level, the dimensional regularization prescription no longer
appears to provide a natural cutoff of the result at small z. Moreover, even if we formally assume
that the 1-loop area factor should be again regularized in exactly the same way as the tree-level
one, we will still not be able to relate the 1-loop coefficient of the 12 pole to the 1-loop coefficient
in the scaling function (1.1). More generally, there appears to be a problem with the usual RGtype relation between the coefficient of the 12 pole and the cusp anomaly when applied order by
order in the inverse string tension expansion.
We shall comment on these problems in Section 3.3. As we shall discuss, it is not a priori clear
why considering string theory in the dimensionally regularized background should provide the
required analog of the IR regularization of gauge theory, with the logarithm of the Wilson loop
expectation value scaling as q 21 + q2 + q3 at each order in the
1 expansion.21 We shall
conclude that the dimensional regularization prescription as formulated in [26] does not appear
to work (at least in the most naive way) order by order in the strong coupling expansion. One
option is that its application may require a resummation of 1 expansion; alternatively, it may
require some modification, for example, a modified relation between the parameters and on
the gauge and the string sides. This important issue needs further clarification.
3.1. 1-loop correction in T -dual picture: Fermionic action
Starting with the AdS5 S 5 superstring action [31] written in a particular -symmetry gauge
and applying 2d duality along 4 isometric directions of AdS5 in Poincar coordinates it was found
in [15] that the resulting action takes a remarkably simple form (m = 0, 1, 2, 3; s = 1, . . . , 6, z2 =
zs zs )22

1 a m
2
d
y a y m + a zs a zs
S=
2
2z2

m
ab
s
+ 2i a y m + a z s b .
(3.1)
Here is a MajoranaWeyl 10d spinor related to the two original fermionic coordinates by a
certain y-dependent rotation and a -symmetry gauge choice involving ( I J 0123 I J ). Note
that the action (3.1) is exactly quadratic in fermions.23 This action was interpreted in [15] as
describing a fundamental superstring propagating in the background which is T -dual to the nearcore D3-brane one, i.e., in the near-core smeared D-instanton [36] background.
Below we shall first rederive the fermionic part of (3.1) by starting with the general form of the
quadratic part of the GS action in a type IIB superstring background and specifying it to the case
of the smeared D-instanton background. We shall then expand that action near the homogeneous
cusp solution (2.19), (2.28) and show that the resulting fermionic spectrum is the same as found
above in Section 2.2 from the quadratic part of the original AdS5 S 5 action (2.37). This is of
21 At the same time, one can give a general argument that the logarithm of the relevant IR singular factor in the perturbative gauge theory amplitude in D = 4 2 should have this particular structure order by order in [22,24].
22 Here we assume conformal gauge and thus ignore the dilaton coupling originating from the 2d duality transformation.
We also use Minkowski signature on the world sheet.
23 The starting point in [15] was the action in a special -symmetry gauge choice found in [34]. An equivalent action
also becoming quadratic in fermions (and having the same structure as (3.1)) after the T -duality along the 4 Poincar
patch coordinates was found in another -symmetry S-gauge in Appendix C of [35].
107
course expected on the basis that the two actions are related by the 2d duality transformation.
The resulting 1-loop partition function is then the same as in (2.40) and thus it leads to the same
1-loop term in the coefficient function f ().24
To quadratic order in fermions the form of the GS superstring action is determined by the generalized covariant derivative that enters the variation of the target space gravitino field. The fact
that we consider a superstring propagating in a background which is a type II supergravity solution guarantees the -symmetry of the classical string action [37]. We shall use the normalization
in which the bosonic part of the string-frame type IIB supergravity Lagrangian is

1 2
1 2
1
1 2
2
2
LIIB = e
(3.2)
FMNK
,
F2
R + 4() HMNK FM
12
2
12
4 5! MNKLP
and the gravitino supersymmetry transformation rule is (I, J = 1, 2)
I
= DIMJ J ,
M

1
1
DM = M + M AB AB s3 HABM AB
4
8

1
1
+ e /
F(1) s0 + /
F(3) s1 + /
F(5) s0 M ,
8
2
with the 2 2 matrices s3 = 3 , s1 = 1 , s0 = i2 and /

F(n) =
quadratic fermionic action is a generalization of (2.37)

LF 2 = i ab I J ab sI J I /ea DIbJ J .
(3.3)
1
A1 ...An .
n! FA1 ...An
Then the
(3.4)
For 1/2 supersymmetric supergravity backgrounds one has D2 = 0 (i.e., [DM , DN ] = 0) which
is the condition of integrability of the Killing spinor equation; shifting by a Killing spinor is
then a global fermionic symmetry of the GS action.
While in this subsection we are ultimately interested in the case when we start with a D3-brane
solution with D = 4 longitudinal directions, in the next section we will consider, following [26],
the case when D = 4 2, 0, so let us keep D arbitrary for generality. The solution T -dual
to a Dp-brane solution (assuming T -duality is formally applied in all D = p + 1 longitudinal
directions) is the D-instanton [36] smeared in D directions

2
2
2
ds10
,
= H 1/2 (z) dyD
+ dz10D
e = H,
F(1) = dC,
2
10D
H =0H =
C = iH 1 ,
cD R 4 4D
,
z8D
e F(1) = i d ln H,
R 4 =
2 .
(3.5)
Here the harmonic function H was taken in the near-core limit and is an analog of the gauge
theory renormalization scale in dimensional regularization.25 Note that the original RR scalar C
is imaginary for the D-instanton solution.26
24 As a by-product, we will thus provide a direct check of the 1-loop finiteness of the action (3.1).
25 In the notation of [26], c = 24 3 (2 + ),
= and
= 1.
D
4 e
26 The dual to F
(1) form F(9) is real in euclidean signature case. Let us note also that one may check that the dilatino
+ 12 e F(1) s0 + other fields) vanishes if 1 = 2 and that the same relation annihilates the
variation ( = 12 /
gravitino variation.
108
The corresponding -invariant quadratic fermion Lagrangian (3.4) then takes the form
(cf. (2.37)):

1
1
LF 2 = i ab I J ab sI J I /ea b + b AB AB J K i J K e /
F(1) /eb K . (3.6)
4
8
Using that the only non-zero components of the connection for the metric in (3.5) are
A
M Ai = M iA = M
i ln H 1/4 ,
i = 9 D, . . . , 9,
(3.7)
it is easy to show that

1
b + b ln H 1/4 J K
2

i J K K ,

LF 2 = i ab I J ab sI J H 1/4 a x A I A

1
+ b x B i ln H 1/4 i B J K
2
(3.8)
where i = 9 D, . . . , 9, x A = (y m , zs ), and A, B = 0, . . . , 9 are flat indicies.

We can eliminate the first Abelian connection term (a + 12 a ln H 1/4 ) by redefining
I = H 1/8 I ,
(3.9)
thus getting

LF 2 = i
ab I J
s
ab I J
1/4

a x A b J K
A I

1
+ b x B i ln H 1/4 i B J K i J K K .
2
(3.10)
Next, we note that the only difference between this action and the one for the AdS5 S 5 supported
by the 5-form flux is that the composite connection term b x B i ln H 1/4 i B ( J K i J K ) is
1
larger by a factor of 8D
4 (i.e., by 1 + 2 if D = 4 2). In the absence of regularization, i.e.,
for = 0 this connection term can be eliminated as in [15] by a rotation of fermions by the same
matrix that appears in the solution of the Killing spinor equation. The same transformation can
be made also for generic D 27
8D I J J
I = 4
.
(3.11)
Consequently, the action becomes simple when written in terms of I .

We are still to fix the -symmetry gauge. Ref. [15] used in the original D3-brane context the
gauge ( I J 0123 I J ) I = 0 which amounts to setting 1 = 0123 2 . The equivalent action is
2 = 1)
found by choosing, as appropriate in the D-instanton context (11
1 = 11 2 ,
i.e., 1 = 2 ,
(3.12)
27 The spinor matrix here is a function of S 5 coordinates which appears in the expression for the Killing spinors
on S 5 .
109
where we used that the type IIB fermions are MajoranaWeyl with the same chirality.28 This
finally leads to the same fermionic Lagrangian as in (3.1)
A b .
LF 2 = 2i ab a x A
(3.13)
The 2d duality derivation of (3.1) in [15] implies that the superstring action does not actually
contain any higher-order fermionic terms.
Finally, we are ready to return to the problem of computing the 1-loop correction in T -dual
string theory expanded near the null cusp solution (2.15) (or (2.19) or (2.28)). The bosonic part of
the action is the same AdS5 S 5 , so the fluctuation spectrum is the same as in Section 2.2. To find
the corresponding fermionic fluctuation spectrum it is easiest to go back to the unrotated form
of the Lagrangian (3.10) and fix the 1 = 2 gauge. We should also take into account that we are
now interested in the euclidean world-sheet metric, i.e., we are to replace ab ab , ab iab .
The resulting fermionic action has then constant coefficients29 and one finds 8 fermionic massive
modes with mass 1, i.e., exactly the same spectrum as in (2.40). Using a direct IR regularization
of the world-sheet area as in [8] or in Section 2.2 (with ln2 ln L ) we then reproduce the
first universal term in the amplitude (1.2) with the same function f () as in the cusp anomaly.
3.2. 1-loop correction in T -dual picture with dimensional regularization
Let us now try to generalize the discussion at the classical string level in [26] and consider the
dimensionally regularized version of the 1-loop computation of the T -dual 1-cusp Wilson loop
expectation value described in the previous subsection.
In the original high-energy scattering set-up we should replace the AdS5 S 5 space by the
near-core D-brane solution with the total number of longitudinal directions being D = 4 2,
i.e., having the metric

2
2
2
,
= H 1/2 (z) dxD
+ H 1/2 (z) dz2 + z2 d9D
ds10
(3.14)
which should be supported by the corresponding dilaton and RR field strength. To consider the
number of longitudinal dimensions D to be continuous may appear a bit odd since the 1/2 supersymmetric type IIB Dp-brane solutions exist only for even integer D. This problem goes away
after we T -dualize along the D directions x m y m in order to switch to the momentum-space
Wilson-loop description [26]. This leads to the smearedD-instanton background already given
in (3.5). Ignoring the overall string tension factor T =

H in (3.5)30 we then get the following dual metric
2

dy42 + dz2
2
210 = z
+
d
ds
5+2 .
z2
cD 2
2
coming from the numerator of
(3.15)
28 There is a subtlety in the discussion of the GS action in D-instanton background in that we should have considered the
euclidean target space and thus formally complexify the fermions (cf. [36]). In particular, the rotation needed to eliminate
the second line in (3.10) is complex and the natural gauge choice appears to be 1 = i2 .
29 Starting with the form of the action in (3.13) to get the simple fermionic fluctuation Lagrangian one would need to
undo the rotation (3.11), i.e., to apply a transformation U = cosh + sinh 0 1 with being linear in world-sheet
coordinates.
30 Note that dimensions are balanced in the string action provided is dimensionless and y and z have the same
dimension as , i.e., the mass dimension (this was the implicit assumption in [26]). This dimension assignment can be
reversed by rescaling by a power of
.
110
In the strict 0 limit we get back to the AdS5 S 5 metric where the AdS part (cf. (2.4)) and the
internal sphere part factorize. This factorization will still be true at the classical level (assuming,
as we will, that one is interested in solutions localized in S 5+2 ), but it will no longer hold at the
level of the quantum fluctuation.
Let us start with generalizing the light-like cusp solution (2.15) from the case of the AdS5
metric (2.13) to the case of the regularized metric (3.15). Assuming that the only non-zero
coordinates are again z, u, (in the notation of (2.13)), choosing the static gauge with (u, ) as
the world-sheet directions and makingthe -homogeneous
ansatz for z, i.e., z = F (u), we find
that the string action is proportional to d du uF 2 F
2 1. This leads to the solution [26]
z = F (u) = 2 + u,
(3.16)
which generalizes the solution (2.14) of [8]. The induced metric is then
2

1+
1 d 2
du
2
+
.
ds2 =
1
2+
1 + u
(2 + )1+ 2 u
(3.17)
Changing the coordinates, it can then be put into the standard conformal-gauge form, i.e., conformal to d 2 + d 2 = d d . In general, one can show that given any function
h = h( ),
= + i,
(3.18)
the background
h h
= 1+
(3.19)
2i
is a conformal-gauge solution, i.e., it solves the string equations and the euclidean version of the
conformal constraints. The simplest choice which
is a direct generalization of the = 0 solution
(2.15) is found if h is linear in , i.e., h = 2 :
= 2(1 + ).
z = 2 + u, u = e 2 ,
(3.20)
z=
2 + u,
u = e 2 (h+h) ,
Then the induced metric is conformally-flat (becoming flat for = 0):31

2

d 2 + d 2 ,
ds22 = N e
N =
2 2 (1 + )
1
(1 + 12 )1+ 2
(3.21)
The value of the classical (euclidean) action is regularized by < 0 at but is still
divergent in other limits, i.e., it depends on a choice of a region in , space. Alternatively,
the action can becomputed in the static gauge as in Eq. (3.29) in [26], i.e., we start with I =

T N d d e 2 and rewrite it in terms of
y = ue = e
2[ 1+ ]
(3.22)
ending up with
u
I = T K
dy+ dy
1
2(2y+ y )1+ 2
T =
1+
cD 2
,
, K =
1
2
(1 + 12 )1+ 2
31 The choice of h that corresponds to flat induced metric is h( ) = 2 ln[1 + (2+)(2+)/4 ].

2 (1+)1/2
(3.23)
111
where one needs to assume a cutoff y u at large y which amounts to specifying the lengths
32
of the null lines forming
the boundary of the cusp surface. As was noticed in [38], choosing
this cutoff as u = 2 s one gets (after formally combining the 4 1-cusp contributions) the
same answer for the two leading singular terms as found in the area for the full 4-cusp surface
in [26].33 Then we get from (3.23)

1 2
2
1
I= 2
+
+ O 0 .
(1
ln
2)

4
|s|
2 |s|
(3.24)
Our aim now will be to compute the 1-loop correction to the -regularized 1-cusp Wilson loop
in the conformal gauge by expanding near (3.20). To study the bosonic fluctuations it is useful
first to change the coordinates in (3.15) from (z, u, y) to (, , y
) as follows:
z=
2 + e ,
yk = e yk
.
u = e+ ,
(3.25)
Then the 10d metric (3.15) takes the form

1
2
= e (2 + )1 2 e2 (d + d)2 + e2 d 2 + (2 + ) d 2
ds10

1
2
,
+ (dy
+ d y
)222 + (2 + ) 2 d5+2
(3.26)
where yk
stand for the rest of the longitudinal coordinates of the metric. Since the 1-cusp
solution is localized in these coordinates and also in the sphere coordinates, the corresponding
parts of the quadratic fluctuation Lagrangian will contain two sets of 2 2 (yk
) and 5 + 2 (Ys )
decoupled massive modes.
Note that since for = 0 the coordinates and are isometric angles in this parametrization,
this explains why the 1-cusp solution was homogeneous. Indeed, the solution (3.20), i.e.,
=
2,
2(1 + ),
=0
(3.27)
has and linear in and .

This homogeneity is apparently broken for = 0 by the overall e z factor in the
metric (3.26). Remarkably, it can be effectively regained at the level of the quadratic fluctuations
if we properly redefine the fluctuation fields when expanding near the classical solution (3.27)
=

2 + e 2 ,
yk
= e

2
yk ,
=e
Ys = e

2

2
Ys ,

2 + 2 + e 2 ,
k = 1, . . . , 2 2,
s = 1, . . . , 5 + 2.
(3.28)
The Euclidean-signature quadratic fluctuation Lagrangian in the conformal gauge will then have
constant coefficients. Indeed, it is given by (up to integration by parts and trivial rescalings)
32 Our notation for , differs from equation in [26] by
2 factor; I is Euclidean action, i.e., I = iS in the notation

of [26]. The 1/2 factor in the integral over y comes from the Jacobian.
33 As was noted in [26], to get this singular part of area one actually needs only the = 0 form of the classical solution.
Ignoring the -dependence in (3.20), the restriction
of (y+ , y ) to the square [(0, u ), (0, u )] can be easily translated
into the domain in (, ) space using that y = e 2( ) . Integrating the square root of the determinant of the induced
metric (3.21) in this region will lead to the same expression for the area as in the y coordinates in (3.23).
112
1
1
1
a a
a
L 2 = (1 + ) a a + (1 + ) a a a
2
2
2

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

1 a
1 2 2 1 a
2
+ yk a yk + 1 + yk + Ys a Ys + Ys2 .
2
2
2
4
(3.29)
For = 0 this Lagrangian becomes equivalent to (2.36) (with S 5 modes added).

Thus even for = 0, the 1-cusp solution is again effectively homogeneous, and, as in the
discussion of the non-regularized case in (2.42), we should then find that the 1-loop quantum
correction is again proportional to (regularized value of) the world-sheet area.
It is, however, clear from the structure of the metric (3.26) that once we go beyond the
quadratic fluctuation (string 1-loop) level the interaction vertices will contain powers of the effective -dependent coupling
proportional to the inverse of the running string tension, i.e.,
1 z 1 e = 1 e 2 . While this effective string tension provided a cutoff at small z (large
negative ) at the string tree level, i.e., in the world-sheet area, this apparently will no longer be
so at higher orders of inverse tension expansion. The same conclusion is then expected also in
the case of the 4-cusp solution of [26]. Thus there appears to be a problem with implementation
of the idea of this dimensional regularization, at least order by order in 1 expansion. We shall
return to the discussion of this issue below.

Given that the action (3.29) has constant coefficients it is straightforward to find the fluctuation
spectrum. We shall first assume that and run in the infinite range, i.e., ignore the presence of
a cutoff on the world-sheet coordinates that implements an IR cut off in spacetime (we will
need to introduce it at the end to regularize the area factor as in the 1-cusp case in (2.42)). We
shall also assume trivial boundary conditions on the 2d fluctuation fields. Then we can use the
standard 2d momentum representation to diagonalize the fluctuation modes.34
The three mixed modes , , in (3.29) lead to the following contribution to the 1-loop effective action, i.e., to the 1/2 of the logarithm of the fluctuation determinant (here p2 = p12 + p22 ;
cf. (2.39), (2.40))

1 2
1 2
d 2p
2
2
ln p1 + p2 +
+ ln p1 + p2
2
2
(2)2

1
+ ln p 2 + 4 + 4 + 2 .
2
1
V2
2
(3.30)
34 A clarification is in order. When discussing a spectrum of quadratic fluctuations of a string sigma model we may

assume that fluctuations of the coordinates are normalized as d 2 g x M x N GMN (x). The redefinition of the fluctuM
ations made in (3.28) removes the GMN (x) factor, i.e., replaces x by the tangent-space vectors x M . The fiducial 2d
metric g ab may be either the original independent metric of Polyakovs action or the induced metric (on the equations of
motion or in the conformal gauge they differ only by a conformal factor). Since the induced metric in (3.21) is not flat for
= 0 one may wonder if we are allowed to consider the standard Fourier-mode basis when computing the spectrum. The
answer is yes, provided all the modes (including the fermions and the conformal gauge ghosts) are normalized using the
same fiducial metric: its conformal factor dependence should then cancel out since the theory should not have nontrivial
Weyl anomaly (see also [17]).
113
The contribution of the first two terms here is equivalent to the contribution of the two massless
modes, i.e., is the same as in the = 0 case. Indeed, using polar coordinates
in p-space and
2
a
1
2
2
integrating over the angle using 0 d ln(a + b cos ) = 2 ln( 2 + 2 a b ) we find that

1 2
1 2
d 2 p ln p12 + p2 +
+ ln p12 + p2
= 2 d 2 p ln p 2 .
2
2
(3.31)
This is exactly what is required to cancel the contribution of the two massless modes of the
diffeomorphism ghosts in the conformal gauge.
The action needed to find the fermionic contribution was already discussed in Section 3.1.
Starting with the action (3.10) where H is now given by (3.5) with D = 4 2, one should
plug in there the solution (3.20) and repeat the same computation of the fermionic fluctuation
frequencies as described at the end of the previous subsection. Using the 1 = 2 gauge and
noticing that the Abelian connection term drops out since A = 0, it follows that the quadratic
fermion Lagrangian is
L = 2 2 i u (0 + 2 + 1 )0 + 1 + 2 1

1
(1 + )(2 + ) 012 .
2
(3.32)
The overall factor of u can be eliminated by rescaling of by u1/2 . The fermion propagator is
then proportional to the inverse of p 2 + (1 + 12 ) implying that we find 8 fermionic modes with
mass squared equal to 1 + 12 .
Including the contribution of the 2 2 longitudinal modes yk and 5 + 2 sphere modes Ys
in (3.29) we end up with the following expression for the 1-loop effective action generalizing the
one in (2.39), (2.40) to the case of = 0

1 2
1 2
d 2p
2
2
ln p + 4 + 4 + + (2 2) ln p + 2 1 +
2
2
(2)2

1
1
+ (5 + 2) ln p 2 + 2 8 ln p 2 + 1 + .
2
2
1
1 = V2
2
(3.33)
As
is 2d UV finite (the degrees of freedom and mass sum rules
expected,
this expression
2 = 0 are satisfied)35 for any : the D-instanton background (3.5) we started
n
=
0,
n
M
i
i
i
i
i
with is -symmetric and thus the GS superstring action should be 1-loop finite.
The integral over the 2-momentum can be computed explicitly and we get (with V2 = 2A2 as
appropriate for a single cusp case as in (2.41))
1 =
1
c1 ()A2 ,
4
(3.34)
35 This implies also that the result cannot depend on a mass scale or the size of the box in which and take values,
i.e., it can depend only of ratios of masses or scales.
114
c1 () =
0

1
1 2
dv ln v + 4 + 4 + 2 + 2(1 ) ln v + 2 1 +
2
2

1 2
1
+ (5 + 2) ln v + 8 ln v + 1 +
2
2

1
= 8 + 8 + 2 ln 8 + 8 + 2 + 2 2 + 3 + 2 ln(2 + )
2
2 (5 + 2) ln ||.
(3.35)
Expanding for || 0 gives

1
c1 () = 12 ln 2 4(1 + 2 ln 2) + (1 + 9 ln 2) 2 5 2 ln || + O 3 .
2
(3.36)
1
c1 (0) in (2.41). Note that the 2 ln || term here
For = 0 we reproduce the value of a1 = 4
5
originates from the S modes that were massless in the D = 4 case.
The 2d area A2 that factorizes here as in (2.39) due to the effective
homogeneity of the back

1
1
2
ground is simply 2 d d , i.e., it does not contain the z e
(the effective string tension)
factor that was present in the classical expression for the area (3.23). Thus we will need to impose
both UV and IR cutoffs as in the 1-cusp solution [8] in (2.42). Assuming a similar conclusion will also apply to the full 4-cusp solution this indicates that the dimensional regularization
prescription of [26] fails to regularize the 1-loop string correction to the logarithm of the dual
Wilson loop.36
Despite the redefinition (3.28) that eliminated the z1 factor from the fluctuation action we
may attempt to consider, at least at a heuristic level, a possibility that the area factor A2 in (3.34)
should still be computed using the -dependent induced metric in (3.21), i.e., should be taken
as in (3.23), (3.24) but without 2 times the string tension factor T (with extra 2 accounting for
the definition of the cusp area as in (2.42), cf. also (3.24)).37 In this case
(and assuming again
the momentum IR cutoff on the length of the cusp sides, like u = 2 |s|) we will have from
36 It is interesting to note that the fact that in the absence of regularization the 1- and 4-cusp solutions are related [26]
by the transformation (2.29) combined with a conformal boost imply that the calculation above applies to the 4-cusp
Wilson loop as well, just that the regulator factor being used in the metric is not the one of [26]. Indeed, it is not hard to
find the image of z1 through the transformations relating the 1- and 4-cusp solutions. Thus, given the 1-cusp solution for
1 (dz2 + dy 2 ) one may construct (through the SO(2, 4) transformations that work at = 0) a 4-cusp
z2+

solution for the Lagrangian R2 (dz2 + dy 2 ) where R stands for (2 2z)/[1 + z2 2(y0 + y1 y2 ) y02 + y12 + y22 + y32 ].
z
the Lagrangian
This observation reiterates the fact that the right string theory counterpart of the regularization needs to be identified for
the correct singular part of the amplitudes to be reproduced correctly.
37 An attempt to justify this suggestion could be based on the fact that the fluctuation fields in (3.29) should be normalized using the induced metric in (3.21). Then the eigenvalue problem for the fluctuation operators in (3.29) should be
defined with an extra conformal factor coming from the conformally flat metric (3.21), and that may at the end produce
the area factor defined with the non-flat induced metric. However, the condition of cancellation of the total Weyl anomaly
means that one should be able to completely get rid of the conformal factor dependence (modulo possibly complications
with boundary conditions that we are ignoring here). Consider, e.g., a 1-loop contribution of a massive scalar in curved
conformally flat 2d space, i.e., 1 = 12 ln det( 1g 2 + m2 ) or, up to a conformal anomaly term, 12 ln det( 2 + M 2 ),

M 2 = gm2 . Then 1 scales as d 2 M 2 + = d 2 gm2 + . In our present case M 2 is actually constant, so
M
we do not get an extra g factor in the integral in front of M 2 ln M1 type terms in the effective action.
2
115
(3.23) (using tilde to indicate the result of this heuristic prescription)

K
1
1
ln(2u2 )
A 2 = 2
= 2
+ O 0 .
1
4
2 (2u2 ) 2
2
(3.37)
Then the product of c1 (3.36) and (3.37) in (3.34) appears to contain a surprising ln || term. This
term should presumably be omitted: it is an artifact of the procedure of regularizing separately
the area and the fluctuation determinant. In practice, one should first take to zero in c1 and then
multiply it by the divergent area, i.e., the product 12 2 ln 2 should be set equal to zero in the
0 limit.38 Under all these assumptions one would finish with the following expression for
the singular part of the 1-loop correction

2 1

3 ln 2 1
1 + 2 ln 2
3
1 =
(3.38)
+
+
ln 2 ln 2u
+ O 0 .
2 2
2
4

3.3. Problems with IR dimensional regularization in string inverse tension expansion
As we have pointed out above, the structure of the D = 10 metric (3.26) with running effective string tension implies that the presence of the dependent factors will no longer regularize
the quantum string expressions in the z 0 area beyond the classical level. A related issue is that
of applicability of the
1 expansion here since the curvature of the background is singular
near z = 0 (cf. [26]).

But even assuming that some modified string theory side version of dimensional regularization prescription will lead to the b 21 + b2 + b3 contributions like (3.38) at each order in the 1
expansion, we will still face the following problem in matching this strong-coupling expansion
with expectations based on the perturbative gauge-theory relations for the IR dimensionally regularized gluon amplitudes.
The general expression [22] for the IR singular part (1.2) of the on-shell dimensionally regularized gluon amplitude applied to the planar limit of the N = 4 SYM theory yields [24]

1 (2)
1 (1)
2
() g
() , ,
Adiv (, s) = exp 2 f
(3.39)
4
|s|
8
where [21,22]
d2
d
(3.40)
f (2) (),
g() =
g (1) (),
d ln
d ln 2
and f () is the cusp (soft) anomalous dimension. These relations were applied at leading strong
coupling order in [26] giving the expressions for f and g in (1.3).
One may argue that the application of these relations at strong coupling is justified because
of the finite radius of convergence of the planar weak coupling perturbation theory. The latter
suggests that Eqs. (3.40) should be valid in a finite disk around the origin in complex space
from where they may be analytically continued to the strong coupling region. It is not clear,
however, why f (2) () should have a regular series expansion at large . It is consequently
unclear whether the relations (3.40) should hold order by order in the strong coupling expansion.
f () =
38 Note in particular that this issue is unrelated to the expansion of the cusp solution in small : the problem comes from
the structure of the dimensionally continued metric in (3.15). This problem is more likely to be a further indication of
problems with the dimensional regularization that we discuss below.
116
Indeed, there is an apparent problem with this suggestion which appears when we consider
the first subleading order in 1 expansion. Given the relation between gluon amplitudes and
momentum-space Wilson loops proposed in [26] we expect that the exponent of Adiv (, s) in
(3.39) should
string perturbative expansion, i.e., that both f (2) and g (1) should
have standard
k2
+ when expanded at large . On the other hand, the same pattern of

scale as k0 + k1 +
strong coupling expansion is known to apply to f () [7,10,12] (and should presumably be true
also for g()). This is, however,
inconsistent with the relations (3.40). Indeed, to reproduce the
expansion (1.1), i.e., f () = a0 + a1 + a2 + using (3.40) we need to assume that f (2) ()
should have the following behavior at large
4a2
f (2) ( 1) = 4a0 + 2a1 (ln )2 + + .
(3.41)
Here
the first term has the expected classical string form [26] but the first subleading term is
(ln )2 instead of a constant.39 Such logarithmic term cannot appear as a perturbative string
sigma correction. One may conjecture that it originates from a resummation of all higher-order
strong coupling corrections.40 Another possibility is that (3.39) with (3.40) do not actually apply
directly to the strong-coupling expansion as defined by the perturbative string theory.
And vice versa, even assuming that (3.38) represents the 1-loop contributions to f (2) ()
and g (1) (), the latter are then constant (-independent) and thus, according to (3.40), do not
change f () (and g()) at all, in contradiction with (1.1).
One abstract possibility to reconcile the strong-coupling expansions of f (2) () and of f ()
while preserving their relation in (3.40) could be to redefine by a constant that
eliminatesthe
constant a1 term in f (). Namely, if we shift the string tension by a constant + aa10
and identify
with the gauge-theory coupling we would have no 1-loop correction in both f ()
and f (2) . Shifting the string tension by a constant seems to lead to problems with various other
comparisons between the AdS5 S 5 string theory and the N = 4 supersymmetric gauge theory
but in this particular setting it may seem we have little understanding of how the two couplings
should actually be related.
In Appendix B we shall present another attempt to modify the dimensional regularization
prescription of [26] by relaxing the identification between the parameters on the gauge theory
and the string theory sides, which may help to resolve the above problem.
Regardless of the problems discussed above, an unwelcome feature of the string-side version of the IR dimensional regularization suggested in [26] is that instead of providing a simple
modification of the known AdS5 S 5 string action (by analogy with what is done to define the
dimensionally regularized gauge theory)41 it instructs us to start with a dimensionally-modified
supergravity background and rederive the superstring action from scratch. This is rather daunting task beyond the quadratic level in fermions which casts doubt on a practical utility of this
prescription. For example, extending the above 1-loop computations to the 2-loop string theory
level would be quite complicated.
39 We can of course add also a constant and as well as ln
term to f (2) () as this will not change the expression for

f ().
40 We acknowledge a discussion with J. Maldacena on this issue.
41 That would be the case if one would simply impose a cutoff at small z, by, e.g., z2 (z2 + 2 )1 (cf. [9,39]).
But then one would need to identify the corresponding regularization on the gauge theory side. Further complications
with such a regularization include difficulties in finding the minimal surface in the presence of the regulator as well as in
finding a consistent GreenSchwarz action.
117
Note added
An interesting suggestion of how to reproduce the (ln )2 term in (3.41) (and also how to
treat higher-loop orders) was suggested to us by J. Maldacena (private communication). The
effective string sigma model coupling that determines the higher-loop order corrections (after we

rescale the fluctuation fields as in (3.28) to put the propagator term into a canonical form) is z
which grows near z = 0 for < 0. At one loop we may then cut off z at a minimal value zo where

z
o is fixed to some given value k, i.e., ln zo = 1 ln(k ). In this case the area factor of a single

cusp surface which we got in (3.34) written in the same way as in (3.23) can be regularized as

dy+ dy
, where the
follows (ignoring terms of subleading orders in ): A2 = 12 d d 12 2(2y
+ y )
integral should be cut off at large y at u and small y by the above cut off at small z. Namely,
u = y+ y > uo = 1 zo implies that one is to integrate over a triangle instead of a square

2
in the ln y plane, ln y+ + ln y = 2 ln u > 2 ln u
o , and that gives an extra factor of 2 compared
to (3.23). As a result, A2 = 14 (ln uuo )2 = 412 (ln )2 + . We then have according to (3.34)
2
a1
) + which matches the result for the contribution of
and (3.36) 1 = a1 A2 + = 4
2 (ln
the one of 4 divergent factors in the amplitude (1.2) that follows from (3.39), (3.41). For higher
orders one may attemptto analytically continue in , effectively reversing its sign; replacing then
z
1 by
with z 2y+ y under the world-sheet integral at each loop order then appears
to lead to coefficients in f (2) () consistent with (3.40). It still remains to be seen if there is
a systematic version of dimensional regularization that may allow one to capture the strongcoupling expansion not only for f () but also for g().
Acknowledgements
We are grateful to L.F. Alday, Z. Bern, L. Dixon, S. Frolov, G. Korchemsky, D. Kosower,
T. McLoughlin and especially J. Maldacena for useful communications and discussions. The
work of M.K. was supported by the National Science Foundation under grant No. PHY-0653357.
R.R. acknowledges the support of the National Science Foundation under grant PHY-0608114.
A.T. was supported in part by the DOE grant DE-FG02-91ER40690. A.A.T. acknowledges the
support of the PPARC, INTAS 03-51-6346, EC MRTN-CT-2004-005104 and the RS Wolfson
award.
Appendix A. Construction of O() correction to the 4-cusp solution
In this appendix we describe the construction of the O() correction to the 4-cusp Wilson loop
solution (found in original AdS5 S 5 unregularized form in [26]) and outline the structure of
the quadratic fluctuation Lagrangian.
To set up the stage, consider the general Lagrangian

L = h() L=0 ()
(A.1)
and let = 0 be a solution of the Lagrangian L=0 (). We are interested in finding the
function 1 such that = 0 + 1 is a solution of the Lagrangian L to leading and nextto-leading order in the expansion in 0. As mentioned previously, it is not a priori clear that
such an expansion is sufficient for testing the conjectures of [26] and [24]; still, it gives some
information on a class of functions that may appear in the complete solution.
118
The strategy is relatively straightforward. First, we expand L=0 around 0 . The inclusion of
the additional factor h modifies the equations of motion of fluctuations by potential terms. The
solution to these deformed equations yield the desired correction.
Indeed, the expansion of L=0 () around the solution = 0 has the following structure:

L=0 (0 + ) = L=0 (0 ) + a ca + 2 K(0 )

+ 2 a V a (0 ) + O 3 .
(A.2)
The first term here is simply the value of the Lagrangian on the classical solution. The second
and fourth terms are total derivatives that integrate to zero at the level of the action. K(0 ) is the
kinetic operator of the small fluctuations around 0 and, in general, it may be a function of the
world sheet coordinates.
With this starting point it is easy to understand the structure of the equations determining 1 .
Indeed, in the presence of h the field configuration 0 is no longer a solution of the equations of
motion. Instead, at the level of the action, the fluctuations exhibit a tadpole which, to leading
order in is just
Ltadpole = ca a ln h(0 ).
(A.3)
Thus, to summarize, the equations determining 1 are the equations of motion of the fluctuations
around 0 at = 0 deformed by a potential generated by the tadpole Lagrangian (A.3).
In our case we have h() 1/z and the undeformed solution 0 is the 4-cusp solution of
[26] in the conformal gauge
a
,
cosh u2 cosh u1 + b sinh u2 sinh u1
a 1 + b2 sinh u2 sinh u1
y00 =
,
a sinh u2 cosh u1
y01 =
,
a cosh u2 sinh u1
y02 =
,
y03 = 0,
z0 =
(A.4)
where the parameters a and b are related to the Mandelstam invariants [26] and (u1 , u2 ) = (, ).
The relation between the 1- and 4-cusp solutions implies that a convenient choice of the basis
of fluctuation fields
z = z0 + z,
yi = y0i + yi ,
x)
is given by (y,
1 , 2 , ,
z =

a
2 1 + b2 y
2
2(cosh u2 cosh u1 + b sinh u2 sinh u1 )
(cosh u1 sinh u2 + b cosh u2 sinh u1 )(1 + 2 )
(cosh u2 sinh u1 + b cosh u1 sinh u2 )(1 2 )

+ 2(b cosh u1 cosh u2 + sinh u1 sinh u2 ) ,
(A.5)
119

a
4(b cosh u1 cosh u2 sinh u1 sinh u2 )y
2
+ sinh(2u1 )(1 + 2 ) + sinh(2u2 )(1 2 )

2 1 + b2 cosh(2u1 ) + cosh(2u2 ) ,

a
2 1 + b2 cosh u1 sinh u2 y
y1 =
2(cosh u2 cosh u1 + b sinh u2 sinh u1 )2
y0 =
+ cosh2 u1 (1 + 2 ) b sinh2 u2 (1 2 )

sinh(2u1 ) b sinh(2u2 ) ,

a
2 1 + b2 cosh u2 sinh u1 y
y2 =
2
+ cosh2 u2 (1 + 2 ) b sinh2 u1 (1 2 )

sinh(2u2 ) b sinh(2u1 ) ,
y3 =
a x
.
(A.6)
This choice leads to constant coefficients in the operator K0 .

It is straightforward (though somewhat tedious) to find the equations corrected by the presence
of the regulator. Only the solution obeying the Virasoro constraints is physically interesting.
Since for = 0 the solution does obeys the constraints, the resulting conditions on the correction
are not sensitive to the regulator. Indeed, it turns out that the vanishing of the 2d stress tensor of
x)
the fluctuation fields requires that (y,
1 , 2 , ,
are related by
(u1 u2 )1 (u1 + u2 )2 = 0,
(u1 + u2 )1 + (u1 u2 )2 = 4.
(A.7)
The solution of these equations can be parametrized by a single arbitrary function G:

2
2 = (u1 u2 )G,
u + u2 G = 2.
1 = (u1 + u2 )G,
1
(A.8)
Further using the Virasoro constraints to simplify the remaining equations of motion it is not
are determined by
hard to see that they in fact decouple. The remaining functions (x,
y,
)

2
sinh u1 sinh u2 + b cosh u1 cosh u2
= 0,
u1 + u22 4

2
u1 + u22 2 x = 0,
2

1 + b2
2
u1 + u2 2 y
= 0.
(A.9)
The solution for is then the source for Eq. (A.8) determining G.
The = 0 solution already obeys the boundary conditions that relate its shape to the Mandelstam invariants of the scattering process. The corrections to the solution must therefore leave
these boundary conditions unchanged. For the particular case of s = t (i.e., b = 0) this translates
into
u2 :
u1 :
2 = 1 + 2 ,
2 = 1 2 ,
y = 0,
y = 0.
(A.10)
120
Once a solution to (A.8), (A.9) and (A.10) is found, it is then straightforward to find the
value of the classical action. Since the solution correctly captures only the O() deformation, the
expansion of the classical action should be truncated to this order. The result reads
L ( ) =

1
2 + 4u1 u2 G + O 2 .

z0
(A.11)
1
1
2
Note that (z0 +z)
= z + O( ) so the regulator does not contribute to this order except for the
0
overall factor.
With the O() correction to the classical solution at hand, one may now proceed to compute
the quadratic fluctuation Lagrangian. Only the terms of the order 0 and 1 are reliable.
The idea is to use as much as possible the fact that the expansion of L=0 around the solution (A.4) hasup to total derivativesconstant coefficients if the fluctuation fields are chosen
similarly to Eq. (A.6). With this in mind, the structure of the quadratic fluctuation Lagrangian
contains three types of terms:
(1) the quadratic fluctuation part of L=0 expanded around the = 0 solution,
(2) the cubic vertices of L=0 expanded around the = 0 solution in which one of the fields is
replaced by the O() correction constructed above,
(3) the derivative of L with respect to evaluated on the undeformed solution and expanded to
quadratic order in fluctuations around it.
The terms of the first type have constant coefficients; the coordinate dependence in the terms
of the second type arises entirely from the O() correction to the solution. The position dependence in third type of terms arises from the regulator as well as from the total derivative terms
that integrate to zero in the absence of the regulator.
The calculation of the determinant of this operator is non-trivial due to a complicated position
dependence. An expansion in is, however, possible.
Still, it is important to stress that it is unclear whether such an expansion yields the expected
result for the poles in . First, the leading order result has the same unfortunate features discussed
in Section 3namely, that it seems not to be regularized by non-zero . Also, the orders of poles
in appearing at higher orders strongly depend on the large distance structure of the correction
to the = 0 solution. It is possible that the higher order corrections bring increasingly higher
order poles and the expected pole structure of the IR singular part of the amplitude [24] arises
only after all these poles are resummed.
To conclude, it appears that a solution exact to all orders in is required for such an approach
to have a chance of leading to a finite contribution to the expectation value of the 4-cusp Wilson
loop.
Appendix B. An attempt of modified dimensional regularization prescription
We have seen in Section 3 that a naive application of the regularization scheme suggested in
[26] to the single-cusp momentum space Wilson line does not appear to reproduce the structure of
the gauge theory Sudakov form-factor Adiv (M[gg1] )1/2 beyond the leading strong-coupling
order. It turns out, however, that it is possible to suggest a formal prescription which accomplishes
this for the leading 12 pole and, simultaneously, reproduces the finite part of the logarithm of the
exponentiated 4-gluon scattering amplitude [24].
121
Let us consider applying the dimensional regularization of [26] not at the level of the classical
superstring action but only to the divergent quantities that arise during the calculation. More precisely, using the = 0 solution for the minimal surface and taking into account its homogeneity
property, we can first factorize its area as was done in Section 2.2 and then regularize the area
using the induced metric containing the z1 factor. Demanding that the coefficient of the second
order pole in in the expectation value of the single-cusp Wilson line matches that in the Sudakov form factor will relate the dimensional continuation parameters on the gauge and the string
sides. Applying this relation to the 4-cusp Wilson loop we may then reproduce the conjectured
exponential structure of the amplitude [23,24] to all orders in string perturbation theory.
As discussed in Section 2, the homogeneity of the single-cusp solution and its relation to the
spinning closed string solution imply that the expectation value of the Wilson loop factor satisfies
lnW = ln Z = f ()A2 .
(B.1)
In general, W may stand for all Wilson loops whose corresponding minimal surfaces are related
by symmetries of AdS5 S 5 to the single-cusp Wilson loop surface. The difference between the
minimal surfaces will manifest itself in the values of their regularized areas A2 .
An interesting observation is that while the expression above does not immediately reproduce
the singular part of the 4-gluon scattering amplitude, it does, however, correctly capture its finite
part. Indeed, the calculation of the regularized area in [26] yields
4-cusp
A2
string
Adiv,s
string
= 2Adiv,s

=
1
2
2string
string
+ 2Adiv,t

1
s 2
+ const,
ln
8
t
1 ln 2
+
4string
2
|s|
(B.2)
1 string
2
(B.3)
Using it in Eq. (B.1) implies that the terms unrelated to the IR singularities are

f ()
s 2

F1 =
+ const,
ln
8
t
(B.4)
which reproduces (up to a constant shift) the expected structure of the finite part of the 4-gluon
amplitude [2426].
Assuming that there should indeed be a relation between the null momentum Wilson loops
and the gluon amplitudes, the observation above suggests that we should aim at identifying
string
f ()Adiv,s with the logarithm of the Sudakov form factor, namely
f ()Adiv,s (string , string ) =

2gauge
|s|
1
2
8gauge
+
f (2) ()
1
4gauge
g (1) (),
gauge
.
(B.5)
122
This identification should hold up to O(string ) corrections. By comparing the second order poles
it is easy to see that they can be made to agree if we choose42

f ()
.
string = gauge
(B.6)
f (2) ()
It appears hard if not impossible, however, to match the simple pole and the finite parts on
the l.h.s. and r.h.s. of Eq. (B.5). To match the simple pole it seems necessary to demand that
q()
(2 /s)string = (2 /s)gauge . Further ad hoc adjustments appear to be needed in order to match the
finite parts of (B.5).
References
[1] D.J. Gross, F. Wilczek, Asymptotically free gauge theories. 2, Phys. Rev. D 9 (1974) 980;
H. Georgi, H.D. Politzer, Electroproduction scaling in an asymptotically free theory of strong interactions, Phys.
Rev. D 9 (1974) 416.
[2] E.G. Floratos, D.A. Ross, C.T. Sachrajda, Higher order effects in asymptotically free gauge theories. 2. Flavor
singlet Wilson operators and coefficient functions, Nucl. Phys. B 152 (1979) 493;
A. Gonzalez-Arroyo, C. Lopez, Second order contributions to the structure functions in deep inelastic scattering. 3.
The singlet case, Nucl. Phys. B 166 (1980) 429;
A.V. Kotikov, L.N. Lipatov, A.I. Onishchenko, V.N. Velizhanin, Three-loop universal anomalous dimension of the
Wilson operators in N = 4 SUSY YangMills model, Phys. Lett. B 595 (2004) 521;
A.V. Kotikov, L.N. Lipatov, A.I. Onishchenko, V.N. Velizhanin, Phys. Lett. B 632 (2006) 754, hep-th/0404092,
Erratum.
[3] A.M. Polyakov, Gauge fields as rings of glue, Nucl. Phys. B 164 (1980) 171;
R.A. Brandt, F. Neri, M.A. Sato, Renormalization of loop functions for all loops, Phys. Rev. D 24 (1981) 879.
[4] N.S. Craigie, H. Dorn, On the renormalization and short distance properties of hadronic operators in QCD, Nucl.
Phys. B 185 (1981) 204;
G.P. Korchemsky, A.V. Radyushkin, Renormalization of the Wilson loops beyond the leading order, Nucl. Phys.
B 283 (1987) 342;
I.I. Balitsky, V.M. Braun, Evolution equations for QCD string operators, Nucl. Phys. B 311 (1989) 541.
[5] G.P. Korchemsky, Asymptotics of the AltarelliParisiLipatov evolution kernels of parton distributions, Mod. Phys.
Lett. A 4 (1989) 1257;
G.P. Korchemsky, G. Marchesini, Structure function for large x and renormalization of Wilson loop, Nucl. Phys.
B 406 (1993) 225, hep-ph/9210281;
A. Bassetto, I.A. Korchemskaya, G.P. Korchemsky, G. Nardelli, Gauge invariance and anomalous dimensions of a
light cone Wilson loop in lightlike axial gauge, Nucl. Phys. B 408 (1993) 62, hep-ph/9303314.
[6] Y. Makeenko, P. Olesen, G.W. Semenoff, Cusped SYM Wilson loop at two loops and beyond, Nucl. Phys. B 748
(2006) 170, hep-th/0602100.
[7] S.S. Gubser, I.R. Klebanov, A.M. Polyakov, A semi-classical limit of the gauge/string correspondence, Nucl. Phys.
B 636 (2002) 99, hep-th/0204051.
[8] M. Kruczenski, A note on twist two operators in N = 4 SYM and Wilson loops in Minkowski signature, JHEP 0212
(2002) 024, hep-th/0210115.
[9] Y. Makeenko, Light-cone Wilson loops and the string/gauge correspondence, JHEP 0301 (2003) 007, hepth/0210256.
42 This identification also leads to a successful comparison of the coefficient of the double-pole in the vev of the singlecusp Wilson line and that of the Sudakov form factor. Indeed, Eq. (3.37) implies that

f ()
1
1
f (2) ()
+O
+O
= 2
.
lnW1 cusp = 2
string
gauge
string
gauge
123
[10] S. Frolov, A.A. Tseytlin, Semiclassical quantization of rotating superstring in AdS5 S5 , JHEP 0206 (2002) 007,
hep-th/0204226.
[11] S. Frolov, A. Tirziu, A.A. Tseytlin, Logarithmic corrections to higher twist scaling at strong coupling from
AdS/CFT, Nucl. Phys. B 766 (2007) 232, hep-th/0611269.
[12] R. Roiban, A. Tirziu, A.A. Tseytlin, Two-loop world-sheet corrections in AdS5 S5 superstring, arXiv: 0704.3638
[hep-th].
[13] J.M. Maldacena, Wilson loops in large N field theories, Phys. Rev. Lett. 80 (1998) 4859, hep-th/9803002;
S.J. Rey, J.T. Yee, Macroscopic strings as heavy quarks in large N gauge theory and anti-de Sitter supergravity, Eur.
Phys. J. C 22 (2001) 379, hep-th/9803001.
[14] N. Drukker, D.J. Gross, H. Ooguri, Wilson loops and minimal surfaces, Phys. Rev. D 60 (1999) 125006, hepth/9904191.
[15] R. Kallosh, A.A. Tseytlin, Simplifying superstring action on AdS5 S5 , JHEP 9810 (1998) 016, hep-th/
9808088.
[16] S. Forste, D. Ghoshal, S. Theisen, Stringy corrections to the Wilson loop in N = 4 super-YangMills theory,
JHEP 9908 (1999) 013, hep-th/9903042.
[17] N. Drukker, D.J. Gross, A.A. Tseytlin, GreenSchwarz string in AdS5 S5 : Semiclassical partition function,
JHEP 0004 (2000) 021, hep-th/0001204.
[18] S. Frolov, A.A. Tseytlin, Multi-spin string solutions in AdS5 S5 , Nucl. Phys. B 668 (2003) 77, hep-th/
0304255.
[19] G. Arutyunov, J. Russo, A.A. Tseytlin, Spinning strings in AdS5 S5 : New integrable system relations, Phys. Rev.
D 69 (2004) 086009, hep-th/0311004.
[20] G.P. Korchemsky, A.V. Radyushkin, Loop space formalism and renormalization group for the infrared asymptotics
of QCD, Phys. Lett. B 171 (1986) 459;
G.P. Korchemsky, On near forward high-energy scattering in QCD, Phys. Lett. B 325 (1994) 459, hep-ph/9311294;
I.A. Korchemskaya, G.P. Korchemsky, High-energy scattering in QCD and cross singularities of Wilson loops,
Nucl. Phys. B 437 (1995) 127, hep-ph/9409446.
[21] A. Sen, Asymptotic behavior of the Sudakov form-factor in QCD, Phys. Rev. D 24 (1981) 3281;
G.P. Korchemsky, Double logarithmic asymptotics in QCD, Phys. Lett. B 217 (1989) 330.
[22] L. Magnea, G. Sterman, Analytic continuation of the Sudakov form-factor in QCD, Phys. Rev. D 42 (1990) 4222;
G. Sterman, M.E. Tejeda-Yeomans, Multi-loop amplitudes and resummation, Phys. Lett. B 552 (2003) 48, hepph/0210130.
[23] C. Anastasiou, Z. Bern, L.J. Dixon, D.A. Kosower, Phys. Rev. Lett. 91 (2003) 251602, hep-th/0309040.
[24] Z. Bern, L.J. Dixon, V.A. Smirnov, Iteration of planar amplitudes in maximally supersymmetric YangMills theory
at three loops and beyond, Phys. Rev. D 72 (2005) 085001, hep-th/0505205.
[25] J.M. Drummond, G.P. Korchemsky, E. Sokatchev, Conformal properties of four-gluon planar amplitudes and Wilson
loops, arXiv: 0707.0243 [hep-th].
[26] L.F. Alday, J. Maldacena, Gluon scattering amplitudes at strong coupling, arXiv: 0705.0303 [hep-th].
[27] D.J. Gross, P.F. Mende, The high-energy behavior of string scattering amplitudes, Phys. Lett. B 197 (1987) 129.
[28] J. Dai, R.G. Leigh, J. Polchinski, New connections between string theories, Mod. Phys. Lett. A 4 (1989) 2073.
[29] A.A. Migdal, Momentum loop dynamics and random surfaces in QCD, Nucl. Phys. B 265 (1986) 594.
[30] A.M. Polyakov, String theory and quark confinement, Nucl. Phys. B (Proc. Suppl.) 68 (1998) 1, hep-th/9711002.
[31] R.R. Metsaev, A.A. Tseytlin, Type IIB superstring action in AdS5 S5 background, Nucl. Phys. B 533 (1998) 109,
hep-th/9805028.
[32] A.A. Tseytlin, K. Zarembo, Wilson loops in N = 4 SYM theory: Rotation in S5 , Phys. Rev. D 66 (2002) 125010,
hep-th/0207241.
[33] M. Kruczenski, Spiky strings and single trace operators in gauge theories, JHEP 0508 (2005) 014, hep-th/0410226;
M. Kruczenski, J. Russo, A.A. Tseytlin, Spiky strings and giant magnons on S5, JHEP 0610 (2006) 002, hepth/0607044;
R. Ishizeki, M. Kruczenski, Single spike solutions for strings on S2 and S3, arXiv: 0705.2429 [hep-th].
[34] R. Kallosh, J. Rahmfeld, The GS string action on AdS5 S5 , Phys. Lett. B 443 (1998) 143, hep-th/9808038.
[35] R.R. Metsaev, A.A. Tseytlin, Superstring action in AdS5 S5 : Kappa-symmetry light cone gauge, Phys. Rev. D 63
(2001) 046002, hep-th/0007036.
[36] G.W. Gibbons, M.B. Green, M.J. Perry, Instantons and seven-branes in type IIB superstring theory, Phys. Lett.
B 370 (1996) 37, hep-th/9511080.
[37] M.T. Grisaru, P.S. Howe, L. Mezincescu, B. Nilsson, P.K. Townsend, N = 2 superstrings in a supergravity background, Phys. Lett. B 162 (1985) 116.
124
[38] E.I. Buchbinder, Infrared limit of gluon amplitudes at strong coupling, arXiv: 0706.2015 [hep-th].
[39] S. Abel, S. Forste, V.V. Khoze, Scattering amplitudes in strongly coupled N = 4 SYM from semiclassical strings in
AdS, arXiv: 0705.2113 [hep-th].
[40] A. Brandhuber, P. Heslop, G. Travaglini, MHV amplitudes in N = 4 super-YangMills and Wilson loops, arXiv:
0707.1153 [hep-th].
Brane-worlds pseudo-Goldstinos
Karim Benakli , Cesar Moura
Laboratoire de Physique Thorique et Hautes Energies, Universit Pierre et Marie Curie,
Paris VI Universit Denis Diderot, Paris VII, France
Abstract
We consider a spacetime with extra dimensions containing sectors, branes and bulk, that communicate
only through gravitational interactions. In each sector, if considered separately, supersymmetry could be
spontaneously broken, leading to the appearance of Goldstinos. However, when taken all together, only certain combinations of the latter states turn out to be true would-be-Goldstinos, eaten by the gravitinos. The
other (orthogonal) combinations, we call pseudo-Goldstinos, remain in the low energy spectrum. We discuss explicitly how this happen in the simplest set-up of five-dimensional space compactified on S 1 /Z2 . Our
results divide into two parts that can be considered separately. First, we build an extension of the bulk fivedimensional supergravity, by a set of new auxiliary fields, that allows coupling it to branes where supersymmetry is spontaneously broken. Second, we discuss in details the super-Higgs mechanism in the R and unitary gauges, in the presence of both of a bulk ScherkSchwarz mechanism and brane localized F -terms. This
leads us to compute the gravitino mass and provide explicit formulae for the pseudo-Goldstinos spectrum.
PACS: 11.10.Kk; 11.25.Wx; 11.30.Pb; 04.50.Cd; 04.65.+e
Keywords: Supersymmetry breaking; Branes; Goldstino; Gravitino
1. Introduction and conclusions

If supersymmetry has to play a role among the fundamental laws governing our world, it has
to appear as spontaneously broken. For the purpose, the world is often described by an effective
four-dimensional supergravity where the spontaneous breaking corresponds to non-vanishing
vacuum expectation values (v.e.v.s) of auxiliary fields, called F -terms and D-terms. In the global
E-mail addresses: kbenakli@lpthe.jussieu.fr (K. Benakli), moura@lpthe.jussieu.fr (C. Moura).

doi:10.1016/j.nuclphysb.2007.09.010
126
K. Benakli, C. Moura / Nuclear Physics B 791 (2008) 125163
supersymmetry limit, the breaking gives rise in the global limit to massless Goldstone fermions,
the Goldstinos [1]. Instead, in the local version, where gravity is taken into account, the (wouldbe)-Goldstinos are absorbed by the massive gravitinos to become their spin 12 components [24].
The last decade has seen the emergence of a popular scenario for the phenomenological implications of the short distance description of spacetime where extra dimensions play an important
role [511]. Some of the light degrees of freedom are confined to live on branes localized at
particular points in a higher-dimensional space. In such a set-up, supersymmetry breaking can
happen either on the branes or in the bulk, and it is usual to discuss the breaking in each sector, separately. For instance, the dynamical breaking of supersymmetry [12] is often studied in
the global limit as due to some non-perturbative gauge dynamics [13,14] that could happen at
different scales on different, spatially separated, branes [15]. The breaking of supersymmetry in
the bulk can be instead achieved through a ScherkSchwarz mechanism [16]. In each of these
sectors would be Goldstinos are predicted. Only certain combinations of the latter are true ones,
eaten by the gravitinos. One asks then about the fate of the remaining states. This work deals
with this issue.
Another problem addressed in this paper is the coupling of the bulk supergravity to the brane,
in the presence of localized F -terms. In the global supersymmetry case, this was studied by
Mirabelli and Peskin [17]. It was found that the five-dimensional super-YangMills theory had
to be extended off-shell by the addition of an appropriate auxiliary field in order to take into account the presence of localized D-terms. Such auxiliary fields can be integrated out, but with the
price of introducing an explicitly singular coupling (0) in the scalar potential, which requires to
be treated carefully as arising from an infinite sum over extra-dimensional momenta. Here, we
propose the adjunction of a new set of auxiliary fields to the minimal five-dimensional supergravity. These fields vanish identically in the supersymmetric limit and allow us to keep track of the
supersymmetry transformations in the case when boundary F -terms are not explicitly put to zero.
There have been a huge number of papers dealing with supersymmetry breaking in extra dimensions (for a few examples, see [1821]). We believe it useful to point out to the reader where
our work stands in this literature. Off-shell extension of minimal five-dimensional supergravity
was built in [22]. This extension was further studied in [2327]. In particular, [25,26] studied
the coupling to boundary branes and discussed in some details supersymmetry breaking through
generalized ScherkSchwarz boundary conditions [28,29] which correspond to a constant superpotential. Their study uses the auxiliary fields obtained in [22]. Our approach here is different.
After tedious computations, we build our Lagrangians from scratch in components fields. We
identify the necessity to introduce a spacetime vector and a scalar as auxiliary fields, and we
derive transformation rules to take into account the presence of arbitrary superpotentials for
breaking supersymmetry. Although we believe that our auxiliary set of fields can be expressed
as a combination of some of those of [22], the relation is not trivial and, for the aim of this work,
we do not find it worth to go through long computations to extract it.
The other issue discussed here is the fate of the would be Goldstinos. The super-Higgs mechanism in the framework of extra dimensions has been discussed in [30] for the case of a bulk
Goldstino and in [31] for the case of generalized ScherkSchwarz mechanism. The on-shell coupling of a brane Goldstino to a bulk supergravity was discusses by [32] in the RandallSundrum
set-up. Our analysis includes both bulk and boundaries would-be-Goldstinos. Finally, it might
be interesting for the reader to make the analogy with bosonic case of pseudo-Goldstones. They
have been introduced by Weinberg in [33], used for electroweak symmetry breaking by [34] and
in the context of large extra dimensions, they were used as Higgs bosons for example in [3537].
Let us summarize our main results:
127
We have introduced an extension with new auxiliary fields in order to keep track of supersymmetry transformation when coupling the five-dimensional supergravity with the branes,
as mentioned above.
The generalized ScherkSchwarz mechanism, as introduced by Bagger et al. [28], allows
localized gravitino masses on boundary branes. We have generalize it to an arbitrary set
of branes suspended in the bulk. The supersymmetry transformations have been derived
for these cases. As a corollary, we obtain the condition for the association of the given
brane-localized gravitino masses with non-trivial ScherkSchwarz twist not to break supersymmetry. We point out explicitely the obstacles when trying to express the F -term breaking
as a generalized twist.
In the case of flat boundary branes and flat bulk, we study in details the gauge fixing for the
super-Higgs mechanism. We discuss both the general R gauge and the unitary gauge. In
the latter, we explicitly obtain the form of the gravitino mass, and show that from the four
original would-be-Goldstinos (two in the bulk and one on each boundary), two are eaten by
the N = 2 bulk gravitinos while two orthogonal combinations remain. We call them pseudoGoldstinos. We explicitly compute the corresponding masses for specific cases, and show
that in the limit of infinite radius they vanish as expected when one decouples gravity.
Our results are obtained with particular assumptions which allow the computations to be carried out explicitly up to the end. The five-dimensional supergravity is taken with the minimal
content on-shell. Only the corresponding states that are even under an appropriately defined Z2
action are assumed to couple on the branes. In deriving the pseudo-Goldstinos spectrum, we will
take all the branes and the bulk to be separately flat. We also assume that we are working in a basis
where the localized superfields providing the would be Goldstinos are canonically normalized.
Moreover, our treatment is at tree-level. We believe that the departure from these assumptions
should not dramatically change the qualitative picture presented in this first work.
The paper is organized as follows. Section 2 displays the Lagrangians corresponding to the
minimal five-dimensional supergravity in the bulk with supersymmetry broken by non-trivial
boundary conditions, as well as the simplest brane action for a set of chiral multiplets with a
priory non-vanishing F -terms. Appendix A summarizes the related conventions. The coupling
of the two sectors, bulk and branes, is performed in Section 3 after the introduction of new auxiliary fields. The corresponding supersymmetry transformations are collected in Appendix B.
Section 4 reviews some issues of the spontaneous breaking of supersymmetry in compactifications on S 1 /Z2 . In particular the interplay between bulk and boundary branes localized gravitino
masses to break (or restore) supersymmetry. These results are generalized in Section 5 to the case
with an arbitrary number of branes suspended at arbitrary points of S 1 /Z2 . Section 6 discusses in
great details the super-Higgs mechanism when supersymmetry is spontaneously broken both in
the bulk and on the brane. We exhibit that combination of would-be-Goldstinos are not absorbed
and form the remaining pseudo-Goldstinos. The explicit form of the latter and their masses are
provided in Section 7, while the general formulae are given in Appendix D.
2. Bulk with boundary branes
Consider a five-dimensional space parametrized by coordinates (x , x 5 ) with = 0, . . . , 3
and x 5 y parametrizing the interval S 1 /Z2 . The latter is constructed as an orbifold from the
circle of length 2R (y y + 2R) through the identification y y. Matter fields live on
branes localized for instance at particular points y = yn . We will assume here that there are only
128
two branes sitting at the boundaries yn = yb {0, R}. The corresponding action can be written
as1 :
2R
S=
dy

1
d x LBULK + L0 (y) + L (y R) .
2
4
(2.1)
2.1. On-shell supergravity action in the bulk

We take the theory in the bulk to be five-dimensional supergravity with the minimal on-shell
A , the gravitino
content being the fnfbein eM
MI and the graviphoton BM . The on-shell La2
grangian is given by [38]:

1
i I MNP
1
DN P I FMN F MN
LSUGRA = e5 R () + M
2
2
4

MI N

1 ABCDE
6
I MNP Q

FAB FCD BE i
QI
FMN 2 I + P
16
6 6
(2.2)
3
and the on-shell supersymmetry transformations are :
A
= i I A MI ,
e M
6 I
I ,
BM = i
2 M
1
MI = 2DM I + F N P (MNP 4gMP N )I ,
(2.3)
2 6
where is the supersymmetry transformation parameter and FMN = M BN N BM . Note that
we use here the symbol for the variation of the fields while the usual symbol will be defined
later to include extra terms (in Section 3).
The five-dimensional spinors MI and I are symplectic Majorana spinors, described in
Appendix A. In Appendix B, we present the Lagrangian (2.2) and the corresponding supersymmetry transformations (2.3) in two-component spinor notation. The five-dimensional gravitino
MI will be written:

M1
M2
,
M2 =
,
M1 =
(2.4)
M2
M1
using the two-component Weyl spinors MI .
Every generic field has a well defined Z2 transformation:
Z2 : (y) P0 (y),
(2.5)
that allows us to define the orbifold S 1 /Z2 from the original five-dimensional compactification
on S 1 . Here P0 is the parity of the field which obeys P02 = 1. The Lagrangian (2.2) and
supersymmetry transformations (2.3) must be invariant under the action of the mapping (2.5).
5
R 4
1 2 R dy d 4 x.
1 The factor 1 in front of L
BULK in Eq. (2.1) comes from: d x = 0 dy d x = 2 0
2
2 Unless stated otherwise, we take , h and c equal to 1. See Appendix A for our conventions.
3 In this paper we make the following approximations: we drop the four-fermions terms in the Lagrangian and the three
and four-fermions terms in the supersymmetry transformations.
129
Table 1
Parity assignments for bulk fields at y = 0
P0 = +1
a
e
e55
B5
52
P0 = 1
e5a
5
e
51
At the point y = 0, we assume the fnfbein to transform as:

a
a
e
(y) = +e
(y),
5
5
(y) = e
(y),
e
e5a (y) = e5a (y),
e55 (y) = +e55 (y).
(2.6)
These assumptions and the invariance of supersymmetry transformations under the Z2 action imply that M1 and M2 must have opposite parities. The Lagrangian (B.4) is also invariant under
an SU(2)R R-symmetry, under which the gravitinos M1 and M2 transform in the representation 2 of SU(2)R :
SU(2)R : N I UI J N J ,
(2.7)
with U SU(2)R . A possible choice of parity assignments is

1 (y) = +1 (y).
(2.8)
Again, at the point y = 0, the other fields parity transformations are determined from equations
(2.6), (2.8) and invariance of (B.8) under the mapping (2.5), and they are shown in Table 1.
As periodicity condition, we impose the following twisted boundary conditions:

M1 (y + 2R)
cos(2) sin(2)
M1 (y)
(2.9)
=
,
M2 (y + 2R)
M2 (y)
sin(2) cos(2)
which correspond for = 0 to implement a ScherkSchwarz supersymmetry breaking in the
bulk [16]. In Section 4, it will be shown that boundary localized masses for gravitinos can be
absorbed in a generalized ScherkSchwarz twist.
We must assign a parity P for each generic field at the point y = R
(R + y) = P (R y),
(2.10)
which keeps the Lagrangian (B.4) and the supersymmetry transformations (B.8) invariant.
For instance, taking for the fnfbein the parities,
a
a
e
(R y) = +e
(R + y),
5
5
(R y) = e
(R + y),
e
e5a (R y) = e5a (R + y),
e55 (R y) = +e55 (R + y),
(2.11)
an agreement with Table 1 and Eq. (2.9) requires imposing:

M+ (R y) = M+ (R + y),
M (R y) = M (R + y),
(2.12)
where:
+ = cos()1 sin()2 ,
5+ = sin()51 + cos()52 ,
= sin()1 + cos()2 ,
5 = cos()51 sin()52 .
(2.13)
130
Table 2
Parity assignments for bulk fields at y = R
P = +1
a
e
e55
B5
5+
P = 1
e5a
5
e
Invariance of the supersymmetry transformations (B.8) under the Z2 mapping (2.10) determines the parities of all other fields. The result is given in Table 2, where the following definitions
have been introduced:
+ = cos()1 sin()2 ,
= sin()1 + cos()2 .
(2.14)
2.2. Boundary branes actions

The bulk supergravity fields presented above are coupled with matter fields living on branes.
Here, we will consider the simplest case where the branes are localized on the boundaries yb =
0, R, with the simplest matter content given by Nb chiral multiplets. The case with many branes
localized on different points of S 1 /Z2 will be discussed in Section 5.
Each chiral multiplet contains (on-shell) a scalar bi and a fermionic bi fields (i = 1, . . . , Nb ).
These fields are coupled to the even parity bulk fields at the point y = yb . For instance, the even
a , e5 , B , ,
parity bulk fields at the point y = 0 are e
5
1
52 and 1 , and they appear in the
5
Lagrangian at the brane 0 as [39]:

1
1
j
j
L0 = e4 gij 0i 0 i gij 0 D 0i
2
2
1
j
j
+ G0j 0 G0j 0
1 1
8

2
1
j
j
eG0 /2 1 1 + i
G0j 0 1 + G0ij + G0i G0j ijk G0k 0i 0
2
2

2
1 G0 ij
j i
(2.15)
gij 0 0 1 e (g G0i G0j 3) + h.c. ,
2
2
where
1
1
j
j
j
D 0i = 0i + ab ab 0i + jik 0 0k G0j 0 G0j 0 0i
(2.16)
2
4
and G0 (0 , 0 ) is a hermitian function of the fields 0 and 0 . Here we have used the notations:
G0j =
G ,
j 0
G0j =
G ,
j 0
G0ij =
2
j
0i 0
G0 .
(2.17)
We remind that the metric gij , its inverse g ij and the Christoffel symbols in the Khler manifold
are given by:
gij =
2
j
0i 0
G0 ,
gij g j
= ik ,
ijk = g kl
0i
gj l .
(2.18)
131
The function G0 (0 , 0 ) is given, in terms of the Khler potential K0 and superpotential W0 at

the brane 0, by:

G0 (0 , 0 ) = K0 (0 , 0 ) + ln W0 (0 ) + ln W0 (0 ) .
(2.19)
In the following, it will be useful to define the action4 :

1
(0)
= d 4 x e4 R(
S4d
)
+ e4 1 D 1 + L0 ,
2
(2.20)
)
where R(
and D 1 are defined in Eqs. (A.21) and (A.22), respectively. This action is invariant under the four-dimensional local transformations:

a
= i 1 a 1 + h.c.,
e
0i = 21 0i ,
0i = i 2 1 0i 2eG0 /2 g ij G0j 1 ,
1
j
j
1 = 2D 1 + G0j 0 G0j 0 1 + ieG0 /2 1 .
(2.21)
2
It is important, for our concern, to note the presence of a localized mass for the gravitino 1
in the Lagrangian (2.15). If

ci g ij G0j is nonzero, then the field ci 0i is the Goldstino associated
with the supersymmetry breaking in the brane 0 as indicated by its non-linear transformation in
Eq. (2.21).
For the N chiral multiplets i , i (i = 1, . . . , N ) at the brane a similar discussion can
be carried over after the following substitutions:

0i i , 0i i ,
G0 G ,
L0 L ,
brane 0 brane :
(2.22)
K0 K , W0 W .
1 + , 1 + ,
3. Coupling the branes to the bulk
Our aim is to study generic configurations where in additional to a possible bulk Scherk
Schwarz mechanism, supersymmetry can also be spontaneously broken through non-vanishing
boundary F -terms for chiral multiplets.
To make the supersymmetry breaking manifestly spontaneous, we will keep the brane action
written as above in terms of the Khler functions Gb . The supersymmetry breaking terms can be
identified with the vacuum expectation values of the auxiliary fields. Coupling of these v.e.v.s to
the bulk supergravity requires then to add new auxiliary fields to the on-shell five-dimensional
supergravity action written in (2.2). This local case version is analogous to the case of spontaneous breaking of global supersymmetry as studied by Mirabelli and Peskin [17], where in order
to keep the rigid supersymmetry manifest, it was necessary to introduce auxiliary fields in the
bulk. Here we will present a partially off-shell extension of the bulk supergravity with only
the minimal required auxiliary fields. These new fields vanish identically in the supersymmetric limit as their boundary values are proportional to the supersymmetry breaking v.e.v.s (see
Eqs. (3.5) and (3.6)). Moreover, integrating these auxiliary fields to go on-shell leads to singular
terms ((0)), again as in [17], which will require careful summation over the KK bulk states in
order to extract sensible results.
4 We take F 5 = 0 on the branes, for simplicity.
132
3.1. The auxiliary fields action

For our purpose, we introduce two auxiliary fields denoted as u and vM . Here u is a real scalar
field and vM is a real five-dimensional vector field.
The bulk supergravity is now written as:
LBULK = LSUGRA + LAUX ,
(3.1)
where LSUGRA is still given in Eq. (B.4) and LAUX is:

LAUX = e5

1
uu + vM v M .
2
(3.2)
Taking into account the auxiliary fields and the brane supersymmetry breaking v.e.v.s, the
bulk supersymmetry transformations of the on-shell fields become:
A
A
= e M
,
eM
BM = BM ,
1 = 1 + iv 1 + iu 1 ,
2 = 2 + iv 2 + iu 2 ,
51 = 51 4eG /2 sin()+ (y R),
52 = 52 4eG0 /2 1 (y) 4eG /2 cos()+ (y R),
(3.3)
where the supersymmetry transformations were defined in Eq. (B.8).

Both u and v are taken to be even under Z2 on both boundaries:
u(y) = u(y),
v (y) = v (y),
u(R + y) = u(R y),

v (R + y) = v (R y).
(3.4)
They also obey the boundary conditions at y = 0 and y = R:

u|y=0 = eG0 /2 ,
u|y=R = eG /2 ,

1
j
j
(3.5)
F 5 y=0 = 0,
G0j 0 G0j 0 ,
2

1
F 5 y=R = 0, (3.6)
iv |y=R = Gj j Gj j ,
2
iv |y=0 =
a and
which allow matching the supersymmetry transformations for e
1 in the brane 0 (given
by Eq. (2.21)), from one side, and the supersymmetry transformations induced by the bulk (given
by Eq. (3.3) calculated at y = 0) from the other side. A similar result is obtained for the brane
after taking into account the substitutions (2.22).
3.2. The auxiliary fields supersymmetry transformations

We will determine here the supersymmetry transformations of the auxiliary fields u and vM
introduced above. They will be chosen such as to keep the full action invariant.
On one side, under the modified transformations given in Eq. (3.3) the bulk supergravity action
transforms as:
133

LSUGRA
LSUGRA
d xLSUGRA = d x i
DN
(v J + u J ) + h.c.
J
(DN J )

d 4 x e4 8eG0 /2 1 D 1 + h.c. y=0

d 4 x e4 8eG /2 + D + + h.c. y=R ,
(3.7)
where the surface terms in Eq. (3.7) come from the terms proportional to (y) and (y R)
in the modified supersymmetry transformation laws for 51 and 52 . For the sake of keeping
compact formulae, we have not explicitly written the variation with respect to the gravitinos.
On the other side Eq. (3.2) and supersymmetry transformations (B.8) lead to

1
d 5 x LAUX = d 5 x e5
uu + vM v M i1 1 + i2 2 + 2 51
2

1 52 + h.c. + u u + vM v M .
(3.8)
3.2.1. Canceling the bulk terms
We must impose transformations laws for the auxiliary fields in such a way that the bulk
variations in (3.7) and (3.8) cancel each other. This is achieved by taking:

1
u = u i1 1 + i2 2 + 2 51 1 52
2

i
LSUGRA
LSUGRA
+
J DN
+ h.c.,
J
e5
J
(DN J )

1
v = v i1 1 + i2 2 + 2 51 1 52
2

i
LSUGRA
LSUGRA
J D N
+ h.c.,
J
e5
J
(DN J )

1
v5 = v5 i1 1 + i2 2 + 2 51 1 52 + h.c.
(3.9)
2
Using Eqs. (3.7) and (3.8) it is easy to check that

d 5 x LSUGRA + d 5 x LAUX = d 4 x e4 8eG0 /2 1 D 1 + h.c. y=0

d 4 x e4 8eG /2 + D + + h.c. y=R .
(3.10)
3.2.2. Canceling the boundary terms
The above bulk variations need to completed to include the variation of the boundary brane
actions. This will determine the final modification of the transformations laws for the
auxiliary fields that make the full action invariant.
To calculate the variation of the brane action under the supersymmetry transformations one
could simply plug (3.3) in (2.15). This is straightforward but quite long and tedious. Here we
exhibit a trick that permits one to find the variation of the brane action in a much shorter way.
134
Invariance of the action (2.20) under the supersymmetry transformations (2.21) implies

d 4 x L0 = SMinimal Sugra

SMinimal Sugra =

1

d x e4 R()
1 D 1 .
+ e4
2
4
(3.11)
Note that the action SMinimal Sugra is invariant under the following supersymmetry transformations:

a
M.S. e
= i 1 a 1 + h.c.,
M.S. 1 = 2D 1 ,
(3.12)
where D 1 is given by Eq. (A.22), which makes it easy to calculate the supersymmetry variation
of the brane action:

d 4 x L0 =
1
j
j
G0j 0 G0j 0 1 D 1

2

+ 4eG0 /2 1 D 1 + h.c. .
d 4 x e4
(3.13)
Given the parity assignments for the fnfbein of Table 1, Eqs. (A.20) and (A.22) imply
D = D on the boundary at y = 0. The boundary conditions (3.5) leads then to:

d 4 x L0 =

d 4 x e4 i v 1 D 1 + 4eG0 /2 1 D 1 + h.c. y=0 . (3.14)
The same analysis can be made for the brane through the substitutions (2.22) in the above
formulae. The result is:

d 4 x L =

d 4 x e4 i v + D +

+ 4eG /2 + D + + h.c. y=R .
(3.15)
To achieve a fully invariant bulk plus branes action the auxiliary fields transformations are
modified as follow:
u = u,
v = v + c0 (y) + c (y R),
v5 = v5 .
(3.16)
The coefficients c0 and c are determined by putting together the different pieces of the variation of the total action given in (3.10), (3.14) and (3.15) to find (at first order in c0 and c ):

S =
135

d x e4 i v 1 D 1 + h.c. y=0

+ e4 i v + D + + h.c. y=R

1 5
1 5

+ e4 e5 v c0 y=0 + e4 e5 v c y=R .
2
2
4
(3.17)
It is easy to check that if we take:

c0 = 2ie5 1 D 1 + h.c.,
c = 2ie5 + D + + h.c.,
(3.18)
the total action variation is zero to first order in c0 and c . Note that the expressions for c0 and
c are quadratic in the spinor fields, so within our approximation, where the four-fermion terms
in the Lagrangians are dropped, we have S = 0.
The bulk plus brane action (2.1) is then invariant if we use the transformations (3.3) and
(3.16), the parity assignments of Tables 1 and 2, as well as the boundary conditions (3.5) and
(3.6). These results are summarized in Appendix B for future reference.
3.3. Discontinuity of spinor bulk fields at the boundaries
An important consequence of the presence gravitino masses localized on the branes is the
appearance of wave functions discontinuities (see [28]). We provide below a straightforward
generalization for the case = 0.
The equations of motion for the gravitinos 1 can be seen from the Lagrangians (B.6) and
(B.10) to take the form5 :

1
1 + 5 2 2 eG0 /2 1 (y)
2

2 eG /2 cos() + (y R) + = 0,

1
2 5 1 + 2 eG /2 sin() + (y R) + = 0, (3.19)
2
where stands for terms which involve other fields that couple to the gravitinos and that we
drop for the purpose of our discussion.6 The four-dimensional equation of motion for gravitinos
I of mass m3/2 :
I = 2m3/2 I ,
leads then to the following equations:

5 2 + m3/2 1 = 2 eG0 /2 1 (y) + 2 eG /2 cos()+ (y R),

5 1 m3/2 2 = 2 eG /2 sin()+ (y R).
(3.20)
(3.21)
It can be clearly seen from Eq. (3.21) that 1 is a continuous field near the point y = 0 and
+ is a continuous field near the point y = R. In contrast, 2 has a jump at the point y = 0
5 We assume e5 = 1.
5
6 This amounts to keep the quadratic terms and consider the interaction terms as perturbations.
136
while has a jump at the point y = R, their first derivative being proportional to a Dirac
distribution (see Fig. 1).
More precisely, integration of Eq. (3.21) near the points y = 0 and y = R, taking into account the parity assignments of Tables 4 and 5, leads to the following discontinuities of the odd
gravitinos wave functions:

lim 2 (y) = 2 (0+ ) = eG0 /2 1 (0) = 2 (0 ),
y0, y>0
lim
yR, y<R

(y) = (R ) = eG /2 + (R) = (R + ).
(3.22)
From the transformations of the gravitinos I in Eq. (B.15) it is easy to see that (3.22) lead
to the following boundary conditions for the supersymmetry parameters:

2 (0+ ) = eG0 /2 1 (0) = 2 (0 ),

(R ) = eG /2 + (R) = (R + ).
(3.23)
Interesting to observe is that these boundary conditions insure that the modified transformations
of 5I are non-singular: the terms proportional to (y) and (y R) cancel with those coming
from the derivatives 5 2 near y = 0 and 5 near y = R.
We end this section by a comment on the relation between the so-called orbifold approach
(used here) and the interval approach. We do not seem to bother about boundary terms that arise
after integration by parts along the fifth dimension, while it is a central issue in the interval approach. Here, we illustrate, through an example, how the previous construction can be understood
in the interval approach.

In order to perform the variation of d 5 x LSUGRA in (3.7) we integrated by parts in the y
direction. If the odd gravitino fields are allowed to be discontinuous in the branes the wave
function 2 (0+ ) and (R ) may be nonzero. So, in the interval approach, one should care
about the following surface terms in d 5 x LSUGRA :

d 5 x LSUGRA Surface Terms

=
y=R
LSUGRA
d x i
.
(v J + u J ) + h.c.
(D5 J )
y=0+
4
The Lagrangian (B.6) leads to

d 5 x LSUGRA
(3.24)

d 4 x e4 + (v + u )

(v + + u + ) + h.c. y=R

i d 4 x e4 1 (v 2 + u 2 )

2 (v 1 + u 1 ) + h.c. y=0+
=i
Surface Terms
and the boundary conditions (3.22) and (3.23) imply:

d 5 x LSUGRA Surface Terms = 0.
(3.25)
(3.26)
137
Fig. 1. The two bulk gravitinos wave functions along the compact dimension. The discontinuities are due to the presence
of brane localized masses for the other gravitino component. In this example = 1/2.
4. Inclusion of a generalized ScherkSchwarz mechanism

An important issue is the relation between bulk and brane localized gravitino masses and the
twists in ScherkSchwarz compactifications. This section collects a few results. Most of them, if
not all, are probably known, but we rederived them as they will be useful in the rest of the paper.
It also introduces some notations.
It is often useful to work in a basis of periodic fields MI (i.e., MI (x, y + 2R) =
MI (x, y)) in contrast to the multi-valued MI used up to now. These are related by the rotation:

cos[f (y)] sin[f (y)]
M1
M1
=
.
(4.1)
sin[f (y)] cos[f (y)]
M2
M2
The function f (y) must obey f (y + 2R) = f (y) + 2 . Here we follow [28] and take:
f (y) =
B
0
0 +
y+
(y) +
(y),
R
2
2
(4.2)
with B + 0 + = . (y) is the sign function on S 1 :

(y) = +1,
2kR < y < (2k + 1)R,
k Z,
(y) = 1,
(2k 1)R < y < 2kR,
kZ
(4.3)
and (y) is the staircase function:

(y) = 2k + 1,
kR < y < (k + 1)R,
k Z.
(4.4)
The supersymmetry breaking mass terms for the gravitinos is then manifest as we perform
this fields transformation in the kinetic terms of the Lagrangian (B.6) to give:

1
LKinetic = e5 1 D 1 + 2 D 2
2

+ e5 1 D5 2 2 D5 1
5
138

2e5 51 D 2 52 D 1

5
B
+ 20 (y) + 2 (y R)
R

e5 1 1 + 2 2 + h.c. .
5
(4.5)
The localized mass terms in (4.5) imply discontinuities for the gravitino wave functions. They
are too singular to apply the variational principle without regularization. In reference [28] it was
shown that the Lagrangian density (4.5) is equivalent to the action:
2R
SKinetic =
dy

d 4x

1
1
e5 1 D 1 + 2 D 2
2
2

+ e5 1 D5 2 2 D5 1
5

2e5 51 D 2 52 D 1

5

B 5
e 1 1 + 2 2
5
R

tan(0 )(y) + tan( )(y R) e5 1 1 + h.c. ,
5
(4.6)
with the fields now being piece-wise smooth.

In order to study the supersymmetry transformations of the fields MI it is convenient to
regularize the field rotation (4.1) by introducing a regularized function freg (y) instead of the discontinuous function f (y). The continuous function freg (y) obeys: freg () = 0 , freg (0) = 0,
freg () = 0 , freg (R ) = 0 + B , freg (R) = 0 + + B , freg (R + ) =
0 + 2 + B . To get the final results it suffices to take the limit 0 in the desired
expression.
Going to the new basis requires then the following redefinition for the supersymmetry transformation parameters,

1
cos[freg (y)] sin[freg (y)]
1
(4.7)
=
,
2
sin[freg (y)] cos[freg (y)]
2
and the supersymmetry transformations (B.15) take now the form:
1 = 2D 1 + iv 1 + iu 1 + ,
= 2D + iv + iu + ,
2
dfreg
51 = 2D5 1 + 2
2 + ,
dy
dfreg
52 = 2D5 2 2
1 4eG0 /2 1 (y) 4eG /2 1 (y R) + ,
dy
(4.8)
where stand for terms which are proportional to F MN .

It is important to note that the fields 51 and 52 transforms non-linearly under supersymmetry transformations: they are the Goldstino fields associated with the supersymmetry breaking in
the bulk, as expected.
139
The new brane field transformations and boundary conditions can be easily obtained after
noticing that these redefinitions (4.1) and (4.7) imply for the brane at y = R:
+ (R) = 1 (R),
(R) = 2 (R),
+ (R) = 1 (R),
(R) = 2 (R).
(4.9)
Of main interest in the generalized ScherkSchwarz mechanism is the interplay between bulk
and brane localized gravitino mass terms in order to keep or break supersymmetry. Given that we
have explicitly obtained all the supersymmetry transformations, it is very easy to us to answer
this question by looking for Killing spinors. Explicitly, we consider the supersymmetry transformations (4.8) evaluated with the appropriate background and search for spinors I which obey
MI = 0. The interesting equations arise from 51 and 52 :
dfreg
2 = 0,
dy

dfreg
5 2
1 = 2 eG0 /2 1 (y) + 2 eG /2 1 (y R).
dy
The parity transformation assignments of Tables 4, 5 and Eq. (4.7) imply:
5 1 +
1 (y) = +1 (y),
2 (y) = 2 (y),
(4.10)
1 (R y) = +1 (R + y),
2 (R y) = 2 (R + y).
(4.11)
Integrating Eq. (4.10) at y = 0 and y = R and taking into account (4.11) we deduce that 1 is
a continuous field near the points y = 0 and y = R and that 2 has a jump at the points y = 0
and y = R:

2 (0+ ) = eG0 /2 1 (0),

2 (R ) = eG /2 1 (R).
(4.12)
The solutions for the Killing equations (4.10) in the interval 0 < y < R with the first boundary condition in (4.12) are given by:

1 (y) = 1 (0) cos freg (y) eG0 /2 sin freg (y) ,

2 (y) = 1 (0) sin freg (y) + eG0 /2 cos freg (y) .
(4.13)
The second boundary condition in (4.12) leads to the following relation:

eG0 /2 +
eG /2
= tan freg (R)
G
/2
G
/2
e

e
1

0 + + B + arctan eG0 /2 + arctan eG /2 = n
It is sometimes useful to introduce the angles b (b = 0, ) defined by
Eq. (4.14) takes the simple form:
tan( + 0 + ) = 0.
(n Z).
eGb /2 = tan
(4.14)
b.
Then,
(4.15)
Eq. (4.14) is one condition that indicates when supersymmetry is not spontaneously broken, other
conditions are obtained by studding the supersymmetry transformations of the fields 0 and
in the branes. They imply N0 + N extra conditions for the existence of Killing spinors:
G /2
G /2

e 0 G0j = 0,
(4.16)
e G j = 0.
140
5. Suspending branes in the bulk

In this section, we generalize the previous results for the case with multiple branes. More
precisely, we consider N + 1 branes placed at the points y = yn , n = 0 N with y0 = 0, yN =
R, and yn < yn+1 . The total action is given by:
2R
S=
dy

N

1
d x LBULK +
Ln (y yn ) .
2
4
(5.1)
n=0
Every brane n will be characterized by the choice of the bulk fields, in particular the gravitino, that couple to its worldvolume. These are in fact determined as being the even fields under
the Z2 action at the point y = yn :
even (yn + y) = Pn even (yn y) = even (yn y).
(5.2)
We adopt the parity transformations shown in Table 3, where the following definitions have been
introduced:
n
+
= cos(n )1 sin(n )2 ,
n
= sin(n )1 + cos(n )2 ,
n
5+
= sin(n )51 + cos(n )52 ,
n
5
= cos(n )51 sin(n )52 ,
+n = cos(n )1 sin(n )2 ,
n = sin(n )1 + cos(n )2 .
(5.3)
The case of boundary branes discussed in previous sections corresponds to 0 = 0 and N = .

The Lagrangian density and supersymmetry transformations for the worldvolume fields living
on the brane n are given by Eqs. (2.15) and (2.21) after the substitutions:

L 0 Ln ,
0i ni , 0i ni ,
G0 Gn ,
brane 0 brane n:
(5.4)
n
n
1 + , 1 + ,
K0 Kn , W0 Wn .
The bulk Lagrangian density is given as before by Eq. (B.2), with supersymmetry transformations given by:
A
A
= e M
,
eM
BM = BM ,
1 = 1 + iv 1 + iu 1 ,
2 = 2 + iv 2 + iu 2 ,
Table 3
Parity assignments for bulk fields at y = yn
Pn = +1
a
e
Pn = 1
e5a
e55
5
e
B5
n
+
n
5+
n
+
n
5
v5
51 = 51 4
N

141
eGn /2 sin(n )+n (y yn ),
n=0
52 = 52 4
N

eGn /2 cos(n )+n (y yn ),
n=0
u = u,
v = v 2i
N

n
e5 +n D +
+ h.c. (y yn ),
5
n=0
v5 = v5 ,
(5.5)
where the transformations are given in Eqs. (B.8) and (3.9).

We also impose boundary conditions at y = yn , these are given by the obvious generalization
of (3.5) and (3.6). With these boundary conditions and the parity assignments of Table 3 the
action (5.1) is invariant under the transformations (5.5).
Consider the case of localized gravitino masses Mn that include the branes F -terms and generalized ScherkSchwarz contribution written in (4.6). Repeating the analysis of Section 3.3 for
n is continuous at the point y = y
the gravitinos equations of motion shows that the field +
n
n
while the field has a jump at this point:
lim
yyn ,y>yn
n
n
n
n
=
(yn+ ) = Mn +
(yn ) =
(yn ).
(5.6)
The corresponding boundary conditions for the supersymmetry transformation parameters at the
point y = yn are:
n (yn+ ) = Mn +n (yn ) = n (yn ).
(5.7)
Supersymmetry can remain unbroken for a peculiar choice of localized and bulk gravitino
masses, following the same lines as Section 4. Again the equations of interest arise from requiring
51 = 0 and 52 = 0:
5 1 2
N

Mn sin(n )+n (y yn ) = 0,
n=0
5 2 2
N

Mn cos(n )+n (y yn ) = 0.
(5.8)
n=0
Integrating Eq. (5.8) near y = yn , taking into account the parity assignments of Table 3, shows
that +n are continuous fields at y = yn while n have jumps at these points given by (5.7).
The solution of Eq. (5.8) in the interval yn < y < yn+1 with the condition (5.7) can be written
as:
1 (y) = 1 |y=yn+ ,

Mn cos(n ) sin(n )
2 (y) =
= tan arctan(Mn ) n 1 y=y + ,
1
n
cos(n ) + Mn sin(n ) y=yn+
(5.9)
then using (5.7) evaluated at yn+1 , gives the following conditions:

n+1 n + arctan(Mn ) + arctan(Mn+1 ) = k
(k Z).
(5.10)
142
These N conditions generalize Eq. (4.14) for the multi-brane case. When one of the relations (5.10) is not satisfied, the Killing spinor equations have no solution and supersymmetry
is spontaneously broken in the bulk by a non-trivial ScherkSchwarz twist. The other necessary
conditions for supersymmetry not to be spontaneously broken are the direct generalization of
(4.16).
6. The super-Higgs mechanism
In Section 4, we studied the supersymmetry breaking induced by non-periodic boundary conditions for the gravitinos. Here, we turn our attention to the F -terms of chiral multiplets living
on the branes worldvolume. More precisely, we will determine the condition for supersymmetry
breaking and study the super-Higgs effect associated.
We will perform our study in the simplest case with no branes in the bulk other than the
boundary ones at y = 0 and y = R, as it contains all the qualitative features. Eqs. (2.21) and
(4.8) show that four fields 51 , 52 , 0 and transform non-linearly under supersymmetry
transformations. These are the local would be Goldstinos associated with breaking of supersymmetry in the bulk and in the two branes respectively. As we have two gravitinos then two
local would be Goldstinos will be absorbed in the super-Higgs effect to give mass to the gravitino fields 1 and 2 , while two linear combinations of the fields 51 , 52 , 0 and remain
as pseudo-Goldstinos.
To keep the formulae explicit, we will make a number of simplifications:
We impose a zero tree level cosmological constant at each brane. This implies that the vacuum expectation values of the bosonic fields are:

g ij G0i G0j = 3,
ij

g Gi Gj = 3,

g ij G0j G0ki kil G0l + G0k = 0,

ij

g Gj Gki kil Gl + Gk = 0,
(6.1)
the second and fourth equalities in Eq. (6.1) come from the extremisation of the scalar potential at the branes 0 and . In Appendix C one explicit example of Khler function which
satisfies (6.1) is presented.
We will consider that the boundaries gravitino masses arise through explicit F -terms for the
boundary supermultiplets as

Mb = eGb /2
with b {0, },
(6.2)
while the terms tan(b ) arise from a generalized ScherkSchwarz mechanism and are ab
= .
sorbed by redefining the bulk twist B by B initial B = B initial + 0 +
We adopt the following notation:

1
0 =
G0i 0i ,
3
1
=
Gi i
3
and we assume that the kinetic terms are canonically normalized: gij = ij + .
From now on, we drop the overscript over the fields defined in (4.1).
(6.3)
143
In order to study the super-Higgs effect we will concentrate on the bilinear terms of the fermionic fields: 1 , 2 , 51 , 52 , 0 and . These can be extracted from Eqs. (2.15) and (4.5)
and they take the form:

1 1
L=
( 1 1 + 2 2 ) + 1 5 2 2 5 1

2 2

+ 2 52 1 51 2
1 1 + 2 2
R

i
6
+ (y) 0 0 M0 1 1 + i
0 1 + 0 0
2
2

i
6
+ (y R) M 1 1 + i
1 +
2
2
+ h.c.
(6.4)
6.1. R gauge
Here we will use the analogous of R gauges of non-Abelian gauge theories. This kind of
gauge fixing in supergravity theories was first discussed in [40]. Our discussion follows and
generalizes the simpler case of pure ScherkSchwarz breaking studied in [31].
Some field redefinitions allow obtaining standard kinetic terms for the fields 5I :
i
1 1 + 52 ,
6
i
2 2 51 ,
6
2
51 51 ,
6
2
52 52 ,
6
leading to the Lagrangian density:

1 1
(1 1 + 2 2 ) + 1 5 2 2 5 1

L=
2 2

i
51 51 + 52 52 + 51 5 52 52 5 51
2

1 1 + 2 2 + 51 51 + 52 52
R

6

i
5 51 1 + 5 52 2 +
52 1 51 2
2
R
i
6
+ (y) 0 0 M0 1 1 + i
( 0 + 52 ) 1
2
2

i
+ (0 + 52 )(0 + 52 ) + (y R)
2
(6.5)
144
6
M 1 1 + i
( + 52 ) 1
2

+ ( + 52 )( + 52 ) + h.c.,
(6.6)
instead of (6.4).
The gauge choice is made by the addition to the Lagrangian density of the R gauge fixing
term:
LGF =

i
h1 h1 + h 2 h2 ,
2
(6.7)
where
6
h1 = 1
2 g 1 ,
2
6
2 g 2 ,
h2 = 2
2
(6.8)
with
52 + 2(y)M0 (0 + 52 ) + 2(y R)M ( + 52 ),

R
g2 = 5 52 51
R
g1 = 5 51 +
(6.9)
and is a free constant gauge parameter.

It is straightforward to check that this gauge fixing term provides the cancellation of mixing
terms between gravitino and Goldstino fields, which is the aim of our gauge choice:

1
1
1 1 ( 1 1 + 2 2 )
L + LGF =
2
2

i
+ 1 5 2 2 5 1 51 51 + 52 52
2

+ 51 5 52 52 5 51
1 1 + 2 2
R

i
+ 51 51 + 52 52 + (y) 0 0 M0 1 1
2

i
+ (0 + 52 )(0 + 52 ) + (y R)
2

M 1 1 + ( + 52 )( + 52 )

3
i g1 2 g 1 + g2 2 g 2 + h.c.
8
(6.10)
As expected the position of the poles in the propagators of the fields MI , 0 and will depend
on the gauge parameter , but of course the gauge invariant operators and S-matrix elements
should not depend on the parameter .
145
6.2. Unitary gauge

The unitary gauge can be recovered from the R gauge in the limit . In this gauge, the
gravitino propagators have poles at their physical mass and the unphysical degrees of freedom
(would-be-Goldstinos) are eliminated, absorbed to provide the longitudinal components for the
gravitinos, through the super-Higgs mechanism.
We first discuss the gravitino equations of motion in the bulk-branes system. The equations
of motion for the gravitinos I (y) in the unitary gauge can be extracted from the Lagrangian
(6.10) in the limit :
1
1 + 5 2 1
2
R
= 2M0 1 (y) + 2M 1 (y R),
1
2 5 1 2 = 0.
2
R
Assuming the gravitinos have a four-dimensional mass m3/2 :
I = 2m3/2 I ,
(6.11)
(6.12)
their equations of motion can take the form:

5 2 + m3/2
1 = 2M0 1 (y) + 2M 1 (y R),
R

2 = 0.
5 1 m3/2
R
(6.13)
Integration of Eq. (6.13) near the points y = 0 and y = R, taking into account the parity assumptions, leads to the following expressions for the discontinuities of the odd gravitino fields:
2 (0+ ) = M0 1 (0) = 2 (0 ),
2 (R ) = M 1 (R) = 2 (R + ).
(6.14)
It is then straightforward to find a solution for Eq. (6.13) in the interval 0 < y < R satisfying
the first condition in (6.14):

1 (y) = cos m3/2

y + M0 sin m3/2
y 1 (0),
R
R

y sin m3/2
y 1 (0).
2 (y) = M0 cos m3/2
(6.15)
R
R
The second condition in (6.14) is then used to determine the gravitino mass:
m3/2 =
1
+
arctan(M0 ) + arctan(M ) + ,
R R
R
n Z.
(6.16)
In remaining of the of this section we will concentrate on the would-be Goldstino fields
51 (y), 52 (y), 0 and . Note that the Lagrangian density (6.10) shows that, in the unitary
gauge , a stationary action (in order to derive of the equations of motion) is possible if
g1 = g2 = 0, i.e.:
146
52 = 2(y)M0 (0 + 52 ) 2(y R)M ( + 52 ),

R
5 52 51 = 0,
R
which imply that the fields 5I (y), in the interval 0 < y < R can be written as:

1
y + 1 + sin
y + 2 ,
51 (y) =
cos
R
R
R

1
y + 1 cos
y + 2 ,
52 (y) =
sin
R
R
R
5 51 +
(6.17)
(6.18)
where 1 and 2 are y independent 4d spinors and is a constant which corresponds to a choice
of basis for 1 and 2 .
Integrating Eq. (6.17) near y = 0 and y = we deduce that:

51 (0+ ) + M0 0 + 52 (0) = 0,

51 (R ) M + 52 (R) = 0,
(6.19)
which implies (for M = 0 and M0 = 0):

1
1
cos( + ) 1
=
sin( + ) +
M
R

1
1
+
sin( + ) 2 ,
cos( + ) +
M
R

1
1
1
1
cos( ) 1 +
sin( ) 2 .
0 =
(6.20)
sin( ) +
cos( )
M0
M0
R
R
Here we see how the super-Higgs mechanism operate, from the original two 5d and two 4d
degrees of freedom (51 (y), 52 (y), 0 and ), an infinity of KaluzaKlein modes is absorbed
to give mass to the fields 1 (y) and 2 (y) and only two degrees of freedom remain in the
unitary gauge: the pseudo-Goldstinos 1 and 2 .
6.3. Comment on F-terms versus generalized ScherkSchwarz mechanism
Eq. (6.16) raises questions about the possibility to express spontaneous breaking with F terms (and all the gravitinos and pseudo-Goldstinos masses generated) as a generalized Scherk
Schwarz twist, in parallel to the case of b in (4.14). This is not possible, as can be seen by the
following arguments.
In order to have an equivalence between the brane mass terms and a generalized Scherk
Schwarz twist one should be able to expresses the discontinuity of the fields 5I at y = 0 and
y = R as an SU(2)R rotation like in (4.1). This means that in order to be associated with a
generalized ScherkSchwarz twist the effects of the brane mass terms must be described by a
generalized twist. So one should be able to find a rotation such that:

51 (0+ )
cos() sin()
51 (0 )
(6.21)
=
.
52 (0+ )
52 (0 )
sin() cos()
But Eq. (6.18) imply:
51 (0+ ) =
1
R

cos( )1 + sin( )2 = 51 (0 ),

1
sin( )1 cos( )2 = 52 (0 ).
52 (0+ ) =
R
147
(6.22)
Note then that matching Eqs. (6.21) and (6.22) for the coefficients of 1 one finds = 2 +
, while if one matches the coefficients for 2 in (6.21) and (6.22) one finds = 2 . This
incompatibility shows that Mb cannot be casted as tan(b ), as in (4.6).
7. The pseudo-Goldstinos spectrum
In the previous section we have shown how some would-be-Goldstinos are absorbed leading
to massive gravitinos. Here, we will discuss the spectrum of the remaining pseudo-Goldstinos.
More precisely, we will analyze some limits or approximations which allow to display compact
formulae. The general case is treated in Appendix D.
We will restore the explicit dependence on the (reduced) five-dimensional Planck mass M5 =
1 . It is related to the four-dimensional Planck mass M4 by7
RM53 = M42 .
(7.1)
The lightest four-dimensional gravitino mass can be read from (6.16):

m3/2 =
1
+
arctan(M0 ) + arctan(M ) ,
R R
(7.2)
where Mb =
e Gb /2 1 , with b {0, } arise from boundary F -terms. In the four-dimensional
limit Mb 1 the approximate gravitino mass is:
m3/2 +
(7.3)
(M0 + M ).
R R
To identify the pseudo-Goldstinos mass eigenstates we shall plug (6.18) and (6.20) in the
Lagrangian (6.10), integrate over the y dimension, diagonalize and canonically normalize the
kinetic terms of the fields 1 and 2 and finally diagonalize their mass matrix. This is a tedious
task, the resulting mass eigenstates are given in Appendix D. Let us discuss in more details some
particular cases.
2
7.1. Supersymmetry breaking on a single brane

Consider the case where the supersymmetry breaking is realized by a combination of a
ScherkSchwarz twist and a single F -term, say on the brane placed at y = 0. This corresponds
in our generic formulae to M = 0 and = 0. Choosing a basis for 1 and 2 corresponding to
= Eqs. (6.18) and (6.19) imply 1 = 0. So, as expected, only one degree of freedom 2
remains in the unitary gauge.
Substitution of (6.18) and (6.20) in (6.10), integration over y and redefinition of the fields
to canonically normalize their kinetic terms allows identifying the eigenstate, we denote as 1 ,
with mass:
m1 =
2M0 sin()[M0 cos() + sin()]

.
RM02 + [M0 cos() + sin()]2
(7.4)
7 We recall that in our conventions the four-dimensional Planck mass M is related to Newtons constant G by
4
8 G = M41 .
148
The original would-be-Goldstinos are written in terms of the pseudo-Goldstino as given by (6.18)
and (6.20) in the unitary gauge, which in the present case reads:

M0
y
sin
1 ,
51 (y) =
R
RM02 + [M0 cos() + sin()]2

M0
y
cos
1 ,
52 (y) =
R
RM02 + [M0 cos() + sin()]2
0 =
M0 cos() + sin()
RM02 + [M0 cos() + sin()]2
1 .
(7.5)
Let us discuss some particular cases which might bring to the reader some more intuition on
what is happening:
Case 0:
We first discuss the case of vanishing twist. Eq. (7.5) become:
2
0,
R +
51 (y) 0,
1
52 (y)
1 ,
1 R + 1
1
0
1 .
1 R + 1
m1
(7.6)
There are two ways to understand these results. First, from a global view, for = 0 the
Z2 projected out the odd zero mode of 51 , 51 being continuous this implies 51 = 0.
The other way is to consider a local five-dimensional description where the gravitino 2
eats the fermion with the same Z2 parity, i.e., 51 . The remaining gravitino 1 absorbs the
linear combination 52 (0) + 0 and reminds the orthogonal combination 52 (0) 0 1
as a pseudo-Goldstino. The only source of mass for this state is the bulk mass term.
Case R 1 :
This limit gives
2 sin()[M0 cos() + sin()]
,
RM0

1
y
51 (y)
1 ,
sin
R
1 R

1
y
1 ,
52 (y)
cos
R
1 R
M0 cos() + sin()
0
1 ,
RM0
m1
(7.7)
51
+
which agrees with the fact that the absorbed Goldstino on the brane y = 0 is given by ( M
0
+
52 + 0 )(0 ) and the one eaten at y = is 51 (R ).
149
Case = 12 :
m1
2M0
,
RM02 + 1
51 (y)

y
1 ,
cos
2R
2
RM + 1
M0

y
1 ,
sin
52 (y)
2R
RM02 + 1
M0
0
1
RM02 + 1
1 ,
(7.8)
in which case one notes that 52 decouples from the brane at y = 0 while the absorbed state
is 51 (0+ ) + M0 0 .
7.2. Hierarchical supersymmetry breaking on the boundaries
In this section we switch on a large supersymmetry breaking F -term in on brane at M , i.e.,
M M0 . Our results assume explicitly that = 0. They are not generically valid for = 0
which will be presented in Section 7.5. We will exhibit the first orders in an expansion in M0
for the pseudo-Goldstinos mass matrix eigenvalues and eigenvectors. At the leading order, this
is diagonal in the basis for 1 and 2 corresponding to = 0. The mass eigenstates are denoted
as 1 and 2 , and have masses given, respectively, by:
m1 =
m2 =
2 sin()[M cos() + sin()]

M0
sin()[M cos() + sin()] + [R + 2 3 sin()2 ]M0 M

+ O 2 M02 ,
2M sin()[M cos() + sin()]
+ O(M0 ).
RM2 + [M cos() + sin()]2
(7.9)
As in the previous section, the fields 51 (y), 52 (y), 0 and are written in terms of the
pseudo-Goldstinos as:

M
51 (y) =
sin
y 2 + O(M0 ),
R
RM2 + [M cos() + sin()]2

M

cos
y 2 + O(M0 ),
52 (y) =
R
RM2 + [M cos() + sin()]2
0 = 1 + O(M0 ),
M cos() + sin()
2 + O(M0 ).
=
RM2 + [M cos() + sin()]2
(7.10)
Note that the results obtained in Section 7.1 can be derived from these formulas by taking
M0 = 0 and interchanging the branes 0 and .
150
7.3. The 5D or large extra dimension radius limit

In this section we consider a very large extra-dimensional radius, R , RM0 1 and
RM 1, such that the set-up is truly five-dimensional. We will compute the pseudo-Goldstinos
mass eigenvalues and eigenvectors to leading order in a perturbation series in /R.
At leading order, the mass matrix is diagonal in a basis for 1 and 2 corresponding to an
angle given by:
tan(2 ) =
M0 M [1 cos(2)] M0 sin(2)
.
M + M0 cos(2) M0 M sin(2)
(7.11)
We choose in the range /4 < < /4. The mass eigenstates 1 and 2 have masses:

3/2
1
1
1
m1 =
+
+ +O
,
R M0 M
R 3/2

3/2
1
1
1
m2 =
+
+O
,
R M0 M
R 3/2
(7.12)
respectively, where

1
cos(2)
1
=
+
sin(2)
2
M
1 + [tan(2 )] M0

sin(2)
.
+ tan(2 ) cos(2)
M
(7.13)
The original would-be-Goldstinos 51 (y), 52 (y), 0 and are written in terms of the pseudoGoldstinos 1 and 2 as in (6.18) and (6.20), which read now:

cos
y + 1 + sin
y + 2 + O
,
51 (y) =
R
R
R
R

sin
y + 1 cos
y + 2 + O
,
52 (y) =
R
R
R
R

1
1
cos( ) 1 +
sin( ) 2 + O
sin( ) +
cos( )
,
0 =
R
M0
R
M0
R

1
cos( + ) 1
sin( + ) +
=
R
M

1
+
sin( + ) 2 + O
(7.14)
cos( + ) +
.
R
M
R
7.4. The 4D or small extra dimension radius limit
In this section, we discuss the four-dimensional limit corresponding to the case of a very small
extra-dimensional radius, RM0 1 and RM 1. At the leading order, the mass eigenstates
1 and 2 have masses given by:
(M0 + M ) sin() + 2M0 M cos() +

,
(M0 + M ) cos() ( 2 M0 M 1) sin()
(M0 + M ) sin() + 2M0 M cos()

,
m2 =
(M0 + M ) cos() ( 2 M0 M 1) sin()
151
m1 =
(7.15)
respectively, where now stands for

= (M0 M )2 sin()2 + 4(M0 M )2 .
(7.16)
Again, at the leading order, the four initial would-be-Goldstinos are expressed in terms of the
pseudo-Goldstinos 1 and 2 as:
51 (y)
M0 [sin( R
y ) M cos( R
y )]0 + M [sin( R
y) + M0 cos( R
y)]
=
,
1
(M0 + M ) cos() (M0 M ) sin()
52 (y)
M0 [cos( R
y ) + M sin( R
y )]0 M [cos( R
y) M0 sin( R
y)]
=
,
1
(M0 + M ) cos() (M0 M ) sin()
[(M0 M ) sin() + ]1 + 2M0 M 2

,
0 =
2[ + (M0 M ) sin() ]
2M0 M 1 + [(M0 M ) sin() + ]2

.
=
(7.17)
2[ + (M0 M ) sin() ]

7.5. No ScherkSchwarz twist
Another simple case corresponds to having the localized F -terms in the branes as the only
source of supersymmetry breaking, i.e. to consider a vanishing ScherkSchwarz twist, = 0.
The mass eigenstates 1 and 2 have masses respectively given by:
m1 = 0,
m2 =
(RM0 M
)2
2M0 M (M0 + M )(R + 2)

.
+ R(2 2 M02 M2 + M02 + M2 ) + (M0 + M )2
(7.18)
The would-be-Goldstinos 51 (y), 52 (y), 0 and are related to the pseudo-Goldstinos 1

and 2 through:
2 + RM0 M
51 (y) =
2 ,
(M M0 )
1
1 +
52 (y) =
2 ,
2 + 1 R

(M0 + M + 1 RM )
1
0 =
1 +
2 ,
2 + 1 R
(M0 + M + 1 RM0 )
1
=
(7.19)
1
2 ,
2 + 1 R
152
where

= (RM0 M )2 + R 2 2 M02 M2 + M02 + M2 + (M0 + M )2 .
(7.20)
Note that 51 (y) is proportional to M0 M . This is expected as 51 (y) is odd at both boundaries
and for = 0 would vanish if there were not both discontinuities at y = 0 and y = R due to M0
and M respectively. Note that one of the pseudo-Goldstinos is massless. This can be understood
from the following arguments. Generically, the pseudo-Goldstinos get masses from boundaries
and bulk. The brane masses are for the combination 0 + 52 (0) at y = 0 and + 52 (R) at
y = R, as seen from Eq. (6.10). In this case of = 0, both these combinations are proportional
to 51 (0+ ) = 51 (R ) 2 as seen from the unitary gauge condition (6.19). The orthogonal
combination, 1 , would have received a mass from the bulk, but this vanishes now as = 0.
Let us discuss some particular limits that connect this case to the previous ones:
Case M M0 :
In Section 7.2 we provided results for M M0 assuming = 0 and warned the reader
that they are not always valid when = 0. In fact, in this case the masses and the respective
eigenstates are given instead by8 :
m1 = 0,
2(R + 2)
M0 ,
m2
R +
M0 1 R + 2
2 ,
51 (y)
1 R + 1

1
1
1 +
2 ,
52 (y)
2 + 1 R
1 + 1 R

1
0
1 + 1 + 1 R2 ,
2 + 1 R

1
1
1
2 .

2 + 1 R
1 + 1 R
(7.21)
Note that if we take in these expression the large radius limit, i.e., with = 0, M M0
and R 1 1, the result is:
m1 = 0,
m2 2M0 ,
51 (y) M0 2 ,

1
1
1
1 +
2
1 ,
52 (y)
1
1
1
R
R
R

1
0
1 + 1 R2 2 ,
1 R

1
1
1
1
2
1 .

1
1
1
R
R
R
8 Here it is also assumed RM 1.
0
(7.22)
153
Case R 1 0:
Another simple limit is obtained by combining both = 0 and R 1 0, in which case
(7.18) and (7.19) lead to:
m1 = 0,
4M0 M
m2
,
M0 + M
2M0 M
51 (y)
2 ,
M 0 + M
1
M M0
52 (y) 1 +
2 ,
2
2(M0 + M )
1
1
0 1 + 2 ,
2
2
1
1
1 2 .
2
2
(7.23)
It is easy to check the agreement of (7.23) with the results presented in Section 7.4 in the
limit = 0. If we add M M0 , they become:
m1 = 0,
m2 4M0 ,
51 (y) 2M0 2 ,
1
52 (y) (1 + 2 ),
2
1
0 (1 + 2 ),
2
1
(1 2 ).
2
(7.24)
Acknowledgements
Work supported in part by the French ANR contracts BLAN05-0079-01 and PHYS@COL&
COS, and in part by the EU contract MRTN-CT-2004-005104.
Appendix A. Conventions
We use lower case letters from the middle of the Greek alphabet (, , , ) for the fourdimensional Minkowski indices ( = 0, . . . , 3) and lower case letters from the beginning of the
. . . , 3).
Capital indices are fiveLatin alphabet for the four-dimensional Lorentz indices (a = 0,
dimensional space indices: M, N, P , Q, R are five-dimensional coordinate space indices (M =
. . . , 3,
5).
0, . . . , 3, 5) and A, B, C, D, E are five-dimensional tangent space indices (A = 0,
Hated numbers (0, 1, 2, 3, 5) are used for tangent space indices. Indices I, J are used as SU(2)
indices (I = 1, 2) and i, j, k, l, i , j , k , l are Khler manifold indices (i = 1, . . . , N for N
chiral multiplets).
154
A and the vierbein ea allow to convert between coordinate space and tangent
The fnfbein eM
space indices:
A B
gMN = eM
eN AB ,
A
M = eM
A ,
a b
g = e
e ab ,
a
= e
a ,
(A.1)
their determinant is denoted:

A
a
e5 = det eM
,
e4 = det e
.
(A.2)
The five-dimensional gamma matrices obey the relations:

{A , B } = 2AB ,
AB = diag(1, 1, 1, 1, 1).
(A.3)
We use the following representation for the gamma matrices:

0 a
i 0
a
5
,
=
,
=
0 i
a 0
where the Pauli matrices are:

1 0
0 1
0
1
=
,
=
,
0 1
1 0
0 = 0 ,
1 = 1 ,
2 = 2 ,
(A.4)
i
0
0
i
3 = 3 .

,
1
0

0
,
1
(A.5)
The gamma matrices obey the following properties

ABCD = ABCDE E ,
ABC = ABCDE DE ,
ABCDE = ABCDE , (A.6)
where ABCDE is the completely antisymmetric tensor
01235 = +1,
M N P
R ABCDE
MNP QR = eA
eB eC eD eE

(A.7)
and

1
1
AB = AB = A , B .
2
4
From the representation (A.4) we find
ab

i
0
0
ab
a 5
=
,
=
0
ab
2 a
The charge conjugation matrix is:

2
0
i
C=
0 i 2
and obeys:
a T
C T = C,
= C a C 1 .
(A.8)
a
0

.
(A.9)
(A.10)
(A.11)
In the five-dimensional Lagrangians we use symplectic Majorana spinors I . We define:

I = I J J ,
I = I J J ,
(A.12)
where I J is the completely antisymmetric tensor: 12 = 21 = 1. The symplectic Majorana

spinors J obey the reality condition [41]:
J = J ,
(A.13)
155
where
J = J 0 ,
J = JT C.
(A.14)
We can express then in the two-component spinor notation9 as follows,

1 = 1 ,
2
(A.15)
where 1 and 2 are two-component Weyl spinors. Eq. (A.14) implies

1 = ( 2 ,
1 ) ,
1 = ( 1 ,
2 )
and the reality condition (A.13) gives:

2
1 = 2 = 1 ,
2 = 1 =
,
2
1
1 = 2 = ( 2 , 1 ) ,
2 = 1 = ( 1 , 2 ) ,
1 = 2 = ( 1 , 2 ) ,
2 = 1 = ( 2 , 1 ) .
(A.16)
(A.17)
The five-dimensional covariant derivative of a spinor is given by

1
DM J = M J + MAB AB J ,
2
the five-dimensional connection and curvature tensors are

1 P N
C
C
eB eMC [P eN]
eP C [N eM]
eN C [M ePC ] ,
MAB = eA
2
RMNAB = M N AB N MAB + MA C N BC N A C MBC ,
RMA = RMNAB eN B ,
R() = eMA RMA .
(A.18)
(A.19)
The five-dimensional covariant derivatives (Eq. (A.18)) expressed in the two-component

spinors notation are:
1
1
DM 1 = M 1 + Mab ab 1 + i Ma 5 a 2 ,
2
2
1
1
ab
DM 2 = M 2 + Mab 2 i Ma 5 a 1 .
2
2
The four-dimensional connection and curvature tensors are denoted:

1
c
c
c
,
ec [ e]
ec [ e]
ab = ea eb ec [ e]
2
R ab = ab ab + a c bc a c bc ,
)
R(
= ea R a ,
R a = R ab eb ,
(A.20)
(A.21)
and the four-dimensional covariant derivative of a spinor is denoted:

1
D = + ab ab .
2
9 Our conventions follow closely those of [39].
(A.22)
156
Appendix B. Supersymmetric action

The space considered here is five-dimensional with the fifth dimension compactified on the
S 1 /Z2 orbifold through the identification y y. Matter fields live on branes localized in the
boundaries yn = yb {0, R}. The total action is:
2R
S=
dy

d 4x

1
LBULK + L0 (y) + L (y R) .
2
(B.1)
5 = 0 on the following formulas.

For simplicity we fix e5a = 0 and e
The bulk fields are composed by the five-dimensional supergravity multiplet and some auxilA , the gravitinos
iary fields. The on-shell supergravity multiplet contains the fnfbein eM
MI and
the graviphoton BM .
The bulk Lagrangian density is given by:
LBULK = LSUGRA + LAUX .
(B.2)
Auxiliary fields are present in the off-shell part of the Lagrangian density,

1
uu + vM v M .
2
Here u is a real scalar and vM is a real five-dimensional vector field.
The on-shell part of the bulk Lagrangian density reads
LAUX = e5
(B.3)
LSUGRA = LBoson + LFermi ,
(B.4)
with the on-shell bosonic Lagrangian in the bulk given by

1
1
1
LBoson = e5 R() + FMN F MN + ABCDE FAB FCD BE .
2
4
6 6
(B.5)
The fermionic part of the bulk Lagrangian expressed in two-component spinor notation reads10 :

1
LFermi = e5 ( 1 D 1 + 2 D 2 )
2

+ e5 1 D5 2 2 D5 1
5

e5 51 D 2 52 D 1 + 1 D 52 2 D 51
5
6 5
i
F (1 51 + 2 52 + i1 2 )
e
8 5

6
+i
F 1 2 + F 5 (1 52 2 51 )
4

6 5
F5 (1 1 + 2 2 ) + h.c. .
e
+i
(B.6)
8 5
The covariant derivatives employed here are defined in Eq. (A.20).
10 We recall that in this paper we use the following approximation, in the Lagrangians we drop the four-fermions terms
and in the supersymmetry transformations we drop the three and four-fermions terms.
157
In order to express the supersymmetry transformations (2.3) in two-component spinor notation

we adopt, in parallel to Eq. (A.17), the following notation for the supersymmetry transformation
parameter:

2
2 = 1 =
,
1 = 2 = 1 ,
2
1
1 = 2 = ( 1 , 2 ) ,
(B.7)
2 = 1 = ( 2 , 1 ) .
With these definitions, the on-shell supersymmetry transformations in two-component spinor
notation are given by:

a
e M
= i 1 a M1 + 2 a M2 + h.c.,
5
eM
= 2 M1 1 M2 + h.c.,
6
(1 M2 2 M1 ) + h.c.,
BM = i
2

1
2
1 = 2D 1 + F i 4g 2 i e55 F 5 ( + g )1 ,
2 6
6

1
2
2 = 2D 2 F i 4g 1 i e55 F 5 ( + g )2 ,
2 6
6
1 5
2
51 = 2D5 1 i e5 F 1 F5 2 ,
6
6
1 5
2
52 = 2D5 2 i e5 F 2 + F5 1 .
(B.8)
6
6
The bulk fields have well defined Z2 parities as described in Tables 4 and 5, where the following definitions are used:
+ = cos()1 sin()2 ,
= sin()1 + cos()2 ,
5+ = sin()51 + cos()52 ,
5 = cos()51 sin()52 ,
+ = cos()1 sin()2 ,
= sin()1 + cos()2 .
(B.9)
Table 4
Parity assignments for bulk fields at y = 0
Even
a
e
e55
B5
52
Odd
e5a
5
e
51
v5
Table 5
Parity assignments for bulk fields at y = R
Even
a
e
e55
B5
5+
Odd
e5a
5
e
v5
158
At each boundary yb , yb {0, R}, Nb chiral multiplets are placed, each containing one
scalar bi and one fermionic field bi (i = 1, . . . , Nb ). The Lagrangian density for the brane b,
b {0, }, is given by:

1
1
j
j
Lb = e4 gij bi b i gij b D bi
2
2
1
j
j
+ Gbj b Gbj b 1 1
8

2
1
j
j
Gb /2
1 1 + i
Gbj b 1 + Gbij + Gbi Gbj ijk Gbk bi b
e
2
2

2
1
j
gij b bi 1 eGb g ij Gbi Gbj 3 + h.c. ,
(B.10)
2
2
where
1
1
j
j
D bi = bi + ab ab bi Gbj b Gbj b bi
(B.11)
2
4
and the hermitian function Gb (b , b ) is given in terms of the Khler potential and superpotential
by

Gb (b , b ) = Kb (b , b ) + ln Wb (b ) + ln Wb (b ) .
(B.12)
We impose also the following boundary conditions at y = 0 and y = R:

1
j
j
G0j 0 G0j 0 ,
(B.13)
F 5 y=0 = 0,
2

1
u|y=R = eG /2 ,
F 5 y=R = 0.
iv |y=R = Gj j Gj j ,
2
(B.14)
The modified supersymmetry transformations for the bulk fields are given by:
u|y=0 = eG0 /2 ,
iv |y=0 =
A
A
eM
= e M
,
BM = BM ,
1 = 1 + iv 1 + iu 1 ,
2 = 2 + iv 2 + iu 2 ,
51 = 51 4eG /2 sin()+ (y R),
52 = 52 4eG0 /2 1 (y) 4eG /2 cos()+ (y R),

1
u = u i1 1 + i2 2 + 2 51 1 52
2

LSUGRA
LSUGRA
i
+ h.c.,
J
D
+
J
N
e5
J
(DN J )

1
v = v i1 1 + i2 2 + 2 51 1 52
2

LSUGRA
LSUGRA
i
J
+ J DN
e5
J
(DN J )
2ie5 1 D 1 (y) 2ie5 + D + (y R) + h.c.,
5
159

1
v5 = v5 i1 1 + i2 2 + 2 51 1 52 + h.c.
(B.15)
2
a and
In the branes at y = 0 and y = R the supersymmetry transformations of the fields e
I are those induced by the bulk (given by Eq. (B.15) calculated at y = 0 and y = R). Together with the supersymmetry transformations of the brane matter fields (0 , 0 , and )
they read at the brane sitting on the boundary yb , yb {0, R}:

a
= i 1 |y=yb a 1 |y=yb + h.c.,
e
y=yb
bi = 21 |y=yb bi ,
bi = i 2 1 |y=yb bi 2eGb /2 g ij Gbj 1 |y=b ,

1
j
j
1 |y=yb = 2D 1 |y=yb + Gbj b Gbj b 1 |y=yb + ieGb /2 1 |y=yb , (B.16)
2
where D 1 and D + are given by Eq. (A.22).
With the parity assignments of Tables 4 and 5 and the boundary conditions (B.13) and (B.14)
the action (B.1) is invariant under the supersymmetry transformations (B.15) and (B.16) up to
four-fermions terms (which is the approximation we use in this paper).
Appendix C. A simple example of bulk-brane supersymmetry breaking
In this appendix we provide a simple example of supersymmetry breaking in both sectors
(bulk and brane) of the 5d spacetime. We consider only one chiral multiplet living in a brane
placed at y = 0. Supersymmetry is broken in the bulk by a non-trivial ScherkSchwarz twist
described by the angle = 0.
To keep things as simple as possible, in the brane the Khlerpotential is the canonical one,
2
K = and the superpotential considered here is W = e /2+ 3 . This implies the following
Khler function for the brane:
2
2
+ 3
+ 3 .
(C.1)
2
2
We will now show that this choice for the Khler function provide supersymmetry breaking
with a vanishing cosmological constant in the brane. The Lagrangian density (2.15) gives the
following brane potential:

G G
3
.
V = eG
(C.2)

G =
From (C.1) it is easy to obtain that V = eG | |2 . Then at the extremum of the potential
Im() = 0 and
V = 0, giving a vanishing brane cosmological constant, as claimed. It is
useful to parametrize the complex field by two real-valued fields and :
1
= [ + i ].
2
(C.3)
From potential (C.2) it is clear that is massless and has mass squared m2 = 4
e 6 .
The fermionic spectrum is
easily calculated with help of formulae (6.16), (7.4) and the F G
/2
terms values M0 =
e =
e 6/2 and M = 0. Taking the v.e.v.
= A, the fermion masses
160
are given as follows, the gravitino tower of KaluzaKlein masses are:

n

1
+
arctan e 6A/2 + ,
R R
R
and the pseudo-Goldstino mass is:
m3/2 =
mP G =
nZ
6A/2 sin()[e 6A/2 cos() + sin()]
Re 6A +[e 6A/2 cos() + sin()]2
2e
(C.4)
(C.5)
Appendix D. Pseudo-Goldstino mass eigenstates

In this appendix we wish to present the eigenstates and masses of the pseudo-Goldstinos
for general M0 , M and . As said at the end of Section 6.2, the procedure to identify the mass
eigenstates of the pseudo Goldstinos is long but straight forward: one must plug (6.18) and (6.20)
in the Lagrangian (6.10), integrate over the y dimension, diagonalize and canonically normalize
the kinetic terms of the fields 1 and 2 and finally diagonalize their mass matrix. To do this we
set = /2 in (6.18) and do all the diagonalizations described above. We now present the
final results.
We call the mass eigenstates 1 and 2 , their masses are respectively:
m11 a22 + m22 a11 2a12 m12 + d

,
m1 =
2 )
2(a11 a22 a12
m11 a22 + m22 a11 2a12 m12 d

,
m2 =
(D.1)
2 )
2(a11 a22 a12
where we defined

1
1
2
a11 = 1 +
+
2 sin
sin()
R
2
M0 M

1
1
2
+
+
cos
,
2
2 M02 2 M2

1
1
2
a22 = 1 +
+
+
2 cos
sin()
R
2
M0 M

1
1
2
+
+
sin
,
2
2 M02 2 M2

1
1
1
1
1
a12 =
sin() +
cos() ,
R 2 2 M2
M
M0
2 M02

1
1
1
2
m11 =
+
2 sin() ,
2
cos
R
M0 M
2

1
1
1
2
m22 =
+
+ 2 sin() ,
2
sin
R
M0 M
2

1
1
1
m12 =
sin()
R M
M0
(D.2)
161
and
d=
2
a11 a22 a12
,
2 |
|a11 a22 a12

2
2
2
m222 + a22
m211 + 4m11 m22 a12
= 2a11 a22 2m212 m11 m22 + a11
4a12 m12 (a11 m22 + m11 a22 ).
(D.3)
The canonically normalized mass eigenstates are:

1 =
a r3 (a11 a22 + r1 ) 2ba12 r4 1
(a 2 + b2 )r2

+ b r4 (a11 a22 + r1 ) + 2aa12 r3 2 ,

1
2 =
b r3 (a11 a22 + r1 ) + 2aa12 r4 1
2
2
(a + b )r2

+ a r4 (a11 a22 + r1 ) 2ba12 r3 2 ,
(D.4)
where we defined

2
2
2
m11 a22
m22 a11
+ 2(a11 + a22 )a12 m12 + d r1 ,
a = (m11 + m22 ) a11 a22 2a12

2 ,
b = 2d a12 (m11 m22 ) m12 (a11 a12 ) t a11 a22 a12
2
r1 = (a11 a22 )2 + 4a12
,
2
2
r2 = (a11 a22 + r1 ) + 4a12
,
r3 = (a11 + a22 + r1 )/2,
r4 = (a11 + a22 r1 )/2,
a11 a22 r1
t=
.
|a11 a22 r1 |
(D.5)
The fields 51 (y), 52 (y), 0 and can be written in terms of the mass eigenstates 1 and
2 with help of (6.18), (6.20) and
1 =

a r4 (a11 a22 + r1 ) 2ba12 r3 1
(a 2 + b2 )r2 r3 r4

b r4 (a11 a22 + r1 ) + 2aa12 r3 2 ,

1
2 =
b r3 (a11 a22 + r1 ) + 2aa12 r4 1
2
2
(a + b )r2 r3 r4

+ a r3 (a11 a22 + r1 ) 2ba12 r4 2 .
(D.6)
References
[1] P. Fayet, J. Iliopoulos, Phys. Lett. B 51 (1974) 461.
[2] D.V. Volkov, V.A. Soroka, JETP Lett. 18 (1973) 312, Pisma Zh. Eksp. Teor. Fiz. 18 (1973) 529;
S. Deser, B. Zumino, Phys. Rev. Lett. 38 (1977) 1433;
E. Cremmer, B. Julia, J. Scherk, S. Ferrara, L. Girardello, P. van Nieuwenhuizen, Nucl. Phys. B 147 (1979) 105.
[3] P. Fayet, Phys. Lett. B 70 (1977) 461.
[4] R. Casalbuoni, S. De Curtis, D. Dominici, F. Feruglio, R. Gatto, Phys. Rev. D 39 (1989) 2281;
R. Casalbuoni, S. De Curtis, D. Dominici, F. Feruglio, R. Gatto, Phys. Lett. B 215 (1988) 313.
162
[5] I. Antoniadis, Phys. Lett. B 246 (1990) 377;

I. Antoniadis, C. Munoz, M. Quiros, Nucl. Phys. B 397 (1993) 515, hep-ph/9211309;
I. Antoniadis, K. Benakli, Phys. Lett. B 326 (1994) 69, hep-th/9310151;
K. Benakli, Phys. Lett. B 386 (1996) 106, hep-th/9509115.
[6] P. Horava, E. Witten, Nucl. Phys. B 475 (1996) 94, hep-th/9603142;
E. Witten, Nucl. Phys. B 471 (1996) 135, hep-th/9602070;
K. Benakli, Phys. Lett. B 447 (1999) 51, hep-th/9805181;
S. Stieberger, Nucl. Phys. B 541 (1999) 109, hep-th/9807124;
Z. Lalak, S. Pokorski, S. Thomas, Nucl. Phys. B 549 (1999) 63, hep-ph/9807503.
[7] J.D. Lykken, Phys. Rev. D 54 (1996) 3693, hep-th/9603133.
[8] N. Arkani-Hamed, S. Dimopoulos, G.R. Dvali, Phys. Lett. B 429 (1998) 263, hep-ph/9803315;
I. Antoniadis, N. Arkani-Hamed, S. Dimopoulos, G.R. Dvali, Phys. Lett. B 436 (1998) 257, hep-ph/9804398.
[9] K.R. Dienes, E. Dudas, T. Gherghetta, Phys. Lett. B 436 (1998) 55, hep-ph/9803466;
K.R. Dienes, E. Dudas, T. Gherghetta, Nucl. Phys. B 537 (1999) 47, hep-ph/9806292.
[10] K. Benakli, Phys. Rev. D 60 (1999) 104002, hep-ph/9809582;
C.P. Burgess, L.E. Ibanez, F. Quevedo, Phys. Lett. B 447 (1999) 257, hep-ph/9810535.
[11] L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 3370, hep-ph/9905221;
L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 4690, hep-th/9906064.
[12] E. Witten, Nucl. Phys. B 188 (1981) 513.
[13] P. Horava, Phys. Rev. D 54 (1996) 7561, hep-th/9608019.
[14] G. Veneziano, S. Yankielowicz, Phys. Lett. B 113 (1982) 231;
H.P. Nilles, Phys. Lett. B 115 (1982) 193;
H.P. Nilles, Nucl. Phys. B 217 (1983) 366;
T.R. Taylor, G. Veneziano, S. Yankielowicz, Nucl. Phys. B 218 (1983) 493;
S. Ferrara, L. Girardello, H.P. Nilles, Phys. Lett. B 125 (1983) 457;
I. Affleck, M. Dine, N. Seiberg, Nucl. Phys. B 241 (1984) 493;
I. Affleck, M. Dine, N. Seiberg, Nucl. Phys. B 256 (1985) 557;
J.P. Derendinger, L.E. Ibanez, H.P. Nilles, Phys. Lett. B 155 (1985) 65;
M. Dine, R. Rohm, N. Seiberg, E. Witten, Phys. Lett. B 156 (1985) 55;
C. Kounnas, M. Porrati, Phys. Lett. B 191 (1987) 91;
K.A. Intriligator, S.D. Thomas, Nucl. Phys. B 473 (1996) 121, hep-th/9603158;
K.I. Izawa, T. Yanagida, Prog. Theor. Phys. 95 (1996) 829, hep-th/9602180.
[15] Excellent guides through recent developpements are K. Intriligator, N. Seiberg, hep-ph/0702069;
Y. Shadmi, Y. Shirman, Rev. Mod. Phys. 72 (2000) 25, hep-th/9907225;
G.F. Giudice, R. Rattazzi, Phys. Rep. 322 (1999) 419, hep-ph/9801271, and references therein.
[16] J. Scherk, J.H. Schwarz, Phys. Lett. B 82 (1979) 60;
J. Scherk, J.H. Schwarz, Nucl. Phys. B 153 (1979) 61;
P. Fayet, Phys. Lett. B 159 (1985) 121;
P. Fayet, Nucl. Phys. B 263 (1986) 649.
[17] E.A. Mirabelli, M.E. Peskin, Phys. Rev. D 58 (1998) 065002, hep-th/9712214.
[18] L. Randall, R. Sundrum, Nucl. Phys. B 557 (1999) 79, hep-th/9810155;
G.F. Giudice, M.A. Luty, H. Murayama, R. Rattazzi, JHEP 9812 (1998) 027, hep-ph/9810442.
[19] I. Antoniadis, M. Quiros, Nucl. Phys. B 505 (1997) 109, hep-th/9705037;
E. Dudas, C. Grojean, Nucl. Phys. B 507 (1997) 553, hep-th/9704177;
H.P. Nilles, M. Olechowski, M. Yamaguchi, Phys. Lett. B 415 (1997) 24, hep-th/9707143;
A. Lukas, B.A. Ovrut, D. Waldram, Phys. Rev. D 57 (1998) 7529, hep-th/9711197;
K. Choi, H.B. Kim, C. Munoz, Phys. Rev. D 57 (1998) 7521, hep-th/9711158;
J.R. Ellis, Z. Lalak, S. Pokorski, W. Pokorski, Nucl. Phys. B 540 (1999) 149, hep-ph/9805377.
[20] A. Pomarol, M. Quiros, Phys. Lett. B 438 (1998) 255, hep-ph/9806263;
I. Antoniadis, S. Dimopoulos, A. Pomarol, M. Quiros, Nucl. Phys. B 544 (1999) 503, hep-ph/9810410;
R. Barbieri, L.J. Hall, Y. Nomura, Phys. Rev. D 63 (2001) 105007, hep-ph/0011311;
D. Marti, A. Pomarol, Phys. Rev. D 64 (2001) 105025, hep-th/0106256;
D. Marti, A. Pomarol, Phys. Rev. D 66 (2002) 125005, hep-ph/0205034;
D.E. Kaplan, G.D. Kribs, M. Schmaltz, Phys. Rev. D 62 (2002) 035010, hep-ph/9911293;
Z. Chacko, M.A. Luty, A.E. Nelson, E. Ponton, JHEP 0001 (2000) 003, hep-ph/9911323;
K. Benakli, Phys. Lett. B 475 (2000) 77, hep-ph/9911517;
163
N. Arkani-Hamed, L.J. Hall, Y. Nomura, D.R. Smith, N. Weiner, Nucl. Phys. B 605 (2001) 81, hep-ph/0102090.
[21] T. Gherghetta, A. Pomarol, Nucl. Phys. B 586 (2000) 141, hep-ph/0003129;
T. Gherghetta, A. Pomarol, Nucl. Phys. B 602 (2001) 3, hep-ph/0012378;
A. Falkowski, Z. Lalak, S. Pokorski, Phys. Lett. B 491 (2000) 172, hep-th/0004093;
R. Altendorfer, J. Bagger, D. Nemeschansky, Phys. Rev. D 63 (2001) 125025, hep-th/0003117;
E. Bergshoeff, R. Kallosh, A. Van Proeyen, JHEP 0010 (2000) 033, hep-th/0007044;
G. von Gersdorff, M. Quiros, Phys. Rev. D 65 (2002) 064016, hep-th/0110132;
J.L. Lehners, P. Smyth, K.S. Stelle, arXiv: 0704.3343 [hep-th].
[22] M. Zucker, Fortschr. Phys. 51 (2003) 899;
M. Zucker, Phys. Rev. D 64 (2001) 024024, hep-th/0009083;
M. Zucker, JHEP 0008 (2000) 016, hep-th/9909144;
M. Zucker, Nucl. Phys. B 570 (2000) 267, hep-th/9907082.
[23] F. Paccetti Correia, M.G. Schmidt, Z. Tavartkiladze, Nucl. Phys. B 709 (2005) 141, hep-th/0408138;
F. Paccetti Correia, M.G. Schmidt, Z. Tavartkiladze, Phys. Lett. B 613 (2005) 83, hep-th/0410281;
F. Paccetti Correia, M.G. Schmidt, Z. Tavartkiladze, Nucl. Phys. B 751 (2006) 222, hep-th/0602173.
[24] T. Kugo, K. Ohashi, Prog. Theor. Phys. 104 (2000) 835, hep-ph/0006231;
T. Kugo, K. Ohashi, Prog. Theor. Phys. 105 (2001) 323, hep-ph/0010288;
T. Fujita, K. Ohashi, Prog. Theor. Phys. 106 (2001) 221, hep-th/0104130;
H. Abe, Y. Sakamura, JHEP 0410 (2004) 013, hep-th/0408224;
H. Abe, Y. Sakamura, JHEP 0602 (2006) 014, hep-th/0512326.
[25] T. Gherghetta, A. Riotto, Nucl. Phys. B 623 (2002) 97, hep-th/0110022.
[26] R. Rattazzi, C.A. Scrucca, A. Strumia, Nucl. Phys. B 674 (2003) 171, hep-th/0305184.
[27] I.L. Buchbinder, S.J.J. Gates, H.S.J. Goh, W.D.I. Linch, M.A. Luty, S.P. Ng, J. Phillips, Phys. Rev. D 70 (2004)
025008, hep-th/0305169.
[28] J.A. Bagger, F. Feruglio, F. Zwirner, Phys. Rev. Lett. 88 (2002) 101601, hep-th/0107128;
J.A. Bagger, F. Feruglio, F. Zwirner, JHEP 0202 (2002) 010, hep-th/0108010;
C. Biggio, F. Feruglio, A. Wulzer, F. Zwirner, JHEP 0211 (2002) 013, hep-th/0209046.
[29] G. von Gersdorff, M. Quiros, A. Riotto, Nucl. Phys. B 634 (2002) 90, hep-th/0204041;
G. von Gersdorff, L. Pilo, M. Quiros, A. Riotto, V. Sanz, Phys. Lett. B 598 (2004) 106, hep-th/0404091;
K.A. Meissner, H.P. Nilles, M. Olechowski, Acta Phys. Pol. B 33 (2002) 2435, hep-th/0205166;
A. Delgado, G. von Gersdorff, M. Quiros, JHEP 0212 (2002) 002, hep-th/0210181;
Z. Lalak, R. Matyszkiewicz, Nucl. Phys. B 730 (2005) 37, hep-ph/0506223.
[30] K.A. Meissner, H.P. Nilles, M. Olechowski, Nucl. Phys. B 561 (1999) 30, hep-th/9905139.
[31] S. De Curtis, D. Dominici, J.R. Pelaez, JHEP 0401 (2004) 052, hep-th/0311226.
[32] J.A. Bagger, D.V. Belyaev, Phys. Rev. D 72 (2005) 065007, hep-th/0406126.
[33] S. Weinberg, Phys. Rev. Lett. 29 (1972) 1698.
[34] H. Georgi, A. Pais, Phys. Rev. D 10 (1974) 539;
H. Georgi, A. Pais, Phys. Rev. D 12 (1975) 508.
[35] N. Arkani-Hamed, A.G. Cohen, H. Georgi, Phys. Lett. B 513 (2001) 232, hep-ph/0105239.
[36] H. Hatanaka, T. Inami, C.S. Lim, Mod. Phys. Lett. A 13 (1998) 2601, hep-th/9805067;
A. Masiero, C.A. Scrucca, M. Serone, L. Silvestrini, Phys. Rev. Lett. 87 (2001) 251601, hep-ph/0107201;
C.P. Bachas, hep-th/9509067;
G.R. Dvali, S. Randjbar-Daemi, R. Tabbash, Phys. Rev. D 65 (2002) 064021, hep-ph/0102307;
I. Antoniadis, K. Benakli, M. Quiros, New J. Phys. 3 (2001) 20, hep-th/0108005;
C. Csaki, C. Grojean, H. Murayama, Phys. Rev. D 67 (2003) 085012, hep-ph/0210133;
C.A. Scrucca, M. Serone, L. Silvestrini, Nucl. Phys. B 669 (2003) 128, hep-ph/0304220.
[37] G. Cacciapaglia, C. Csaki, C. Grojean, M. Reece, J. Terning, Phys. Rev. D 72 (2005) 095018, hep-ph/0505001.
[38] E. Cremmer, Invited paper at the Nuffield Gravity Workshop, Cambridge, Eng., June 22July 12, 1980;
A.H. Chamseddine, H. Nicolai, Phys. Lett. B 96 (1980) 89.
[39] J. Wess, J. Bagger, Supersymmetry and supergravity, second ed., Princeton Univ. Press, 1992.
[40] L. Baulieu, A. Georges, S. Ouvry, Nucl. Phys. B 273 (1986) 366.
[41] A. Van Proeyen, hep-th/9910030.
2D Heisenberg model from rotating membrane

Wen-Yu Wen
Department of Physics and Center for Theoretical Sciences, National Taiwan University, Taipei 106, Taiwan
Received 1 June 2007; received in revised form 21 August 2007; accepted 21 September 2007
Abstract
We study a rotating probe membrane in S 3 inside AdS4 S 7 background of M-theory. With a particular
gauge choice and given ansatz, we show that in the fast limit the worldvolume of tensionless membrane
reduces to either the XXX1/2 spin chain or the two-dimensional SU(2) Heisenberg spin model. Later we
introduce the anisotropy and couple it to the external magnetic field. We also establish the correspondence
for higher-dimensional (D)p-branes.
Keywords: M-theory; Membrane dynamics; 2D Heisenberg model
1. Introduction
The anti-de Sitter/conformal field theory (AdS/CFT) correspondence has revealed deep relation between string theory and gauge theory, in particular the correspondence between IIB strings
on AdS5 S 5 and N = 4 super-YangMills theory (SYM) at the t Hoofts large N limit [13].
While the correspondence was mostly tested for those BogomolnyiPrasadSommerfield (BPS)
states where partial supersymmetry protects it against quantum correction [4], semiclassical
analysis on the near-BPS sector is also considered in Ref. [5]; the integrability of this Berenstein
MadalcenaNastase (BMN) limit allows for quantitative tests of correspondence beyond BPS
states, where the energy of classical string solutions is compared to the anomalous dimension of
SYM operators with large R-charge. On the other hand, intimate relation between SYM dynamics and integrable spin chain was realized in Ref. [6], because the planar limit of the dilatation
operator was identified with the Hamiltonian of integrable spin chains.
E-mail address: steve.wen@gmail.com.

doi:10.1016/j.nuclphysb.2007.09.027
W.-Y. Wen / Nuclear Physics B 791 (2008) 164174
165
The string/spin chain correspondence was first demonstrated in Ref. [7], where a classical
rotating string on S 3 inside S 5 was identified with the semiclassical coherent state in the SU(2)
Heisenberg spin chain model. Later this identification was explored in the full SU(3) sector [8]
and SL(2) sector [9]. A few examples of generalization were discussed in the past: the fast
spinning string in the marginally deformed AdS3 S 3 [10] and -deformed N = 4 [11], each
corresponds to the anisotropic XXZ spin chain [12,13]. The Melvins magnetic-deformed background was also studied in Ref. [14].
Strings, however, are not the only elementary objects since we have learnt that there are
more extended objects such as D(Dirichlet)p-branes in the string theory as well as membranes
and five-branes in the M-theory. Those extended objects are soliton-like objects in the low energy description of supergravity, resisting any naive perturbative analysis applicable to strings.
Nevertheless, they may serve as a good probe to explore non-perturbative aspects of string/Mtheory, complementary to our perturbative knowledge based on strings. In addition to their
non-perturbative nature, there is no unique way to gauge away degrees of freedom in the worldvolume of p-branes such as what we usually do to the worldsheet of strings. Despite of these
apparent difficulties, it is still an educated guess that there might be similar correspondence between those extended objects and higher-dimensional spin systems in some specific limit. In
this paper, we take the first step to support this conjecture with concrete examples. In particular, with careful choice of Lagrangian multipliers and proper embedding, we are able to achieve
the correspondence between rotating membrane and two-dimensional (2D) Heisenberg model
and later generalize to arbitrary p dimensions. The outline of this paper is as follows: In Section 2, a rotating tensionless membrane with a particular embedding ansatz is shown to give
rise to the 2D Heisenberg model. The condition of integrability for this spin model is briefly
mentioned. The anisotropy is introduced via the same deformation as shown in Ref. [12] and
vortices-like excitation is discussed. At the end, we show that upon a partial embedding the
action can also reduce to the XXX1/2 spin chain. In Section 3, the correspondence between pbranes and higher-dimensional spin models is also established. In particular, for Dp-branes there
exist some new features due to nontrivial dilaton field and two-form flux. We also show that the
external magnetic field can be generated geometrically in all the SU(2) spin models. In the conclusion, we summarize our results and comment on possible directions for future investigation.
In Appendix A we review the double scaling limit constructed in Ref. [12] for the anisotropic
Heisenberg model.
2. 2D Heisenberg model/membrane correspondence
2.1. p-brane action
The probe branes approach is widely used in the string/M-theory to study different aspects
of brane itself or specific background, from classical to quantum levels. It takes the assumption
that the back-reaction of the brane on the background is negligible. In this paper we only study a
single brane in the tensionless limit, thus qualifying the probing assumption. The brane action in
general falls into two types: the DiracNambuGoto type action and the Polyakov type action.
The former one is intuitively simple but carries nonlinearity due to the square root. The latter
one introduces non-dynamical Lagrangian multipliers in order to linearize the former action. It
is, however, easier for the purpose of calculation and suitable for taking the tensionless limit. To
166
be specific, we adopt a Polyakov type action proposed for general p-branes [15]:

2

1
Sp = d 2
G00 2j G0j + i j Gij 20 Tp |Gij |
0
4

+ Tp b0 p 0 X 0 p X p ,
(1)
= {X 0 , . . . , X D } are (D + 1)where = { 0 , . . . , p } are worldvolume coordinates and X

dimensional target spacetime. s are the Lagrangian multipliers. i, j = {1, . . . , p} only run for
spatial indices. Gij i X j X g is the pull back metric on the worldvolume and |G| is for
determinant. This action can be shown equivalent to the DiracNambuGoto type action with the
following equations of motion of s substituted in,
2

G00 2j G0j + i j Gij + 20 Tp |Gij | = 0,
G0j i Gij = 0.
(2)
Without loss of generality, we may set i = 0 but 0 = 0 as a gauge choice. Then the action
is simplified and equations of motion of s reduce to the Virasoro-like constraints
2

G00 + 20 Tp |Gij | = 0,
(3)
G0j = 0.
(4)
2.2. Embedding ansatz and tensionless limit

We will restrict our discussion on membranes in the M-theory for the moment and come back
to general p-branes in the next section. We first make an embedding ansatz as follows,
X 0 = 0 ,
X 9 = c2 2 ,
X 10 = c1 1 ,
(5)
and the determinant decomposes into

ij | + c22 g99 G
11 + c12 g[10][10] G
22 + c12 c22 g99 g[10][10] ,
|Gij | = |G
(6)
ij is the pull back metric of subspace {X 1 , . . . , X 8 }. We also make an assumption

where the G
that there is no cross term such as g9[10] , gi9 , gi[10] . Then the action becomes

0 2
1
g1 2 g2 2
3
2
2
G00 2 T2 |Gij | g1 G11 g2 G22 0

Sm2 = d
40
2 T2

+ T2 b 0 X 1 X 2 X ,
g1 20 c2 T2 g99 ,
g2 20 c1 T2 g[10][10] .
(7)
In particular, we are looking for the limit where the membrane is tensionless but effective couplings gi s are finite. This can be achieved by sending T2 0 and ci but keeping their
product finite. With this embedding and scaling, we only pay attention to the local property of
the membrane regardless its global topology. In the next section, we will see that the tensionless
membrane in this embedding provides an appropriate setting for the 2D Heisenberg model.
167
2.3. 2D Heisenberg model from fast membrane

In order to reproduce the Heisenberg model, a fast membrane limit has to be taken in analogy
to Ref. [7]. We first make the membrane rotate in one angular direction then take the fast limit. In
this limit, we send and X 0 but keep X finite, here X is partial derivative w.r.t. 0 .
The physics behind implies that at this time scale, along those direction transverse to the rotation
plane, the membrane moves very slowly and can be seen as almost frozen. Although the conjugated energy and momentum at this scale in fact blow up, the finite part of worldvolume action
is recognized as the sigma model of Heisenberg spin system. To illustrate this, we will assume
that a tensionless probe membrane, constructed in the previous subsection, rotates inside a S 3 ,
which could be part of the eleven-dimensional vacuum of M-theory, such as AdS4 S 7 . The metric g99 , g[10][10] are assumed to take constant values on the sphere for simplicity throughout the
paper, but this assumption can be easily relaxed. The relevant background metric after rotating,
i.e. t + , is given by
ds 2 =

1
2 dt d + d 2 + d 2 + d 2 + 2 cos d d + 2 cos dt d .
4
After taking the fast limit, the finite worldvolume action with pull back metric reads1

2

1
3
d
Sm 2 =
+
2
cos
g2i (i )2 + (i )2 + (i )2
0
16
i=1

+ 2 cos i i .
(8)
(9)
Notice that only derivatives of first order in 0 and second order in i are survived in this limit,
which is the very differential structure underlying the Heisenberg spin model. Applying the constraint (4) in this limit, which reads
i + cos i = 0,
(10)
to replace i , one achieves the action of Heisenberg model:

2

1
d 3 2t + 2 cos t
g2i (i )2 + sin2 (i )2 ,
S m2 =
160
(11)
i=1
or the Hamiltonian after the Legendre transformation,

g2
2.
d 1 d 2 ( S)
H2D =
160
(12)
Here we have made gi = g for the universal coupling and defined a SO(3) vector S =
(sin sin , sin cos , cos ), which smoothly rotates as we move around the 1 2 plane.
Strictly speaking, this Hamiltonian is the long wavelength or continuous limit of that of the
discrete lattice model, where the only interaction is among nearest neighbors. Therefore this is
1 The complete action also includes a divergent constant term which is inversely proportional to T . It is generically
2
finite but seen as an artifact in the tensionless limit. Anyway, this constant term will not affect the equations of motion.
168
straightforward generalization of one-dimensional spin chain to higher dimensions. The nearest diagonal sites are next-nearest neighbors and interaction among them only appears in the
second loop mixing. In another words, that mixing enters the nonlinear sigma model of worldsheet/worldvolume only via higher order correction, if desired.
2.4. Membrane excitation and integrability
In Ref. [16], magnon-like excitation was studied for the membrane in the same Polyakov
action. In our limit, the membrane action is greatly simplified and excitation on the membrane
has its description in the corresponding Heisenberg model. The equation of motion derived from
Eq. (A.2) is the very (1 + 2)-dimensional LandauLifshitz equation, i.e.

t S = g2 S 2 S,
(13)
which, different from its (1 + 1)-dimensional counterpart, is not integrable in general. This system is gauge equivalent to the (1 + 2)-dimensional nonlinear Schrdinger-type equation [17]. It
has been shown that, however, for the following travelling wave ansatz it is still integrable,

S , 0 = S v 0 , = 1 , 2 .
(14)
We will follow Ref. [18] to discuss its integrability by taking the advantage of isomorphism
SO(3) SU(2). To do so, we set g = 1 for simplicity and rewrite S = S a a , where a are
the Pauli matrices and introduce complex variables z = 12 ( 1 + i 2 ) and = v 1 + iv 2 , and their
complex conjugates z , .
Then one can form a Lax pair (C , C + ) with a 22 complex matrix ,
z = C ,
z = C + ,
[S, z S]
x vS
C
,
+
2i(x i) (x i)2
C+
[S, z S]
xvS
.
+
2i(x + i) (x + i)2
(15)
The integrability condition thus reads Cz+ + Cz [C + , C ] = 0. The arbitrary real spectral
parameter x relates to a rotation on v in the following way,
v = ei v,
x = cot /2.
In fact, one can construct a new travelling wave solution with velocity v

S = S v 0 ,
(16)
(17)
where S = S with S = S a a .
2.5. Anisotropy and vortex dynamics
It has been shown in Ref. [12] that in the limit of fast string and small deformation,
a one-parameter deformation on the target S 3 reproduces the anisotropic Heisenberg spin chain
XXZ1/2 . This generalization is also applicable to the fast membrane considered in the present
paper. We will summarize the derivation in Appendix A and only present the result here. The
modified LandauLifshitz equation is now with an anisotropic 3 3 matrix J , i.e.

t S = g2 S 2 S + S J S,
J11 = J22 = 1,
J33 = 1 .
(18)
169
is the deformed parameter which rescales the U (1) fiber in the Hopf fibration of S 3 . It reduces to XY -model and isotropic XXX1/2 model for = 1 and = 0, respectively. The same
model has been applied to the study of two-dimensional magnets for decades and we expect it
could be carried over, with some caution, to the study of tensionless membrane in the fast/small
deformation limit. The excitation in the two-dimensional system is much richer than that in
the one-dimensional spin chain. To our particular interests, topological objects such as vortices
can be excited [19,20]. The vortices here carry two topological charges under the homotopy
group 1 and 2 , respectively. They are the vorticity, q = 1, 2, . . . and the polarization p,
where p = 1 for non-planar excitation and p = 0 for planar one. The dynamics of vortices is
governed by both the Coulomb force and Magnus force. The latter is a gyro force coupled to
the vortex velocity, thus its behavior is non-Newtonian. A single static vortex is non-local for
its naked topological charge. A vortexantivortex pair, however, can be localized in the form of
dipole charge. Other than vortices, one also expects the meson-like excitation, magnons, appears
as usual. A domain wall could even form if several of them are excited coherently.
In fact, the shape of vortex can be given by a more generic travelling wave ansatz [20],

S , 0 = S x, x , x , . . . , x(n) ,
(19)
where x is the trajectory of vortex center. Given the anisotropic Hamiltonian H = H2D +
g2 2
d (S 3 )2 , Eq. (18) can be rewritten as
160
H
S
,
= S
0
S
(20)
which yields an (n + 1)th order differential equation for x( 0 ). We direct the readers to Ref. [20]
for more discussion on dynamics of those excitations.
2.6. Spin chain from partial embedding ansatz
In the previous discussion we have succeeded in establishing the correspondence between the
2D Heisenberg model and a rotating membrane, here we would like to explore another possible
correspondence associated with the membrane. In the case of partial embedding, one may instead
obtain the XXX1/2 spin chain from the same limit. To illustrate this, let us consider the following
ansatz,
X 0 = 0 ,
X 10 = c 1 ,
(21)
and the determinant decomposes into

ij | + c2 g[10][10] G
22 ,
|Gij | = |G
(22)
ij is now the pull back metric of subspace {X 1 , . . . , X 9 }. Then the action becomes
where the G

0 2
1
2
Sm 2 = d 3
G
+
T
2
T
|
G
|
g
b
X
,
G
00
2
ij
22
2 0
1
2
40
g 20 cT2 g[10][10] .
(23)
In the tensionless limit, i.e. T2 0 but g kept finite, one maps the membrane to the target
space S 3 and takes the fast spinning limit as before. It is not difficult to identify the resulting
170
action with the Heisenberg XXX1/2 spin chain and the Hamiltonian is given by

g2

d 2 2 S 2 S.
H1D =
160
(24)
Here we see that the membrane loses its dependance on 1 and behaves just like a string extending along 2 . From the viewpoint of double dimensional reduction where M-theory reduces to
type IIA string and membranes to strings, we might think that the same spin chain system can
also be obtained from rotating D1 string inside S 3 , in comparison with the case of fundamental
string [7]. However, this is puzzling since there is no D1 brane in the IIA string. A better interpretation would be that a compactified D2 brane behaves like a one-dimensional string. We are
also tempted to make a connection with the matrix model interpretation of M-theory [21], where
the dynamics is translated into (1 + 0) matrix quantum mechanics and degrees of freedom are
those of D0 particles. Here we conjecture that those D0 particles form a chain and each of them
possesses a spin on S 2 .
3. Heisenberg model/p-brane correspondence
3.1. p-dimensional SU(2) spin model from rotating p-brane
In principle, one may start with a rotating p-brane inside a S 3 out of D-dimensional spacetime. With an embedding ansatz as follows2 :
X 0 = 0 ,
X Dp = cp p , . . . , X D1 = c1 1 .
(25)
It is not difficult to show that in the tensionless limit and fast brane limit as before, we reproduce
the p-dimensional SU(2) Heisenberg model with the action,

p

1
p+1
2
2
2
2
d
,
2t + 2 cos t
gi (i ) + sin (i )
Sp =
160
i=1
gi = 20 Tp
p

cj g[Dj ][Dj ] / ci g[Di][Di] ,
(26)
j =1
where the limit has been taken with Tp 0, ci but finite gi . As for Dp-branes, more care
is needed for its non-vanishing dilaton and antisymmetric B-field Bmn from the closed string
sector3 as well as a U (1) gauge field Fmn from the open string sector. A Polyakov type action is
given by [15]

2

ea
G00 2i G0i + i j i j Gij 20 TDp |Gij |
SDp = d p+1
0
4

+ 2 i F0i j Fj i + TDp c0 p 0 X 0 p X p ,
(27)
where Fmn Bmn + 2 Fmn = [m An] . Equations of motion for those Lagrange multipliers
i , i give the constraints,

2

G00 2j G0j + i j i j Gij + 20 TDp |Gij | + 2 i F0i j Fj i = 0,
2 We always make the assumption that there is enough space for a S 3 after embedding.
3 Technically speaking, dilaton field and antisymmetric B-field are in the same NSNS sector as graviton, namely both
left- and right-moving modes satisfy NeveuSchwarz periodic condition.
171
Goj i Gij = i Fij ,

F0j i Fij = i Gij .
(28)
The choice of vanishing

decouples F from the metric and gives us the same Heisenberg
model as in Eq. (26). Another nontrivial choice is to keep nonzero i . In the following, for
simplicity we will discuss the case with only 1 = 0 and A0 = 0. The action then simplifies as

2
ea
SDp = d p+1
G00 + 1 1 G11 20 TDp |Gij |
0
4
+ TDp c0 p 0 X 0 p X p .
(29)
i ,
With the embedding (25) and target space given in Eq. (8), the action becomes

a
p+1 e
SDp = d
20 + 2 cos 0 + 1 0 A1
160

1
2
p
p

g2i
g2i (i )2 + sin2 (i )2 ,
F1i
2
i=2
(30)
i=1
where those constraints in (28) have been used to replace and G11 . We should take 1 in
order for F to survive the fast limit as . After the Legendre transformation, one obtains
the Hamiltonian of Heisenberg model with an extra magnetic field F1i , that is,

1 2
p
a
2 + 1 F1i F 1i .
d
HDp =
(31)
g
e
(
S)
2
160
Here introduction of the nontrivial dilaton field and two-form flux complicates the original
LandauLifshitz equation. We simply comment that the effective coupling now varies with location and an external magnetic field goes through the p-volume. Their functions are to be
determined by the modified equations of motion derived from the action (30).
3.2. Lower-dimensional Heisenberg model from partial embedding
In the previous section, we have learnt that with partial embedding, a rotating membrane can
also have its correspondence in the spin chain. The same technique can easily apply to p-branes,
where a lower-dimensional Heisenberg model is obtained. Here we simply mention the scheme
without detail. First we make the embedding ansatz,
X 0 = 0 ,
X Dq = cq q , . . . , X D1 = c1 1 ,
(32)
where 0 < q < p. Then the determinant decomposes into

ij | +
|Gij | = |G
q

cj2 g[Dj ][Dj ]

jj | + ,
ci2 g[Di][Di] |G
(33)
j =1
ij is the pull back metric of subspace {X 1 , . . . , X Dq1 }. The rest terms on the rightwhere the G
hand side is irrelevant after both tensionless and fast limit are taken. Then it is straightforward to
show that q-dimensional Heisenberg model can be derived.
172
3.3. Anisotropy and external magnetic field

As shown in Ref. [12] and summarized in Appendix A, similarly one can also deform the
target S 3 for the p-brane. This corresponds to the p-dimensional anisotropic SU(2) Heisenberg
model. In addition to anisotropy, one may wonder if a Zeeman term can be added for interaction with the internal field. We have just demonstrated that a local coupling and magnetic flux
occurs typically for the Dp-branes. However, this magnetic flux is somehow expected from the
full spectrum of string theory and takes arbitrary form unless specified by equations of motion.
In fact, there is an alternative way to switch on external magnetic field, say along z-direction,
for both string and general p-branes. This is done geometrically by mixing two angles of S 3
unevenly. Start with the usual parametrization,
ds 2 = d 2 + sin2 d12 + cos2 d22 ,
and then perturb the mixing matrix with a deformation parameter b, such that

1/2
0
0
1 =
0 1/2(1 + b) 1/2
.
2
0 1/2(1 b) 1/2
(34)
(35)
The metric after transformation becomes

1
ds 2 = d 2 + d 2 + d 2 + 2 cos d d 2b cos d 2 2b d d + b2 d 2 . (36)
4
After taking the fast limit, in the same spirit of the small deformation limit, we then send b 0
but only keep 2 b finite. This results in an additional term 2b cos to the action (26), here
has been absorbed into redefinition of X X and 2 b b. It can be further put into the form
B S and precisely interpreted as the interaction with external magnetic field B = 2bz. This
uneven mixing of angles was already observed in [22], there appears as mixing of angular velocities. In fact, it can be shown that our double scaling limit of small deformation is equivalently to
their decoupling limit.
4. Conclusion
We have realized in the fast membrane limit the 2D Heisenberg model, which is integrable at
least for the ansatz of travelling wave. In Ref. [23], the author found several types of membrane
embedding into the AdS4 S 7 background, which are related to the Neumann and Neumann
Rosochatius integrable systems. Our result supports his conjecture that there might have various
kinds of integrable system dual to membranes in the M-theory.
In this paper, we also proposed the correspondence between p-branes and p-dimensional
Heisenberg spin model. For Dp-branes, in particular, the nontrivial dilaton field generates a
position-varying coupling and the two-form flux couples to the system magnetically. Later we
also provide a geometric realization of external magnetic field in the SU(2) Heisenberg model in
arbitrary dimension.
Several directions may deserve further investigation. Here we only mention some of them:
In the string/spin chain correspondence, one is able to identify the anomalous dimension of
single trace operator Tr(ZZ . . . Z), calculated from the dual super-YangMills theory, with the
Hamiltonian of spin chain. The two or higher-dimensional spin model may not have such a
correspondence due to lacking the knowledge of hypersurface operators. However, the spin
173
chain model obtained from partial gauge fixing can still have its correspondence on the dual
field theory side. In the present case with membranes, calculation of single trace operators in a
three-dimensional N = 8 super-YangMills (CFT3 as its IR fixed point) may support this correspondence.
In the view of 2D lattice, it is tempted to also switch on interaction among diagonal sites,
which counts as next-nearest neighbors as previously mentioned. It could be interesting to investigate the way it appears in the worldvolume action, in comparison to the correction in the
string worldsheet.
Finally, in Ref. [24], we have learnt that under a partial embedding ansatz, a membrane can
be seen as a perturbation around string like configuration, where the membrane tension acts like
the coupling constant. It is tempting to relax our tensionless limit and study the deviation from
the Heisenberg model perturbatively.
Acknowledgements
I am grateful to invitation of NCTS/NTHU to present this work at its final stage. I would
like to thank K. Furuuchi, P.M. Ho, S. Teraguchi, D. Tomino, Q.S. Yan, and S. Zeze for useful
discussion and comments. The author is supported in part by the Taiwans National Science
Council under Grant Nos. NSC95-2811-M-002-013 and NSC96-2119-M-002-001.
Appendix A. Anisotropic Heisenberg model from deformed S 3
The deformed metric is given by

1
dt 2 + d 2 + sin2 d 2 + (1 )(d + cos d)2 .
(A.1)
4
We request 0 1 to avoid non-unitary gauge field ( < 0) and closed time-like geodesics
( > 1). For trivial = 0, one recovers the unit round S 3 as Hopf fibration. For maximal = 1,
the S 1 fiber degenerates and we are left with S 2 .
After sending + t and taking fast limit as the undeformed one, then the relevant part
becomes

2

1

3
2
2
2
2
2
Sm 2 =
d cos + 2 + 2 cos
,
gi (i ) + (i )
160
ds 2 =
i=1
(1 ) sin
.
1 cos2
2
(A.2)
In order to keep only the anisotropic term, we have to take the small deformation limit by sending
0 but keep 2 finite. In this limit, the sin2 as desired. After both fast and small
The equations of
deformation limits are taken, it is convenient to rescale 2 and X X.
motion for and are then given by
g2 sin g2 sin cos ( )2 + sin cos = 0,

sin + g2 sin2 = 0,
which gives rise to the general LandauLifshitz equation,

t S = g2 S 2 S + S J S,
(A.3)
174
J11 = J22 = 1,
J33 = 1 ,
(A.4)
with the same spin vector S = (sin cos , sin sin , cos ).
This general LandauLifshitz equation can be seen as continuous limit of the inhomogeneous
Heisenberg spin chain model.
References
[1] J.M. Maldacena, The large N limit of superconformal field theories and supergravity, Adv. Theor. Math. Phys. 2
(1998) 231, Int. J. Theor. Phys. 38 (1999) 1113, hep-th/9711200.
[2] S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Gauge theory correlators from non-critical string theory, Phys. Lett. B
428 (1998) 105, hep-th/9802109.
[3] E. Witten, Anti-de Sitter space and holography, Adv. Theor. Math. Phys. 2 (1998) 253, hep-th/9802150.
[4] O. Aharony, S.S. Gubser, J.M. Maldacena, H. Ooguri, Y. Oz, Large N field theories, string theory and gravity, Phys.
Rep. 323 (2000) 183, hep-th/9905111.
[5] D. Berenstein, J.M. Maldacena, H. Nastase, Strings in flat space and pp waves from N = 4 super-YangMills,
JHEP 0204 (2002) 013, hep-th/0202021.
[6] N. Beisert, S. Frolov, M. Staudacher, A.A. Tseytlin, Precision spectroscopy of AdS/CFT, JHEP 0310 (2003) 037,
hep-th/0308117.
[7] M. Kruczenski, Spin chains and string theory, Phys. Rev. Lett. 93 (2004) 161602, hep-th/0311203.
[8] R. Hernandez, E. Lopez, The SU(3) spin chain sigma model and string theory, JHEP 0404 (2004) 052, hep-th/
0403139.
[9] S. Bellucci, P.Y. Casteill, J.F. Morales, C. Sochichiu, SL(2) spin chain and spinning strings on AdS5 S 5 , Nucl.
Phys. B 707 (2005) 303, hep-th/0409086.
[10] D. Israel, C. Kounnas, D. Orlando, P.M. Petropoulos, Electric/magnetic deformations of S 3 and AdS3 , and geometric, Fortschr. Phys. 53 (2005) 73, hep-th/0405213.
[11] O. Lunin, J.M. Maldacena, Deforming field theories with U (1) U (1) global symmetry and their gravity duals,
JHEP 0505 (2005) 033, hep-th/0502086.
[12] W.Y. Wen, Spin chain from marginally deformed AdS3 S 3 , Phys. Rev. D 75 (2007) 067901, hep-th/0610147.
[13] S.A. Frolov, R. Roiban, A.A. Tseytlin, Gauge-string duality for superconformal deformations of N = 4 super-Yang
Mills theory, JHEP 0507 (2005) 045, hep-th/0503192.
[14] W.H. Huang, Spin chain with magnetic field and spinning string in magnetic field background, Phys. Rev. D 74
(2006) 027901, hep-th/0605242.
[15] P. Bozhilov, Probe branes dynamics: Exact solutions in general backgrounds, Nucl. Phys. B 656 (2003) 199, hep-th/
0211181.
[16] P. Bozhilov, R.C. Rashkov, Magnon-like dispersion relation from M-theory, Nucl. Phys. B 768 (2007) 193, hep-th/
0607116.
[17] Q. Ding, On the (1 + 2)-dimensional LandauLifshitz equation, math/0504288.
[18] N. Papanicolaou, Duality rotation for two-dimensional classical ferromagnets, Phys. Lett. A 84 (1981) 151154.
[19] M.E. Gouva, G.M. Wysin, A.R. Bishop, F.G. Mertens, Vortices in the classical two-dimensional anisotropic
Heisenberg model, Phys. Rev. B 39 (1989) 11840.
[20] F.G. Mertens, A.R. Bishop, Dynamics of vortices in two-dimensional magnets, in: P.L. Christiansen, M.P. Sorensen
(Eds.), Nonlinear Science at the Dawn of the 21st Century, Springer, 1999.
[21] T. Banks, W. Fischler, S.H. Shenker, L. Susskind, M theory as a matrix model: A conjecture, Phys. Rev. D 55 (1997)
5112, hep-th/9610043.
[22] T. Harmark, K.R. Kristjansson, M. Orselli, Magnetic Heisenberg-chain/pp-wave correspondence, hep-th/0611242.
[23] P. Bozhilov, Neumann and NeumannRosochatius integrable systems from membranes on AdS4 S 7 , arXiv:
0704.3082 [hep-th].
[24] J. Bjornsson, S. Hwang, The membrane as a perturbation around string-like configurations, Nucl. Phys. B 689
(2004) 37, hep-th/0403092.
Background field equations for the duality

symmetric string
David S. Berman , Neil B. Copland, Daniel C. Thompson
Queen Mary College, University of London, Department of Physics, Mile End Road, London, E1 4NS, England, UK
Received 7 September 2007; accepted 24 September 2007
Abstract
This paper describes the background field equations for strings in T-duality symmetric formalisms in
which the dimension of target space is doubled and the sigma model supplemented with constraints. These
are calculated by demanding the vanishing of the beta-functional of the sigma model couplings in the
doubled target space. We demonstrate the equivalence with the background field equations of the standard
string sigma model.
1. Introduction
T-duality is one of the cornerstones of string theory. It is an intrinsically stringy effect which
relates small to large manifolds exchanging winding with momentum modes. From a spacetime
perspective T-duality is a solution generating symmetry of the low energy equations of motion
but from the world sheet point of view, T-duality is a non-perturbative symmetry. Importantly, the
presence of T-duality allows for the construction of non-geometric manifolds where locally geometric regions are patched together with T-duality transformations. Such constructions, known
as T-folds [1], may play a crucial role in moduli stabilisation and certainly are an important part
of any string landscape.
Given the importance of T-duality it is desirable that this symmetry is made manifest in the
string sigma model. There have been various attempts in the past to develop a formalism where
E-mail addresses: d.s.berman@qmul.ac.uk (D.S. Berman), n.b.copland@qmul.ac.uk (N.B. Copland),

d.c.thompson@qmul.ac.uk (D.C. Thompson).
doi:10.1016/j.nuclphysb.2007.09.021
176
D.S. Berman et al. / Nuclear Physics B 791 (2008) 175191
T-duality is a symmetry of the action [2]. We will use the form most recently championed by
Hull [3] as our starting point. This doubles the dimension of the target space with the doubled dimensions being related by T-duality. Additional constraints are then needed to reduce the degrees
of freedom. These take the form of a set of chirality constraints. The result is that the formalism
has manifest T-duality with a doubled target space. The key difficulty is dealing with the constraints in the correct way. Demonstrating the classical equivalence of this formalism to the usual
sigma model is straightforward, however, showing quantum equivalence is less trivial. This has
been discussed in [79].
In this paper we would like to examine the beta-functional of the string in this formalism so
as to determine the background-field equations for the doubled space. This project was prompted
by various questions. The most important was to determine if the background-field equations
arising from the one-loop beta-functional for the doubled formalism were the same as for the
usual string. A priori this did not have to be the case. In fact, one may imagine that they will
be different since the doubled formalism naturally incorporates the string winding modes which
could in principle correct the usual the beta-function. We know that world sheet instantons correct
T-duality [10,11] and so since the doubled space contains the naive T-dual one may think of all
sorts of possibilities that could arise for corrections to the doubled geometry.
We begin by briefly introducing the doubled formalism before incorporating the constraints
into the action so that we can use the background-field method. We then carry out the
background-field expansion to find the resulting one loop beta-functional for the background
doubled metric. Finally the relation to the non-doubled geometry is described. This shows that
the constraints on the doubled geometry required for conformal invariance are equivalent to the
usual background-field equations for the standard string sigma model.
2. The doubled formalism
The doubled formalism [1,3,4] is an alternate description of string theory on target spaces that
are locally T n bundles, with fibre coordinates X i , over a base N with coordinates Y a . The fibre
is doubled to be a T 2n with 2n coordinates XA . The doubled sigma model then has Lagrangian1
1
L = HAB (Y ) dXA dXB + L(Y ) + Ltop (X),
(1)
4
where L(Y ) is the standard Lagrangian for a string on the base and H(Y ) is a metric on the
fibre.2 Ltop is a purely topological term described in [4]. It is vital to obtaining the equivalence
of the doubled to the non-doubled partition functions [7] but it will play no role in the betafunctions and so will be dropped from now on. One may choose a frame where the metric H has
an O(n, n)/O(n) O(n) coset form as follows:

h bh1 b bh1
HAB (Y ) =
(2)
.
h1 b
h1
h and b are the target space metric and B-field on the fibre of the undoubled space. In this
frame XA = (X i , X i ) with {X i } the coordinates on the T-dual torus. We must then supplement
1 The complete formalism also introduces a 1-form connection for the fibration AA (Y ) which we set to zero throughout
this paper.
2 We adopt the conventions of [4]; the worldsheet signature is (+, ), = , = 1 = 01 and for convenience
0
1 01
we have dropped an overall factor of 2 . The factor of 14 in (1) is half the usual normalisation and is required to make
contact with the standard sigma model.
177
this doubled sigma model action with a set of constraints so as to obtain the correct number of
degrees of freedom and be equivalent to the usual non-doubled sigma model. The constraints are
given by
dXA = LAB HBC dXC ,
(3)
where the L is an O(n, n) invariant metric that can be used to raise and lower indices on H which
in this basis is given by

0 1
LAB =
(4)
.
1 0
To understand this constraint it is helpful to introduce a vielbein to allow a change to a chiral
frame (denoted by barred indices) where:

1 0
1 0
HA B (y) =
(5)
,
LA B =
.
0 1
0 1
In this frame the constraint (3) is a chirality constraint ensuring that half the XA are chiral Bosons
and half are anti-chiral Bosons.
As a simple example let us consider the n = 1 case, i.e. a circle, with constant radius R. The
doubled action on the fibre is

1
1
Sd = R 2 dX dX + R 2 d X d X.
(6)
4
4
As in [7], we make the change to basis in which the fields are chiral:
P = RX + R 1 X,
1
Q = RX R X,
P = 0,
(7)
+ Q = 0.
(8)
In this basis the action becomes

1
1
dP dP +
dQ dQ.
Sd =
8
8
(9)
One may then incorporate the constraints into the action using the method of Pasti et al. [12]. We
define one-forms
P = dP dP ,
Q = dQ + dQ,
(10)
which vanish on the constraint. These allow us to incorporate the constraint into the action via
the introduction of two auxiliary closed one-forms u and v as follows:

1
1
1
(Pm um )2 (Qm v m )2
2
SPST =
(11)
dP dP +
dQ dQ
d
+
.
8
8
8
u2
v2
The PST action works by essentially introducing a new gauge symmetry, the PST symmetry,
that allows the gauging away of fields that do not obey the chiral constraints. Thus only the fields
obeying the chiral constraints are physical.
There are now two ways to proceed. One may either gauge fix the PST-style action immediately which will break manifest Lorentz invariance or try to quantise covariantly and introduce
ghosts to deal with the PST gauge symmetry. In this paper we choose the non-covariant option
and immediately gauge fix to give a FloreaniniJackiw [13] style action. Picking the auxiliary
178
PST fields (u and v) to be time-like produces two copies of the FJ action (one chiral and one
anti-chiral)

1
d 2 (1 P P 1 Q+ Q).
S=
(12)
4
We re-expand this in the non-chiral basis to give Tseytlins [5] duality-symmetric formulation of
the action

2

1
d 2 (R1 X)2 R 1 1 X + 20 X1 X .
S=
(13)
2
Notice that the unusual normalisation of (1) was exactly what was needed for this form of the
action to have the correct normalisation. The constraints
0 X = R 2 1 X,
0 X = R 2 1 X
(14)
motion3
then follow after integrating the equations of

and the string wave equation is given by
combining the constraint equations.
Returning to the general case, the PST procedure yields an action

1
d 2 G 1 X 1 X + L 1 X 0 X + K 0 X 0 X ,
S=
(15)
2
where X = (XA , Y a ) = (X i , X j , Y a ) and4

H 0
L 0
G=
,
L=
,
0 g
0 0

K=
0
0

0
.
g
(16)
For the fibre coordinates we have the equation of motion

1 (H1 X) = L1 0 X,
(17)
which integrates to give the constraint (3). This form of the action will be our starting point for
calculating the beta-functional of the theory.
3. The background-field expansion
To perturbatively study ultraviolet divergences in the doubled formalism we expand quantum
fluctuations around a classical background. We make a choice of coordinates that leaves the
general coordinate invariance of the action manifest in the perturbative expansion [14,15].
and a quantum fluctuation ,
First one writes the fields X as the sum of a classical piece Xcl
however, the expansion in this fluctuation would not be general covariant. Instead one takes the
to X + and finds its tangent vector at X with length equal to that of the
geodesic from Xcl
cl
cl
geodesic. We call this tangent vector and it is contravariant, so the coefficients in an expansion
in terms of are tensors. In general, terms linear in are proportional to the equations of
motion of the classical background and so vanish. The quadratic terms will give rise to the one
loop corrections and so are the relevant terms for calculating the one-loop beta-function.
3 We fix the arbitrary function of introduced by integration by observing that (13) has X = f ( ) gauge invariance.
4 In our notation we reserve the Greek characters and to denote worldsheet indices.
179
We will use the algorithmic method of calculating the background-field expansion developed
in [16]. To obtain the nth order background-field expanded action we simply act on the Lagrangian with the operator

d 2 ( )D
(18)
n times and divide by n! (D is the covariant functional derivative with respect to X ( )). The
action of this operator can be summarised as

d 2 ( )D ( ) = 0,
(19)

d 2 ( )D X ( ) = D ( ),
(20)

d 2 ( )D D ( ) = R X ( ),
(21)

d 2 ( )D T1 2 ...n X( ) = D T1 2 ...n ( ),
(22)
where R is the target space Riemann tensor and T1 2 ...n is a rank n tensor and these are
understood to be evaluated at the classical value Xcl . The form of (22) is particularly relevant,
leading to simplification when dealing with the metric.
Expanding the first term in (15) is exactly the same as the standard sigma model calculation
(albeit without world sheet covariance), at first order we have5
G 1 X D1
(23)
and at second order

1
G D1 D1 + R 1 X 1 X ,
(24)
2
where R is the Riemann tensor constructed from the metric G. The expansion of the L term
in the action is more complex giving

1
(25)
L 0 X D1 + L D0 1 X + D L 0 X 1 X
2
at first order and

1
1
L D0 D1 + D D L + L R + L R 0 X 1 X
2
2

+ D L 0 X D1 + D0 1 X
(26)
at second-order. This is the general expansion for any second rank tensor L so the K term in (15)
may be expanded in a similar way.
The first-order terms in vanish as they should (using the equations of motion of Xcl ) leaving
the second-order Lagrangian which is given by
5 From now on X refers to the classical field X .
cl
180
2L(2) = G D1 D1 + L D0 D1 + K D0 D0

R 1 X 1 X + L; D0 1 X + 0 X D1

1
1
+ D D L 0 X 1 X + L R + L R 0 X 1 X
2
2
+ 2K; D0 0 X
1
+ D D K 0 X 0 X + K R 0 X 0 X .
(27)
2
4. Simplification strategy
Now we have the background-field expanded action we may simplify it using the equations of
motion of X (recall X is the classical field configuration which we are expanding around and
so obeys its equation of motion). Then we use Wick contractions on the quantum field . The
equation of motion is

D1 G 1 X = L 1 0 X + D 0 K 0 X ,
(28)
where D 0 is a covariant derivative constructed only from the base metric g (D 0 K = 0 and will
similarly refer to the connection constructed from g). This equation is required to show vanishing
of the first-order action and we now use it to remove all L terms (except the L fluctuation kinetic
term) from the action. This means (using integration by parts) we can substitute

L; 0 1 X + 1 0 X + L 1 X 0 + L 0 X 1

1
= L 1 X 0 X + 0 X 1 X
2

2G 1 1 X G 1 X 1 X

+ 2K 0 0 X + K 0 X 0 X

+ G 1 X 1 X K 0 X 0 X .
(29)
This leads to a dramatically simplified Lagrangian given by
2L(2) = G 1 1 + L 0 1 + K 0 0
1
2 G 1 X 1 G 1 X 1 X
2
1
+ 2 g 0 0 X + K 0 X 0 X .
(30)
2
Note that we have chosen to proceed by expanding covariant derivatives and simplifying using
the equations of motion, rather than leaving things expressed in terms of covariant derivatives.
We now proceed to introduce vielbeins so that we can work in the chiral frame where we know
how to find the fluctuation propagators.
Once vielbeins are introduced there are of course terms with derivatives acting on the vielbeins. Normally such terms are accounted for by exchanging the usual connection for the spin
connection. The pull back of the spin connection to the world sheet transforms as a gauge field.
There is then a general argument that this gauge field, which is minimally coupled, cannot contribute to the Weyl anomaly. We have a modified action where the gauge connection is no longer
181
minimally coupled and there is no such argument (indeed we find contributions from the gauge
terms).
Introducing the vielbein has the effect of moving all indices on s in the second-order action
to the chiral frame at the expense of also introducing the following terms:
2LV = 2G
1 V 1 G 1 V 1 V
+ L
1 V 0 + L
0 V 1 + L 1 V 0 V
+ 2K
0 V 0 + K 0 V 0 V
2 G 1 X 1 V + 2 K 0 X 0 V .
(31)
5. Wick contraction
In this chiral frame L and H are diagonal and one can calculate the propagators for the fluctuations from the kinetic terms in the Lagrangian. On the fibre these kinetic terms are FJ style
Lagrangians of the form (12) for n chiral and n anti-chiral Bosons in flat space. The propagators
for such chiral Lagrangians have previously been considered by Tseytlin [5]. The sum of a chiral
and anti-chiral propagator is proportional to a standard Boson propagator 0 . The difference of
chiral and anti-chiral propagators gives a phase . Full details of this are given in Appendix A.
The general result for our action is that
A B
= 0 HAB + LAB .
(32)
0 contains UV divergence that needs regularisation and renormalisation. The coefficients of
0 will thus contribute to the Weyl anomaly and in turn, to the beta-functionals.
does not contain any divergence and does not therefore contribute to the Weyl anomaly.
Instead parameterises any Lorentz anomaly. One would demand such anomaly vanishes by
setting Seff = 0. This would place additional constraints on the background fields beyond that
of the beta-functionals vanishing. However, when the dust settles all occurrences of in the
effective action cancel out leaving no Lorentz anomaly. This is as expected since we have an
equal number of Bosons of each chirality.
Given (32) we can deduce the form of more complicated Wick contractions which are quartic
2 ) term in the
in and contain derivatives of fluctuations.6 These contractions arise in the O(Seff
exponential of the effective action and must be included since we need to count all logarithmic
divergences (see for example [15]).
When evaluating all contributing terms it will be useful to distinguish base terms which
contain only the base metric g and its derivatives and vielbein terms which come from (31) and
contain derivatives of the vielbein prior to any integrations by parts.
5.1. Single contraction terms
The terms with a single contraction are

1
1
a b gbg a b Y g Y d = a a ggd Y g Y d 0
2
2
6 See Appendix A.
(33)
182
from the base,

1
1
a b HGD 1 XG 1 XD a b = a a HGD 1 XG 1 XD 0
2
2
on the fibre and

B A
1
A
B
A
A
AB
HAB 1 V B 1 V A = 1 V A 1 V A + 1 HAB 1 H
0 ,
2

LAB 1 V B B 0 V A A B A = 1 V A A 0 V A B B C LC A 0 ,

2a HBG 1 XG 1 V B B a B = 0,

b a
1
b
a
a
a
ab
0 ,
gab V b V a = V a V a gab g
2

2a gbg Y g V b b a b = 2a ggb d V b b a b Va a 0
(34)
(35)
(36)
(37)
(38)
(39)
from the vielbeins.

5.2. Double contraction terms
These occur when expanding the exponential of the effective action to second order. Although
there seem myriad possible terms that could contribute, especially from vielbein terms, many
vanish trivially. This because new divergent diagrams must still be one-loop in fluctuations so
contain one loop of indices: the block diagonal form of the metrics and vielbeins ensure the
terms mainly separate into base and fibre terms, with a few cross-terms. We use the propagator
contractions described in Appendix A and note that these terms are a factor of a half down due
the exponential, and a further factor of a half down due to the two sitting on the left-hand side of
(30).
On the base we get

a gg b Y g c gd e Y d a 1 b c 1 e

1
= a ggb g bc a gcd a gbg b gad Y g Y d 0
(40)
2
and on the fibre

a HGB 1 XG c HD E 1 XD a 1 B c 1 E
1
= a HGB HBC a HCD 1 XG 1 XD 0 .
(41)
2
Purely from the vielbein piece of the Lagrangian (31) we have

A
1
1
B C
D
A
B
D
A
AB
= 1 V A 1 V A + 1 HAB 1 H
L 1 V B LCD
0 ,
1 V D 0 0
4 AB
8

A
1
1
B C
D
B
D
= 0 HAB 0 HAB 0 ,
LAB
0 V B LCD
0 V D 1 1
4
8

1
A
B C
D
B
D
= 1 V A A 0 V A B A
L 0 V B B LCD
1 V D 1 0
0 ,
2 AB

A
1
B C
D
A
B
D
A
AB
HAB
= 21 V A 1 V A 1 HAB 1 H
0 ,
1 V B LCD
1 V D 1 0
2
183
B C
D
B
D
= 0,
HAB
1 V B LCD
0 V D 1 1

A
B C
D
B
D
HAB
= 0,
1 V B HCD
1 V D 1 1

1
b c
d
b
d
a
ab
a
a
=
gab
V
g
a
a 0 , (42)
ab
b cd
d
4
=0,1
with one cross-term

a
b c
d
e
2a ggb V b b V a a Y g gce
V d
=0,1

= 2a ggb d V b b a b Vaa + a ggb d g ba Y g Y d 0 .
(43)
5.3. The Weyl divergence

The total Weyl divergence will be given by the coefficient of 0 which we denote by W so
that

1
d 2 WGD 1 XG 1 XD + Wgd Y g Y d 0 .
SWeyl =
(44)
2
On the base the divergence Wgd is given by
1
1
1
Wgd = a a ggd g gab d g ab a ggb g bc a gcd
2
4
2
1 a
1
b
ba
+ gbg gad + a ggb d g g HAB d HAB .
2
8
The divergence on the fibre is
(45)

1
1
WGD = 2 HGD (a H)H1 a H GD .
(46)
2
2
The divergence on the base, (45), can be rewritten as

Wgd = g ab ggs b s ad + s bt t ad
1
1
g gab d g ab + a ggb d g ba g HAB d HAB ,
(47)
2
8
where we recognise the first two terms as part of the Ricci tensor. We now, using the base components of the equation of motion for the fields, add zero to the divergence in the form

1
t ab g ab D gtd Y d t HGD 1 XG 1 XD .
(48)
2
The base divergence becomes
1
Wgd = R gd g HAB d HAB ,
(49)
8
where R gd is the Ricci tensor constructed from the base metric g alone. The fibre components of
the divergence become

1
1
1
WGD = 2 HGD (a H)H1 a H GD t ab g ab t HGD .
2
2
2
(50)
184
5.4. Relation to the doubled Ricci tensor

If we calculate the Ricci tensor of the doubled space (the Ricci tensor of G) for comparison,
and drop terms proportional to HAB d HAB = 0, it also has block diagonal form with

1
1
1
RGD = 2 HGD + (a H)H1 a H GD + t ab g ab t HGD ,
(51)
2
2
2

1
Rgd = R ab + tr g Hd H1 ,
(52)
4
for the fibre and base parts respectively. We see that the Weyl divergence are almost equal
to minus the Ricci tensor except that the term on the base containing the doubled metric H
have an extra factor of 1/2. We note also that the fibre divergence is contracted with 1 X1 X,
whereas if we considered an ordinary sigma model with metric G the fibre piece would be contracted with X X. However, we can use the fibre equations of motion to make WGD contract
X X at the expense of introducing a factor of 1/2. Then, comparing W with R, all terms
containing the doubled metric H would be a factor of 1/2 down. We will see that writing the
doubled metric in terms of the standard sigma model fields h and b takes care of these extra
factors.
6. Doubled renormalisation
One may now proceed directly to regularise and renormalise the divergences coming from 0 .
In the standard way one would dimensionally regularise and introduce a mass scale through,
say, minimal subtraction and the introduction of counter terms. We then absorb all scale dependence to define the renormalised couplings

{G, K, L} G R (), KR (), LR ()
(53)
producing the renormalised action

1
SR =
d 2 G R () 1 X 1 X + LR () 1 X 0 X
2

+ KR () 0 X 0 X .
(54)
We can calculate the beta-functions from this by differentiating the renormalised couplings with
respect to the log of the mass scale giving

0
WAB
0
0
G
K
L
,
(55)
=
=
= 0.
0
Wab
0 Wab
Demanding the vanishing of these beta-functions gives the background field equations.
7. Equivalence with standard sigma model
Instead of working directly with these doubled beta-functions we shall show the equivalence
to the standard sigma model by expanding out the Weyl divergence in terms of the non-doubled
metric and B-field and eliminate the extra doubled coordinates before renormalisation. Since
this can be cast in a well known form for trivial base metric we will proceed putting gab = ab .
Expanding the Weyl divergence W using (2) we obtain

1 (r + bh1 rh1 b bh1 s sh1 b)ij (sh1 bh1 rh1 )i j
WAB =
,
2
(h1 s h1 rbh1 )i j
(h1 rh1 )ij
1
Wab = tab ,
4
where we have defined

rij = 2 h a hh1 a h a bh1 a b ij ,

sij = 2 b a bh1 a h a hh1 a b ij ,

tab = tr h1 a hh1 b h h1 a bh1 b b .
185
(56)
(57)
(58)
(59)
(60)
Recall that X A = XA = (X i , X i ). We now wish to eliminate the dual coordinates X i from the
Weyl divergence using the constraint
dXA = LAB HBC dXC ,
(61)
which implies that

1 X i = hij 0 X j + bij 1 X j .
(62)
We can observe that the right-hand side of the above has a sensible interpretation in terms of
the standard sigma model; it is proportional to the canonical momentum. On using the constraint
we find that
1
1
WAB 1 XA 1 XB = rij X i X j + sij X i X j .
(63)
2
2
Thus, we find that prior to renormalisation, the Weyl divergence part of the effective action is

1
SWeyl =
(64)
d 2 WAB 1 XA 1 XB + Wab Y a Y b 0
2

1
tab
d 2 rij X i X j + sij X i X j +
=
(65)
Y a Y b 0 .
4
2
Demanding that this divergence vanishes constrains the background fields to obey r = s = t = 0.
We now wish to compare this to the standard sigma model in conformal gauge

1
d 2 G X X + B X X
S=
(66)
2
with metric and B-field

hij (Y ) 0
bij (Y ) 0
G =
(67)
,
B =
.
0
ab
0
0
The beta-functionals for this sigma model are [17]7
1
G
= R H H ,
4
1
B
= D H ,
2
7 We have set the dilaton to a constant.
(68)
(69)
186
where H = B + B + B . On substitution of our ansatz for B and G we find that

the non-vanishing components are

1
1
ijG = r + tr h1 a h a h ,
(70)
2
2
ij

1 tab
G
=
ab
(71)
+ a tr h1 b h ,
2 2

1
1
ijB = s + tr h1 a h a b ,
(72)
2
2
ij

1
B
= tr h1 a bh1 b h + h1 a hh1 b b .
ab
(73)
4
The Weyl divergent part of the effective action which produces these beta-functions after renormalisation is,

1
G
d 2 ijG X i X j + ab
Y a Y b
SWeyl =
2

B
+ ijB X i X j + ab
Y a Y b 0 .
(74)
B.
The anti-symmetry of Y a Y b allows us to cancel the divergence that gave rise to ab
The equation of motion for the base coordinate Y is
2 2 Y a = a hij X i X j + a bij X i X j ,
(75)
tr(h1 a h)
so upon multiplying both sides

and integrating by parts we have

1
d 2 tr h1 a h a hij + a bij X i X j
2

1
2 a

2
= d tr h a h Y = d 2 a tr h1 b h Y a Y b
(76)
(77)
so that (74) reduces to

1
tab
i j
i
j
a b
rij X X + sij X X +
Y Y 0 .
SWeyl =
(78)
4
2
This agrees with what we found previously from the doubled formalism in (65). Thus after
integrating out the dual coordinate the doubled formalism gives exactly the same divergent terms
as the standard string sigma model.
The construction can be straightforwardly extended to include a non-trivial base metric g(Y ).
In this case the following additional terms
1 t
(79)
ab g ab t hij ,
2

1
1
R ab + tr h1 t h t gab tr h1 t h a gtb ,
(80)
4
2
1 t
ab g ab t bij ,
(81)
2
are required to reproduce the usual beta-functionals (70), (71) and (72), respectively. These terms
do indeed follow from the doubled geometry Weyl divergences (49) and (50) after application of
the equations of motion. In fact one can immediately see R ab , the Ricci tensor of the base metric
g in (49).
187
8. Conclusion
In summary, we have been able to calculate the one-loop Weyl divergence of the duality
symmetric string which upon renormalisation gives rise to the beta-functionals for the doubled
geometry metric couplings. For the fibre coordinates these are the obvious geometric quantity,
the doubled target space Ricci tensor. For base coordinates, the terms in the Ricci tensor that
contain the fibre metric H pick up an extra factor of a half. However, when we interpreted the
results in terms of the non-doubled fields these factors are taken care of and indeed we are left
with exactly the same Weyl divergence as for the standard sigma model.
In this calculation there are some notable features that we wish to draw attention to. First,
the topological term which is crucial in establishing equivalence of partition functions with the
standard sigma model, played no role in this calculation. Second, the non-covariant structure of
the action (15) meant that unlike the calculation for the standard string, the gauge terms do
make a contribution to the divergence and increase the computational difficulty. Third, the Bfield is incorporated into the doubled metric and there is no anti-symmetric term in the action.
Fourth, the chiral nature of the fibre coordinates suggests that one should be concerned about
any potential Lorentz anomaly. This anomaly actually vanished, cancelling between the Bosons
of opposite chirality. Finally, we found that the L coupling containing the O(n, n) fibre metric L
does not get renormalised.
8.1. Discussion
There are a number of assumptions in this work that would be interesting to explore further.
We assumed that the connection in the fibration was identically zero. To include such a connection would add off diagonal elements to the doubled-space metric and would also require a
suitably generalised constraint (with derivatives promoted to covariant derivatives). On applying
the PST procedure to produce an action akin to (15) one finds that terms involving the connection
appear in both the metric G and the invariant O(d, d) metric L. Evaluation of the divergence
in this case would be more challenging.
As with other treatments of the duality symmetric formalism we had to specify that the fibre
metric depended only on the base coordinates. It would be nice to relax this assumption. The
difficulty with doing so is that the chirality constraint would have to be modified as would its
implementation with a suitable generalisation of the PST action. Another interesting and perhaps
more democratic generalisation along this line would be the doubling of all coordinates. It is
remarkable that the background field equations obtained required no use of the presence of any
Killing directions implying that the doubled formalism is more general than one might have
been first led to believe. In this paper we have not included the doubled space dilaton, , which
is related to the standard dilaton, , through
1
ln det h.
2
This is introduced into doubled formalism with the usual FradkinTseytlin action

1
d 2 (Y )R (2) .
Sdil =
8
=
(82)
(83)
The doubled dilaton is T-dual invariant. However integrating out the dual coordinates from the
doubled action (15) produces a determinant factor which correctly reproduces the transformation
188
of the standard dilaton under T-duality [4]. Similar invariant dilatons have been used in other
treatments of T-duality as reviewed in [18].
The relation between the two dilatons (82) tells us that if we wish to set the doubled dilaton
to be a constant it is inconsistent to also set to be a constant as we did when showing the
equivalence of Weyl divergence. Instead, one should have = 12 ln det h and consider the full
beta-functions for the metric and B-field in the standard sigma model
G
= R + D D ,
(84)
1
1
B
(85)
= D H + D H .
2
2
The addition of these dilaton terms actually produces an exact match to the doubled space betafunctions without further manipulation using the equations of motion as we did in Section 7.
A further generalisation is to consider a general non-constant dilaton in the doubled theory
producing a beta-functional . It is clear that the leading term in will be proportional to
26 c as is the case for the standard string. The 26 comes from an integration over world sheet
metrics which remains unchanged upon doubling the target space. c is the central charge of
the theory which remains the same after doubling since although we now have d + 2n bosons
(d being the dimension of the non-doubled space), 2n of these are chiral and so contribute only
a half each to the central charge. To be concrete about the full equivalence of the beta-functions
with a general dilaton would require extending our analysis through to two loops. Indeed, higher
loop analysis could still provide interesting corrections that are not present at leading order.
Acknowledgements
D.S.B. is supported by EPSRC grant GR/R75373/02 and would like to thank DAMTP and
Clare Hall college Cambridge for continued support. N.B.C. would also like to thank DAMTP
for continued support. This work was in part supported by the EC Marie Curie Research Training
Network, MRTN-CT-2004-512194. D.C.T. is supported by a STFC studentship. We would like
to thank Chris Hull, James Lucietti, Andrew Low and Malcolm Perry for discussions.
Appendix A. Propagators
To find the propagators for the fluctuations we look at the kinetic terms for the scalars when
we have rotated the Lagrangian (30) to the chiral frame. The fluctuations have either chiral (+)
or anti-chiral () FloreaniniJackiw style kinetic terms with action

1
d 2 1 .
S =
(A.1)
2
This action yields momentum-space loop propagators

1 ip.( )
d 2p
e
,
(, ) = i
(A.2)
2
(2) p1 p
and we will normally write to indicate the limit. Simply by examining the integrals
we see
+ + = 20 ,
(A.3)
+ = 2,
(A.4)
189
d 2 p 1 ip.( )
where 0 ( ) = i (2)
is the propagator for a non-chiral boson8 and we
2 p2 e
take this as the definition of . Of course these propagator integrals are divergent and we must
regularise and then renormalise to find the beta-function.
The propagators can be calculated in z-space after Wick rotation [5] with z = + it = +
= .
Using a z 0 regularisation such that z
1 = (2) (z) one
and = + ,
finds
1
ln(z z ),
2
1
ln(z z ),
(z, z ) =
2
1
ln |z z |2 = 20 (z, z ),
+ (z, z ) + (z, z ) =
2
1
i
z z
+ (z, z ) (z, z ) =
= arg(z z ) = 2 ,
ln
2 z z
+ (z, z ) =
(A.5)
(A.6)
(A.7)
(A.8)
where in (A.7) we have noted after regularisation we have the same relation as (A.3) to the
standardly normalised two-dimensional scalar propagator in this regularisation scheme.
Terms in the path integral of the effective action that are proportional to 0 will be those
related to a breakdown in Weyl invariance whereas terms proportional to will correspond to a
breakdown in worldsheet Lorentz invariance [5,6]. Looking at the form of (A.7) and (A.8) we
can see scaling of z shifts 0 and not and rotation by a phase shifts and not 0 . One can
also see that in flat space the propagator between two X coordinates or two X coordinates is 0
whereas between an X and an X it is . The beta-function should come from terms proportional
to 0 in the exponential of the action and one can obtain it by regularising and renormalising 0
using dimensional regularisation as is standard. Terms proportional to would mean additional
constraints on the background to ensure worldsheet Lorentz invariance and would indicate a
difference from the ordinary sigma model. We find such terms cancel giving agreement with
the standard formulation. We would expect this as we have equal numbers of each chirality of
boson.
Looking again at our general d-dimensional doubled action for the fibre coordinates in the
chiral frame (indicated by barred indices) we have fluctuation kinetic terms given by

1

S=
(A.9)
HA B 1 A 1 B + LA B 1 A 0 B ,
2
where in this frame the metrics are diagonal:

1 0
1 0
H=
,
L=
.
0 1
0 1
(A.10)
Thus the general propagator for A is given by

A
1 0
0 0
+ +
(z) B (z) =
0 0
0 1
1
1
= (H + L)+ + (H L)
2
2
8 We use (+, ) signature on the worldsheet.
(A.11)
(A.12)
190
1
1
= HAB (+ + ) + LAB (+ )
2
2
= 0 HAB + LAB .
(A.13)
(A.14)
We can use this result and a Wick contraction procedure, described for the ordinary string in [15],
to determine the divergent behaviour of higher-order propagator contractions which appear in the
expansion of the exponential of our action.
For example

i
d 2 ( )A 1 ( )B ( )C 0 ( )D

1
1

A C
A C
=i d
(H + L) + (p) + (H L) (p) eip.( )
2
2

1
1

(H + L)B D + (q) + (H L)B D (q) q1 q0 eiq.( )

2
2

1
1

+ i d 2 (H + L)AD + (p) + (H L)AD (p) p1 eip.( )
2
2

1
1

B C
B C
(H + L) + (q) + (H L) (q) q0 eiq.( ) .

2
2
2
(A.15)
The integral gives (2)2 (p + q) and putting in the forms of the integrals in from
(A.2) gives

i
d 2 ( )A 1 ( )B ( )C 0 ( )D

A C
1
1
d 2p
1
1
+
(H
L)
(H
+
L)
p1 p+
p1 p
(2)2

B D
(H L) (H + L)
(A B)
p+
p
(HH LH HL + LL)
1
+(HH + LH HL LL)( + 0 )

(A B)
8 +(HH LH + HL LL)(+ 0 )
=
i
8
(A.16)
(A.17)
+(HAC HB D + LH + HL + LL)+

= 0 HAC LB D + LAC HB D LAC LB D (A B),
(A.18)
where
indicates equality up to convergent terms which are irrelevant for our purpose.
A similar procedure can be used to calculate the two-propagator contractions with any combination of worldsheet derivatives. One can also allow indices on the base; when all four indices
are on the base the calculations are as for the standard sigma model (see for example [15]) and
since there is no basefibre propagator the only other allowed possibility is to have two indices
on the base and two on the fibre. The results can be compactly summarised in terms of the total
191
space metric G and L as9

1 1 = 0 (GG LL)( ) ,

1 0 = 0 (GL + LG)() LL() ,

0 0 = 0 (GG + 3LL)() (LG + GL)() ,

where GG
)
(
(A.19)
(A.20)
(A.21)
represents G G G G and () understood in the same way.
References
[1] C.M. Hull, Global aspects of T-duality, gauged sigma models and T-folds, hep-th/0604178;
A. Dabholkar, C. Hull, Generalised T-duality and non-geometric backgrounds, JHEP 0605 (2006) 009, hep-th/
0512005;
J. Shelton, W. Taylor, B. Wecht, Nongeometric flux compactifications, JHEP 0510 (2005) 085, hep-th/0508133;
J. Shelton, W. Taylor, B. Wecht, Generalized flux vacua, hep-th/0607015;
C.M. Hull, R.A. Reid-Edwards, Flux compactifications of string theory on twisted tori, hep-th/0503114;
K. Becker, M. Becker, C. Vafa, J. Walcher, Moduli stabilization in non-geometric backgrounds, hep-th/0611001.
[2] M.J. Duff, Duality rotations in string theory, Nucl. Phys. B 335 (1990) 610;
J.H. Schwarz, A. Sen, Duality symmetric actions, Nucl. Phys. B 411 (1994) 35, hep-th/9304154.
[3] C.M. Hull, A geometry for non-geometric string backgrounds, JHEP 0510 (2005) 065, hep-th/0406102.
[4] C.M. Hull, Doubled geometry and T-folds, hep-th/0605149.
[5] A.A. Tseytlin, Duality symmetric formulation of string world sheet dynamics, Phys. Lett. B 242 (1990) 163.
[6] A.A. Tseytlin, Duality symmetric closed string theory and interacting chiral scalars, Nucl. Phys. B 350 (1991) 395.
[7] D.S. Berman, N.B. Copland, The string partition function in Hulls doubled formalism, Phys. Lett. B 649 (2007)
325, hep-th/0701080.
[8] E. Hackett-Jones, G. Moutsopoulos, Quantum mechanics of the doubled torus, JHEP 0610 (2006) 062, hepth/0605114.
[9] S.P. Chowdhury, Superstring partition functions in the doubled formalism, arXiv: 0707.3549 [hep-th].
[10] R. Gregory, J.A. Harvey, G.W. Moore, Unwinding strings and T-duality of KaluzaKlein and H-monopoles, Adv.
Theor. Math. Phys. 1 (1997) 283, hep-th/9708086;
J.A. Harvey, S. Jensen, Worldsheet instanton corrections to the KaluzaKlein monopole, JHEP 0510 (2005) 028,
hep-th/0507204.
[11] D. Tong, NS5-branes, T-duality and worldsheet instantons, JHEP 0207 (2002) 013, hep-th/0204186.
[12] P. Pasti, D.P. Sorokin, M. Tonin, On Lorentz invariant actions for chiral p-forms, Phys. Rev. D 55 (1997) 6292,
hep-th/9611100.
[13] R. Floreanini, R. Jackiw, Selfdual fields as charge density solitons, Phys. Rev. Lett. 59 (1987) 1873.
[14] J. Honerkamp, Chiral multiloops, Nucl. Phys. B 36 (1972) 130;
L. Alvarez-Gaume, D.Z. Freedman, S. Mukhi, The background field method and the ultraviolet structure of the
supersymmetric nonlinear sigma model, Ann. Phys. 134 (1981) 85.
[15] E. Braaten, T.L. Curtright, C.K. Zachos, Torsion and geometrostasis in nonlinear sigma models, Nucl. Phys. B 260
(1985) 630.
[16] S. Mukhi, The geometric background field method, renormalization and the WessZumino term in nonlinear sigma
models, Nucl. Phys. B 264 (1986) 640.
[17] C.G. Callan, E.J. Martinec, M.J. Perry, D. Friedan, Strings in background fields, Nucl. Phys. B 262 (1985) 593.
[18] A. Giveon, M. Porrati, E. Rabinovici, Target space duality in string theory, Phys. Rep. 244 (1994) 77, hep-th/
9401139.

9 We will simplify notation by using A B C D = i d 2 ( )A ( )B ( )C ( )D it will always
1
0
1
0
appear in the expansion of the exponential of the effective action in this form.
Nuclear Physics B 791 [FS] (2008) 193230

www.elsevier.com/locate/nuclphysb
Noncompact sigma-models: Large N expansion and

thermodynamic limit
A. Duncan a , M. Niedermaier b,,1 , P. Weisz c
a Department of Physics, 100 Allen Hall, University of Pittsburgh, Pittsburgh, PA 15260, USA
b Laboratoire de Mathematiques et Physique Theorique, CNRS/UMR 6083, Universit de Tours,
Parc de Grandmont, 37200 Tours, France

c Max-Planck-Institut fr Physik, Fhringer Ring 6, 80805 Mnchen, Germany
Received 3 July 2007; accepted 20 July 2007

Available online 31 July 2007
Abstract
Noncompact SO(1, N) sigma-models are studied in terms of their large N expansion in a lattice formulation in dimensions d 2. Explicit results for the spin and current two-point functions as well as for the
Binder cumulant are presented to next to leading order on a finite lattice. The dynamically generated gap is
negative and serves as a coupling-dependent infrared regulator which vanishes in the limit of infinite lattice
size. The cancellation of infrared divergences in invariant correlation functions in this limit is nontrivial
and is in d = 2 demonstrated by explicit computation for the above quantities. For the Binder cumulant the
thermodynamic limit is finite and is given by 2/(N + 1) in the order considered. Monte Carlo simulations
suggest that the remainder is small or zero. The potential implications for criticality and triviality of the
theories in the SO(1, N) invariant sector are discussed.
1. Introduction
In quantum field theories with non-Abelian symmetries and dynamical mass generation the
large N expansion often provides a qualitatively correct and quantitatively reasonable description
of the physics of the systems. Specifically in sigma-models with a compact global symmetry
group the expansion is known to be an asymptotic expansion [1] and when slightly ad hoc applied
E-mail address: max@phys.univ-tours.fr (M. Niedermaier).

1 Membre du CNRS.
doi:10.1016/j.nuclphysb.2007.07.020
194
A. Duncan et al. / Nuclear Physics B 791 [FS] (2008) 193230
to low orders at fixed small N sometimes gives surprisingly accurate results, see, e.g., [2,3] for
the renormalized coupling. In a lattice formulation one starts off on a finite lattice, the associated
finite volume mass gap then is uniformly bounded away from zero, and in the large N series
for invariant correlation functions the limit of infinite lattice size (also called the thermodynamic
limit) can safely be taken termwise.
A study of the large N expansion in sigma-models having a noncompact SO(1, N) internal
symmetry group has been initiated in [4,5]. A large value of N in this case is also physically
relevant for the granular limit of random Hamiltonians describing disordered electrons with N
orbitals per site [6]. In a lattice formulation of the SO(1, N) sigma-models again a gap is dynamically generated in the large N expansion, which is however negative and vanishes as the size of
the lattice goes to infinity. Effectively the gap now acts as a subtle, coupling-dependent, infrared
regulator and the technicalproblem consists in studying the coordinated limit V of lattice sums of the form V1n k1 ,...,kn fV (k1 , . . . , kn ), where fV carries an explicit V -dependence
via the gap. The sums associated with individual Feynman diagrams of the large N expansion
will typically diverge in the limit. The issue whether or not in the combinations entering invariant
correlation functions the infrared divergences cancel is analogous to the one encountered in the
perturbation theory of compact sigma-models [7,8] and is the subject of the present paper. Since
this issue is most critical in the two-dimensional systems we examine the limit specifically in this
case, although our finite volume results are valid in all dimensions d 2. We compute a number
of physically interesting quantities to leading and subleading order and show that they indeed do
have a well-defined thermodynamic limit. Concretely we consider the spin two-point function,
the two-point function of the Noether current, and the Binder cumulant.
The Binder cumulant U is defined in terms of the zero momentum limit of the connected
four-point function. In massive scalar field theories it serves to define an intrinsic measure of the
interaction strength and has been used to explore triviality issues. In a massless theory, like the
systems considered here, there is no obvious reason why U should have a finite thermodynamic
limit. Somewhat surprisingly we find that U does have a finite and nonzero limit to leading and
subleading order, which is moreover independent of and given by 2/(N + 1). Supported also by
Monte Carlo simulations we conjecture that the infinite volume limit of the exact U is also very
close to 2/(N + 1). The potential implications for criticality and triviality in the SO(1, N)
invariant sector of the theory will be discussed in the conclusions.
The rest of the article is organized as follows. In the next section we review a result from
a previous paper [5] which on a finite lattice allows one to do large N computations in the
simpler compact models and then transfer the results to the noncompact ones via a large N
correspondence. This correspondence has an interesting interplay with the SchwingerDyson
equations and instead of going through the (fairly routine) diagrammatic computations we merely
present the results as solutions of the large N expanded SchwingerDyson equations with the
correct initial data. Expressions for the two- and four-point functions to leading and subleading
order are given (valid on a finite lattice in all dimensions d 2), from which also the two-point
function of the Noether current and the Binder cumulant can be obtained in the same order. The
thermodynamic limit in d = 2 of the local quantities and of the Binder cumulant are studied in
Sections 4 and 5, respectively.
2. Large N expansions of compact and noncompact models
In compact sigma-models the large N expansion is a saddle point expansion based on a generating functional obtained by dualizing the spins, i.e., by imposing the constraint via a Lagrange
195
multiplier field and performing the Gaussians. The counterpart of this duality transformation is
somewhat ill-defined in the noncompact models. The large N expansion can nevertheless be justified and on a finite lattice the expansion coefficients for invariant correlation functions can be
inferred from those in the compact model [5]. This large N correspondence allows one to do
computations in the compact model, where no gauge-fixing is required, and the familiar framework can be used. Here we briefly summarize the correspondence and present explicit results
for two and four-point functions in Section 3. The results of Sections 2 and 3 are valid in all
dimensions d 2.
2.1. Definitions
Here we recall the notation and the definitions for the invariant correlation functions considered and their generating functionals. We consider the SO(N + 1) spherical and the SO(1, N)
hyperbolic sigma-models in two dimensions with standard lattice action, defined on a hypercubic lattice Zd of volume V = || = Ld . The dynamical variables (spins) will be
denoted by nax , x , a = 0, . . . , N , in both cases, and periodic boundary conditions are
assumed throughout nx+L = nx . The constraint is n n = 1 in both cases, but with different dot products; namely a b := a 0 b0 + a 1 b1 + + a N bN in the compact model, and
a b := a 0 b0 a 1 b1 a N bN in the noncompact model. Clearly S N = {n RN+1 | n n = 1}
is the N -sphere and H N = {n R1,N | n n = 1, n0 > 0} is the upper half of the two-sheeted
N -dimensional hyperboloid. The lattice actions are

(nx nx+ 1) =
nx (n)x 0,
S =
(2.1)
2 x
x,
where the upper sign refersto the compact model and the lower sign to the noncompact model.
The Laplacian is xy = [2x,y x,y+ x,y ], as usual. We write
d+ (n) = dN+1 n (n n 1),

d (n) = 2dN+1 n (n n 1) n0 ,
(2.2)
for the invariant measure on S N and H N , respectively. Further (n, n ) is the invariant point
measure on S N , H N , and n = (1, 0, . . . , 0). Note that the measure d+ (n) is normalized while
H N has infinite volume.
In the compact model we consider the generating functional,

1
d+ (nx ) exp S+ +
Hxy (nx ny 1) ,
exp W+ [H ] = N
(2.3)
2 x,y
x
where Hxy 0 is a source field and the normalization N is such that W [0] = 0. For the noncompact model we consider the generating functional

d (nx ) nx0 , n
exp W [H ] = N
x

1
exp S +
Hxy (nx ny 1) ,
2 x,y
(2.4)
where now Hxy < 0 sources give damping exponentials, and one spin at site x0 is fixed in order
to make the generating functional well defined.
196
Connected 2r point functions are defined by

1
W [H ] =
W,r (x1 , y1 ; . . . ; xr , yr )Hx1 y1 . . . Hxr yr ,
r!2r
r1
W,r (x1 , y1 ; . . . ; xr , yr ) := hx1 y1 . . . hxr yr W [H ]|H =0 ,
hxy :=
.
Hxy
(2.5)
In particular W,1 (x, y) =

nx ny 1, W,2 (x1 , y1 ; x2 , y2 ) :=
nx1 ny1 nx2 ny2
nx1
ny1
nx2 ny2 , where
are the functional averages with respect to N 1 eS . Note that
W,r (. . . ; x, x; . . .) = 0.
2.2. The 1/N expansion
The goal in the following is to construct these invariant correlation functions in a large N
asymptotic expansion. That is, := (N + 1)/ is kept fixed and the coefficient functions W,r
in

1
r
(s)
W,r (x1 , . . . , yr ) =
(2.6)
W (x1 , . . . , yr ),
r1
(N + 1)s ,r
(N + 1)
s=0
are sought, with the understanding that the right-hand side of (2.6) provides a valid asymptotic
expansion of the exact Wr , initially on a finite lattice.
(s)
is rather
The diagrammatic algorithm for the computation of the coefficient functions W+,r
straightforward in the compact model, see, e.g., [9]. From [1] it is also known to provide a valid
(s)
asymptotic expansion. Direct computation of the functions W,r in the noncompact model is also
possible [5], although due to the gauge fixing the computations are considerably more tedious
than in the compact model. In [10] it will be shown that this algorithm also provides a valid
asymptotic expansion (2.6) for the W,r .
One of the advantages of a lattice formulation for these systems is that there is an exact
(s)
correspondence [5] between the functions W,r in the noncompact model and their counterparts
(s)
W+,r in the compact model, valid on a finite lattice in all dimensions d 2:
(a) The coefficient functions W,r are translation invariant and can be expressed in terms of
D (x) = D(x)| , with D(x) the free propagator of squared mass and with (, V )
the solutions of the gap equations D(0) = 1 discussed further in Section 3.1.
(s)
(s)
(b) For all r 1, s 0, there exists unique functionals Xr [D]() of D such that Wr,+ =
(s)
(s)
(s)
Xr [D+ ]() are the coefficients in the compact model and Wr, = (1)r Xr [D ]()
are the coefficients in the noncompact model.
As a consequence the computations only have to be done in the compact model, and the result
in the noncompact model can be obtained via (b). In the next section we will compute a subset
of correlation functions, but instead of doing the computation using the 1/N Feynman rules we
present the results and verify that they solve the associated SchwingerDyson equations.
3. SchwingerDyson equations
In the compact model the SchwingerDyson equations for the functions (2.5) have, to our
knowledge, first been formulated by M. Lscher [11]. The derivation is readily extended to non-
(0)
W1
(0)
W2
(0)
(1)
W3
197
(1)
W1
W2
Fig. 1. Recursion pattern for the solution of the large N expanded SD equations.
compact models and is reproduced in [5]. Here we just record the basic equations:

z hzy W hzx W hzx hxy W (hzx W )(hxy W ) z=x

+
Hxz hzy W hzx W hxy W hxz hxy W (hzx W )(hxy W )
z =x
N(1 xy )(hxy W + 1) = 0,
(3.1)
where the upper sign corresponds to the compact model and the lower sign to the noncompact
model. In terms of the multi-point functions (2.5) these equations amount to an infinite coupled
system of nonlinear partial differential equations. As such boundary conditions have to be specified; without them even the exact equations (3.1) do not determine their solution uniquely, see
[5] for a counter example. However, since (3.1) does not contain a closed equation for any of the
Wr , it is difficult to impose such boundary conditions in practice.
In contrast, the large N ansatz (2.6) effectively converts the Wr equations into a hierarchy
which can be solved recursively and where initial conditions can be specified. The recursion
(s)
pattern for the Wr , r + s > 1, functions is given in Fig. 1. To compute a given coefficient all
quantities having arrows pointing towards it are needed.
(s)
The first few equations for the Wr are spelled out in Appendix A; a closed formula for the
generic equation can also be given and used to show the recursion pattern in Fig. 1 by induction.
The key assumption in our use of the large N expanded SchwingerDyson equations will be
that at each recursion step in Fig. 1 there exists a solution and that the solution is unique. This
(0)
assumption implies that, once W1 has been specified, there will be an infinite sequence of
(s)
functions Wr , r + s > 1, uniquely associated with it, which in turn determine the series (2.6)
(0)
uniquely for each Wr . The choice of W1 is ultimately determined by the physics problem one
seeks to study; different choices are possible [5] for the same (initial) equation (A.1).
The existence and uniqueness of a solution at each recursion step of Fig. 1 is presumably
difficult to establish directly from the equations. In terms of the underlying discretized functional
integral (which solves the exact equation (3.1) by construction) existence and uniqueness of
the solution to the recursive equations amount to the existence of a well-defined asymptotic
expansion of the form (2.6). For the generating functionals W+ [H ] and W [H ] the latter is
guaranteed by the results of [1] and [10], respectively. This provides an indirect justification of
the assumption stated in the preceding paragraph.
It also provides the rationale for the procedure adopted in Sections 3.2 and 3.3: we present the
(s)
(s)
expressions for W,1 and W,2 to leading and subleading order (s = 0, 1) and claim that they
(0)
solve the SchwingerDyson equations with the correct initial conditions, W1, , respectively.
(s)
The required equations are tabulated in Appendix A. The verification that the W,r presented
198
indeed solve these equations is straightforward and is omitted. It is however far shorter than the
diagrammatic computation in either the compact or the noncompact model.
Though in this paper we shall be concerned with the large N expansion exclusively, let us add
that the SchwingerDyson equations (3.1) can also be subjected to a perturbative expansion, i.e.,
the ansatz (2.6) is replaced with one in terms of powers of 1/. The recursive pattern determining
the coefficients is similar to that in Fig. 1, but the differential part of the equations now involves
a linear differential operator with constant coefficients. This has been used by Lscher [11] to
show directly from the equations that each recursion step has at most one solution. The existence
of a solution of course follows from the diagrammatic algorithm as described (on the lattice)
by Hasenfratz [12]. Once the results are known the perturbatively expanded SchwingerDyson
equations provide an efficient way to verify them.
3.1. 2- and 4-functions to leading order
(0)
The leading order two-point functions W,1 provide the starting point for the recursion. They
have to solve Eq. (A.1) but in order to pick a specific solution further information has to be added.
The appropriate solutions turn out to be given by
W,1 (x, y) = D (x y) 1 ,
(0)
(3.2)
with
D (x) =
1 eipx
,
V p Ep +
where the sum goes over p = 2

L (n1 , . . . , nd ), n = 0, 1, . . . , L 1 and Ep =
p
p = 2 sin 2 . Further are the particular solutions to the gap equations
1 = D (0) =
1
1
,
V p E p +
(3.3)
d
2
=1 p
with
(3.4)
4
sin2 /L in the noncompact
obeying + > 0 in the compact model and 0 > > 2d+1
model [5]. The rationale for the choice of these solutions of (A.1) is that their properties are
necessary for the stability of the expansions, see [1] for the compact and [10] for the noncompact
(s)
(0)
model. All W,r , r + s > 1, are then in principle uniquely determined by the W,1 , and we shall
simply present the solutions of the associated SchwingerDyson equations.
The solution of (A.3) for the 4-point function in leading order is
(0)
(x1 , y1 ; x2 , y2 )
W,2
= D (x1 x2 )D (y1 y2 ) + D (x1 y2 )D (y1 x2 )

D (x1 u)D (y1 u) (u v)D (v x2 )D (v y2 ),
2
(3.5)
u,v
where (u v) is defined by

D (x u)2 (u v) = xv ,
u
(3.6)
199
so that
(u) =
(k) :=
1 eiku
,
V
(k)
k
1
V
1
.
(Ep + )(Ek+p + )
(3.7)
Obviously in the compact model + (k) > 0 for all k. In contrast in the noncompact model
one has (k) < 0, k = 0 and (0) > 0, a fact of importance for the validity of the large N
expansion [5]. (u) is the propagator of the auxiliary field in the functional treatment of the
1/N expansion, and correspondingly the last term in (3.5) corresponds to a tree diagram with an
intermediate auxiliary field propagator.
3.2. 2- and 4-point functions to next-to-leading order
The solution of (A.2) for the 2-point function in next-to-leading order is
(1)
W,1
(x, y) = 2q
W (0) (x, y)
,1

D (x u)D (u v) (u v)D (v y),
2
(3.8)
u,v
where
q =
=
1
D (u) (u v)D (u v)D (v)
(0) u,v

1
1
(0)V 2 p q (Ep + )2 (Ep+q + ) (q)
and the partial derivative
(3.9)
means 2 (0)
(at fixed volume). In particular
(0)
W (x, y) = D,2 (x y)
,1

1 eip(xy)
:=
D (w x)D (w y) =
.
V p (Ep + )2
w
(3.10)
Note (0) = D,2 (0). The first term on the rhs of (3.8) corresponds to the tadpole diagram and
the second term to the nontrivial self-energy diagram.
Finally the solution for the next-to-leading 4-point function is given by
(1)
(0)
W,2 (x1 , y1 ; x2 , y2 ) = 2q
W (x1 , y1 ; x2 , y2 )
,2

(0)
W,2 (x1 , y1 ; u, v)D (u w) (u w)
2
u,v,w,z
1
(v z)W,2 (x2 , y2 ; w, z)
D

(0)
+2
W,2 (x1 , y1 ; u, v)D (u v) (u w)
(0)
u,v,w,z
200

(0)
(v z)D (w z)W,2 (x2 , y2 ; w, z)

(0)
(0)
W,2 (x1 , y1 ; u, v) (u v)W,2 (x2 , y2 ; u, v).
(3.11)
u,v
Again the separate contributions in (3.11) become more apparent when drawn as corresponding
Feynman diagrams.
3.3. Two-point current correlation function
In both models the Noether currents are given by

Jab (x) = nax nbx nbx nax , 0 a, b N.

The invariant two-point function of these currents is

J ab (x)Jab (y)+ ,
J, (x, y) := a,b
ab
cd
a,b,c,d
ac bd J (x)J (y) .
It obeys the Ward identity

J, (x, y) = 2NE (x,y x,y+ ),
(3.12)
(3.13)
(3.14)
where f (x) = f (x) f (x )

and in the noncompact case (3.14) holds for x = x0 only.
Further
E =
nx nx+ ,
(3.15)
is independent of x because of translation invariance. One way of obtaining (3.14) is by specializing the pre-SchwingerDyson equation (3.39) of [5] to O = Jcd (y) and using the completeness
relations (3.38) in [5].
The two-point function can be expressed in terms of the 2- and 4-point functions of the spins
according to
J, (x, y) = 2 2 W,2 (x, y; x + ,

y + ) W,2 (x, y + ; x + ,
y)
+ W,1 (x, y)W,1 (x + ,
y + ) W,1 (x, y + )W,1 (x + ,
y)
+ W,1 (x, y) + W,1 (x + ,
y + ) W,1 (x, y + )

W,1 (x + ,
y) .
The current correlation function has accordingly a 1/N expansion of the form:

1
(s)
J
(x, y).
J, (x, y) = 2N (N + 1)
(N + 1)s ,
(3.16)
(3.17)
s0
In the lowest order we have
(0)
(0)
(0)
(0)
y + ) W,1 (x, y + )
J, (x, y) = 1 W,1 (x, y) + W,1 (x + ,

(0)
(0)
(0)
W,1 (x + ,
y) + W,1 (x, y)W,1 (x + ,
y + )
(0)
(0)
W,1 (x, y + )W,1 (x + ,

y).
(3.18)
201
Inserting the solution (3.2) one gets

(0)
J, (x, y) = D (x y)D (x + y ) D (x y )D (x + y),
(3.19)
the Fourier transform of which is

(0)
(0)
eiqx J, (x, 0)
J, (q) =
x

sin(p + q/2) sin(p + q/2)
i
2
= exp [q q ]
.
2
V p
(Ep + )(Ep+q + )
It is seen to satisfy

(0)

1 eiq J, (q) = 1 eiq D ( ),
(3.20)
(3.21)
which is the Ward identity (3.14) to lowest order 1/N .

In next order we have
(1)
(0)
(0)
(0)
J,
(x, y) = J,
(x, y) + W,2
(x, y; x + ,
y + )
W,2
(x, y + ; x + ,
y)

(1)
(0)
y + ) + 1
+ W,1 (x, y) W,1 (x + ,

(0)
(1)
y + ) W,1 (x, y) + 1
+ W,1 (x + ,

(0)
(1)
(x, y + )
W,1
(x + ,
y) + 1
W,1

(0)
(1)
y) W,1 (x, y + ) + 1 .
W,1 (x + ,
(3.22)
Using the solutions (3.2), (3.5), (3.8), this becomes
(1)
J, (x, y) = 2q D,2 (x y)D (x + y ) + D (x y)D,2 (x + y )

D,2 (x + y)D (x y ) D (x + y)D,2 (x y )
(w)D (w) D,2 (y x + w)D (x y + )

2
w
+ D,2 (y x + + w)D (x y)
D,2 (y x + w)D (x y )

D,2 (y x + + w)D (x y + )

D (x u)D (y u) (u v)D (x + v)D (y + v)
2
u,v
+2
D (x u)D (y + u) (u v)D (x + v)D (y v).
u,v
(3.23)
Its Fourier transform is simpler in form:

i
(1)
J, (q) = exp [q q ] X,1; (q) + X,2; (q) + X,3; (q) ,

2
with
X,1; (q) =
8q sin(p + q/2) sin(p + q/2)

,
V p (Ep + )2 (Ep+q + )
(3.24)
202
X,2; (q) =
sin(p1 + q/2) sin(p1 + q/2)

8
,
2
V p ,p (Ep1 + )2 (Ep1 +q + )(Ep2 + ) (p1 p2 )
1
X,3; (q)
sin(p1 + q/2) sin(p2 + q/2)
4
. (3.25)
= 2
V p ,p (Ep1 + )(Ep1 +q + )(Ep2 + )(Ep2 +q + ) (p1 p2 )
1
It can be checked to satisfy the Ward identity in the next order in the large N expansion.
3.4. The Binder cumulant
In scalar field theories with a mass gap a renormalized 4-point coupling is defined in terms of
the Binder cumulant
U := 1 +
Here a =

x
( )2
1
=
N +1
nx1 ny1 nx2 ny2 c .
(3.26)
x1 ,x2 ,y1 ,y2
nax and
nx1 ny1 nx2 ny2 c = W2 (x1 , y1 ; x2 , y2 )
nx nx2
ny1 ny2
N +1 1
nx ny2
ny1 nx2
N +1 1
(3.27)
is the usual connected 4-point function, related to the previously used second H -moment as
indicated. In terms of W1 and W2 the Binder cumulant reads

2
x1 ,x2 ,y1 ,y2 W,2 (x1 , y1 ; x2 , y2 )
U =
(3.28)
.
N +1
[ x,y (W,1 (x, y) + 1)]2
The Binder cumulant has accordingly a large N expansion of the form
U =

s=0
1
U,s (, V ).
(N + 1)s+1
(3.29)
The coefficients can obviously be expanded in terms of the coefficients of the 2- and 4-point
functions summed over all arguments:

(s)
w,s (, V ) :=
W,2 (x1 , y1 ; x2 , y2 ),
x1 ,x2 ,y1 ,y2

(s)

W,1 (x, y) + 1 s0 .
,s (, V ) :=
(3.30)
x,y
The two lowest orders

2
U,0 = 2 ,0
w,0 ,
2
3
U,1 = ,0
w,1 + 2,1 ,0
w,0 ,
involve only functions already computed in the previous subsections. We have
(3.31)
2V
,0 =
V
,
,1 =

2V

(x)D
(x)
,
q
2
w,0 = 2,0
4 (0)
203
(3.32)
and
w,0 + 2
r (u, v)r (w, z)
u,v,z,w

1
(v z) + D (u v) (v z)D (w z)
(u w) D (u w)D
r (u, v)2 (u v),

(3.33)
w,1 = 2q
u,v
where
r (x, y) :=
(0)
W,2 (z, w; x, y) =
z,w
1 (0)1 D,2 (x y) .
2
(3.34)
4. TD limit of spin and current two-point functions

Up to this point we have been considering both the compact and noncompact models in arbitrary dimensions d 2. In the following we restrict attention to d = 2. Also numerous articles
have dealt with the 1/N expansion of the compact model so we will in this and in the next section
restrict attention to the noncompact case and drop the minus () suffix on all functions.
The results summarized in Section 2 seemingly suggest a simple relation between the compact and the noncompact models. It is important to stress, however, that the relations hold only
on a finite lattice and for invariant correlators. Physical quantities arising after taking the limit
of infinite lattice size (thermodynamic limit) turn out to be very different in both systems. We
will illustrate this fact by studying the thermodynamic (TD) limit of the coefficients in the 1/N
expansions of the correlators computed in the last section in the noncompact model. The very
existence of the limit is nontrivial in this case because 0 as V , specifically
= (, V )

4
1
+O
.
V ln V
V ln2 V
(4.1)

This means a coordinated limit of lattice sums of the form V1n k1 ,...,kn fV (k1 , . . . , kn ) has to
be taken, where fV via carries an explicit V -dependence. The gap equation (3.4) effectively
acts as a subtle infrared regulator whose usefulness is underlined by the result summarized in
Section 2.2. As mentioned, the sums associated with individual Feynman diagrams will typically
(s)
diverge in the limit. The issue is whether the infrared divergences cancel in the Wr and the
quantities computed in terms of them.
In this section we discuss the limit of the spin and current two-point functions; the limit of the
Binder cumulant is computed in Section 5.
204
4.1. TD limit of the spin two-point function

In the leading order the 2-point function has an infinite volume limit
ipx
e 1
1 eipx 1
,
D(x) :=
W1(0) (x, 0) =
V
Ep +
Ep
p =0
(4.2)
2 d2 p
where here and in the following p means integration over the Brillouin zone 0 (2)
2 . The
infinite volume lattice propagator D(x) is a remarkable function which has been discussed in
detail by Shin [13]. At every lattice point it is given by an expression of the form r1 (x) + r2 (x)/
where ri are rational numbers. As |x| it diverges logarithmically:

1 2
ln x + 2 + 3 ln 2 + O |x|2 ,
4
where 0.577 is Eulers constant.
The next-to-leading term is given by
D(x)
W1(1) (x, 0) =
(1)
1 ipx
e 1 W1 (p),
V
(4.3)
(4.4)
p =0
with
W1(1) (p) =

1
2
1
,
q +
V
(k)(Epk + )
(Ep + )2
(4.5)
(1)
and q = q as in (3.9). For p = 0 the TD limit of W1 (p) does not exist (similarly to the situation
(1)
for the leading order): W1 (0) V ln V ln ln V , reflecting the fact that W1 (x, 0) is an increasing
function of distance |x|.
For p = 0, however the limit exists. To see this we first note that (p) has a TD limit for
p = 0. Indeed using one insertion of (5.17b) below and the gap equation one can rewrite (p)
as

1
2
(p) =
+ J (p) , p = 0,
(Ep + 2)
1 Ek + Epk Ep
J (p) :=
(4.6)
.
V
(Ek + )(Epk + )
k
Throughout we often use the symbol J to denote lattice sums which give rise to convergent
integrals over the Brillouin zone upon taking the infinite volume limit. The limit (p) of
(p) is then given by
1

1
2
(p) =
(4.7)
,
, p = 0, v(p) :=
+ J (p)
Ep v(p)
with

J (p) =
k
Ek + Epk Ep
.
Ek Epk
(4.8)
The properties of the function J (p) will be important later on; we mostly need:

J (, ) = 0,
J (p) 0 for p = 0,
205
(4.9)
and the behavior for p 0

1
(4.10)
ln p 2 5 ln 2 + O p 2 ln p 2 .
2
Eq. (4.10) follows from [13], the positivity in (4.9) follows from Appendix A of [5]. A direct way
to see J (p) 0 is by performing one of the integrations explicitly. This leads to the integral
representation
J (p) =
1
J (p) =
2
2
0

2
dk1
a
b sh(a + b) 4sh2 ( a+b
2 ) p 1
th th +
,
2
2
2
2
shashb 4sh2 ( a+b
2 ) + p 2
(4.11)
1
where a, b > 0 are determined by sh a2 = | sin k21 |, sh b2 = | sin k1 p
2 |. For the numerator of the
last term in the integrand one then has

a+b
4 sinh2
p 12
2

a
b
a+b
k1
k1 p1
p1
= 8 sinh sinh cosh
+ sin
sin
cos
2
2
2
2
2
2

b
a+b
p1
a
+ cos
,
= 8 sinh sinh cosh
(4.12)
2
2
2
2
where = 1. For fixed p1 , k1 the integrand of (4.11) therefore is a monotonically decreasing

function of p2 for 0 < p2 < . By symmetry the same must hold with the roles of p1 and p2
interchanged, so that J (p) J ((, )) = 0, for all p = 0.
Returning to (4.5) we rewrite the sum as
1
1
1
= J (p)
,
V
(k)(Epk + )
(p)
k

1
1
1
1
2
J (p) :=
+
.
2V
Ek + (p k) (p + k) (p)
(4.13)
Then using the properties of the function (4.9) and (4.10), it follows that J has a finite TD
limit which we denote by J, . In Section 5 we show that also the TD limit of q exists,

q = v(k) = J, (0).
(4.14)
k
Putting the results together one sees that the limit of (4.5) is

2
j (p) + 1 v(p) , p = 0,
Ep

1
J, (p) J, (0) .
j (p) :=
Ep
(1)
W1,
(p) =
(4.15)
206
From (4.13) one also gets the sum rule

2
(1)
Ep W1, (p) = 2 + q ,
(4.16)

(1)
(1)
while p W1, (p) = 0 = W1, (x, x), as required.
It is instructive to compare now the continuum (i.e., small ap behavior where a is the lattice
(1)
spacing) of W1, (p) with its counterpart in the compact model. In Appendix B the small p
asymptotics of j (p) is determined. In terms of the (non-universal) constants g2 (), g3 () the
small p behavior comes out as

2g3 () 4/
(1)
Ep W1,
(p) ln ln p2 /T + 2g2 () +
ln(p 2 /T )

2 2
+ O ln p /T
,
(4.17)
where

4
.
T = 32 exp
(4.18)
Eqs. (4.17), (4.18) illustrate in particular the nonperturbative nature of the large N expansion
in the noncompact model, despite the fact that in infinite volume the expansion is effectively
performed with respect to massless fields, as it is the case in perturbation theory. The infrared
regulator (, V ) clearly works very differently from a constant small mass regulator.
The subleading logarithmic terms in (4.17) are difficult to determine on the lattice. Using a
continuum cutoff instead and cutoff-normalized continuum momenta

1
32p
d2 k 2
2
(4.19)
[, ] platt pc :=

k ,
,
2
V
(2)
k
the continuum counterpart of (4.15) can be found in closed form:

2
2T
p
4
1
Epc W1,c (pc ) = 2 Li c 1 +
T
ln(T /pc2 )
pc
1
ln(T /pc2 )
x
2
32/p
c
ds
0
ln s
1
,
|1 s| ln(T /pc2 ) ln s
(4.20)
where Li(x) = 0 ds/ ln s. The integral in the last term is singularity free as the divergence of the
1/|1 s| factor at s = 1 is removed by the ln s, and ln T /pc2 > ln s holds over the entire range of
the integration. The asymptotic expansion of (4.20) comes out as

2
2
Epc W1,c (pc ) ln
ln pc /T +
4
ln(T /pc2 )

1
cn n+2
+ O pc2 ,
+
2
ln (T /pc )
n0

cn := (1)n+1 (n + 2) 2 (1)n+1 + 1 (n + 2) .
(4.21)
207
We add some remarks. First, note that (4.21) contains factorially growing terms both with oscillating signs (Borel-summable) and with constant phase (non-Borel-summable). Second, (4.21)
is an expansion in terms of the running coupling
1
(p)
=
:=
2
ln(T /pc2 ) 1 +
2
4 ln p 2
(4.22)
which can be re-expanded in terms of positive powers of the bare coupling . Comparing the
result with its counterpart in the compact model one may verify that both perturbative expansions
are related simply by flipping the sign of ; see [5] for a general proof of this perturbative
correspondence. Third, the expressions (4.17), (4.21) do not suggest a nontrivial continuum
limit reachable merely by a multiplicative field and a coupling renormalization. For example
defining a renormalized coupling r by 2 e4/ = 2 e4/r , the infinite cutoff limit can only be
taken by allowing negative bare couplings. For all > 0 both r and (p) vanish for .
Finally, the expressions (4.20), (4.21) can also directly be compared with their analogues in the
compact model, see [14] for the former. In the compact model the bare mass gap m20 = 2 e4/
enters and for m20 /p 2 = 32e4/ /pc2 1 (i.e., small at fixed pc ) expressions for the selfenergy Epc W1,c (pc ) are obtained related to (4.20), (4.21) formally by flipping the sign of . The
attempt to take the sign flip beyond the asymptotic small expansion, however, would on the
bare level produce a hyperviolet mass scale 2 e+4/ , difficult to interpret. In contrast, in a
lattice formulation the exact large N correspondence summarized in Section 2.2 exists.
4.2. TD limit of the Noether current two-point function
To obtain the infinite volume limit of the leading order contribution to the current correlation
function (3.20) we decompose it according to:

i
(0)
J (q) = 2 exp [q q ] A (q) + sin(q /2) sin(q /2)(q) ,

(4.23)
2
with
A (q) =
1 sin(p + q/2) sin(p + q/2) sin(q /2) sin(q /2)

.
V p
(Ep + )(Ep+q + )
(4.24)
Now we have seen in the last subsection that (q), q = 0, has a finite limit, and so obviously
does A (q):

1
A (q) A, (q) =
Ep Ep+q
p

(4.25)
sin(q /2) sin(q /2) ,
(0)
which is independent of . Thus the infinite volume limit of J (q) exists.

In the next order we have the result (3.24). We now show consecutively: (i) that X3 has a finite
TD limit, and (ii) that X1 + X2 has a finite TD limit.
(i) Noting the identity

sin(p + q/2)
= 0,
(4.26)
(Ep + )(Ep+q + )
p
208
we can write X3 in (3.25) as

sin(p1 + q/2) sin(p2 + q/2) Y (p1 , p2 )
4
X3; (q) = 2
,
V p ,p (Ep1 + )(Ep1 +q + )(Ep2 + )(Ep2 +q + )
1
(4.27)
where
Y (p1 , p2 ) =
1
1
1
,
(p1 p2 ) (p1 ) (p2 )
(4.28)
which vanishes when either p1 = 0 or p2 = 0. Then we break up X3

X3; =
1
[X4; X5; X5; X6; ],
(Eq + )2
(4.29)
with
X4; (q) =
4 sin(p1 + q/2) sin(p2 + q/2) Y (p1 , p2 )(Eq Ep1 )(Eq Ep2 )

,
(Ep1 + )(Ep1 +q + )(Ep2 + )(Ep2 +q + )
V 2 p ,p
1
4 sin(p1 + q/2) sin(p2 + q/2) Y (p1 , p2 )(Eq Ep1 )

X5; (q) = 2
,
(Ep1 + )(Ep1 +q + )(Ep2 +q + )
V p ,p
4 sin(p1 + q/2) sin(p2 + q/2) Y (p1 , p2 )
X6; (q) = 2
.
(Ep1 +q + )(Ep2 +q + )
V p ,p
1
(4.30)
Now the limit of X4 exists. Next for X5 we write

X5; (q) = X7; (q) + X8; (q),
(4.31)
with
X7; (q)
4 sin(p1 + q/2) sin(p2 + q/2) [Y (p1 , p2 ) Y (p1 , q)](Eq Ep1 )
= 2
,
(Ep1 + )(Ep1 +q + )(Ep2 +q + )
V p ,p
1
X8; (q) = r (q)
1 sin(p + q/2) Y (p, q)(Eq Ep )

,
V p
(Ep + )(Ep+q + )
where (using the gap equation)

1 sin(p + q/2) 1
= 1 + (4 + )1 sin(q /2).
r (q) :=
V p
Ep+q +
4
(4.32)
(4.33)
It is clear that both X7 , X8 and so also X5 have finite limits. Finally we have
X6; (q) = X9; (q) + X10; (q) + X10; (q) + 4Y (q, q)r (q)r (q),
with
X9; (q) =
4
sin(p1 + q/2) sin(p2 + q/2)
V 2 p ,p
1
[Y (p1 , p2 ) Y (p1 , q) Y (p2 , q) + Y (q, q)]
,
(Ep1 +q + )(Ep2 +q + )
(4.34)
X10; (q) = r (q)
4 sin(p + q/2) [Y (p, q) Y (q, q)]

.
V p
Ep+q +
209
(4.35)
It follows that X8 has a limit, and the demonstration that X3 has a finite TD limit is complete.
(ii) We first rewrite the sum
X1; + X2; = 8q A2; + X11; ,
(4.36)
with
A2; (q) =
1 F (p, q)
,
V p (Ep + )2
X11; (q) =
F (p1 , q)
8
.
2
V p ,p (p1 p2 )(Ep1 + )2 (Ep2 + )
(4.37)
sin(q/2) sin(q/2)
.
Ep+q +
Eq +
(4.38)
Here
F (p, q) =
To proceed we again write X11 as a sum of terms:

X11; (q) = X12; (q) + X13; (q) + 8f1 A2; (q),
(4.39)
where
X12; (q) =
8 Y (p1 , p2 )F (p1 , q)
,
V 2 p ,p (Ep1 + )2 (Ep2 + )
1
X13; (q) =
81
V

p
F (p, q)
,
(p)(Ep + )2
and (f2 will appear in the next section):

1
1
fs :=
.
V p (p)(Ep + )s
(4.40)
(4.41)
We still need to consider the limits of X12; , X13; , A2; (actually the contribution in X1 + X2
involving A2 has a coefficient proportional to q f1 which vanishes in the TD limit). We note
that these three functions can be written in the form T2; [g](q) where functions Ts; [g](q) are
defined by
1 g(p)F (p, q)
,
Ts; [g](q) =
(4.42)
V p
(Ep + )s
with g(p) a regular periodic function with g(p) = g(p) and finite at p = 0. Now for functions
of the type T1 we have
1 g(p)F (p, q)
T1; [g](q) =
V p
Ep +

g(p)F (p, q)(Eq Ep Ep+q )
1
1
=
Eq + 2 V p
Ep +
210
1 g(p)F (p, q)(Ep+q + )

V p
Ep +

1
1
+
(4.43)
g(p) g(q) F (p, q) + g(q)
F (p, q) ,
V p
V p

which has a finite limit since V1 p F (p, q) does, as can easily be seen using the gap equation.
Next
1 g(p)F (p, q)
T2; [g](q) =
V p
(Ep + )2
=
Here

1
T1; [g](q) + T3; [g](q) + T4; [g](q) .
Eq + 2
T3; [g](q) =
1 g(p)F (p, q)(Eq Ep Ep+q )

,
V p
(Ep + )2
T4; [g](q) =
1 g(p)F (p, q)
1
,
(Eq + ) V p
(Ep + )2
(4.44)
(4.45)
with
F (p, q) := (Eq + )(Ep+q + )F (p, q).
Decomposing T3 further gives

1
T5; [g](q) + T6; [g](q) ,
T3; [g](q) =
Eq +
with
T5; [g](q) =
1 g(p)F (p, q)(Eq Ep Ep+q )(Eq Ep+q )

,
V p
(Ep + )2
T6; [g](q) =
1 g(p)F (p, q)(Eq Ep Ep+q )

1
,
Eq + V p
(Ep + )2
(4.46)
(4.47)
(4.48)
where T5 clearly has a TD limit. Now

F (p, q) = F; (p, q) + F+; (p, q),
with F (p, q) = F (p, q):

F; (p, q) = (Eq + ) sin p cos p cos(q/2) sin(q/2) + ( )

sin p cos q sin(q/2) sin(q/2) ,
2
(4.49)
(4.50a)
F+; (p, q) = (Eq + ) sin p sin p cos(q/2) cos(q/2)

1
1 2 2
2
2
sin(q/2) sin(q/2) (Eq + ) p + p p p
2
2

+ 2Ep
p 2 q2 .
(4.50b)
211
So
T6; [g](q) =

1
T7; [g](q) + T8; [g](q) + g(0)T9; (q) ,
Eq +
T7; [g](q) =
1 2 1 g(p)F+; (p, q)p 2

q
,
2 V p
(Ep + )2
(4.51)
with
T8; [g](q) = 2
cos q
T9; (q) = 2
cos q
1 [g(p) g(0)]F; (p, q) sin p

,
V p
(Ep + )2
1 F; (p, q) sin p
.
V p
(Ep + )2
(4.52)
All of these quantities have a finite limit, in particular T9; (q) on account of the gap equation.
A similar decomposition can be performed for T4; (q), showing that it likewise has a TD limit.
So T2; [g](q) has a TD limit. It follows that X12; , X13; , A2; all have infinite volume
limits.
To summarize, we have shown that the TD limit of the contribution to the two leading orders
(s)
in the 1/N expansion of the current correlator J , s = 0, 1, exists.
5. TD limit of the Binder cumulant
In the following we evaluate the infinite volume limit of U in the 2-dimensional noncompact
model to sub-leading order, i.e., the coefficients U0 , U1 in

1
U0 (, V ) U1 (, V )
+O
+
,
U (, V ) =
(5.1)
N +1
(N + 1)2
(N + 1)3
are computed for large volumes. Somewhat surprisingly we will find that the limit V
exists and is independent of ! In Section 5.1 the large volume asymptotics will be evaluated
analytically. As a test of the estimates and in order to have finite volume results to compare
Monte Carlo data with, we also directly evaluated the multiple lattice sums numerically up to
L = 1024. The results are reported in Section 5.2. Finally, to preclude that the large N results are
misleading we performed a Monte Carlo study of U (, V ) up to L = 384 for N = 8.
5.1. Analytical analysis of large V asymptotics
Evaluation of the leading order coefficient is straightforward. From (3.31) and (3.32) one
obtains
U0 (, V ) =
2
V 2 (0)
(5.2)
Now the TD limit of (0) can be evaluated from

(0) =

1
Ek
1
1
1
1
=
+
ln
V
+
+
a()
+
O
,
V
ln V
(Ek + )2 4
k =0
(5.3)
212
where a() > 0. Note that (taking the -derivative of the gap equation (3.4)) this translates into
a large volume asymptotics for = of the form

1
a(s)
(4)2 1
(4)3
1
4
2
ds
+
1
+
O
.
V (, V ) =
ln V
ln2 V
2
s 2 ln3 V
ln4 V
(5.4)
From (5.3) one has

8a()
1
+O
,
U0 (, V ) = 2
ln V
ln2 V
U0 (, ) = 2,
(5.5)
in stark contrast to the compact model (as discussed in the conclusions).
The discussion of the sub-leading order is more involved and is best done in Fourier space.
We prepare the following auxiliary functions
1
1
st (p) =
, s, t 1,
V
(Ek + )s (Epk + )t
k
s (p) = s1 (p),
s 1,
(5.6)
(p)
where (p) := 1 (p) =

is the inverse of the Fourier transform of (x). Further we
shall need in addition to (4.41)
1
1
1 st (k)
fst = 2
(5.7)
=
.
s
t
(k)(Ep + ) (Epk + )
V
(k)
V
p,k
Note
f21 = (0)q.
(5.8)
The expression (3.29) for U1 we break up according to

U1 = U11 + U12 ,
U11 = 21 03 w0 ,
U12 = 02 w1 =
1
[A + B + C + D].
2 V 2
Using 1 = 2V 2 (q + f1 ) and w0 = 2V 2 (V 2 (0)1 ) one has

8
1
U11 = (q f1 ) 1 2
.
V (0)
In U12 the term A comes from the w0 / derivative in w1 and is given by

2V
2 (0)
1
2
2
=
8qV
+
A = 2q4
V
.
2
V (0) V (0)2
2 (0)
The term B corresponds to the W2 DD 1 W2 piece in w1 and is given by

8V
2
f31 .
Vf1
B = 8 V
(0)
(0)2
(0)
(5.9)
(5.10)
(5.11)
(0)
(5.12)

(0)
213
(0)
The term C arises from the W2 DDW2 structure and reads

2
1 2 (k)
1
.
C=8
Ek +
(k)2 (0)
(5.13)
(0)
(0)
Finally D comes from the W2 W2 term

4
1
1
D=
f22 .
1 2f2 +
(0)
V 2 2
V 2 (0)
(5.14)
In the V limit certain terms are superficially divergent, so it is best to combine terms where
such divergences cancel. Specifically we rewrite A + B as

8
2 (0)
1
(A
+
B)
=
2U
+
)
1
(q
f
11
1
2 V 2
V (0)2
8
1
(f1 2 (0) f31 ).
+
(5.15)
2
V (0) (0)
Further useful combinations to discuss the limit turn out to be

1 1
1
2 (k)
q f1 =
,
V
(k) (0)
Ek +
k

1 1
2
22 (k)
1
f22 2f2 =
,
(0)
V
(k) (0)
(Ek + )2
k

1 1
2 (0)
3 (k)
.
f31 2 (0)f1 =
V
(k)
Ek +
(5.16a)
(5.16b)
(5.16c)
For the evaluation of the TD limit then mainly the combinations in square brackets in (5.16)
have to be studied, for large volumes. A naive replacement of the lattice sums by integrals over
the Brillouin zone with = 0 would produce infrared divergent integrals. The strategy in the
following will be to evaluate the large volume asymptotics of the functions s , s = 1, 2, 3, and
22 by repeated insertion of one of the following decompositions of unity

1
Ep Epk + (Epk + ) ,
Ep +

1
1=
Ep Ek Epk + (Epk + ) + (Ek + ) ,
Ep + 2
1=
(5.17a)
(5.17b)
until terms corresponding to infrared convergent integrals over the Brillouin zone arise. The
volume and the momentum dependence of the additional pieces picked up in the process (which
may diverge as V ) can then be studied analytically. For 1 = itself only one insertion
of (5.17b) was needed and no divergent piece arose, see (4.6). Proceeding similarly we derive
the relations

1
st (p) =
Xst (p) + (s1)t (p) + s(t1) (p) , s, t 1,
(Ep + 2)
1 Ep Ek Epk
.
Xst (p) =
(5.18)
V
(Ek + )s (Epk + )t
k
214
Applying it for the case s = 2, t = 1 (and noting s0 (p) = s1 (0)) we get

2 (p) =

1
X21 (p) + (p) + (0) .
Ep + 2
(5.19)
This can be rewritten as

1
2 (p)
1
(p) (0)
Ep +

1
1
X21 (p)
=
+
,
(0) Ep + 2 (Ep + 2)(p)
(p)(Ep + )(Ep + 2)
(5.20)
from which q f1 can be evaluated. In a first step one finds

1 2 (0)
1
(0)(q f1 + f2 ) =
V (0)
1 J3 (k)
1
1
+
.
+
V
Ek + 2 V
(Ek + )2
k =0
(5.21)
Here
1
1
,
f2 =
V
(k)(Ek + )(Ek + 2)
k =0
J3 (p) =
Ek Ep Epk
1
.
V
(k)(Epk + )(Ek + 2)
(5.22)
k =0
The function J3 (p) has a finite limit which for small p behaves as

J3, (p) = Ep j3 + O 1/ ln p 2 ,

v(k)
j3 =
(cos k1 cos k2 )2 .
Ek2
(5.23)
This can be used to show that

q2
1
ln ln V ln V +
ln V + O(ln ln V ),
8
4
q1
1
ln ln V ln V +
ln V + O(ln ln V ),
(0)(q f1 ) =
8
4
where the constants are related by
(0)f2 =
(5.24)
q 1 + q 2 = j3 .
Indeed, using the fact that Ek (k) has a finite limit which scales like
one readily verifies the first equation. Further

1
2 (0)
,
=O
1
(0)
ln2 V

1
1
,
f2 +
= f2 + O
V 2 (0)
ln2 V
(5.25)
1
2
ln k 2 /T
for
k2
0,
215

1
1
1
1
1
=
+O
,
V
(Ek + 2)
V
ln V
k =0
1
1 J3 (k)
=
j3 ln V + O(1).
2
V
4
(Ek + )
(5.26)
Inserted into (5.21) gives

1
ln V + O(ln ln V ),
4
and hence the first equation in (5.24). Combined with (5.3) we arrive at

ln ln V
1
U11 = 16a()
+O
,
ln V
ln V

2 (0)
ln ln V
8
1 = U11 + O
(q f1 )
,
V (0)2
ln2 V
using

1
1
2 (0)
+
O
=
1
1
.
V (0)2
V 2 (0)
ln2 V
(0)[q f1 + f2 ] = j3
(5.27)
(5.28)
(5.29)
Since U1 = U11 (A + B + C + D)/(2 V 2 ) one concludes from (5.28) and (5.15) that the softly
decaying U11 terms in U1 cancel.
We proceed with the evaluation of C. To this end a more detailed evaluation of X21 defined
in (5.18) is needed. In a first step one obtains
(Ep + )2 X21 (p)

p
1 (sin k1 + sin k2 )2
cos2
sin2 p
+2
= (Ep + )(0)
2
V
(Ek + )2
k

Ek2
E p + 2 p
p
1
+
cos
cos2
+ J2 (p) +
cos p
2
2
V
(Ek + )2
k
k
k
p1 p2 2 p1 + p2 1 sin2 21 sin2 22
64 sin2
sin
,
2
2
V
(Ek + )2
(5.30)
where
1 (Ep Ek Epk )(Ep Epk )2
,
V
(Ek + )2 (Epk + )
k

2 J2, (p) 0, J2, (p) = 2 + j2 p 2 + O p 4 ,
J2 (p) =
j2 0.93.
(5.31)
As V the first curly bracket diverges logarithmically while the second one is convergent.
Using

k
k
1
1
1 (sin 21 sin 22 )2
=
+O
,
V
32
V
(Ek + )2
k

1
1
1
1
1 (sin k1 + sin k2 )2
=
ln V + a() +
+O
,
V
4
4 4
ln V
(Ek + )2
k
216

Ek2
1
1
=
1
+
O
,
V
V
(Ek + )2
(5.32)
with a() as in (5.3) one finds

1
1
2 1
ln V + a() +
X21 (p) = 2 (cos p1 cos p2 )
4
2
Ep

1
1
1 2
+ 2 J2, (p) +
(cos p + cos 2p ) +
sin p
2
2
Ep

1
.
+O
ln V
For small momenta this behaves like

1
1
1
p1 p2 2
X21 (p) =
+
+ o(p1 , p2 )
ln V + a() +
4
2 4
p2 p1

1
5
1
+ 2 j2 +
+ o(p1 , p2 ) .
4 2
p
For the evaluation of C it is useful to rewrite (5.18) for p = 0 as

1
1
1
2 (p) X21 (p)
2 (p)
=
1
+
,
(p) (0)
Ep +
(Ep + )(0)
(p)
(p)
Ep
2 (p) X21 (p)
X21 (p) Ep +
1
+
=
X2 (p) +
,
(p)
(p)
Ep + 2
(p) Ep + 2
where
X2 (p) := 1 +

(p) (0) .
Ep (p)
(5.33)
(5.34)
(5.35a)
(5.35b)
(5.36)
In the first term the fact that the numerator is proportional to Ep is important for the eventual
decay properties of C. Further X2 (p) is bounded by a function of order ln V in the volume and
it has at most logarithmic singularities for p 0. The latter is consistent with
(0)
2 (p)
+ O(1/V ),
(p)
Ep (p)
(5.37)
where the O(1/V ) piece comes from X21 (p).

Inserting (5.35b) into (5.13) gives

8
1
8V 1
2 (0) 2
2 (k) X21 (k) 2
C= 2
+
.
+
1
1
(0)
(k)
(k)
(0)2
(0)2 V
(Ek + )2
k =0
(5.38)
Using the known behavior of the constituent functions for large V and small momenta, one can
verify a decay of the form2

C
1
1
=
O
(5.39)
+
O
.
2 V 2
ln2 V
ln3 V
2 Here and later on we indicate the form of the sub-leading term, without however (in a slight abuse of the O symbol)
presupposing that its coefficient is nonzero.
217
We proceed with the D term, where 22 (p) enters. A useful representation is

6
2
(p) + (0) +
X21 (p)
(Ep + 2)2
(Ep + 2)2

1
J4 (p) + 2J (p) 2X1 (p) ,
+
(Ep + 2)3
1 (Ep Epk Ek )3
,
J4 (p) =
V
(Ek + )2 (Epk + )2
22 (p) =
(5.40a)
(5.40b)
X1 (p) =
1 Ep Ek Epk
.
V
(Ek + )2
(5.40c)
This can be used to determine the large V behavior of the term D in (5.14). We begin by separating the zero mode in the relevant combination

1
1
2 22 (0)
1
f22 2f2 + 1 = 1
(0)
(0)
V 2 (0)
V 2 (0)

1
1
2
22 (k)
+
(5.41)
.
V
(k) (0)
(Ek + )2
k =0
On account of

1
1
=O
,
1 2
ln V
V (0)
1

2 (0)
1
2 22 (0)
=1
+O
,
(0)
(0)
ln3 V
the zero mode pieces are O(1/ ln V ). For the last term on the right-hand side of (5.41) we introduce the shorthand S1 + S2 . Upon insertion of (5.40a) we write S1 for the part coming from the
(p) + (0) piece in (5.40a) and S2 for the rest,

1
2 1
2 1
1
1
S1 =
+
(0) V
(k) (Ek + 2)2 (Ek + )2
(Ek + 2)2 V
k =0
k =0

1
=O
(5.42)
.
ln2 V
The term S2 reads
S2 =

1
1 1
J4 (k) + 2J (k) 2X1 (k)
(0) V
(k)(Ek + 2)3
k =0
6 1
X21 (k)
+
,
(0) V
(k)(Ek + 2)2
(5.43)
k =0
and is checked to behave as

1
1
+
O
.
S2 = O
ln2 V
ln3 V
(5.44)
218
Together

D
1
1
=O
+O
.
ln V
2 V 2
ln2 V
(5.45)
It remains to consider A + B. In view of (5.15) and (5.28) we know

A+B
ln ln V
8
1
,
=
U
+
O
(0)
f
f
+
11
1
2
31
V 2 2
V 2 (0) (0)
ln2 V
(5.46)
so that only f31 2 (0)f1 is still needed. Using (5.18) for the case s = 3, t = 1 we obtain
1
2
X31 (k)
1
f31 2 (0)f1 =
3 (0) 2 (0) +
2
V
(k)(Ek + 2)
V (0)
1
2 (k)
+
V
(k)(Ek + 2)
k =0
2 (0)
k =0
1
1
.
V
(k)(Ek + 2)(Ek + )
(5.47)
k =0
Since

2 3 (0) 2 (0) = O V 2 ,
the zero mode piece scales like O(V / ln V ). For the second term we observe
X31 (k)
1 J3 (p)
1
,
=
V
(k)(Ek + 2) V p (Ep + )3
(5.48)
(5.49)
k =0
with J3 as in (5.22). As a consequence this term in (5.47) scales like O(V ) for large V . In the
last two terms we insert (5.19) to get

1
1
X21 (k)
1
+
V
(k)
(Ek + 2)2
k =0

1
1
Ek + 2
+
(5.50)
.
(0) 2 (0)
V
Ek +
(k)(Ek + 2)2
k =0
The very first term is O(V ln V ), the one involving X21 (p) is O(V / ln V ), and the last one is
O(V ln ln V ). Together
f31 2 (0)f1 = O(V ln V ) + O(V ln ln V ).
(5.51)
For A + B this results in

1
ln ln V
1
(A
+
B)
=
U
+
O
+
O
.
11
ln V
V 2 2
ln2 V
Combining (5.9) with (5.52), (5.39) and (5.45) we arrive at the conclusion:

1
ln ln V
1
U1 (, V ) = O
+O
+
O
,
ln V
ln2 V
ln2 V
U1 (, ) = 0.
(5.52)
(5.53)
219
Table 1
Quantities entering U11 for = 3; all given digits are significant
L
106
(q f1 )/
f2
106 (0)
64
128
256
512
768
1024
233.007495171
52.7199137013
12.0362079518
2.76867400687
1.17557207628
0.64094863000
0.25538202207
0.23716419150
0.21754473457
0.19723910189
0.18522986021
0.17669596989
0.42884117188
0.37935605329
0.33427610710
0.29289048252
0.27016715051
0.25464412160
0.00451330538
0.02202558370
0.10558882967
0.49868470422
1.22915930848
2.32558377444
For the TD limit of the full Binder cumulant the result (5.53) amounts to

1
2
+O
.
U (, ) =
N +1
(N + 1)3
(5.54)
This result will be backed by Monte Carlo simulations in Section 5.2. Potential implications for
criticality and triviality of the theories are discussed in the conclusions.
5.2. Direct evaluation of lattice sums
Both as a check on the previous analysis and in order to have finite volume data to compare
Monte Carlo data with, we also evaluated the lattice sums defining U1 and several other quantities numerically up to L = 1024. Since O(L4 ) terms have to be summed and both very small
(e.g., ) and very large numbers (e.g., (0)) enter high precision is needed. The summations
were performed to 96 bit (26 significant figures) accuracy using the publicly available arbitrary
precision MPFR library (http://www.mpfr.org) and for moderate L also with Mathematica.
The results were found to vary with such that for smaller the presumed large V asymptotics sets in later. Below we present the results for = 3; qualitatively those for other values
are similar. Due to the predicted occurrence of very slowly varying terms (e.g., of ln ln V / ln V
type) one cannot expect that the genuine large V asymptotics can be unambiguously probed by
direct summation. Nevertheless two or three parameter fits of the L 1024 sums to the expected
decay form are generally convincing. Table 1 summarizes results for , (0) and some slowly
varying quantities entering U11 . Here U11 is defined in Eq. (5.10), q f1 is evaluated directly
from (3.9), (4.41) and from (5.16a), f2 is defined in (4.41). For example the leading asymptotics
1
(0) 4
ln V and the coefficients in
f2
1
1
ln ln V (q f1 ),
2
(5.55)
come out well in fits to the data.

Table 2 presents results for the terms used in the breakup of U1 , see Eq. (5.9), and the final
result for U1 . The column for A + B again illustrates the need for high precision, as individually A and B are 2 5 orders of magnitudes larger that their sum. It also highlights that
the analytical evaluation of the large V asymptotics is crucial. Even at L = 1024 the normalized A + B contribution is still increasing. On account of (5.52) the decay of the combination
(A + B)/(V )2 U11 should be faster. Indeed these data have a maximum at around L = 300
and then decay monotonically in a way fitted well by the predicted functional form.
Finally we present a fit of the U1 data to the predicted decay form in (5.53) in Fig. 2.
220
Table 2
Quantities contributing to U1 for = 3; all given digits are significant
L
102 U11
103 (A + B)/(V )2
103 C/(V )2
103 D/(V )2
U1
64
128
256
512
768
1024
0.748704
0.565668
0.431162
0.330143
0.282636
0.253114
4.65746
2.24328
0.73430
0.23478
0.63551
0.86471
9.13703
7.25786
5.92122
4.92648
4.45707
4.16358
0.387281
0.275149
0.189879
0.132507
0.108460
0.094567
0.012353890
0.010946420
0.009688426
0.008595209
0.008027416
0.007654026
Fig. 2. U1 vs. V . Fit to O(ln ln V / ln2 V ) + O(1/ ln V ) + O(1/ ln2 V ).
5.3. MC results for U

Since the large N expansion is only an asymptotic expansion the higher order coefficients in
(3.29) are not bound to be small, even in finite volume. At any given N the truncated series could
in principle misrepresent the exact U (, V ). In order to preclude this possibility we estimated
U (, V ) via Monte Carlo simulations.
We have chosen to simulate a SO(1, 8) theory at = 3 on lattices of linear dimensions
L = 32, 64, 128, 256, 384. The simulations were performed in a fixed spin gauge (the spin at
the origin was held fixed). The variable spins were updated by a Metropolis procedure tuned
to achieve a roughly 50% acceptance rate. Equilibration and autocorrelation times for various
observables have an enormous range: non-gauge-invariant observables in particular (e.g.,
n0 )
require extremely long runs, and on larger lattices fail to reach equilibrium even after billions of
Monte Carlo sweeps. The situation is much better for gauge-invariant observables, such as
entering U1 , see (3.26). The fluctuations in this latter quantity determine the Binder cumulant,
2
U versus latand are typically stable after a few million sweeps. Results for the quantity N+1
tice size are shown in Fig. 3. This quantity is not monotonic, but reaches a maximum near L = 64
and then decreases quite rapidly. The decrease appears to be faster than the log-type decay found
for U1 in Section 5.1, suggesting that the termwise large V asymptotics of the large N series
(when formally treated as convergent) sums to a power-like large volume decay.
221
Fig. 3. Monte Carlo results for N 2+1 U vs. V .
In summary, the numerical evidence suggests that the large N contributions to the Binder
cumulant beyond leading order (i.e., U0 (, V )) may indeed vanish in the thermodynamic limit.
6. Conclusions
Noncompact SO(1, N) sigma-models are expected to be massless in contrast to their compact
counterparts. The infrared problem therefore is nontrivial, especially in dimension d = 2, and
the goal of the present paper has been to gain computational control over the limit of vanishing
infrared regulator. The large N expansion is well suited for this; in a lattice formulation the dynamically generated gap is negative and serves as a coupling dependent infrared regulator which
vanishes in the limit of infinite lattice size. The cancellation of infrared divergences has been
demonstrated in d = 2 by explicit computation of a number of physically interesting quantities
defined in terms of invariant correlation functions: the spin and current two-point functions as
well as the Binder cumulant, all to next to leading order. A complementary result is [15] where
a noninvariant observable was shown to have a finite thermodynamic limit in d 3 beyond
large N . In d = 2 we expect that a large N counterpart of Davids theorem [7] can be established, showing that infrared divergences cancel termwise in the large N expansion of invariant
correlation functions to all orders.
To discuss our result for the Binder cumulant let us first recall the situation in the compact
model. In the notation of Section 3.4 one has there 1/+ (0) 4+ , in the thermodynamic
(and continuum) limit, so that V U = 8/[(N + 1)+ ]. Taking = 1/ + as the definition of

the correlation length, this gives the familiar result for the renormalized coupling gr = V U/ 2 =
8/(N + 1), to leading order. See also Ref. [3] for a direct continuum computation to sub-leading
order, with the result (N + 1)gr = 8[1 0.602033/(N + 1) + O(1/(N + 1)2 )].
In the noncompact model the zero momentum limits of invariant correlation functions are
expected to diverge in the thermodynamic limit. Indeed, mostly this reflects the fact that they are
increasing functions of the lattice distance (recall nx ny 1, always). Our results of Section 5
suggest however that the ratio entering U is finite, independent of , and very close to 2/(N + 1).
222
One can view this result as a manifestation of a concentration of measure phenomenon.

For the 1D lattice model with L sites it was shown in [16] that the functional measure has
support mostly on configurations boosted by an amount increasing at least powerlike with L.
In the thermodynamic limit the measure (or mean) is therefore concentrated at infinity, i.e., in
the disc model of the hyperbolic geometry at the boundary of the disc. Though not proven in
dimensions d > 1 it is very plausible that a similar concentration phenomenon will hold for the
d-dimensional functional measures. Indeed our result on the Binder cumulant can
be put into
a := a / , with
this context:
First
note
that
in
terms
of
the
normalized
average
spins

a = x nax , one has

2
Var 2 :=
=
2
U 0.
N +1
(6.1)
In the thermodynamic limit Var( 2 ) has been argued to vanish, which is natural if the components of a are typically very large rendering the relative fluctuations ensuring
= 1
and
( )2 1, negligible. Indeed to leading order of the large N expansion one finds
0 ln V . Note that the constant 2/(N + 1) can be interpreted as the value of U in a constant
configuration and that the indefinite dot product is crucial here.
Alternatively 2/(N + 1) U is given by the ratio of the susceptibilities defined from the partially connected 4-point and 2-point functions W2 and W1 , respectively. Both diverge as V ,
but the ratios entering the large N expansion of U (viz. w1 /02 , w2 /02 , see Eqs. (3.28)(3.31))
vanish in the thermodynamic limit. This is compatible with a genuine factorization of the 4-point
function but does not entail it. (Since the connected four point function
nx1 ny1 nx2 ny2 c entering (3.26) does not take into account the nonzero one-point functions, the fact that it must be
nonzero is only indirectly relevant for this.)
Concerning local quantities, the analysis of the TD limit for the subleading term of the spintwo point function in Section 4.1 does not suggest the existence of a nontrivial limit as the UV
cutoff is removed. Positive bare couplings are required for the large N series to be an asymptotic expansion, in which case a naturally defined renormalized coupling vanishes as the UV
cutoff is removed. The situation should be similar for the two-point function of the Noether
current.
Together, our results may be taken as an indication for triviality of the theory in the sector comprising SO(1, N) invariant observables. If corroborated beyond the large N expansion
this would be of significance in a number of other contexts, e.g., for a class of KaluzaKlein
theories or for the widely studied systems with AdS5 S5 target spaces. The focus on invariant
observables is certainly natural from the viewpoint of the compact models. In the context of the
OsterwalderSchrader reconstruction [4,16], however, invariant correlators are not ideal for the
noncompact systems, and the situation may well be different when noninvariant observables are
considered.
Acknowledgements
We wish to thank E. Seiler for many discussions, and M. Lscher for correspondence. The
research of A.D. is supported in part by the National Science Foundation under grant PHY0554660.
223
Appendix A. Leading and next-to-leading order SD equations

Here we tabulate the first few of the hierarchy of SchwingerDyson equations for the large N
(s)
coefficients Wr , r + s > 1. They can be obtained, e.g., by first converting (3.1) into a system
of equations for the exact Wr and then inserting the large N ansatz (2.6). The ensued recursive
structure is summarized in Fig. 1.
The leading 2-point function (3.2) satisfies

(0)
(0)
(0)
(x, y) W,1
(x, y) + 1 x W,1
(x, z)|z=x = 1 xy .
[x ]W,1
(A.1)
In the next order we have
(1)
(1)
[x ]W,1 (x, y) D (x, y)x W,1 (x, z)|z=x

(0)
(0)
= (1 xy ) W,1 (x, y) + 1 x W,2 (x, z; z, y)|z=x ,
(A.2)
where we have used the solution (3.2) to the leading order equation (A.1) to simplify some terms.
We see that to solve (A.2) we first need to solve the equation for the leading order 4-point
function:
(0)
(0)
(0)
(x1 , y1 ; x2 , y2 ) W,1
(x1 y1 )x1 W,2
(x1 , z; x2 , y2 )|z=x1
(x1 )W,2

= x1 x2 D (y1 y2 ) + x1 y2 D (y1 x2 )
+ [x1 x2 + x1 y2 ]D (x1 y1 )D (x2 y2 ).
(A.3)
(0)
(1)
The solution for W,2 is given in (3.5), from which can verify that W,1 in (3.8) solves (A.2).
In the next order the equation for the 4-point function is
(1)
(1)
[x1 ]W,2 (x1 , y1 ; x2 , y2 ) D (x1 y1 )x1 W,2 (x1 , z; x2 , y2 )|z=x1

(0)
= x1 W,3 (x1 , u; v, y1 ; x2 , y2 )|u=v=x1

(0)
(1)
W,2 (x1 , y1 ; x2 , y2 )x1 W,1 (x1 , z)|z=x1

(1)
(0)
W,1 (x1 , y1 )x1 W,2 (x1 , z; x2 , y2 )|z=x1

(0)
(1)
(1)
W,2 (x1 , y1 ; x2 , y2 ) x1 x2 W,1 (y1 , y2 ) x1 y2 W,1 (x2 , y1 )

(0)
+ (x1 x2 + x1 y2 ) W,2 (x1 , y1 ; x2 , y2 )

(1)
(1)
D (x2 y2 )W,1 (x1 , y1 ) D (x1 y1 )W,1 (x2 , y2 ) .
(A.4)
One sees the pattern summarized in Fig. 1 emerging, in that the solution of (A.4) requires knowledge of the leading order 6-point function. The latter satisfies the equation:
(0)
(x1 )W,3 (x1 , y1 ; x2 , y2 ; x3 , y3 )

(0)
D (x1 y1 )x1 W,3 (x1 , z; x2 , y2 ; x3 , y3 )|z=x1

(0)
(0)
= W,2 (x1 , y1 ; x2 , y2 )x1 W,2 (x1 , z; x3 , y3 )|z=x1

(0)
(0)
W,2 (x1 , y1 ; x3 , y3 )x1 W,2 (x1 , z; x2 , y2 )|z=x1

(0)
(0)
(0)
(0)
x1 x2 W,2 (y1 , y2 ; x3 , y3 ) x1 x3 W,2 (y1 , y3 ; x2 , y2 )

x1 y2 W,2 (y1 , x2 ; x3 , y3 ) x1 y3 W,2 (y1 , x3 ; x2 , y2 )
224
(0)
[x1 x2 + x1 y2 ] D (x1 y1 )W,2 (x2 , y2 ; x3 , y3 )

(0)
+ D (x2 y2 )W,2 (x1 , y1 ; x3 , y3 )
(0)
[x1 x3 + x1 y3 ] D (x1 y1 )W,2 (x2 , y2 ; x3 , y3 )

(0)
+ D (x3 y3 )W,2 (x1 , y1 ; x2 , y2 ) ,
(A.5)
where all the functions on the rhs are known from solutions of (A.1) and (A.3). The solution is
simply given by
(0)
W,3
(x1 , y1 ; x2 , y2 ; x3 , y3 )

(0)
(0)
(0)
=
W,2 (x1 , y1 ; w1 , z1 )W,2 (x2 , y2 ; w2 , z2 )W,2 (x3 , y3 ; w3 , z3 )
w1 ,w2 ,w3 z1 ,z2 ,z3
1
1
1
(w1 z2 )D
(w2 z3 )D
(w3 z1 ).
D
(A.6)
(1)
Using (A.6) one can verify that W,2 as given in (3.11) solves (A.4).
Appendix B. Continuum limit behavior of j (p)
In this appendix we consider the small p behavior of the function j (p) entering the spin
two-point function (4.15) to subleading order. We have

sin p j2; (p) +
p 2 j3; (p),
J (p) J (0) = Ep j1 (p) +
(B.1)
with
v(k) v(p) ,
Ekp
j1 (p) =
k

j2; (p) =
k
j3; (p) =
1
4

sin k
v(p k) v(p + k) ,
Ek
2
k
Ek

v(p k) + v(p + k) ,
(B.2)
k
k
where v(p) is defined in (4.7) and k = 2 sin 2 , as usual. Note first j3; (p) is non-singular at
p = 0:
1
j3; (0) = J (0).
4
Next

j2; (p) = 2
(B.3)

sin(k p)
v(k) v(p)
Ekp

= sin p 2j1 (p) + j4; (p) 2 cos p j5; (p),
(B.4)
225
with

k2
j4; (p) =
Ekp
k

j5; (p) =
k

v(k) v(p) ,

sin k
v(k) v(p) = 2
sin p j6; (p),
Ekp

sin k sin k
v(k) v(p) .
Ekp Ek+p
j6; (p) =
(B.5)
Noting
1
j4; (0) = J (0),
2
(B.6)
we have for small p 2 :

j (p) j1 (p) 4

sin p sin p
1
j6; (p) + J (0) + O 1/ ln p 2 .
Ep
4
(B.7)
From (4.10) we have

v(p) = (p) + O p2 ,
(p) :=
2
,
ln(p 2 /T )
(B.8)
with T defined in (4.18). We now consider two corresponding integrals

j1 (p) =

k
j6; (p) =

(c2 k 2 )
(k) (p) ,
2
(k p)

k

(c2 k 2 )k k
(k) (p) ,
(k p)2 (k + p)2
(B.9)

where k denotes d2 k/(2)2 and c is a momentum cutoff T > c2 > p 2 . These give the
leading small p 2 contribution because

j1 (p) j1 (p) = v1 (c) + O 1/ ln p 2 ,

1
1
j6; (p) j6; (p) = v1 (c) + v2 + O 1/ ln p 2 ,
(B.10)
2
4
with

v1 (c) =
k
v2 =

(k) 2
v(k)
2
|k | 2 c k ,
Ek
k
4
k v(k)
k
Ek2
(B.11)
226
First using
2
d
0
2
1
=
,
(t + cos )
t2 1
t 2 > 1,
(B.12)
we can do the angular integrations in the j functions, setting without loss of generality p
(p, 0):
1
j1 (p) =
2
c2
dx
0

1
1
1
,
|x p 2 | ln(x/T ) ln(p 2 /T )
c2
1
j6;00 (p) = 2
8p
x + p2
dx 1
|x p 2 |

1
1
.
ln(x/T ) ln(p 2 /T )
(B.13)
Noting c2 > p 2 we obtain

j1 (p) =
2

1
S1 p , T + S2 p 2 , T + S3 p 2 , c 2 , T ,
2
2 ln(p /T )
(B.14)
with

S1 p 2 , T =
1
dy
ln y
,
(1 y) ln(p 2 y/T )
dy
ln(1 + y)
,
y ln[(1 + y)p 2 /T ]

S2 p 2 , T =
1
0

S3 p 2 , c 2 , T =
2
c2 /p
1
dy
1
Now for small
ln(1 + y)
.
y ln[(1 + y)p 2 /T ]
(B.15)
p2

S1 p 2 , T
(n)
1
1
s1
,
ln(p 2 /T )
[ln(p 2 /T )]n
n=0

S2 p 2 , T
1
ln(p 2 /T )
(n)
s2
n=0
1
,
[ln(p 2 /T )]n
(B.16)
with
s1(n) = (1)n
1
dy
[ln y]n+1
,
(1 y)
dy
[ln(1 + y)]n+1
,
y
0
(n)
s2
1
= (1)
n
0
(B.17)

2
(0)
227
(0)
giving s1 = 6 , s2 = 12 , . . . . Next

S 3 p 2 , c 2 , T = S 4 p 2 , c 2 , T + S5 p 2 , c 2 , T ,
c2 < T .
(B.18)
Here
2
c2 /p
1
S4 p 2 , c 2 , T =
1
ln(1 + y)
=
dy
(1 + y) ln[(1 + y)p 2 /T ]
2 2
ln(c
/p )
dx
ln 2
2 /p 2
c
dz
ln(z)
z ln(zp 2 /T )
x
x + ln(p 2 /T )

ln(2p 2 /T )
= ln 2p 2 /c2 + ln p 2 /T ln
,
ln(c2 /T )
(B.19)
and
2
c2 /p
1
S5 p 2 , c 2 , T =
dy
1
ln(1 + y)
y(1 + y) ln[(1 + y)p 2 /T ]
(n)
1
1
s5
,
2
ln(p /T )
[ln(p 2 /T )]n
(B.20)
n=0
with
(n)
s5

= (1)
dy
1
giving
(0)
s5
2
12
[ln(1 + y)]n+1
,
y(1 + y)
(B.21)
+ 12 (ln 2)2 . . . . So by (B.10) and (B.14)

1

1
j1 (p) ln ln 2p 2 /T + g1 + O ln p2 /T
,
2
1

1
g1 = v1 (c) + ln ln c2 /T + ,
2
2
which is independent of c. Similarly
j6;00 (p) =

2
1
S 6 p , T + S2 p 2 , T + S3 p 2 , c 2 , T ,
2
4 ln(p /T )
(B.22)
(B.23)
with
1
S6 p , T =
2
dy
(n)
1
1
y ln y
s6
,
2
2
2
(1 y) ln(p y/T ) ln(p /T )
[ln(p /T )]n
(B.24)
y[ln y]n+1
,
(1 y)
(B.25)
n=0
where
(n)
s6
1
= (1)
dy
0
228

(0)
giving s6 = 1 6 , . . . .
Putting all the results together we obtain (4.17) with
1
1
g2 = g1 + v2 + J (0).
2
4
(B.26)
Appendix C. Large N with two auxiliary fields

The results for the large N expanded correlation functions in the noncompact model have
in Section 3 been obtained via the large N correspondence summarized in Section 2.2. Direct
large N computations in the noncompact model can be based on the following generating functional [5]

1
Hxy N
dx exp (N + 1)S [, H ] ,
exp Wd [H ] = exp
2 x,y
S [, H ] =
1
Tr ln A + i
2

x =x0
Axy = xy + 2ix xy +
x =x0
1 1 1
A xx,
0 0
2
Hxy = A xy + 2ixy xx0 x0 .

N +1
(C.1)
The formal
Here A is the matrix obtained by deleting the x0 -th row and column of A or A.
expansion based on (C.1) is not a valid saddle point expansion but it does produce the correct
expansion coefficients and is related to its counterpart W+d [H ] in the compact model by the
involution x x , . The functional (C.1) thus provides a simple heuristic way to
understand the large N correspondence. In contrast to the compact model, however, Wd [H ] is
not equivalent to the original generating functional W [H ].
Here we outline how (C.1) can formally be obtained from the formulation of the large N
expansion with two auxiliary fields introduced in [4]. We begin by dualizing the spatial spin
components nx , x , as one would do in the compact model. Indeed, the N spatial components nx , x = x0 , enter (2.4) with the good sign; their dualization gives

1
Hxy
dn0x n0x0 1
exp W [H ] = exp
2 x,y
x

N
dx exp Tr ln A i(N + 1)
x
2
x =x0
x =x0

N +1 0 0
n Axy ny .
exp +
(C.2)
2 x,y x
Here A arises due to the constrained Gaussian integration. Note the small but crucial differences

to the compact model [5]: only N copies of Tr ln A occur so far and the sign of the x =x0 x
term is flipped, as is the sign of the Hxy term in Axy . Most importantly the kinetic term in the
last exponential has the wrong sign, which is why one cannot naively interchange the order of
the integrations over n0x and x , x = x0 . To proceed we assume that in a large N expansion the
replacement

(C.3)
dn0x n0x0 1 = ()||/2
dx (x0 ), n0x = n x + ix ,
x
229
is legitimate, for certain saddle point configurations n x , with n x0 = 1. With this replacement
the kinetic term acquires the good sign. After the additional re-routing x = ix /(2) + x the
saddle point conditions S/x = S/x = 0 lead to
n 2x = 1 + D xx ,
(n)
x = x n x ,
x = x0 .
(C.4)
(See Eq. (5.18) of [4], with x = 2 x , and correcting the sign in the second formula.) Here
Dxy = (M 1 )xy with Mxy := xy + xy x . Since is a positive operator it follows from
the second
that the position dependent x must be predominantly negative:
equation in (C.4)
x = x x n 2x .
0 x n x (n)
To leading order the spin two-point function is given by [4]
nx ny f.s. = n x n y D x,y ,
Dx,x0 Dy,x0
,
D x,y := Dx,y
Dx0 ,x0
(C.5)
where we write momentarily

f.s for the average computed with the fixed spin measure in (2.4).
The quantity n x then is the nonzero expectation value
n0x f.s. to leading order in 1/(N + 1).
To make contact to the gap equation (3.4) in Section 3.1 we now first replace (C.4) by a
simpler gap equation with constant and n,
n 2 D (0) = 1,
n 2 =
,
V
with D (x) :=
1 eipx
.
V
E(p) +
(C.6)
p =0
These are the saddle point conditions arising from a translation invariant gauge fixing of the
functional integral (see Eq. (5.6) of [4]) and imply D(0) = 1, in accordance with (3.4). Given
a solution , n of (C.6) we claim that
n x := D(x x0 ),
x := + x,x0 ,
(C.7)
is a solution of (C.4). The equation (n)

x = ( + x,x0 )n x is checked using D(0) = 1. To
verify the first equation in (C.6) it suffices to observe that
D x,y = D(x y) D(x x0 )D(y x0 )/D(0).
(C.8)
This can be seen as follows: suppose that invertible matrices M and M are related by Mxy =
M xy cxy x0 x . Then the inverse of M is related to the inverse of M by
M 1

xy

= M 1 xy +
1 1
c
M
M
.
xx0
yx0
1
1 c(M )x0 x0
(C.9)
Applied to Mxy = xy + x xy = M xy + xy x,x0 , for x = + x,x0 , without yet assuming

the gap equation (C.6), this gives first
Dxy = D(x y)
D(x x0 )D(y x0 ),
1 + D(0)
(C.10)
and then (C.8) from (C.5). In particular no pole occurs for the quantities in (C.8) at D(0) = 1.
Using (C.8) and D(0) = 1, it follows 1 + D x,x = 2 D(x x0 )2 = n 2x , while the sign in (C.7)
is fixed by D(x) 1.
Inserting (C.7) into (C.5) we arrive at
nx ny f.s. = D(x y) = n 2 D (x y) =
nx ny trans ,
(C.11)
230
where the right-hand side coincides with the spin two-point function computed to leading order
in 1/(N + 1) in the translation invariant gauge [4] and with that of Section 3.1.
In summary, the leading order results with two auxiliary fields and the two gauge fixings
considered (fixed spin and translation invariant gauge) are related via (C.7). Both saddle point
equations (C.4) and (C.6) imply the version D(0) = 1 used here, but in addition provide
the interpretation of n x =
n0x f.s. and n =
n0x trans , as the averages of n0x with respect to the
respective gauge-fixed
functional measures. Note that n x approaches n 2 as |x x0 | becomes

large and that x n x = V n 2 . Although the invariant two-point functions coincide to leading
order in the two gauges, the results for the noninvariant quantity
n0x are very different,
n0x f.s. =
n0x 2trans .
Equipped with this interpretation of n x we return to (C.2). Subject to the assumption (C.3)
one can proceed by interchanging the order of integrations, which results in the Gaussian

N +1
N +1
dx (x0 ) exp
x Axy y +
x
i Axy n y
2 x,y
x
x
y

1/2 exp N + 1 n 2x A 1 1
.
= Const(det A)
A
(C.12)
x xy y
x0 x0
0
2
x,y
Using also n x0 = 1 and substituting back into (C.2) we arrive at (C.1).
References
[1] A.J. Kupiainen, On the 1/n expansion, Commun. Math. Phys. 73 (1980) 273.
[2] J. Balog, M. Niedermaier, F. Niedermayer, A. Patrascioiu, E. Seiler, P. Weisz, The intrinsic coupling in integrable
quantum field theories, Nucl. Phys. B 583 (2000) 614, hep-th/0001097.
[3] M. Campostrini, A. Pelissetto, P. Rossi, Four point renormalized coupling constant in O(N ) models, Nucl. Phys.
B 459 (1996) 207.
[4] A. Duncan, M. Niedermaier, E. Seiler, Vacuum orbit and spontaneous symmetry breaking in hyperbolic sigmamodels, Nucl. Phys. B 720 (2005) 235;
A. Duncan, M. Niedermaier, E. Seiler, Nucl. Phys. B 758 (2006) 330, Erratum.
[5] M. Niedermaier, E. Seiler, P. Weisz, Perturbative and non-perturbative correspondences between compact noncompact sigma-models, hep-th/0703212, Nucl. Phys. B, in press.
[6] L. Schfer, F. Wegner, Disordered systems with n orbitals per site: Lagrange formulation, hyperbolic symmetry,
and Goldstone modes, Z. Phys. B 38 (1980) 113.
[7] F. David, Cancellation of infrared divergences in the 2D nonlinear sigma-model, Commun. Math. Phys. 81 (1981)
149.
[8] S. Elitzur, The applicability of perturbation expansion to two-dimensional Goldstone systems, Nucl. Phys. B 212
(1983) 501.
[9] G. Cristofano, R. Musto, F. Nicodemi, R. Pettorino, F. Pezzella, 1/N contribution to physical quantities in the lattice
O(N ) sigma-model, Nucl. Phys. B 257 (1985) 505.
[10] M. Niedermaier, E. Seiler, On the large N expansion in hyperbolic sigma-models, in preparation.
[11] M. Lscher, SchwingerDyson equations in a finite volume, unpublished notes 1980/81.
[12] P. Hasenfratz, Perturbation theory and zero modes in O(N ) lattice sigma models, Phys. Lett. B 141 (1984) 385.
[13] D.S. Shin, Application of a coordinate space method for the evaluation of lattice Feynman diagrams in two dimensions, Nucl. Phys. B 525 (1998) 457.
[14] H. Flyvbjerg, Some exact results for the O(N ) symmetric nonlinear sigma-model, Nucl. Phys. B 348 (1991) 714.
[15] T. Spencer, M. Zirnbauer, Spontaneous symmetry breaking of a hyperbolic sigma model in three dimensions, Commun. Math. Phys. 252 (2004) 167187.
[16] M. Niedermaier, E. Seiler, Non-amenability and spontaneous symmetry breakingThe hyperbolic spin chain, Ann.
Henri Poincar 6 (2005) 1025.

Real forms of complex higher spin field equations

and new exact solutions
Carlo Iazeolla a,b, , Ergin Sezgin c , Per Sundell b
a Dipartimento di Fisica, Universit di Roma Tor Vergata, INFN, Sezione di Roma Tor Vergata,
Via della Ricerca Scientifica 1, 00133 Roma, Italy

b Scuola Normale Superiore and INFN, Piazza dei Cavalieri 7, 56100 Pisa, Italy
c George P. and Cynthia W. Mitchell Institute for Fundamental Physics, Texas A&M University, College Station,
TX 77843-4242, USA
Received 29 June 2007; accepted 6 August 2007
Available online 14 August 2007
Abstract
We formulate four-dimensional higher spin gauge theories in spacetimes with signature (4 p, p) and
non-vanishing cosmological constant. Among them are chiral models in Euclidean (4, 0) and Kleinian
(2, 2) signature involving half-flat gauge fields. Apart from the maximally symmetric solutions, including
de Sitter spacetime, we find: (a) SO(4 p, p) invariant deformations, depending on one continuous and
infinitely many discrete parameters, including a degenerate metric of rank one; (b) non-maximally symmetric solutions with vanishing Weyl tensors and higher spin gauge fields, that differ from the maximally
symmetric solutions in the auxiliary field sector; and (c) solutions of the chiral models furnishing higher
spin generalizations of type D gravitational instantons, with an infinite tower of Weyl tensors proportional
to totally symmetric products of two principal spinors. These are apparently the first exact 4D solutions
with non-vanishing massless higher spin fields.
1. Introduction
Given the impact YangMills theory and gravity have had on the development towards our
present understanding of fundamental interactions, as formulated within quantum field theory
* Corresponding author at: Dipartimento di Fisica, Universit di Roma Tor Vergata, INFN, Sezione di Roma Tor
Vergata, Via della Ricerca Scientifica 1, 00133 Roma, Italy.
E-mail address: carlo.iazeolla@roma2.infn.it (C. Iazeolla).
doi:10.1016/j.nuclphysb.2007.08.002
232
C. Iazeolla et al. / Nuclear Physics B 791 [FS] (2008) 231264
(QFT) and to some extent string field theory (SFT), it is natural to explore higher spin (HS)
extensions of gauge symmetries (i.e., non-Abelian gauge groups containing generators in representations of the Lorentz group with spins higher than one).
Presently, the only known full models of interacting higher-spin gauge fields are those based
on the Vasiliev equations [1]. These equations are naturally formulated in terms of SL(2; C)
spinor oscillators in Lorentzian signature (3, 1). In this paper, we shall formulate them using
spinor oscillators in Euclidean signature (4, 0) and Kleinian signature (2, 2) as well, and present
non-trivial exact solutions with novel properties such as the excitation of all higher spin fields.
Before we state our motivations for this work, let us first highlight some key elements of the HS
theory.
To begin with, the Vasiliev equations that describe the HS theory involve two features that
are relatively novel from the point-of-view of lower-spin QFT as well as the standard formulation of SFT. Firstly, they are written in a frame-like language, closely related to the constraint
formulation of supergravity, known as free differential algebra (FDA), or unfolded dynamics.
Here, all fields are differential forms, which live on an a priori unspecified base manifold. Moreover, for each differential form there is a, in general non-linear, differential constraint, written
using the exterior derivative (and no contractions of curved indices using the metric). Thus, diffeomorphism invariance is manifest without the need to single out a metric or other component
field.
Secondly, in order to accommodate an infinite number of physical as well as auxiliary fields,
one works with master fields that, in addition to being differential forms, are functions of oscillator variables. The functions belong to, or, depending on taste, define a fiber over the base
manifold consisting of representations of an underlying non-Abelian higher-spin algebra. In particular, the master zero-form is directly related to the (massless) spectrum via the theorem of
Flato and Fronsdal [2]. One may go further and associate the oscillators to a particle or other extended objects, perhaps related to discretization of tensionless strings and membranes in AdS [3],
though these considerations are of course not crucial for setting up Vasilievs formalism.
The simplest higher spin gauge theories of Vasiliev type, and indeed the first ones to appear
in the literature [1], are based on higher-spin extensions of SO(3, 2) realized using oscillators
that are SL(2, C) doublets (coordinatizing the phase space of Diracs Sp(4) singleton). Here, the
master fields are an adjoint one-form and a twisted-adjoint zero-form, sometimes referred to as
the Weyl zero-form. The master field equations, which we again stress are manifestly background
independent and diffeomorphism invariant, can then be written in a remarkably simple closed
form. These equations can be treated in two almost opposite ways, namely by projecting to
the fiber or by projecting to the base. In the latter case one can make contact with lower-spin
field theory by taking the base manifold to be an ordinary spacetime and eliminate the auxiliary
fields, treating only the Lorentz connection and the vierbein exactly. This well-defined, albeit
tedious, approach yields manifestly reparametrization and locally Lorentz invariant physical field
equations in a perturbative expansion in curvatures as well as higher-spin gauge fields (see, for
example, [21]).
The projection to the fiber, on the other hand, is a more tractable operation, since the Vasiliev
equations can be solved locally on the base manifold using gauge functions [4]. This leaves
equations on the non-commutative fiber, which are thus purely algebraic from the point-of-view
of the base manifold. The simplest exact solution to these equations is the four-dimensional
anti-de Sitter spacetime. In a recent paper, [5], two of us have given an exact SO(3, 1)-invariant
solution to these equations. Locally, the solution describes a scalar field in a FRW-like metric
with a space-like singularity, that can be resolved by the method of patches. The solution is
233
asymptotically AdS and periodic in time, so that one may think of it as an instanton universe
inside AdS [6]. More recently, the gauge function method has been used to describe the BTZ
black hole metric as a solution to full three-dimensional HS gauge theory [7].
This raises the question how to Wick rotate solutions of the Lorentzian theory into solutions of
a Euclidean theory. The main difficulty is to impose proper reality conditions given the doubling
of the spinor oscillators due to the Euclidean signature. We resolve this by taking the master fields
to be holomorphic functions of the left-handed and the right-handed spinor oscillators subject to
pseudo-reality conditions.
In addition to the Euclidean signature, we shall consider the Kleinian signature as well. While
in all signatures there is the possibility of a chiral asymmetry, in Euclidean and Kleinian signatures the extreme case of parity violation involving half-flat gauge fields can also arise. We refer
to the latter ones as chiral models. In HS gauge theory, the HS algebra valued gauge-field curvatures can be made, say, self-dual, but the model nonetheless contains the anti-self-dual gauge
fields through the master zero-form which contains the corresponding Weyl tensor obeying the
appropriate field equation. Although this is contrary to what happens in ordinary Euclidean gravity, where the field equations can contain only self-dual fields, it is not a surprise in higher spin
theory since the underlying higher spin algebra, which is an extension of SO(5), does not admit
a chiral massless multiplet.1
There are several reasons that make the investigation of HS theory in Euclidean and Kleinian
signatures worthwhile. To begin with, just as the Euclidean version of gravity plays a significant
role in the path integral formulation of quantum gravity, it is reasonable to expect that this may
also be the case in the quantum formulation of HS theory, despite the fact that an action formulation is yet to be spelled out (see [9] for recent progress). For reviews of Euclidean quantum
gravity, see, for example, [10] and [11].
Another well-known aspect of self-dual field theories is their capability to unify a wide class
of integrable systems in two and three dimensions. It would be interesting to extend these mathematical structures to self-dual HS gauge theories to find new integrable systems.
The chiral HS theories in Kleinian signatures may also be of considerable interest in closed
N = 2 string theory in which the self-dual gravity in (2, 2) dimensions arises as the effective
target space theory [12]. However, there are some subtleties in treating the picture-changing
operators in the BRST quantization which have raised the question of whether there are more
physical states [13], and in the case of open N = 2 theory an interpretation in terms of an infinite
tower of massless higher spin states has been proposed [14]. It would be very interesting to
establish whether these theories or their possible variants admit self-dual HS theory in the target
space. While the N = 2 string theories may seem to be highly unrealistic, it should not be ruled
out that they may be connected in subtle ways to all the other string theories which themselves
are connected by a web of dualities in M theory.
In this paper, we shall take the necessary first steps to start the exploration of the Euclidean and
Kleinian HS theories. We shall start by determining the real forms of the complex higher spin
algebra based on infinite-dimensional extension of SO(5; C) and formulate the corresponding
higher spin gauge theories in four-dimensional spacetime with signature (4 p, p). Maximally
symmetric four-dimensional constant curvature spacetimes, including de Sitter spacetime, defined by the embedding into five-plane with signature (5 q, q) are readily exact solutions.
1 We shall leave the group-theoretical analysis to [8], where we also give the spinor-oscillator formulations of the
four-dimensional minimal bosonic models with H4 , dS4 and H3,2 vacua.
234
Table 1
The minimal bosonic higher-spin algebras ho(p , 5 p ) so(5 p , p ) in signature (p, 4 p) can be realized with
spinor oscillators transforming as doublets under the groups listed in the third column. These realizations obey reality
conditions (MAB ) = (MAB ), with Hermitian subalgebras listed below [8]. The symmetric spaces with unit radius
have cosmological constant = 32
HSA
Signature ab
Spinors
Reality
Symmetric space
Hermitian isometries
ho(5)
ho(4, 1)
ho(4, 1)
ho(3, 2)
ho(3, 2)
(4, 0)
(4, 0)
(3, 1)
(3, 1)
(2, 2)
SU(2)L SU(2)R
SU(2)L SU(2)R
SL(2, C)diag
SL(2, C)diag
SL(2, R)L SL(2, R)R
1
+1
1
+1
1
id
id
S4
H4
dS4
AdS4
H3,2
so(2) so(3)
so(3, 1)
so(3, 1)
so(3, 2)
so(3, 2)
Fluctuations about these spaces arrange themselves into all the irreducible representations of
SO(5 q, q) contained in the symmetric two-fold product of the fundamental singleton representation of this group, each occurring once. The details of this phenomenon will be provided in
a separate paper [8].
We then devote the rest of the paper to finding a class of non-trivial exact solutions of these
models, including the Euclidean and chiral cases. The key information about these solutions is
encoded in the master zero-form, which contains a real ordinary scalar field and the Weyl tensors
1 2s and 1 2s for spin s = 2, 4, 6, . . . in the minimal bosonic model and s = 1, 2, 3, 4, . . .
in a non-minimal bosonic model [15,16]. Vasilievs full higher spin field equations assume a form
reminiscent of that of open string field theory, with master fields that are functions of spacetime
as well as an internal non-commutative space of oscillators. Our new exact solutions are constructed by using the oscillators to build suitable projectors, with slightly different properties in
the minimal and non-minimal models.
Our exact solutions fall into the following four classes:
Type 0: These are maximally symmetric solutions (see Table 1) with
1 2s = 0,
(x) = 0,
1 2s = 0,
a
4
a
=
,
W a1 as1 = 0,
e
(1 2 x 2 )2
(1.1)
describing the symmetric spaces S 4 , H4 , AdS4 , dS4 , H3,2 = SO(3, 2)/SO(2, 2), where || is the
inverse radius of the symmetric space, x 2 = x a x b ab , and ab is the tangent space metric. In
the above the Weyl tensors have spin s = 2, 4, 6, . . . in the minimal model, and s = 1, 2, 3, 4, . . .
in the non-minimal model, while for W a1 as1 , s = 4, 6, . . . in the minimal model, and s =
1, 3, 4, 5, 6, . . . in the non-minimal model.
Type 1: These solutions, which arise in the minimal models (and therefore are evidently solutions also to the non-minimal models with vanishing odd spins), are SO(p, 4 p) invariant
deformations of the maximally symmetric solutions with

(x) = 1 2 x 2 ,
1 2s = 0, 1 2s = 0 (s = 2, 4, . . .),
a
= f1 a + 2 f2 x x a ,
e
W a1 as1 = 0
(s = 4, 6, . . .),
(1.2)
where is a continuous parameter and f1 , f2 (see (3.85)) are highly complicated functions
of x 2 , , and a set of discrete parameters corresponding to whether certain projectors are switched
on or off. The metric is Weyl-flat conformal to the maximally symmetric solution with a complicated conformal factor, and note that all the higher spin gauge fields vanish. Interestingly,
235
a particular choice of the discrete parameters yield, in the 0 limit, the degenerate metric:
g =
x x
1
.
2
2
(1 x ) 2 x 2
(1.3)
Degenerate metrics are known to play a role in topology change in spacetime (see, for example, [17], and references therein). Interestingly, here they arise in a natural way by simply taking
a certain limit in the parameter space of our solution.
Type 2: These are solutions of the non-minimal model that are not solutions to the minimal model. The spacetime component fields are identical to those of the maximally symmetric
type 0 solutions, but, unlike in the type 0 solution, the spinorial master one-form is non-vanishing
(see (3.102)). Even though all odd spin fields are vanishing, the solution exists only for the
non-minimal model because the spinorial master field violates the kinematic conditions of the
minimal model. In particular, this means that this type of solution cannot be a 0 limit of the
type 1 solutions. Furthermore, the spinorial master field is parametrized by discrete parameters,
again associated with projectors.
Type 3: These are solutions of the non-minimal chiral models in Euclidean and Kleinian signatures, in which all gauge fields are non-vanishing. These solutions also depend on an infinite set
of discrete parameters and for simple choices of these parameters we obtain two such solutions
in both of which
(x) = 1,
1 2s = 0,
W a1 as1 = 0.
In one of the solutions the Weyl tensors and the vierbein take the form
2

h 1 s
U( 1 U s V s+1 V 2s ) ,
1 2s = 22s+1 (2s 1)!!
h2

a
2
a
e
= 2
g3 + g4 2 x x a + g5 2 (J x) (J x)a ,
h (1 + 2g)
(1.4)
(1.5)
(1.6)
where h, g, g3 , g4 , g5 are functions of x 2 defined in (B.3), (3.126), and the almost complex
structure Jab and spinors (U, V ) are defined in (B.8) and (B.11), and = 1 as explained in
Section 3.5. For the other solution we have

1 s
2s+1
1 2s = 2
(1.7)
(2s 1)!!
( 1 s s+1 2s ) ,
h2
a

2
a
+ g 4 2 x x a + g 5 2 (Jx) (Jx)a ,
e
(1.8)
= 2
h (1 + 2g)
where the functions g,

g 4 , g 5 are defined in (3.134), and the almost complex structure Jab is
defined in (B.10).
These are remarkable solutions in that they are, to our best knowledge, the first exact solution
of higher spin gauge theory in which higher spin fields are non-vanishing. We also note that the
Weyl tensors in these solutions corresponds to higher spin generalization of the type D Weyl
tensor that takes the form ( )
up to a scale factor [18]. Type D instanton
solutions of Einsteins equation in Euclidean signature with and without cosmological constant
have been discussed in [19]. Our solution provides their higher spin generalization.
After we describe the HS field equations in diverse signatures in Section 2, we shall present the
detailed construction of our solutions in Section 3. We shall comment further on these solutions
and open problems in Section 4.
236
2. The bosonic 4D models in various signatures

We shall first describe the field equations without imposing reality conditions on the master
fields. These conditions will then be discussed separately, leading to five different models in
four-dimensional spacetimes with various signatures (see Table 1).
2.1. The complex field equations
To formulate the complex field equations we use independent SL(2; C)L doublet spinors
(y , z ) and SL(2; C)R doublet spinors (y , z ) generating an oscillator algebra with noncommutative and associative product defined by
y y = y y + i ,
y z = y z i ,
(2.1)
z y = z y + i ,
z z = z z i ,
(2.2)
y y = y y + i ,
z y = z y i ,
(2.3)
y z = y z + i ,
z z = z z i ,
(2.4)
and
where the juxtaposition denotes the symmetrized, or Weyl-ordered, products. For example,
y y = 12 (y y + y y ). Equivalently, Weyl-ordered functions obey2
f(y, y,
z, z ) g(y,
y,
z, z )
4 4
d d i +i
=
e
f (y + , y + , z + , z )g(y
+ , y + ,
z , z + ),
(2)4
(2.5)
where the hats are used to denote functions of all oscillators, while functions of only y and y
will be unhatted.
The complex master fields are the adjoint one-form A and the twisted-adjoint zero-form
defined by
A = dx A (x; y, y,
z, z ) + dz A (x; y, y,
z, z ) + d z A (x; y, y,
z, z ),
(2.6)
= (x;
y, y,
z, z ),
(2.7)
where x are coordinates on a commutative base manifold (which can, but need not, be fixed to
be four-dimensional spacetime). One also defines the total exterior derivative
d = dx + dz
+ d z ,
z
z
(2.8)
with the property d(f g)

= (d f) g + (1)degf f d g for general differential forms. In
what follows we shall suppress the . The master fields can be made subject to the following
2 The integration measure is defined by d 4 = d 2 1 d 2 2 , where d 2 z = i dz d z = 2 dx dy for z = x + iy. With
this normalization, I f = f.
237
discrete symmetry conditions3 [15,16]:

Minimal model (s = 0, 2, 4, . . .):
Non-minimal model (s = 0, 1, 2, 3, . . .):
= A,
(A)
= A,
(A)
= (),
()
= ,
()
(2.9)
(2.10)
where is the -product algebra anti-automorphism defined by

f(y, y;
z, z ) = f(iy, i y;
iz, i z ),
(2.11)
and and are two involutive -product automorphisms defined by

f(y, y;
z, z ) = f(y, y;
z, z ),
f(y, y;
z, z ) = f(y, y;
z, z).
(2.12)
We note that
(f g)
= (1)deg(f )deg(g)
(g)
(f),
(f g)
= (f) (g),
(2.13)
(2.14)
(
f g)
= (
f) (g),
(2.15)
= . The automorphisms are inner and can be generated by conjugation with the
and that
functions and given by

= exp(i y z ),
= exp iy z ,
(2.16)
2
such that
f(y, z) = f(z, y),
f(y, z) = f(z, y),
f(y,
z ) = f(z, y),
f(y,
z ) = f(z, y),
f = (f),
f = (
f).
(2.17)
(2.18)
The full complex field equations are

i
F = c1 dz dz + c2 d z d z ,
4
D = 0,
(2.19)
(2.20)
where c1 and c2 are complex constants and the curvatures and gauge transformations are given
by
F = d A + A A,
A = D ,
]
,
,
= [ , ]
D = d + [A,
(2.21)
(2.22)
with
[f, g]
= f g (1)deg(f )deg(g)
g (f).
(2.23)
Since is defined up to rescalings by complex numbers, the model only depends on one complex
parameter, that we can take to be
c2
c= .
(2.24)
c1
3 The exterior derivative obeys d = d and d = d , and the and maps do not act on the commutative coordinates.
238
In components, the constraints read

F = 0,
= 0,
D + [A , ]
(2.25)
F = 0,
F = 0,
(2.26)
ic1
F = ,
2
F = 0,
ic2
F = ,
(2.27)
(2.28)
D + A + (A ) = 0,
(2.29)
D + A + (A ) = 0,
(2.30)
where (2.30) can be derived using (A ) = A . Introducing [1]

S = z 2i A ,
S = z 2i A ,
(2.31)
the component form of the equations carrying at least one spinor index now take the form
S + [A , S ] = 0,
S + [A , S ] = 0,
[S , S ] = 2i (1 c1 ),
[S , S ] = 2i (1 c2 ),
(2.32)
(2.33)
[S , S ] = 0,
(2.34)
S + (S ) = 0,
(2.35)
S ) = 0.
S + (
(2.36)
This form of the equations makes the following Z2 Z2 symmetry manifest:

S S ,
S S
(2.37)
(where the two transformations can be performed independently) keeping A and fixed. We
note that S S is equivalent to A A iz , idem S and A .
All component fields are of course complex at this level. Next we shall discuss various reality
conditions on the (hatted) master fields that will lead to models with real physical fields living in
spacetimes with different signatures.
2.2. Real forms
In order to define the real forms of the field equations one has to impose reality conditions
on both adjoint one-form and twisted-adjoint zero-form, corresponding to suitable real forms of
the higher-spin algebra and signatures of spacetime. There are three distinct real forms of the
complex higher-spin algebra itself. In two of these cases there are two distinct reality conditions
that can be imposed on the zero-form, leading to five distinct models in total, as shown in Table 1.
The reality conditions are
A = (A),

,
= ()
(2.38)
239
where the possible actions of the dagger4 on the spinor oscillators and consequential selections
of real forms of SO(4; C) SL(2; C) SL(2; C) are given by

SU(2)L SU(2)R :
y
= y ,
z
= z ,

= y ,
z = z ,
y
(2.39)

y
= y ,
z = z ,
SL(2; C)diag :
(2.40)

= y ,
z = z ,
Sp(2; R)L Sp(2; R)R : y

= y ,
z = z ,
y
(2.41)
and the map is given in Table 1, with the isomorphism given by

f y , y , z , z = f(y , y , z , z )
(2.42)
in the case of (4, 0) signature. Note that is an oscillator-algebra automorphism in signatures

(3, 1) and (2, 2), while it is an isomorphism in signature (4, 0). Here, the SU(2) doublets are
pseudo real in the sense that from (y ) = y idem (z ) , (y ) and (z ) it follows that
(y , y ; z , z ) and (y , y ; z , z ) generate equivalent oscillator algebras with isomorphism .
The reality property of the exterior derivative takes the following form in different signatures:
Signature (3, 1) and (2, 2):
d = d,
(2.43)
Signature (4, 0):
d = d .
(2.44)
We note that the Euclidean case is consistent in the sense that

dz = d z = d z = dz
(2.45)
is compatible with representing d f using f/z = 2i [z , f] , which yields

i

i
= dz f , z = dz f , z
2
2

i
= dz z , f .
(2.46)
2
Demanding compatibility between the reality conditions (2.38) and the master field equations
(2.19) and (2.20), and using

() = ,
(2.47)
i dz dz
= i dz dz ,
i
dz [z , f]
2
one finds the following reality conditions on the parameters

Signature (3, 1):
c1 = c2 ,
c1
= c1 ,
(2.48)
c2
= c2 .
(2.49)
As a result, the parameter c is a phase factor in Lorentzian signature and a real number in Euclidean and Kleinian signatures. The parameters can be restricted further by requiring invariance
4 The dagger acts as usual complex conjugation on component fields; in this paper we shall denote the conjugate of a
complex number x by x , while reserving the bar for denoting quantities associated with the R-handed oscillators.
240
under the parity transformation

P (y ) = y ,
P d = d P,
P 2 = Id.
(2.50)
Taking A to be invariant and assigning intrinsic parity = 1 to ,

= A,
P (A)
= ,
P ()
one finds that the master equations are parity invariant provided that [20]

1
type A model (scalar),
c= =
1 type B model (pseudoscalar).
(2.51)
(2.52)
In Lorentzian signature, there is no loss of generality in choosing c1 = c2 = 1 in the type A

model and c1 = c2 = i in the type B model, while in Euclidean and Kleinian signatures one
may always take c1 = c2 = 1 in the type A model and c1 = c2 = 1 in the type B model. More
generally, the parity transformation maps different models into each other as follows,
1
P (c) = ,
c
leaving invariant the types A and B models. The maximally parity violating cases are
P (c1 ) = c2 ,
P (c2 ) = c1 ,
(2.53)
Signature (3, 1):
c = exp(i/4),
(2.54)
c = 0.
(2.55)
The case with c = 0 shall be referred to as the chiral model, that we shall discuss in more detail
below.
The HS equations in Lorentzian signature have the Z2 symmetry acting as (S , S )
( S , S ), and Z2 Z2 symmetry in (4, 0) and (2, 2) signatures acting as (S , S )
( S , S ), where = 1 and = 1.
Finally, let us give the reality conditions at the level of the SO(5; C) algebra and its minimal bosonic higher-spin extension. The adjoint representation of the complex minimal bosonic
higher-spin Lie algebra is defined by5

ho(5; C) = Q(y, y):
(Q) = Q ,
(2.56)
and the corresponding minimal twisted-adjoint representation by

T ho(5; C) = S(y, y):
(S) = (S) .
The real forms are defined by

ho(5 q, q) = Q(y, y)
ho(5; C): Q = (Q) ,

T ho(5 q, q) = S(y, y)
T ho(5 q, q) : S = (S) .
(2.57)
(2.58)
(2.59)
The finite-dimensional SO(5; C) subalgebra is generated by MAB , that we split into Lorentz
rotations and translations (Mab , Pa ) defined by
(Mab ) = Mab ,
(Pa ) = Pa .
5 A more detailed description of the complex higher-spin algebra and its representations is given in [8].
(2.60)
241
For these generators, which by convention arise in the expansion of the master fields together
with a factor of i (see, for example, [21]), the reality condition (2.38) implies
(MAB ) = (MAB ).
(2.61)
This condition is solved by

Mab =

1
(ab ) y y + ( ab ) y y ,
8
Pa = (a ) y y ,
4
(2.62)
where the van der Waerden symbols are defined in Appendix A and 2 is proportional to the
cosmological constant, as shown in Table 1. The van der Waerden symbols encode the spacetime signature ab , and the commutation relations among the MAB then fix the signature of the
ambient space to be

AB = ab ; 2 .
(2.63)
2.3. The chiral model
In the chiral model with c = 0, the master field can be eliminated using (2.27), and expressed as

i
= 1 + S S ,
(2.64)
2
where we have chosen c1 = 1 and S is given by (2.31). The remaining independent master-field
equations now read
F = 0,
D S = 0,
D S = 0,
[S , S ] = 0,
[S , S ] = 2i ,
(2.66)
S S S + S S S = 4i S .
(2.67)
(2.65)
We note that (2.36) holds identically in virtue of S + (

S ) = [S , 1+ 2i S S ]
= 0,
where we used S = S and [S , S ] = 0. The chiral model can be truncated further

by imposing
A = 0,

A = 0,
z

A = 0.
z
(2.68)
In general, the chiral model also has interesting solutions with non-vanishing A , since flat connections in non-commutative geometry can be non-trivial.
2.4. Comments on the weak-field expansion and spectrum
The procedure, described in great detail in [21], for obtaining the manifestly diffeomorphism
and locally Lorentz invariant weak-field expansion of the physical field equations can be extended straightforwardly to arbitrary signature. The expansion is in terms of spin-s physical fields
with s = 2 as well as higher derivatives of all fields, while the vierbein and Lorentz connection
are treated exactly.
In this approach one first solves (2.26)(2.30) subject to the initial condition

= Z=0 ,
(2.69)
242

A = A Z=0 = e + + W + K ,
(2.70)
where
1 a
1
(2.71)
= ab Mab ;
e Pa ,
2i
4i
W contains the higher-spin gauge fields (and also the spin s = 1 gauge field in the non-minimal
model); and the field redefinition

1
1
K = S S Z=0 + S S Z=0
(2.72)
4i
4i

= i A A A
+i A A A
.
(2.73)
y
Z=0
Z=0
e =
One also imposes the gauge condition

(0)
A = 0,
(2.74)
where we have defined the internal flat connection

(0)

A = A =0 .
A (0)
= A =0 ,
(2.75)
A (0)
= 0,
One then substitutes the resulting and A , which can be obtained explicitly in a perturbative
expansion in , into (2.25) and sets Z = 0, which yields a manifestly spin-2 covariant complex
HS gauge theory on the base manifold. Up to this point the local structure of the base-manifold,
nor the detailed structure of the gauge fields, have played any role. To proceed, one may refer
to an ordinary spacetime, take e a to be an (invertible) vierbein, and treat W as a weak field.
This allows one to eliminate a large number of auxiliary fields in and W , leaving a model
a
consisting of a physical scalar = |y=y=0
, the vierbein e , and an infinite tower of (doubly
traceless) HS gauge fields a(s) residing in W .
The gauge choice (2.74) is convenient since it implies y A |Z=0 = 0 that simplifies the
expansion [21]. However, there are also other gauges where A |=0 is a flat but non-trivial
internal connection, and indeed this will be the case for the type 1 and type 2 solutions that we
shall present in Section 3.
In the leading order in the weak fields, the two-form and one-form constraints for the minimal
model read
= 0,
R, = c2 , R ,
R, = 0,
R ,
s = 2:
(2.76)
= 0,
R
=
0,
R
=
c
1
,
,

(1)
(1)
F, ...
= c2 1 ...2s2 , F ,
1 ...k k+1 ...2s2 = 0,
1
2s2
s = 4, 6, . . . :
(2.77)
(1)
(1)
F
F ,
= c1 1 ...2s2 ,
...
,1 ...k k+1 ...2s2 = 0,
1
0-forms:
2s2
1 ...n
1 ...m
1
= i 1 ...m 1 ...n mn (1 (
2 ...m ) 2 ...n ) ,
(2.78)
where for higher spins s = 4, 6, . . . and k = 0, . . . , 2s 3, and for 0-forms |m n| = 0 mod 4.

In all cases, the zero-form system contains a physical scalar with field equation
2

+ 22 = 0.
(2.79)
243
In the Lorentzian case, where both c1 and c2 = c1 are non-zero, the spin-2 sector consists of
gravity with cosmological constant 32 , and the spin-s sectors with s = 4, 6, . . . consist of
higher-spin tensor gauge fields with critical masses proportional to 2 . The criticality in the
masses, that implies composite masslessness6 in the case of AdS, holds in the dS case as well,
where thus the physical spectrum is given by the symmetric tensor product of two (non-unitary)
SO(4, 1) singletons [8].
In the Euclidean and Kleinian cases, the parameters c1 and c2 are real and independent. In
case c1 c2 = 0, the Lorentzian analysis carries over, leading to a composite massless spectrum
given by symmetric tensor products of suitable singletons [8]. However, unlike the Lorentzian
case, the spin-s sector of the twisted adjoint representation can be decomposed into left-handed
and right-handed sub-sectors of real states, corresponding to {1 ...m , 1 ... n } with m n = 2s
[8]. These sub-sectors mix under HS transformations.
In case either c1 or c2 , but not both, vanishes, that we shall refer to as the chiral models, the
metric and the higher-spin gauge fields become half-flat. For definiteness, let us consider the case
c2 = 0. The components of the zero-form that drop out in the two-form constraint, i.e., 1 ...2s ,
now become independent physical fields, obeying field equations following from (2.78).
3. Exact solutions
In this section we shall give four types of exact solutions to the 4D HS models given in
the previous section. The salient features of these are summarized in the Introduction. Here we
stress that (a) the type 0 solutions are maximally symmetric spaces; (b) the type 1 solutions are
SO(4 p, p) invariant deformations of type 0; (c) the type 2 solutions, which exist necessarily in
the non-minimal model, have vanishing spacetime component fields but non-vanishing spinorial
master one-form; (d) the type 3 solutions, which exist in the non-minimal chiral model only, have
the remarkable feature that all higher spin gauge fields are non-vanishing in such a way that the
Weyl zero-forms are covariantly constant, in a certain sense that will be explained below. Before
we give these four types of solutions we shall describe briefly the method for solving the master
field equations using gauge functions.
3.1. The gauge function ansatz
In order to construct an interesting class of solutions we shall use the Z-space approach [5,22]
in which the constraints carrying at least one curved spacetime index, viz.
F = 0,
D = 0,
(3.1)
F = 0,
F = 0,
(3.2)
are integrated in simply connected spacetime regions given the spacetime zero-forms at a point p,

S = S p ,
S = S p ,
= p ,
(3.3)
and expressed explicitly as
6 By composite masslessness we mean that the massless states are composites of two singletons [2]. This is well known
for SO(3, 2) and similar situations arise for other signatures as well.
244
A = L 1 L,
1
S = L S L,
= L 1 (L),
1
S = L S L,
where L = L(x,
z, z ; y, y)
is a gauge function, and

L p = 1,
= 0,
S = 0,
S = 0.
(3.4)
(3.5)
(3.6)
The internal connections A and A can be reconstructed from S and S using (2.31). In particular note the relation
A = L L + L 1 A L,
(3.7)
from which it follows that

S = z 2i A .
(3.8)
The remaining constraints in Z-space, viz.

[S , S ] = 2i (1 c1 ),
[S , S ] = 2i (1 c2 ),
(3.9)
[S , S ] = 0,
(3.10)
S + (S ) = 0,
S + (S ) = 0,
(3.11)
(3.12)
are then to be solved with an initial condition

C (y, y)
= Z=0 ,
(3.13)
and some assumption about the topology of the internal flat connections

(0)
S(0) = S C =0 ,
S = S C =0 .
(3.14)
In what follows, we shall restrict the class of solutions further by assuming that
L = L(x; y, y).
(3.15)
The gauge fields can then be obtained from (2.70), (2.73) and (3.5), viz.
e + + W = L1 L K ,
(3.16)
where

1 1
(3.17)
L S S + S S LZ=0 .
4i
Hence, the gauge fields, including the metric, can be obtained algebraically without having to
solve any differential equations in spacetime.
K =
3.2. Ordinary maximally symmetric spaces (type 0)

The complex master-field equations are solved by
= 0,
S = z ,
S = z ,
A = L1 L,
(3.18)
where the gauge function [22]

2h
ix y y
L(x; y, y)
=
exp
,
1+h
1+h
245
(3.19)
gives
4 dx 2
,
(1 2 x 2 )2
2
=
ds(0)
(3.20)
which we identify as the metric of the symmetric spaces listed in Table 1 for the different real
forms of the model, in stereographic coordinates with inverse radius ||. This metric is invariant
under the inversion

x a x a / 2 x 2 ,
(3.21)
and H4 is covered by a single coordinate chart, while the remaining symmetric spaces
two charts, related by the inversion. If we let x a = x a /(2 x 2 ), the atlases are given by

x : 0 2 x 2 1 x : 0 2 x 2 1 ,
S 4 2 = 1 :

H 4 2 = 1 :
x : 0 2 x 2 < 1 ,

dS4 2 = 1 :
x : 1 < 2 x 2 1 x : 1 < 2 x 2 1 ,

x : 1 2 x 2 < 1 x : 1 2 x 2 < 1 ,
AdS4 2 = 1 :

H3,2 2 = 1 : x : 1 < 2 x 2 1 x : 1 < 2 x 2 1 ,
require
(3.22)
(3.23)
(3.24)
(3.25)
(3.26)
where the overlap between the charts is given by {x : 2 x 2 = 1} in the cases of S 4 , dS4 ,
AdS4 and H3,2 , and the boundary is {x : 2 x 2 = 1} in the case of H4 and {x : 2 x 2 = 1}
{x : 2 x 2 = 1} in the cases of dS4 , AdS4 and H3,2 . The H3,2 space can be described as the coset
SO(3, 2)/SO(2, 2).
3.3. SO(4 p, p) invariant solutions to the minimal model (type 1)
3.3.1. Internal master fields
A particular class of SO(4; C)-invariant solutions is given by the ansatz
= ,
S = z S(u),
u),
S = z S(
(3.27)
where
u = y z ,
u = y z .
(3.28)
The above ansatz solves (3.10)(3.12). There remains to solve (3.9), which now takes the form

S , S = 4i 1 c2 ei u .
S , S = 4i 1 c1 eiu ,
(3.29)
Following [4], we use the integral representation
1
S(u) =
1
ds n(s)e 2 (1+s)u ,
(3.30)
246
u)
S(
=
1
2 (1+s)u
ds n(s)e
(3.31)
which reduces (3.29) to

c1
(1 t),
2
c2
(1 t),
(n n)(t)
= (t 1)
2
with defined by [4]
(n n)(t) = (t 1)
1
(f g)(t) =
1
ds
(3.32)
(3.33)
ds (t ss )f (s)g(s ).
(3.34)
Even and odd functions, denoted by f (t), are orthogonal with respect to the product. Thus,
one finds

+ +
c1
c1
n n (t) =
n n (t) = +
(3.35)
0 (t) 2 ,
0 (t) + 2 t,

+ +
c2
c2
n n (t) =
n n (t) = +
(3.36)
0 (t) 2 ,
0 (t) + 2 t,
where

1
(3.37)
0 (t) = 2 (1 t) (1 + t) .
One proceeds [4], by writing
n (t) = m (t) +
k pk ,
(3.38)
k=0
where m are expanded in terms of 0 (t) and the functions (k 1)

()
k (t) =
1 (1 )
sign(t) 2
1
1
ds1

= sign(t)
1
2 (1 )
dsk (t s1 sk )
(log t12 )k1

(k 1)!
(3.39)
obeying the algebra (k, l 0)

k l = k+l ,
and
pk (t)
(3.40)
(k 0) are the -product projectors
pk (t) =
(1)k (k)
(t),
k!
= (1)k ,
(3.41)
obeying
1
pk
= Lk [f ]pk ,
Lk [f ] =
1
dt t k f (t).
(3.42)
247
In particular,
pk pl = kl pl .
(3.43)
Substituting the expansion (3.38) into (3.35) and (3.36), one finds, in view of (3.40), (3.42)
and (3.43), manageable algebraic equations. Transforming back one finds, after some algebra [5],
m(t) = (1 + t) + q(t),

c1
1
c1
1
1
1
c1
; 2;
log 2 + t 1 F1 ; 2;
log 2 ,
q(t) =
1 F1
4
2
2
2
2
t
t
(3.44)
k = 2k Lk [m],
(3.46)
(3.45)
and
k {0, 1},
where
Lk [m] = (1)k + Lk [q],

1 + (1)k
c1
c1
1 (1)k
Lk [q] =
1 1
1 1+
.
2
1+k
2
2+k
(3.47)
(3.48)
The overall signs in m have been fixed in (3.45) by requiring that

S(u) = 1 for = 0 and k = 0.
(3.49)
Treating n the same way, one finds

m(t)
= (1 + t) + q(t),

c2
c2
1
c2
1
1
1
q(t)
=
F
F
;
2;
log
;
2;
log
+
t
,
1 1
1 1
4
2
2
2
2
t2
t2
k {0, 1},
k = 2k Lk [m],
(3.50)
= (1) + Lk [q],
Lk [m]

c2
1 + (1)k
c2
1 (1)k
Lk [q]
=
1 1
1 1+
.
2
1+k
2
2+k
(3.53)
(3.51)
(3.52)
(3.54)
Thus, the internal solution is given by

= ,
together with
(3.55)
S
and
S
as given in (3.8) with
(reg)
(proj)
A = A
+ A
,
i
(reg)
= z
A
2
1
(proj)
A
= i z
dt q(t)e 2 (1+t)u ,
(proj)
= iz
A
(reg)
(proj)
A = A
+ A
,

k=0

k=0
i
(reg)
A
= z
2
(3.56)
1
2 (1+t)u
dt q(t)e
(3.57)
k (1)k Lk [m]Pk (u),

Pk (u),
k (1)k Lk [m]
(3.58)
248
where
1
Pk (u) =
ds e 2 (1s)u pk (s) =
Pk (u)
=
1

1 iu k iu
e2,
k!
2
(3.59)

1 i u k i u
e 2
k! 2
(3.60)
ds e 2 (1s)u pk (s) =
are projectors in the -product algebra given by functions of u and u,

viz.
Pk F = Lk [f ]Pk ,
(3.61)
Pk Pl = kl Pk ,
Pk F = Lk [f]Pk ,
(3.62)
Pk Pl = kl Pk ,

1
i
i
1
= 1 ds e 2 (1s)u f(s) with Lk [f ] and Lk [f] given
for F (u) = 1 ds e 2 (1s)u f (s) and F (u)
in (3.42). The projectors also obey (u 2ik) Pk = 0 and y Pk z = i(k + 1)(Pk1 +
Pk+1 ) with P1 0. We note the opposite signs in front of s in the exponents of (3.30), (3.31)
and (3.59), (3.60), resulting in the (1)k in the projector part (3.58) of the internal connection,
which we can thus write as

c1
(proj)
= iz
A
2k P2k
k Pk 1 1
1 + 2k
k=0

c1
(3.63)
+ 1 1+
2k+1 P2k+1 ,
3 + 2k

c2
(proj)
= i z
2k P2k
k Pk 1 1
A
1 + 2k
k=0

c2
+ 1 1+
2k+1 P2k+1 ,
(3.64)
3 + 2k
which are analytic functions of in a finite region around the origin. For example, for c1 =
c2 = 1, they are real analytic for 3 < Re < 1, where also the particular solution can be shown
to be real analytic [5]. The reality conditions on the k and k parameters are as follows:
(4, 0) and (2, 2) signature: k , k independent,
(3, 1) signature:
k = k .
(3.65)
(3.66)
Taking = 0 there remains only the projector part, leading to the following vacuum solutions
= 0,
A = iz

k=0
1 iu
k!
2
k
iu
e2,
A = i z

k=0
1 i u
k! 2
k
(3.67)
i u
e 2 .
(3.68)
The Z2 Z2 symmetry (2.37) acts by

k 1 k ,
k 1 k .
(3.69)
249
The maximally symmetric spaces discussed in Section 3.2 are recovered by setting k = and
k = for all k. In Euclidean and Kleinian signatures, and are independent, leading to four
solutions related by Z2 Z2 transformations. In Lorentzian signature, = leading to two
solutions related by Z2 symmetry.
3.3.2. Spacetime component fields
The calculation of the component fields follow the same steps as in [5]. The spin s 1 Weyl
tensors vanish, while the scalar field is given by

(x) = h2 x 2 = 1 2 x 2 .
(3.70)
In order to compute the gauge fields, we first need to compute the quantity K given in (3.17).
This calculation is formally the same as the one spelled out in the case of k = k = 0 in [5], and
the result is

Q
Q
K =
(3.71)
v v +
v v ,
4i
4i
where
(1 a 2 )2
Q=
4
2 2
= (1 a )
Q
4
and
1
1
ds
1
1
ds
ds
(1 + s)(1 + s )n(s)n(s )
,
(1 ss a 2 )4
(3.72)
ds
(1 + s)(1 + s )n(s)
n(s
)
,
(1 ss a 2 )4
(3.73)

v = 1 + a 2 y + 2(a y)
,

v = 1 + a 2 y + 2(ay)
,
with a
= a defined in (B.3). We can simplify Q using n(t) = (1 + t) + q(t) +
with pk (t) given by (3.41) and k by (3.46) and (3.47). After some algebra we find

Q ; {k } = Q(reg) () + Q(proj) ; {k } ,
Q(reg) =
(1 a 2 )2
4
2

Q(proj) = 1 a 2
1

k=0
1
ds
1
ds
(1 + s)(1 + s )q(s)q(s )
,
(1 ss a 2 )4
(3.74)
k k pk (t),
(3.75)
(3.76)

4k a 2k
(k k+1 )2 (1)k + Lk (q) (1)k+1 + Lk+1 (q) , (3.77)
k!
where we note that Q depends on k only via k k+1 . The same expression with q q and
The regular part, which was computed in [5], is given by
k k holds for Q.
(reg)
Q(reg) = Q+
(reg)
+ Q ,
2

(1 a 2 )2 4 4p
c1
c1
(reg)
Q+ =
1
,
1+
a
4
2p + 1
2p + 3
2p
(reg)
Q
p=0

2
2

(1 a )
p=0
2

c1
c1
4
4p+2
1
,
1+
a
2p + 3
2p + 3
2p + 1
(3.78)
(3.79)
(3.80)
250
while a similar expression, obtained by replacing c1 c2 , holds for Q.

Since K is bilinear in the y and y oscillators, it immediately follows that all higher spin
fields vanish. Moreover, after some algebra, we find that the vierbein and so(4; C) connection
are given by

ea = f1 x 2 dx a + f2 x 2 x a dx b xb ,
(3.81)
2 (0)
2 (0)
= f x ,
(3.82)
= f x ,
where
1 + (1 a 2 )2 Q
,
16a 4 QQ
[1 + (1 + a 2 )2 Q][1 + (1 + a 2 )2 Q]
1 + (1 a 2 )2 Q
f =
,
16a 4 QQ
[1 + (1 + a 2 )2 Q][1 + (1 + a 2 )2 Q]
f=
(3.83)
(3.84)
and
f 1 + 2 x 2 f 2 =
2
,
h2
f2 =
2(1 + a 2 )4
(f Q + fQ).
(1 a 2 )2
(3.85)
By a change of coordinates, the metric can be written locally, in a given coordinate chart, as a
foliation

ds 2 = d 2 + R 2 d32 ,
(3.86)
R 2 ( ) = 2 sinh2 ( ),
where x 2 = tan2 2 with = 1, and d32 is a three-dimensional metric of constant curvature
with suitable signature, and [6]
f1 h2
.
2
One has the following simplifications in specific models:
=
Type A model: Q = Q,
Chiral model:
= 0,
Q
1 + (1 a 2 )2 Q
,
1 + (1 + 6a 2 + a 4 )Q
1 + (1 a 2 )2 Q
=
.
1 + (1 + a 2 )2 Q
=
(3.87)
(3.88)
(3.89)
The metric may have conical singularities, namely zeroes R(0 ) = 0 for which R|0 = 1 (we
note that | =0 = 1, so that = 0 is not a conical singularity). The scale factor depends heavily
on as well as the choice of the infinitely many discrete parameters k and k . This makes the
analysis unyielding, and we shall therefore limit ourselves to the case of vanishing discrete parameters and || 1. The resulting analysis was performed in [6] in Lorentzian signature, and it
generalizes straightforwardly to Euclidean and Kleinian signatures. To this end, one examines the
integrals (3.72) and (3.73) in the limits a 2 1, where there are potentially divergent contributions from the region of the integration domain where s and s approach 1. These contributions
are actually finite for a 2 = 1, while they diverge as

c1
Q(reg) =
(3.90)
log 1 + a 2 + O 2 ,
6
when a 2 1 (and the O( 2 ) contributions are finite).
Focusing
on a single chart, as listed
in (3.22)(3.26), a 2 is bounded from below by (1 2 )(1 + 2 )1 , and hence, if || 1,
251
then |Q| 1, and consequently the factor defined in (3.87) remains finite. Thus, for small
enough , there are no conical singularities within the coordinate charts. However, they may
appear for some finite critical .
While the Q functions are highly complicated for = 0, they simplify drastically at = 0,
where we find

2
4k a 2k
Q = 1 a2
(3.91)
(k k+1 )2 .
k!
k=0
Setting (k k+1 )2 = 1, yields

An analogous expression can be found for Q.
Q=
1
.
(1 a 2 )2
(3.92)
= (1 a 2 )2 , which is necessarily the case in the Lorentzian models, then the

If Q = Q
equation system for and becomes degenerate, and one finds
(1 a 2 )2 (0) ( ab ) dxa xb
=
,
8a 2
2x 2
ab
(1 a 2 )2 (0) ( ) dxa xb
=
=
,

8a 2
2x 2
leading to the degenerate vierbein
=
e =
(3.93)
(3.94)
x x a dxa
,
x 2 h2
(3.95)
and metric
ds 2 =
4(x a dxa )2
.
2 x 2 h2
(3.96)
3.4. Solutions of non-minimal model (type 2)

The non-minimal model admits the following solutions
= 0,
S = z (y, y),
S = z (y, y),
(3.97)
provided that
= = 1,
[, ] = 0,
( ) = ,
( ) = .
(3.98)
The elements and can be written as

= 1 2P ,
= 1 2P ,
(3.99)
where P (y, y)
and P (y, y)
are projectors obeying
P P = P ,
[P , P ] = 0,
(P ) = P .
(3.100)
A set of such projectors is described in Appendix C, where we also explain why the projectors
can be subject to the -conditions of the non-minimal model, given in (2.10), but not those of the
minimal model, given in (2.9), unless one develops some further formalism for handling certain
divergent -products.
P P = P,
(P ) = P ,
252

Turning to the computation of the space components of the master fields, since z starcommutes with L, it immediately follows from (3.16), (2.73) and (3.97) that
K = 0.
(3.101)
From (3.16) this in turn implies that all HS gauge fields and the spin-1 gauge field vanish, while
the metric is that of maximally symmetric spacetime. To that extent, the type 1 solution looks like
the type 0 solution, but it does differ in an important way, namely, here the internal connection,
i.e., the spinor component A of the master 1-form, is non-vanishing. Indeed, (3.97), (3.98)
and (3.8) give the result
A = iz V (x; y, y),
A = i z V (x; y, y),
(3.102)
where the quantities V and V , which shall be frequently encountered in what follows, are defined
by
V = L1 P L,
V = L1 P L.
(3.103)
Their explicit evaluation is given in Appendix D, with the result (D.21).

Whilst the internal connection does not turn on any spacetime component fields, it does, however, affect the interactions as it does not obey the physical gauge condition normally used in the
weak-field expansion [21], namely that the internal connection should vanish when the zero-form
vanishes. In this sense, the internal connection may be viewed as a non-trivial flat connection in
the non-commutative space.
3.5. Solutions of non-minimal chiral model (type 3)
In the case of the non-minimal chiral model, defined in Section 2.3, it is possible to use
projectors P (y, y)
to build solutions with non-vanishing Weyl zero-form and higher spin fields.
They are
= (1 P ) ,
S = z P ,
S = z ,
(3.104)
where
P P = P,
= 1,
[P , ] = 0,
(P ) = P ,
( ) = .
(3.105)
These elements of the -product algebra can be constructed as in Section 3.4 and Appendix C.
For the purpose of exhibiting explicitly the spacetime component fields, we shall choose to
work with the simplest possible projectors, namely
P+ (y) = 2e2 uv = 2e yby ,
P (y)
= 2e2 u v = 2e
y b y
where = 1, and u, v, u,
v,
b and b are defined in Appendices B and C.
(3.106)
(3.107)
253

The master gauge field and zero-form is given by
(0)
(0)
+
+
e + + W = e
and
2 V
,
4i y y

= L1 (1 P ) (L) Z=0 = 1 V |y =0 ,
(3.108)
(3.109)
where V is given by (D.21) and we have used (2.17). Remarkably, since there is no y-dependence
in the Weyl zero-form , it is covariantly constant in the sense that (m)(n)
vanishes unless
m = 0. Moreover, using (D.21), it is straightforward to compute the constant value of the physical
scalar field, with the result

1 + 2
(1)n1 +n2 2 n1 ,n2 ,
(x) = 1 4
(3.110)
n1 ,n2 Z+ 12
where n1 ,n2 are constrained as

in (C.15). Summing over all n2 , as explained in Appendix C, and
1
k
using (C.17) with x = 0, i.e.,
k=0 (1) = 2 , one finds that for the reduced projector (C.19),
the scalar field is given by

(x) = 1 2
(3.111)
(1)n 2 n ,
nZ+ 12
where n obey the condition given below (C.19). Finally, setting n = (n), one ends up with
P = 1, i.e., in the type 0 case, where indeed (x) = 0.
In the special cases of (3.106) and (3.107), one finds

[2y a (1 + a 2 )y]b[2a y + (1 + a 2 )y]
,
V+ = L1 P+ L = 2 exp
(3.112)
(1 a 2 )2

ay
b[2
+ (1 + a 2 )y]
[2ya (1 + a 2 )y]
,
V = L1 P L = 2 exp
(3.113)
(1 a 2 )2
where a and b are defined in Appendix B, and is defined in (3.106) and (3.107). The
physical scalar is now given in both cases by
(x) = 1,
(3.114)
and the self-dual Weyl tensors in both cases by (s = 1, 2, 3, . . .)

(2s) = 0,
while the anti-self-dual Weyl tensors take the form
2

h 1 s
+
2s+1
(2s 1)!!
U( 1 U s V s+1 V 2s ) ,
1 2s = 2
h2

1 s
2s+1
=
2
(2s
1)!!
( 1 s s+1 2s ) ,
1 2s
h2
(3.115)
(3.116)
(3.117)
with spinors (U, V ) defined in (B.11).

In the case of 2 = 1 in Euclidean signature, we only need to use one coordinate chart, in
which 0 h2 1. The Weyl tensors blow up in the limit h2 0, preventing the solution from
254
approaching H4 in this limit. In this sense the above solution is a non-perturbative solution without weak-field limit in any region of spacetime. Indeed, in the perturbative weak-field expansion
around the H4 solution, the scalar field has non-vanishing mass, preventing the linearized scalar
field from being a non-vanishing constant.
In the case of 2 = 1 in Euclidean signature, the base manifold consists of two charts,
covered by the coordinates in (3.22). Thus, in each chart we have 1 h2 < 2, and so the local
representatives (3.116) and (3.117) of the Weyl tensors are well defined throughout the base
manifold.
Finally, in the case of 2 = 1 in Kleinian signature, one also needs two charts, with 0
2
h 2, and hence the Weyl tensors blow up in the limit h2 0 preventing the solution from
approaching H3,2 in this limit.
From the Weyl tensors, which are not in themselves HS gauge invariant quantities, one can
construct an infinite set of invariant (and thus closed) zero-forms [5], namely
4 4

d y d z
p .
C2p
(3.118)
=
()
4
(2)
Remarkably, on our solution they all assume the same value, given by the constant value of the
scalar field, viz.

1 + 2
= (1 V )2p y=y=0
=14
(1)n1 +n2 2 n1 ,n2 .
C2p
(3.119)
n1 ,n2 Z+ 12
The calculation of the metric in the two models proceeds in a parallel fashion as follows:
The P+ solution: From (3.108) and (3.112) a straightforward computation yields the result
(0)
4
e
= e
+ 12(1 + h)h b( (ba) ) ,
(0)
=
+ 12h
(3.120)
b( b )
,
(3.121)

(0)

= + 4(1 + h)2 h4 (aba)

.
(ab)
b + 2(ab)
(3.122)
First we solve for the spin connection from (3.121) by inverting the hyper-matrix that multiplies
(0) , obtaining the result
(0)
(0)
(0)
b + g2 b b ,
= g1 8g b
(3.123)
where
g1 =
1
,
1 4g 2
g2 =
4g
,
(1 2g)(1 4g)
g = h4 .
(3.124)
Substituting this result in (3.120) then gives the vierbein

a
e
=

a
2
g3 + g4 2 x x a + g5 2 (J x) (J x)a ,
h2 (1 + 2g)
(3.125)
where
g3 = 1 + 2h2 ,
g4 = 2g,
g5 =
6g
,
1 4g
(3.126)
255
a eb takes
and the spin connections are given in (3.123) and (3.122). Thus, the metric g = e
ab
the form
g =

g32 + g4 2 x 2 g4 + 2g3 x x
h4 (1 + 2g)2

+ g5 2 x 2 g5 + 2g3 (J x) (J x) .
(3.127)
The vierbein thus has potential singularities at h2 = 0 and h2 = 2. The limit h2 0 is a boundary in the case of 2 = 1 in Euclidean signature and 2 = 1 in Kleinian signature. At these
boundaries e a h2 x x a , i.e., a scale factor times a degenerate vierbein. In the limit h2 2
one approaches the boundary of a coordinate chart in the case of 2 = 1 in Euclidean signature and 2 = 1 in Kleinian signature. Also in this limit, the vierbein becomes degenerate, viz.
e a h2 (J x) (J x)a .
The P solution: A parallel computation that uses (3.108) and (3.113) yields the result
(0)
2 2
4
e
= e
+ 12 x (1 + h)h b( (ba) ) ,
2

(0)
= + 12 2 x 2 h4 b( b )
,

(0)
= + 4(1 + h)2 h4 (a ba)

b + 2(ba) (ba) ,
(3.128)
(3.129)
(3.130)
where b is defined in (B.9). As before, solving for the spin connection from (3.129) by inverting
the hyper-matrix that multiplies (0) , we obtain
(0)
(0)
(0)
b
+ g 2 b b ,
= g 1 8g b
(3.131)
where
g 1 =
1
,
1 4g
g 2 =
4g
,
(1 2g)(1
4g)
2

g = 2 x 2 h4 .
(3.132)
Substituting this result in (3.128) then gives the vierbein

a
=
e

a
2
2
a
2
x)a ,
+
g
x
x
+
g
(
J
x)
(
J
h2 (1 + 2g)
(3.133)
where Jab is defined in (B.10)

g 4 = 22 x 2 h4 ,
g 5 =
62 x 2 h4
,
1 4g
(3.134)
a eb takes
and the spin connections are given in (3.131) and (3.130). Thus, the metric g = e
ab
the form
g =
4
h4 [1 + 2g]
2

+ g 4 2 x 2 g 4 + 2 x x + g 5 2 x 2 g 5 + 2 (Jx) (Jx) . (3.135)
The vierbein has potential singularities at h2 = 0, h2 = 2 and h2 = 23 . The singularities at h2 = 0

and h2 = 2 are related to degenerate vierbeins exactly as for the P + solution. The singularity
at h2 = 23 , which arises in the case of 2 = 1 in Euclidean and Kleinian signatures, also gives a
degenerate vierbein. This is an intriguing situation since the latter degeneration occurs inside the
coordinate charts.
256
4. Conclusions
Starting from HS gauge theories in four dimensions based on infinite-dimensional extensions of SO(5; C), we have determined their real forms in spacetimes with Euclidean (4, 0)
and Kleinian (2, 2) signature, in addition to the usual Lorentzian (3, 1) signature. We have then
found three new types of solutions in addition to the maximally symmetric ones. Type 1 solutions, which are invariant under an infinite-dimensional extension of SO(4 p, p), give us a
non-trivial deformation of the maximally symmetric solutions, and depend on a continuous real
parameter as well as an infinite set of discrete parameters. Interestingly, a particular choice of the
discrete parameters, in the limit of vanishing continuous parameter, gives rise to a degenerate,
indeed rank one, metric. Given that degenerate metrics are known to play an important role in
topology change in quantum gravity [17], it is remarkable that such metrics emerge naturally in
HS gauge theory. Type 2 solutions, which provide another kind of deformation of the maximally
symmetric solutions, have a non-vanishing spinorial master one-form. Type 3 solutions are particularly remarkable because all the higher spin fields are non-vanishing, and the corresponding
Weyl tensors furnish a higher spin generalization of type D gravitational instantons. It would be
interesting to apply the framework we have used in this paper to finding pp-wave, black hole and
domain wall solutions with non-vanishing HS fields.
We stress that our models in Euclidean and Kleinian signatures are formulated using the 4D
spinor-oscillator formulation. It would be interesting to compare these models to the vectoroscillator formulation [23]. The latter exists in any dimension and signature, and relies on the
gauging of an internal Sp(2) symmetry. At the full level, the vector-oscillator master field equations, in any dimension and signature, are formulated using a single Sp(2)-doublet Z-oscillator,
leaving, apparently, no room for parity violating interactions. The precise relation between the
spinor and vector-oscillator formulations in D = 4 therefore deserves further study.
In the context of supersymmetric field theories, including supergravity, the non-Lorentzian
signature typically presents obstacles since the spinor properties are sensitive to the spacetime
signature. Here, however, we have considered bosonic HS gauge theories in which the spinor
oscillators play an auxiliary role, and we have formulated the non-Lorentzian signature theories
with suitable definition of the spinors without having to face such obstacles. Remarkably, nonsupersymmetric 4D theories in Kleinian signature describing self-dual gravity arise in worldsheet
N = 2 supersymmetric string theories, known as N = 2 strings. For reasons mentioned in the
Introduction, it is an interesting open problem to find a niche for Kleinian HS gauge theory in a
variant of an N = 2 string.
There are several other open problems that deserve investigation. To begin with, we have not
determined the symmetries of type 2 and type 3 solutions. While it may be useful in its own right
to determine whether our type 3 solutions support a complex, possibly Khler, structure up to
a conformal scaling, such results may be limited in shedding light to the geometry associated
with infinitely many gauge fields present in HS gauge theory. The correct interpretation of the
singularities or degeneracies in the metrics we have found also requires sufficient knowledge of
the HS geometry. Furthermore, a proper formulation of the HS geometry would also provide a
framework for constructing invariants that could distinguish the gauge inequivalent classes of
exact solutions.
It would also be interesting to study the fluctuations about our exact solutions, and explore their potential application in quantum gravity and cosmology. Similarities between the
frameworks for studying instanton and soliton solutions of the non-commutative field theories
257
(see, for example, [24]), and in particular open string field theory, are also worth investigating.
Acknowledgements
We are grateful to J. Engquist and A. Sagnotti for discussions. P.S. thanks the Physics Department at Texas A&M University, and E.S. thanks the Scuola Normale Superiore, where part of
this work was done, for hospitality. The research of C.I. was supported in part by INFN; by the
MIUR-PRIN contract 2003-023852; by the EU contracts MRTN-CT-2004-503369 and MRTNCT-2004-512194; by the INTAS contract 03-51-6346; and by the NATO grant PST.CLG.978785.
The research of E.S. was supported in part by NSF Grant PHY-0555575. The research of
P.S. was supported in part by a visiting professorship issued by Scuola Normale Superiore;
by INFN; by the MIUR-PRIN contract 2003-023852; by the EU contracts MRTN-CT-2004503369 and MRTN-CT-2004-512194; by the INTAS contract 03-51-6346; and by the NATO
grant PST.CLG.978785.
Appendix A. General conventions and notation
We use the conventions of [25] in which the real form of the SO(5; C) generators obey
[MAB , MCD ] = iBC MAD + 3 more,
(MAB ) = (MAB ),
(A.1)
where AB = (ab ; 2 ). The commutation relations above decompose as

[Mab , Mcd ] = 4i[c|[b Ma]|d] ,
[Mab , Pc ] = 2ic[b Pa] ,
[Pa , Pb ] = i Mab .
2
(A.2)
The corresponding oscillator realization is taken to be

Mab =

1
(ab ) y y + ( ab ) y y ,
8
Pa = (a ) y y .
4
(A.3)
Our spinor conventions are
= 2 ,
and

( ) =
= ,
(A.4)
for SU(2),
for SL(2, C),
for Sp(2).
(A.5)
Oscillator indices are raised and lowered according to the following conventions, A = A ,
A = A . The reality conditions on oscillators have been summarized in (2.39), (2.40) and
(2.41). The van der Waerden symbols obey
= ab + ( ab ) ,

1
abcd cd = (ab ) ,
2

= ab + ab ,

1
abcd cd = ( ab ) ,
2
(A.6)
(A.7)
258
det ab , and the reality conditions
a
a ) for SU(2),
a (a ) = (
= ( )
= ( a )
for SL(2, C),
a
a
for Sp(2),
( )
= ( )
where =
and

ab
ab
( )
= ( ab )
ab
( )
(A.8)
for SU(2),
for SL(2, C),
for Sp(2).
(A.9)
Convenient representations are:

SU(2):
a = i, i , a = i, i , = i 2 ;

SL(2, C): a = i 2 , i i 2 , a = i 2 , i 2 i ,

Sp(2):
a = 1, i , a = 1, i , = i 2 ,
(A.10)
= i 2 ;
(A.11)
(A.12)
where in the last case i = ( 1 , i 2 , 3 ). Combining (2.71) with (A.3), the real form of the
so(5; C)-valued connection can be expressed as
=

1
dx y y + y y + 2e y y ,
4i
(A.13)
where
1
1
= (ab ) ab ,
= ( ab ) ab ,
4
4
Likewise, for the curvature R = d + one finds
e = (a ) ea .
2
R = d + + e e ,
= d + + e e ,
R
(A.14)
(A.15)
(A.16)
R = de + e + e ,
(A.17)
and
Rab = dab + a c cb + 2 ea eb ,
Ra = dea + a b eb .
(A.18)
Appendix B. Further notation used for the solutions

The gauge function L(x; y, y)
defined in (3.19) can be written as
L=
2h
exp(iya y),
1+h
(B.1)
where
a =
x
,
1+h
x 2 = ab x a x b ,

x = a xa ,

h = 1 2 x 2 .
(B.2)
(B.3)
259
Useful relations that follow from these definitions are

a2 =
1h
,
1+h
h=
1 a2
.
1 + a2
(B.4)
The MaurerCartan form based on L defined in (3.19) yields the vierbein and Lorentz connection
( a ) dxa
,
h2
with Riemann tensor given by
e(0) =
(0) =
2 ( ab ) dxa xb
,
h2
R(0), = 2 (g(0) g(0) g(0) g(0) ).
(B.5)
(B.6)
In the case of type 3 solutions, a useful definition is

i
= .
2
b = 2( ) ,
(B.7)
It obeys the relation (b2 ) = 14 and it defines an almost complex structure via the relations
(see, for example, [26])
1 ab
Jab ,
8
Similarly, using the definition
b =
Jab = (ab ) b ,
,
b = a 2 (a b a)
Ja c Jc b = ab .
(B.8)
(B.9)
we have the relations

1
b = ab Jab , Jab = (ab ) b ,
(B.10)
Ja c Jc b = ab .
8
Finally, we have the following definition for spinors used in describing a type 3 solution:
xa
U = ( a ) ,
x2
xa
V = ( a ) .
x2
(B.11)
Appendix C. Weyl-ordered projectors

Weyl-ordered projectors P (y, y)
can be constructed by recombining (y, y)
into a pair of
Heisenberg oscillators (ai , bj ) (i, j = 1, 2) obeying

j
a i , b j = i .
(C.1)
For example, one can take
a1 = u = y ,
a2 = u = y ,
b 1 = v = y ,
b = v = y ,
2
(C.2)
(C.3)
where the constant spinors are normalized as

i
= ,
2
i
= .
2
(C.4)
260
The projectors, obeying the appropriate reality conditions, take the form

n1 ,n2 Pn1 ,n2 ,
P =
n1 ,n2 Pn1 ,n2 ,
P=
n1 ,n2 Z+ 12
(C.5)
n1 ,n2 Z+ 12
where n1 ,n2 {0, 1} and n1 ,n2 {0, 1}, with

(3, 1) signature:
(4, 0) and (2, 2) signatures:
n1 ,n2 = n1 ,n2 ,
n1 ,n2 , n1 ,n2 independent,
(C.6)
(C.7)
and
Pn1 ,n2 = 4(1)n1 +n2
wi = bi ai = bi ai +
1 + 2
2
e2
i i wi
Ln1 1 (4 1 w1 )Ln2 2 (4 2 w2 ),
2
1
1
= ai b i
2
2
(no sum),
(C.8)
(C.9)
1 x d
with i = ni /|ni | and Ln (x) = n!
e dx n (ex x n ) are the Laguerre polynomials. The projector
property of P and P follows from
Pm1 ,m2 Pn1 ,n2 = m1 n1 m2 n2 Pn1 ,n2 ,
(C.10)
(wi ni ) Pn1 ,n2 = 0,
(C.11)
(Pn1 ,n2 ) = Pn1 ,n2 .
(C.12)
Here, wi
1
2
is the Weyl-ordered form of the number operator, and

i
for ni > 0,
|ni ni |
1
2(1)ni 2 e2 i wi Lni i (4 i wi ) =
n
i
2 |ni ni |
2
for ni < 0,
(1)
where |ni =
n 1
(bi ) i 2
(ni 12 )!
(C.13)
|0 with ni > 0 belongs to the standard Fock space, built by acting with
bi on the ground state |0 obeying ai |0 = 0, while |ni =
n 1
(ai ) i 2

|0
(ni 12 )!
for ni < 0 are anti-Fock
= 0 obeying bi |0
= 0. Eq. (C.10)
space states, built by acting with ai on the anti-ground state |0
holds formally, since the inner product between a Fock space state and an anti-Fock space state
vanishes. However, the corresponding Weyl-ordered projectors have divergent -products, as can
be seen from the lemma

1
s +t
exp
uv
.
esuv etuv =
(C.14)
1 + st4
1 + st4
Thus, lacking, at present, a suitable regularization scheme that does not violate associativity and
other basic properties of the -product algebra, we shall restrict our attention to projectors that
are constructed in either the Fock space or the anti-Fock space, i.e.,
n1 ,n2 = 1 only if (n1 , n2 ) Q,
(C.15)
where Q is anyone of the four quadrants in the (n1 , n2 ) plane. From (C.12), it follows that these
projectors are not invariant under the map, and therefore the master fields type 2 and type 3
solutions will be those of the non-minimal model, where the conditions are relaxed to
conditions, which are certainly satisfied.
We also note that in order to solve the higher-spin equations it is essential that

(n1 ,n2 n1 ,n2 n1 ,n2 n1 ,n2 )Pn1 ,n2 = 0,
[P , P ] =
261
(C.16)
n1 ,n2
which holds for independent n1 ,n2 and n1 ,n2 parameters (in the Euclidean and Kleinian signatures). Moreover, one can work with a reduced set of oscillators, say a1 = u and b1 = v, by
taking n1 ,n2 = (n2 )n1 , where (x) = 0 for x < 0 and (x) = 1 for x > 0, and summing over
all values of n2 using

t k Lk (x) = (1 t)1 exp xt (1 t)1 .
(C.17)
k=0
Setting n = n1 and = 1 , this leads to

n Pn ,
P =
n Pn ,
P=
nZ+ 12
(C.18)
nZ+ 12

Pn = 2(1)n 2 e2 uv Ln (4 uv),
(C.19)
where n and n obey suitable reality conditions, and we note that n = (n)n idem n as a
consequence of (C.15). Finally, setting n = (n), and using (C.17) once more, one finds that
setting all -parameters equal to 1 gives P = 1.
Appendix D. Calculation of V = L1 P L
In this appendix we compute V = L1 P L where L is the gauge function given in (B.1)
and P is a projector of the form given in (C.5). Let us begin by considering the case of P = P 1 =
2e2uv , i.e.,
V=
8h2
eiya y eyby eiya y ,
(1 + h)2
(D.1)
where ya y = y a y and yby = y b y , with a and b given by (B.2) and (B.7). The first
-product can be performed treating the integration variables ( , ) and ( , ) as separate
real variables. Using the formulae (B.1) provided in [5], we find
V=
8h2
ya)b(y+a
y)
eiya y+(y
eiya y .
(1 + h)2
The remaining -product leads to the Gaussian integral

4 4
8h2
d d 1 I MI J J + I NI +(yya)b(y+a
y)
V=
e2
,
(1 + h)2
(2)4
(D.2)
(D.3)
where I = ( , ; , ) and I = ( , ; , ) = J J I , with block-diagonal symplectic metric = , and

A i
M=
,
(D.4)
i B

2b
ia + 2ba
0
ia
,
A=
,
B=
(D.5)
i a
0
ia 2ba 2aba
262
i(1 2ib)a y + 2by

y + i a(1
+ 2ib)y
2aba
N =
.
ia y
i ay
(D.6)
The Gaussian integration gives

V=
1 I
8h2
1 J
y)
e 2 N (M )I NJ +(yya)b(y+a
.
(1 + h)2 det M
From det M = det(1 + AB), and noting that the matrices defined as

2
2
AB a 2
BA a 2
a b a 2 ba
a b
,
C
C
=
=
ab
aba
aba
2
2i
2i
are nilpotent, i.e., C 2 = C 2 = 0, one finds

4
det M = 1 a 2 ,
(D.7)

ba
,
aba
(D.8)
(D.9)
and, using 1 a 2 = 2h/(1 + h), the pre-factor in V is thus given by

8h2
= 2.
(1 + h)2 det M
(D.10)
Next, using geometric series expansions, one finds

i
i(1 a 2 )B + 2B C (1 a 2 ) 2iC
1
M =
,
i(1 a 2 )A + 2AC
1 a 2 + 2i C
(1 a 2 )
(D.11)
and
y
4a 2 yby + 2(1 + 4a 2 a 4 )yba y (3 a 2 )(1 + a 2 )y aba
1 I 1 J
N
=
. (D.12)
N M
J
I
2
(1 a 2 )2
Adding the classical term in the exponent in (D.3) yields the final result

[2y a (1 + a 2 )y]b[2a y + (1 + a 2 )y]
.
V = 2 exp
(1 a 2 )2
(D.13)
The projector property V V = V follows manifestly from

V = 2 exp(2u v),
[u,
v]
= 1,
(D.14)
where
u = ,
v = ,
(D.15)
with

[(1 + a 2 )y + 2a y]
(D.16)
,
[ , ] = 2i .
1 a2
Thus, the net effect of rotating the projector P 1 (u, v) given in (C.19) is to replace the oscillators
2
u and v by their rotated dittos u and v.
We claim, without proof, that this generalizes to any n,
viz.
=
v).
L1 Pn (u, v) L = Pn (u,
(D.17)
263
Similarly, for Pn (u,

v)
we have
v),
L1 Pn (u,
v)
L = Pn (u,
(D.18)
where
u = ,
v = ,
(D.19)
with
=

[(1 + a 2 )y + 2ay]
,
1 a2
[ , ] = 2i .
Finally, using [ , ] = 0, we deduce that

v).
V = L1 P L =
n1 ,n2 Pn1 ,n2 (u,
v;
u,
(D.20)
(D.21)
n1 ,n2
References
[1] M.A. Vasiliev, Consistent equations for interacting gauge fields of all spins in 3 + 1 dimensions, Phys. Lett. B 243
(1990) 378.
[2] M. Flato, C. Fronsdal, One massless particle equals two Dirac singletons: Elementary particles in a curved space.
6, Lett. Math. Phys. 2 (1978) 421.
[3] J. Engquist, P. Sundell, L. Tamassia, Singleton strings, hep-th/0701081.
[4] S.F. Prokushkin, M.A. Vasiliev, Higher-spin gauge interactions for massive matter fields in 3D AdS spacetime,
Nucl. Phys. B 545 (1999) 385, hep-th/9806236.
[5] E. Sezgin, P. Sundell, An exact solution of 4D higher-spin gauge theory, Nucl. Phys. B 762 (2007) 1, hep-th/
0508158.
[6] E. Sezgin, P. Sundell, On an exact cosmological solution of higher spin gauge theory, hep-th/0511296.
[7] V.E. Didenko, A.S. Matveev, M.A. Vasiliev, BTZ black hole as solution of 3d higher spin gauge theory, hep-th/
0612161.
[8] C. Iazeolla, P. Sundell, in preparation.
[9] J. Engquist, O. Hohm, Higher-spin ChernSimons theories in odd dimensions, arXiv: 0705.3714 [hep-th].
[10] G.W. Gibbons, S.W. Hawking (Eds.), Euclidean Quantum Gravity, World Scientific, 1993.
[11] G.W. Gibbons, Euclidean quantum gravity: The view from 2002, in: The Future of Theoretical Physics and Cosmology, Cambridge Univ. Press, 2002, p. 351.
[12] H. Ooguri, C. Vafa, Selfduality and N = 2 string magic, Mod. Phys. Lett. A 5 (1990) 1389.
[13] K. Junemann, O. Lechtenfeld, Chiral BRST cohomology of N = 2 strings at arbitrary ghost and picture number,
Commun. Math. Phys. 203 (1999) 53, hep-th/9712182.
[14] C. Devchand, O. Lechtenfeld, Extended self-dual YangMills from the N = 2 string, Nucl. Phys. B 516 (1998) 255,
hep-th/9712043.
[15] M.A. Vasiliev, Higher-spin gauge theories in four, three and two dimensions, Int. J. Mod. Phys. D 5 (1996) 763,
hep-th/9611024.
[16] J. Engquist, E. Sezgin, P. Sundell, On N = 1, 2, 4 higher spin gauge theories in four dimensions, Class. Quantum
Grav. 19 (2002) 6175, hep-th/0207101.
[17] G.T. Horowitz, Topology change in classical and quantum gravity, Class. Quantum Grav. 8 (1991) 587.
[18] E.J. Flaherty, Hermitian and Khlerian Geometry in Relativity, Springer, 1976, p. 175.
[19] A.S. Lapedes, M.J. Perry, Type D gravitational instantons, Phys. Rev. D 24 (1981) 1478.
[20] E. Sezgin, P. Sundell, Holography in 4D (super) higher spin theories and a test via cubic scalar couplings, JHEP 0507
(2005) 044, hep-th/0305040.
[21] E. Sezgin, P. Sundell, Analysis of higher spin field equations in four dimensions, JHEP 0207 (2002) 055, hep-th/
0205132.
[22] K.I. Bolotin, M.A. Vasiliev, Star-product and massless free field dynamics in AdS(4), Phys. Lett. B 479 (2000) 421,
hep-th/0001031.
264
[23] M.A. Vasiliev, Nonlinear equations for symmetric massless higher spin fields in (A)dS(d), Phys. Lett. B 567 (2003)
139, hep-th/0304049.
[24] F.A. Schaposnik, Noncommutative solitons and instantons, Braz. J. Phys. 34 (2004) 1349, hep-th/0310202.
[25] E. Sezgin, P. Sundell, Higher spin N = 8 supergravity, JHEP 9811 (1998) 016, hep-th/9805125.
[26] C.N. Pope, Khler manifolds and quantum gravity, J. Phys. A 15 (1982) 2455.

Confinement in the q-state Potts field theory

Gesualdo Delfino a,b, , Paolo Grinza c,d
a International School for Advanced Studies (SISSA), via Beirut 2-4, 34014 Trieste, Italy
b Istituto Nazionale di Fisica Nucleare, sezione di Trieste, Italy
c Dipartimento di Fisica Teorica dellUniversit di Torino, via P. Giuria 1, I-10125 Torino, Italy
d Istituto Nazionale di Fisica Nucleare, sezione di Torino, Italy
Received 18 June 2007; accepted 12 September 2007

Abstract
The q-state Potts field theory describes the universality class associated to the spontaneous breaking of
the permutation symmetry of q colors. In two dimensions it is defined up to q = 4 and exhibits duality and
integrability away from critical temperature in absence of magnetic field. We show how, when a magnetic
field is switched on, it provides the simplest model of confinement allowing for both mesons and baryons.
Deconfined quarks (kinks) exist in a phase bounded by a first order transition on one side, and a second
order transition on the other. The evolution of the mass spectrum with temperature and magnetic field is
discussed.
PACS: 05.50.+q; 03.70.+k; 11.10.Kk
Keywords: Kinks; Confinement; Potts model
1. Introduction
Confinement is that property of quantum field theory for which excitations which are genuine
asymptotic particles in a region of coupling space become unobservable in isolation in another
region, where they leave the place to new asymptotic particles (mesons, baryons, . . .) of which
they can be seen as constituents (quarks). The case of two-dimensional spacetime provides
the framework in which this, as other general properties of quantum field theory, can be studied in
E-mail address: delfino@sissa.it (G. Delfino).

doi:10.1016/j.nuclphysb.2007.09.003
266
G. Delfino, P. Grinza / Nuclear Physics B 791 [FS] (2008) 265283
their simplest form. It is known that the gauge theory setting in which confinement is ordinarily
discussed in four dimensions becomes somehow redundant in two dimensions. Indeed, due to the
absence of transverse spatial dimensions, massless gauge fields do not carry particle degrees of
freedom in d = 2, so that an alternative description of the theory exists which relies on physical
excitations only. For example, two-dimensional quantum electrodynamics can be exactly mapped
onto the theory of a self-interacting neutral boson which makes quite transparent the presence
of quark confinement [1,2]. In absence of electromagnetic interaction the quarks correspond to
the solitons interpolating between the vacua of a periodic bosonic potential. Quark interaction
destroys the degeneracy of the bosonic vacua and removes the topologically charged excitations
from the spectrum of asymptotic states. What remains is a spectrum of mesons originating from
confinement of solitonantisoliton pairs.1
This mechanism of confinement through breaking of degeneracy of discrete vacua is quite
general in two dimensions and exhibits its most essential features in the case of a finite number of vacua originating from the spontaneous breaking of a discrete symmetry, with the kinks
interpolating between these vacua playing the role of the quarks. Then it is not surprising that
two-dimensional Ising field theory (i.e., the field theory describing the scaling limit of the twodimensional Ising model) provides the simplest model of confinement (only two vacua). The
associated mesonic spectrum was first studied in [3].
In theories, like Ising, with a one-component order parameter, the confined particles are made
of an even number of quarks. Indeed, in this case the vacua are located along a line in order
parameter space,2 so that the kink sequences starting from and going back to the true vacuum
(the only ones generating bound states via confinement) consist of a number 2j of kinks. The
lightest bound states are mesons (kinkantikink composites) corresponding to j = 1.
On the other hand, when the order parameter has more than one component we can find three
vacua located on a plane in order parameter space. Now we can have a three-kink sequence
making a loop through the three vacua. Such a sequence is confined into baryons, a finite number
of which must be stable sufficiently close to the deconfining point. Indeed, if m is the mass of
the kink, the lightest baryons will have mass 3m and will not be able to decay into two mesons
with mass 2m each.
In this paper we consider the simplest model of confinement allowing for baryons, i.e., the
field theory describing the scaling limit of the two-dimensional q-state Potts model. The latter
generalizes the Ising model to the case in which each site on the lattice can take q different
colors, and has an order parameter with q 1 components [5,6]. In absence of magnetic field,
the ferromagnetic q-state Potts model undergoes an ordering transition which is continuous up to
a number of colors qc , which in two dimensions equals 4 [7]. The scaling limit then can be taken
up to this value of q and produces a field theory in which a magnetic field confines the kinks and
allows for baryons at q = 3, 4.
In absence of the confining field, the quarks (kinks) behave as free neutral fermions in the Ising
case (q = 2), but become interacting for q = 3, 4. For q = 4 the zero-field theory is equivalent
to the sine-Gordon model at a specific value of the coupling ( 2 = 2 , see, e.g., [8]), while
for q = 3 the ultraviolet fixed point is non-trivial. In any case the confining interaction is nonlocal with respect to the quarks. The possibility of a quantitative study in weak field comes from
1 We refer to the generic case in which the -angle is not fine-tuned to the specific value which partially preserves
vacuum degeneracy.
2 This applies also to two-dimensional quantum electrodynamics, with the notion of one-dimensional order parameter
referred to the bosonic version. See [4] for a discussion of confinement in theories with more general bosonic potentials.
267
integrability of the q-state Potts field theory in zero-field [9]. Hence, form factor perturbation
theory [10] allows to express mass corrections in terms of the matrix elements of the magnetic
operator computed in [11].
Here, however, our main interest will be in a qualitative characterization of the evolution of
the mass spectrum for generic values of the temperature and of a magnetic field chosen to act on
a single color. For q = 3, 4 this choice allows for an extended phase on the parameter plane in
which the quarks are deconfined. Such a phase is bounded by the confining (first order) transition
on one side, and by a spontaneous breaking (second order) transition on the other. Outside this
region, the spectrum of asymptotic particles is made of mesons (everywhere) and baryons (at
least sufficiently close to the deconfining transition).
The paper is organized as follows. In Section 2 we discuss the q-state Potts model with our
choice of magnetic field, starting with the lattice definition and then switching to the field theoretical description of the scaling limit. In Section 3 we focus on the two-dimensional case and
recall the exact scattering solution for the low-temperature phase in zero-field, before showing
how this is related by duality to the scattering solution for the high-temperature phase. Section 4
is devoted to the weak field analysis, while the spectrum evolution as a function of temperature
and magnetic field is discussed in Section 5. Few final remarks are collected in Section 6.
2. Potts model with magnetic field
In this section we discuss the lattice definition of the q-state Potts model and the field theoretical description of the scaling limit. Some remarks about d > 2, and about q > 4 in d = 2 are
included although they are not used in the rest of the paper.
2.1. Lattice model
The q-state Potts model [5,6] is a generalization of the Ising model in which each site variable
s(x) at site x on the lattice can assume q different values (colors). In absence of magnetic field the
interaction only distinguishes whether nearest neighbor sites have equal or different color, so that
the Hamiltonian is invariant under the group Sq of permutation of the colors. If we add a magnetic
field H acting only on the sites with a specific color (say s = q), the reduced Hamiltonian can be
written as

1
s(x),s(y) H
s(x),q ,
H=
(1)
T
x
(x,y)
and is invariant under the group Sq1 of permutations of the first q 1 colors (the first sum is
over nearest neighbors).
In the ferromagnetic case at H = 0, the q configurations in which all the sites have the same
color minimize the energy and the system exhibits spontaneous magnetisation for sufficiently
low values of the temperature T . Above a critical temperature Tc the thermal fluctuations become
dominant and the system is in a disordered phase. If we introduce the variables
(x) = s(x),
1
,
q
= 1, 2, . . . , q,
(2)
satisfying the condition

q

=1
(x) = 0,
(3)
268
Fig. 1. Phase diagram of the model (1) for q > 2. The values H = 0 and H = correspond to the zero-field q- and
(q 1)-state Potts model, respectively. The ordered phase possesses q 1 degenerate ground states and is separated
from the disordered region by two phase transition lines (dashed and thick line). The dashed line is a first order transition
line along which q ground states are degenerate. The nature of the transition along the thick line depends on q and d.
the expectation values differ from zero only in the low-temperature phase and can be used
as order parameters.
When the magnetic field is switched on with a positive value, the ground state at T = 0
is unique (all sites have color q), and there can be no phase transition as the temperature is
increased.
Different is the situation for H < 0, q > 2. As H the color q becomes forbidden and
a zero-field (q 1)-Potts model is obtained. Then the critical points at H = 0 and H = are
the endpoints of a phase transition line which in the T H plane separates a low-temperature,
spontaneously magnetized phase with q 1 degenerate ground states from a high-temperature,
disordered phase (Fig. 1).
The nature of the transition at T = Tc (H ), H 0, depends on q and on the dimensionality d.
It is well known (see [6]) that in the zero-field q-state Potts model there exists a value qc (d) (not
necessarily integer3 ) such that the transition is continuous for q qc and first order for q > qc .
Accordingly, three cases can be distinguished for the transition going from Cq to Cq1 in Fig. 1:
(i) q 1 > qc . The transition is first order with q phases coexisting along the transition line.
Only at Cq q + 1 phases coexist.
(ii) 2 < q qc . The transition is continuous.
(iii) q 1 qc < q. The correlation length is finite at Cq and infinite at Cq1 . The nature of the
transition depends on d.
The value qc (2) is exactly known to be 4 [7], while qc (3) lies in between 2 and 3. Hence, in
d = 2 the transition induced by the field is continuous for q = 3, 4 and first order for q > 5; in
d = 3 it is first order for q 4. The cases q = 5 in d = 2 and q = 3 in d = 3 are of the type (iii)
above and will be discussed in a moment.
3 One can make sense of the Potts model for non-integer values of q through the mapping onto the random cluster
model [12]. Although we will have mainly in mind integer values, most of our discussion can be done treating q as a
continuous parameter, and this will be understood in the following.
269
2.2. Field theory description

A continuous, field theoretical description at scales much larger than the lattice spacing is
possible at and around those points of the phase diagram where the correlation length diverges.
For q qc , the transition point at H = 0 (Cq in Fig. 1) corresponds in the scaling limit to a
(q)
fixed point of the renormalization group, i.e., to a conformal field theory with action ACFT . The
scaling limit of (1) around Cq is described by the action

(q)
d
A = ACFT d x (x) h d d x q (x),
(4)
(q)
where (x) is the leading Sq -invariant operator in ACFT , and q (x) is the leading Sq1 (q)
preserving magnetic operator. If X denotes the scaling dimension of an operator (x) at the
Sq -invariant fixed point and m a mass scale, the couplings and h behave dimensionally as
(q)
mdX ,
(q)
h mdX ,
(5)
and measure the deviation from critical temperature and the magnetic field, respectively.
For 2 < q qc the transition along the line joining Cq to Cq1 in Fig. 1 is continuous and the
(q)
(q1)
scaling action (4) with = 0, h < 0 describes a massless flow from ACFT to ACFT . The scaling
limit around the infrared fixed point Cq1 is described by the action

(q1)
AIR = ACFT d d x (x) + d d x (x) + ,
(6)
(q1)
(q1)
where mdX
is proportional to T Tc and mdX
is proportional to 1/H . All
the operators in the r.h.s. of (6) are Sq1 -invariant: is relevant, while is the most relevant of
the infinitely many irrelevant operators (dots) which specify the massless flow at = 0.
In d = 3 the condition 2 < q qc is not satisfied for integer values of q and the transition at
H < 0 is first order for q > 3. For q = 3 the correlation length is finite at Cq and infinite at Cq1 .
Since the latter is an Ising critical point for which is the only symmetry-preserving relevant
operator, C2 must be an infrared fixed point whose scaling region is described by the action (6).
The ultraviolet endpoint of the massless flow at = 0 must be a fixed point P located on the
transition line (Fig. 2(a)). The nature of the transition then requires that P is an Ising tricritical
point.4
In d = 2 the field theoretical description can rely on a number of exact results. Baxter solved
the zero-field q-state Potts model at T = Tc on the square lattice and found that qc = 4 [7]. The
scaling limit of the critical point up to qc was later identified [13] to correspond to the conformal
field theory with central charge [14]
6
,
t (t + 1)
where the parameter t is related to q by the formula
c(q) = 1
q = 2 sin
(t 1)
.
2(t + 1)
(7)
(8)
4 Since Ising tricriticality is described by a 6 LandauGinzburg potential for which d = 3 is the upper critical dimension, P is a Gaussian fixed point.
270
(a)
(b)
Fig. 2. Phase diagrams of the model (1) for d = q = 3 (a), and for d = 2, q = 5 (b). The correlation length is infinite
along P C2 and at C4 ; q phases coexist at generic points along the dashed lines, q + 1 at the points C3 and C5 .
The scaling dimensions of the leading thermal and magnetic operators coincide with those of the
operators 2,1 and (t1)/2,(t+1)/2 in the conformal theory, and read [13,15]

1
(t 1)(t + 3)
3
X(q) =
(9)
1+
,
X(q) =
.
2
t
8t (t + 1)
The action (4) with d = 2 describes the scaling region around Cq for q 4 and is known to
correspond to an integrable quantum field theory for h = 0 [9] (see Section 3). For = 0, h < 0
and 2 < q 4 it describes a massless flow between an ultraviolet fixed point with central charge
c(q) and an infrared fixed point with central charge c(q 1). Around this latter fixed point we can
use the action (6) with the irrelevant operator identified with the operator 3,1 of the conformal
classification. Its scaling dimension at an Sq -invariant critical point is

2
(q)
.
X = 2 1 +
(10)
t
A field theory description around Cq1 is still possible at q = 5. Since X(4) = 2, is in this case
a marginal operator and the scaling region around C4 is described by the action

(4)
ACFT d 2 x (x) + d 2 x (x).
(11)
Since the two-dimensional 4-state Potts model does not admit a tricritical point5 [16,17], acts
as a marginally relevant perturbation and the transition is first order (Fig. 2(b)). The action (11)
is integrable also for = 0 [18,19] and can be used to describe exactly the transition close to C4 .
3. Exact scaling solution in zero field
In the two-dimensional case, to which we restrict our attention from now on, the action (4) is
integrable for h = 0 and the solution can be found in the form of an exact, elastic and factorized
S-matrix for the relativistic particles of the associated (1 + 1)-dimensional theory [20]. The
scattering theories above and below Tc must describe two different physical situations and, at the
same time, must reflect the existence of a duality transformation [5,6] relating the ordered and
disordered phases.
5 A phase diagram analogous to that of Fig. 2(a), with P corresponding to a (q 1)-Potts tricritical point, should be
obtained for 4 < q < 5. P tends to Cq as q 4 and to Cq1 as q 5.
271
3.1. Ordered phase

The S-matrix in the case of spontaneously broken symmetry ( < 0) was determined6 in [9].
The q ferromagnetic ground states correspond in the field theory to degenerate vacua labelled
by an index = 1, 2, . . . , q. The elementary excitations are then provided by kinks7 K ( )
interpolating between the vacua and ( = ). The space of asymptotic states consists of
multi-kink configurations of the type K0 1 (1 )K1 2 (2 ) . . . Kn1 n (n ) (i = i+1 ) interpolating between the vacua 0 and n . As a consequence of Sq -invariance, all the n-kink states fall
into two topological sectors: the neutral sector, corresponding to 0 = n , and the charged sector,
corresponding to 0 = n .
Integrability implies that the scattering processes are completely elastic and factorised into the
product of two-kink interactions. An outgoing two-kink state can only differ from the ingoing one
by the vacuum state between the kinks. Hence, the two-kink scattering can formally be described
through the FaddeevZamolodchikov commutation relation

K (1 )K (2 ) =
(12)
S (12 )K (2 )K (1 ),
=,
where 12 1 2 , and S (12 ) denote the two-body scattering amplitudes (Fig. 3(a)). Sq
invariance reduces to four the number of independent amplitudes, two for the charged and two
for the neutral topological sector

K (1 )K (2 ) = S0 (12 )
K (2 )K (1 ) + S1 (12 )K (2 )K (1 ), = ,
=
K (1 )K (2 ) = S2 (12 )
K (2 )K (1 ) + S3 (12 )K (2 )K (1 ).
(13)
Using the commutation relation (12) twice one obtains the unitarity constraint

S ( )S
( ) = ,
(14)
=,
which amounts to the set of equations

(q 3)S0 ( )S0 ( ) + S1 ( )S1 ( ) = 1,
(15)
(q 4)S0 ( )S0 ( ) + S0 ( )S1 ( ) + S1 ( )S0 ( ) = 0,
(16)
(q 2)S2 ( )S2 ( ) + S3 ( )S3 ( ) = 1,
(17)
(q 3)S2 ( )S2 ( ) + S3 ( )S2 ( ) + S2 ( )S3 ( ) = 0.
(18)
Crossing symmetry provides the relations

S0 ( ) = S0 (i ),
(19)
S1 ( ) = S2 (i ),
(20)
6 The first exact S-matrix for -perturbed conformal field theories was determined in [21]. The relation between
2,1
this solution, which does not exploit Sq invariance and relies on a different particle basis, and the one of [9] is explained
in [22].
7 We parameterise on-shell momenta as p = (m cosh , m sinh ), m being the mass of the kink.
272
S3 ( ) = S3 (i ).
(21)
Using these constraints together with the YangBaxter and bootstrap equations (that we do not
reproduce here) the following expressions for the four elementary amplitudes were determined
in Ref. [9]

sinh sinh ( i)
S0 ( ) =
(22)
2i
i
i
sinh ( 3 ) sinh ( 3 )

sin 2
3 sinh ( i)
,
S1 ( ) =
(23)
2i
i
sin 3 sinh ( 3 )

sin 2
3 sinh
,
S2 ( ) =
(24)
i
i
sin 3 sinh ( 3 )

sin
,
S3 ( ) =
(25)
i
sin 3
where is related to q as
q = 2 sin
,
3
and

sinh ( + i 3 ) A()
,
=
e
i
sinh ( i)

dx sinh x2 (1 1 ) sinh x2 ( 1 53 )
x
sinh .
A( ) =
x
x
sinh 2
cosh x2
i
(26)
(27)
(28)
The above solution is well defined for real values of , and one sees that (26) implies 0 q 4.
The pure scaling Potts model in this range of q is identified with the values of in the range
going from8 0 to 3/2.
The function (/i) is free of poles in the physical strip Im (0, ) for q < 3 (i.e.,
< 1). Hence, in this range of q the only poles of the scattering amplitudes in the physical strip
are those located at = 2i/3 and = i/3 and correspond to the appearance of the elementary
kink itself as a bound state in the direct and crossed channel, respectively.
For q > 3 ( > 1) a direct channel (positive residue) pole located at = 2i,

1
1
,
=
(29)
2
enters the physical strip in the amplitudes S2 ( ) and S3 ( ). Such a pole must be accordingly
associated to a (topologically neutral) kinkantikink bound state B with mass
mB = 2m cos .
(30)
The amplitudes S1 ( ) and S3 ( ) exhibit the corresponding crossed channel (negative residue)
pole at = i 2i. The amplitudes SKB ( ) and SBB ( ) describing the kink-bound state scattering and the bound state self-interaction are determined by the bootstrap equations
SKB ( ) = (q 2)S2 ( i)S1 ( + i) + S3 ( i)S3 ( + i),
8 The range 3/2 < < 3 corresponds to the thermal perturbation of the tricritical q-state Potts model [9].
(a)
273
(b)
Fig. 3. Pictorial representation of the kink scattering amplitudes S (a), and of the high-temperature amplitudes
pj,j
Sk,pk (b).
SBB ( ) = SBK ( i)SBK ( + i),
(31)
and read
SBK ( ) = t1/ ( )t2/3/ ( ),
(32)
SBB ( ) = t2/3 ( )t12/ ( )t2/32/ ( ),
(33)
in terms of the functions

ta ( ) =
tanh 12 ( + ia)
tanh 12 ( ia)
(34)
The poles located at = i( ) in SBK and at = 2/3 in SBB are bound state poles corresponding to K and B, respectively.
It has been shown [23] that the remaining poles in the amplitudes SKB and SBB are associated
to multi-scattering processes rather than to new particles. Hence, the elementary kinks and their
neutral bound state B are the only particles in the spectrum of the field theory describing the
scaling zero-field Potts model below the critical temperature.
3.2. Disordered phase
We now show how the amplitudes (22)(25) determine also the scattering theory in the disordered phase.
Above the critical temperature there is a unique ground state and the excitations of the scaling
limit must be ordinary particles rather than kinks. Observing that the group Zq of cyclic permutations is a subgroup of Sq , we can take these particles to have a well defined Zq charge. The simplest possibility is then to conjecture a basis of elementary excitations Ak ( ), k = 1, . . . , q 1,
each carrying k units of Zq charge. A multi-particle state Ak1 (1 ) . . . Akn (n ) will carry a charge
k1 + + kn (mod q). The integrable scattering theory is characterized by the two-particle amplitudes defined through the FaddeevZamolodchikov algebra (Fig. 3(b))
pj,j
Ak (1 )Apk (2 ) =
(35)
Sk,pk (12 )Apj (2 )Aj (1 ),
j =p
where all the particle indices are taken mod q. The neutral channel corresponds to p = q, while
p = 1, . . . , q 1 yields charged channels. Full Sq invariance is then recovered requiring that the
interaction does not distinguish between the charged channels; within the neutral channel and
the charged channel, the need to have well defined crossing properties forces to distinguish the
274
case j = k from the case j = k. This leaves us with a total number of four different amplitudes
corresponding to

Ak (1 )Apk (2 ) = S0 (12 )
Aj (2 )Apj (1 ) + S1 (12 )Ak (2 )Apk (1 ), p = q,
j =k,p
Ak (1 )Aqk (2 ) = S2 (12 )
Aj (2 )Aqj (1 ) + S3 (12 )Ak (2 )Aqk (1 ).
(36)
j =k
These identifications require that all the particles Ak have the same mass m.
The unitarity equations

pl,l
pj,j
j pj
Sk,pk ( )Sl,pl ( ) = k pk
(37)
l =p
are obtained iterating (35) and take a form identical to (15)(18) with the substitution Si Si .
The crossing relations coincide with (19)(21) under the same substitution.
This correspondence between the scattering theories above and below the critical temperature
extends to the factorization and bootstrap equations and simply expresses the fact that the two
phases are related by duality. The elementary excitations of the disordered phase are particles
Ak , k = 1, . . . , q 1, with the same mass of the kinks of the low-temperature phase and whose
scattering is expressed in terms of the same amplitudes which specify the kink S-matrix, i.e.,
m
= m,
Si ( ) = Si ( ),
i = 0, 1, 2, 3.
(38)
For 3 < q 4, the high-temperature theory contains a neutral bound state B with the mass (30)
and whose scattering is specified by amplitudes SBAk and SBB coinciding with (32) and (33).
Duality allows to compute correlation functions above and below Tc within the form factor
approach which relies on the knowledge of the S-matrix. This was done in [11] using the kink
S-matrix. It can be checked that the same results are obtained through the high-temperature
scattering theory.
4. Particle spectrum in weak magnetic field
The action (4) describes the renormalization group trajectories flowing out of the fixed point
located at the origin of the h plane. Such trajectories can be labelled by the dimensionless
parameters
=
(39)
,
(dX
)/(dX )
(h)
where the upper and lower signs are used for h > 0 and h < 0, respectively, in such a way that
+ parameterizes the trajectories in the upper half-plane and those in the lower half-plane;
the two trajectories at h = 0 correspond to = + and = (see Fig. 4).
In this section we discuss the evolution of the mass spectrum in d = 2 starting from the case of
small magnetic field for which perturbation theory around integrable quantum field theories [10]
can be used.
4.1. Weak field above critical temperature
For small h at > 0 (i.e., for +) the corrections to the mass spectrum are determined
by the matrix elements of the magnetic operator q (x) on the asymptotic states of the unperturbed
275
Fig. 4. Some renormalization group trajectories associated to the action (4).
(a)
(b)
Fig. 5. First order mass corrections in weak magnetic field above critical temperature (a), and below (b).
( = +) theory. At leading order, the correction to the mass matrix is [10] (Fig. 5(a))

2

m j,k 2h Aj (0) q (0) Ak (0) , j, k = 1, . . . , q 1
(40)
(here and below the matrix element on the particles Ak are intended at h = 0).
Let us introduce the operators
k (x) =
q
qk (x),
k = 1, . . . , q 1,
(41)
=1
where n exp(2i/n). The generator q of cyclic permutations acts as

q (x) = +1(mod q) (x),
q k (x) = qk k (x),
(42)
showing that k carries k units of Zq charge and can be taken as interpolating operator of particle
Ak at h = 0. Using
n
j k
= nk,n
(43)
j =1
and (3), one also has

1
k (x).
q
q1
q (x) =
k=1
(44)
276
Denoting

= Aj (0) (0) Ak (0) ,

Mj,k
(45)
and taking into account conservation of Zq charge as well as full Sq symmetry at h = 0, we have
l
= fq j,k+l(mod q)
Mj,k
(46)
and
fq
1 l
=
Mj,k = (1 j,k ),
q
q
q1
Mj,kq
(47)
l=1
for the matrix which determines the mass corrections (40). Diagonalization of this matrix gives
for the spectrum at + a particle
q1

1
Ak ,
A0 =
q 1 k=1
(48)
with square mass

m20 m2 2fq
q 2
h,
q
(49)
and a degenerate multiplet

1
A k = (Ak Aq1 ),
2
k = 1, . . . , q 2,
(50)
with square mass

m 2 m2 + 2
fq
h.
q
(51)
Comparison with (44) shows that A0 is interpolated by q and is a singlet of the Sq1 symmetry
surviving at h = 0; suitable linear combinations of the A k yield a multiplet in which each of the
q 2 particles carries a definite (non-zero) Zq1 charge.
For 3 < q 4 the theory also contains the additional Sq1 singlet B with a mass correction
with respect to (30) given by

m2B 2h B(0) q (0) B(0) .

(52)
This matrix element, as well as fq above, can be obtained from the form factor results of [11] (for
q = 4 see also [8]). Notice that the first order correction to the mass of A0 vanishes for q = 2;
the leading correction in this case is of order h2 and was computed in [24].
4.2. Weak field below critical temperature
A small magnetic field acting on the sites with color q affects the ground state degeneracy at
< 0. Denoting by |0 , = 1, . . . , q the ferromagnetic vacua, Sq symmetry gives at =
0 | (x)|0 =
v
(q , 1),
q 1
(53)
277
with v positive. At first order in h, the energy density difference between a vacuum |0 =q and
|0q is then

vq
E = E Eq h q q q =
(54)
h,
q 1
so that |0q is the unique true vacuum at h > 0, and the unique false vacuum at h < 0.
Since no finite-energy topological excitation can begin or end on a false vacuum, the space
of asymptotic states of the theory does not contain kinks for h > 0. For large and negative,
instead, the kinks K with , = q survive as the elementary excitations of the theory. The first
order correction to their mass is (Fig. 5(b))

q
(i),
m2K 2h K (0) q (0) K (0) conn = 2hF
(55)
where

q
F
(1 2 ) = 0 q (0) K (1 )K (2 ) ,
(56)
is the two-kink form factor at = [11]. In general this function has an annihilation pole
whose residue
vq
q
i Res=i F
(57)
( ) = q q =
(,q ,q )
q 1
vanishes precisely when both and differ from q, giving a finite mass correction9 (55) (which
does not depend on , = 1, . . . , q 1). The divergence of (55) when or equal q simply
reflects the decoupling of a kink interpolating between vacua which are no longer degenerate.
Although for h positive |0q is the only true vacuum and no single-kink state survives, nkink states beginning and ending on the true vacuum, i.e., Kq1 (1 )K1 2 (2 ) . . . Kn1 q (n ), do
not decouple. Here we consider all the intermediate vacua to be false, since insertion of a true
intermediate vacuum would simply amount to breaking the sequence into two states of the first
type. Then we have n q.
In the non-relativistic limit valid for small rapidities, the total energy of such a configuration
consist of the rest mass term nm, the kinetic term, a contribution coming from kink interaction
which decays exponentially with interkink distance, and the false vacuum contribution
n1

V (x1 , . . . , xn ) = E
(xi+1 xi ) = (xn x1 )E,
(58)
i=1
where x1 < x2 < < xn are the spatial positions of the kinks. The positive linear potential (58)
confines the n kinks into a finite spatial interval and prevents the observation of isolated kinks. In
this sense the kinks of the = theory play the role of quarks at h > 0. The asymptotic
particles are instead the topologically neutral bound states produced by the confinement of the
n-kink state. When + the confining potential is extremely shallow and the kinks are
very loosely bound; the average interkink distance is large and kink interaction is negligible in
first approximation. The n-kink bound states form an infinite tower of levels which are dense
above the value nm as + . It is natural to call mesons the n = 2 (kinkantikink) bound
9 In this case (55) is finite irrespectively of the sign of h. We should recall, however, that for h > 0 we are computing
a mass gap above a false vacuum whose energy separation from the true vacuum in a system of spatial size L is EL.
Hence, all the single-kink states decouple in the infinite system with a positive, however small, magnetic field.
278
(a)
(b)
Fig. 6. Spatial kink configurations which are confined into mesons (a) and baryons (b) by a positive magnetic field.
states (Fig. 6(a)), and baryons the bound states with n = 3 (Fig. 6(b)). Recalling the conditions
n q and q qc = 4, we have that the mesons can occur for q = 2, 3, 4, and the baryons for
q = 3, 4; tetraquark confined states are allowed for q = 4.
These particles organize themselves into multiplets of the residual Sq1 symmetry. The number of different n-kink sequences coincides with the possible ways of coloring the intermediate
vacua. The q 1 mesonic sequences can be combined into the states
(j )
k (0)
q1
k
q1
Kq ( )Kq ( ),
k = 0, 1, . . . , q 2
(59)
=1
with j = 1, 2, . . . labelling in order of increasing energy the levels originated by the confinement
of the kink-antikink superposition. Since
(j )
(j )
k
q1 k = q1
k ,
(60)
the mesons (59) are eigenstates under cyclic permutations of the first q 1 colors with Zq1
charge k. In the non-relativistic limit valid for the lowest levels in weak field, we can think of the
kink and antikink inside a meson as experiencing elastic reflection on the walls of the confining
potential and elastic scattering among themselves. As the effect of the latter, the intermediate
vacuum can either remain unchanged with a probability amplitude 3 (2 ), or switch to a different color with a probability amplitude 2 (2 ) which, by Sq1 -invariance, does not depend on
the new color.10 The superpositions in the right-hand side of (59) are eigenstates of the S-matrix
with scattering amplitudes

3 (2 ) + (q 1)k,0 1 2 (2 ).
(61)
Hence, the quark interaction is different for the neutral mesons (k = 0) and for the charged
(j )
(j )
(j )
mesons (k = 0). For fixed j this will lead to a siglet 0 and a multiplet 1 , . . . , q2 with
energies which differ very slightly in weak field.
Only a finite number of these confined states are stable. For q = 2, 3 all those states with
energy larger than twice the mass of the lightest mesons (i.e., larger than a value close to 4m
for weak field) lie in the continuum and can decay. This means that for + sufficiently large and
negative we certainly have stable mesons in the energy interval (2m, 4m) for q = 2, 3, and stable
baryons in the energy interval (3m, 4m) for q = 3. At q = 4 the decay thresholds
are lowered by
11 mass m = 3m. Hence the stability

the presence of the neutral particle
B
with
unperturbed
B
intervals in weak
(2m, (2 + 3)m) for the charged
field are (2m, 2 3m) for the neutral mesons,
mesons, (3m, 2 3m) for the neutral baryons and (3m, (2 + 3)m) for the charged baryons; as
for the tetraquark states, they all lie in the continuum and are expected to be unstable.
10 At = the scattering amplitudes and coincide with (24) and (25), respectively. Corrections in weak
2
3
field are determined by the matrix elements of the magnetic operator q [10].
11 The first order correction to m2 coincides with (52).
B
279
The number of 3-kink sequences giving rise to baryons is (q 1)(q 2). For q = 3 we then
have the two series of baryons
(j )
p ( ) K31 (1 )K12 (2 )K23 (3 ) K32 (1 )K21 (2 )K13 (3 ),
(62)
with even or odd parity with respect the residual Z2 symmetry which interchanges the colors 1
and 2. For q = 4 the residual S3 symmetry can be seen as the product of the group Z3 of
cyclic permutations times the topological charge conjugation CT which transforms the kink K ,
, = 4, into its antikink K . Then the six series of baryonic states
(j )
pk, ( )
3

3k K4, (1 )K,+1(mod 3) (2 )K+1(mod 3),4 (3 )
=1

K4,+1(mod 3) (1 )K+1(mod 3), (2 )K,4 (3 ) ,
k = 0, 1, 2
(63)
are S3 -eigenstates:
(j )
(j )
3 pk, = 3k pk, ,
(j )
(j )
CT pk, = pk, .
(64)
For a given j they give rise to two singlets (k = 0) and two doublets (k = 0) with opposite
CT -parity.
As already said, a small negative magnetic field confines only the kinks K with or
equal q. Since now the false vacuum is unique, the only confined states are mesons of type
Kq Kq . For q > 2, however, these can decay into two asymptotic kinks through expansion of
bubbles of true vacuum into the false vacuum.
5. Evolution of the spectrum with temperature and magnetic field
5.1. Positive magnetic field
We discussed in the previous section how the particle spectrum of the scaling two-dimensional
q-state Potts model changes when a small magnetic field acting on the sites with color q is
switched on. For positive field, below critical temperature we have a complete confinement of
kinks and the generation of a dense spectrum of mesons and (for q > 2) baryons; the number of
such particles which are stable tends to infinity as + . Above critical temperature, on the
other hand, the effect of the magnetic field is much less dramatic, simply amounting to a partial
removal of the degeneracy of the mass spectrum of the h = 0 particles.
It must be possible to interpolate continuously between these two limiting cases following the
evolution of the spectrum as + grows from to + on the plane of Fig. 4. The simplest scenario is that, as + increases, more and more mesons and baryons cross the decay thresholds and
become unstable. This process of depletion of the spectrum of stable excitations would continue
until the only stable particles surviving as + + are those of the high-temperature theory
in zero field.
We expect that, during the evolution of the spectrum as a function of the parameters , energy levels corresponding to states originating from the confinement of a same number of kinks
do not cross for h = 0. Indeed, the additional degeneracy at a crossing point of this nature should
normally be related to a symmetry enhancement. In the field theory (4) the only symmetry enhancement (from Sq1 to Sq ) occurs at h = 0. If we add to this that the baryons should decay
more easily than the lightest mesons, it seems natural to expect that the particles surviving in the
280
Fig. 7. Seven-kink matrix element responsible for the decay of baryons above threshold into two mesons in positive
magnetic field.
limit + + should be identified with the lightest among the mesons produced by kink con(1)
(1)
finement at + very large and negative. In particular, the lightest meson multiplet 1 , . . . , q2
should evolve for increasing + into the multiplet (50). As for the particle which evolves into the
(1)
singlet (48), it should be identified with 0 for q = 2, 3. At q = 4 the zero-field spectrum also
includes the neutral particle B with mass 3m, which is the lightest particle for + . If
one assumes that it does not cross mesonic levels, then it should evolve into the neutral parti
cle (48) for + +; then the meson 0(1) could evolve into the particle with mass 3m in the
same limit.
According to this scenario, all the mesons (59) with j > 1 and all the baryons must have
become unstable by the time + approaches +. For each of this particles there should exist
a finite critical value of + for which they reach the lowest decay threshold compatible with
their charge, and above which they become unstable. Such critical value is expected to decrease
as the mass of the particle increases, so that the critical trajectories accumulate in the limit
+ . Notice that, due to the non-locality of the magnetic operator with respect to the
kinks, the magnetic term of the action has infinitely many matrix elements on kinks which are
non-zero in zero-field. In particular, the vertex responsible for the decay of the baryons which
reach the two-meson threshold is shown in (Fig. 7).
Also in view of considerations to be made below about the spectrum evolution at h < 0, we
expect the quantity fq determining the first order mass corrections (49) and (51) to be positive.12
Together with the previous identifications, this leads to the expectation that for any h > 0 the
different quark interactions inside the neutral and the charged mesons (see (61)) induce a mass
(1)
(1)
(1)
splitting between 1 , . . . , q2 and 0 which is positive for q = 3. If the previous speculations
about q = 4 should turn out to be correct, the mass splitting would be negative in this case.
When specialized to q = 2 this scenario coincides with that originally proposed for the scaling
Ising model by McCoy and Wu [3]. In this case the pattern is simplified by the absence of baryons
and charged mesons, as well as by the absence of interaction at h = 0. Moreover, integrability
at = 0 [25] also allows analytic investigation for strong magnetic field. Several studies, both
analytic and numerical, have confirmed the McCoyWu scenario and provide us with a detailed
description of the mass spectrum of Ising field theory in the full h plane [3,10,2431] (see
also [32] for an introductory review of Ising field theory).
12 This quantity can be exactly computed within the integrable field theory at h = 0 (see [11]).
281
5.2. Negative magnetic field

We now discuss the evolution of the spectrum as a function of for q = 3, 4 (for q = 2
the spectrum does not depend on the sign of the magnetic field). We saw in the previous section
that the elementary excitations for are the kinks K interpolating between the q 1
residual vacua. For +, on the other hand, due to the negative sign of h in (49) and (51),
the lightest excitations are the charged particles that in the previous subsection we identified with
(1)
(1)
the mesons 1 , . . . , q2 . Passing from these topologically neutral particles to the topologically
charged kinks requires a phase transition at an intermediate value of . We already know that,
for q = 3, 4, a second order phase transition takes place at = 0.13 The mass of the kinks
decreases as goes from to zero, and vanishes at = 0. Similarly, the mass of the mesons
(1)
(1)
1 , . . . , q2 decreases as goes from + to zero, and vanishes at = 0. Deconfined kinks
exist only for [, 0); for (0, +] and for all positive values of h the spectrum can
be seen as consisting of kink bound states. The trajectory = 0 is a massless flow between
the Sq -invariant ultraviolet fixed point at h = 0 and the Sq1 -invariant infrared fixed point at
h = .
(1)
The ratio between the mass of the neutral meson 0 and that of the charged mesons
(1)
(1)
1 , . . . , q2 , which is 1 at = +, increases as decreases until it reaches the value 2 for
(1)
(1) (1)
some positive value of . Beyond this point the opening of the decay channel 0 1 q2
makes the neutral meson unstable. The second neutral particle present at q = 4 becomes unstable in the same way. We then expect that the charged mesons are the only stable particles
surviving for sufficiently small and positive. Similarly for the kinks at sufficiently small
and negative.
The conjectured evolution of the mass spectrum with the parameters + and is illustrated
in Fig. 8 for q = 3.
6. Conclusion
In this paper we considered the field theory describing the scaling limit of the two-dimensional
q-state Potts model in a magnetic field acting on one of the q colors. This field breaks the Sq
symmetry of color permutations down to Sq1 and, for 2 < q qc = 4 allows for an extended
region in the plane of temperature and magnetic field in which the quarks (kinks interpolating
between
degenerate vacua) are deconfined. If the analysis is extended to a more general magnetic
term q1
1 h , a renormalization group trajectory corresponding to generic values of the components h will possess no internal symmetry and will contain only particles made of confined
quarks with unequal masses. Of course, suitable relations among the magnetic parameters identify regions in parameter space with a residual Sq1 or (for q = 4) Sq2 symmetry, which in turn
contain phases with deconfined quarks.
We saw how form factor perturbation theory allows to compute mass corrections in weak field
above critical temperature, and below critical temperature for the quarks in the deconfined phase.
As for the mass spectrum of the particles originating from confinement, the BetheSalpeter
approach proved to be remarkably effective for the Ising mesons [26,29], but needs to be gener13 An analogous transition from ordinary particles to kinks, involving the spontaneous breaking of the symmetry S
q
rather than Sq1 , takes place when going through = 0 at h = 0.
282
Fig. 8. Conjectured qualitative evolution of the mass spectrum for q = 3. The left-hand half corresponds to the evolution
in + , the right-hand half to the evolution in . The values = +, as well as the values = , describe the
same renormalization group trajectory. Only three lightest mesons and the lightest neutral baryon are shown for positive
magnetic field. The dashed line is the lowest decay threshold (twice the mass of the lightest particle). Particles whose
mass reaches the threshold become unstable. Deconfined kinks exist for negative only.
alized to the case of quarks which interact already in absence of field in order to deal with q = 2.
Hopefully, it will become possible to study also the baryonic spectrum along similar lines.
Numerical methods will be eventually needed for a quantitative study of the mass spectrum in strong field. Particularly promising in this respect seems the truncated conformal space
approach [33], which amounts to the numerical diagonalization of the Hamiltonian on a finitedimensional subspace of the conformal basis of states of the ultraviolet fixed point. This approach
has been successfully used for the Ising case14 [10,26,29,34] and can be used also for q = 2.
Acknowledgements
The work of G.D. is partially supported by the ESF grant INSTANS and by the MUR project
Quantum field theory and statistical mechanics in low dimensions.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
S. Coleman, R. Jackiw, L. Susskind, Ann. Phys. 93 (1975) 267.

S. Coleman, Ann. Phys. 101 (1976) 239.
B.M. McCoy, T.T. Wu, Phys. Rev. D 18 (1978) 1259.
G. Delfino, G. Mussardo, Nucl. Phys. B 516 (1998) 675.
R.B. Potts, Proc. Cambridge Philos. Soc. 48 (1952) 106.
F.Y. Wu, Rev. Mod. Phys. 54 (1982) 235.
R.J. Baxter, Exactly Solved Models of Statistical Mechanics, Academic Press, London, 1982.
G. Delfino, P. Grinza, Nucl. Phys. B 682 (2004) 521.
14 In [26,29] the free nature of the zero-field Ising model is exploited to diagonalize the Hamiltonian on a truncated
basis of massive fermionic states.
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
L. Chim, A.B. Zamolodchikov, Int. J. Mod. Phys. A 7 (1992) 5317.

G. Delfino, G. Mussardo, P. Simonetti, Nucl. Phys. B 473 (1996) 469.
G. Delfino, J. Cardy, Nucl. Phys. B 519 (1998) 551.
P.W. Kasteleyn, E.M. Fortuin, J. Phys. Soc. Jpn. Suppl. 26 (1969) 11;
P.W. Kasteleyn, E.M. Fortuin, Physica 57 (1972) 536.
Vl.S. Dotsenko, V.A. Fateev, Nucl. Phys. B 240 (1984) 312.
A.A. Belavin, A.M. Polyakov, A.B. Zamolodchikov, Nucl. Phys. B 241 (1984) 333.
B. Nienhuis, J. Stat. Phys. 34 (1984) 781.
B. Nienhuis, A. Berker, E. Riedel, M. Shick, Phys. Rev. Lett. 43 (1979) 737.
J. Cardy, M. Nauenberg, D. Scalapino, Phys. Rev. B 22 (1980) 2560.
G. Delfino, Nucl. Phys. B 554 (1999) 537.
G. Delfino, J. Cardy, Phys. Lett. B 483 (2000) 303.
A.B. Zamolodchikov, Al.B. Zamolodchikov, Ann. Phys. 120 (1979) 253.
F.A. Smirnov, Int. J. Mod. Phys. A 6 (1991) 1407.
P. Fendley, N. Read, hep-th/0207176.
P. Dorey, A. Pocklington, R. Tateo, Nucl. Phys. B 661 (2003) 425.
P. Fonseca, A.B. Zamolodchikov, hep-th/0309228.
A.B. Zamolodchikov, Adv. Stud. Pure Math. 19 (1989) 641;
A.B. Zamolodchikov, Int. J. Mod. Phys. A 3 (1988) 743.
P. Fonseca, A.B. Zamolodchikov, J. Stat. Phys. 110 (2003) 527.
G. Delfino, P. Grinza, G. Mussardo, Nucl. Phys. B 737 (2006) 291.
S.B. Rutkevich, Phys. Rev. Lett. 95 (2005) 250601.
P. Fonseca, A.B. Zamolodchikov, hep-th/0612304.
P. Grinza, A. Rago, Nucl. Phys. B 651 (2003) 387.
M. Caselle, P. Grinza, A. Rago, hep-lat/0408044.
G. Delfino, J. Phys. A 37 (2004) R45.
V.P. Yurov, Al.B. Zamolodchikov, Int. J. Mod. Phys. A 6 (1991) 4557.
B. Pozgay, G. Takacs, hep-th/0604022.
283

Gauge theory of the star product

A. Pinzul a , A. Stern b,
a Instituto de Fsica, Universidade de So Paulo, C.P. 66318, Sao Paulo, SP 05315-970, Brazil
b Department of Physics, University of Alabama, Tuscaloosa, AL 35487, USA

Abstract
The choice of a star product realization for non-commutative field theory can be regarded as a gauge
choice in the space of all equivalent star products. With the goal of having a gauge invariant treatment, we
develop tools, such as integration measures and covariant derivatives on this space. The covariant derivative
can be expressed in terms of connections in the usual way giving rise to new degrees of freedom for noncommutative theories.
PACS: 11.10.Nx; 02.40.Gh
Keywords: Star product; Non-commutative gauge theory
1. Introduction
Deformation quantization [1] replaces the commutative algebra of functions on a Poisson
manifold with a non-commutative algebra, where multiplication for the latter is given by some associative star product. Kontsevich [2] showed that this program can be carried out for any smooth
Poisson manifold. An explicit construction for the star product (the Kontsevich star product) was
given in [2] which was completely determined by the Poisson bi-vector . The Kontsevich star
product belongs to a very large equivalence class {} of star products. The different star products in {} are related by gauge transformations, where the gauge group G is generated by all
differential operators. G also includes transformations, such as standard non-commutative U (1)
gauge transformations, which leave the star product invariant [3,4].
E-mail address: astern@bama.ua.edu (A. Stern).

doi:10.1016/j.nuclphysb.2007.09.012
A. Pinzul, A. Stern / Nuclear Physics B 791 [FS] (2008) 284297
285
Other ingredients, in addition to the non-commutative algebra, are needed in order to write
down field theories on the non-commutative spaces. Among them are the trace and derivative.
Concerning the former, theorems have been given which show the existence of the trace for the
deformation of a symplectic manifold [5] and, more generally, for any regular Poisson manifold for which a Poisson trace exists [6,7]. It was shown, for example, that the usual integral
with the commutative measure satisfies the necessary conditions for a trace when the topology
of the manifold is R2d and one restricts to the Kontsevich star. In another example, corrections were computed in [8] to the commutative measure for a star product constructed from
deformed coherent states [9]. Concerning the derivative, much is known for the case of constant non-commutativity, where the Kontsevich star reduces to the familiar GroenewoldMoyal
star [10,11]. For that case, the standard partial derivative can be realized as an inner derivative on
the algebra. This is not true for non-constant non-commutativity where the star product is position dependent. Derivatives have nevertheless been defined in the general case, after specializing
to the Kontsevich star product [12].
As most previous treatments of non-commutative field theory have relied heavily on one particular star product in {} , namely the Kontsevich star product, it is of interest to search for gauge
invariant approaches, where here the gauge group is G . This is addressed in the current article.
One approach is to simply map the non-commutative field theory written with the Kontsevich
star to a non-commutative field theory associated with an arbitrary star product in the equivalence class, resulting in no new physical degrees of freedom. Alternatively, one can introduce
the notion of covariance with respect to G , where functions {f, g, . . .} and their star products
transform in the same manner, and are hence covariant. In addition, one can define the trace of
the functions to be gauge invariant. Like in YangMills theories, one can then also introduce a
covariant derivative, now associated with gauge transformations between different star products
in {} , where the covariant derivative of functions {f, g, . . .} transforms in the same manner as
the functions {f, g, . . .}. In such an approach, which is what we follow here, one thereby obtains
new degrees of freedom associated with the connections.1 As G is an infinite dimensional extension of the non-commutative U (1) gauge group, there are in principle an infinite number of such
degrees of freedom, which contains the standard non-commutative U (1) gauge degrees of freedom. It then becomes possible to consider an infinite dimensional extension of non-commutative
U (1) gauge theory, with the dynamics for the gauge fields and matter fields written on the entire
equivalence class {} .
The plan of the paper is as follows. In Section 2 we introduce the integration measure,
covariant derivative, connection and curvature for the special case where {} contains the
GroenewoldMoyal star product, while the generalization to an arbitrary equivalence class is
given in Section 3. Star products in {} can be expanded in the non-commutative parameter, which we denote by h , and each order can be expressed in terms of an infinite number of
bi-differential operators. Furthermore, G is generated by an infinite number of differential operators at each order in h . For practical purposes, we examine a restricted gauge group in Section 4
which is generated by a finite number of differential operators at each order in h . We can then
write down explicit formulae for components of the connection, curvature and field equations.
We summarize the results and indicate possible future developments in Section 5.
1 Only after restricting the connection to a certain pure gauge, will the covariant derivative be gauge equivalent to the
derivative [12] written for the Kontsevich star.
286
2. Gauging the GroenewoldMoyal star product

We first review well-known facts about the GroenewoldMoyal star product. Here the Poisson
bi-vector on R2d coordinatized by x , = 1, 2, . . . , 2d, is
,
(2.1)
where = are constants on R2d . and are left and right derivatives = x ,
respectively. Constant non-commutativity results after deformation quantization. Denote by A
the non-commutative algebra of functions f0 , g0 , . . . on R2d with multiplication given by the
GroenewoldMoyal star product [10,11]. is the bi-differential operator

i h
= exp
(2.2)
.
2
Then at lowest order in h , the star commutator of functions reduces to their Poisson bracket
3
[f0 , g0 ] f0 g0 g0 f0 = i h{f
(2.3)
0 , g0 } + O h ,
where {f0 , g0 } = f0 g0 . The derivative satisfies the usual Leibniz rule when acting
on the GroenewoldMoyal star product of two functions. Using the standard measure d 2d x
on R2d , the integral serves as a trace for A . Moreover, the integral of the GroenewoldMoyal
star product of two functions can be replaced with the integral of the pointwise product of the
two functions, provided these functions vanish sufficiently rapidly at infinity

2d
d x f0 g0 = d 2d x f0 g0 ,
(2.4)
from which the trace property easily follows,

d 2d x [f0 , g0 ] = 0.
(2.5)
The GroenewoldMoyal star product is an element of the equivalence class of star products {} . The equivalence class is generated from the set of all invertible operators T of the
form
T =1+
h k Tk ,
(2.6)
k=1
where Tk are arbitrary differential operators. Under the action of T , functions f0 , g0 , . . . are
mapped to
f = T (f0 ),
g = T (g0 ),
...,
(2.7)
while is mapped to another associative star product {} , such that [2]
f g = T (f0 g0 ).
The new star commutator has the same h 0 limit as in (2.3),

[f, g] f g g f = i h{f,
g} + O h 2 .
(2.8)
(2.9)
As before, the integral serves as the trace. However, the measure associated with the star product is, in general, no longer d 2d x. Call the transformed measure d . Invariance of the trace
implies

287

d f =
d 2d x f0 ,
(2.10)
for functions f0 vanishing sufficiently rapidly at infinity. From (2.6), d can differ from the flat
measure d 2d x at order h.
(An explicit expression for d for restricted gauge transformations
is given in Section
4.3.)
The analogue
of (2.4) does not hold for an arbitrary in the equiva

lence class; i.e., d f g = d f g. However, the trace property easily follows from (2.8)
and (2.10)

d [f, g] = d T [f0 , g0 ] = d 2d x [f0 , g0 ] = 0.
(2.11)
Subsequent transformations can be performed to map between any two star products in {} .
Say that is obtained from using invertible operator , which we assume to have a form
analogous to (2.6),
=1+
h k k ,
(2.12)
k=1
where k are arbitrary differential operators. Then functions, as well as star products of functions, transform covariantly if:
f f = (f ),
f g f g = (f g).
(2.13)
In order that the trace be an invariant for {} , the measure should, in general, transform, d
d , such that

d f = d f ,
(2.14)
for all functions f that vanish sufficiently rapidly at infinity. d and d correspond to the flat
measure d 2d x at zeroth order in h , while from (2.12), they, in general, differ at order h . Since
the measure transforms nontrivially for general , gauge transformations cannot be considered
to be internal transformations beyond zeroth order in h .
There exists a subset of transformations (2.12) which leaves the star product and the measure
invariant, i.e., = and d = d . This is the case for inner automorphisms [3,4] = ,
parametrized by functions on R2d , where
(f ) = f 1
,
(2.15)
and 1
= 1. It is not surprising that general gauge transformations given by (2.13) are not
internal beyond zeroth order in h,
because the same is true for the subset of inner automorphisms,
even though the measure is invariant under the latter. For the case of the GroenewoldMoyal star,
the inner automorphisms are known to contain (global) translations [13]. Here we need that [ ]
has an inverse:
1 c x
i
e
f
e
i 1 c x
x e
= 1 + f + 12 f f +
where
translations of order h.
= x + h c ,
1
3! f
(2.16)
f f + . So for c of zeroth order in h , one gets
288
We denote the derivative associated with any {} by D[A] , and require that it is covariant under the gauge transformations (2.13),

D[A] f D[A ] f = D[A] f ,
(2.17)
or
D[A ] = D[A] .
(2.18)
Since is a differential operator of arbitrary order, so in general should be D[A] . As usual, let
us write the covariant derivative in terms of potentials A , which we expand according to
A =
h k Ak, ,
(2.19)
k=1
where Ak, are differential operators.2 If we require D[A] to reduce to the standard derivative
in the absence of the potentials, then we can write the usual expression
D[A] = + A ,
and the potentials A gauge transform as

A A = , 1 + A 1 .
(2.20)
(2.21)
Derivative-valued field strengths

F =
h k Fk, ,
(2.22)
k=1
where Fk, are arbitrary differential operators, can also be introduced

F = D[A] , D[A] = [ , A ] [ , A ] + [A , A ].
(2.23)
They satisfy the Bianchi identity

D[A] , F + D[A] , F + D[A] , F = 0,
(2.24)
and gauge transform according to

F F
= F 1 .
(2.25)
For the special case where A is the pure gauge A = T [ , T 1 ], D[A] satisfies the usual
Leibniz rule when acting on the star product of two functions, as we can use the inverse of (2.8)
to map back to the GroenewoldMoyal star product. This, however, is not true for arbitrary
connections A .
U (1) gauge theory on the non-commutative plane is contained in this system. Here we write
A = A acting on functions f [which gauge transform as inner automorphisms (2.15)] according to A (f ) = [a , f ] , where a are the non-commutative U (1) potentials. So the
2 Derivative-valued gauge fields have been considered previously in different contexts [3,4,15].
derivative-valued potentials acting on covariant functions can be written as3

h
A = 2ia sin .
2
289
(2.26)
Upon restricting to the GroenewoldMoyal star , (2.15) leads to the usual non-commutative
U (1) gauge transformations for a ,
1
a a = a 1
.
(2.27)
From (2.23), the field strength operator F = F acting on a function is F () = [f , ] ,

where f is the non-commutative U (1) field strength tensor f = a a + [a , a ] .
Thus

h

F = 2if sin ,
(2.28)
2
and upon restricting in (2.25) to (2.15), f gauge transform as inner automorphisms, f
= f
1
f
.
Field theory actions can now be written down which are invariants for the equivalence class
{} . If we assume the field on R2d transforms covariantly with respect to the above gauge
transformations, = (), then an invariant action is

1
d D[A] D[A] ,
S,A =
(2.29)
2
where is the flat metric. It is equivalent to the commutative action for a free massless scalar
field after restricting A to the pure gauge A = T [ , T 1 ] and making the field redefinition
from = T (0 ) to 0 . This is since then D[A] = T 0 , and we can re-express the action in
terms of the GroenewoldMoyal star product and use (2.4).
More exciting is the possibility of writing down a kinetic term for A . This would require a
trace over the operator-valued fields. A possible candidate is the Wodzicki residue [14]. Alternatively, one can adopt the usual YangMills form for the field equations:

D[A] , F = J .
(2.30)
The right-hand side represents a matter current source which can be expanded
J =
h k Jk, ,
(2.31)
k=1
where Jk, are differential operators and J gauge transforms as the field strengths F ,
J J = J 1 .
Moreover, it must be covariantly conserved,

D[A] , J = 0.
(2.32)
(2.33)
3 Alternatively, on can define the action of A on fields

fund in the fundamental representation. Such fields are not
covariant, in that they do not gauge transform according to (2.13) with = , but rather with the left action fund

fund
= fund . On such fields one has

i h
A = a exp
.
2
290
Since J takes values in an infinite dimensional vector space, (2.33) then corresponds to infinitely many conservation laws.
3. Generalization to the Kontsevich star product
Now we go to the case of a general Poisson bi-vector
,
(3.1)
where
=
, = 1, 2, . . . , 2d, are functions on an open subset
of
Corresponding star products can be given in terms of series expansions in the non-commutativity
parameter h , where the terms in the expansions are bi-differential operators Bn , n = 1, 2, 3, . . . ,

f g = fg +
M 2d
h n Bn (f, g).
R2d .
(3.2)
n=1
In the Kontsevich construction of the star product [2], which we denote using , B1 is proportional to the Poisson bi-vector field. Acting between functions f0 and g0 , is, up to second
order in h , given by4
i h
h 2
f0 g0 , f0 , g0
2
8
2

h
(, f0 g0 f0 , g0 ) + O h 3 ,
12
f0 g0 = f0 g0 +
(3.3)
where ,,..., = x x x . The Poisson bracket is again recovered at lowest order in h

from the star commutator
3
[f0 , g0 ] f0 g0 g0 f0 = i h{f
(3.4)
0 , g0 } + O h ,
where {f0 , g0 } = f0 g0 .
The integral with measure d0 = d 2d x 0 (x) can serve as a trace for a star product associated
with the Poisson bi-vector (3.1), provided that 0 (x) satisfies [57]

0 = 0.
(3.5)
From this relation the cyclicity property easily follows at first order in h,

d0 [f0 , g0 ] = O h 2 ,
(3.6)
provided functions f0 and g0 vanish sufficiently rapidly at infinity. For the special case of symplectic manifolds, 0 is proportional to |det |1/2 . More generally, it is known [6] that there
exists a star product 0 , that is gauge equivalent to , for which the cyclicity property is guaranteed to all orders in h using measure d0 . Call T the map from to 0 , and define a measure
d = d 2d x (x) associated with star product such that

d f0 = d0 T (f0 ),
(3.7)
4 Third order terms were computed in [16].
291
for all f0 that vanish sufficiently rapidly at infinity. Thus

d [f0 , g0 ] = 0,
(3.8)

and so d serves as a trace for the star product .
can be defined for the star product satisfying the standard Leibniz rule
Derivations X
X
(f0 g0 ) = X
f0 g0 + f0 X
g0 ,
(3.9)
provided the Lie derivative LX of vanishes,

LX = X X + X = 0,
(3.10)
for some vectors X =

This is the same condition that is needed for X to be a derivation
of the Poisson bracket; i.e., X{f0 , g0 } = {Xf0 , g0 } + {f0 , Xg0 }, and it also corresponds to a
was
vanishing SchoutenNijenhuis bracket of X with the Poisson bivector. An expansion for X
given in [12]. Up to second order in h,

X was found to be
X .

h 2
h 2
, X , , X ,, + O h 3 .
(3.11)
12
24
and is nonvanishing, with the zeroth order
The commutator of any two such derivatives X
Y
2
being the Lie bracket, [X , Y ] = LX Y + O(h ).
As with the GroenewoldMoyal star product, belongs to an equivalence class of star products which we denote as {} . The equivalence class is once again generated from the set of all
invertible differential operators T , mapping functions f0 , g0 , . . . to f, g, . . . in (2.7), and the star
product to , whose general form is given by (3.2), with
= X +
X
f g = T (f0 g0 ).
The measure d associated with the star product is related to d by

d f = d f0 ,
(3.12)
(3.13)
for functions f0 vanishing sufficiently rapidly at infinity. Subsequent transformations can again
be performed to map between any two star products in the equivalence class {} given by (2.13).
The corresponding measures are related by (2.14). For covariant derivatives we again need (2.17).
in the absence of the potentials, as is
Now say that the covariant derivative D[A]X reduces to X
the case for
D[A]X = X
+ AX .
(3.14)
The derivative-valued potentials AX gauge transform as

1
AX A X = X
+ AX 1 .
,
(3.15)
and , one can define field strengths

Given independent derivatives X
Y

FXY = D[A]X , D[A]Y = X , Y + X , AY Y , AX + [AX , AY ],
(3.16)
which gauge transform as in (2.25). For the special case where AX is the pure gauge AX =
, T 1 ], D[A] satisfies the usual Leibniz rule when acting on the star product of two
T [X
X
functions. Gauge invariant actions analogous to (2.29) can be written down after introducing a
metric over the space of vector fields {X, Y, . . .}.
292
4. h
expansion
The most general Tk and k in (2.6) and (2.12), respectively, contain an infinite number of
derivatives, and map to star products using (2.8), which then also contain an infinite number of
derivatives at each order in h beyond the zeroth order. Here for simplicity we shall restrict to
operators Tk and k which have a finite number of derivatives. More specifically, terms of order
n in h in the equivalence map will be, at most, of order 2n in derivatives. As a result of this the
star product , connections A and curvature F can be written in terms of a finite number of
derivatives at each order in h.
4.1. Gauge group

We parametrize the set of all differential operators {Tk = Tk(s) } with an infinite number of
symmetric tensors s = (s 1 , s 1 2 , s 1 2 3 , . . .) which are functions on R2d and are polynomials

(s)
in h starting with order zero. The resulting expression for T (s) = 1 +
k Tk should be
k=1 h
consistent with closure

T (s ) T (s) = T (s ) .
(4.1)
A possible solution is
(s)
Tk
(s)
(s)
= 2k1 + 2k ,
1
n(s) = s 1 2 n 1 ,2 ,...,n .
n
(4.2)
The identity corresponds to s = 0, T (0) = 1. From (4.1) one gets

2
(s )
s = s + s + hT
1 s + O h ,

(s )
s = s + s + h T1 s + s ( s ) + s ( s ) + O h 2 ,

1
s = s + s + s ( s ) + s ( s ) + s ( s ) + O(h ),
4
1
s = s + s + s ( s ) + O(h ),
24
,
(4.3)
where y () = y + all symmetric combinations. Denoting the inverse of T (s) by T (sinv ) =

1
T (s) , we get that

(s)
sinv = s + h T1 s + O h 2 ,

sinv = s + h T1(s) s + s ( s ) + s ( s ) + O h 2 ,
1
1
sinv = s + s ( s ) + s ( s ) + O(h ),
2
4
1
sinv = s + s ( s ) + O(h ),
24
.
(4.4)
293
4.2. Star product

Using the operator T (s) in the equivalence relation (3.12), the Kontsevich star product is
mapped to the star product given by (3.2), with the first two bi-differential operators B1 and B2
given by
B1 (f, g) = b f g,
B2 (f, g) = b1 f g

1
1
[]
b
+ 2b + b
b b
+ b
, f g
6
2

1
1
+ b b + 2b b[] b b f , g
6
2
+ b (,, f g + f ,, g)

1
+ 3b + b b , f , g,
2
are expressed in terms of and s according to

where the tensors b
b = s ,
i
b = + s ,
2

i (s)
i
i
+s
+s
s
s ,
b1 = T1 s s
2
2
2
1
b = s s ( s ) ,
4
1
b = s s ( s ) ,
48
(4.5)
(4.6)
and b[] = b b . We introduced the vector field b for the sake of completeness. Although
it does not appear directly in the star product, it does appear in the measure. (4.5) reduces to (3.3)
when s = 0, corresponding to the Kontsevich gauge. The star commutator is now

h 2
[]
[f, g]B = h b[] + h b1
f g +
b b b b , f , g
2

h 2 ( )
+
(4.7)
b[] + b[ ] b (, f g f , g) + O h 3 .
b
2
As in (3.4), the leading term is i h{f,
g}.
A subsequent gauge transformation can be performed using in (2.12) to map the star product
()
()
to . Now write k = Tk , with Tk given by (4.2) and denoting symmetric tensors =

( 1 , 1 2 , 1 2 3 , . . .). is again of the form (3.2), with b in (4.5) replaced by b
defined by
b b = b + ,
b b = b + ,
b1 b1
()
= b1 + T1 b b b ( ) ,
294

1 (
b) ( ) ,
4
1
b
(4.8)
b
=b
+
( ) .
48
A gauge invariant antisymmetric tensor can be constructed from the fields b . It is simply
the zeroth order term in an h expansion expression for re-expressed in terms of b .
We remark that the subset of inner automorphisms (2.15) are contained in the group generated
by (4.2). For example, in that case T1 = 2b 1
.
b b = b + +
4.3. Measure
The measure d associated with the above star product should satisfy (3.13) (for functions f0
vanishing sufficiently rapidly at infinity) and reduce to d0 = d 2d x 0 (x) in the commutative
limit. If we expand d in h ,

2
d = d 2d x 0 + h
(4.9)
1 + O h .
0 is gauge invariant. We can substitute into (2.14) to obtain the gauge transformations of the
higher order corrections to the measure. For example, if under the action of , 1 goes to 1
then

()
d 2d x 1 1 f = d 2d x 0 T1 f.
(4.10)
Upon integrating by parts and assuming functions f vanish sufficiently rapidly at infinity one
gets

1

1 = 1 + 0 , 0 .
2
(4.11)
4.4. Connection and curvature

For simplicity, here and in the following section, we work in the of equivalence class {}
containing the GroenewoldMoyal star product. Now write the differential operator-valued po(a )
tentials Ak, in (2.19) according to Ak, = Tk given in (4.2), where here a denote the tensors
1 1 2 1 2 3
a = (a , a , a
, . . .). From (2.21), using (4.3) and (4.4), we then deduce the following

gauge transformations for a
:

()
(a )
a a = (a ) + h T1 a T1 + O h 2 ,

(a )
a a = (a ) + h T1() a T1

1 ( )
(
)
(
)
+ (a ) + a + O h 2 ,
2
1
a a = (a ) + ( (a ) ) (a )( )
4
(

)
+ O(h ),
+

1
a a = (a ) + ( ) + O(h ).
(4.12)
48

(f )
Using (2.23) we can construct the field strengths Fk, = Tk

tensors f1 , f1 2 , f1 2 3 , . . . ,
295
where here f denotes the

(a )
= a + h T1 a ( ) + O h 2 ,
f

(a )
= a + h T1 a + a( a ) ( ) + O h 2 ,
f
1

= a + a( a ) ( ) + O(h ),
f
4

= a ( ) + O(h ).
f
(4.13)
They gauge transform according to

()
(f )
f
= f
+ h T1 f
T1 + O h 2 ,
f

()
(f )
)
(
f
= f
+ h T1 f
T1 + ( f
f
) + O h 2 ,
f

1

)
(
f
= f
+ ( f
f
) + O(h ),
f
4

= f + O(h ).
f f
(4.14)
4.5. Field equations
We can now substitute the above expansion for the field strength tensor into the sourceless
YangMills type equation (2.30)

(a )
f
+ h T1 f
+ O h 2 = 0,

(a )
(
)
+ O h 2 = 0,
+ h T1 f
+ a
f
f
1
(

)
+ a
f
+ O(h ) = 0,
f
4

+ O(h ) = 0,
f
(4.15)
where a is obtained from a assuming a flat metric.
From the action (2.29) we can also easily obtain the field equation for the scalar field in a
background A . For simplicity, choose to be the GroenewoldMoyal star. Variations of lead
to an exactly conserved current
k = 0,
(4.16)
where

2
h
k = D[A] + ha
(4.17)
D[A] a D[A] + O h .
2
One recovers the commutative result = 0 at zeroth order in h.
On the other hand, in
order to couple to the above gauge theory one should search currents J which are covariantly conserved, i.e., satisfy (2.33), and gauge transform as (2.32). Let us assume they exists
(j )
and can be expanded as in (2.31), with Jk, given by Jk, = Tk . j denote the tensors
1 1 2 1 2 3
j = (j , j , j
, . . .), which now enter in the right-hand sides of (4.15). From (2.33),
(j )
T1
is exactly conserved at zeroth order in h.

Candidates for first two currents (up to an overall
296
factor) are
j = + O(h ),
j = , , + O(h ).
(4.18)
The next order expressions for these currents will contain the potentials a and they are expected
to be nonlocal.
5. Conclusion
In the previous sections we developed tools for writing gauge theories on the space {}
of equivalent star products associated with any given Poisson bi-vector . The gauge theories
can be regarded as an extension of non-commutative U (1) gauge theory. Since general gauge
transformations induce O(h ) corrections in the integration measure, they cannot be regarded as
purely internal transformations.
Although it is not difficult to write down matter field actions, as for example in (2.29), a
final ingredient is needed in order to introduce kinetic terms for the infinitely many gauge fields
in F , namely the trace Tr over differential operators. In additional to satisfying the usual trace
property, Tr F F should reduce to the usual action for non-commutative U (1) gauge fields
upon restricting F to (2.28). Other familiar non-commutative field theories may be contained
in the full action. In order to make contact with physical theories, mechanisms, such as the Higgs
mechanism, should be applied to give (large) masses to all but a finite number of the gauge fields.
This may then involve introducing additional derivative-valued fields.
Finally, a more ambitious project would be to write down field theories on the space of all
equivalence classes {} of star products. This means making the Poisson bi-vector dynamical,
and then as a result all of the bi-differential operators Bn in the star product (3.2) dynamical.
Variations of these operators must then include diffeomorphisms on the underlying manifold, at
all orders in h,
setting the possible framework for a quasi-classical approximation to quantum
gravity [17].
Acknowledgement
The work of A.P. has been supported by FAPESP grant number 06/56056-0.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
For an overview, see G. Dito, D. Sternheimer, math.QA/0201168.

M. Kontsevich, Lett. Math. Phys. 66 (2003) 157.
B. Jurco, P. Schupp, Eur. Phys. J. C 14 (2000) 367.
B. Jurco, P. Schupp, J. Wess, Nucl. Phys. B 604 (2001) 148.
S. Gutt, J. Rawnsley, J. Geom. Phys. 42 (2002) 12.
G. Felder, B. Shoikhet, math.QA/0002057.
V. Dolgushev, math.QA/0504420.
A. Stern, Nucl. Phys. B 745 (2006) 236.
G. Alexanian, A. Pinzul, A. Stern, Nucl. Phys. B 600 (2001) 531.
H.J. Groenewold, Physica 12 (1946) 405.
J.E. Moyal, Proc. Cambridge Phil. Soc. 45 (1949) 99.
W. Behr, A. Sykora, Phys. Rev. D 66 (2002) 010001.
[13]
[14]
[15]
[16]
[17]
297
See for example, R.J. Szabo, Class. Quantum Grav. 23 (2006) R199.
M. Wodzicki, in: Yu.I. Manin (Ed.), Lecture Notes in Mathematics, vol. 1289, Springer-Verlag, Berlin, 1987, p. 320.
M. Dimitrijevic, F. Meyer, L. Moller, J. Wess, Eur. Phys. J. C 36 (2004) 117.
A. Zotov, Mod. Phys. Lett. A 16 (2001) 615.
See for example, S. Doplicher, K. Fredenhagen, J.E. Roberts, Commun. Math. Phys. 172 (1995) 187.

Nucl - Phys.B v.791

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.791

Uploaded by

Copyright:

Available Formats

Nuclear Physics B 791 (2008) 119

Higgs boson production at the LHC:

(1428) Pabelln 1 Ciudad Universitaria, Capital Federal, Argentina

E-mail address: massimiliano.grazzini@cern.ch (M. Grazzini).

G. Bozzi et al. / Nuclear Physics B 791 (2008) 119

G. Bozzi et al. / Nuclear Physics B 791 (2008) 119

G. Bozzi et al. / Nuclear Physics B 791 (2008) 119

The logarithmically-enhanced contributions are embodied in the resummed component d a1 a2 .

G. Bozzi et al. / Nuclear Physics B 791 (2008) 119

(N1 , N2 )-moments with respect to the two variables

dz2 z2N2 1 Wa1 a2 (y,

be considered as a function of the two variables z1 = e+y z and z2 = ey z. Thus, throughout

dz2 z2N2 1 h(y; z),

where N1 = N + i/2, N2 = N i/2.

G. Bozzi et al. / Nuclear Physics B 791 (2008) 119

G. Bozzi et al. / Nuclear Physics B 791 (2008) 119

H(N1 ,N2 ) = [HN1 HN2 ]1/2 .

G. Bozzi et al. / Nuclear Physics B 791 (2008) 119

meaning that is different from ours.

G. Bozzi et al. / Nuclear Physics B 791 (2008) 119

G. Bozzi et al. / Nuclear Physics B 791 (2008) 119

G. Bozzi et al. / Nuclear Physics B 791 (2008) 119

dNNLL+NLO /(dqT dy)

G. Bozzi et al. / Nuclear Physics B 791 (2008) 119

G. Bozzi et al. / Nuclear Physics B 791 (2008) 119

G. Bozzi et al. / Nuclear Physics B 791 (2008) 119

G. Bozzi et al. / Nuclear Physics B 791 (2008) 119

G. Bozzi et al. / Nuclear Physics B 791 (2008) 119

G. Bozzi et al. / Nuclear Physics B 791 (2008) 119

the process-independent form factor and the process-dependent hard

= Gc + Gi1 ,N1 + Gcb1 ,N1 + Gi2 ,N2 + Gcb

G. Bozzi et al. / Nuclear Physics B 791 (2008) 119

[4] S. Dawson, Nucl. Phys. B 359 (1991) 283;

G. Bozzi et al. / Nuclear Physics B 791 (2008) 119

G. Bozzi, S. Catani, D. de Florian, M. Grazzini, Phys. Lett. B 564 (2003) 65.

Nuclear Physics B 791 (2008) 2059

Long wavelength limit of evolution of cosmological

E-mail address: yj4t-hmzk@asahi-net.or.jp.

T. Hamazaki / Nuclear Physics B 791 (2008) 2059

1. Introduction and summary

T. Hamazaki / Nuclear Physics B 791 (2008) 2059

T. Hamazaki / Nuclear Physics B 791 (2008) 2059

where n is an arbitrary vector field satisfying

T. Hamazaki / Nuclear Physics B 791 (2008) 2059

where the energymomentum transfer vector Q A is defined by

TAj = a(A + PA )(vA B)Yj ,

T. Hamazaki / Nuclear Physics B 791 (2008) 2059

and the perturbation part of fluid variables is given by

total energymomentum tensor T = A TA , we can define the gauge invariant perturbation

T. Hamazaki / Nuclear Physics B 791 (2008) 2059

and perturbation equations as

This Ya has been called the SasakiMukhanov variable [25,26].

L1 (DT , A) = (D) + 3H (D) +

As for the fluid components, T = Q gives the background equations as

and the perturbation equations as

T. Hamazaki / Nuclear Physics B 791 (2008) 2059

T. Hamazaki / Nuclear Physics B 791 (2008) 2059

T. Hamazaki / Nuclear Physics B 791 (2008) 2059

By summing (2.78) with respect to all the fluid components, we obtain

Integrating (2.59) gives

T. Hamazaki / Nuclear Physics B 791 (2008) 2059

If we assume that c is nonvanishing, since c = O(1), A O(1), therefore

T. Hamazaki / Nuclear Physics B 791 (2008) 2059

I (1) = (1) = (1) = 0.

I (1) = (1) = (1) = 0.

In the limits p 1, p 1, this empirical S formula gives

As for Fa , F , Ga defined by