Nucl - Phys.B v.667

Nuclear Physics B 667 (2003) 354
www.elsevier.com/locate/npe
Gauge/string duality for QCD conformal operators

A.V. Belitsky a , A.S. Gorsky b,c , G.P. Korchemsky d
a Department of Physics, University of Maryland at College Park, College Park, MD 20742-4111, USA
b Institute of Theoretical and Experimental Physics, B. Cheremushkinskaya 25, 117259 Moscow, Russia
c Institut des Hautes Etudes Scientifique, Le Bois-Marie, F-91440 Bures-sur-Yvette, France
d Laboratoire de Physique Thorique, 1 Universit de Paris XI, 91405 Orsay cedex, France
Received 28 April 2003; accepted 18 June 2003
Abstract
Renormalization group evolution of QCD composite light-cone operators, built from two and
more quark and gluon fields, is responsible for the logarithmic scaling violations in diverse physical
observables. We analyze spectra of anomalous dimensions of these operators at large conformal
spins at weak and strong coupling with the emphasis on the emergence of a dual string picture.
The multi-particle spectrum at weak coupling has a hidden symmetry due to integrability of the
underlying dilatation operator which drives the evolution. In perturbative regime, we demonstrate
the equivalence of the one-loop cusp anomaly to the disk partition function in two-dimensional
YangMills theory which admits a string representation. The strong coupling regime for anomalous
dimensions is discussed within the two pictures addressed recentlyminimal surfaces of open strings
and rotating long closed strings in AdS background. In the latter case we find that the integrability
implies the presence of extra degrees of freedomthe string junctions. We demonstrate how the
analysis of their equations of motion naturally agrees with the spectrum found at weak coupling.
2003 Elsevier B.V. All rights reserved.
PACS: 11.25.Hf; 11.15.Pg; 12.38.Cy
1. Introduction
The formalism of path-ordered exponentials, or Wilson loops, is an indispensable tool in
QCD. It allows one to formulate complicated QCD dynamics in terms of gauge invariant
degrees of freedom [1] and express correlation functions as a sum over random walks,
E-mail address: belitsky@physics.umd.edu (A.V. Belitsky).
1 Unit Mixte de Recherche du CNRS (UMR 8627).
0550-3213/$ see front matter 2003 Elsevier B.V. All rights reserved.
doi:10.1016/S0550-3213(03)00542-X
A.V. Belitsky et al. / Nuclear Physics B 667 (2003) 354
e.g.,
0|J (x)J (0)|0 =

C
mL[C]

[C]0| tr P exp i dx A (x) |0,
(1.1)
where J (x) = (x) (x) is the electromagnetic current of a quark with mass m. L[C]
is the length of a closed path C = C[0, x] that passes through the points x and 0. [C]
is a geometrical phase, the so-called Polyakov spin factor, that takes into account the variation of the quark spin upon parallel transport along the path C. To evaluate (1.1), one has to
calculate the (nonperturbative) expectation value of the Wilson loop for an arbitrary path C
and perform resummation in the right-hand side of (1.1). Both tasks are extremely difficult
and cannot be performed in full at the current stage. Recently, a significant progress has
been achieved in understanding the strong coupling dynamics of supersymmetric gauge
theories [2,3] based on the gauge/string correspondence [46]. One of the goals of the
present paper is to establish a relation between certain QCD observables and their counterparts in string theory.
There exists a special class of QCD observables, for which the sum over paths in the
right-hand side of (1.1) can be performed exactly. As a relevant physical example, let us
consider a propagation of an energetic quark through a cloud of soft gluons. In the limit
when its energy goes to infinity, the quark behaves as a point-like charged particle that
moves along a straight line and interacts with soft gluons. This means that the sum over
all paths in (1.1) is dominated in that case by a saddle point describing a propagation of a
quark along its classical path. The Wilson loop corresponding to this path has the meaning
of the eikonal phase acquired by the quark field upon interaction with gluons. In this way,
the Wilson loop encodes universal features of soft radiation in QCD. Let us point out two
important QCD observables, in which similar semiclassical regime occurs: the IsgurWise
heavy-meson form factor, ( ), and parton distributions in a hadron, f (x), at the edge of
the phase space, x 1. As we will demonstrate below, both observables are given by an
expectation value of a Wilson loop with the integration contour C fixed by the kinematics
of the process. A unique feature of the contour C is that it contains a few cusps at points in
Minkowski spacetime where the interaction with a large momentum has occurred in the
underlying process.
In this way, Wilson loops with cusps, being fundamental objects in gauge theories,
have a direct relevance for QCD phenomenology. Their calculation in the strong coupling
(nonperturbative) regime is one of the prominent problems in gauge theories. In the present
paper, we make use of a recent progress in understanding the strong coupling behavior
of the N = 4 supersymmetric (SUSY) YangMills (YM) theory to get some insights into
properties of Wilson loops in QCD. Our analysis relies on the gauge/string duality between
N = 4 SUSY gauge theory and a string theory on AdS5 S 5 background [46].2 A natural
question arises: what is in common between QCD and N = 4 SUSY YangMills theory?
The two theories have quite different dynamics at large distances, while at short distances
they have many features in common. For instance, anomalous dimensions of twist-two
2 Note that recently there were several studies which aimed on the derivation of strong coupling results for
high-energy QCD observables, most notably Refs. [713].
operators, contributing to high-energy QCD processes, have a similar form in two theories
including their behavior at large Lorentz spin. Having this relation in mind, we will study
expressions for resummed anomalous dimensions in the N = 4 SUSY YangMills theory.3
We concentrate on two observations relevant to our present discussion. Recently, it was
proposed that anomalous dimensions of twist-two composite operators with large Lorentz
spin J are equal in the strong coupling limit to the energy spin of a folded closed
string rotating in AdS5 and having the shape of a long rigid rod (mimicking the adjoint
QCD string of glue with heavy quarks at the folding points) [7]

s Nc
ln J + O J 0 .
J (s ) = E J = 2
(1.2)
Another observation comes from the calculation of the Wilson loop in the strong coupling
regime via the minimal surface, Amin , swept by an open string which goes into the fifth
AdS dimension and whose ends trace its contour in Minkowski space [2]. This picture
naturally embeds the color flux tubes between the color sources, albeit penetrating into
an extra Liouville dimension [14] as compared to the conventional four-dimensional setup.
From the QCD perspective, one is mostly interested in calculating Wilson loops with cusps.
Such contours were considered in a number of studies [810]. For the -shaped Wilson
loop with two cusps (see Eq. (2.14) below) the result reads [9,10]

s Nc 2
ln (iv n ).
W (v n ) = exp(iAmin ), Amin = i
(1.3)
Comparing (1.2) and (1.3), one notices that J and Amin depend on the coupling constant
in the same manner. The coincidence is not accidental, of course. Identifying s -factor
as the leading term in the expression for the cusp anomalous dimension

s Nc
+ O ( s )0 ,
cusp (s ) =
(1.4)
one can show that Eqs. (1.2) and (1.3) hold in a conformal gauge theory for arbitrary
coupling constant [15]. Eq. (1.4) defines the asymptotic behavior of cusp(s ) in the N = 4
SUSY YM theory in the strong coupling regime.
In the present paper, we will extend these results to composite QCD operators of higher
twist, built from an arbitrary number of fields and having autonomous renormalization
scale evolution. Such operators are known in QCD as multi-particle conformal operators.
We determine the spectrum of their anomalous dimensions at large Lorentz spins, J 1,
both in weak and strong coupling regimes. At weak coupling, the spectrum has a hidden
symmetry due to integrability of the dilatation operator in the underlying YangMills
theory. At strong coupling, the two different picturesthe rotating folded long string and
the minimal surface swept by an open string with ends attached to the cuspresult into
the same asymptotic expression for the anomalous dimension of conformal operators. We
argue that integrability at weak coupling implies the presence of extra stringy degrees
3 Since the anomalous dimensions originate from short distances, we find it appropriate to refer to them in the
strong coupling regime as resummed anomalous dimensions rather then nonperturbative ones.
of freedom at strong couplingthe string junctionsand elucidate the relation between

the anomalous dimensions of multi-particle conformal operators at strong coupling and
solutions to the classical equations of motion for the string junctions.
To sew together the expressions for the anomalous dimensions at weak and strong coupling, one needs the stringy description of the weak coupling regime in YangMills theory. One approach to deriving such stringy picture, based on the hidden integrability of
evolution equations for the light-cone operators, has been developed in [13]. It relies on
the identification of the underlying YangMills dilatation operator as the Hamiltonian of
SL(2, R) Heisenberg spin chain. The SL(2, R) group naturally appears in this context as a
subgroup of the four-dimensional conformal group acting on the light-cone. Due to complete integrability of the spin chain model, the spectrum of the anomalous dimensions of
multi-particle light-cone operators can be found exactly in terms of the Riemann surfaces
whose genus is related to the number of the particles involved. As a consequence, the twist
expansion on the light-cone corresponds to the summation over the genera of the corresponding Riemann surfaces.
Our consequent presentation is organized as follows. In Section 2 we review the relation
of certain QCD observables to expectation values of Wilson lines and elucidate the physical
meaning of the cusp anomaly. We also give there results for the two-loop cusp anomalous
dimensions in supersymmetric theories. In Section 3, we turn to multi-particle operators
and show how conformal symmetry in gauge theory simplifies the problem of finding the
spectrum of their anomalous dimensions. We demonstrate the way the integrability of the
evolution equations arises through the identification of the underlying dilatation operator
with the Hamiltonian of a Heisenberg spin chain. In the subsequent section we address
the stringy interpretation of the gauge theory results. Our analysis suggests that the cusp
anomaly at weak coupling is described by a string which is different from the Nambu
Goto string. To identify this string we show that the cusp anomalous dimension to one-loop
order is equal to the transition amplitude for a test particle on the SL(2, R) group manifold,
which in its turn is given by the partition function of two-dimensional YangMills theory
on a disk which admits a stringy representation. Next, in Section 5, we discuss the strong
coupling computation within the open and closed string theory context for multi-particle
operators extending earlier results. Section 6 contains concluding remarks. In Appendix A,
we calculate the contribution of vacuum polarization to the two-loop cusp anomaly in
dimensional regularization and dimensional reduction schemes.
2. Wilson loops as QCD observables
As was emphasized in the introduction, there are several QCD observables directly
related to expectation values of Wilson lines.
2.1. IsgurWise form factor
The IsgurWise form factor ( ) describes the electromagnetic transition of a heavy
meson |M(v) with mass m and momentum p mv , built from a heavy quark and a
light component, to the same meson with momentum p mv (with v2 = v 2 = 1) [16]

M(v ) (0) (0) M(v) = ( )(v + v ) .

(2.1)
In the heavy-quark limit, m , it depends only on the product of velocities v v =

cosh , or equivalently on the angle between them in Minkowski spacetime. The
operator (0) annihilates the heavy quark inside the meson |M(v). For m , the
heavy quark behaves as a classical particle with the velocity v interacting with the light
0
component of the meson through its eikonal current, Ja,eik (x) = d v t a (4) (x v )
with t a being the quark color charge. This allows one to replace
(x) eim(vx)bv v [x; ],
0

d v A(x + v ) ,
v [x; ] P exp i
(2.2)
where v [x; ] is the eikonal phase of a heavy quark in the fundamental representation
of the SU(Nc ) and bv amputates this quark inside the heavy meson. Applying similar
transformation to the quark field in the final state meson |M(v ), one obtains the following
expression for the form factor [17]

) v [; 0]v [0; ] M(v)
( ) = M(v
(2.3)
P exp i dx A (x) ,
with |M(v)
= bv |M(v) standing for the light component of the meson with the amputated
heavy quark. Here the net effect of nonperturbative interaction with the light component
of the heavy meson is accumulated only via the Wilson line4 evaluated along the contour
consisting of two rays that run along the meson velocities v and v . It is important to
notice that the contour has a cusp at the point 0, in which the interaction with the external
probe has occurred.
The IsgurWise form factor ( ) is a nonperturbative observable in QCD [19]. It
depends on hadronic, long-distance scales as well as on the ultraviolet cut-off m,
which sets up the maximal energy of soft gluons. Although ( ) cannot be calculated
at present in QCD from the first principles, its dependence on can be found from the
renormalization group equation [1,20]

(2.4)
+ (g)
+ cusp ( ; s ) ( ) = 0,
g
where s = g 2 /(4) is the QCD coupling constant and cusp ( ; s ) is the cusp anomalous
dimension. To the lowest order in s
cusp ( ; s ) =

s CF
( coth 1) + O s2 ,
(2.5)
3
d k (k)ak |0, the
transition of the light cloud from the initial to the final state can be described by an exact light-quark propagator
in an external gluon field. In this manner the IsgurWise form factor will be rewritten as a correlation function of
Wilson loop along the contour formed by the straight-line trajectories of heavy quarks and the phase of the light
spectator quark in the world-line expression of its propagator [18].
4 Representing the brown-muck of the heavy meson by a wave function |M =
.
where CF = (Nc2 1)/(2Nc ) is the Casimir operator of the SU(Nc ) group in the
fundamental representation. The two-loop correction to (2.5) has been calculated in [20]
and its dependence on is more involved.
Eq. (2.4) follows from renormalization properties of the Wilson line in the right-hand
side of (2.3). It acquires the anomalous dimension due to the presence of a cusp on
the integration contour. The cusp anomalous dimension cusp( ; s ) determines universal
features of soft-gluon radiation and is known as the QCD bremsstrahlung function. As
such, it is a positive definite function of the cusp angle (for real Minkowski angle ) at
arbitrary value of the coupling constant
cusp ( ; s ) 0.
(2.6)
To see this we recall that at the cusp point the heavy quark suddenly changes its velocity
from v to v and, due to instantaneous acceleration, it starts to emit soft (virtual and real)
gluons with momentum k < with a cut-off m. Denoting the eikonal phase of the
heavy quark as v [; 0]v [0; ] and using its unitarity, = 1, one calculates
the total probability for the heavy quark to undergo the scattering (the Bjorken sum rule) as
2

,
M(v)
M X (v ) M(v)
= ( ) +
1 = M(v)
(2.7)
X
where in the right-hand side we inserted the decomposition of the unity operator over the
physical hadronic states and separated the contribution of the ground state meson, |M(v),
from excited states |MX (v). The Wilson

line (2.3) defines the probability of the elastic
transition, | |2 exp(w) with w = 2 (dk/k)cusp( ; s (k)). For = 0, depending on
the sign of cusp ( ; s ), it either vanishes or goes to infinity for . In order to preserve the unitarity condition | |2 1 that follows from (2.7), one has to require that 0
for leading to (2.6). At = 0 the cusp vanishes, that is the heavy meson stays
intact, the sum in (2.7) equals zero and ( = 0) = 1. This implies that the cusp anomalous
dimension vanishes for 0.
2.2. Parton distributions at x 1
Our second example is provided by deeply inelastic scattering of a hadron H (p) with
momentum p off a virtual photon (q) with momentum q2 = Q2 p2 in the exclusive limit xBj = Q2 /(2p q) 1, i.e., when the invariant mass of the final state system becomes small (q + p)2 Q2 . In the scaling limit, Q2 , the cross section of the process
is expressed in terms of the twist-two quark distribution function [21], see also Ref. [22],

f (x) =
d ix
e
H (p) ( n) n [ ; 0] (0) H (p) ,
2
(2.8)
describing the probability to find a quark inside the hadron H (p) with the fraction x of
its momentum p. The Wilson line stretched in between the quark fields makes the bilocal
operator gauge invariant. It goes along the light-like direction n = (q + p xBj )/(p q),
so that n2 = 0 and n p = 1.
The matrix = n
/ in (2.8) serves to select the quark states with opposite helicities. For
our purposes, we will not specify and treat it as a free parameter. The Mellin moments of
the distribution function (2.8) are related to the matrix elements of local twist-two operators
1

dx x J f x; 2 = H (p) (0) (in D)J (0) H (p) OJ 2 ,
(2.9)
where D = iA is a covariant derivative. Their dependence on the ultraviolet cut-off

is described by an evolution equation, whose solution reads in terms of the anomalous
dimensions
(s )
J
2
2
OJ =
(2.10)
OJ 0 ,
0
where we assumed for simplicity that the coupling constant does not run, = 0. The
anomalous dimension of the twist-two operators, J (s ), depends on the choice of the
matrix . In particular, in case when selects the same helicities of the quark fields in
(2.9), = (1 + 5 )/
n , the anomalous dimension is

s
3
J (s ) = CF 2(J + 2) + 2E
(2.11)
+ O s2 ,
2
where (J ) = d ln (J )/dJ is the Euler psi-function and E is the Euler constant. For
other choices of , the anomalous dimensions have extra (rational in J ) terms in addition
to the -function, see, e.g., [23]. As we will argue below, Eq. (2.11) has a hidden symmetry
which is responsible for integrability of evolution equations for three-quark (baryonic)
composite operators. Going over to the N = 4 SUSY YangMills theory, one finds that
the same expression (2.11) defines (up to redefinition of the color factor CF Nc ) the
anomalous dimensions of multiplicatively renormalizable twist-two operators [24].
As follows from (2.9), the asymptotics of the distribution function for x 1 is related
to the contribution of twist-two operators of large Lorentz spins J 1/(1 x) 1. One
finds from (2.11) that the anomalous dimension scales in this limit as

1
B2n
s
3
2n
J (s ) = CF ln(J + 2) + E
+ ,
(J + 2)
4 2(J + 2)
2n
n=1
(2.12)
where Bn s are the Bernoulli numbers, B2 = 1/6, B4 = 1/30, . . ., and the ellipsis stands
for higher order terms in s . It turns out that the leading scaling behavior J (s ) ln J
is a universal property of the anomalous dimensions of the twist-two operators (2.9) for
arbitrary . It holds to all orders in s and is intrinsically related to the cusp anomaly of
the Wilson loops. The reason for this is that analyzing deeply inelastic scattering for x 1
one encounters the same physical phenomenon as in the case of the IsgurWise form factor,
i.e., the struck quark carries almost the whole momentum of the hadron and, therefore, it
interacts with other partons by exchanging soft gluons. In these circumstances, in complete
analogy to the previous case, Eq. (2.2), the quark field can be approximated by an eikonal
10
phase evaluated along the classical path in the direction of its velocity p = mv ,

f (x) =
d i(1x)
e
H (p) W (v n i0) H (p) ,
2
(2.13)
where |H (p) is the state of the target hadron with amputated energetic quark and the
causal i0 prescription ensures the correct spectral property, f (x) = 0 for x > 1. The shaped Wilson line in Eq. (2.13) consists out of two rays and one segment: a link from
to 0 along the velocity of the incoming quark, next along the light-cone direction n to the
point n and, then, along v from 0 to ,
W (v n ) = v [ ; ]n [ ; 0]v [0; ].
(2.14)
Substituting (2.13) into (2.9), one finds the following relation between the matrix elements
of local composite operators at large spin J and the Wilson loop expectation value [15]

2

OJ = H (p) W (iJ ) H (p) P exp i dx A (x) .

(2.15)
Here, the large Lorentz spin of the local operator defines the length of the light-cone
segment:
v n iJ.
(2.16)
We would like to stress that Eq. (2.15) holds only for J 1.

According to Eq. (2.15), the -dependence of the twist-two operators follows from
the renormalization of the Wilson line (2.14). The latter has two cusps located at the
points 0 and n . In distinction with the previous case, one of the segments attached
to the cusps lies on the light-cone, n2 = 0, and the corresponding cusp angle is infinite,
(1/2) ln[(v n)2 /n2 ] . In this limit, the cusp anomalous dimension scales to all
orders in s as [20]

cusp ( ; s ) = cusp(s ) + O 0 .
(2.17)
Here cusp (s ) is a universal anomalous dimension independent on . At weak coupling,
it has the following form in QCD
2

5
s
s
67 2
cusp (s ) = CF +
(2.18)
CF Nc
nf
+ O s3 ,
36 12
18
where nf is the number of quark flavors. This expression was obtained within the
dimensional regularization scheme (DREG) by using the MS-subtraction procedure, s
sMS .
The divergence of the anomalous dimension (2.17) for indicates that the Wilson
line with a light-like segment satisfies an evolution equation different from (2.4). The
modified equation looks like [15]

+ (g)
+ 2cusp(s ) ln[i(v n) ] + (s ) W (v n ) = 0. (2.19)
11
Here the factor of 2 stems from the presence of two cusps on the -shaped line
contour and (s ) is a process-dependent anomalous dimension. The explicit dependence
of the anomalous dimension on the renormalization scale implies the absence of
the multiplicative renormalizability of the light-like Wilson line. Combining together
Eqs. (2.19) and (2.15), we obtain the renormalization group equation for local composite
operators OJ (2 ) at large J . Matching its solution into (2.10), we find the asymptotic
behavior of the anomalous dimensions of the twist-two quark operators for J

(qq)
J (s ) = 2cusp(s ) ln J + O J 0 .
(2.20)
Repeating a similar analysis for the twist-two gluon operators, one can show that their
matrix elements satisfy (2.15) with the Wilson line defined in the adjoint representation.
Therefore, their anomalous dimension satisfies (2.20) upon replacing CF Nc leading
to [15]

Nc (qq)
(gg)
J (s ) =
(2.21)
(s ) + O J 0 .
CF J
In general, the quark and gluon operators mix with each other. However, at large J
the mixing occurs through the exchange of a soft quark with momentum 1/J . Its
contribution to the corresponding anomalous dimensions is suppressed by a power of 1/J
leading to

1
1
(gq)
(qg)
,
J (s ) = O
.
J (s ) = O
(2.22)
J
J
We would like to stress that the relations (2.20)(2.22) are valid to all orders in s .
Remarkably enough, they hold both in QCD and its supersymmetric extensions. In
the latter case, the mixing matrix has a bigger size due to the presence of additional
scalar fields. Nevertheless, this matrix remains diagonal at large J . Since the fields in
supersymmetric YM theories belong to the adjoint representation, the diagonal matrix
elements are the same

J(ab) = 2ab cusp (s ) ln J + O J 0 ,
(2.23)
with a, b = (q, g, s) and cusp (s ) defined in the adjoint representation. Notice that one
cannot use for cusp(s ) the two-loop expression (2.18) for CF = Nc , since it was obtained
within the dimensional regularization scheme which breaks supersymmetry and also lacks
the contribution of possible scalars.
2.3. Cusp anomaly in supersymmetric theories
To calculate cusp (s ) in a supersymmetric YangMills theory, we use the regularization
by dimensional reduction (DRED) [2527]. In analogy to compactification, one gets this
scheme by dimensionally reducing the four-dimensional theory down to d = 4 2 < 4
dimensions. In comparison with the DREG scheme, the Lagrangian involves now the socalled epsilon-scalars generated by 2 components of the four-dimensional gauge field. To
two-loop accuracy, these scalar fields contribute to the Wilson loop by modifying the selfenergy of a gluon at the level of O(0 ) corrections. The calculation of the corresponding
12
Feynman diagram is straightforward and details can be found in Appendix A. It leads to

the two-loop correction (sDR /)2 CF Nc /12 to the right-hand side of (2.18), resulting
into

DR 2

3
16 2
5
1
DR
s
cusp = s Nc +
nf
ns
+ O sDR ,
Nc Nc
9
12
18
9
(2.24)
where s = sDR is the coupling constant in the DRED scheme with modified minimal
subtractions. Here, in comparison with (2.18), we added the contribution of ns = Ns Nc
real scalars and set CF = Nc since in a supersymmetric gauge theory fields belong to
the adjoint representation of the SU(Nc ). Notice that one can obtain the same expression
(2.24) by expanding (2.18) in powers of the coupling constant in the dimensional reduction
scheme, sDR , which is related to the coupling constant in the dimensional regularization
scheme, sMS , by a finite renormalization [28,29]5

Nc sDR
MS
DR
DR 2
+ O s
s = s
(2.25)
1
.
12
Eqs. (2.18) and (2.24) define the cusp anomalous dimensions in two different renormalization schemes, based on dimensional regularization and dimensional reduction, respectively.
It is the latter scheme that does not break supersymmetry.
Using (2.23) and (2.24) we obtain the asymptotic behavior of the twist-two anomalous
dimension in the DR scheme (s sDR ) in various supersymmetric theories:
In the N = 1 YM theory, one has one Majorana fermion in the adjoint representation,
nf = Nc , and no scalars ns = 0

s Nc 2 3 2
s Nc
N =1
+ O s3 .
+
cusp (s ) =
(2.26)
2 12
In the N = 2 YM theory, one has two Majorana fermions in the adjoint representation,
nf = 2Nc , and two real scalar fields in the adjoint representation, ns = 2Nc

s Nc 2
2
s Nc
N =2
+
+ O s3 .
1
cusp
(s ) =
(2.27)
12
In the N = 4 YM theory, one has four Majorana fermions in the adjoint representation,
nf = 4Nc , and six real scalar fields in the adjoint representation, ns = 6Nc

s Nc 2
2
s Nc
N =4
+
+ O s3 .
cusp
(s ) =
(2.28)
12
Notice that the two-loop correction to cusp (s ) is positive in all cases except the N = 4
theory. It becomes smaller as one goes from QCD to the N = 1 and N = 2 theory and,
then, it becomes negative at N = 4. We recall that cusp(s ) > 0 for arbitrary s , Eq. (2.6).
5 This simple rule of substitutions can be understood by noticing that
cusp governs the scale dependence
of a physical observable, the IsgurWise form factor, and, therefore, it is renormalization scheme invariant.
13
Together with (2.23), the expressions (2.26)(2.28) establish the large-J asymptotics of
the anomalous dimensions of the twist-two operators in supersymmetric theories. Namely,
the matrix of anomalous dimension is diagonal at large J with the entries on the main diagonal equal to 2cusp(s ) ln J . Eq. (2.26) agrees with the results of explicit two-loop calculations in Ref. [30]. Eq. (2.28) is in disagreement with the results of Ref. [24].6 Eq. (2.27)
is a prediction since a two-loop calculation in that case has not been performed yet.
3. Conformal operators and integrable spin chains

So far we have discussed two-particle composite operators. Let us now generalize
our consideration to operators involving many fields. Such operators are of great
phenomenological interest as their matrix elements define, e.g., baryon distribution
amplitudes [31] and higher-twist corrections in various high-energy processes [32]. To
start with, we define first a framework which solves partially the expected complications
in the mixing problem in this case.
3.1. Two-particle conformal operators
The local twist-two operators OJ can be obtained from the expansion of a nonlocal
light-cone operator, cf. Eq. (2.8),

(i )J
OJ (0).
tr X1 (0)n [0, ]X2 ( n)n [, 0] =
J!
(3.1)
J =0
Here Xi (n ) Xia (n )t a denotes a general primary operator in the gauge theory defined
in the adjoint representation of the SU(Nc ) group, living on the light-cone n2 = 0 and
having definite quantum numbers with respect to transformations of the conformal group
(see Eq. (3.3) below). The latter condition implies that Xia can be identified as the quasipartonic operator [23], that is a scalar field, or a specific component of the quark field and
the gluon strength tensor.
Two Wilson lines in (3.1) run along the light-cone in opposite directions between
the points 0 and and ensure the gauge invariance of the operator. The local twist-two
operators OJ were introduced in the previous section. Let us reinstate their definition again,

OJ ( ) = tr X1 (n )(in D)J X2 (n ) = tr X1 ( )(i)J X2 ( ) ,
(3.2)
where in distinction with the previous case, Eq. (2.9), the covariant derivative is defined in
the adjoint representation, D X = ig[A , X]. Here in the second relation we have
chosen the gauge n A(x) = 0 and simplified the notation for the argument of the fields.
The twist-two operators defined in this way mix under renormalization with the operators
containing total derivatives (i)l OJ l ( ) with 1 l J . Although the mixing can be
neglected for forward matrix elements like (2.8), it survives in the case one considers
6 Eq. (2.28) were in agreement with the results of Ref. [24] if one would assume that their result is given in
the MS scheme and transforms the coupling constant to the DR scheme via (2.25).
14
matrix elements with different momenta in the initial and final state, or when they are a part
of multi-parton operators. In conformal theories, one can construct linear combinations of
such operators, the so-called conformal operators [3338], in such a way that they are
renormalized multiplicatively to all orders in the coupling constant.7 The mixing between
these operators is protected by the SO(4, 2) conformal symmetry of the gauge theory, more
precisely, by its collinear SL(2, R) SU(1, 1) subgroup, which acts on the primary fields
living on the light-cone as follows

a + b
a + b
2ji
,
Xi ( ) (c + d)
(3.3)
Xi
,
c + d
c + d
with a, . . . , d real such that ad bc = 1. Here ji = (si + li )/2 is the conformal spin of
the field Xi ( ) equal to one half of the sum of its canonical dimension, l, and projection of
the spin onto the light-cone, + Xi ( ) = si Xi ( ). By definition, the conformal operators
O J ( ) are composite operators built from the primary fields and satisfying (3.3) with ji
replaced by J + j1 + j2 . It is easy to see that the operators (3.2) do not obey the latter
condition.
To simplify the analysis, one chooses the axial gauge n A(x) = 0. Then, the operators
in the left-hand side of (3.1) are given by the product of two primary fields. It is transformed
under (3.3) in accordance with the tensor product of two SL(2, R) representations8 labelled
by the spins j1and j2 . Decomposing this product into the sum of irreducible components,
[j1 ] [j2 ] = J 0 [J + j1 + j2 ], one can identify the spin-J component as defining the
conformal operator O J ( ). It has the following form

J
J (1 ,2 ) 2 1
tr X1 (1 )X2 (2 ) = = ,
OJ ( ) = i (2 + 1 ) PJ
(3.4)
1
2
2 + 1
where a = /a , a = 2ja 1 and PJ(1 ,2 ) are the Jacobi polynomials. To restore gauge
invariance, one has to substitute X( ) = (n D)X( ). Going back to (3.1), one obtains
the operator product expansion for a nonlocal light-cone operator over the conformal
operators, see, e.g., Refs. [3941],

tr X1 (0)n [0, ]X2 ( n)n [, 0]
=

J =0
(i )J
CJ (1 , 2 )
J!
du uJ +1 (1 u)J +2 O J (u ),
(3.5)
where
CJ (1 , 2 ) = (2J + 1 + 2 + 1)
(J + 1) (J + 1 + 2 + 1)
.
(J + 1 + 1) (J + 2 + 1)
(3.6)
7 In gauge theories with broken conformal symmetry, like QCD, this holds only to the lowest order in .
s
Beyond the leading order O J s start to mix with (i)l O J l . The corresponding mixing anomalous dimensions
are determined solely by the conformal anomalies, see the last paper of Ref. [37], Eqs. (51) and (113) for explicit
expressions.
8 These are unitary representations of the SL(2, R) group of the discrete series.
15
The Lorentz operators (3.2) can be re-expressed in terms of conformal operators O J as

OJ =
J
cjJ (1 , 2 )i J j (2 + 1 )J j O j ,
(3.7)
j =0
1
( , )
with the expansion coefficients cjJ (1 , 2 ) = Cj (1 , 2 ) 0 du(1 u)1 u2 +J Pj 1 2 (2u
1), which can be calculated in terms of the hypergeometric function 3 F2 . As was already
mentioned, the conformal operators evolve autonomously under renormalization and their
anomalous dimensions have a universal scaling behavior at large J , Eqs. (2.20) and (1.2).
Let us consider the forward matrix element of the both sides of (3.5) for large lightcone separations, 1. In this limit, typical wavelength of gluons exchanged between
two fields X1 (0) and X2 ( n) scales as and the eikonal approximation (2.2) is justified.
This allows us to replace the quantum operators Xi ( ) in the left-hand side of (3.5)
by Wilson lines in the adjoint representations of the SU(Nc ). Then, the left-hand side
of (3.5) can be approximated as p|W ( )|pei(pn) , where the Wilson line in the
adjoint representation, W ( ), is evaluated along the same -like contour as in (2.14).
In the right-hand side of (3.5), one can neglect the contribution of operators with total
derivatives since their forward matrix elements vanish. In addition, at large the sum is
dominated by the contribution of operators with large spin J . In this way, p|O J (u )|p
p|OJ (0)|p (p n)J 2cusp (s ) ln J and
(i )J
p|W ( )|pei(pn)
(3.8)
(p n)J 2cusp (s ) ln J .
J!
J
For 1 the sum in the right-hand side receives the dominant contribution from J
i(p n) leading to [15]

p|W ( )|p p|O J (0)|p J =i(pn) 2cusp (s ) ln[i(pn)] .
(3.9)
This relation establishes the correspondence between the Wilson line in the adjoint
representation and the matrix element of the conformal operator analytically continued
to large complex values of the spin J . Notice that in the multi-color limit, Nc , one
has W ( ) = W ( )2 with W ( ) defined in the fundamental representation.
3.2. Multi-particle conformal operators
Let us generalize the above analysis to conformal operators built from three and more
primary quasi-partonic fields. As before, we construct a nonlocal operator containing N
primary fields on the light-cone and expand it in powers of light-cone separations

tr X1 (0)n [0, 2 ]X2 (2 ) XN (N )n [N , 0]
(i2 )j2
(iN )jN
Oj2 jN (0).
=
(3.10)
j2 !
jN !
j2 jN 0
Here the Wilson lines run between two adjacent fields along the light-cone and ensure the
gauge invariance. The local composite operators have the form

Oj2 jN ( ) = tr X1 ( )(in D)j2 X2 ( ) (in D)jN XN ( ) ,
(3.11)
16
Fig. 1. Leading Feynman diagram of cylinder topology for a non-local N -particle operator in the large-Nc limit
in the light-cone gauge when only nearest-neighbour interactions survive (left). The array of -shaped Wilson
lines resulting from the figure on the left in the Feynman gauge (right).
where, as before, the covariant derivative is defined in the adjoint representation. The
evolution of these operators under renormalization group transformations is much more
complicated as compared with the previous case due to larger number of particles involved
and complicated color flow. The latter can be simplified by going over to the multi-color
limit Nc . In this limit, in the axial gauge n A(x) = 0, the planar Feynman diagrams
contributing to the left-hand side of (3.10) have the topology of a cylinder9 as shown in
Fig. 1.
In the multi-color limit, the operators (3.11) mix under renormalization among
themselves and with operators containing total derivatives.
The mixing occurs between

operators with the same total conformal spin J + N
k=1 jXk with J = j2 + + jN and
their number grows rapidly with N , even for the forward matrix elements. In the latter case,
the dimension of the mixing matrix scales as J N2 for large J . At N = 2 this matrix
has a single element for arbitrary J . For the number of particles N 3 the problem of
constructing multiplicatively renormalizable operators O J is reduced to diagonalization of
the mixing matrix whose size growes with J . As we will argue below, the same problem is
equivalent to solving a Schrdinger equation for a Hamiltonian of a Heisenberg spin chain
model. Before doing this, let us analyze the picture from the point of view of the Wilson
line formalism.
3.2.1. Wilson line approach
Following the Wilson line approach [15], one can obtain the scaling behavior of the
anomalous dimensions of conformal operators at large spin J . To this end, we examine
Eq. (3.10) for large light-cone separations 2 3 N 1. As in the previous
case, the nonlocal light-cone operator in the left-hand side of (3.10) is dominated by
the contribution of soft gluons with the wavelength k . This allows one to apply the
eikonal approximation and replace the quantum fields by eikonal phases, the Wilson lines
in the adjoint representation evaluated along rays that run along momenta of particles
9 For N = 3 the result is exact for arbitrary N .
c
17
and terminate at the light-cone. Assuming for simplicity that all particles have the same
momentum p we apply the eikonal approximation10
Xi ( ) ei(pn) p [, ]bi (p)p [, ],
(3.12)
where bi (p) = bia (p)t a is the annihilation operator of the ith particle and p [, ] =
p [, ]. Then, the matrix element of the operator entering the left-hand side of (3.10)
between the vacuum and the N -particle state tr{b1 (p)b2 (p) bN
(p)}|0 is given in the
multi-color limit by (see Fig. 1)

tr W [1 , 2 ] tr W [2 , 3 ] tr W [N , 1 ] ei(pn)(2 ++N ) ,
(3.13)
with 1 = 0 and |k+1 k | 1. (In arriving at this relation we applied the vacuum
dominance property, tr W1 tr W2 = tr W1 tr W2 + O(1/Nc2 ).) Here W [2 , 3 ] is the
Wilson line in the fundamental representation evaluated along the -like contour that runs
along the momentum p from to the point 2 n , then along the light-cone to 3 n
and returns to infinity along p . Due to the presence of two cusps on the integration
contour, tr W [1 , 2 ] acquires the cusp anomalous dimension cusp (s ) ln[i(pn)] .
In this way, one finds from (3.13) that the left-hand side of (3.10) scales as
Ncusp (s ) ln[i(pn)] ei(pn)(2++N ) .
(3.14)
Let us now examine the scaling behavior of the right-hand side of (3.10). As before, for
large light-cone separations, 1, the sum is
dominated by the contribution of terms
with j2 jN (i ), or equivalently J = k jk i N . According to (3.14), the
corresponding composite operator Oj2 jN (0) is renormalized multiplicatively and has the
anomalous dimension
J(max) = Ncusp (s ) ln J.
(3.15)
We recall that this result was obtained in the multi-color limit Nc for J 1.
Anomalous dimensions of the N -particle conformal operators are defined as eigenvalues of the mixing matrix. As we argue in the next section, they can be parameterized by the
set of N 2 non-negative integers . = (.1 , . . . , .N2 ) such that 0 .1 .N2 J .
Their total number equals the size of the matrix and grows at large J as J N2 . For given
spin J , the anomalous dimensions of conformal operators occupy the band
J(min) J (.1 , . . . , .N2 ) J(max) .
(3.16)
Eq. (3.15) sets up the upper bound in the spectrum of the anomalous dimensions.
We recall that (3.15) was obtained from the analysis of the nonlocal operator in the lefthand side of (3.10) at large light-cone separations between the fields. To establish the lower
bound in (3.16), one has to relax the latter condition by allowing two or more fields to be
closely located on the light-cone. It is convenient to choose the axial gauge n A(x) = 0 and
consider tr {X1 (0)X2 (2 ) XN (N )} in the region 2 1 and |k+1 k | 1 with
10 Here we used the relation between the eikonal phases defined in the fundamental () and adjoint ()
ab = t a .
representations, t b []
18
k 2. The fields X1 (0) and X2 (2 ) are separated along the light-cone by a (relatively) short
distance. They interact with each other by exchanging particles with short wavelengths,
thus invalidating the eikonal approximation. At the same time, the interaction of these two
fields with the remaining fields still occurs through soft gluon exchanges. Since the soft
gluons with the wavelength cannot resolve the fields X1 (0) and X2 (2 ), they couple to
their total color charge. This means that one can replace the bilocal operator X1 (0)X2 (2 )
by its expansion over the two-particle conformal operators O j (2 ) and apply the eikonal
approximation to O j (2 ) and remaining fields X3 (3 ), . . . , XN (N ) afterwards. This leads
to the same expression as before, Eq. (3.13), with the only difference that the factor
tr W [1 , 2 ] is missing. As a consequence, the anomalous dimension of the nonlocal
light-cone operator under consideration is given at large J by (3.15) with N replaced by
N 2. In other words, the coefficient in front of cusp(s ) ln J in Eq. (3.15) counts the
number of fields separated along the light-cone by large distances iJ .
This suggests that the minimal anomalous dimension in (3.16) corresponds to the
configuration when all fields in the left-hand side of (3.10) are grouped into two clusters on
the light-cone located at the points 0 and iJ . In this case, the left-hand side of (3.10)
is reduced to a single Wilson line tr W [0, ] whose anomalous dimension equals
(min)
= 2cusp(s ) ln J.
(3.17)
At the
same time, going over to the limit , one finds that for a given total spin
J = k jk the right-hand side of (3.10) receives the dominant contribution from the
conformal operators with the minimal anomalous dimension. The latter is given by (3.17).
This result is in agreement with Refs. [4248].
As a function of spin J , the anomalous dimensions J (.1 , . . . , .N2 ) form the family
of (nonintersecting) trajectories labelled by the integers .1 , . . . , .N2 . Since the width
of the band (3.16) scales as ln J , while the total number of anomalous dimensions
J (.1 , . . . , .N2 ) grows as a power of J , one obtains that for J 1 the distribution of the
trajectories inside the band (3.16) can be described by a continuous function. Its explicit
form depends on the coupling constant. Going over from the weak to the strong coupling
one finds that the trajectories do not intercept as s increases although their distribution
inside the band (3.16) is modified.
3.2.2. Hamiltonian approach to evolution equations
The above analysis relied on the properties of Wilson loops. Let us now develop
a dual picture based on the properties of conformal operators. In the multi-color
limit, the conformal operators are given by linear combinations of composite operators
(3.11) including the operators with total derivatives. As was already discussed above, to
construct N -particle conformal operators, one has to decompose the tensor product of N
representations of the SL(2, R) group labelled by conformal spins of the fields, jk jXk ,
over the irreducible components and project out the nonlocal operator (3.10) onto the spinJ representation

[j1 ] [j2 ] [jn ] =
(3.18)
[J + j1 + j2 + + jN ].
J 0
19
Subsequently
applying the rule for the sum of two SL(2, R) spins, [j1 ] [j2 ] =

j12 0 [j1 + j2 + j12 ], one finds that the spin-J representation has a nontrivial multiplicity
nJ = (J + N 2)!/[J !(N 2)!]. It is uniquely specified by the external conformal spins
j1 , . . . , jN and internal spins 0 j12 j123 j12N1 J with j12k defining
the total spin in the (12 k)-channel. Each irreducible spin-J component gives rise to the
following local composite operator

{j }
{j }
O J ( ) = PJ (i1 , . . . , iN ) tr X1 (1 )X2 (2 ) XN (N ) == = ,
(3.19)
1
{j }
where {j } (j12 , . . . , j12N1 ) and PJ (x1 , . . . , xN ) is a homogeneous polynomial of

degree N in momentum fractions xk . This polynomial is the highest weight vector of the
spin-J SL(2, R) representation of the discrete series. It satisfies the system of differential
equations
2
{j }
{j }
(L1 + + Lk ) PJ = J12k (J12k 1)PJ

(L+
1
{j }
+ + L+
N )PJ
(k = 2, . . . , N),
= 0,
(3.20)
with J12k = j1 + + jk + j12k and j12N J being the total Lorentz spin. Here
0
Lk = (L+
k , Lk , Lk ) are the SL(2, R) generators in the momentum representation
L
k = xk ,
2
L+
k = 2jk xk + xk xk ,
L0k = jk + xk xk ,
(3.21)
with L2 = (L+ L + L L+ )/2 + L20 . At N = 2 the solution to (3.20) is given by Jacobi

polynomials (see Eq. (3.4)). For higher N , it can be constructed iteratively as a product of
the N = 2 solutions [44,45].11
The operators (3.19) are transformed under
the SL(2, R) transformations (3.3) as
primary fields with the conformal spin J + k jk . For given J their total number equals
the multiplicity nJ . The operators (3.19) do not have autonomous evolution and mix
under renormalization with each other. To construct multiplicative conformal operators,
one has to diagonalize the corresponding nJ nJ mixing matrix. Its eigenstates define the
coefficients c.,{j } of the expansion of the conformal operators over the basis (3.19)

{j }
O J,. ( ) =
c.,{j } O J ( )
{j }

PJ,. (i1 , . . . , iN ) tr X1 (1 )X2 (2 ) XN (N )
1 ==N =
(3.22)
and the corresponding eigenvalues provide the anomalous dimensions. Here the subscript
. = (.1 , . . . , .N2 ) with 0 .1 .N2 J enumerates different conformal
operators, or equivalently homogeneous polynomials

{j }
PJ,. (x1 , . . . , xN ) =
c.,{j } PJ (x1 , . . . , xN ).
{j }
11 One can obtain the same expression by subsequently applying the fusion rules (3.5) to the product of N
primary fields.
20
Thus, in distinction with the N = 2 case, the conformal symmetry alone does not
allow one to construct multiplicatively renormalizable conformal operators for N 3.
Nevertheless, it reduces the problem to diagonalizing the mixing matrix of the dimension
nJ . This matrix has a number of remarkable properties. To begin with, we notice that the
homogeneous polynomials entering (3.19) are orthogonal to each other with respect to the
SL(2, R) scalar product

J, {j } J , {j }
2j 1
2j 1 {j }
{j }
d N x x1 1 xN N PJ (x1 , . . . , xN )PJ (x1 , . . . , xN )
J J jj ,
(3.23)
where [d N x] = dx1 dxN (1 k xk ) and integration goes over 0 xk 1. Eq. (3.23)

follows from the fact that the SL(2, R) generators (3.21) are self-adjoint operators on
the vector space endowed with the scalar product (3.23). Then, the mixing matrix can
be interpreted as a Hamiltonian acting on the Hilbert space (3.23), J, {j }|HN |J , {j }.
Denoting |J, . PJ,. (x1 , . . . , xN ), one can determine the anomalous dimensions J (.)
of the conformal operators (3.22) as solutions to the N -particle Schrdinger equation
HN |J, . = J (.)|J, .
(3.24)
under the additional (highest weight) condition

N

L+
L0tot |J, . = J +
jk |J, .,
tot |J, . = 0,
(3.25)
k=1

with Ltot N
k=1 Lk being the total conformal spin. The Hamiltonian HN commutes
with the total SL(2, R) spin, [Ltot , HN ] = 0 and is a self-adjoint operator on the Hilbert
space (3.23). This ensures that the anomalous dimensions J (.) take real values and the
eigenstates |J, . are orthogonal to each other with respect to (3.23).
In perturbation theory, the Hamiltonian HN can be obtained from explicit calculation of
Feynman diagrams in the multi-color limit Nc [23]. In this limit, due to cylinder-like
topology of the planar diagrams, the interaction occurs only between nearest neighbors
leading to the lowest order in s (see Fig. 1)

s Nc
HN =
(3.26)
H12 + H23 + + HN1 + O (s Nc )2 ,
where the two-particle Hamiltonian Hk,k+1 acts on the coordinates xk and xk+1 only.12
Conformal invariance implies that Hk,k+1 depends on the sum of two SL(2, R) spins
(Lk + Lk+1 )2 Jk,k+1 (Jk,k+1 1)
= x1 x2 (1 2 )2 + 2(j2 x1 j1 x2 )(1 2 )
+ (j1 + j2 )(j1 + j2 1),
(3.27)
12 We recall that x has the meaning of the momentum fraction carried by the particle described by the field
k
Xk (k ).
21
where k = /xk . The general expression for Hk,k+1 looks like

Hk,k+1 = (Jk,k+1 ) + ,
(3.28)
where (x) = d ln (x)/dx is the Euler psi-function. Here ellipses denote additional,
rational in Jk,k+1 terms which are subleading for Jk,k+1 1. In distinction with the first
term in the right-hand side of (3.28), they depend on the type of particles involved. Notice
that for large spins Jk,k+1 1 the two-particle Hamiltonian (3.28) has a universal form

Hk,k+1 = ln Jk,k+1 + O (Jk,k+1 )0 .
(3.29)
At N = 2 the Hamiltonian (3.26) equals HN=2 = 2(s Nc /)H12 and its eigenvalues,
N=2 (J ) 2(s Nc /) ln J , define the anomalous dimension of the two-particle conformal
operator of spin J 1 at weak coupling, Eq. (2.20). To find the spectrum of anomalous
dimension for N 3, one has to solve the Schrdinger equation (3.24) for the N -particle
Hamiltonian (3.26). It turns out the quantum-mechanical system with the Hamiltonian
(3.24) possesses a hidden symmetry and is intrinsically related to the Heisenberg spin
magnets.
We would like to emphasize that the equivalence between evolution equations and
dynamical Hamiltonian systems is a rather general phenomenon in YangMills theories
although the interpretation of the evolution time is different. In particular, similar integrable
structures appear in the Regge asymptotics of multi-color QCD [49] and in the low-energy
behavior of the N = 2 effective action [50]. In the former case, the evolution time is the
logarithm of the energy scale tRegge = ln s, while in the latter case tSYM = ln QCD .
3.2.3. Heisenberg spin chains
Let us substitute (3.28) into (3.26) and define the following Hamiltonian
quan
HN
N
s Nc
(Jk,k+1 ),
(3.30)
k=1
with JN,N+1 JN,1 . Notice that, in general, it differs from the exact QCD Hamiltonian
(3.26) by terms that vanish for Jk,k+1 1. Therefore one expects that in spite of the fact
that the fine structure of the energy spectrum of the Hamiltonians (3.30) and (3.26) may
be different, their asymptotic behavior for J 1 is the same. Exact calculations at N = 3
confirm such expectations [4248]. The main advantage of dealing with (3.30) is that the
Hamiltonian (3.30) possesses a set of integrals of motion, q = (q2 , . . . , qN ),
quan
HN , qn = [qn , qm ] = 0,
(3.31)
quan
and, as a consequence, the Schrdinger equation for HN turns out to be completely

integrable [4349]. The Hamiltonian (3.30) is well known in the theory of integrable
models. It has been constructed in [5154] as a generalization of the celebrated spin-1/2
XXX Heisenberg magnet to higher spin representations of the SU(2) and SL(2, R) groups.
Eq. (3.30) defines periodic Heisenberg spin chain model of length N and the spin
operators being the SL(2, R) generators. The value of the spin at the kth site is given by
the conformal spin jk of the corresponding primary field. Such identification allows one
to solve the spectral problem for the Hamiltonian (3.30) exactly by the quantum inverse
22
scattering method [5156]. In particular, using the Lax operator for the XXX Heisenberg
spin magnet

iL+
u + iL0k
k
Lk (u) =
(3.32)
,
iL
u iL0k
k
with L0k and L
k being the SL(2, R) generators (3.21), one can obtain the explicit form
of the integrals of motion q2 , . . . , qN from the expansion of the transfer matrix tN (u) in
powers of the spectral parameter u

tN (u) = tr L1 (u)L2 (u) LN (u) = 2uN + q2 uN2 + + qN ,
(3.33)

with q2 = L2tot + N
k=1 jk (jk 1) depending on the total spin of the system J . Due
to complete integrability of the model, the energy spectrum is uniquely specified by their
eigenvalues, EN = EN (q2 , . . . , qN ). Applying the Bethe Ansatz [5156], one can calculate
explicitly both the energy spectrum and the corresponding eigenfunctions. We recall that
the former defines the anomalous dimensions of conformal operators, while the latter
determine the polynomials PJ,. (x1 , . . . , xN ) entering (3.22).
For our purposes, we are interested in finding the large J asymptotic behavior of the energy spectrum. Assuming for simplicity that the particles have the same SL(2, R) spin, j1 =
= jN j one obtains the following asymptotic expression for the energy [42,44,46]

N

s Nc
1
J (.) =
(3.34)
(j + ik ) + E + O 2N
,
ln 2 + "e
J
k=1
where k are roots of the transfer matrix (3.33), tN (u) = 2 k (uk ). According to (3.33),
the quantum numbers qk are given by symmetric polynomials of degree k in 1 , . . . , N .
For J 1 one finds that the roots are real and they scale differently with J in the upper
and lower part of the spectrum. In the upper part of the spectrum, all roots scale as k J
with k = 1, . . . , N leading to qk = O(J k )
(max)
s Nc
s Nc
s Nc
ln |1 2 N | =
ln qN = N
ln J.
(3.35)
In
1 2 O(J ) and k J 0 for k 3 (notice that
the lower part of the spectrum,N1
-term in the right-hand side of (3.33)). This leads to
k k = 0 due to absence of the u
qk J 2 and
s Nc
s Nc
ln |1 2 | = 2
ln J.
(3.36)
We observe a perfect agreement of these expressions with Eqs. (3.15) and (3.17) obtained
within the Wilson line approach.
In the Wilson line approach, the asymptotic behavior of the anomalous dimensions,
Eqs. (3.15) and (3.17) is tied to the scaling behavior of the two-particle spins Jk,k+1 while
in the Hamiltonian approach Eqs. (3.35) and (3.36) follow from similar behavior of the
roots k of the transfer matrix (3.33). Let us demonstrate that k Jk,k+1 at large J .
By the definition, the operator Jk,k+1 is the sum of the SL(2, R) spins of two
neighbouring particles. Its eigenvalues satisfy jk + jk+1 Jk,k+1 J , where J is the total
(min)
23
spin of N particles. Since [HN , Jk,k+1 ] = 0, one cannot assign a definite value of Jk,k+1 to
the eigenstates of the Hamiltonian HN . Nevertheless, for J the system approaches a
quasiclassical regime in which quantum fluctuations are frozen and the spins Jk,k+1 can be
treated as classical variables. To see this one notices that for quantum-mechanical system
defined by the Hamiltonian (3.30) the effective Planck constant equals unity, h = 1, and
the energy scale is defined by the total spin J . This suggests that for J h one can solve
the Schrdinger equation (3.24) by the WKB methods [42].
In the WKB approach, one looks for the solution to (3.24) in the form

i
|J, . PJ,. (x1 , . . . , xN ) = exp S0 (x) + iS1 (x) + O(h ) ,
(3.37)
h
where x = (x1 , . . . , xN ). To find the functions S0 (x), S1 (x), . . . , one requires that the
wave function (3.37) has to be an eigenstate of the transfer matrix (3.33), or equivalently
diagonalize simultaneously the integrals of motion q2 , . . . , qN . In this way, one obtains that
the leading term S0 (x) satisfies the HamiltonJacobi equations in the underlying classical
system, while subleading terms can be expressed in terms of S0 (x). To go over to the
classical limit, one applies the operator of the two-particle spin defined in (3.27) to the
WKB wave function (3.37)

h 2 (Lk + Lk+1 )2 eiS0 /h = xk xk+1 (pk pk+1 )2 + O(h ) eiS0 /h ,
(3.38)
where pk = i h xk S0 (x) is a classical momentum of the kth particle and we restored the
dependence on the Planck constant. In a similar manner, the SL(2, R) spin operators (3.21)
can be replaced by classical functions on the phase space of N particles
L
k, cl = xk ,
2
L+
k, cl = xk pk ,
L0k, cl = ixk pk .
(3.39)
Notice that the dependence on a single-particle spin disappears since jk = O(h ). One
verifies that, in agreement with (3.38),
cl 2
= xk xk+1 (pk pk+1 )2 .
(Lk, cl + Lk+1, cl )2 = Jk,k+1
(3.40)
cl
defines the classical limit of the two-particle spin Jk,k+1 . Then, one replaces
Here Jk,k+1
quan
cl
Jk,k+1 hJ
k,k+1 in (3.30), expands the Hamiltonian HN in powers of h and identifies
the leading term of the expansion as the Hamiltonian of the classical model
cl
HN
=
N
s Nc cl 2
ln Jk,k+1 .
2
(3.41)
k=1
Remarkably enough, this Hamiltonian inherits integrability properties of the quantum

model. It contains a set of integrals of motion qkcl (k = 2, . . . , N)
cl cl cl cl
HN , qk = qn , qk = 0,
(3.42)
with the Poisson bracket defined as {f, g} = xk f pk g pk f xk g. To obtain the
explicit form of qkcl , one replaces the SL(2, R) spin operators in (3.32) by their classical
counterparts (3.39) and substitutes the resulting expression for the Lax operator into (3.33).
24
This leads to
qncl =
xj1 xjn1 xjn (pj1 pj2 ) (pjn1 pjn )(pjn pj1 ), (3.43)
1j1 <<jn N
2
with n = 2, . . . , N . One observes that (qNcl ) =

Hamiltonian (3.41) can be written as
cl
HN
=
2
N
cl
k=1 (Jk,k+1 )
s Nc
ln qNcl .
and, therefore, the classical
(3.44)
cl defines a classical limit of the Heisenberg SL(2, R) spin magnet

By construction, HN
cl
model. Evaluating HN
along the orbits of classical motion of N particles, one obtains
the energy spectrum of the quantum magnet to the leading order of the WKB expansion,
or equivalently the large-J behavior of the anomalous dimensions J (.) of conformal
operators, Eq. (3.24). According to (3.44), this behavior is controlled by the large-J scaling
of the highest integral of motion qNcl . In the WKB approach, the spectrum of qN is
obtained by imposing the BohrSommerfeld quantization conditions on the periodic orbits
of classical motion of N particles [42]. As was shown in Refs. [4447], the WKB spectrum
of anomalous dimensions derived in this manner is in agreement with the exact expressions
(3.35) and (3.36).
Finally, let us establish the relation between the roots k of the transfer matrix and
cl
the two-particle spins Jk,k+1
. Substituting tN (u) = 2 k (u k ) into (3.33) one obtains
that the roots parameterize the eigenvalues of the integrals of motion qn . At large J ,
replacing qn by their classical counterparts defined in (3.43), one finds that qncl are given by
symmetric polynomials in 1 , . . . , N of degree n (with n = 2, . . . , N ). One deduces from
(3.43) that the large-J behavior of the momenta pk pk+1 is determined by the scaling
behavior of the roots k . In particular, in the upper part of the spectrum, Eq. (3.35), one gets
cl
J for k = 1, . . . , N . In similar
from k J that pk pk+1 J , or equivalently Jk,k+1
manner, in the lower part of the spectrum, Eq. (3.36), k J 0 leads to pk pk+1 J 0
cl
and Jk,k+1
J 0 . We recall that the xk -variables entering (3.40) have the meaning of
momentum fractions carried by N particles. Then, conjugated to them pk -variables are
the light-cone coordinates of the corresponding quantum fields Xk (k ), pk k . Thus, in
the upper and the lower part of the spectrum the light-cone coordinates scale at large J as
k k+1 J and k k+1 J 0 , respectively. These properties are in agreement with
the results obtained in Section 3.2.1 within the Wilson line approach.
We notice that the anomalous dimensions of N -particle conformal operators, Eqs. (3.35)
and (3.36), were obtained using the lowest-order expression for the QCD evolution
kernels whereas Eqs. (3.15) and (3.17) hold to all orders in the coupling constant. The
above analysis suggests that at large J the higher-order corrections modify the classical
Hamiltonian (3.41) and (3.44) in the following way

2
1
cl
cl
= cusp(s )
ln Jk,k+1
= cusp (s ) ln qNcl ,
HN
2
N
(3.45)
k=1
cl
and qNcl given by the same expressions as before, Eqs. (3.40) and (3.43),
with Jk,k+1
respectively.
25
4. Cusp anomaly at weak coupling

Let us revisit the computation of the cusp anomalous dimension to the lowest order of
perturbation theory aiming on an analogy with the stringy computation of Wilson loops
within the AdS/CFT framework. As we will see momentarily, the cusp anomaly in the
weak coupling regime can be interpreted as a quantum transition amplitude for a test
particle propagating in the radial time ln r and the angular coordinate . This should be
compared with the strong coupling calculation [810], in which the same quantity is given
by a classical action function for a particle propagating on the same phase space.
4.1. Cusp anomaly in perturbation theory
To the lowest order in the coupling constant, the Wilson line expectation value is given
by

W = P exp i dx A (x)
=1+
(ig)2
2
a a
t t

dx

dy D (x y) + O g 4 ,
(4.1)
where 0|T Aa (x)Ab (y)|0 = g 2 ab D (x y) is a gluon propagator and t a t a = Nc is

the Casimir operator in the adjoint representation of the SU(Nc ). To calculate the cusp
anomaly we choose the integration contour C, see Fig. 2 (left), to be the same as for the
IsgurWise form factor, Eq. (2.3). In this way, we obtain
W (v v ) = 1

s Nc
w(v v ) w(1) + O s2 ,
(4.2)
Fig. 2. Cusp anomaly in one-loop approximation (left) and its interpretation in radial quantization formalism
(right).
26
where v and v are tangents to the integration contour in the vicinity of the cusp,
v 2 = v 2 = 1, v v = cosh , w(1) = w(v v) = w(v v ) and
0
w(v v ) =

ds
dt
0
v v
,
(vs v t)2
(4.3)
with s and t being proper times. Going over to higher orders in s , one takes into account
that the Wilson loop possesses the property of non-Abelian exponentiation [57]
s Nc k
ln W =
(4.4)
wk ,
where the weights wk receive contribution from diagrams to the kth order in s with
maximally non-Abelian color structure. In our case, the exponentiation property states that
w1 = w(v v ) w(1).
It is straightforward to perform the integration in (4.3). For our purposes, however, we
change the integration variables to r> = max(t, s) and r< = min(t, s) and apply the
identity

n
v v
(n + 1)2
1
1
n r<

Un (v v ) +
,
(4.5)
= 2
(1)
r> n(n + 2)
4
(vs v t)2
r>
n=1
where Un (cosh ) = sinh((n + 1) )/ sinh are Chebyshev polynomials of the second kind
[58]. Its substitution into (4.3) leads to

rmax

dr>
1

n n+1
Un (cosh ) +
(1)
w(v v ) = 2
n(n + 2)
4
r>
n=1
= coth ln(rmax ),
rmin
(4.6)
where rmin 1/ and rmax are ultraviolet and infrared cut-offs, respectively. Combining
together (4.2) and (4.6), one verifies that the Wilson line satisfies the evolution equation
(2.4).
Eqs. (4.5) and (4.6) have a simple interpretation within the radial quantization approach
[59,60]. In this formalism, one performs a quantization procedure using four-dimensional
(Euclidean) polar coordinates r 2 = x2 and v = x /r. This allows one to separate the
dynamics in the radial and angular coordinates by decomposing the propagators of fields
over partial waves defined as eigenstates of the operator of the angular momentum
1
L2 = l l ,
8
l = i(x x ),
(4.7)
which has the meaning of the LaplaceBeltrami operator on a sphere v2 = 1. Then, in the
radial quantization the lowest-order contribution to the cusp anomaly, Eq. (4.6), takes the
factorized form [60]

1
1
d ln r,
w(v v ) = v | 2 |v + 1
(4.8)
2
L
27
where |v denotes a point on the (hyper)sphere SO(3, 1)/SO(3) defined by the unit vector
v and additional minus sign inside v | takes into account that two tangents have
opposite orientation at the cusp. Thus at the weak coupling, the cusp anomalous dimension
is given by

s Nc
1
1
cusp ( ; s ) =
(4.9)
v | 2 |v v| 2 |v + O s2 .
2
L
L
Let us demonstrate that the matrix elements entering this expression coincide with the
propagator of a test particle on the time-like hyperboloid v2 = v02 v12 v22 v32 = 1.
4.2. Particle propagation on a sphere
To avoid complications due to the infinite volume of the (noncompact) SL(2, R) group
manifold, we will calculate the propagator of a particle on a unit sphere S 3 and perform an
analytic continuation from Euclidean to Minkowski signatures afterwards.
The transition amplitude for the particle on the S 3 sphere to go from the point v to
v is equal to the sum over all paths P connecting these two points, Fig. 3,

eiA[P ] ,
G[v, v ] =
(4.10)
P S 3
where summation goes over connected paths P of the length A[P ] on the S 3 sphere. With
an arbitrary point on the sphere, v S 3 , one can associate an element of the SU(2) group
gv = v0 + i
3
va a ,
tr[gv1
gv2 ] = 2(v1 v2 ),
1
(4.11)
a=1
with a being Pauli matrices. Eq. (4.10) defines a quantum dynamics of a particle on
the SU(2) group manifold. Introducing local (angular) coordinates on the sphere Xa =
(, , )
v0 = cos cos ,
v1 = sin sin ,
v2 = sin cos ,
v3 = cos sin ,
(4.12)
with = and =
Fig. 3. Transition amplitude on a sphere between two points v and v
2
1
2
1
(left). Classical trajectories which saturate the amplitude are multiple windings around the sphere over principles
circles, e.g., . = 2 trajectories (right), separated from each other for illustration purposes only.
28
one obtains the metric on this manifold as

2
1
ds 2 = tr gv1 dgv = d 2 sin2 d 2 cos2 d 2
2
GMN (X) dXM dXN .
(4.13)
The action of a particle has the meaning of the length of the path on this manifold
1
A[P ] =
GMN (X) XM XN
1
=

1
e( )
GMN (X) XM XN ,
4
e( )
(4.14)
where is a local coordinate on the trajectory. The two expressions coincide upon
extremizing with respect to the einbein field e( ).
Following, for instance, [61], one can express this path integral as an integral over the
1
proper time = 0 d e( ) of the matrix element of the transition operator

G[v, v ] = i
d v |ei H |v = v |
1
|v.
H
(4.15)
Here the Hamiltonian H = L2 is the LaplaceBeltrami operator on the S 3 . It defines

a quantum-mechanical model (symmetric top) with the SU(2) dynamical symmetry.
Since the Hamiltonian coincides with the Casimir operator of this group, its eigenstates
correspond to unitary irreducible SU(2) representations of spin j = 0, 12 , 1, . . .
j
HYm1 m2 (g) = j (j + 1)Ym1 m2 (g),
Ym1 m2 (g) = (2j + 1)1/2 dm1 m2 (g),
(4.16)
where dm1 m2 (g) = j, m1 |d j (g)|j, m2, the matrix elements of the spin-j group representation j m1 , m2 j , are the well-known Wigner functions [62]. The transition matrix
element can be expanded over characters of the SU(2) representation

1 j
ij (j +1)
v |ei H |v =
(2j + 1) tr d j gv
d (gv ) e
j =0, 21 ,1,...
1 ij (j +1)
(2j + 1)j gv
.
gv e
(4.17)
j =0, 21 ,1,...
Here, the SU(2) character is defined as [62]

j [gv ] =
j

m=j
dmm (gv ) =
sin(2j + 1)/2
,
sin /2
(4.18)
where the Euler angle is defined as cos(/2) = tr gv /2 with gv given by (4.11) leading
to = 2( ). Substituting (4.18) into (4.17) and making use of the second relation in
29
(4.11), one calculates the matrix element entering (4.15) as

v |ei L |v =
2
(2j + 1)eij (j +1)
j =0, 12 ,1,...
sin((2j + 1) )
,
sin
(4.19)
where = is the angle between the vectors v and v . Substituting (4.19) into
(4.15), one notices that the j = 0 term in (4.19) leads to a divergent contribution upon
integration in the right-hand side of (4.15). It does not depend, however, on the cusp angle
and cancels in the difference G[v, v ] G[v, v] leading to
s Nc
s Nc
G[v, v ] G[v, v] =
( cot 1),
(4.20)
2
where is the Euclidean cusp angle (v v ) = cos . Going over to Minkowski space, we
continue i and reproduce the correct expression for the cusp anomalous dimension,
Eq. (2.5), upon identification of the Casimirs CF = Nc .
The following comments are in order. The sum in (4.19) can be expressed, via the
Poisson summation formula, as a derivative of Jacobi theta functions, namely,

#
"
2
, q + 3
,q ,
v |q L |v = 1/4
2
2q sin
cusp ( ; s ) =
where q = ei and = . One can rewrite the same expression as

v |ei L |v = 2
2

1
2
(i )3/2
( + 2.)ei(+2.) / i/4
sin
.=
(i )1/2
sin

ei(+2.)
2 / i/4
(4.21)
.=
Eq. (4.21) has a remarkably simple physical meaning. As was shown in [6366], see also
Refs. [67,68], the semiclassical expression for the transition amplitude on the SU(2) group
manifold coincides with the exact solution, Eq. (4.21), i.e., the path integral collapses from
a sum over all paths to a sum over classical paths. In the semiclassical approach, the righthand side of (4.21) comes about as a sum over classical trajectories dressed by quadratic
fluctuations. The classical trajectories are geodesics and run along the principal circle on
the unit sphere between the two points, v and v , and wrap around this circle .-times
in the (anti-)clockwise direction depending on the sign of .. The trajectories fall into two
homotopy classes depending on the .-parity. Denoting ( ) the angular variable on this
circle, one can parameterize classical trajectories as ( ) = ( + 2.)/ with 0 .
The metric (4.13) on the classical trajectories equals
ds 2 = GMN dXM dXN = d 2
(4.22)
and the classical action (4.14) is given in the gauge e( ) = const by

Acl [P ] =
d
0

1
1
+ ( )2 = + ( + 2.)2 .
4
4
(4.23)
30
Coming back to the original sum (4.10), we conclude that the exact expression for
the transition amplitude (4.21), and as a consequence the cusp anomalous dimension
(4.20), is given by the sum over classical trajectories. This property of the path integral
is a manifestation of the DuistermaatHeckman localization phenomenon in quantum
dynamical systems on Lie groups [69].
The derivation of (4.19) was based on the identification of the unit sphere S 3 as
the SU(2) group manifold. Going over from Euclidean to Minkowski kinematics, one
has to substitute the sphere S 3 by the time-like hyperboloid H3+ , or equivalently the 3dimensional Lobachevsky space AdS3 . The appropriate group manifold is provided by
the SO(3, 1)/SO(3) coset. In distinction with the previous case, the dynamical symmetry
group is noncompact and we have to deal with quantum mechanics on the space of constant
negative curvature. The analysis goes along the same line as above with the only difference
that the SU(2) representations of spin-j have to be substituted by the unitary, continuous
representations of the SO(3, 1) group:
fundamental series: j = 1/2 + i/2 with < < ,
complementary series: 1 < j < 0.
It turns out that the resulting expression can be obtained from (4.19) through analytical
continuation in spin j . To see this, one applies the BarnesMellin transformation to rewrite
the sum over half-integer j in (4.19) as a contour integral over the complex spin that runs
parallel to the imaginary axis to the right from j = 0

(2j + 1)2 sin((2j + 1) )

dj
1 ,
G[v, v ] G[v, v] =
2i sin(2j ) j (j + 1) (2j + 1) sin
i
(4.24)
with 0 < < 1. One verifies that moving the integration contour to the right and picking
up the residues at half-integer j one reproduces the known expression for the SU(2)
propagator, Eq. (4.20). Let us now move the integration contour in (4.24) to the left parallel
to the imaginary axis until it reaches "e j = 1/2. Since the integrand has a pole at j = 0,
the deformed contour will contain an additional addendum that encircles the segment
1/2 < j < 0 on the real axis. Changing the integration variable as j = 1/2 + i/2
and going over to Minkowski kinematics, i , one finds

+i

G[v, v ] G[v, v] =
1
+
2

sin( )
d
.
2
2(1 + ) sinh() sinh
(4.25)
Here in the second integral we made use of the symmetry of the integrand under j
1 j and extended the integration to the contour , which encircles the segment [i, i]
in the anticlockwise direction. The two integrals in the right-hand side of (4.25) describe
the contribution of the fundamental and complimentary series, correspondingly. Since the
integrand in (4.25) is an odd function of , the former integral vanishes, while the latter
is given by the residue at the poles = i, or equivalently j = 0, 1. The resulting
expression for the propagator (4.25) coincides with (4.20) upon replacing i . Similar
31
to the previous case, one can expand the cusp anomalous dimension as the sum over
classical trajectories on the hyperplane v02 v12 = 1 defined by the time-like vectors v
and v in the AdS space v02 v12 v22 v32 = 1. Since the trajectories do not penetrate
into transverse (n2 , n3 )-directions, the sum will be the same if one changes the metric of
the AdS3 space from Euclidean to Lorentzian signature, v3 iv3 . It is this version of the
AdS3 space that one encounters in the strong-coupling calculation of the cusp anomaly [7,
9,10].
4.3. Analytic structure of cusp anomaly
Examining (4.20) one faces a paradox. By definition, cos = (v v ), the cusp angle is
defined up to + 2n with n an arbitrary integer. At the same, (4.20) is not invariant
under this transformation indicating that cusp ( ; s ) is a multivalued function of the cusp
angle. To understand the origin of this non-analyticity we observe that the sum in (4.19)
diverges at , or equivalently v = v . This divergence has a simple meaning in
terms of the sum over random paths on the S 3 -sphere, Eq. (4.10). For 0 < the
minimal length path connecting the points v and v is unique. For = the points
v and v are opposite poles on the sphere, the length of the minimal path equals
and the number of such paths is infinite. The same singularity has a clear meaning in
QCD in context of the heavy quark form factor. We recall that v and v are velocities
of the heavy quark before and after interaction with space-like external momentum q,
q 2 Q2 = m2 (v v )2 = 2m2 [(v v ) 1] > 0. Re-expressing the cusp anomalous
dimension (2.5) as a function of x 4m2 /Q2 = 1/ sin2 (/2),

1
1+x +1
s Nc
x
1+x 1+
,
cusp ( ; s ) =
ln
2
1+x
1+x 1
(4.26)
one finds that it has singularity at x = 1, i.e., in the non-physical point Q2 = 4m2 ,
corresponding to = . The heavy quark form factor, analytically continued from Q2 > 0
to Q2 < 0, describes the threshold creation of a pair of heavy quarks. For Q2 > 4m2 , i.e.,
above the threshold, the cusp anomalous dimension acquires an imaginary part.
4.4. Cusp anomalous dimension and 2D gauge theory
In the previous section, we identified the one-loop cusp anomaly with a transition
amplitude for a test particle on the SL(2, R) group manifold. As a next step, we will express
this amplitude in terms of a disk partition function in a two-dimensional gauge theory and
use the stringy representation of the latter. We will demonstrate that the emerging stringy
description of the cusp anomaly at weak coupling is different from the one dictated by the
NambuGoto string.
In two dimensions, the YangMills theory does not contain transverse gauge degrees
of freedom and, therefore, it can be reduced to a quantum-mechanical model. Its partition
function on an arbitrary Euclidean two-dimensional manifold of genus G with the metric
32
tensor g can be calculated through the heat kernel expansion [70]

2
2
12 d 2 x det g tr F 2
g
=
(dim R)22G eg AC2 (R)/2 , (4.27)
Z g A = DA e
R
a [A]t a is the only nontrivial component of the strength tensor F a [A] with
where F F01
generators in the fundamental representation normalized as tr(t a t b ) = ab /2. A is the

area of the target manifold . The sum in the right-hand side of (4.27) runs over unitary
representations R of the gauge group of dimension dim R and quadratic Casimir C2 (R). As
in Section 4.2, we will consider the SU(2) gauge group and perform analytical continuation
to the SL(2, R) group afterwards. In that case, C2 (R) = j (j + 1) and dim R = (2j + 1)
with j being non-negative (half)integer. Then, the sum in (4.27) has a striking resemblance
with a similar sum defining the transition amplitude (4.17).
To make the correspondence exact, one introduces the amplitude of the two-dimensional
YangMills theory on a disk, with radial coordinate x 0 , 0 x 0 T , and angular x 1 ,
0 x 1 L, of area A = LT /2, and a holonomy at its boundary C = ,

U = P exp i dx A(x) .
C
The partition function of the disk is [70,71]

$

1 d 2 x det g tr F 2

Z U ; g 2 A = DA P ei C dx A(x) , U e g2

2
=
(2j + 1)j [U ]eg Aj (j +1)/2,
(4.28)
where j [U ] is the character for the spin-j representation of the gauge group. The
path integral representation is due to Ref. [72], where the conjugation
invariant delta

function13 is defined by a group Fourier transform (U, U ) = R R [U 1 ]R [U ]. The
disk transition amplitude (4.28) is used to build the transition amplitudes for arbitrary
manifolds of genus G by a gluing procedure [71]. Eq. (4.28) reduces to the partition
function via (4.27) by Z = dU Z[U ]. Thus,

Z U ; g 2 A = 2 = v |e H |v,
(4.29)
1
c.f., Eq. (4.17), where U = gv

gv and tr[U ( )] = 2 cos with = . Thus, the oneloop cusp anomalous dimension (4.20) is given by the integral of the wave functional in
two-dimensional YangMills theory on the disc with respect to its area
s Nc
cusp ( ; s ) =
2

d Z[U ; 2 ] Z[1; 2 ] .
(4.30)
As we have already mentioned, the transition amplitude of the particle on the SU(2)
group manifold has two different representations, Eqs. (4.19) and (4.21). The first one
13 It equates the eigenvalues of two unitary matrices.
33
coincides with (4.29), while the second one is related to the saturation of the partition
function on a Riemann surface by a sum over classical saddle points in the path integral
[73] (see also [74])a consequence of the DuistermaatHeckman localization. It can be
rephrased in physical terms as instanton mechanism of confinement in two-dimensional
YangMills theory [75]. The instantons under consideration are solutions to the Yang
Mills equations of motion on the two-dimensional disk with the boundary conditions
set by the holonomy tr U [A(x 0 = T , x 1 )] = 2 cos . The classical configurations in the
A0 (x 0 , x 1 ) = 0 gauge correspond to straight paths connecting the initial and final points
and read in the topological charge-. sector A1. (x 0 , x 1 ) = x 0 (3 /2)( +2.)/A. The action
evaluated on these instanton solutions reads
2( + 2.)2
,
S[A. ] =
g2 A
and the weight factor w. = + 2. in w. exp(S. ). These properties are in a perfect
agreement with our findings in Section 4.2, Eq. (4.21).
It is well known that the two-dimensional SU(N) YangMills theory is a string
theory [76,77]. Its partition function is given by the sum of maps from two-dimensional
worldsheet to two-dimensional target manifold . The explicit relation between the
partition functions in two theories looks as follows [78]
"
#

1
1
ln Z g 2 A = Zstr gs = , =
(4.31)
N
g 2 N
with N = 2. Combining together Eqs. (4.30) and (4.31), we conclude that the one-loop
cusp anomalous dimension (4.30) admits the same stringy representation.
Eq. (4.31) is a counterpart of the relation between the transition amplitude of firstly
quantized particle and partition function of secondly quantized field theory. Namely, it
relates firstly quantized string with two-dimensional YangMills theory. According to
(4.31), the partition function in the latter theory is equal to the transition amplitude in
the string theory with the boundary conditions specified by the holonomy U ( ) on the disk
boundary. Moreover, one can establish a one-to-one correspondence between the gauge
invariant states in the Hilbert space of two-dimensional YangMills theory, tr[U n ( )] with
n = 1, 2, . . . , and stringy excitations (the oscillatory modes). The relation between the two
looks as follows [78,79]

tr U n ( ) (n + n )|),
(4.32)
where n and n satisfy the commutation relations [n , m ] = [ n , m ] = nn+m,0 and
is the stringy coherent state
[n , m ] = 0 and |)

n

n
in
in
exp
|0.
| ) = exp
e
e
n
n
n=1
n=1
The holonomy U ( ) is defined in the fundamental representation of the SU(2) group.14

14 The closed string picture for SU(N ) two-dimensional YM theory is valid perturbatively only for N .
c
c
However, the stringy representation exists for arbitrary Nc [80].
34
1
n
We recall that U ( ) = gv
gv with gv given in (4.11) so that tr[U ()] = 2 cos(n ).
n
Notice that tr[U ] and the characters j [U ] provide two linear independent bases on the
space of invariant functions on the SU(2) group and, therefore, they are related to each
other by a linear transformation [80], a form of the Frobenius character formula [81]. This
allows one to rewrite Eq. (4.29) in terms of tr[U n ] and make use of (4.32) in order to reexpress the transition amplitude for a particle on a cylinder, deduced by gluing the opposite
arcs of the boundary holonomies of the disk amplitude [71],

Z[U1 , U2 ; 2 ] = d Z U1 U2 ; 2 ,
(4.33)
in terms of a string amplitude

Z U ( ), U (0); 2 = v |e H |v = ( |e Hstr | = 0),
(4.34)
where | = 0) corresponds to a trivial holonomy U (0) = 1, thus resulting merely to a disk

U (0); 2 ] = Z[U ();
2 ]. The group theory Hamiltonian C2 (R) is not
amplitude Z[U (),
diagonal in the basis spanned by string states and involves splitting and joining of strings.
The string Hamiltonian is defined for the SU(2) group as [82]
1
Hstr = 2(L0 + L 0 ) + (L0 L 0 )2 + (V + V ),
2
where the Virasoro generator and interaction vertex operator read
L0 =
1
n n ,
2 n
V=
(4.35)
1
(nm n m + n m n+m ),
2
n,m>0
with L 0 and V given by similar expressions. Substituting 4.34 into (4.30) one could obtain
the representation for one-loop cusp anomalous dimension in terms of string propagator
#
"
s Nc
1
1
cusp ( ; s ) =
(4.36)
|0) (0|
|0) .
( |
2
Hstr
Hstr
We would like to stress that the string corresponding to the one-loop cusp anomaly is
not of the NambuGoto type [78]. At the same time, the cusp anomaly at strong coupling
is described by the NambuGoto string on the AdS background whose tension scales as
(s Nc )1/2 (see next section). Going over from weak to strong coupling regime one expects
to find the transition from the former string to the latter. The mechanism governing such
transition remains unclear. One of possible scenarios was proposed in Ref. [83]. It is based
on the identification of the two-dimensional YangMills theory (4.27), rewritten as

1
tr F 2 tr 2F + g 2 2 ,
2
g
as a topological string (for g 2 = 0) perturbed by a rigidity term [61,84]. The NambuGoto

action was conjectured to arise through the dimensional transmutation mechanism at strong
coupling.
It is worth mentioning on the relation of YangMills theory with AdS background.
Its appearance can be understood by noticing that after analytical continuation of the
35
cusp anomaly from the Euclidean to Minkowski kinematics, one has to deal with a twodimensional YangMills theory with the SL(2, R) gauge group. It is well known [85] that at
g 2 = 0 such theory is equivalent to topological JackiwTeitelboim gravity with the action

!

S = d 2 x det g R(g) ,
(4.37)
where is the dilaton field and is the cosmological constant. Solutions to the classical
equations of motion give rise to the AdS2 gravity coupled to the dilaton.
5. Cusp anomaly at strong coupling

The main goal of our previous discussion of the cusp anomaly at weak coupling was to
emphasize its quantum nature as a transition amplitude for a particle on the AdS3 space. In
this section, we will argue that at strong coupling the cusp anomaly is given by Hamilton
Jacobi action function corresponding to a classical mechanical system defined on the same
space.
According to the AdS/CFT correspondence [46], the strong coupling regime in gauge
theories is related to the supergravity limit of a string theory on the AdS5 S 5 background.
In the present discussion we are interested in operators with large angular momentum J
where the conventional (supergravity field)/(YangMills operator) correspondence is not
applicable, and one has to solve the string theory semiclassically [2,7]. For the light-cone
observables discussed here, like quark distribution functions (2.8) and light-like Wilson
loops (2.13), the full conformal QCD group SO(4, 2) is effectively reduced to its collinear
subgroup SU(1, 1). It is only the latter which acts non-trivially on the field operators
living on the light-cone. The group SL(2, R) SL(2, R) is an isometry of the AdS3 .
Therefore, applying the gauge/string correspondence, instead of the full AdS5 space it will
be enough to consider only its AdS3 subspace.
Let us remind a few elementary facts about anti-de Sitter space. The AdS3 space with
the Lorentzian signature is a hypersurface embedded in flat R2,2
X02 X12 X22 + X32 = R 2 .
(5.1)
We set its radius to be R = 1 for simplicity. The SU(1, 1) group structure becomes manifest
via the following parametrization

g=
X0 + iX3
X1 + iX2
X1 iX2
X0 iX3

=
3
Xa a ,
(5.2)
a=0
with a = {1, 1 , 2 , i3 }. To parametrize the hypersurface (5.1), we will use two different
sets of coordinates [86]. In the global (, , )-coordinates the AdS space looks like
X0 + iX3 = cosh ei ,
X1 + iX2 = sinh ei
(5.3)
and in local Poincar (u, x, t)-coordinates

X0 + X1 = u,
X2 = ux,
X3 = ut.
(5.4)
36
The transformation from one set of coordinates to the other is achieved via the map
u = cosh cos + sinh cos ,
x=
sinh sin
,
cosh cos + sinh cos
t=
cosh sin
.
cosh cos + sinh cos
(5.5)
For the discussion which follows we introduce the Rindler coordinates on the conformally
flat part R1,1 of AdS3 in Poincar parametrization
t = r cosh ,
x = r sinh .
(5.6)
The metric on the AdS space in the global coordinates reads, see, e.g., [86],
ds 2 =
1 1 2
tr g dg = cosh2 d 2 + sinh2 d 2 + d 2
2
(5.7)
and in the PoincarRindler coordinates

ds 2 =

du2

du2
+ u2 dx 2 dt 2 = 2 + u2 dr 2 + r 2 d 2 .
2
u
u
(5.8)
Here, the two sets of coordinates have different physical interpretation. In Eq. (5.7),
0 < defines the radial coordinate on the AdS space, 0 < 2 is the azimuthal
angle and < < sets up the AdS time. Notice that the latter is different from the
time variable in (5.8). In Eq. (5.8), 0 u2 < is the Liouville coordinate, t and x are the
time and the spacial coordinates, respectively, on the hyperplane in Minkowski space to
which the contour entering the definition of the Wilson loop (2.3) (see also Fig. 2) belongs
to. In the polar coordinates, choosing r = 0 at the cusp, one identifies in (5.8) as the cusp
angle. Since the relation (5.5) between two sets of the coordinates is nonlinear, the same
trajectory of a test particle on the AdS space looks differently in the (, , ) and (r, , u)
coordinates.
Let us now turn to the analysis of the cusp anomaly in the strong coupling regime. As
was mentioned in Section 2, there are two apparently different approaches to calculation
the cusp anomalous dimension within the gauge/string correspondence. One of them relies
on the relation between the large-J behavior of the anomalous dimensions of twist-two
composite operators and the cusp anomaly, Eq. (2.20). In this way, following [7], one
can calculate cusp (s ) as the energy minus spin of a long folded closed string rotating
in the AdS space in the (, , ) coordinates, Eq. (1.2). In the second approach [810],
one calculates cusp(s ) from the minimal surface of the worldsheet of an open string
propagating in the AdS space in the (r, , u) coordinates with the ends sliding along
two rays at the boundary u = with the cusp angle . In both approaches, the
result for the cusp anomaly is expressed in terms of solutions to classical equations of
motion. The main difference between the two cases is the form of the classical Hamiltonian
37
Fig. 4. The minimal surfaces swept by open strings whose ends move along the contour of the Wilson loop. The
surface of the Wilson loop in the adjoint representation (left) is a double of the Wilson loop in the fundamental
representation (right).
and underlying picture of classical motion.15 In this section, we will demonstrate the
equivalence between the two approaches.
5.1. Open string and Wilson loop
To begin with, we recall the calculation of a Wilson loop with a cusp. The NambuGoto
action for a string propagating in the AdS3 target space looks like

%
&
&
R2
2
&GMN a XM b XN &,
S =2
(5.9)
det
d
2
!
where R 2 / = g 2 Nc and GMN is the metric tensor on the AdS3 space, ds 2 =
GMN dXM dXN , Eqs. (5.7) and (5.8). The Wilson loop is defined by a classical
configuration that minimizes this action, W exp(iSmin ). The additional factor 2 in the
right-hand side of (5.9) takes into account that the Wilson loop is taken in the adjoint
representation of the SU(Nc ) and, in multi-color limit Nc , it is just the square of the
same loop in the fundamental representation. Also, the minus sign under the square-root in
(5.9) ensures that S is real in Minkowski signature.
The main contribution to the cusp anomaly comes from the vicinity of the cusp, see
Fig. 4. In the PoincarRindler (r, , u) coordinates the minimal surface can be written as
[8]
f ( )
.
u=
(5.10)
r
Choosing (1 , 2 ) = (, r) as local coordinates on the worldsheet, one finds the induced
metric as

f/(f r)
(f/f )2 + f 2
M
N
GMN a X b X =
(5.11)
,
f/(f r)
(1 f 2 )/r 2
where f = f ( )/ . Then, the action (5.9) becomes

%
rmax
dr
s Nc
Smin ( ) = 2
,
d f2 f 2 + f 4 icusp(, s ) ln
r
rmin
(5.12)
15 Another difference is that the calculation of Ref. [7] can be performed only in the AdS space with Lorentzian
3
signature, whereas in the case of the Wilson loop, Refs. [810], the result can be reproduced by an analytical
continuation from Euclidean space.
38
where rmin 1/ is a UV cut-off and the cusp anomalous dimension is given by

cusp ( ; s ) = 2
s Nc
%
d
f2 + f 2 f 4 .
(5.13)
This allows us to interpret cusp(, s ) as a classical action of a particle with the Lagrangian
L[f ] = (f2 + f 2 f 4 )1/2 , where f ( ) and play the rle of the coordinate and the
time, respectively. The energy E and momentum P ( ) of the particle take the form
L[f ]
f 2 + f 4
,
E = f
L= !
f
f2 + f 2 f 4
f
L[f ]
= !
.
P=
f
f2 + f 2 f 4
(5.14)
Being the integral of motion, E = const, the energy determines the solutions to the classical
equations of motion. The action calculated along the classical trajectory satisfies the
HamiltonJacobi equation and is given by

s Nc
cusp ( ; s ) = 2

d P ( )f( ) E .
(5.15)
Since the minimal surface ends at the boundary of the AdS space, u = , the classical
trajectories have to satisfy the boundary condition f (0) = . As was shown in Refs. [9,
10], the asymptotic behavior of the cusp anomaly at large is governed by the contribution
of classical solutions with the energy E = 1/2. In this
case, the classical trajectory starts
at infinity, f (0) = , and approaches f ( ) 1/ 2 as . Since P ( ) f( )
vanishes for , one finds from (5.15)

s Nc
,
cusp ( ; s ) =
(5.16)
in agreement with (2.17) and (1.4). The corresponding minimal surface can be translated
into the global coordinates by noticing that f 2 ( ) = X32 X22 1/2. Then, one gets from
the equation for AdS embedding (5.1),
1
1
X02 X12 = ,
X32 X22 = ,
2
2
which is the result of Ref. [9].
(5.17)
5.2. Rotating closed string

Let us now recapitulate the consideration of a rotating folded closed string in the AdS3
space around its center-of-mass following [7], see also [87]. In distinction with the previous
case, one chooses to work in the global (, , )-coordinates (5.7) and assumes that the
center of the string lies at = 0. The string action is given by the same NambuGoto
expression (5.9) but with different boundary conditions. Choosing the gauge 1 = and
2 = , one finds the induced metric as

cosh2 + 2 sinh2
M
N
GMN a X b X =
0

0
,
1
39
(5.18)
where and are the AdS time and radial coordinates, respectively, = ( ) is an
azimuthal angle of a point on the string and / the corresponding angular velocity.
As a consequence, the action of the rotating stretched string looks like
R2
S =4
2
%
d
cosh 2 ( ) sinh2
2

d L[].
(5.19)
Here the additional factor 4 counts the number of segments of the folded string rotating
around = 0 and the maximal radial coordinate 0 is determined by
coth2 2 ( ) 0.
(5.20)
Eq. (5.19) defines a classical mechanical model of a rotating rod with the Lagrangian L[].
Its energy and angular momentum are

0
s Nc
cosh2
L[]
,
d %
E =
(5.21)
L = 4

2
2 sinh2
cosh
0

0
L[]
s Nc
sinh2
J=
(5.22)
.
d %
= 4

cosh2 2 sinh2
0
Both quantities are integrals of motion so that the classical trajectories are specified by the
values of E and J . The action (5.19) evaluated along the classical trajectory is given by16

rmax
Scl = d (J E) = J (s ) ln
,
(5.23)
rmin
where max/min = ln rmax/min and
2
J (s ) =
d
(J E) = E + J
2
(5.24)
with = being the angular velocity of the rod. The additional factor 2 in the righthand side of (5.23) counts the number of end-points of the folded string. The anomalous
dimension defined in this way is the coefficient in front of the AdS time in the expression
for the action function Scl . The latter is the solution to the HamiltonJacobi equations for
the system (5.19). In the limit of long strings,

1
1
1,
= 1 + 2,
0 = ln
(5.25)
2
16 Note that the pair (J, ) defines the action-angle variables for the system under consideration.
40
for 0, one finds the energy and angular momentum of the folded string as

s Nc 1
s Nc 1
E=
J =
ln ,
+ ln .
Substituting these relations into (5.24), one obtains

s Nc
ln J.
J (s ) = 2
(5.26)
(5.27)
This expression defines the anomalous dimension of the twist-two operators O J (0) at
strong coupling.
5.3. Multi-particle operators: minimal surfaces
As we have seen in the previous sections, the anomalous dimensions of the twist-two
operators at strong coupling can be obtained using two different approaches based on
the calculation of the Wilson loop with a cusp and the classical energy of a long string
rotating on the AdS background. In this section we will generalize these results to N particle conformal operators of higher twist.
We have demonstrated in Section 3.2.1, that the anomalous dimensions of such
operators at large spins J occupy the band (3.16) whose boundaries, Eqs. (3.15) and
(3.17), are defined by the cusp anomalous dimension. Since this result holds for arbitrary
coupling constant, one makes use of (1.4) to replace cusp (s ) by its asymptotic behavior at
strong coupling. Remarkably enough, the same result can be obtained from the gauge/string
duality.
Following the approach described in Section 5.1, one has to construct the minimal
surface on the AdS3 target space whose boundary involves multiple cusps. The number
of cusps, k, varies along the band. On the upper and lower boundary it equals N and 2,
respectively. At large Nc , the expectation value of the product of Wilson loops factorizes
into the product of their expectation values (see Fig. 1)

tr W [1 , 2 ] tr W [k , 1 ] = tr W [1 , 2 ] tr W [k , 1 ] .
(5.28)
This implies that the area of the minimal surface corresponding to the Wilson loop with k
cusps is given by the sum of k elementary areas derived in Section 5.1

tr W [j , j +1 ] = exp iS(j ) + iS(j +1 )
1
2 cusp (j ;s ) 2 cusp (j+1 ;s )

= 2 (j +j+1 )cusp (s ) ,
1
(5.29)
where S(j ) = Smin (j )/2, Eq. (5.12). Substituting this relation into (5.28), we calculate
the total area of the minimal surface
S(1 , . . . , k ) =
k
S(j ) icusp(s )
j =1
= k cusp(s ) i ln
k
j ln
j =1
(5.30)
41
for 1 k 1. As before, at k = 2 and k = N the coefficient in front of

(i ln ) can be identified for ln J as the lower and the upper bounds in the spectrum of
anomalous dimensions of the N -particle conformal operators, Jmin and Jmax , respectively.
5.4. Multi-particle operators: revolving closed string
Let us turn to the picture of a rotating closed string. We remind that the anomalous
dimension of the twist-two conformal operators at strong coupling, Jtw-2 (s ), is related
to the energy of the rotating NambuGoto string evaluated on a classical configuration
with the minimal energy for a given angular momentum J 1. In the AdS background
such configuration corresponds to a folded rotating long string. It is worth mentioning
that the emerging picture is a generalization of the well-known hadronic string for the
meson states from flat to curved background, see, e.g., [88]. Attempting to extend the
GubserKlebanovPolyakov approach to N -particle conformal operators, one immediately
encounters the following difficulty. In distinction with the N = 2 case, the anomalous
dimensions occupy the band (3.16). On the stringy side, this indicates the existence of
additional stringy degrees of freedom. One expects that their total number should be N 2,
in accordance with the total number of integrals of motion q2 , . . . , qN in the Heisenberg
spin chain (both in classical and quantum cases). The spectrum of the integrals of motion is
specified by the total angular momentum J and the set of integers .1 , . . . , .N2 .17 Going
over to the strong coupling regime, the integrability properties of the evolution equations
may be lost, but the analytical structure of the energy spectrum remains intact. In other
words, for large J the anomalous dimensions of N -particle operators are parameterized
by the same set of integers, J = J (s ; .1 , . . . , .N2 ) although the explicit form of this
dependence may be different at strong and weak coupling. As we will argue below, these
additional degrees of freedom can be identified as string junctions.
Let us consider the simplest case of the N = 3 conformal operators. Similar to the
N = 2 case, we expect to recover a folded closed string rotating on the AdS background.
Its total angular momentum equals the Lorentz spin of the conformal operator. Since the
logarithmically enhanced contribution, ln J , to the energy of the string originates from
the boundary region (5.25), the large J asymptotics of the anomalous dimension depends
on how many bits of the folded string approach the boundary (see Fig. 5). The fact that
the anomalous dimensions on the upper boundary (3.15) scale as 3 ln J suggests that the
corresponding stringy configuration consists of three long bits, which are interconnected at
some point close to the center of the AdS. Similar Y-shaped configurations are well known
in QCD as describing baryonic string and following [88] we will refer to the string vertex
as the string junction. There is however a number of important differences.
Since the quarks in QCD belong to the fundamental representation of the SU(Nc )
group for Nc = 3, the color-singlet baryonic operators are built from Nc quarks and the
baryonic vertex in the corresponding Y-shaped hadronic string contains the same number
of string bits. Within the AdS/CFT framework, at large Nc , similar baryonic vertices have
17 These integers appear in the BohrSommerfeld quantization conditions imposed on the orbits of classical
motion.
42
Fig. 5. Baryon string with a junction.
been constructed in Ref. [89]. In supersymmetric theories, fermions belong to the adjoint
representation of the SU(Nc ), which allows one to construct color-singlet composite
operators containing an arbitrary number N 2 of fermions, see, e.g., Eq. (3.11). This
leads to important differences in the renormalization properties of such operators both at
the weak and strong coupling. Namely, at large Nc , the interaction between fermions in the
adjoint representation occurs only between nearest neighbours while in the fundamental
representation, due to antisymmetry of the baryonic vertex under permutation of quarks,
the interaction between any pair of quarks is allowed. In the string representation, one can
effectively replace a Wilson line in the adjoint representation by a pair of Wilson lines in
the fundamental representation running in opposite directions. Then, one can construct the
Y-shaped baryonic vertex as shown in Fig. 5. In distinction with the previous case, this
vertex contains six string bits, but, as before, we shall call it the string junction.
The Y-shaped string on the AdS background is described by the action
%
N=3
&
R2
(k)
(k)
(k)
(k) &
&,
S =2
det&GMN a XM
b XN
d1 d2
2
(5.31)
k=1
(k)
where the superscript (k) enumerates three different arms and a are local coordinates
on the worldsheet of the kth arm. Making use of the reparameterization invariance and
assigning 2(k) = 0 and 2(k) = to the folding point and string junction, respectively,
(1) (1)
one can write the boundary condition along the string junction as XM
(1 , ) =
(2) (2)
(3) (3)
XM (1 , ) = XM (1 , ). The string equations of motion corresponding to the action
(5.31) take the following form in the conformal gauge [88,90,91]

(k)
(k)
+ XM + + X(k) X(k) XM = 0,

(k)
T = X(k) X(k) = 0,
(5.32)
where (X Y ) GMN XM YN is the scalar product on the AdS space and /

with = 1 2 being light-cone coordinates on the world-sheet. Eqs. (5.32) are
invariant under reparameterization f ( ). To fix this ambiguity we identify the
local coordinate on the world-sheet as the time coordinate on the AdS space 1 = . Then,
43
the energy and the angular momentum of the string are given by
3 k ( )

R2
E=2
d2(k) G X(k) ,
2
k=1 0
3 k ( )
(k)
R2
(k)
J =2
d
G
X ,
k
2
2
(5.33)
k=1 0
where G = cosh2 (k) ( ) and G = sinh2 (k) ( ) define the AdS3 metric (5.7), while
(k) and (k) are the radial and angular AdS coordinates of the kth arm. Notice that in this
gauge the local parameters 2 corresponding to each arm take the same value at the folding
(k)
(k)
point 2 = 0 and different values at the junction 2 = k ( ). To find the explicit form
of k ( ) one has to impose the junction conditions. In covariant form they take the form
[88]
3

dX(k) (, k ( ))
M
k=1
3

(k)
(, 2 )|2 =k ( ) = 0,
+ k ( )2 XM
(5.34)
k=1
together with
(junction)
XM

(1)
(2)
(3)
( ) XM
, 1 ( ) = XM
, 2 ( ) = XM
, 3 ( ) .
(5.35)
Solving the system of equations (5.32), (5.34) and (5.35), one can find the classical
motion of the Y-shaped string on the AdS background and apply (5.33) to calculate the
corresponding energy and angular momentum.
In the flat space, for hadronic QCD string, this analysis has been carried out in Ref. [88].
In that model, the Y-shaped string describes the spectrum of baryons and the string junction
plays the rle of their additional degree of freedoms. It was found that the dependence
of the mass of baryons, m = E, on their angular momentum J takes the Regge form,
J /m2 = 2 / with depending on the classical dynamics of the junction and taking
the values within the band 2 3. The maximal value = 3 corresponds to the
configuration when the string junction is at rest and three bits of the string have the form of
the rods of the same length, rotating with the same angular velocity and forming the same
angle 2/3 against each other (recall the analogy with the interior minimal surface of
three joint soap bubbles). The minimal value = 2, corresponds to the meson-like Regge
trajectory, i.e., N = 2 in Eq. (5.31). In this case, the baryon has the diquarkquark structure,
that is the end-points of two bits are located close to each other and to the string junction.
As we will argue in a moment, similar picture emerges in the AdS geometry.
To start with, we notice that short strings rotating around the center of the AdS5 do not
feel its curvature and, therefore, look the same as strings in a flat background. That is, the
dependence of the energy, E, and the angular momentum, J , of the string on its angular
velocity is the same in two cases. The only difference is due to different representation of
the fermionsthe open hadronic string in QCD is replaced by a folded closed string in the
supersymmetric case. In the long-string limit, as was first shown in Ref. [7], a long folded
string rotating on the AdS background gives rise to the anomalous dimensions of local
44
composite operators of large spin J . According to (1.2), the anomalous dimension scales as
ln J with the prefactor depending on the number of string bits reaching the boundary (5.25).
In particular, meson-like long folded string gives rises rise to the anomalous dimension of
two-particle conformal operators, Eq. (5.27).
Generalizing this picture to the case of baryon-like folded strings, we consider the same
Y-shaped configuration as in the hadronic string with the only difference that each bit of the
fundamental string is replaced by two bits of the folded string. One can verify that such
configuration satisfies the classical equations of motion on the AdS background. Since
the string junction vertex is at rest, the energy of the rotating folded Y-shaped string is
given by the sum of energies of three arms, i.e., we can choose the gauge 1(k) = (k) and
2(k) = (k) . The same is true for the total angular momentum. Due to symmetry of the
configuration, three arms have the same energy and the angular momentum which in their
turn are equal to half of the energy and the angular momentum of the meson-like folded
string discussed in Section 5.2. As a consequence, the energy spectrum of the Y-shaped
baryonic string with the string junction at rest and the spectrum of the mesonic string are
related to each other as
3
3
JY () = JI ().
EY () = EI (),
(5.36)
2
2
For 1, in the limit of short string [7], (5.36) coincides with the known relation between
Regge trajectories of mesons and baryons described by the hadronic QCD string. For
1, in the limit of long strings, one calculates the anomalous dimension of the N = 3
particle conformal operators as
3
3
EI () JI () = Jtw-2 = 3cusp(s ) ln J. (5.37)
2
2
We observe that this expression coincides with the upper bound in the spectrum of the
anomalous dimensions of N = 3-particle operators, Eq. (3.16).
We would like to stress that (5.37) corresponds to the Y-shaped string with the string
junction at rest. In general, the classical solutions to the string equations of motion are
(junction)
parameterized by the classical trajectory of the junction, XM = XM
( ). For given
total angular momentum J the minimal classically allowed energy of the Y-shaped string
depends on the junction trajectories. It is well-known that on the flat background and
junction moving, the minimal energy of the string with the total angular momentum J = JY
is smaller than the energy EY () defined in (5.36). Obviously, the same property holds
for short strings on the AdS background. Going over to the limit of long strings, one
expects that the energy levels do not collide and, as a consequence, the same hierarchy is
preserved. In other words, the minimal energy of the Y-shaped long string with the string
junction moving is smaller than the energy with the junction at rest. This implies that the
anomalous dimensions of the corresponding N = 3-particle conformal operators is smaller
than the anomalous dimension JN=3 defined in (5.37). Moreover, the minimal energy of
the Y-shaped string for the given total angular momentum J corresponds to the diquark
quark configuration when the string junction is located near the folding point. In that case,
the energy and the angular momentum of the string approaches the energy and the angular
momentum of the mesonic string, EI () and JI (), respectively, and, as a consequence, the
JN=3 = EY () JY () =
45
Fig. 6. String picture of multi-particle operators.
anomalous dimension of the N = 3-particle conformal operator coincides with the twisttwo anomalous dimension, 2cusp(s ) ln J . These properties are in a perfect agreement
with the expression obtained before within the Wilson line approach Eq. (3.16).
We recall that at N = 3 the spectrum of the anomalous dimensions, J (.1 ), is
parameterized by integer 0 .1 J . We have demonstrated that the two extreme
classical trajectories of the junction, that is the junction at rest and rotating along the AdS
boundary, are mapped into the upper and lower boundaries of the band, .1 = 0 and .1 = J ,
respectively. We expect that similar correspondence exists for arbitrary 0 < .1 < J .
Let us now consider the N -particle conformal operators. As was shown in Section 3.2,
the spectrum of their anomalous dimensions at weak coupling is parameterized by the set of
N 2 integers .i . Going over to the strong coupling limit, we expect that the same structure
should be present. In other words, the string configuration describing such operators has
to manifest the (N 2) additional degrees of freedom. At N = 3 such degree of freedom
is provided by the string junction. For N 4 one can use the Y-shaped folded string as a
building block to construct the classical string with an arbitrary number of bits. An example
is shown in Fig. 6. Notice that the number of junctions for the string with N folding points
equals N 2. It worth mentioning that similar configurations in hadronic QCD string
describe exotic mesons and baryons [88].
A general analysis of such string configurations is rather involved, even on the flat
background. Nevertheless, the asymptotic behavior of the anomalous E J Ncusp ln J
can be easily derived by considering the limiting case when all N folding points approach
the AdS boundary. As in the N = 3 case, the (N 2) junctions are at rest so that the
energy and angular momentum of the string with N arms receives an additive contribution
from each arm. This result agrees with the upper bound in the spectrum of the N -particle
conformal operator, Eq. (3.15).
A natural question arises about the possible physical interpretation of the string junction.
If the string junction is a genuine physical degree of freedom of N -particle operators,
it should be also found at weak coupling. As we have discussed in Section 3.2.3, the
anomalous dimensions at weak coupling are identified as the eigenvalues of the oneloop QCD dilatation operator which coincides at large Lorentz spin J with spin chain
46
Hamiltonian, Eq. (3.34). In the quasiclassical approach, the anomalous dimensions are
calculated by imposing the BohrSommerfeld quantization conditions on the orbits of
classical motion of N particles on the light-cone. The latter can be significantly simplified
by going over from the original, light-cone -coordinates to the separated z-coordinates.
The HamiltonJacobi equations for the action function S0 (z) take the following form in
the separated coordinates (see [13] and references therein)
y 2 = tN (z) 4z2N
(5.38)
where y = 2zN sinh S0 (z) and tN (z) = 2zN + q2 zN2 + + qN with q2 = J 2 at large
J and qk being the higher integrals of motion. This hyperelliptic curve of genus N 2 is
a surface of equal energy for a given set of the integrals of motion qk which define the
coordinates on the moduli space of the complex structures of the Riemann surface (5.38).
Quasiclassical calculation of the anomalous dimension at weak coupling, Eq. (3.34),
amounts to the quantization of the moduli space of these complex structures.
In particular, at N = 3 the Riemann surface (5.38) corresponding to the baryonic
operator has the topology of a torus. In that case, the collective degrees of freedom live
on this surface. The BohrSommerfeld quantization conditions allow one to find quantized
values of the integral of motion q3 = q3 (J, .1 ) and calculate the corresponding anomalous
dimension J (J, .1 ). It turns out that the upper and the lower bounds in their spectrum,
J (J, .1 = 0) and J (J, .1 = J ), correspond to q32 = J 6 /27 and q3 = 0, respectively. At
these values of q3 the Riemann surface (5.38) becomes degenerate, that is one of the cycles
shrinks into a point. From the point of view of classical mechanics this corresponds to
freezing out the collective degrees of freedom. We observe that the same phenomenon
occurs at the strong coupling. Namely, the string junction is at rest at the upper bound
of the spectrum and at the diquark center-of-mass at the lower bound of the spectrum.
This suggests that the classical dynamics of the string junction is governed by yet another
Riemann surface of the same genus. Indeed, it is known that the general solutions to the
string equations of motion (5.32) are parametrized by a hyperelliptic curve of higher genus
[91]. The explicit form of this curve is fixed by the boundary conditions. In the case of the
string with N 2 junctions such conditions are given by Eqs. (5.34). It would be interesting
to compare (5.38) with the curve emerging at strong coupling.
6. Concluding remarks
The present paper was devoted to studies of the anomalous dimensions of conformal
operators at weak and strong coupling. We have demonstrated that in the both regimes
the anomalous dimensions behave asymptotically as cusp(s ) ln J at large Lorentz spin
J while the dependence of the cusp anomalous dimension on the coupling constant is
different. At weak coupling, we calculated the first two terms of perturbative expansion of
cusp(s ) in a generic gauge theory involving scalars. At large coupling, we obtained the
leading asymptotic behavior of cusp(s ) using the classical limit of string propagating on
the AdS background.
In perturbative regime, we have found the one-loop cusp anomaly corresponds, in the
radial quantization approach, to angular gluon propagation on a cylinder. This allowed us
47
to establish a relation of the former to the quantum transition amplitude of a spherical

top. Due to localization phenomena, it is saturated by classical trajectories, i.e., multiple
windings of paths around the principle circles of a sphere. These properties are naturally
incorporated into the two-dimensional gauge theory on a disk. There, the cusp anomaly is
expressed as an integral of the partition function with a boundary holonomy with respect
to the area of the disk. The well-known relation of YangMills in two dimensions to a
string theory, give us the opportunity to give a stringy representation of the cusp anomalous
dimension at weak coupling.
At strong coupling, we extended the GubserKlebanovPolyakov results for the twisttwo operator with large Lorentz spin as a rotating long closed string to the multiparticle case. The integrability of one-loop dilatation operator implies that the anomalous
dimension of N -particle conformal operator is a function of N 2 additional parameters
the conserved charges. We argued that the anomalous dimension at strong coupling has
a similar analytical structure. In the stringy picture, these new degrees of freedom are
encoded into the string junctions. Solutions to the equations of motion for the latter are
parametrized by a hyperelliptic curve, i.e., the classical dynamics of the string is driven by
a Riemann surface. This will be discussed elsewhere.
To establish the relation between the weak and the strong coupling expressions for
cusp(s ) within the gauge/string correspondence, one has to provide the explicit mapping
between the conformal operators in YangMills theory and eigenstates of the stringy
Hamiltonian on some background. Then, one can identify the anomalous dimensions of the
conformal operators for arbitrary s as the energy of the corresponding stringy excitations.
To go over from the strong coupling regime to arbitrary coupling constant in gauge theory,
one needs to know the whole spectrum of the quantum string. At present this problem
cannot be solved in full due to lack of the quantization of the strings on AdS5 S 5
background.
It is known that this difficulty can be avoided by considering the Penrose limit of
the AdS5 S 5 background. It is relevant to calculation of the anomalous dimension of
local operators in the N = 4 YM theory with large R-charge [92]. The string theory on
this background is exactly solvable and, as a consequence, the spectrum of the stringy
excitations can be found. In this case, the gauge/string correspondence looks as follows.
There are six adjoint scalars i in N = 4 theory and the R-symmetry rotates two
of them, say 1 and 2 . The stringy oscillator states are mapped into the so-called
BNM operators constructed from the complex field Z(x) = 1 (x) + i2 (x). Namely, the
operator tr[Z J (0)] with the large R-charge J 1 is dual to the ground state of the string of
the length J , |0, J , while the operator with two impurities is dual to excited oscillatory
stringy state
j
ani an |0, J
J

e2iln/J tr i Z l j Z J l ,
(6.1)
l=0
where i, j = 1, . . . , 6. The exact spectrum of the string Hamiltonian in the pp-wave

background gives rise to the anomalous dimensions of the BMN operators with large Rcharge for arbitrary coupling constant. At strong coupling they coincide with expressions
48
obtained in the quasiclassical approximation while at weak coupling they match the first
few terms of perturbative expansion [93,94].
It turned out that the quantum string on the pp-background is intrinsically related to
integrable spin chains. The latter appear when one examines renormalization of the local
operators tr[i1 (0) ik (0)] at weak coupling in the multi-color limit. These operators
mix under renormalization already at one-loop level and their eigenvalues can be found
by diagonalizing the corresponding mixing matrix [95]. As was found in [96], the oneloop mixing matrix coincides with the Hamiltonian of a completely integrable SO(6)
Heisenberg spin chain defined in an appropriate basis. The length of the spin chain is
equal to k and the spin operators belong to the fundamental representation of the SO(6)
group. The appearance of this group can be traced back to the fact that the same group is
the isometry group of the S 5 .
We observe a striking similarity between renormalization properties of such operators
and conformal operators discussed above. In both cases, the one-loop mixing matrix gives
rise to an integrable spin chain. The dynamical symmetry group of the spin chainthe
SO(6) group for the local scalar operators and the SL(2, R) group for the conformal
operatorsis dictated by the isometry of the relevant part of the background, the S 5
and the AdS parts, respectively. In spite of the fact that two spin chains are different
their energy spectrum can be obtained within the Bethe Ansatz in a similar manner by
quantizing their spectral curves. For the SL(2, R) magnet the spectral curve, Eq. (5.38),
is hyperelliptic and its genus depends on the number of fields involved. For the SO(6)
magnet the curve is more complicated and it can be reduced to hyperelliptic curve of the
genus J if one considers only its SO(3) subgroup. Having these properties in mind, one
may consider a more general case of renormalization of a local composite operator built
from an arbitrary number of scalar fields and covariant derivatives acting along different
light-cone directions, (nj D)i (0) with n2j = 0 and i = 1, . . . , 6. One might expect that
the corresponding one-loop mixing matrix is related to the spin chain with the symmetry
group SO(2, 4) SO(6), which is the isometry group of the AdS5 S 5 background.
Going over to the strong coupling regime, one should ask about the fate of integrability
of the mixing matrix. For the scalar, BMN like operators it has been suggested that
integrability holds to higher loop orders [97]. Would it be the case, the transition of the
anomalous dimensions from weak to strong coupling regime would correspond to the
flow in the space of integrable Hamiltonians with respect to the coupling constant s .
Moreover, if the same property was valid for the conformal operators, it would allow one
to calculate their anomalous dimensions at large spin J for arbitrary s . This question
certainly deserves further studies.
We would like to stress that the origin of integrability of the one-loop dilatation operator
remains obscure. A possible explanation could come from the stringy picture of the cusp
anomaly at weak coupling discussed in Section 4. We have argued that the corresponding
string picture emerges from the two-dimensional YangMills theory which in its turn is
equivalent to topological theory at g 2 = 0 with the gauge group SL(2, R). The latter theory
is the limit of the SL(2, R) ChernSimons theory at the level k . It is known that
the correlation functions of Wilson lines in the ChernSimons theory exhibit integrable
structure related to the XXZ Heisenberg spin chain with symmetry group SL(2, R) and
anisotropy q = e2i/(k2) . In this way, for k one recovers the homogeneous XXX
49
spin chain. The use of ChernSimons approach for the calculation of the anomalous
dimensions of the operators with arbitrary conformal spins and the corresponding stringy
picture behind will be discussed elsewhere.
Acknowledgements
We would like to thank I. Kogan, Yu. Makeenko, K. Zarembo for useful discussions and
M. Axenides, A. Kotikov, V. Velizhanin for correspondence on their two-loop computation
of anomalous dimensions. A.G. thanks TPI at University of Minnesota, TPI at Uppsala
University where a part of the work was done and IHES where it has been completed
for the kind hospitality. This work was supported by the US Department of Energy under
contract DE-FG02-93ER40762 (A.B.) and in part by Grants INTAS-00-334 and RFBR01-01-00549 (A.G.).
Appendix A. Cusp anomaly in dimensional reduction

The difference between two-loop expressions for the cusp anomalous dimension in the
dimensional regularization and dimensional reduction schemes, Eqs. (2.18) and (2.24),
respectively, is solely due to difference of the corresponding one-loop gluon polarization
operators. In gauge theory it receives contribution from gauge bosons, nf fundamental
fermions and ns scalars. In the momentum representation in two different schemes one
obtains (with the Feynman gauge)
dimensional regularization (DREG):

s 42 () (1 ) (2 ) 2 (d)
DREG
(q) =
q g q q
2
2 q
(4 2)
"
#
ns
Nc (5 3) 2nf (1 )
(A.1)
,
2
dimensional reduction (DRED):

s 42 () (1 ) (2 )
DRED
(q) =
2 q 2
(4 2)
#
"

ns
2 (4)
(d)
q g g Nc nf +
2
#
"
2 (4)

ns
+ q g q q Nc (5 4) 2nf (1 )
,
2
(4)
(d)
(A.2)
where g and g are metric tensors in the Minkowski spacetime of dimension 4 and
d = 4 2 with > 0, respectively, and s is a bare coupling constant. Note that in
supersymmetric theories the d-dimensional Lorentz noncovariant part vanishes in the righthand side of (A.2)
ns
= 0,
Nc nf +
(A.3)
2
50
which is easy to very at N = 1 (nf = Nc , ns = 0), N = 2 (nf = 2Nc , ns = 2Nc ) and

N = 4 (nf = 4Nc , ns = 6Nc ).
The regularized polarization operators in the DREG and DRED schemes have the same
residue at the pole in and differ by a finite O(0 )-term due to contribution of -scalars in
the DRED scheme. To renormalize the obtain expressions, we apply the modified minimal
DRED , one
subtraction procedure. In this way, subtracting the ultraviolet pole from
defines the so-called DR renormalization scheme for the coupling constant. The counterterm in the DR scheme is given by
DRED
divDR

sDR 2 (4)
ns
1
=
q g q q 5Nc 2nf
E + ln 4 .
12
2
(A.4)
One can define yet another renormalization scheme by adding a finite term to the right-hand
side of (A.4)
DRED
divMS
=

sMS 2 (4)
q g q q
12

ns
1
E + ln 4 Nc .
5Nc 2nf
2
(A.5)
In this scheme, the renormalized polarization operator RDRED = DRED div DRED
coincides with the polarization operator (A.1) renormalized within the conventional
dimensional regularization MS scheme, that is RDRED = DREG . That is the reason
MS
why one usually refers to (A.5) as the dimensional reduction MS scheme. The coupling
constants in two schemes are related to each other through the scheme transformation
sMS
= sDR

Nc sDR
DR 2
+ O s
1
.
12
(A.6)
The polarization operator modifies the gluon propagator by the term

(1)
D (x) D
(x) = i
d d q iqx ,R (q)
e
,
(2)d
q4
(A.7)
with R div . Its substitution into (4.1) yields the following contribution to the
Wilson loop evaluated along the contour shown in Fig. 2

W (1) = ig 2 t a t a
d d q v R (q)v
.
(2)d q 4 (q v)(q v )
(A.8)
(4)
(d)
Since the velocity vectors are two-dimensional, (g g )v = 0, so that the first term
in Eq. (A.2) does not contribute. Calculating this integral in the DR-scheme, one can
determine the contribution to the two-loop cusp anomalous dimension (2.24), coming from
nf fermions, ns scalars and part of the Nc term. The remaining terms Nc originate
from other two-loop Feynman diagrams. In the dimensional reduction, the only difference
51
between W (1) evaluated in the MS- and DR-schemes comes from (Nc ) term in the righthand side of (A.5). A simple evaluation using the integral

2i (m d/2 + 1) coth
v v
dd q
=
, (A.9)
d
2
2
m

(4) [q + ] (q v)(q v ) (4)d/2
(m)
2md+2
with 2 being an infrared cut-off, gives
DR 2
CF Nc
s
DRED
DRED
coth ln .
WDR
WMS
=
12
(A.10)
Notice that, by construction, W DRED = W DREG to two-loop level. At large , Eq. (A.10)
MS
MS
is translated into similar relation between the cusp anomalous dimension in two schemes,
Eqs. (2.18) and (2.24).
References
[1] A.M. Polyakov, Nucl. Phys. B 164 (1980) 171.
[2] J.M. Maldacena, Phys. Rev. Lett. 80 (1998) 4859;
S.J. Rey, J. Yee, Eur. Phys. J. C 22 (1998) 379.
[3] J.K. Erickson, G.W. Semenoff, K. Zarembo, Nucl. Phys. B 582 (2000) 155.
[4] J.M. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231.
[5] S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Phys. Lett. B 428 (1998) 105.
[6] E. Witten, Adv. Theor. Math. Phys. 2 (1998) 253.
[7] S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Nucl. Phys. B 636 (2002) 99.
[8] N. Drukker, D.J. Gross, H. Ooguri, Phys. Rev. D 60 (1999) 125006.
[9] M. Kruczenski, JHEP 0212 (2002) 024.
[10] Yu.M. Makeenko, JHEP 0301 (2003) 007.
[11] R.A. Janik, R. Peschanski, Nucl. Phys. B 565 (2000) 193;
R.A. Janik, R. Peschanski, Nucl. Phys. B 586 (2000) 163;
R.A. Janik, R. Peschanski, Nucl. Phys. B 625 (2002) 279;
R.A. Janik, Phys. Lett. B 500 (2001) 118;
M. Rho, S.J. Sin, I. Zahed, Phys. Lett. B 466 (1999) 199.
[12] J. Polchinski, L. Susskind, String theory and the size of hadrons, in: What Comes Beyond the Standard
Model, Vol. 2, 2001, p. 105, hep-th/0112204;
J. Polchinski, M.J. Strassler, Phys. Rev. Lett. 88 (2002) 031601;
J. Polchinski, M.J. Strassler, Deep-inelastic scattering and gauge/string duality, hep-th/0209211;
S.B. Giddings, High-energy QCD scattering, the shape of gravity on an IR brane, and the Froissart bound,
hep-th/0203004;
R.C. Brower, C.I. Tan, Hard scattering in the M-theory dual for QCD string, hep-th/0207144;
H. Boschi-Filho, N.R.F. Braga, QCD/string holographic mapping and high-energy scattering amplitudes,
hep-th/0207071.
[13] A. Gorsky, I.I. Kogan, G.P. Korchemsky, JHEP 0205 (2002) 053.
[14] A.M. Polyakov, Nucl. Phys. B (Proc. Suppl.) 68 (1998) 1;
A.M. Polyakov, Int. J. Mod. Phys. 16 (2001) 4511.
[15] G.P. Korchemsky, Mod. Phys. Lett. A 4 (1989) 1257;
G.P. Korchemsky, G. Marchesini, Nucl. Phys. B 406 (1993) 225.
[16] N. Isgur, M. Wise, Phys. Lett. B 237 (1990) 527;
M.A. Shifman, M.B. Voloshin, Sov. J. Nucl. Phys. 45 (1987) 292.
[17] S.V. Ivanov, G.P. Korchemsky, A.V. Radyushkin, Sov. J. Nucl. Phys. 44 (1986) 145;
G.P. Korchemsky, A.V. Radyushkin, Phys. Lett. B 279 (1992) 359.
52
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
[50]
[51]
A.M. Polyakov, Mod. Phys. Lett. A 3 (1988) 325.

A.V. Radyushkin, Phys. Lett. B 271 (1991) 218.
G.P. Korchemsky, A.V. Radyushkin, Nucl. Phys. B 283 (1987) 342.
J.C. Collins, D.E. Soper, Nucl. Phys. B 194 (1982) 445.
A.V. Belitsky, X. Ji, F. Yuan, Nucl. Phys. B 656 (2003) 165.
A.P. Bukhvostov, G.V. Frolov, E.A. Kuraev, L.N. Lipatov, Nucl. Phys. B 258 (1985) 601.
A.V. Kotikov, L.N. Lipatov, DGLAP and BFKL equations in the N = 4 supersymmetric theory, hepph/0208110;
A.V. Kotikov, L.N. Lipatov, V.N. Velizhanin, Phys. Lett. B 557 (2003) 114.
W. Siegel, Phys. Lett. B 84 (1979) 193.
D.M. Capper, D.R.T. Jones, P. van Nieuwenhuizen, Nucl. Phys. B 167 (1980) 479.
W. Siegel, Phys. Lett. B 94 (1980) 37;
L.V. Avdeev, G.I. Chochia, A.A. Vladimirov, Phys. Lett. B 105 (1981) 272;
L.V. Avdeev, A.A. Vladimirov, Nucl. Phys. B 219 (1983) 262.
G. Altarelli, G. Curci, G. Martinelli, S. Petrarca, Nucl. Phys. B 187 (1981) 461.
G.A. Schuler, S. Sakakibara, J.G. Krner, Phys. Lett. B 194 (1987) 125.
M. Axenides, E. Floratos, A. Kehagias, Scaling violations in YangMills theories and strings in AdS5 , hepth/0210091.
G.P. Lepage, S.J. Brodsky, Phys. Rev. D 22 (1980) 2157.
R.L. Jaffe, Spin, twist and hadron stricture in deep-inelastic processes, in: F. Lenz (Ed.), Lectures on QCD,
in: Lecture in Notes Physics, Vol. 496, Springer, 1997, p. 179, hep-ph/9602236;
A.V. Belitsky, Leading order analysis of the twist-three spacelike and timelike cut vertices in QCD, in:
V.A. Gordeev (Ed.), Proceedings of the 31st PNPI Winter School on Nuclear and Particle Physics, St.
Petersburg, 1997, p. 369, hep-ph/9703432;
J. Kodaira, T. Tanaka, Prog. Theor. Phys. 101 (1999) 191.
A.V. Efremov, A.V. Radyushkin, Phys. Lett. B 94 (1980) 245.
S.J. Brodsky, G.P. Lepage, Phys. Lett. B 87 (1979) 359;
S.J. Brodsky, Y. Frishman, G.P. Lepage, C. Sachradja, Phys. Lett. B 91 (1980) 239;
S.J. Brodsky, P. Damgaard, Y. Frishman, G.P. Lepage, Phys. Rev. D 33 (1986) 1881.
Yu.M. Makeenko, Sov. J. Nucl. Phys. 33 (1981) 440.
Th. Ohrndorf, Nucl. Phys. B 198 (1982) 26.
D. Mller, Phys. Rev. D 49 (1994) 2525;
A.V. Belitsky, D. Mller, Nucl. Phys. B 527 (1998) 207;
A.V. Belitsky, D. Mller, Nucl. Phys. B 537 (1999) 397.
A.S. Gorsky, Sov. J. Nucl. Phys. 50 (1989) 498.
S. Ferrara, R. Gatto, A.F. Grillo, Conformal Algebra in SpaceTime and Operator Product Expansion, in:
Springer Tracts in Modern Physics, Vol. 67, Springer, 1973.
I.I. Balitsky, V.M. Braun, Nucl. Phys. B 311 (1989) 541.
D. Mller, Phys. Rev. D 58 (1998) 054005;
A.V. Belitsky, D. Mller, Phys. Lett. B 417 (1998) 129;
A.V. Belitsky, A. Schfer, Nucl. Phys. B 527 (1998) 235.
G.P. Korchemsky, Nucl. Phys. B 462 (1995) 333.
V.M. Braun, S.E. Derkachov, A.N. Manashov, Phys. Rev. Lett. 81 (1998) 2020.
V.M. Braun, S.E. Derkachov, G.P. Korchemsky, A.N. Manashov, Nucl. Phys. B 553 (1999) 355.
A.V. Belitsky, Phys. Lett. B 453 (1999) 59;
A.V. Belitsky, Nucl. Phys. B 558 (1999) 259.
A.V. Belitsky, Nucl. Phys. B 574 (2000) 407.
S.E. Derkachov, G.P. Korchemsky, A.N. Manashov, Nucl. Phys. B 566 (2000) 203.
V.M. Braun, G.P. Korchemsky, A.N. Manashov, Nucl. Phys. B 603 (2001) 69;
V.M. Braun, G.P. Korchemsky, A.N. Manashov, Phys. Lett. B 476 (2000) 455.
L.N. Lipatov, JETP Lett. 59 (1994) 596;
L.D. Faddeev, G.P. Korchemsky, Phys. Lett. B 342 (1995) 311.
A.S. Gorsky, I.M. Krichever, A. Marshakov, A. Mironov, A. Morozov, Phys. Lett. B 355 (1995) 466.
E.K. Sklyanin, L.A. Takhtajan, L.D. Faddeev, Theor. Math. Phys. 40 (1979) 688.
53
[52] L.D. Faddeev, in: S.P. Novikov (Ed.), Mathematical Physics Reviews, Vol. 1, Harwood Academic, 1980,
p. 107.
[53] L.D. Faddeev, L.A. Takhtajan, Russian Math. Surveys 34 (1979) 11.
[54] P.P. Kulish, N.Yu. Reshetikhin, E.K. Sklyanin, Lett. Math. Phys. 5 (1981) 393.
[55] V.O. Tarasov, L.A. Takhtajan, L.D. Faddeev, Theor. Math. Phys. 57 (1983) 1059.
[56] L.D. Faddeev, Int. J. Mod. Phys. A 10 (1995) 1845;
L.D. Faddeev, How algebraic Bethe ansatz works for integrable models, hep-th/9605187.
[57] G. Sterman, AIP Conf. Proc. 74 (1981) 22;
J.G.M. Gatheral, Phys. Lett. B 133 (1983) 90;
J. Frenkel, J.C. Taylor, Nucl. Phys. B 246 (1984) 231.
[58] H. Bateman, A. Erdlyi, Higher Transcendental Functions, Vol. 2, McGraw-Hill, New York, 1953.
[59] S. Fubini, A. Hansson, R. Jackiw, Phys. Rev. D 7 (1973) 1732.
[60] J. Galayda, Phys. Rev. D 29 (1984) 1175;
C. Lovelace, Phys. Lett. B 271 (1991) 213.
[61] A.M. Polyakov, Gauge Fields and Strings, in: Contemporary Concepts in Physics, Vol. 3, Harwood
Academic, Chur, 1987.
[62] N.J. Vilenkin, Special Functions and the Theory of Group Representations, in: Translation of Mathematical
Monographs, Vol. 22, American Mathematical Society, Providence, 1968.
[63] L. Schulman, Phys. Rev. 176 (1968) 1558.
[64] J.S. Dowker, Ann. Phys. 62 (1971) 361.
[65] M.S. Marinov, M.V. Terentev, Fortschr. Phys. 27 (1979) 511;
N. Krausz, M.S. Marinov, J. Math. Phys. 41 (2000) 5180.
[66] P. Menotti, E. Onofri, Nucl. Phys. B 190 (1981) 288.
[67] I.H. Duru, Phys. Rev. D 30 (1984) 2121.
[68] M. Bohm, G. Junker, Phys. Lett. A 117 (1986) 375;
M. Bohm, G. Junker, J. Math. Phys. 28 (1987) 1978.
[69] J.J. Duistermaat, G.J. Heckman, Invent. Math. 69 (1982) 259;
J.J. Duistermaat, G.J. Heckman, Invent. Math. 72 (1983) 153.
[70] A.A. Migdal, Sov. Phys. JETP 42 (1975) 413.
[71] B.E. Rusakov, Int. J. Mod. Phys. A 5 (1990) 693.
[72] M. Blau, G. Thompson, Int. J. Mod. Phys. A 7 (1992) 3781.
[73] E. Witten, Commun. Math. Phys. 141 (1991) 153;
E. Witten, J. Geom. Phys. 9 (1992) 303.
[74] J.A. Minahan, A.P. Polychronakos, Nucl. Phys. B 422 (1994) 172.
[75] D.J. Gross, A. Matytsin, Nucl. Phys. B 429 (1994) 50.
[76] D.J. Gross, Nucl. Phys. B 400 (1993) 161.
[77] D.J. Gross, W.I. Taylor, Nucl. Phys. B 400 (1993) 181.
[78] S. Cordes, G.W. Moore, S. Ramgoolam, Nucl. Phys. B (Proc. Suppl.) 41 (1995) 184.
[79] M.R. Douglas, Conformal field theory techniques in large-N YangMills theories, hep-th/9311130.
[80] J. Baez, W.I. Taylor, Nucl. Phys. B 426 (1994) 53.
[81] W. Fulton, J. Harris, Representation Theory: A First Course, in: Graduate Texts in Mathematics, Vol. 129,
Springer, New York, 1991.
[82] J.A. Minahan, A.P. Polychronakos, Phys. Lett. B 312 (1993) 155.
[83] P. Horava, Nucl. Phys. B 463 (1996) 238;
P. Horava, JHEP 9901 (1999) 016.
[84] A.M. Polyakov, Nucl. Phys. B 268 (1986) 406;
A.M. Polyakov, A few projects in string theory, hep-th/9304146.
[85] T. Fukuyama, K. Kamimura, Phys. Lett. B 160 (1985) 259;
K. Isler, C.A. Trugenberger, Phys. Rev. Lett. 63 (1989) 834;
A.H. Chamseddine, D. Wyler, Nucl. Phys. B 340 (1990) 595.
[86] O. Aharony, S.S. Gubser, J.M. Maldacena, H. Ooguri, Y. Oz, Phys. Rep. 323 (2000) 183.
[87] S. Frolov, A.A. Tseytlin, JHEP 0206 (2002) 007;
J.G. Russo, JHEP 0206 (2002) 0038;
G. Mandal, N.V. Suryanarayana, S.R. Wadia, Phys. Lett. B 543 (2002) 81;
54
[88]
[89]
[90]
[91]
[92]
[93]
[94]
[95]
[96]
[97]
J.A. Minahan, Nucl. Phys. B 648 (2003) 203;

A. Loewy, Y. Oz, Phys. Lett. B 557 (2003) 253.
X. Artru, Nucl. Phys. B 85 (1975) 442;
P.A. Collins, J.F.L. Hopkinson, R.W. Tucker, Nucl. Phys. B 100 (1975) 157;
M.S. Plyushchai, G.P. Pronko, A.V. Razumov, Theor. Math. Phys. 63 (1985) 389.
E. Witten, JHEP 9807 (1998) 006.
H.J. de Vega, N. Sanchez, Lectures on string theory in curved spacetimes, hep-th/9512074.
I.M. Krichever, Functional Anal. Appl. 28 (1994) 21.
D. Berenstein, J.M. Maldacena, H. Nastase, JHEP 0204 (2002) 013.
D.J. Gross, A. Mikhailov, R. Roiban, Ann. Phys. 301 (2002) 31.
A. Santambrogio, D. Zanon, Phys. Lett. B 545 (2002) 425.
N. Beisert, C. Kristjansen, J. Plefka, G.W. Semenoff, M. Staudacher, Nucl. Phys. B 650 (2003) 125;
N.R. Constable, D.Z. Freedman, M. Headrick, S. Minwalla, JHEP 0210 (2002) 068.
J.A. Minahan, K. Zarembo, The Bethe ansatz for N = 4 super-YangMills, hep-th/0212208.
N. Beisert, C. Kristjansen, M. Staudacher, The dilatation operator of N = 4 super-YangMills theory, hepth/0303060.

Massive IIA string theory and Matrix theory

compactification
David A. Lowe, Horatiu Nastase, Sanjaye Ramgoolam
Brown University, Providence, RI, 02912, USA
Received 19 April 2003; accepted 20 June 2003
Abstract
We propose a Matrix theory approach to Romans massive Type IIA supergravity. It is obtained
by applying the procedure of Matrix theory compactifications to Hulls proposal of the massive Type
IIA string theory as M-theory on a twisted torus. The resulting Matrix theory is a super-YangMills
theory on large N three-branes with a space-dependent noncommutativity parameter, which is also
independently derived by a T-duality approach. We give evidence showing that the energies of a
class of physical excitations of the super-YangMills theory show the correct symmetry expected
from massive Type IIA string theory in a lightcone quantization.
PACS: 11.25.-w; 11.25.Mj
1. Introduction
Sometime ago Romans [1] found a massive deformation of ten-dimensional Type IIA
supergravity. This ten-dimensional theory has remained something of a mystery from the
string theory viewpoint. Polchinski [2] argued this supergravity theory should lift to a
massive Type IIA string theory, corresponding to ordinary Type IIA string theory in the
background of a constant 10-form RamondRamond field strength.
The problem of lifting this theory into M-theory has been considered by a number of
authors, including Hull [3]. His proposal is similar in spirit to the idea of [4,5] of obtaining
ten-dimensional Type IIB by compactifying M-theory on a 2-torus of vanishing size but
fixed complex structure. Instead one considers M-theory compactified on a 2-torus bundle
E-mail address: lowe@het.brown.edu (D.A. Lowe).
doi:10.1016/S0550-3213(03)00547-9
56
D.A. Lowe et al. / Nuclear Physics B 667 (2003) 5589
over S 1 , B(A, R), and takes the limit of zero size. We review this construction in detail in
Section 2.
We propose a nonperturbative formulation of M-theory in this background using Matrix
theory techniques. In Section 3 we generalize the construction of Seiberg and Sen [6,7],
which provides us with a formulation of the discretized light-cone quantization (DLCQ)
of M-theory on the twisted torus. The end result is a decoupled system of D3-branes in a
background B field, which we represent as a noncommutative YangMills theory with 8
linearly realized supercharges. An important novel feature of our construction is the spacedependent noncommutativity parameter .
In Section 4, we construct the noncommutative YangMills degrees of freedom directly
by compactifying an infinite system of D0-branes on the twisted torus, following the
general Matrix theory [8,9] procedure of [8,1013]. Since we do not have the full zerobrane action in the original curved background we cannot proceed to derive the superYangMills theory as in the commutative case or the case with constant . Nevertheless,
we obtain some useful information about the nature of fields in the theory. Concretely, we
give a construction of the covariant derivatives acting on an appropriate space of fields, and
obeying the compactification constraints of the twisted torus.
In Section 5, we take advantage of known results about the star products in the
presence of space-dependent noncommutativity and the result of Section 3 concerning the
emergence of space-dependent noncommutative YangMills in a generalized SenSeiberg
limit in order to elaborate on the form of the action.
The spectrum of states for a D8-brane background of massive Type IIA is examined in
Section 6 and we provide evidence for an SO(7)-invariant spectrum of states, as expected
for DLCQ string theory in this background. This provides further evidence supporting the
Matrix formulation of the DLCQ string theory. In Section 7 we consider a holographic dual
spacetime to the noncommutative gauge theory, generalizing [14,15], and we end with
conclusions and discussion in Section 8. We comment on the extension of these Matrix
compactification methods to other massive reductions of M-theory which admit de Sitter
space solutions.
2. Review of Hulls duality

It is unknown how to lift the massive Type IIA string theory [3] and its D8 background
solution directly into M-theory. While M-theory does not seem to admit a cosmological
constant, a direct lifting of Romans massive ten-dimensional supergravity [1] would
yield an eleven-dimensional cosmological constant. One possible way around this is to
obtain the ten-dimensional mass via a generalized ScherkSchwarz reduction on a circle.
The standard implementation of such a reduction requires a global symmetry in eleven
dimensions. The action of the eleven-dimensional supergravity does not have such a
symmetry but the equations of motion do have a scaling symmetry, which was exploited
in [16] to reduce to a massive ten-dimensional supergravity. However, one obtains not
Romans massive supergravity but a different supergravity in ten dimensions. That massive
supergravity can also be obtained as a usual reduction of a modified M-theory [17].
57
Hull [3] was able to embed the massive supergravity [1] and the D8 background in
M-theory by introducing two extra T-dualities, one of which is a massive T-duality as
defined in [18]. Let us describe this in detail. ScherkSchwarz reduction is a mechanism
for generating masses by compactification in the presence of a global symmetry g(),
by an ansatz

(x , y) = gy (x ) .
(1)
For the simplest case, of a U (1) invariance, we can write (x, y) = e2iqmy (x), and
obtain a mass qm for (x).
In [18] it was shown that the ScherkSchwarz reduction of 10d IIB supergravity, using
a U (1) subgroup of the SL(2, R) global invariance is T-dual to massive IIA supergravity,
using a modified set of massive T-duality rules. The reduction is given by

1 my/R
gy =
(2)
0
1
which implies
my
(3)
.
R
The monodromy (obtained for y = R) must be a symmetry of the full quantum theory,
that is it must be an element of SL(2, Z), which implies that m must be an integer. Then
this compactification is mapped by massive T-duality into the usual compactification of the
massive IIA supergravity in ten dimensions.
On the other hand, a Type IIB compactification on S 1 with nontrivial (x, y) is
equivalent to M-theory compactified on a space B which is a T 2 bundle over S 1 , where
the T 2 has modulus (x, y) fixed and area A 0. Equivalently, it is an F-theory
compactification on B where A is fixed.
We consider ten-dimensional massive Type IIA string theory, so the T-dual (Type IIA)
radius must go to infinity, hence the IIB radius R goes to zero. If we also impose that
(x) = 0 = iR2 /R1 , then massive IIA supergravity is equivalent to M-theory on the space
B(A, R), in the limit A 0, R 0. The metric is (renaming y as x3 and R as R3 )
2
A
dsB2 = R32 (dx3 )2 +
dx1 + (x3 ) dx2
Im( )
(x, y) a + ie = (x) +
= R32 (dx3 )2 + R22 (dx2 )2 + R12 (dx1 + mx3 dx2 )2 ,
(4)
with all the radii going to zero, and the xi with periodicity 1, xi xi + 1. In the limit, we
should keep the massive IIA quantities fixed, so
gsA
ls
R1 ls
=
=
= fixed,
Im(0 )R3 R2 R3
3/2
ls =
lP
1/2
R1
= fixed,
m fixed.
(5)
A comment is in order regarding the quantization of the 10d IIA mass m and the massive
T-duality. The relevant terms in the string frame supergravity actions are, for IIA

1
2
2 +
g e
R+M
SIIA = 2
k10
58

2 2
1
2(0 )
+ ,
g e
R + gsA M
(6)
2
k
whereas on the IIB side we have, similarly,

2
1
2(0 )
SIIB = 2
(7)
g e
R + gsB ( a)2 + .
k
the supergravity mass parameter, is quantized in units of ls , and it remains so when we
M,
3
reduce to 9d, whereas on the IIB side, a = a0 + mx
R3 , so the string frame masses are indeed
equal
=
gsA m mgsB
=
= mB
(8)
(9) .
ls
R3
When talking about a duality, we have to specify the background as well. The question
is nontrivial, as the massive supergravity does not admit a Minkowski background, not
even a maximally supersymmetric one. It does admit a half supersymmetric background,
namely, the D8 brane solution.
2 is the (8 + 1)-dimensional Minkowski
The D8 has the string metric and dilaton (d8,1
metric)
2
+ H 1/2 dx 2,
ds 2 = H 1/2 d8,1
A
mA
(9) M = gs M =
e = H 5/4,
m

H = c + |M||x|
= c + |x|,
(9)
ls
where c is an arbitrary constant of integration or (by the usual rescaling for p-branes)
2
+ H 1/2 d x 2,
ds 2 = H 1/2 d 8,1
e = e0 H 5/4 = gs H 5/4 ,
H
gs m
x|
H =
(10)
= 1 + gs |M||
=1+
|x|,
c
ls
where gs is defined as the coupling constant at the position of the D8 brane. The solution
to a field M(x)

is obtained by promoting M
and dualizing it to a 10-form field strength
i1 ...i10 dx i1 dx i10 . Then M(x)

= H , so the mass is piecewise constant,
F(10) = M'
and jumps at the positions of the D8 branes. The in the mass corresponds to D8 branes vs.
is the field strength for D8s), so for a D8 the supergravity
anti-D8 branes (since F10 = M
mass jumps by a positive amount, whereas for an anti-D8 by a negative amount. Note
though that the tension of both is positive (the metric is the same for both).
On a compact space we should think of the D8s as being part of a D8O8 system,
with 16 D8s canceling the charge of the orientifold 8plane O8. If the transverse space
is noncompact we can assume that the O8 and the rest of the D8s are far away, and
concentrate on the local physics of single, or coincident D8s.
We now have to find the M-theory dual of the D8 solution. Dimensionally reducing to
eight dimensions, one finds a 6-brane solution (domain wall), which can be oxidized on
the space B(A, R) to the Ricci-flat M-theory background

2
2
+ H 1/2 dx 2 + dsB2 = d6,1
+ H dx 2 + dsB2 ,
ds 2 = H 1/2 H 1/2 d6,1
(11)
59
where the moduli parameters of dsB2 are

R3 = H 1/2 ,
= mx3 + iH.
(12)
Equivalently, introducing the constants ri , we define
R1 = r1 / H ,
R2 = r2 H ,
R3 = r3 H ,
(13)
and the limit becomes ri 0 with

r1 ls
= fixed,
r2 r3
and ls fixed.
(14)
Counting parameters, we find 5 parameters in the M-theory compactification, lP , Ri and

m. This limit sends 2 parameters to zero (e.g., A = R1 R2 0 then R3 0), so that we
are left with the 3 parameters of massive IIA, ls , gs and m.
3. Matrix theory description in D8 background and T-duality approach

Hulls prescription tells us how to relate massive IIA string theory to M-theory. In this
section we construct a Matrix description of the M-theory compactified on B(A, R).
The problem is nontrivial for two reasons. The first is that the space is curved, and
moreover, if we write
ds 2 = (dz3 )2 + (dz1 + z3 dz2 )2 + (dz2 )2 ,
(15)
we find the Ricci tensor components1

2
3 z3
,
R12 =
,
2
2

2
1 2 (z3 )2 ,
R22 =
2
The curvature scalar is
R11 =
R=
R33 =
2
.
2
2
.
2
(18)
(19)
1 We have g 11 = 1 + 2 (z )2 , g 22 = 1 and g 12 = z3 , and

3
3 = ,
12
2
3 = 2 z ,
22
3
2 z3
,
2
2 = ,
13
2
1 =
13

1 2 (z3 )2 ,
2
2
2 = z3 ,
23
2
1 =
23
(16)
and all the rest are zero. Then we use

c 1 ln g + 1 c ln g c d
Rab = c ab
a b
c
ad cb
2
2 ab
and the fact that g = 1.
(17)
60
The metric (15) is invariant under the following isometries

T1 :
z1 z1 + a1 ,
z2 z2 ,
z3 z3 ,
T2 :
z2 z2 + a2 ,
z1 z1 ,
z3 z3 ,
T3 :
z3 z3 + a3 ,
z1 z1 z2 a3 ,
z2 z2 ,
(20)
with Killing vectors V1 = 1 , V2 = 2 and V3 = 3 z2 1 . We also note that [T2 , T3 ] = 0.

By identifying under the isometries with ai = Ri we obtain the space B(A, R), and then
we have (since zi = Ri xi )
mR1
= M.
R2 R3
We note, therefore, that we can trust supergravity as long as
=
lP =
mR1 lP
1,
R2 R3
(21)
(22)
which is true in our limit ( RR2 R1 3 fixed, lP 0).

In the following we choose to work with the D8 background, corresponding to the
M-theory metric (11) with radii (13). We propose a Matrix description is obtained by
considering the action of N D0-branes in the D8 background (11). Since the radii of B
go to zero, we have to make T-dualities in the 3 directions of B, and so the Matrix model
describing massive IIA will be the action of N D3 branes in the T-dual background. It is
understood that the general procedure used to obtain the dual Matrix model will be the
same for any massive IIA background with a light-like symmetry.
First, however, we must define correctly the limit taken on the M-theory, and see what
kind of limit we obtain for the D3 brane. This is described in detail in the appendix, but
we will give here only the relevant facts. Sen [7] and Seiberg [6] give a prescription for
the discretized light-cone quantization (DLCQ) of M-theory (with light-like radius R and
-theory with lP 0 and
finite lP ) on a torus of finite radii Ri . One goes to an equivalent M
spacelike 11th direction of radius Rs 0 and compactification radii R i 0 such that
Rs
R
= 2,
l 2
lP
P
Ri
R i
= ,
lP
lP
(23)
are held fixed in the lP 0 limit. Then one makes T-dualities in the compact directions
and gets a decoupled theory of Dp-branes (D3 in our case) with finite gYM and dual radii
l3
l2
R i = s = P ,
Ri R
R i
2
gYM
= gs =
R3
Ri .
lP6 i
(24)
As we can see, the parameters of the dual D3 matrix model do not depend on the
-theory, which was introduced just to prove the duality. Therefore,
parameters of the M
we can apply another limit to this construction (independent of the SenSeiberg lP 0
limit), namely lP 0 and Ri 0, with lP3 /R1 = (lsA )2 = ls2 and R1 /(R2 R3 ) kept fixed.
We also need to make a 911 flip, namely, to reinterpret the lightcone coordinate R as
the 11th direction (since in the M-theory construction of massive IIA R1 takes the role of
11th coordinate).
61
The parameters of the super-YangMills are

l2
R 1 = s ,
R
R1
R 2 =
R1 ,
R2
R1
R1 ,
R 3 =
R3
and the inverse relations are, if R = Nls ,

N
R 2 R 3 ,
R1 =
R = Nls ,
gYM

A
2 R2 R3 ,
gs = gYM
ls = N R 1 .
R 2
2
gs = gYM
=
ls gsA
,
R1
(25)
(26)
In order to still have decoupling of the string theory from the D3 brane theory, we need
-theory,
to have ls 0 and the S-dual string length gs ls2 0, which is satisfied in the M
since
ls2 =
lP3 lP
l 3 g A lP
g s ls2 = s s
0.
R lP
R lP
(27)
Let us now follow this procedure in order to find the Matrix model description of the
-theory, dimensional reduction to string theory,
background (11): 911 flip, going to M
-theory is
followed by 3 T-dualities. The string theory background in the M
2
+ H dx 2 + dsB2 ,
ds 2 = d5,1
(28)
with the radii given in (13) and constant dilaton 0 , and now we need to perform 3
T-dualities. We will concentrate on the space B, with metric
r 2

dsB2 = H r32 dx32 + r22 dx22 + 1 (dx1 + mx3 dx2 )2 ,
(29)
H
and work with string metrics, on which the T-dualities act in a simple way. We will work
in units of ls . If we want to restore the ls dependence we can formally put ri ri /ls ,
xi xi ls , m m/ls .
The Buscher T-duality rules [1921] are
g0i g0j B0i B0j
1
B0i
,
g0i =
,
gij = gij
,
g00
g00
g00
g0i B0j B0i g0j
g0i
,
B ij = Bij +
,
B 0i =
g00
g00
1
= log(g00 ).
(30)
2
Here the coordinate 0 of the T-duality is defined such that 0 is the Killing vector of
an isometry. It is worth noting here that one might be worried that we have to use the
massive T-duality rules at some point, however the m-dependent terms are only in the
transformation rules of the RR fields (see [18]).
As we saw, we have 3 isometries, T1 , T2 , T3 . T2 and T3 do not commute, so the order
of T-dualities matters. We will choose to do T1 , then T2 , then T3 . We begin by considering
g00 =
62
the simpler case of T-dualities on the twisted torus with the radii and ls , H set to 1
corresponding to the core of the D8 background x = 0, and later we will generalize this to
the complete background. After T1 we have:
ds 2 = dx32 + dx22 + dx12 ,
B12 = mx3 H123 = m,
e = e0 .
(31)
After T2 we have

ds 2 = dx32 + dx12 + (dx2 + mx3 dx1 )2 ,
e = e0 .
(32)
T3 is generated by the vector V3 = 3 mx1 2 . Transforming to coordinates

x3 = x3 ,
x2 = x2 + mx1 x3 ,
(33)
implies V3 = 3 so that we can apply the usual T-duality rules.

The metric in the new coordinates (after dropping primes on coordinates)

ds 2 = dx32 + dx12 + (dx2 mx1 dx3 )2 ,
e = e0 .
(34)
After the third T-duality, we have

ds 2 = dx12 +
dx22 + dx32
,
1 + m2 x12
mx1
B23 dx 2 dx 3 =
dx2 dx3 ,
1 + m2 x12
e =
e0
.
(1 + m2 x12 )1/2
(35)
The open string metric and -field are

ds 2 = dx12 + dx22 + dx32 ,
23 = mx 1 .
(36)
The closed string metric in (35) is no longer periodic in x1 . The metric in (32) has the
property that the (23) torus at x1 + 1 is related to that at x1 by an SL(2, Z) transformation

1
0
A=
(37)
.
m 1
This 2 2 matrix is embedded in the full O(2, 2; Z) T-duality group as

A
0
.
S=
0 (AT )1
as explained for example in the review [22].
(38)
63
The closed string metric and B-field in (35) obey the property that the background
matrix E = G + B of the (23) torus and the dilaton are related
aE(x1) + b
,
cE(x1 ) + d

det g(x1 + 1) 1/4
e(x1 +1) = e(x1)
,
det g(x1 )
E(x1 + 1) =
where a, b, c, d are 2 2 matrices entering a 4 4 O(2, 2) matrix

a b
M=
.
c d
(39)
(40)
One easily checks (39) when x1 = 0 and with a little more work for general x1 . The
a, b, c, d are calculated by observing that the shift by x1 in (35) can be accomplished
by first T-dualizing to (34), doing the shift and T-dualizing back. The O(2, 2; Z) matrix T3
for the T-duality along x3 in (34) is
1 0 0 0
0 0 0 1
T3 =
(41)
0 0 1 0
0 1 0 0
and M = T ST 1 . This gives

1 0
0 0
a=
,
b=
,
0 1
0 0

0
m
1 0
c=
,
d=
.
m 0
0 1
(42)
It is also interesting to observe that the open string background in (36) characterized by the
matrices G and transforms under a shift of x1 by the same O(2, 2; Z) matrix M when
we use the action of O(2, 2; Z) given by SeibergWitten:

T
G(x1 + 1) = G(x1 ) = a + b(x1 ) G(x1 ) a + b(x1) ,

1
(x1 + 1) = c + d(x1 ) a + b(x1 ) ,

1/4
gYM (x1 + 1) = gYM (x1 ) det(a + b(x1 ))
(43)
.
The final background is to be viewed as a T 2 bundle over S 1 where the T 2 is twisted
by an element of the full T-duality group of the torus upon transport along the S 1 . This
structure of the closed string background obtained after T-dualizing the twisted torus has
been observed recently in [23] and related work appears in [24,25].
We now describe the T-dualities on the full background. The above remarks on the
O(2, 2) carry over. After T-duality on T1 we have (the full 10d metric)

2
ds 2 = d5,1
+ H dx 2 + r32 dx32 + r22 dx22 + 1/r12 dx12 ,
B12 = mx3 H123 = m,
e0
H,
e =
r1
(44)
64
which we recognize as nothing but the NS5 brane metric smeared over the transverse
directions 1, 2, 3.
After T-duality on T2 we have

1
ds 2 = H r32 dx32 + 1/r12 dx12 + 2 (dx2 + mx3 dx1 )2 ,
r2 H
e0
(45)
r1 r2
which is the same metric as we started from, with inverted radii r1 , r2 and with 1 and 2
interchanged.
Then in these dual coordinates, T3 has Killing vector V3 = 3 mx1 2 . Applying the
coordinate transformation (33) and T3 T-duality we get (restoring also the ls dependence
for later use)

dx12 H 1 dx22/r22 + dx32 /r32
2
4
,
ds = ls H 2 +
ml 2 x 2
r1
1 + H rs2 r31
e =
mr1
x1 /r1
dx2 dx3
B23 dx 2 dx 3 = ls4
,

2 x 2
2
m
l
r2
r3
r2 r3 H 1 +
s 1
e =
ls3 e0 1/2
H
r1 r2 r3
H r2 r3
2 2 1/2
mls x1
1+
.
H r2 r3
(46)
Let us now define yi = ls2 xi /ri and calculate open string variables, to find the metric G and
noncommutativity parameter the D3-brane sees [26]. Using
ij

ij
1
G+
(47)
=
,
ls2
g + ls2 B
we get for the full 10d metric
d y 2 + d y32

2
ds 2 = d5,1
,
+ H dx 2 + d y12 + 2
H

ls3 e0
r1 2
e =
[y2 , y3 ] = i m
ls y1 = i y1 ,
r1 r2 r3 H
2 r3
mr1 ls2
mr1
|x| = 1 +
|X| = 1 + |X|,
H =1+
(48)
r2 r3
r2 r3
where we have defined X = x/ls2 .
Recalling that ls goes to zero in the infinite boost limit, and then, with yi and fixed,
we have
gij
gij ls4 ',
Gij 4 = fixed,
ls2 ' 1/2 ,
(49)
ls
which is nothing other than the SeibergWitten limit for noncommutative geometry, which
means that the theory on the D3 branes is nothing other than noncommutative superYangMills theory with variables (metric, dilaton and noncommutativity) given in (48).
65
The noncommutativity parameter is

=
mg A l 3
mR 2 R 3
mr1 ls2
mr1 lP3
= As P =
=
.
r2 r3
r2 r3 R
ls R
R 1
(50)
4. Space-dependent noncommutativity and solution of Matrix theory constraints

Let us now see that the super-YangMills defined on (48) can also be obtained from an
algebraic approach in Matrix theory. Our goal will be to reproduce the noncommutative
structure of the Matrix degrees of freedom Xi (t). We follow the general procedure for
compactification [8,1013]. We need to find T-dual variables yi such that the matrices Xi
can be represented as covariant derivatives. For the simple case of circle compactification,
one represents the algebra
X 1 = X + R
(51)
by X = iDy iy + A, = eiRy , since [iy + A, eiRy ] = ReiRy .

Let us review this in a bit more detail. We start with a finite number k of zero branes
on a circle. This is equivalent to having an infinite number of copies of k zero branes
along a line, with zero branes separated by a constant periodic shift R related by a gauge
a b where a , b run
transformation in U () as in (51). One writes Xab matrices as Xmn
over the k zero branes and m, n are integers. These matrices are now operators on states
labeled by an integer m corresponding to eimRy = |m, tensored by finite k k matrices.
The operators on the states
|m can be viewed as the algebra of functions on the T-dual
space. The form A(y) = p Ap eipRy , allows us to read off the T-dual radii. The matrix
elements of X and are given by Xn,m = nRn,m + Ap n,m+p and nm = n,m+1 . So
we have the T-dual Matrix model description in terms of D1-branes (and by generalization,
Dp-branes). Fluctuations in the compact X are mapped to fluctuations in the gauge field
A, and part of the original Matrix degrees of freedom were used to generate functions of
the worldvolume direction y.
We will now try and apply this procedure to our case. We will treat first the case when
the harmonic function H = 1, and will see later what complications H introduces. We can
think of it as working in the near core region x 0. We will also put for simplicity Ri = 1
for the moment, and return to the general case later on.
In our case we have 3 isometries, T1 , T2 , T3 , which means that we need to impose
constraints on the D0 Matrix model analog to (51) and try to solve them in terms of a
T-dual space. The constraints are
T1 :
1 X1 11 = X1 + 1,
1 X2 11 = X2 ,
1 X3 11 = X3 ,
(52)
where 1 is a transformation acting on the Xs which corresponds to the isometry T1 , and

similarly
T2 :
2 X1 21 = X1 ,
66
2 X2 21 = X2 + 1,
2 X3 21 = X3 ,
T3 :
(53)
3 X1 31 = X1 mX2 ,
3 X2 31 = X2 ,
3 X3 31 = X3 + 1.
(54)
We noted that T2 and T3 do not commute, and we can see that therefore 2 and 3 do not
commute. Namely, if we put M = 21 31 2 3 , then we have
MX1 M 1 = X1 m,
MX2 M 1 = X2 ,
MX3 M 1 = X3 .
(55)
From the relations defining T1 and T2 we can see that we can put X1 = iD1 , X2 = iD2
and 1 = eiy1 , 2 = eiy2 just as in flat space. The commutation of relation of 3 with X3
can also be solved by X3 = iD3 and 3 = eiy3 (here = m). The relations (55) allow us
to solve for M = eimy1 . Hence, we deduce that y2 and y3 do not commute, and we get
exactly the noncommutativity relations
[y2 , y3 ] = i23 = iy1 ,
[y1 , y3 ] = [y1 , y2 ] = 0,
(56)
as we obtained from the T-duality approach of the last section.

The relations (56) also imply nontrivial commutations for derivatives and coordinates.
Indeed, by taking the commutator with various derivatives of the relations (56), we get a
set of equations for [i , yj ]. We will not list them here but just mention a solution, namely
[i , yi ] = 1 as usual, but also [1 , y3 ] = i2 . The relevant equation is obtained from
[1 , [y2 , y3 ]]
[[1 , y2 ], y3 ] + [y2 , [1 , y3 ]] = i,
(57)
and we can see that it is indeed solved by [1 , y3 ] = i2 .

It is useful to observe that a change of variables maps the noncommutativity parameter
to a constant. Indeed, in the open string metric
ds 2 = dy12 + dy22 + dy32 + dstr2
(58)
with the noncommutativity 23 = y1 we can make the change of variables y1 = y1 ,

y2 = y2 y1 , y3 = y3 , after which the theory has metric
2

2
2
ds 2 = dy1 + dy3 + d y1 y2
(59)
and constant noncommutativity 23 = . The closed string metric and B-field in (46)
become, under the transformation:

dy3 2 + y1 2 dy2 2 + 2y2 y1 dy2 dy1
y2 2
+
,
ds 2 = dy1 2 1 +
1 + 2 y1 2
1 + 2 y1 2
ls4 B =
y1 (y1 dy2 + y2 dy1 ) dy3

1 + 2 y1 2
(60)
67
To put back the Ri and ls dependence, we only need to substitute = /

ls2 in the above.

Since the noncommutativity is constant in the new coordinates yi , we can realize the
commutation relations
[yi , yj ] = iij ,
[i , yj ] = ji ,
(61)
as we explain further below. The commutation relations (61) imply the following relations
for the unprimed coordinates
[yi , yj ] = iij (y1 ) = iij (y1 ),
[i , yj ] = ji + ji (),
(62)
where ij (y) has nontrivial components 23 = 32 = y1 , ji has nontrivial components

31 = i y 2 , as we wanted (see (56) and (57)). These guarantee that if we set Xi = i y i and
eiyi , we correctly obey the constraints in (52), (53), (54). This provides the foundation for
the general solution including gauge fields but we first need to review the construction of
the covariant derivatives including gauge fields in the case of constant noncommutativity.
We recall some facts about the construction of a noncommutative gauge theory from
covariant derivatives in the context of an ordinary constant noncommutativity ij . To
have notation which agrees with our set-up we will let i, j run over 2, 3 and we will
let the noncommutative torus algebra be generated by y2 , y3 . Consider compactification

constraints generated by 2 = eiy2 and 3 = eiy3 , where i j (i )1 (j )1 = eiij ,

= = . The and X are represented
with the only nontrivial components being 23
i
i
32
in a Hilbert space where there is nontrivial commutant generated by 2 , 3 (i.e.,

[ , ] = 0) which have a noncommutativity parameter = . Writing = ei y2 and
i

3 = ei y3 , we have [y2 , y3 ] = i . Explicit construction of the yi and yi s or equivalently
the i and i in terms of coordinates w2 , w3 which commute with each other and satisfy
standard commutation relations with their derivatives w

, w
, and together describe a
2
3
four-dimensional phase space are given in ([26])
i
2
i
y2 = w2
2
y2 = w2 +
,
w3
,
w3
i
2
i
y3 = w3 +
2
y3 = w3
,
w2
.
w2
(63)
These formulas can be used to check that the correct mutual noncommutativity of yi and
of yi are reproduced, as well as the vanishing commutators of any yi with yi .
We can define derivatives with respect to yi and yi

j

, y = i ,
, y = 0,
(64)
yi j
yi j
and
j

,
y
= i ,
yi j

,
y
= 0.
yi j
(65)
68
These constraints can be solved by defining the derivatives in terms of appropriate

commutator actions with elements in the algebra of y, y:
= i( 1 )ij yj ,
yi
= +i( 1 )ij yj ,
yi
and the nontrivial commutation relations of derivatives follow:

1 ij
,
)
,
,
=
i(
= +i( 1 )ij .
yi yj
yi yj
(66)
(67)
The fact that the derivatives can be expressed in terms of commutator action with
elements in the algebra plays an important role in [27] in the context of a discussion of
solutions of Matrix theory describing extended objects in R 2 and having a noncommutative
worldvolume theory derived from Matrix theory.
The presence of the commutant generated by the y is important in getting solutions
to the constraints with nontrivial gauge fields. The simplest gauge theories are in fact
obtained when we take the covariant derivatives to be yi + yi iAi (ei yi ). Such a choice
of derivative was implicit for example in [13]. It is useful to note that

+
,
+ = 0,
(68)
yi yi yj
yj
which means that there is no background magnetic field. As far as solving the compactification constraints we could work with a more general set of partial derivatives (yi + ij yi ).
We can now write a solution for the X operators acting on a Hilbert space of functions.
Since the periodicities are simple in the y-coordinates, we are lead to consider functions
of generated by eiy1 , eiy2 and eiy3 . Recalling the discussion above, where we saw that
the constraints are expressed in terms of y variables whereas the fields are functions of
variables y , we are lead to work with the Hilbert space of functions of the form

n1 ,n2 ,n3 ein1 y1 ein2 y1 y2 ein3 y3 ,
=
(69)
n1 ,n2 ,n3
where ni are arbitrary integers. The yi and yi , for i = 2, 3 are constructed in terms of a
four-dimensional phase space as in (63).
On this Hilbert space we can write operators

iy iy y i y
1

1
1
2
3
,e
X1 = i iy2 iA2 e , e
y1
y1 y2

i y

2 + A1 eiy1 , eiy1 y2 , ei y3 ,
y1 y2

i

X2 = + A2 eiy1 , eiy1 y2 , ei y3 ,
y1 y2

X3 = i + A3 eiy1 , eiy1 y2 , ei y3 ,
y3

a
< = <a eiy1 , eiy1 y2 , ei y3 .
(70)
69
The constraints are generated by

1 = eiy1 ,
2 = eiy1 y2 ,
3 = eiy3 .
(71)
With these expressions we can check that the constraints in (52), (53), (54) are satisfied
and that the Xs correctly act in the Hilbert space. We elaborate on some aspects of these

i y
properties. The combination i y y 2 y is necessary in X1 because it allows X1
1
to be well defined on the Hilbert space. Consider for example i y acting on eiy1 y2 . It
1
gives y2 eiy1 y2 which does not belong to the Hilbert space we defined. The combination

i y
i y y 2 y does map elements in the Hilbert space back to themselves. For similar
1
1
2

iy
reasons, the appearance of i y y 2 y guarantees that the constraint 2 X1 21 = X1
1
1
2
is satisfied. The appearance of A2 in X1 may seem surprising but is necessary to make
sure that the conjugation of X1 by 3 does correctly reproduce the shift mX2 in (53).
It is also worth noting that the change of variables to y coordinates is a useful guide in
constructing the solution but the periodicity conditions are not simple in these coordinates.
A consequence is that what we might call the gauge fields in the primed coordinates,
deduced from those in the unprimed coordinates are not good operators that act in the
Hilbert space. For example, A 2 = y1 A2 acting on the Hilbert space gives functions of the
1
form y1 which do not belong to the Hilbert space. Finally, while the above is a relatively
1
simple solution including gauge fields, it is not the most general. Just as in the commutative
case, we can also consider k-vectors k acted on by the above operators. By analogy to
the commutative case [11] or the case of constant [12,13], where the appropriate Hilbert
spaces could be generalized to include magnetic fluxes we expect similar generalizations
here.
Let us see what happens now in the presence of the harmonic function H (48), and let
us restore also the Ri factors. The isometries of the metric continue to be the same, H does
not affect the identifications, so we can write down the same constraints as before, where
now Xi correspond to the dimensionless coordinates x i .
1 X1 11 = X1 + 1,
2 X2 21 = X2 + 1,
3 X3 31 = X3 + 1,
3 X1 31 = X1 mX2 .
(72)
We have only written the nontrivial relations above. These have solutions described above.
If we rescale X i = ri Xi and correspondingly yi = ls2 yi /ri for the dual variables, then
2
X i = ils2 D i , i = eiri yi /ls , and one obtains
[y2 , y3 ] = im
ls2 r1
,
r2 r3
y1 = i y1 ,
[y1 , y3 ] = [y1 , y2 ] = 0.
(73)
We have thus obtained the same noncommutativity as from the T-duality approach (50).
Unfortunately, now there is no independent way to verify the open string metric and dilaton,
which are nontrivial in the presence of the harmonic function H .
Still, this fact gives us some useful information about other backgrounds. We notice
that the constraints were not modified by the presence of the D8-brane background (i.e., by
70
the nontrivial H ). We can guess that for a general massive IIA background, the M-theory
lift will be again a dressing of the same space B(A, R) with the same identifications
for the dimensionless xi s, therefore the constraints (72) are unmodified. So by the above
procedure we will get a super-YangMills on a space with the same noncommutativity.
Again we will have a nontrivial metric and dilaton as well as possible other RR fluxes,
which will have to be derived from the T-duality approach. The identifications in M-theory
and correspondingly the noncommutativity of the D3 brane space are of a topological
nature, and so insensitive to local modifications.
5. NCSYM action and stability

In this section we describe the D3 brane action we are getting for the Matrix model.
First let us check that we can put D0 branes at x = 0 in the background (28) (and they are
stable). We will also check whether they can stay at nonzero x (in the D3 Matrix model,
whether we can have a nonzero vev for X).
A probe D0 brane in the background (28) will have the action

S1D0 = dt e< 1 + gij (X i X j ).
(74)
The equations of motion of this action in the background (28) are
d
[H X 3] = 2r12 mX 2 H 1 (X 1 + mX3 X 2 ),
dt

d
= H X 2 + r32 X 32 + r22 X 22 r12 H 1 (X 1 + mx3 X 2 )2
2 [H X]
dt

d 1
H (X1 + mX3 X 2 ) = 0,
dt

d 2mX3
2
(X1 + mX3 X2 ) + H r2 X2 = 0.
dt
H
2r32
(75)
It follows that if X 1 = X 2 = X 3 = 0 and X small (so that H 1), then the only remaining
equation is
2,
2X = H (X)
(76)
hence the static potential vanishes.

Let us compare this with what happens for the D0D8 system. There we have a 1-loop
ChernSimons term k dt (X + A0 ), k Z, which gives a potential for the D0s. One
can calculate it in string theory [28] or directly from arguments about the supersymmetric
quantum mechanics [2931]. But one can understand it from charge conservation. When a
D0 passes through a D8 charge conservation requires the creation of an elementary string
(HananyWitten effect), which will generate a linear potential. In its absence, the D0D8
is not supersymmetric and has a linear repulsive potential V (R) = T0 R. Such a Chern
Simons term (and consequently the linear potential) are absent in the geometric background
we consider (28). One can also see this from the D0 worldline perspective. The CS term of
71
the D0D8 system appeared by integrating out the massive (0, 8) fermion, which is absent
from our case.
Now that we have established that we can have D0 branes at fixed x (and correspondingly D3 branes in the T-dual picture), we would like to describe the action of these D
branes in more detail. The prescription of Myers [32] for the (bosonic part of the) Dirac
BornInfeld (DBI) action in a general background is

SDBI = Tp d p+1

STr e det P Eab + Eai (Q1 )ij Ej b + Fab det Qij ,
Qij = ji + i[Xi , Xj ]Ekj ,
E = (g + B)ij ,
(77)
and a corresponding ChernSimons piece. Here the fields are in closed string variables
and if the fields depend on the transverse scalars, the prescription is to write the fields as
functions of the adjoint-valued scalars and then take a completely symmetric trace over all
adjoint indices.
So the DBI action in our background (46) will be

3

dxi
S = T3 dt
i=1

mn
STr e(X ,x1 ) det gab (X mn , x1 ) + Bab (X mn , x1 ) + Fab ,
(78)
here X mn is the Matrix scalar corresponding to the coordinate transverse to the D8-brane,
and x1 is as defined before. If we assume that the SeibergWitten map continues to hold in
the presence of the nontrivial X mn (which is not entirely obvious, but should probably be
true in a Taylor expansion for small values of X mn ), then we get

3

mn
S = T3 dt
(79)
dxi STr e(X ) det Gab (X mn ) + Fab ,
i=1
where F is the noncommutative field strength. Moreover, we saw that the SenSeiberg
procedure implied that we take the SeibergWitten limit for noncommutative geometry on
the D3 action, so we are left with noncommutative super-YangMills with X-dependent
metric and dilaton.
The D8 brane had 16 supersymmetries, and correspondingly the M-theory background
had also 16 supersymmetries, which means that the D3 brane action (noncommutative
super-YangMills) will have 8 linearly realized supersymmetries. The fermionic field
content is the same as for the flat D3 brane, but half the supersymmetries are broken by the
nonzero Xmn and the nontrivial noncommutativity.
In the near horizon region (at x = 0), the D8 background is flat, so it has 32
supersymmetries. Correspondingly, the DBI action has 16 supersymmetries if we put
and Gab constant (and keep only the noncommutativity), as in the constant case.
72
Let us now examine the star product, since it is nontrivial (space-dependent). For
constant noncommutativity, the noncommutativity of the space can be traded for a modified
product, the star product,

ij
f A g = ei i j f (x)g(x )x=x ,
(80)
but when is spacetime-dependent, we have to be more careful.
The first observation we can make is that our ij satisfies the associativity condition
[33,34]
il l j k + j l l ki + kl l ij = 0,
(81)
and so we can define an associative star product. As an aside, we have a nonzero H123 , yet
the product is still associative. This is possible because ij is not invertible in the whole
space (1, 2, 3), but just in (2, 3) (if it would be, then associativity and zero H field would
be the same, see [34]). Associative star products in the case of space-dependent can be
defined with the prescription given by Kontsevich [33].
The abstract formula is

w B, (f, g),
f Ag =
h n
n
w =
Gn
1
n!(2)2 n

ni=1 de1 de2 ,
k
(82)
Hn
and where explicitly, derivatives which can act either on f and g, or on , are contracted
with other s. For example, up to second order in we have
ij
f A g = fg + h
i f j g +
h 2 ij kl
i k f j l g
2

h 2 ij
(83)
j kl (i k f l g k f i l g) + O(h 3 ).
3
But in our case we have not only the associativity condition (81) but also the more
restrictive condition
+
ij j kl = 0,
(84)
which implies that there will be no corrections (since derivatives on will always appear
in the above combination, as the only object with contravariant indices is ). Then the
Kontsevich product will be the same as the usual star product, a fact which is obvious in
the expanded form. The exponential form will also be the same, and we therefore have

ij
ij
f A g = ei (x)i j f (x)g(x )x=x = ei (x )i j f (x)g(x )x=x .
(85)
As we saw, we can change coordinates by x2 = x2 x1 , x1 = x1 , and then 23 = ,
but then the metric is not flat anymore. We can, however, obtain a third form for the star
product. Since in these new coordinates the product is

ij
f A g = exp i xi yj f (x )g(y ) ,
(86)
x =y
by going back to unprimed coordinates we get

!
m
n

ij x y
f A g = exp i
f (x)g(y) .
i
j xm yn

x y
x=y
73
(87)
As this point it is interesting to observe the similarity of our noncommutative theory

to the one described by Hashimoto and Sethi in [35]. Moreover, if we take a Penrose-like
limit (infinite boost, while taking a relevant mass parameter to zero) we obtain half their
solution. Indeed, take an infinite boost in the x1 direction, and also take to zero as
x1
e'
e' +
(x1 + t ) = x ,
2
2
e'
,
2
(88)
and drop the primes. Then the open string variables (48) become the flat metric (and
constant string coupling), with 23 = x + (notice that H = 1 in this limit, since 0).
Their solution has also 3 = x 2 . In these coordinates (with spacetime-dependent
noncommutativity), their closed string variables (metric, B-field and dilaton) are

R 2 dx22 + dx32
x 2 (dx + )2 2x2x + dx2 dx +
+
ds = 2 dx dx +
2 + 2 +
,
2
+
2
R + (x )
(x )
R 2 + (x + )2
+

Rx dx2 dx3
Rx2 dx + dx3
B=
,
R 2 + (x + )2
R 2 + (x + )2
"
R2
e = gs
,
R 2 + (x + )2
2
(89)
where the terms in brackets correspond to the Penrose limit of our solution. In these
coordinates, their open string metric is flat and the open string dilaton constant, just as in
our case. So we are obtaining half the solution in [35], which seems to suggest that both
are part of a 1-parameter set of solutions.
Another observation is that in [35] there is also a transformation of coordinates which
makes ij constant, namely,
+
x+ = x ,
x2 = x2 x ,
1 +
x = x + x x22 ,
2
(90)
whereas for the Penrose limit of our solution it is just

+
x+ = x ,
x2 = x2 x .
(91)
However, in their case (84) is not satisfied, while (81) is still satisfied, so in their case
the Kontsevich product is different from the usual star product, even though there is a
coordinate transformation which makes constant.
Finally we note that an example of spacetime-dependent noncommutativity has been
analyzed in [36], and the SeibergWitten analysis still holds (even though is varying).
74
6. Spectrum of states
Now we take a step toward deriving the duality between the noncommutative superYangMills on (48) and massive IIA in the D8 background, by studying the spectrum
of BPS states. Type IIB string theory in ten dimensions can be obtained by compactifying
M-theory on a 2-torus of vanishing area, but fixed complex structure. In this case the Sethi
Susskind [5] and BanksSeiberg [4] constructions gave evidence for the duality. We will
follow the SethiSusskind construction, which is defined in 3 + 1 dimensions, setting it
up so that we can go smoothly to our case. The IIB Matrix model is 2 + 1 super-Yang
Mills which has naturally an SO(7) invariance, but the claim is that at strong coupling it
develops an SO(8) invariance (which is consistent with the supersymmetry algebra and is
the maximal R symmetry). The easiest way to see it [5] is to go to (3 + 1)d super-Yang
Mills and use electricmagnetic duality. There we have only an SO(6) manifest invariance
(6 scalars), which will be enhanced to SO(8). In our case we naturally have (3 + 1)d superYangMills, so it should be our starting point. In the massless case (m = 0), we still have
an SO(6) (6 scalars) enhanced to SO(8), but in the massive case we have an SO(5) (the
scalar X vev corresponding to the direction transverse to the D8 brane in IIA is special),
which should be enhanced to SO(7).
Let us then set up (3 + 1)d super-YangMills for our use. The super-YangMills lives
on a dual torus of lengths R 1 , R 2 , R 3 . Sethi and Susskind have R2 , R3 0, R1 .
The mass of a membrane on the shrinking torus R2 , R3 is identified via the M-theoryIIB
duality with the momentum mode on another direction Y in IIB,
R2 R3
lP3
1
RY
(92)
with the limit RY to infinity, and we set RY = R1 . By the above formula, we see that
RY = ls2 /R2 (if R3 is the M-theory direction), and so as we said RY is the extra transverse
direction in the lightcone IIB theory which appears when the M-theory torus shrinks to
zero size. To obtain SO(8) invariance, we indeed need to choose RY = R1 = R , so that
all the IIB lightcone coordinates, X1 , XY , X4 , . . . , X9 have length R . Then the (3 + 1)d
super-YangMills coupling,
lP3
= 1,
(93)
R1 R2 R3
so we are at the self-dual point, and we have electricmagnetic duality. As usual by
-theory was introduced in order to show massive string
the SenSeiberg procedure, M
degrees of freedom decouple from the super-YangMills, but the T-dual super-YangMills
variables depend only on M-theory quantities.
The electric flux along R 1 corresponds to the momentum conjugate to X1 under Tduality A1 X1 , and so it goes together with the other momenta to increase SO(6) to
SO(7) invariance. Because of electricmagnetic duality however, SO(7) becomes SO(8).
In our case, when the 3d space B in M-theory shrinks to zero size, we have 2 extra
transverse lightcone coordinates appearing in Type IIB,
2
=
gYM
R1 R2
lP3
1
RY2
and
R1 R3
lP3
1
,
RY3
(94)
75
and so SO(8) invariance (in the massless case) should be recovered when XY2 , XY3 , X4 ,
. . . , X9 have the same length,
RY2 = RY3 = R .
(95)
For Sethi and Susskind, the magnetic flux F23 in the D2 theory (and correspondingly
in the D3 theory as well) was T-dual to wrapping number of membranes, which by the
MIIB duality (and 911 flip) was identified with momentum on RY (see (92)). For our
case, the invariance we seek is with the momentum in RY2 and RY3 , which corresponds in
the D0 theory by (94) to wrapping number on 12 and 13, respectively. By T-duality, in the
D3 theory, this is magnetic flux F12 and F13 .
So let us see the SO(8) invariance in the 2 cases from the YM energy. The energy of
magnetic fluxes and electric fluxes is deduced from

2
3
23
2
dx 2 dx 3 tr F01 = n23
dx dx tr F23 = nm ,
(96)
e gYM
(for magnetic flux on 23 and electric flux on 1) so that
tr F23 =
n23
m
,
R 2 R 3
tr F01 =
2
n23
e gYM
.
R 2 R 3
If we add momentum modes on x1 , x2 , x3 (pi = ni /R i ), the energy

2
1
tr Fj2k + Xi
E= 2
dx 1 dx 2 dx 3 tr F0i2 +
2gYM
j <k
(97)
(98)
2 = tr
2
2
becomes (using that trU (N) F
SU(N) F + (1/N)(tr F ) and concentrating on the
U (1) piece)
2

2

nm12
nm23
R 1
R 3
2
2
2
2
NE =
+ ne12 gYM +
+ ne23 gYM
2
2
2R 1 R 2 gYM
2R 2 R 3 gYM
2
" 2 2 2
nm13
n2
n3
n1
R 2
2
+
(99)
+ n2e13 gYM
+
+
.
+
2
R1
R2
R 3
2R1 R3 gYM
This formula is also in accord with [13,37]. The elementary excitations of the theory are
the momentum modes ni , but the Matrix theory prescription tells us to look at the energy of
excitations on the moduli space, in other words, for excitations with energy much smaller
than that of momentum modes.
In the SethiSusskind case (gYM = 1), the smallest elementary excitation (momentum
mode) is of order 1/R 2,3 (R 1 R 2,3 ), and the 12 and 13 fluxes have energy much bigger
than that, whereas the 23 fluxes have much smaller energy, so they should be thought of as
moduli.
In our case, (gYM , R 1 $ R 2,3 ), the smallest elementary excitation is of order
1/R 1 , and the 23 fluxes have energy much bigger than that. For the 12 and 13 fluxes,
choosing R 2 = R 3 (as we have seen we need in (94)), the prefactor (energy scale) of
the fluxes is also 1/R 1 , as for the momentum modes. However, since gYM is infinite, the
76
electric fluxes have energy much larger than the momentum modes, whereas the magnetic
fluxes have energy much smaller than the momentum modes, so are real moduli.
Now let us see what should we compare the energy of those moduli against. The energy
on the moduli space of the 6 scalar fields is
NE =
p%2
n%2
n%2 lP2
=
=
,
2M
2 l 2
2M0 2R
2M0 R
0
P
(100)
where we have put the transverse space in a box of size R , to be equated with RY2 = RY3
as above, and M0 is the mass of those moduli, but we have taken into account that we are
-theory. M0 comes from the fact that we are really expanding
calculating energies in the M
the DBI action of the D3 in order to get (100). But then
M0 = Vp Tp =
p% 2
R 1 R 2 R 3
1
n%2 gs ls lP2
n% 2
=
NE = =
=
R
.
2 l2
2
2M0
g s ls4
gs ls
2R
2R
P
(101)
-theory. Then in the SethiSusskind case we have the

Here ls is the string length for the M
energy of the moduli (using (93))

l 2 ls gs n2m12 + n2e12
R n2m12 + n2e12
n% 2
n% 2
=
,
+
+
E= P 2
(102)
2
2
2N lP
2N
R12
R
R12
R
which is SO(8)-invariant if we put Ry = R1 = R , and in our case (with m = 0) we get
2 = l 2 /(R R ), R
1 = ls2 /R and (94))
(using that gYM
2 3
s

2
2
R nm,13 nm,12
n%2
E=
(103)
+ 2 + 2 ,
2N RY2
RY3
R
2
which is SO(8)-invariant if we put RY2 = RY3 = R . We notice that the formulas (102) and
(103) are exactly what we expect from supergravity and from the BFSS model [8] for the
free supergravitons. Of course it would be more interesting to derive the interaction piece.
Finally, what happens in the massive case m = 0 (in the X = 0 sector, which has still 16
supersymmetries)? As we mentioned, one of the scalars (X) corresponds to the direction
transverse to the D8 brane, so we have manifest SO(5) invariance of the scalars which
should be lifted to a SO(7) invariance, of the lightcone string theory in the D8 background.
So the above formulas should apply to 5 of the scalars, but not to the X direction.
The noncommutative super-YangMills is obtained by replacing the usual product with
the star product. To first order in , the action (see [26], Eq. 4.27) is

1
S=
Fij Fmn Gim Gj n 1 Fij ij 2 kl Fki Flj Fmn Gim Gj n
2

2

=
Fij 1 + 3F23 23 ,
(104)
and we see that if F23 = 0, the action is unmodified, and so the energy formula is
unmodified as well, as expected. The higher order terms will just contain terms with
derivatives of F , and so a constant F23 and F13 will still be a solution, and the energy
formula of the magnetic fluxes will again be unmodified. As for the momentum modes p% ,
77
they are modes on the moduli space, not in worldvolume, so there is no reason for their
energy to be modified. We can easily verify that there are no corrections to the energy in
our case by applying the general formula in [13,37] to the moduli space of our theory.
As for the action of the elementary strings, that is easy to understand. In the Type
IIB case, [4] considered the limit of small coupling, when R3 R2 R 2 R 3 , so
YangMills moduli space excitations occur only in the 3 direction. Then the string action
is just the sigma model action on the moduli space, with worldvolume given by time
and the 3 direction. In our case, there is no need to take small coupling, since already
R1 R2,3 R 1 $ R 2,3 , and so YangMills moduli space excitations already occur just
in the worldvolume 1 direction, and the string action is again the sigma model.
What about the supergravity mass? Massive IIA supergravity [1] has a massless
graviton, a dilaton and an antisymmetric tensor B which acquire a mass proportional
to m, a massless 3-form A(3) , and massive fermions (gravitino and spin-1/2). The 1-form
A(1) is gauged away, since it appears in the combination F(2) + mB(2) . In the D8 brane
background, one has to study the wave equation for each field. The massless fields (graviton
and A(3) ) can still have a constant wavefunction in the x direction (transverse to the D8),
and then
2
= 0 E = p =
p2 = 2p+ p + p%
2
p%
2p+
(105)
(where p is the momentum along the D8) which reproduces (103). A nontrivial
wavefunction in the x direction implies that p2 = 0, and correspondingly an extra term
in the energy.
For one of the massive fields, we have to study the wave equation in the D8 background.
The Einstein metric is
dsE2 = H 1/8 dyi2 + H 9/8 dx 2 ,
(106)
which means that the wave equation for a scalar of mass aM in this background (the dilaton
is such a scalar), with a a constant, is

a 2 M 2 = 0 i2 + H 1 x2 a 2 M 2 = 0,
(107)
and then for a separated solution,
= eipi yi (x),
we get
(108)

x2 p2 (1 + M|x|) a 2 M 2 (1 + M|x|)9/8 (x) = 0.
(109)
Notice that this equation does not admit a constant wavefunction, since (x) = c, p2
a 2 M 2 = 0 is not a solution. At large x, it has the asymptotic solutions
(x) = e 25 a(Mx)
16
25/16
(110)
so we can keep only the decaying solution and at small x it becomes a combination of the
oscillatory solutions
2
2 2
(x) = eix p a M
(111)
78
(for p2 < a 2 M 2 ). So, from matching the wavefunction and its derivative over x = 0 we
get a condition on p2 , which will be that the coefficient of the sin should be zero, which
will imply a quantization condition, of the type
p2 = Mn2 (a, M).
(112)
The same type of analysis should hold for every massive field in supergravity, so in general
we will get a formula for the lightcone energy of the type
p%2 + Mn2 (a, M)
(113)
.
2p+
Correspondingly, we expect to find in super-YangMills that various moduli have such
an additional mass term, which will depend on the detailed structure of the interactions
permitted by the 8 linearly realized supersymmetries. However, these massive moduli will
appear only when we look at nontrivial wavefunctions for the super-YangMills scalar
X (other than X = constant and small), so it is hard to analyze. The moduli with trivial
wavefunctions in the X direction will correspond to the massless supergravity modes, and
as we saw, they have the right lightcone energy. Moreover, even if we would find the
massive super-YangMills moduli, on the supergravity side it is also hard to get any results
(although one could of course use numerical methods to find the mass terms).
But we can make one observation. On the supergravity side, all the fermions are
massive, so we expect that also fermionic super-YangMills moduli will have a mass term
in the energy. One hint that this might happen is that we expect the D3 brane fermions to
have a worldvolume mass term of the type aM. Indeed, although we do not know how
to write down the D3 fermionic action for a general supergravity background, we know
that in some backgrounds (like super-coset manifolds), the kinetic term for the fermions
is of the type D,
where D is the spacetime Killing spinor operator pulled back on the
worldvolume [38]. The kinetic term then contains a term of the type
E=
H ,
(114)
which would imply a mass proportional to ls2 Habc (flat indices). But for our closed string
background in the SeibergWitten limit (46), B23 = 1/(x
1 ) and g22 = g33 = (ls2 /(x
1 ))2 ,
and so the fermion mass will be proportional to
= = mr1 = mr1 lP = M lP ,
M
ls2 r2 r3 r2 r3 lP
lP
(115)
=
n (a, M)
with M being the supergravity mass. Then the moduli mass will be M
2
2

Mn (a, M)lP /lP and if we would get an energy Mn (a, M)/2NM0 = Mn (a, M)/2p+ , it
would be as desired. It would be, of course, very interesting to see whether one can recover
all the supergravity mass terms for the lightcone energy, but as we saw, the analysis looks
quite difficult.
7. Holographic dual
Let us try to write down the holographic dual of our noncommutative super-YangMills
defined on (48) in the spirit of the AdS/CFT correspondence. We have to write down a
79
solution for D3 branes in the closed string background (46) corresponding to (48) and then
take a decoupling limit. It turns out, however, to be easier to start with D1 branes in the
background (44) and then make two T-dualities. Indeed, as we saw, the background (44)
corresponds to NS5 branes smeared over the directions 1, 2, 3. We have to put D1 branes at
x = 0 along time and x1 . But this background is one of D1 ending on NS5s, smeared over
the D1 direction, as well as 2 others, transverse to both NS5 and D1. The S-dual of this
(Type IIB) configuration is F1 ending on D5, which we know that exists. Then the original
IIA metric is D0s parallel to KK monopoles, and after T1 and T2 we have D2 ending on
KK monopoles, and finally after T3 we have D3 ending on an unusually T-dualized KK
monopole (an 8-dimensional worldvolume).
The solution for D1 ending on NS5s, depending only on the coordinate x can be found
pretty easily, namely,
1/2
ds 2 = dt 2 H1
B12 = mx3 ,
1/2
+ H1
e =
1/2
d %52 + H H1
dx12
r12
1/2

dx 2 + r22 dx22 + r32 dx32 ,
+ H H1
e0 1/2 1/2
H H1 .
r1
(116)
T-dualizing on T2 we get
1/2
ds 2 = dt 2 H1
1/2
+ H1
1/2
d %52 + H H1
dx12
r12
1/2
+ H H1
dx 2 + r32 dx32
1/2
H 1 H1
r22
(dx2 + mx3 dx1 )2 ,
e0 1/4
(117)
H .
r1 r2 1
And finally, after the coordinate transformation (33) and T3 T-duality, we get (putting back
the ls dependence)
#
2
1/2
1/2
1/2 dx1
1/2
+ H1 ls4 d %52 + H H1
+ H H1 ls4 dx 2
ds 2 = ls4 dt 2 ls4 H1
2
r1
2
$
1/2
2
1
dx3
H H1
dx2
+
,
r1 2 m2 x12 /r12 r 2 + r 2
2
3
1 + ls4 r2 r3
H 2H
e =
mx1
B23 = ls4 2 2
r2 r3 H 2 H1 1 + l 4
1
,
2 2
r1 2 m2 x1 /r1
s r2 r3
H 2 H1

r1 2 m2 x12 /r12 1/2
ls3 e0
4
1
+
l
e =
.
s
r1 r2 r3 H 1/2
r2 r3
H 2 H1
(118)
However, we need to generalize this to the fully localized solution, where the D3-branes
are not smeared over the transverse directions.
The first thing we can do is to look in the near core region. In the near core region, H
is constant ( c), and then there is no obstruction for making the harmonic function H1
80
depend on all its transverse coordinates. Indeed, since H is constant, we can ask to find a
D1 brane solution in the corresponding flat background, and that is just the usual D1 brane
with a nontrivial B-field, i.e., (116), with H = c and H1 (%
5 , x2 , x3 , x) the usual harmonic
function. Then after the two T-dualities one gets the solution (118), where H = c and
H1 1 +
4gs N 2
4gs N 2

1
+
,
(%
2 + cx 2 )2
% 4
(119)
where in the last line we have used the fact that we work near x 0.
Let us now derive the equation for the full solution (outside the core). Partially localized
intersections, where brane 1 with harmonic function H1 lives on t, w,
% x% , and brane 2 with
harmonic function H2 lives on t, w,
% y% , with overall transverse space z%, are written in terms
of harmonic functions H1 and H2 in the usual way, except that now H1 and H2 satisfy the
equations (e.g., [39,40])
z2 H1 (z, y) + H2 (z)y2 H1 (z, y) = 0,
z2 H2 = 0,
or z2 H2 (z, x) + H1 (z)x2 H2 (z, x) = 0,
z2 H1 = 0.
(120)
In other words, we delocalize one brane (say, brane 2) over the worldvolume coordinates
of the other brane (1), and then H1 is harmonic (obeys the Laplace equation) in the
background of brane 2. This is true for any kind of branes, but in particular [40] has
derived explicitly this equations for the 11d intersection of M2 and M5 (over a string).
This intersection is related to our D1NS5 solution as follows. Dimensionally reduce to
Type IIA on the common string, to an F1D4(0) solution, T-dualize to IIB on a transverse
direction to a F1D5(0), and then S-dualize to D1NS5(0).
In our case, the harmonic function H is delocalized over the D1, that is over x1 , as well
as over x2 , x3 , over which we need to T-dualize, and H1 is delocalized over x2 , x3 . So the
full solution is given by (116), where H1 satisfies the equation

, x) = Q(%
)(x),
x2 + H (x)%2 H1 (%
(121)
where we have put explicitly the source term Q = 16 4 gs N( )2 . Then also (117) and
(118) are the corresponding T-dual solutions. We notice that near the core x = 0, H c,
so the solution is indeed (119).
In order to solve (121), we separate variables, by writing

d 5 p i p%
e % H1,p (x)
(2)5

1 1
2 sin(pr)
cos(pr) H1,p (x),
=1+
dp p
8 3 r 2
pr
H1 (%
, x) = 1 +
(122)
and get the equation

H1,p
(x) p2 H (x)H1,p (x) = Q(x).
(123)
81
By putting H = c + m|x| (we will keep this form for now and replace it later with c = 1
and m = /
ls2 ) and
2/3
p
x =
(124)
(c + m|x|),
m
we get the Airy equation

d 2 Hp
(125)
xH
p (x)
= Qm1/3 p2/3 x c(p/m)2/3 ,
d x 2
which has solutions in terms of the Bessel functions I1/3 and K1/3 . We choose K1/3 which
decays exponentially at infinity, and get

2
Hp (x)
(126)
= cp x 1/2 K1/3 x 3/2 .
3
The coefficient cp can be fixed by matching with the normalization of the function source.
We get
Q c
cp =
(127)

2 p
p
p
.
2p1/3 m2/3 K1/3 3 m c3/2 m
c3/2 K4/3 23 m
c3/2
Therefore, the final formula for the harmonic function is
sin(pr)
2

Q c 1/3
pr cos(pr) K1/3 3 p
2
H1 (r, x) = 1 +
dp p
p
p
p

8 3 m2/3 r 2
K1/3 23 m
c3/2 m
c3/2K4/3 23 m
c3/2
(128)
with
(c + m|x|)2/3
.
(129)
m
So, we have found the full solution for the D3 branes in the background.
We can now write down the decoupling limit for the holographic dual in the near-core
(x = 0), namely,
=
|%
|
= fixed,

but we have to supplement it with
0,
U=
ri 0,
yi =
ls2 xi
= fixed,
ri
X=
x
= fixed,

mr1 ls2
= fixed,
r2 r3
and then we have the holographic dual

2

d y22 + d y32
2
U
2
2
ds = dt + d y1 +
1 + 2 y12 U 4 /

2
2
2
2
+ 2 dX + dU + U d4 .
U
gs N = = fixed,
(130)
(131)
(132)
82
This metric is then dual to super-YangMills with

[y2 , y3 ] = i y1 ,
ds 2 = dt 2 + d y22 + d y12 + d y32 .
(133)
We note that the holographic dual in the near-core region (132) is just what we would have
expected from the usual noncommutative case [14,15], with holographic dual

2

dy22 + dy32
2
2
U
2
2
2
2
ds = dt + dy1 +
(134)
+ 2 dU + U d5 ,
1 + 4 U 4 /
U
and 2 = 23 .
To get the full holographic dual, since (remembering just for the purpose of next formula
that what we call ls is really ls , whereas lsA still appears in H and also m denoting the
integer = D8 number)
H =1+
mgsA ls2
mgsA
|x| = 1 +
|X| = 1 + |X|,
ls
ls
(135)
we replace c = 1, m = /
ls2 , Q = 16 4gs Nls4 , together with the rest of the limit into (128),
and rescaling the integration variable as p = P / ls2 we get
H1 (r, x)
h1 (U, X)
,
ls4
(136)
where

2gs N 1/3
dP P 2
h1 (U, X) =
2/3 U 2
sin(P U )

cos(P U ) K1/3 23 P
,

K1/3 23 P P K4/3 23 P
PU
(137)
and
2/3
(1 + |X|)
Then the full holographic dual is

1/2
2

ds = h1 (U, X) dt 2 + H d y12 + H 1

1/2
+ h1 H dX2 + dU 2 + U 2 d42 .
(138)
d y22 + d y32
1 + 2 y12 U 4 /
(139)
We note that the SeibergWitten limit is a subset of the holographic limit, as it should be.
8. Conclusions and discussion

We have proposed a new nonperturbative formulation of massive Type IIA string theory
in terms of a noncommutative YangMills theory with space-dependent noncommutativity
parameter. There remains much to study. In particular, it would be very interesting to
construct in more detail the interaction terms in the action, the energies of physical
83
excitations and to study the S-duality properties of this noncommutative gauge theory.
A more direct derivation of the noncommutative YangMills in Section 5 starting from the
solution of the Matrix theory constraints of Section 4, using information about the action
of zero branes in the curved space of the twisted torus will be useful. In fact, it may be
easier to try and guess the form of the zero brane action which would lead to the actions
in Section 5, using the X matrices constructed in Section 4. Progress in these directions
is likely to also be useful in flux compactifications since T-duality of the twisted torus
gives a background with H -flux as discussed in Section 3. These compactifications offer
promising avenues toward the problem of fixing moduli in string phenomenology [23].
We note that we have described massive IIA theory in terms of a matrix model of D3
branes with noncommutativity, a theory which has a holographic dual. As a limit, massless
IIA theory is described by a matrix model of D3 branes, which is dual to string theory in
AdS5 S5 . But there are two things we should observe:
(1) The D3 branes are on a torus, which translates in making identifications in AdS5 (in
Poincar coordinates, ds 2 = y 2 (dt 2 + d x% 2 ) + dy 2 /y 2 , and the x% coordinates are
identified on a torus).
(2) There are different observables in the D3 brane theory which describe flat space IIA
string theory and AdS5 S5 string theory. For AdS5 S5 , we look at gauge-invariant
observables in the D3 brane theory, whereas for the IIA matrix model we look at
wavefunctions on the moduli space, thereby spontaneously breaking gauge invariance.
Holographic duals in the context of 8-brane solutions have also been discussed recently
in [41].
We comment on the relation of this construction to Type IA string theory [42], where
D8-branes and O8-planes coexist. The massive Type IIA physics is recovered by focusing
on the local physics between a pair of separated D8-branes, or equivalently, by sending the
D8-branes and O8-planes off to infinity. There exists a Matrix proposal for the complete
nonperturbative Type IA system [4346] which is related by S-duality to the E8 E8
heterotic string. It would be interesting to recover the noncommutative theory described in
this paper by integrating out degrees of freedom in these heterotic Matrix models.
As we mentioned in Section 2, a generalized ScherkSchwarz reduction based on a
scaling symmetry of the equations of motion gives a ten-dimensional supergravity which
has de Sitter solution [47]. It was observed in [16] that these can be viewed in terms a
Euclidean radial reduction from M-theory. This suggests that a Matrix model could be
found by generalizing the dimensional reduction methods of Matrix theory that we have
used to radial reductions. This is of course a nontrivial generalization since the spacetime
of M-theory, and hence a Euclidean radial direction, appears very indirectly in Matrix
theory. Rather than imposing the constraints directly on a few X-fields corresponding to
the compactified directions, one has to scale all the X matrices as well as the worldline
time coordinate. This approach appears nontrivial and very different from proposals made
for a Matrix model for de Sitter made so far [48,49], and is an interesting avenue for the
future.
84
Acknowledgements
We are happy to acknowledge useful discussions with Steve Corley, Atish Dabholkar,
Laurent Freidel, Aki Hashimoto, Antal Jevicki, Robert de Mello Koch, Joao Rodrigues.
This research was supported in part by DOE grant DE-FE0291ER40688-Task A.
Appendix A. Limits
In this appendix we review the various Matrix theory limits, and derive the correct limit
in our case. For completeness, let us recall the formulas relating the M-theory parameters
on a spatial circle to the IIA string theory parameters. They are obtained from
1
1
=
,
gs ls
R11
1
R11
= 3 .
2
ls
lP
(A.1)
Sen [7] and Seiberg [6] used a construction for M-theory compactified on T p in a limit of
-theory, taking
vanishing radii. We refer to this as M
lP , R 11 , R i 0 gs , ls 0,
(A.2)
such that
ai =
R i
,
lP
M=
R 11
lP2
(A.3)
are held fixed. After dimensionally reducing on R 11 to string theory and making T-dualities
on all the R i , the T-dual variables are
l2
1
R i = s =
,
Ma
Ri
i
p

1p/3
ai1 ,
= g s
1/3
ls = ls = M 1 g s ,
gs = %p
gs
(R i /ls )
i=1
(A.4)
i=1
and moreover
1
2
gYM
ls3
,
gs
1
2
gYM
3p
ls
,
gs
(A.5)
such that
2
= M 3,
gYM
2
gYM
= M 3p
p

ai1 .
(A.6)
i=1
So the limit was chosen to decouple string theory both in the original and in the dual
theory (gs , ls , g s , ls 0), while keeping the YangMills couplings (gYM of the D0-branes
and gYM of the Dn-branes) and the dual radii finite.
Let us now review the BFSS point of view, which is also advocated for the Matrix string
of Type IIA, and the Matrix theory of Type IIB, and then apply it for our case. After that,
85
we will look at the relation between SenSeiberg and BFSS and apply it to the Matrix
models, finally deriving our limit.
BFSS [8] chose the limit
R11 N ,
so that
gs =
R11
lP
lP = fixed,
3/2
(A.7)
3/2
ls =
lP
1/2
R11
(A.8)
and thus one obtained an U () D0 brane theory. The argument being that in the limit,
string theory does decouple (even though gs is infinite), because there are no string states
other than D0 branes which have momentum on the 11th direction (that is the D0 charge),
and so if we look at fixed momentum N/R11 , strings decouple.
The analogous statement happened for the IIA Matrix string [4,50,51]. One added to the
above construction a compactification on a finite R9 , and then made a 911 flip, meaning
one reinterprets 9 as the 11th direction. Since lP was finite, after the flip
3/2
3/2
l
R9
= finite,
ls = P1/2 = finite.
gs =
(A.9)
lP
R9
The IIB Matrix theory [4] was similar. Add to the BFSS two extra radii R1 , R2 0,
with R1 /R2 = finite. We know that M-theory on this space gives IIB with finite coupling.
Then take the BFSS construction and consider lP 0, but independent of N (which is
consistent with the BFSS limit), such that one holds R1 / lP3 and R2 / lP3 fixed (the (p, q)
Type IIB string tensions fixed), then flip 911.
Then
l3
R1
,
ls2 = P
(A.10)
R2
R1
are fixed in this limit.
Similarly for our case, for the new Matrix theory of Type IIA obtained by compactifying
on a radius of zero size, we have, first for the massless case: compactify on an extra R3 0
and make a T-duality so that
gsB =
l3
gsB ls
R1 ls
(A.11)
=
= fixed,
ls2 = P = fixed,
R3
R2 R3
R9
where we have as before flipped 911 (so this corresponds to IIA with R11 = N ), and
we note that now gsB goes to zero, and it is gsA which is finite. For the massive IIA case,
everything is similar, with the addition of the new parameter m.
The equivalence of the BFSS limit and the SeibergSen limit [6,7] was derived as
follows. The light-like circle compactification for finite N (DLCQ, see [52]) with p+ =
N/R finite (BFSS corresponds to N , R infinite, keeping p+ finite, with other possible
compactified directions of fixed radii Ri ),

x
x
R/ 2
+
(A.12)
t
t
R/ 2
gsA =
86
is understood as the Rs R limit of

& 2

R /2
+ Rs2
x
x
,
+
t
t
R/ 2
&
which is the infinite boost limit ( = R/ R 2 + 2Rs2 ) of

x
x
Rs
.
+
0
t
t
(A.13)
(A.14)
So the light-like compactification of M-theory on R is related to the Rs 0 limit of the

spatial compactification of another M-theory. If we subsequently rescale Plancks constant
-theory described above. So,
such that p and pi are held fixed we obtain the M
P )3/2 ,
gs = (R 11 M
p+ = N/R,
p11 = N/R 11 ,
s2 = R 11 M
3 ,
M
P
p RMP2 ,
pi Ri MP ,
P2 ,
P ,
p R 11 M
p i R i M
(A.15)
and then
P2 = RMP2 ,
R 11 M
P = Ri MP
R i M
(A.16)
P limit. Then
are held fixed in the M

P )3/2 = R 3/4 RMP2 3/4 0,
gs = (R 11 M
11

3 = R 1/2 RMP2 3/2 ,
s2 = R 11 M
M
P
11
(A.17)
so string theory decouples and the D0 coupling is fixed

2
s3 = R 11 M
P2 3 = RMP2 3 .
= gs M
gYM
(A.18)
If the (BFSS) M-theory is compactified on a torus of fixed radii Ri then one T-dualizes
-theory and gets
the string theory coming from the M
1
1
1
=
=
,
s2 R i M
3 R 11 Ri RM 3
R i M
P
P

sp3 R 3 MP6
sp
R i = M
R i 0 if p < 3,
gs = gs M
R i =

g s
2
gYM,Dp
= p3 = R 3 MP6
R i .
s
M
(A.19)
So again string theory decouples and one gets a Dp brane theory of fixed YangMills
coupling and dual radii.
Applying this to the Matrix string theory, one relates again M-theory on light-like
P , R 9 0,
-theory with Rs 0, M
R N , with lP fixed and R9 fixed to the M
that is a D0-brane theory on a vanishing circle. After T-duality, it becomes a D1-brane
theory with fixed R 9 .
87
On the M-theory side, one flips 911, reinterpreting it as string theory with a light-like
coordinate R, so
3/2
gs = (R9 MP )3/2,
1/2
ls = lP R9
R Nls .
-theory side, string theory decouples:

Then on the M
3/2
1/2
gs = Rs RMP2
R 9 0,

s2 = Rs1/2 RMP2 3/2
M
(A.20)
(A.21)
and the YangMills parameters are

l2
R 9 = s ,
R
2
gYM
=
1
gs2 R 92
(A.22)
In this way a D1 theory with fixed parameters is related to a string theory with fixed
parameters.
For our IIA Matrix model, M-theory with Ri 0, lP 0, R Nls with
gsA =
R1 ls
,
R2 R3
3/2
1/2
ls = lP R1
(A.23)
-theory we see we are left with a decoupled D3-brane theory on the

fixed. Passing to the M
T-dual space with parameters
g A ls R3
R 2 = s
,
R
ls gsA
2
=
.
gs = gYM
R1
l2
R 1 = s ,
R
g A ls R2
R 3 = s
,
R
(A.24)
We notice though that the YangMills coupling and 2 of the radii are actually not finite,
so there is probably a better description, but one has to find it. In particular, since the Yang
Mills coupling is going to infinity, one should S-dualize, but the problem is in the presence
of the noncommutativity it is not very obvious what that means, so we will stick with this
description. At = 0 the S-dual is a good description, and
2
gs,D = gYM,D
=
R1
0.
ls gsA
(A.25)
But under S-duality, the dimensionless Newton constant k 2 /R 8 g s2 (ls /R)8 (with R a
fixed length scale in the metric) is invariant (since gs 1/gs , R R/ gs ) and we would
like to have ls /R 0 as well as ls /RD 0. (Then, the S-dual theory is decoupled, and
therefore so is the original theory.) The condition can be written as gs (ls /R)2 0, that
is, the coefficient of the first loop correction to the action should be negligible, and this
-theory.
condition is satisfied if we have an M
We also notice that R1 fixed, but R 2,3 0, but all R i /ls (so we are talking
about a (3 + 1)d super-YangMills!) and there are 2 fixed quantities, a dimensionless one,
R 2 R 3 2
A
A
2 g YM , and a dimensionful one, R1 , which will be related to gs and ls , respectively.

R1
88
References
[1] L.J. Romans, Massive n = 2a supergravity in ten dimensions, Phys. Lett. B 169 (1986) 374.
[2] J. Polchinski, Dirichlet-branes and RamondRamond charges, Phys. Rev. Lett. 75 (1995) 47244727, hepth/9510017.
[3] C.M. Hull, Massive string theories from M-theory and F-theory, JHEP 11 (1998) 027, hep-th/9811021.
[4] T. Banks, N. Seiberg, Strings from matrices, Nucl. Phys. B 497 (1997) 4155, hep-th/9702187.
[5] S. Sethi, L. Susskind, Rotational invariance in the M(atrix) formulation of type IIB theory, Phys. Lett. B 400
(1997) 265268, hep-th/9702101.
[6] N. Seiberg, Why is the matrix model correct?, Phys. Rev. Lett. 79 (1997) 35773580, hep-th/9710009.
[7] A. Sen, D0 branes on Tn and matrix theory, Adv. Theor. Math. Phys. 2 (1998) 5159, hep-th/9709220.
[8] T. Banks, W. Fischler, S.H. Shenker, L. Susskind, M-theory as a matrix model: A conjecture, Phys. Rev.
D 55 (1997) 51125128, hep-th/9610043.
[9] M. Claudson, M.B. Halpern, Supersymmetric ground state wave functions, Nucl. Phys. B 250 (1985) 689.
[10] W. Taylor, D-brane field theory on compact spaces, Phys. Lett. B 394 (1997) 283287, hep-th/9611042.
[11] O.J. Ganor, S. Ramgoolam, W. Taylor, Branes, fluxes and duality in M(atrix)-theory, Nucl. Phys. B 492
(1997) 191204, hep-th/9611202.
[12] A. Connes, M.R. Douglas, A. Schwarz, Noncommutative geometry and matrix theory: compactification on
tori, JHEP 02 (1998) 003, hep-th/9711162.
[13] P.-M. Ho, Twisted bundle on quantum torus and BPS states in matrix theory, Phys. Lett. B 434 (1998) 4147,
hep-th/9803166.
[14] A. Hashimoto, N. Itzhaki, Non-commutative YangMills and the AdS/CFT correspondence, Phys. Lett.
B 465 (1999) 142147, hep-th/9907166.
[15] J.M. Maldacena, J.G. Russo, Large N limit of noncommutative gauge theories, JHEP 09 (1999) 025, hepth/9908134.
[16] I.V. Lavrinenko, H. Lu, C.N. Pope, Fibre bundles and generalised dimensional reductions, Class. Quantum
Grav. 15 (1998) 22392256, hep-th/9710243.
[17] P.S. Howe, N.D. Lambert, P.C. West, A new massive type IIA supergravity from compactification, Phys.
Lett. B 416 (1998) 303308, hep-th/9707139.
[18] E. Bergshoeff, M. de Roo, M.B. Green, G. Papadopoulos, P.K. Townsend, Duality of type II 7-branes and
8-branes, Nucl. Phys. B 470 (1996) 113135, hep-th/9601150.
[19] T.H. Buscher, Quantum corrections and extended supersymmetry in new sigma models, Phys. Lett. B 159
(1985) 127.
[20] T.H. Buscher, A symmetry of the string background field equations, Phys. Lett. B 194 (1987) 59.
[21] T.H. Buscher, Path integral derivation of quantum duality in nonlinear sigma models, Phys. Lett. B 201
(1988) 466.
[22] A. Giveon, M. Porrati, E. Rabinovici, Target space duality in string theory, Phys. Rep. 244 (1994) 77202,
hep-th/9401139.
[23] S. Kachru, M.B. Schulz, P.K. Tripathy, S.P. Trivedi, New supersymmetric string compactifications, hepth/0211182.
[24] A. Dabholkar, C. Hull, Duality twists, orbifolds, and fluxes, hep-th/0210209.
[25] S. Hellerman, J. McGreevy, B. Williams, Geometric constructions of nongeometric string theories, hepth/0208174.
[26] N. Seiberg, E. Witten, String theory and noncommutative geometry, JHEP 09 (1999) 032, hep-th/9908142.
[27] N. Seiberg, A note on background independence in noncommutative gauge theories, matrix model and
tachyon condensation, JHEP 09 (2000) 003, hep-th/0008013.
[28] O. Bergman, M.R. Gaberdiel, G. Lifschytz, Branes, orientifolds and the creation of elementary strings, Nucl.
Phys. B 509 (1998) 194215, hep-th/9705130.
[29] D.A. Lowe, Bound states of type I D-particles and enhanced gauge symmetry, Nucl. Phys. B 501 (1997)
134142, hep-th/9702006.
[30] T. Banks, N. Seiberg, E. Silverstein, Zero and one-dimensional probes with N = 8 supersymmetry, Phys.
Lett. B 401 (1997) 3037, hep-th/9703052.
[31] N. Kim, S.-J. Rey, M(atrix) theory on an orbifold and twisted membrane, Nucl. Phys. B 504 (1997) 189213,
hep-th/9701139.
89
[32] R.C. Myers, Dielectric branes, JHEP 12 (1999) 022, hep-th/9910053.

[33] M. Kontsevich, Deformation quantization of Poisson manifolds, I, q-alg/9709040.
[34] L. Cornalba, R. Schiappa, Nonassociative star product deformations for D-brane worldvolumes in curved
backgrounds, Commun. Math. Phys. 225 (2002) 3366, hep-th/0101219.
[35] A. Hashimoto, S. Sethi, Holography and string dynamics in time-dependent backgrounds, hep-th/0208126.
[36] L. Dolan, C.R. Nappi, Noncommutativity in a time-dependent background, Phys. Lett. B 551 (2003) 369
377, hep-th/0210030.
[37] B. Pioline, A. Schwarz, Morita equivalence and T-duality (or B versus ), JHEP 08 (1999) 021, hepth/9908019.
[38] R. Kallosh, J. Rahmfeld, A. Rajaraman, Near-horizon superspace, JHEP 09 (1998) 002, hep-th/9805217.
[39] D. Youm, Partially localized intersecting BPS branes, Nucl. Phys. B 556 (1999) 222246.
[40] H. Lu, C.N. Pope, Interacting intersections, Int. J. Mod. Phys. A 13 (1998) 44254443, hep-th/9710155.
[41] H. Singh, Note on (D6, D8) bound state, massive duality and noncommutativity, hep-th/0212103.
[42] J. Polchinski, E. Witten, Evidence for heterotictype I string duality, Nucl. Phys. B 460 (1996) 525540,
hep-th/9510169.
[43] D.A. Lowe, Heterotic matrix string theory, Phys. Lett. B 403 (1997) 243249, hep-th/9704041.
[44] T. Banks, L. Motl, Heterotic strings from matrices, JHEP 12 (1997) 004, hep-th/9703218.
[45] S.-J. Rey, Heterotic M(atrix) strings and their interactions, Nucl. Phys. B 502 (1997) 170190, hepth/9704158.
[46] D. Kabat, S.-J. Rey, Wilson lines and T-duality in heterotic M(atrix) theory, Nucl. Phys. B 508 (1997) 535
568, hep-th/9707099.
[47] A. Chamblin, N.D. Lambert, De Sitter space from M-theory, Phys. Lett. B 508 (2001) 369374, hepth/0102159.
[48] M. Li, Matrix model for de Sitter, JHEP 04 (2002) 005, hep-th/0106184.
[49] A. Chamblin, N.D. Lambert, Zero-branes, quantum mechanics and the cosmological constant, Phys. Rev.
D 65 (2002) 066002, hep-th/0107031.
[50] L. Motl, Proposals on nonperturbative superstring interactions, hep-th/9701025.
[51] R. Dijkgraaf, E. Verlinde, H. Verlinde, Matrix string theory, Nucl. Phys. B 500 (1997) 4361, hepth/9703030.
[52] L. Susskind, Another conjecture about M(atrix) theory, hep-th/9704080.

2D string theory as normal matrix model

Sergei Yu. Alexandrov a,b,c , Vladimir A. Kazakov a,d ,
Ivan K. Kostov b,1
a Laboratoire de Physique Thorique de lEcole Normale Suprieure, 2
24 rue Lhomond, 75231 Paris cedex, France

b Service de Physique Thorique, CNRSURA 2306, C.E.A.Saclay, F-91191 Gif-sur-Yvette cedex, France
c V.A. Fock Department of Theoretical Physics, St. Petersburg University, St. Petersburg, Russia
d Department of Physics and Astronomy, Rutgers University, Piscataway, NJ 08855, USA
Received 7 March 2003; accepted 20 June 2003
Abstract
We show that the c = 1 bosonic string theory at finite temperature has two matrix-model
realizations related by a kind of duality transformation. The first realization is the standard one
given by the compactified matrix quantum mechanics in the inverted oscillator potential. The second
realization, which we derive here, is given by the normal matrix model. Both matrix models exhibit
the Toda integrable structure and are associated with two dual cycles (a compact and a non-compact
one) of a complex curve with the topology of a sphere with two punctures. The equivalence of the
two matrix models holds for an arbitrary tachyon perturbation and in all orders in the string coupling
constant.
PACS: 11.25.Pm
E-mail addresses: alexand@spht.saclay.cea.fr (S.Yu. Alexandrov), kazakov@physique.ens.fr

(V.A. Kazakov), kostov@spht.saclay.cea.fr (I.K. Kostov).
1 Associate member of Institut za dreni Izsledvami i drena Energetika, Bulgarian
Academy of Sciences, Sofia, Bulgaria.
2 Unit de Recherche du Centre National de la Recherche Scientifique et de lEcole Normale Suprieure et
lUniversit de Paris-Sud.
doi:10.1016/S0550-3213(03)00546-7
S.Yu. Alexandrov et al. / Nuclear Physics B 667 (2003) 90110
91
1. Introduction
The c = 1 string theory3 has been originally constructed in the early 90s as the
theory of random surfaces embedded into a one-dimensional spacetime [2]. Since then
it became clear that this is only one of the realizations of a more universal structure, which
reappeared in various mathematical and physical problems. Most recently, it was used for
the description of the 2D black hole [36] in the context of the topological string theories
on CalabiYau manifolds with vanishing cycles [7] and N = 1 SYM theories [8].
The c = 1 string theory can be constructed as the collective field theory for a
one-dimensional N N Hermitian matrix field theory known also as matrix quantum
mechanics (MQM). This construction represents the simplest example of the strings/matrix
correspondence. The collective excitations in the singlet sector of MQM are massless
tachyons with various momenta, while the non-singlet sectors contain also winding
modes.
The singlet sector of MQM can be reduced to a system of N non-relativistic fermions in
the upside-down gaussian potential. Thus the elementary excitations of the c = 1 string can
be represented as collective excitations of free fermions near the Fermi level. The tree-level
string-theory S-matrix can be extracted by considering the propagation of infinitesimal
pulses along the Fermi sea and their reflection off the Liouville wall [9,10]. The
(time-dependent) string backgrounds are associated with the possible profiles of the Fermi
sea [11].
An important property of the c = 1 string theory is its integrability. The latter has been
discovered by Dijkgraaf, Moore and Plesser [12] in studying the properties of the tachyon
scattering amplitudes. It was demonstrated in [12] that the partition function of the c = 1
string theory in the case when the allowed momenta form a lattice, as in the case of the
compactified Euclidean theory, is a tau function of Toda hierarchy [13]. The operators
associated with the momentum modes in the string theory have been interpreted in [12] as
Toda flows. A special case represents the theory compactified at the self-dual radius R = 1.
It is equivalent to a topological theory that computes the Euler characteristic of the moduli
space of Riemann surfaces [14]. When R = 1 and only in this case, the partition function
of the string theory has alternative realization as a Kontsevich-type model [12,15].
In this paper we will show that there is another realization of the c = 1 string theory
by the so-called normal matrix model (NMM). This is a complex N N matrix model in
which the integration measure is restricted to matrices that commute with their Hermitian
conjugates. The normal matrix model has been studied in the recent years in the context
of the Laplacian growth problem, the integrable structure of conformal maps, and the
quantum Hall droplets [1621]. It has been noticed that both matrix models, the singlet
sector of MQM and NMM, have similar properties. Both models can be reduced to systems
of non-relativistic fermions and possess Toda integrable structure. There is however an
essential difference: the normal matrix model describes a compact droplet of Fermi liquid
while the Fermi sea of the MQM is non-compact. Furthermore, the perturbations of NMM
are introduced by a matrix potential while these of MQM are introduced by means of time3 As a good review we recommend [1].
92
dependent asymptotic states. In this paper we will show that nevertheless these two models
are equivalent.
The exact statement is that the matrix quantum mechanics compactified at radius R
is equivalent to the normal matrix model defined by the probability distribution function
exp[ h1 WR (Z, Z )], where

R

R1
k
tr log ZZ
WR Z, Z = tr ZZ h
tk tr Z k
tk tr Z .
2
k1
k1
(1.1)
More precisely, the grand canonical partition function of MQM is identical to the canonical
partition function of NMM at the perturbative level, by which we mean that the genus
expansions of the two free energies coincide. This will be proved in two steps. First, we
will show that the non-perturbed partition functions are equal. Then we will use the fact
that both partition functions are -functions of Toda lattice hierarchy, which implies that
they also coincide in presence of an arbitrary perturbation.
We also give a unified geometrical description of the two models in terms of a complex
curve with the topology of a sphere with two punctures. This complex curve is analog of the
Riemann surfaces arising in the case of Hermitian matrix models. However, for generic R it
is not an algebraic curve. The curve has two dual non-contractible cycles, which determine
the boundaries of the supports of the eigenvalue distributions for the two models. The
normal matrix model is associated with the compact cycle while the MQM is associated
with the non-compact cycle connecting the two punctures. We will construct a globally
defined one-form whose integrals along the two cycles give the number of eigenvalues N
and the derivative of the free energy with respect to N .
The paper is organized as follows. In the next section we will remind the realization of
the c = 1 string theory as a fermionic system with chiral perturbations worked out in [11].
We will stress on the calculation of the free energy, which will be needed to establish the
equivalence with the NMM. In Section 3 we construct the NMM having the same partition
function. In Section 4 we consider the quasiclassical limit and give a unified geometrical
description of the two models.
2. Matrix quantum mechanics, free fermions and integrability

2.1. Eigenfunctions and fermionic scattering
In absence of winding modes, the 2D string theory is described by the singlet sector of
matrix quantum mechanics in the double scaling limit with Hamiltonian

2
0 = 1 tr h 2 X2 .
H
(2.1)
2
X2
The radial part of this Hamiltonian is expressed in terms of the eigenvalues x1 , . . . , xN
of the matrix X. The wave functions in the SU(N)-singlet sector are completely
antisymmetric and thus describe a system of N non-relativistic fermions in the inverse
93
Gaussian potential. The dynamics of the fermions is governed by the Hamiltonian

0 = 1
p i2 xi2 ,
H
2
N
(2.2)
i=1
where pi are the momenta conjugated to the fermionic coordinates xi .

To describe the incoming and outgoing tachyonic states, it is convenient to introduce
the light-cone coordinates in the phase space
x p
x =
2
(2.3)
satisfying the canonical commutation relations

[x+ , x ] = i h .
(2.4)
In these variables the one-particle Hamiltonian takes the form

0 = 1 (x+ x + x x+ ).
H
(2.5)
2
We can work either in x+ or in x representation, where the theory is defined in terms of
E (x , t) =
fermionic fields (x ), respectively. The solutions with a given energy are
E (x ) with
e h Et
i
i E 1
1
E
e 2h 0 x h 2 ,
(x ) =
2 h
(2.6)
where the phase factor 0 (E) will be determined below. These solutions form two complete
systems of -function normalized orthonormal states under the condition that the domain
of the definition of the wave functions is a semi-axis.4 The left and right representations are
related to each other by a unitary operator which is the Fourier transform on the half-line
[
S+ ](x ) =

dx+ S(x , x+ )+ (x+ ).
(2.7)
The integration kernel S(x , x+ ) is either a sine or a cosine depending on the boundary
conditions at the origin. For definiteness, let us choose the cosine kernel

1
2
S(x , x+ ) =
(2.8)
cos x+ x .
h
h
4 In order to completely define the theory, we should define the interval where the phase-space coordinates
x are allowed to take their values. One possibility is to allow the eigenvalues to take any real value. The whole
real axis corresponds to a theory whose Fermi sea consists of two disconnected components on both sides of the
maximum of the potential. To avoid the technicalities related to the tunneling phenomena between the two Fermi
seas, we will define the theory by restricting the eigenvalues to the positive real axis. The difference between the
two choices is seen only at the non-perturbative level.
94
Then the operator

S is diagonal on the eigenfunctions (2.6) with given energy
E
i
E

(x ),
S+ (x ) = e h 0 R(E)
(2.9)
where

R(E) = h
iE

1
i
2
i
1
cosh

.
E
E+
2 h
2
2
h
(2.10)
Since the operator

S is unitary, i.e.,
R(E)R(E) = R(E)R(E) = 1,
(2.11)
it can be absorbed into the function 0 (E). This fixes the phase of the eigenfunctions as
0 (E) = i h log R(E).
(2.12)
In fact, the operator

S appears in our formalism as the fermionic S-matrix describing
the scattering off the inverse oscillator potential and the factor R(E) gives the fermionic
reflection coefficient.
Let us introduce the scalar product between left and right states as follows
|
S|+ =

dx+ dx (x )S(x , x+ )+ (x+ ).
(2.13)
Then the in- and out-eigenfunctions are orthonormal with respect to this scalar product

E E
S + = (E E ).
(2.14)
It is actually this relation that defines the phase factor containing all information about the
scattering.
The ground state of the Hamiltonian (2.5) can be constructed from the wave functions
(2.6) and describes the linear dilaton background of string theory [10]. The tachyon
perturbations can be introduced by changing the asymptotics of the wave functions at
x to
i
E (x ) e h
k/R
tk x
i E 12
x h
(2.15)
The exact phase contains also a constant mode 21h (E) as in (2.6) and negative powers
1/R
of x
which are determined by the orthonormality condition
E
E
S + = (E E ).
(2.16)
The constant mode (E) of the phase of the fermion wave function contains all essential
information about the perturbed system.
95
2.2. Cut-off prescription and density of states

To find the density of states, we introduce a cut-off by confining the phase space to a
periodic box
x + 2 x ,
(2.17)
which can be
interpreted as putting
a reflecting wall at distance . This means that at the
points x+ = and x = the reflected wave function coincides with the incoming
one
[
S ]( ) = ( ).
(2.18)
Applied to the wave functions (2.15), this condition gives an equation for the admissible
energies

i
i
i
V () =
(tk + tk )k/2R ,
e h (E) = e h V () h E ,
(2.19)
where we neglected the negative powers of . It is satisfied by a discrete set of energies
En defined by the relation
En log (En ) = 2 h n + V (),
n Z.
(2.20)
Taking the limit we find the density of states (in units of h )

(E) =
1 d(E)
log
.
2
2 dE
(2.21)
2.3. The partition function of the compactified 2D string theory

Knowing the density of states, we can calculate the grand canonical partition function
Z of the quantum-mechanical system compactified at Euclidean time = 2R. If is the
chemical potential, then the free energy F = h 2 log Z of the ensemble of free fermions is
given by

F (, t) = h
1
dE (E) log 1 + e h (+E) .
(2.22)
Integrating by parts and dropping out the -dependent non-universal piece, we obtain
F (, t) =
2

dE
(E)
1
1 + e h (+E)
(2.23)
We close the contour of integration in the upper half plane and take the integral as a sum
of residues of the thermal factor5

n+ 1
2
.
F (, t) = i h
i h
(2.24)
R
n0
5 We neglect the non-perturbative terms associated with the cuts of the function (E).
96
We will be interested only in the perturbative expansion in 1/ and neglect the non
perturbative terms e h , as well as non-universal terms in the free energy (regular in ).
Therefore, for the case of zero tachyon couplings we can retain only the -function in the
reflection factor (2.10) and choose the phase 0 as

1
i
E+
0 (E) = i h log
(2.25)
.
h
2
It is known [12] that the tachyon scattering data for MQM are generated by a -function
of Toda lattice hierarchy where the coupling constants tk play the role of the Toda times.
The -function is related to the constant mode (E) by [11]
i
e h () =
h
)
0 ( + i 2R
h
0 ( i 2R
)
(2.26)
Taking into account (2.24) we conclude that the partition function of the perturbed MQM
is equal to the -function:
Z(, t) = 0 (, t).
(2.27)
The discrete space parameter s Z along the Toda chain is related to the chemical
potential . More precisely, s corresponds to an imaginary shift of [6,11]:

s
s (, t) = 0 + i h , t .
(2.28)
R
This fact was used to rewrite the Toda equations as difference equations in rather than in
the discrete parameter s. It will be also used below to prove the equivalence of 2D string
theory to the normal matrix model.
3. 2D string theory as normal matrix model

In this section we will show that the partition function of 2D string theory with tachyonic
perturbations can be rewritten as a normal matrix model. It is related to the original matrix
quantum mechanics by a kind of duality transformation. In fact, we will give two slightly
different normal matrix models fulfilling this goal.
Consider the following matrix integral

(R1)/2+
NMM
h
dZ dZ det ZZ
Zh (N, t, ) =
[Z,Z ]=0
e h tr[(ZZ
1
)R V
+ (Z)V (Z
where the integral goes over all complex matrices satisfying

are given by

V (z) =
tk zk .
k1
)]
[Z, Z ] = 0
(3.1)
and the potentials
(3.2)
97
We made the dependence on the Planck constant explicit because later we will need to
analytically continue in h . The integral (3.1) defines a -function of Toda hierarchy [20].
We remind the proof of this statement in Appendix A, where we also derive the string
equation specifying the unique solution of Toda equations. This string equation coincides
(up to change h i h ) with that of the hierarchy describing the 2D string theory [6,11].
Therefore, if we identify correctly the parameters of the two models, the -functions should
also be the same. Namely, we should relate the chemical potential of 2D string theory
to the size of matrices N and the parameter of the normal matrix model. There are two
possibilities to make such identification.
3.1. Model I
The first possibility is realized taking a large N limit of the matrix integral (3.1).
Namely, we will prove that the full perturbed partition function of MQM is given by the
large N limit of the partition function (3.1) with = R hN
and a subsequent analytical
continuation h i h
NMM
Zh (, t) = lim Zi h (N, t, R i h N).

N
(3.3)
The necessity to change the Planck constant by the imaginary one follows from the
comparison of the string equations of two models as was discussed above. More generally,
the -function (2.28) for arbitrary s is obtained from the partition function (3.1) in the
following way
NMM
s,h (, t) = lim Zi h (N + s, t, R i h N).

N
(3.4)
Let us stress that in spite of the large N limit taken here, we obtain as a result the full (and
not only dispersionless) partition function of the c = 1 string theory.
To prove this statement we will use the fact that both partition functions are -functions
of the Toda lattice hierarchy with times tk , k = 1, 2, . . . . Since the unperturbed -function
provides the necessary boundary conditions for the unique solution of Toda equations, it is
sufficient to show that the unperturbed partition functions of the two models coincide and
the s-parameters of two -functions are identical.
The integral (3.1) reduces to the product of the normalization coefficients of the
orthogonal polynomials, Eq. (A.5). From the definition of the scalar product (A.4) it
follows that when all tk = 0 the orthogonal polynomials are simple monomials and the
normalization factors hn are given by

1
R
1
hn (0, ) =
(3.5)
d 2 z e h (zz) (zz)(R1)/2+ h +n .
2i
C
Up to constant factors and powers of h , we find

n+
+
hn (0, )
h R
R
1
2

1
+
.
2
(3.6)
98
We see that the result coincides with the reflection coefficient given in (2.25). Therefore,
taking the analytical continuation of parameters as in (3.3), we have
NMM
lim h log Zi h
2

N n
(N, 0, R i h N) = lim i h
0 i h
N
R
n=0

n + 12
= i h
.
0 i h
R
N1
1
2
(3.7)
n=0
This coincides with the unperturbed free energy F (, 0) from (2.24), what proves (3.3) for
all tk = 0.
Thus, it remains to show that the s-parameter of the -function describing the normal
matrix model is associated with . From (A.17) we see that it coincides with N . Hence, it
trivially follows from (3.4) that

s
s
NMM
s,h (, t) = lim Zi h
N, t, R + i h
i h N = 0,h + i h , t , (3.8)
N
R
R
what means that the -function defined in (3.4) possesses the characteristic property (2.28).
3.2. Model II
In fact, one can even simplify the representation (3.3) what will provide us with the
second realization of 2D string theory in terms of NMM. Let us consider the matrix model
(3.1) for = 0. We claim that the partition function Zi h (N, t, 0) in the canonical ensemble,
analytically continued to imaginary Planck constant and
i
N = R,
h
(3.9)
coincides with the partition function of 2D string theory in the grand canonical ensemble:6

i
NMM
R, t .
Zh (, t) = Zi h
(3.10)
h
Indeed, as for the first model, they are both given by -functions of Toda hierarchy.
After the identification (3.9), the s-parameters of these -functions are identical due to
(A.17) and (2.28). Therefore, it only remains to show that the integral Zi h (N, 0) without
potential is equal to the unperturbed partition function (2.24). In this case the method of
orthogonal polynomials gives
NMM
Zi h (N, 0) =
N1

n=0

n + 12
,
R i h
R
6 From now on we will omit the last argument of Z NMM corresponding to the vanishing parameter .
(3.11)
99
where the R factors are the same as in MQM, as was shown in (3.6). Then we write
NMM
Zi h (N, 0) = (0)/ (N),
where (N) =
n=N

1
R i h n +
R . (3.12)
2
(0) is a constant and can be neglected, whereas (N) can be rewritten as

(N) =

n=0

1
R .
R i h N/R i h n +
2
(3.13)
Taking into account the unitarity relation (2.11) and substituting N from (3.9), we obtain
NMM
Zi h

1
i
i
R + i h n +
R, 0 1 R =
R .
2
h
h
n=0
(3.14)
NMM
Thus h 2 log Zi h ( hi R, 0) coincides with the free energy (2.24) with all tk = 0. Note
that the difference in the sign of does not matter since the free energy is an even function
of (up non-universal terms).7 Since the two partition functions are both solutions of the
Toda hierarchy, the fact that they coincide at tk = 0 implies that they coincide for arbitrary
perturbation.
We conclude that the grand canonical partition function of the c = 1 string theory
equals the canonical partition function of the normal matrix model (3.1) for = 0 and
N = hi R. It is also clear that we can identify the operators of tachyons in two models
n/R
Tr X+
Tr Z n ,
n/R
Tr X
n
Tr Z .
(3.15)
4. Geometrical meaning of the duality between the two models

The quasiclassical limit of most of the solvable matrix models has a nice geometrical
interpretation. Namely, the free energy in this limit can be parameterized in terms of the
periods of a holomorphic 1-form around the cycles of an analytical curve [2225]. Recently
this geometrical picture appeared also in the context of the topological strings on singular
CalabiYau manifolds and supersymmetric gauge theories [2628].
In this section we will show that both MQM and NMM have in the quasiclassical limit
a similar geometrical interpretation in terms of a one-dimensional complex curve, which
is topologically a sphere with two punctures. Here by complex curve we understand a
complex manifold with punctures and given behavior of the functions on this manifold at
the punctures. Each of the two matrix models corresponds to a particular real section of
this curve coinciding with one of its two non-contractible cycles, and the duality between
them is realized by the exchange of the cycles of the curve.
7 Moreover, this sign can be correctly reproduced from (2.23) if closing the integration contour in the lower
half plane.
100
4.1. The dispersionless limit

The quasiclassical limit h 0 corresponds to the dispersionless limit of the Toda
hierarchy where it has a description in terms of a classical dynamical system. The Lax
operators are considered as c-functions of the two canonically conjugated coordinates
and , where is the quasiclassical limit of the shift operator = e/s . Moreover, the two
Lax operators can be expressed as series in . The solutions of the dispersionless hierarchy
correspond to canonical transformations in the phase space of the dynamical system. In the
case of MQM this is the transformation from the coordinates and log to x+ and x .
The particular solution of the Toda hierarchy that appears in our problem also satisfies
the dispersionless string equation. In the case of MQM this is nothing but the equation of
the profile of the Fermi sea in the phase space [11]
1
1
k/R
k/R
ktk x + +
vk x .
x+ x =
(4.1)
R
R
k1
k1
To get the string equation for NMM it is enough to make the substitution following (3.15)
x+ z R ,
x z R
(4.2)
and = h N/R as explained in the previous section. (We took into account that h from
(3.9) should be replaced here by i h so that the factor i is canceled.) Then the string
equation describes the contour bounding the region D in the complex z-plane filled by
the eigenvalues of the normal matrix.8
4.2. NMM in terms of electrostatic potential
In this section we will introduce a function of the spectral variables z and z , which plays
a central role in the NMM integrable structure. This function has several interpretations in
the quasiclassical limit. First, it can be viewed as the generating function for the canonical
transformation mentioned above, which maps variables and log to the variables z and
z . Second, it gives the phase of the fermion wave function at the Fermi level after the
identification (4.2).
There is also a third, electrostatic interpretation, which is geometrically the most explicit
and which we will follow in this section.9 According to this interpretation, the eigenvalues
distributed in the domain D can be considered to form a charged liquid with the density
R2
1
z z WR (z, z ) =
(zz)R1 .
(4.3)
Let (z, z ) be the potential of the charged eigenvalue liquid, which is a harmonic function
outside the domain D and it is a solution of the Laplace equation inside the domain

(z) + (
z), z
/ D,
(z, z ) =
(4.4)
z D.
(zz)R ,
(z, z ) =
8 Here we consider the case when the eigenvalues are distributed in a simply connected domain.
9 See also [29] for another interesting interpretation which appears useful in string field theory.
101
To fix completely the potential, we should also impose the asymptotics of the potential at
infinity. This asymptotics is determined by the coupling constants tn , n = 1, 2, . . . and can
be considered as the result of placing a dipole, quadrupole, etc. charges at infinity.
The solution of this electrostatic problem is obtained as follows. The continuity of the
potential (z, z ) and its first derivatives leads to the following conditions to be satisfied on
the boundary = D
(z) + (
z) = (zz)R ,
(4.5)
zz (z) = z z (
z) = RzR z R .
(4.6)
Each of two equations (4.6) can be interpreted as an equation for the contour . Since we
obtain two equations for one curve, they should be compatible. This imposes a restriction
on the holomorphic functions (z) and (
z). The solutions for these chiral parts of the
potential can be found comparing (4.6) with the string equation (4.1). In this way we have

1
1
(z) = h N log z + +
vk zk ,
tk zk
2
k
k1
k1

1
1
(
z) = h N log z + +
vk z k .
tk z k
2
k
k1
(4.7)
k1
The zero mode is fixed by the condition (4.5). However, it is easier to use another
interpretation. As we mentioned above, the holomorphic functions (z) and (
z) coincide
with the phases of the fermionic wave functions and in particular their constant modes are
the same. Then, taking the limit h 0 of (2.26), in the variables of NMM we find
= h
NMM
.
log Z
N
(4.8)
4.3. The complex curve

= C/D,
Instead of considering the function (z, z ) harmonic in the exterior domain D
The
we can introduce one holomorphic function defined in the double cover of D.
is topologically equivalent to a two-sphere with two punctures
doubly covered domain D
(at z = and z = ), which we identify with the north and the south poles. It can be
covered by two coordinate patches associated with the north and south hemispheres and
parameterized by z and z . They induce a natural complex structure so that the sphere can
be viewed as a complex manifold. The patches overlap in a ring containing the contour
and the transition function between them z (z) (or z(z)) is defined through the string
equation (4.6).
The string equation can also be considered as a defining equation for the manifold. We
start with the complex plane C2 with flat coordinates z and z . Then the manifold is defined
as solution of Eq. (4.6), thus providing its natural embedding in C2 .
To completely define the complex curve, we should also specify the singular behavior
at the punctures of the analytic functions on this curve. We fix it to be the same as the one
of the holomorphic parts of the potential (4.7).
102
We define on the complex curve the following holomorphic field

R
1
z(z) ,
in the north hemisphere,
def + (z) = (z) 2 z
=

R
1
z) + 2 z(z)z , in the south hemisphere.
(z) = (
(4.9)
Its analyticity follows from Eq. (4.5), which now holds on the entire curve, since z and z
are no more considered as conjugated to each other.
The field gives rise to a closed (but not exact) holomorphic 1-form d, globally
defined on the complex curve. Actually, this is the unique globally defined 1-form with
given singular behavior at the two punctures. Therefore, we can characterize the curve
by periods of this 1-form around two conjugated cycles A and B defined as follows. The
cycle A goes along the ring where both parameterizations overlap and is homotopic to the
contour . The cycle B is a path going from the puncture at z = to the puncture at
z = .
The integral around the cycle A is easy to calculate using (4.7) and (4.6) by picking up
the pole

1
1
1
(4.10)
d =
d+ (z) =
d (z) = h N,
2i
2i
2i
1
where we indicated that in z -coordinates the contour should be reversed.

To find the integral along
cycle B, one should introduce a regulariza the non-compact
tion by cutting it at z = and z = . Then we obtain

d =
B
z0

d+ (z) +
d (z) = ( ) + ( ),
(4.11)
z (z0 )
where z0 is any point on the cycle and we used (4.5). Taking into account the definition
(4.9) and the explicit solution (4.7), we see that the result contains the part vanishing at
, the diverging part which does not depend on N and, as non-universal, can be
neglected, and the contribution of the zero mode . Since enters + and with
different signs, strictly speaking, it is not a zero mode for . As a result, it does not
disappear from the integral, but is doubled. Finally, using (4.8) we get

NMM
(4.12)
log Z
d = h
.
N
B
In Appendix B we derive this formula using the eigenvalue transfer procedure similar to
one used in [28], or in [25] in a similar case of the two matrix model equivalent to our
R = 1 NMM. Thus, the free energy of NMM can be reproduced from the monodromy of
the holomorphic 1-form around the non-compact cycle on the punctured sphere.
Since the string equations coincide, it is clear that the solution of MQM is described by
the same analytical curve and holomorphic 1-form. To write the period integrals in terms
of MQM variables, it is enough to change the coordinates according to (4.2) and h N = R
in all formulas. In this way we find10

1
1 F
d = R,
d =
.
2i
R
A
103
(4.13)
The last integral can be also obtained from the integral over the Fermi sea. Indeed, we
can choose a point on the contour of the Fermi sea and split the integral, which actually
calculates the area of the sea, into three parts
1 F
=
R
F.s.

dx+ dx =
x (x+ ) dx+ +
x+ (x ) dx + x0 x (x0 ).
x0
(4.14)
x (x0 )
Since + = x (x+ ) and = x+ (x ) (see (4.6)), the last expression is equal to the
integral (4.11) of d around the (reversed) cycle B. This derivation gives an independent
check that the free energies of NMM and MQM do coincide.
4.4. Duality
We found that the solutions of both models are described by the same complex curve.
The curve is characterized by a pair of conjugated cycles: the compact cycle A encircling
one of the punctures and the non-compact cycle B connecting the two punctures. The
parameter of the free energy or N and the derivative of the free energy itself are given by
the integrals of the unique globally defined holomorphic 1-form along the A and B cycles,
correspondingly.
Now one can ask: what is the difference between the two models? Is it seen at the level
of the curve? In fact, both models, NMM and MQM can be associated with two different
real sections of the complex curve. Indeed, in NMM the variables are conjugated to each
other, whereas in MQM they are real. Therefore, let us take the interpretation where the
curve is embedded into C2 and consider its intersection with two planes. The first plane is
defined by the condition z = z and the second one is given by z = z, z = z . In the former
case we get the cycle A, whereas in the latter case the intersection coincides with the cycle
B (see Fig. 1). One can think about the planes as the place where the eigenvalues of NMM
and MQM, correspondingly, live (with the density given by (4.3)). Then the intersections
describe the contours of the regions filled by the eigenvalues. We see that for NMM it is
given by the compact cycle A, and for MQM the Fermi sea is bounded by the cycle B of
the curve.
Therefore, the duality between the two models can be interpreted as the duality
exchanging the cycles of the complex curve describing the solution. Under this exchange
the real variables of one model go to the complex conjugated variables of another model,
and the grand canonical free energy is replaced by the canonical one.
This duality can be seen even more explicitly if one rewrites the relations (4.13) in terms
F is
of the canonical free energy of MQM defined as F = F + h RM, where M = h1R
10 To get the second integral from (4.12), one should take into account that also h i h.
104
Fig. 1. Symbolic representation of the complex curve and the two real sections along the cycles A and B. The
filled regions symbolize the Fermi seas of the two matrix models.
the number of eigenvalues. Then they take the following form

1 F
1
,
d =
d = h M.
h M
2i
A
(4.15)
They have exactly the same form as (4.10) and (4.12) provided we change the cycles. In
particular, the numbers of eigenvalues are given in the two models as integrals around the
dual cycles.
5. Conclusions and problems

We have shown that the tachyonic sector of the compactified c = 1 string theory is
described in terms of a normal matrix model. This matrix model is related by duality to
the traditional representation in terms of matrix quantum mechanics. More precisely, the
canonical partition function of NMM is equal to the grand canonical partition function
of MQM, provided we identify the number of eigenvalues N of NMM with the chemical
potential in MQM. This holds for any compactification radius. Moreover, the duality
between the two matrix models has a nice geometrical interpretation. In the quasiclassical
limit the solutions of the two models can be described in terms of a complex curve with the
topology of a sphere with two punctures. The matrix quantum mechanics and the normal
matrix model are associated with two real sections of this curve which are given by the two
non-contractible cycles on the punctured sphere. The duality acts by exchanging the two
cycles.
The duality we have observed relates a fermionic problem with non-compact Fermi sea
and continuous spectrum to a problem with compact Fermi sea and discrete spectrum. The
105
solutions of the two problems are obtained from each other by analytic continuation. Thus,
the reason for the duality can be seen in the common analytic structure of both problems.
We also emphasize that the duality relates the partition functions in the canonical and
the grand canonical ensembles, which are related by Legendre transform. This is natural
regarding that it exchanges the cycles of the complex curve.
Let us mention two among the many unsolved problems related to the duality between
these two models. First, it would be interesting to generalize the above analysis to the case
when the eigenvalues of NMM form two or more disconnected droplets. For R = 1 this is
the analog of the multicut solutions of the two-matrix model, considered recently in [25].
In terms of MQM this situation would mean the appearance of new, compact components
of the Fermi sea. The corresponding complex curve has the topology of a sphere with two
punctures and a number of handles.
Second, we would like to generalize the correspondence between the two matrix models
in such a way that it incorporates also the winding modes. For this we should understand
how to introduce the winding modes in the normal matrix model. Whereas in MQM they
appear when we relax the projection to the singlet sector, in NMM this could happen when
= 0.
we relax the normality condition [Z, Z]
Up to now there were two suggestions how to include both the tachyon and winding
modes within a single matrix model. In [11], a 3-matrix model was proposed with
interacting two Hermitian matrices and one unitary matrix. This model can be seen as
a particular reduction of Euclidean compactified MQM. The model correctly describes the
cases of only tachyon or only winding perturbations, though its validity in the general case
is still to be proven. For the particular case of the self-dual radius and the multiples of it,
a 4-matrix model was recently proposed in [8]. It is based on the old observation of [30]
about the geometry of the ground ring of c = 1 string theory. It would be interesting to find
the relation of this model to our approach, at least in the above-mentioned particular cases.
The general understanding of this important problem is still missing.
Acknowledgements
We are grateful to M. Aganagic, R. Dijkgraaf, S. Gukov, M. Mario, C. Vafa and
P. Wiegmann for valuable discussions. We thank for the hospitality Jefferson Physical
Laboratory of Harvard University, where a part of this work has been done. The work of
S.A. and V.K. was partially supported by European Union under the RTN contracts HPRNCT-2000-00122 and -00131. The work of S.A. and I.K was supported in part by European
networks EUROGRID HPRN-CT-1999-00161 and EUCLID HPRN-CT-2002-00325.
Appendix A. Toda description of the perturbed system

Here we show that the partition function (3.1) is a -function of the Toda lattice
hierarchy (for details see [31]). We will do it following the standard method of orthogonal
106
polynomials. First, we write the integral (3.1) as integral over the eigenvalues z1 , . . . , zN

N
1
NMM
Zh (N, t, ) =
w (zk , z k )(z)(z)
(A.1)
N!
k=1
with measure
d 2 z 1 [(zz)R V+ (z)V(z)]
e h
(zz)(R1)/2+ h .
2i
Then we introduce a set of bi-orthogonal polynomials

n
n

w (zk , z k )
1
n
P+ (z) =
(z)(z) (z zk ),
n! hn k=1 hk1 (t, )
k=1
w (z, z ) =
and similarly for Pn (z). Here (z) =
(A.2)
(A.3)
(zk zl ) is the Vandermond determinant and
k<l
normalization factors hk are determined by the orthonormality condition

n m

P P+ w (z, z ) Pn (z)P+m (z) = nm .
w
(A.4)
Then the partition function (A.1) reduces to the product of the normalization factors
NMM
Zh
(N, t, ) =
N1

hn (t, ).
(A.5)
n=0
The operators of multiplication by z and z are represented in the basis of orthogonal

polynomials by the infinite matrices

Lnm P+m (z),
z Pn (z) =
Pm (z)L mn
zP+n (z) =
(A.6)
m
with

Ln,n+1 = L n+1,n = hn / hn+1 ,
Lnm = L mn = 0, m > n + 1.
(A.7)
Differentiation of the orthogonality relation (A.4) with respect to coupling constants tk

gives
h
n1

k
P+n (z)
1
L nm P+m (z) Lk nn P+n (z),
=
tk
2
m=0
n1

k
P+n (z)
1
L nm P+m (z) Lk nn P+n (z)
=
tk
2
(A.8)
m=0
and similarly for Pn (z). Let us define now a wave function which is a vector = {n }
with elements
1
n (t; z) = P+n (z)e h V+ (z) .
(A.9)
107
Then Eq. (A.8) lead to the following eigenvalue problem

z = L,
= Hk ,
tk
= Hk ,
tk
(A.10)
where we introduced the Hamiltonians

1
1
Hk = Lk + + Lk 0 ,
(A.11)
Hk = L k L k 0 .
2
2
Here the subscripts 0/ /+ denote diagonal/lower/upper triangular parts of the matrix.
From the commutativity of the second derivatives, it is easy to find the ZakharovShabat
zero-curvature condition
Hk
Hl
h
(A.12)
h
+ [Hk , Hl ] = 0.
tl
tk
It means that the Hamiltonians generate commuting flows and the perturbed system is
described by the Toda lattice hierarchy.
In addition, one can obtain the string equation for the hierarchy. It follows from the
Ward identity

R R R
= L L (R + 1)/2 +
z
(A.13)
z
h
h
and can be written as
R R
L , L = h.
(A.14)
Here we should understand the operators LR as analytical continuation of the operators in

integer powers.
The Toda structure leads to an infinite set of PDEs for the coefficients of the operators
The first of these equations can be written for the normalization factors and is
L and L.
known as Toda equation
h 2

hn
hn+1
log hn =
.
t1 t1
hn1
hn
(A.15)
The quantity hn is known to be related to the -function of the Toda hierarchy by

hn =
NMM
n+1
nNMM
(A.16)
Taking into account (A.5), this gives for the partition function
NMM
Zh
NMM
(N, t, ) = N,
h (t, ).
(A.17)
Appendix B. Derivation of the formula for the cycle B

The formula (4.12) can be obtained by means of the procedure of transfer of an
eigenvalue from the point (z1
, z 1 = z (z1 )) belonging to the boundary of the spot to
(or rather to the cut-off point ), worked out for the case R = 1 in [25]. On the double
108
cover described in Subsection 4.3, z1 and z 1 are considered as complex coordinates of the
point in two patches. We find from (A.1) that the free energy in the large N limit changes
during this transfer as follows:
h
N

NMM
log Z
= (z1 z 1 )R + V+ (z1 ) + V (z1 ) + h
log (z1 zm )(z1 z m ) ,
N
m=2
(B.1)
where all zm s are taken at their saddle point values. Here we took into account that the
determinant in the matrix integral (3.1) does not contribute in the quasiclassical limit. In
this limit the integral (A.1) leads to the following saddle point equations [25]
z),
RzR z R1 V (z) = G(
RzR1 z R V+ (z) = G(z),
(B.2)
where G(z) and G(z) are the resolvents of the one-dimensional distributions of zk s and
z k s, respectively.11 They have the following asymptotics at large z and z :
h N
z) h N .
G(z)
(B.3)
,
G(
z
z
Rewriting the sum in (B.1) as an integral with the measure given by the eigenvalue density
and then expressing it through the resolvents, we find

dz
NMM
log Z
G(z) log(z1 z)
= (z1 z 1 )R + V+ (z1 ) + V (z1 ) +
h
N
2i

d z
+
(B.4)
G(z) log(z1 z ),
2i
where the contours of integration encircle the whole support of the distribution of
eigenvalues on the physical sheets of the functions z (z) and z(z), respectively.12
Due to (B.2) we do not have any singularities at z (z ) except poles. Blowing
up the contour we pick up the logarithmic cut and obtain
NMM
log Z
= (z1 z 1 )R R
N
dz
z
zz(z)
R

R
z1
z 1
R
d z
z(z)z .
z
(B.5)
The first term in the r.h.s. can be regrouped with the other two terms, giving (up to a nonuniversal term R )

z1

R dz 1
NMM
R
h
=
log Z
d(zz)
R zz(z)
N
z
2

R d z 1
R
d(zz) ,
R z(z)z
z
2
(B.6)
z 1
11 We view z (z) and z(z) as analytical functions similarly to the interpretation of Subsection 4.3.
12 Note that the point of infinite branching at the origin related to the singularities of zR and z R should not
appear on the physical sheets.
109
which immediately yields (4.11) if we take into account (4.6) and the definition (4.9).
Note that (B.2) is essentially the same as (4.6) if one uses the explicit expansion (4.7) for
the holomorphic parts of the potential. Therefore, they define the same functions z (z) and
z(z). On the complex curve described in the Subsection 4.3, the formula (B.6) reduces to
the period integral (4.12) of the holomorphic differential d.
References
[1] I. Klebanov, Lectures delivered at the ICTP Spring School on String Theory and Quantum Gravity, Trieste,
April 1991, hep-th/9108019.
[2] V.A. Kazakov, A.A. Migdal, Recent progress in the non-critical strings, Nucl. Phys. B 311 (1988) 171.
[3] V. Kazakov, I. Kostov, D. Kutasov, A matrix model for the two-dimensional black hole, Nucl. Phys. B 622
(2002) 141, hep-th/0101011.
[4] S. Alexandrov, V. Kazakov, Correlators in 2D string theory with vortex condensation, Nucl. Phys. B 610
(2001) 77, hep-th/0104094.
[5] S. Alexandrov, V. Kazakov, Thermodynamics of 2D string theory, JHEP 0301 (2003) 078, hep-th/0210251.
[6] I. Kostov, String equation for string theory on a circle, Nucl. Phys. B 624 (2002) 146, hep-th/0107247.
[7] D. Ghoshal, C. Vafa, c = 1 string as the topological theory of the conifold, Nucl. Phys. B 453 (1995) 121,
hep-th/9506122.
[8] R. Dijkgraaf, C. Vafa, N = 1 supersymmetry, deconstruction and bosonic gauge theories, hep-th/0302011.
[9] J. Polchinski, What is string theory, Lectures presented at the 1994 Les Houches Summer School Fluctuating
Geometries in Statistical Mechanics and Field Theory, hep-th/9411028.
[10] A. Jevicki, Developments in 2D string theory, hep-th/9309115.
[11] S.Yu. Alexandrov, V.A. Kazakov, I.K. Kostov, Time-dependent backgrounds of 2D string theory, Nucl.
Phys. B 640 (2002) 119, hep-th/0205079.
[12] R. Dijkgraaf, G. Moore, M.R. Plesser, The partition function of 2D string theory, Nucl. Phys. B 394 (1993)
356, hep-th/9208031.
[13] K. Takasaki, T. Takebe, Integrable hierarchies and dispersionless limit, Rev. Math. Phys. 7 (1995) 743, hepth/9405096.
[14] S. Mukhi, C. Vafa, Two-dimensional black-hole as a topological coset model of c = 1 string theory, Nucl.
Phys. B 407 (1993) 667, hep-th/9301083.
[15] C. Imbimbo, S. Mukhi, The topological matrix model of c = 1 string, Nucl. Phys. B 449 (1995) 553, hepth/9505127.
[16] I. Kostov, I. Krichever, M. Mineev-Veinstein, P. Wiegmann, A. Zabrodin, -function for analytic curves,
in: Random Matrices and Their Applications, in: MSRI Publ., Vol. 40, Cambridge Univ. Press, Cambridge,
2001, p. 285, hep-th/0005259.
[17] A. Zabrodin, Dispersionless limit of Hirota equations in some problems of complex analysis, Theor. Math.
Phys. 129 (2001) 1511;
A. Zabrodin, Theor. Math. Phys. 129 (2001) 239, math.CV/0104169.
[18] A. Boyarsky, A. Marshakov, O. Ruchhayskiy, P. Wiegmann, A. Zabrodin, On associativity equations in
dispersionless integrable hierarchies, Phys. Lett. B 515 (2001) 483, hep-th/0105260.
[19] P. Wiegmann, A. Zabrodin, Conformal maps and dispersionless integrable hierarchies, Commun. Math.
Phys. 213 (2000) 523, hep-th/9909147.
[20] P. Wiegmann, A. Zabrodin, Large scale correlations in normal and general non-Hermitian matrix ensembles,
hep-th/0210159.
[21] O. Agam, E. Bettelheim, P. Wiegmann, A. Zabrodin, Viscous fingering and a shape of an electronic droplet
in the quantum Hall regime, Phys. Rev. Lett. 88 (2002) 236802, cond-mat/0111333.
[22] F. David, Non-perturbative effects in matrix models and vacua of two-dimensional gravity, Phys. Lett. B 302
(1993) 403, hep-th/9212106.
[23] G. Bonnet, F. David, B. Eynard, Breakdown of universality in multi-cut matrix models, J. Phys. A 33 (2000)
6739, cond-mat/0003324.
110
[24] I.K. Kostov, Conformal field theory techniques in random matrix models, hep-th/9907060.
[25] V. Kazakov, A. Marshakov, Complex curve of the two matrix model and its tau-function, hep-th/0211236.
[26] R. Dijkgraaf, C. Vafa, Matrix models, topological strings, and supersymmetric gauge theories, Nucl. Phys.
B 644 (2002) 3, hep-th/0206255.
[27] R. Dijkgraaf, C. Vafa, On geometry and matrix models, Nucl. Phys. B 644 (2002) 21, hep-th/0207106.
[28] R. Dijkgraaf, C. Vafa, A perturbative window into non-perturbative physics, hep-th/0208048.
[29] A. Boyarsky, O. Ruchayskiy, Integrability in SFT and new representation of KP tau-function, hepth/0211010.
[30] E. Witten, Ground ring of two-dimensional string theory, Nucl. Phys. B 373 (1992) 187, hep-th/9108004.
[31] M. Adler, P. van Moerbeke, String-orthogonal polynomials, string equations, and 2-Toda symmetries,
Commun. Pure Appl. Math. 50 (1997) 241, hep-th/9706182.

Low energy dynamics of self-dual A1 strings

Andreas Gustavsson
Institute of Theoretical Physics, Chalmers University of Technology, S-412 96 Gteborg, Sweden
Received 25 April 2003; received in revised form 26 May 2003; accepted 16 June 2003
Abstract
We examine the interrelation between the (2, 0) supersymmetric six-dimensional effective action
for the A1 theory, and the corresponding low-energy theory for the collective coordinates associated
to selfdual BPS strings. We argue that this low energy theory is a two-dimensional N = 4
supersymmetric sigma model.
2003 Published by Elsevier B.V.
PACS: 11.10.Lm
1. Introduction
We do not have a microscopic understanding of the interacting (2, 0) supersymmetric
quantum theories in six dimensions, although many years now have passed since they
where first discovered by Witten in 1995 [1]. But whatever the microscopic theory is, we
may always consider the low-energy effective action obtained by integrating out all massive
degrees of freedom, provided that we have given a non-zero vacuum expectation value to
say the fifth scalar field in the (2, 0)-supermultiplet, so that the selfdual strings and all other
subtle degrees of freedom become massive. We will in this letter only consider the A1
effective theory in which we have just one (2, 0)-supermultiplet, which on-shell consists
of a selfdual two-form gauge potential B + with selfdual field strength H + = dB + , five
scalars a , and a symplectic Majorana spinor (we do not know the off-shell multiplet if
any such exists).
The low energy dynamics of supersymmetric magnetic monopoles in super-YangMills
theories has been derived in [5] for N = 2 SYM and in [6] for N = 4 SYM. A natural
next step would be to examine the low energy dynamics of supersymmetric selfdual string
E-mail addresses: f93angu@fy.chalmers.se, a.r.qustavsson@swipnet.se (A. Gustavsson).
0550-3213/$ see front matter 2003 Published by Elsevier B.V.
doi:10.1016/S0550-3213(03)00533-9
112
A. Gustavsson / Nuclear Physics B 667 (2003) 111118
solitons in the (2, 0) supersymmetric theories in six dimensions. So far we have not derived
the fully supersymmetric effective action for the A1 theory. But already the terms we know
should be in that action can be used to derive a formula for the moduli space metric (which
requires only the kinetic terms in the effective action) and to derive that the moduli space
must be a hyper-Khler manifold.
The microscopic theory is of course far out of reach at present, contrary to the situation
for super-YangMills theories. We know that, for instance, the hedgehog monopole
solutions in SU(2) gauge theory can be gauge transformed so that only the massless scalar
field is non-zero. The analogue in six dimensions to the fields that have acquired mass due
to the non-zero vacuum expectation value of the Higgs scalar are not known. We think that
they correspond to the selfdual strings, but we have no microscopic understanding of these
objects. But in analogy with the four-dimensional case we will make the assumption that
there is a gauge in which the massive degrees of freedom (whatever they are) are zero
and the string soliton solutions are entirely described in terms of a massless scalar field 5
(and the gauge field B + ). The other scalar fields must then vanish, A = 0 (A = 1, 2, 3, 4)
for any 12 -BPS configuration as we showed in [7]. Furthermore we have the Bogomolnyi
equation H = d 5 for the spatial components of H and where is with respect to the
transverse space to the string world-sheets. These results should hold in any interacting
theory that possesses linear (2, 0) supersymmetry as explained in [7].
2. Effective action and collective coordinates

We will assume that we have flat six-dimensional Minkowski space and linear (2, 0)
supersymmetry. To construct a Hamiltonian invariant under this supersymmetry we can
anti-commute two supercharges given by

1
Q=
(1)
d 5 x MNP 0 HMNP + 2 d 5 x M 0 a M a .
6
Here M and a denote gamma matrices of SO(1, 5) and SO(5)R respectively and
H = dB + A. If we take A to be an external field we will get the free theory in which only
the selfdual part of H couples to A. To get the interacting A1 theory we take A = A() to
be a function of the five scalars. This will bring in a new term in the anti-commutator of
We will content ourselves with
two supercharges, arising from the commutator [A(), ].
constructing an action that is classically supersymmetric. Then commutators mean Poisson
A() . We take the gauge potential as in [7]1
brackets and [A, ]
A()MNP =
( 5 2||)
ABCD A M B N C P D .
2V4 ||3 ( 5 ||)2
1
(2)
Here || := ( a a ) 2 and A, B, . . . = 1, 2, 3, 4. V4 = 83 . Carrying out the above

construction we get the following supersymmetric A1 effective action (after a Legendre
1 This gauge potential A will get corrections if the five branes are not embedded as flat parallel planes in
eleven-dimensional spacetime.
transformation of the Hamiltonian),

1
1
H H d a d a + H A() + i T c M M
2
2

1 a
2
T
b
c
d MNP e
i cabcde 5 M N P
+ .
+
6V4
||
113
(3)
We have written + for terms involving four and more fermions, as well as terms that we
have to add in order to archive invariance under A A + d, B B . In particular
we must add the term A where d A = dA A dA [2]. This term should then also be
supersymmetrized. But the action is still supersymmetric if we skip the terms in + . This
action is also invariant under the Lorentz group SO(1, 5) and under a global R-symmetry
group SO(5)R . To make the SO(5)R symmetry manifest we must make an integration by
parts to get B dA(). The spinors transform in (4, 4) of SO(1, 5) SO(5)R and are
constrained by a symplectic Majorana condition
= T c,
(4)
where := 0 and where c and are charge conjugation matrices of SO(1, 5) and
T
SO(5), respectively. We choose the conventions so that M = c M c1 and a T =
a 1 .
The moduli spaces we will consider are those of k = 1, 2, . . . parallel BPS saturated
selfdual strings. These moduli spaces should be given by Mk R4 where the center of
mass moduli space is R4 and corresponds to the four transverse directions to the (center of
mass of the) strings. It seems very plausible that the dimension of the moduli space is 4k
(corresponding, in the classical picture of widely separated strings, to the four transverse
coordinates of each string). But to prove this one has to count, e.g., the number of fermionic
zero modes in a configuration with total charge 2k. Parallel BPS-saturated strings are
necessarily straight. To describe their motion tangent the moduli space we thus need
just one parameter, which we will take to be the time-coordinate. But the generic string
configuration is not a BPS configuration, and to describe such a configuration we also need
a parameter running along the strings, which we take to be x 5 . The tangent space to a BPS
string configuration should thus be a two-dimensional plane in the configuration space of
all physically permitted configurations. This should be the natural generalization of [3]. If
we let Xi (i = 1, . . . , dim Mk ) denote the bosonic moduli parameters on Mk and assume
that the BPS strings are aligned in the x 5 direction, then we make the following ansatz (we
restrict ourselves to the case that A = 0 for A = 1, 2, 3, 4),

5 x 0 , . . . , x 5 = 5 x 1 , . . . , x 4 , Xi x 0 , x 5 + ,

+ 0
+ 1
x , . . . , x 5 = BMN
x , . . . , x 4 , Xi x 0 , x 5 +
BMN
(5)
for the bosonic fields. Here (x 1 , . . . , x 4 , Xi ) (for constant Xi ) are the static BPS
solutions, which we assumed could be parametrized by Xi . The + involve terms that
has to be added in order to satisfy the equations of motion to higher orders in n = nt + nf ,
where nt are the number of time derivatives and nf the number of fermions, following
the logic of [4]. When we let the moduli parameters depend on x 0 , x 5 , they will be called
114
collective coordinates. We will use the indices , , . . . = 0, 5 for the string world-sheets,
and I = 1, 2, 3, 4 for the transverse coordinates.
For a single string, k = 1, the fermionic zero modes corresponds to broken supersymmetries u which are constant spinors such that 05 5 u = u . The zero modes are
given by
0 = I I 5 u .
(6)
That these really are zero modes is justified in Appendix A. We could promote the moduli
u to fermionic collective coordinates u just by letting them depend on x 0 , x 5 and expand
the fermion field as

= I I 5 x I , XI x 0 , x 5 u x 0 , x 5 .
(7)
However we want to label the fermionic and the bosonic collective coordinates with the
same label. That is, for k = 1 we want to convert u into I . u transforms in the
representation ( 12 , 2, 2 ) (+ 12 , 2 , 2) of SO(1, 1) SO(4) SO(4)R . We may find
one such u such that I := 5 I u constitute four linearly independent spinors. These
will be four two-component Majorana spinors under SO(1, 1). But we may choose other
representations for the SO(4) gamma matrices. If we transform I J JJ I then the

Clifford algebra is preserved if and only if JI I JJ J I J = I J . Demanding J to have
real entries (which is needed in order to preserve the hermiticity property of the gamma
matrices) we find that J is a unit quaternion. We thus have the freedom to make the
redefinition I J JJ I of the collective coordinates.
For a general k we make the following ansatz for the collective coordinates,

x M = i 5 x I , Xi x 0 , x 5 5 i x 0 , x 5
(8)
or with i replaced by Jj i j where Ji j is the quaternion induced by JI J .
Only for widely separated strings do we have the interpretation of Xi as being four
transverse coordinates of k solitonic strings. We have no proof that given above for a
constant spinor is a zero mode other than for widely separated strings (with a separation
1
much larger than 5 () 2 ). But, as we will see in the next section, if we make this ansatz
we get for instance a covariant derivative of in the low-energy action just as one should
expect. It seems very unlikely that there could be any other ansatz that also would give that
result. The ansatz we make combines bosonic and fermionic zero modes in such a simple
and natural way that it would be surprising if this formula did not give a fermionic zero
mode for constant in any point in the moduli space. So this is what we will assume, and
we will get no contradionary result when we assume this, which straightens our belief that
the ansatz for the fermionic collective coordinates is correct, not only for well separated
strings but everywhere in the moduli space.
3. Low energy dynamics

We get the action for the collective coordinates by expanding the fields about the BPS
configuration, inserting this expansion into the 6d Wilsonian low-energy effective action
(3) and integrating over the transverse coordinates.
115
When inserting any on-shell solution, H becomes self-dual and hence H H = 0.

Inserting the static fieldconfiguration
0. Then we
use the
equation of
we get A0MN =
motion dH = dA to get H A = R 2 R+ H S 3 A = R 2 R+ H S 3 H = d 5 dx 0

dx 5 S 3 H = dx 0 dx 5 d 5 H = dx 0 dx 5 QD where Q = 2k and D = 5 ().

Similarly we get, using the Bogomolnyi equation, that R 4 d 5 4 d 5 = d 5 H .
The total tension of the strings is 2QD with our
conventions. Finally the two-fermion
interaction term vanishes due to a factor I d 4 x x I = 0. Using these results when
inserting the collective coordinate expansions into (3), we get the action

1
T
dx 0 dx 5 Gij Xk Xi Xj + ii c D j 4kD + ,
2
(9)
where

Gij = d 4 x i 5 j 5
(10)
is the metric on the moduli space and
D i = i + kli Xk l ,
where
jik

=G
il

d 4 x j k 5 l 5 .
(11)
(12)
It is easily seen that jik is the Christoffel symbol associated with the metric Gij , so D is
a covariant derivative. The terms in + in (9) arise from those in + in (3).
To get this metric we relied on the assumption that all the massive degrees of freedom
can be put to zero in any string soliton solution. We think this is a plausible assumption
given the analogy with the situation for the t Hooft monopole in four dimensions where
one can find a gauge in which only the massless scalar field is non-zero. The string solution
we constructed in [7] also confirms that this is a valid assumption. A problem is that it is
not sufficient that this assumption is valid only at far distances from the strings, since we
integrated over the entire transverse space to the strings to get the metric above. So what
happens close to the strings cannot be ignored. Furthermore the form of the metric would
not cease to hold just because to strings happened to be close to each other since, as we
already have said, the ansatz we made for the collective coordinates we make everywhere
on the moduli space. Either the assumption we have made about the possibility of gauging
all the massive degrees of freedom to zero is correct everywhere, and then the metric is
given by (10) everywhere on the moduli space. Or else it is wrong (incomplete) everywhere
on the moduli space, and should be completed by adding some subtle degree of freedom
that have become massive upon Higgsing. But as we get both a consistent low-energy
theory (at least as far as we could go) as well as supersymmetry variations (to be shown
in a moment) by assuming the form of the metric above, we feel quite confident that the
metric on the moduli space is indeed given by (10) and that this formula is valid everywhere
on the moduli space.
The unbroken supersymmetry + relates the bosonic and fermionic collective coordinates. The bosonic zero modes may be obtained by an unbroken supersymmetry variation
116
as
T
T
c a 0 = +
c a 5 i i 5 .
a = +
(13)
Expanding a = (Xi )i a , we see that in a BPS configuration where i A = 0 the righthand side must also vanish. Indeed, it does identically (which can be seen by inserting
05 5 and first let it act on i and then on + ). Taking a = 5 we see that the bosonic
collective coordinates Xi are related to the fermionic collective coordinates i as
T
ci .
Xi = +
(14)
Similarly, by inserting the expansions

= (X = const) + Xi i (X = const) + Ordo (X)2 ,

B = B(X = const) + Xi i B(X = const) + Ordo (X)2
into the (unbroken) supersymmetry variation

1 MNP +
1 M
a
HMNP + a M +
= i
24
2
(15)
(16)
and using that in the BPS configuration (that is for the fields at X = const) this variation
vanishes, we get only a contribution from term involving derivatives of Xi . To first order
this is

= i a Xi i a + .
(17)
Noting that i A = 0, we get

i 5 i = i Xi i 5 + .
Multiplying by j
and integrating over the transverse
i = i Xi + jik Xk j .
(18)
coordinates x i
we get, using (10),

(19)
The 2d spinors are two-component Majorana spinors. These are the supersymmetries of
a non-linear sigma model. The SO(4) SO(4)R spinor indices on and i are always
trivially contracted and we can forget about them from the two-dimensional point of view.
We have four supersymmetries obtained by substituting Jj i j for i . So the 2d action
should be the N = 4 supersymmetric sigma model. As shown in [8] the quaternions Ji j
comprise the three complex structures of a hyper-Khler manifold. So we conclude that
the moduli space Mk is a hyper-Khler manifold.
We know that the N = 4 supersymmetric sigma model [8] contains an additional term
which is proportional to
Rij kl i k j l

m n i k j l
= 2 [ j [ k Gi ] l] + Gmn k[i
j ]l

5
5
pq
4
5
5
4
5
5 i k j l
=2
k [i j ] l + G
d x p k [i
d y q j ] l

= 4 k [i 5 j ] l 5 i k j l + ,
(20)
117
where in the last step we have used the completeness and orthonormality conditions of the
modes to rewrite the second term. To get this term we would need the 6d effective action
up to four-fermion interactions (which we do not have yet). This could presumably be done
easier once a superfield formalism has been developed for the tensor multiplet.
Appendix A. Fermionic zero modes

The equations of motion for the fermions associated to a string solitons aligned in the
5-direction read (I = 1, 2, 3, 4)
2 1 I J KL

abcde a I b J c K d e 05 L ,
(A.1)
6V4 r 5

where will let = A A (A = 1, . . . , 4) and r = 2 + ( 5 )2 .
A
For k = 1 the map x A A has winding number one, and A = (R) xR . We can then
write the equation of motion as

2 4 x I
05 5
I I =
(A.2)
.
I
V4 r 5 R 4
I I =
Here + are terms proportional to A 05 A and to xA 05 A .

Inserting that = I I 5 u wherethe BPS solution [7] is given by 5 = (D
|Q|
) (R R0 ) with Q = 2 and R0 = |Q|/(2V3 D), we get the l.h.s.
2V R 2
3

|Q|
J J I I 5 u = I I 5 u =
(R R0 )u .
V3 R0 3
(A.3)
To compute the r.h.s. we need the following results,

4 4 5
= ,
0 r 5
3
lim
(R R0 ) V3 R0 3
(R R0 ).
5 =
=
d 5
|Q|
(R
)
0
dR
(A.4)
The terms + should combine into something that is proportional to ( 5 ) 5 + ( 5 ) =

(( 5 ) 5 ) (with denoting d/d 5 ) and thus vanishes identically. We then get the r.h.s.
2 4 x I
2 4 5 x I
|Q|
05 5
I =
(R R0 )u .
I
5
4
V4 r R
V4 3
R
V3 R0 3
(A.5)
The r.h.s. is then equal to the l.h.s. only for the choice u . These are thus fermionic moduli
for k = 1.
References
[1] E. Witten, Some comments on string theory dynamics, in: Future Perspectives in String Theory, Los Angeles,
1995, pp. 501523, hep-th/9507121.
[2] O. Aharony, String theory dualities from M-theory, Nucl. Phys. B 476 (1996) 470483, hep-th/9604103.
[3] N.S. Manton, A remark on the scattering of BPS monopoles, Phys. Lett. B 110 (1982) 54.
118
[4] J. Harvey, A. Strominger, String theory and the Donaldson polynomial, Commun. Math. Phys. 151 (1993)
221232, hep-th/9108020.
[5] J.P. Gauntlett, Low-energy dynamics of N = 2 supersymmetric monopoles, Nucl. Phys. B 411 (1994) 443
460, hep-th/9305068.
[6] J. Blum, Supersymmetric quantum mechanics of monopoles in N = 4 YangMills theory, Phys. Lett. B 333
(1994) 9297, hep-th/9401133.
[7] A. Gustavsson, Classical selfdual BPS strings in d = 6, (2, 0) theory from afar, JHEP 0301 (2003) 019,
hep-th/0212167.
[8] L. Alvarez-Gaume, D. Freedman, Geometrical structure and ultraviolet finiteness in the supersymmetric
sigma-model, Commun. Math. Phys. 80 (1981) 443.

Gauge-invariant second-order perturbations

and non-Gaussianity from inflation
Viviana Acquaviva a,1 , Nicola Bartolo b,c,2 , Sabino Matarrese b,c ,
Antonio Riotto c
a Dipartimento di Fisica, Universit di Pisa, via Buonarroti 2, Pisa I-56100, Italy
b Dipartimento di Fisica di Padova G. Galilei, Via Marzolo 8, Padova I-35131, Italy
c INFN, Sezione di Padova, Via Marzolo 8, Padova I-35131, Italy
Received 24 December 2002; received in revised form 10 June 2003; accepted 23 June 2003
Abstract
We present the first computation of the cosmological perturbations generated during inflation
up to second order in deviations from the homogeneous background solution. Our results, which
fully account for the inflaton self-interactions as well as for the second-order fluctuations of the
background metric, provide the exact expression for the gauge-invariant curvature perturbation
bispectrum produced during inflation in terms of the slow-roll parameters. The bispectrum represents
a specific non-Gaussian signature of fluctuations generated by quantum oscillations during slow-roll
inflation. However, our findings indicate that detecting the non-Gaussianity in the cosmic microwave
background anisotropies emerging from the second-order calculation will be a challenge for the
forthcoming satellite experiments.
PACS: 98.80.Cq
1. Introduction
Inflation represents a successful mechanism for the causal generation of primordial
cosmological perturbations in the early Universe [1]. These fluctuations are then amplified
by the gravitational instability to seed structure formation in the Universe, and to produce
E-mail address: antonio.riotto@pd.infn.it (A. Riotto).
1 Address after November 2002: SISSA/ISAS, Via Beirut 4, I-34014, Trieste, Italy.
2 Address after November 2002: Astronomy Centre, University of Sussex, Brighton BN1 9QJ, UK.
doi:10.1016/S0550-3213(03)00550-9
120
V. Acquaviva et al. / Nuclear Physics B 667 (2003) 119148
cosmic microwave background (CMB) anisotropies. Since the primordial cosmological

perturbations are tiny, the generation and evolution of fluctuations during inflation
has always been studied within linear theory. On the other hand, there exist physical
observables, such as the three-point function of scalar perturbations, or its Fourier
transform, the bispectrum, for which a perturbative treatment up to second order is
required, in order to obtain a self-consistent result
The importance of the bispectrum comes from the fact that it represents the lowest order
statistics able to distinguish non-Gaussian from Gaussian perturbations for which oddorder correlation functions necessarily vanish. An accurate calculation of the primordial
bispectrum of cosmological perturbations has become an extremely important issue, as a
number of present and future experiments, such as MAP and Planck, will allow to constrain
or detect non-Gaussianity of CMB anisotropy data with high precision.
So far, the problem of calculating the bispectrum of perturbations produced during
inflation has been addressed by either looking at the effect of inflaton self-interactions
(which necessarily generate non-linearities in its quantum fluctuations) in a fixed de Sitter
background [2], or by using the so-called stochastic approach to inflation [3],3 where
back-reaction effects of field fluctuations on the background metric are partially taken
into account. An intriguing result of the stochastic approach is that the dominant source
of non-Gaussianity actually comes from non-linear gravitational perturbations, rather than
by inflaton self-interactions.
In this paper we provide for the first time the computation of the scalar perturbations
produced during single-field slow-roll inflation up to second order in deviations from
the homogeneous background. We achieve different goals. First, we provide a gaugeinvariant definition of the comoving curvature R at second order of perturbation theory.
The importance of the second-order comoving curvature perturbation R comes from
the fact that it allows to compute the three-point correlation function for the primordial
scalar perturbationsor its Fourier transform, the bispectrumwhich represent the lowest
order statistics able to distinguish non-Gaussian from Gaussian perturbations. Secondly,
we show that the second-order comoving curvature perturbation is conserved on superhorizon scales, like its first-order counterpart.4 Third, we obtain the expression for the
gauge-invariant gravitational potential bispectrum during inflation, in terms of slow-roll
parameters or, equivalently, of the spectral indices of the scalar and tensor power-spectra.
An accurate calculation of the primordial bispectrum of cosmological perturbations has
become a crucial issue, as a number of present and future experiments, such as MAP and
Planck, will allow to constrain or detect non-Gaussianity of CMB anisotropy data with
high precision.
The plan of the paper is as follows. In Section 2 we write the perturbations of the
metric for a spatially flat RobertsonWalker background up to second order and we derive
consistently the fluctuations of the energymomentum tensor of a scalar field. In Section 3
we demonstrate how to find a second-order gauge-invariant definition of the comoving
curvature perturbation. Section 4 will be devoted to the perturbed Einstein equations up to
3 See also Ref. [4].
4 For related results see also Ref. [5].
121
second order in the metric and in the inflaton fluctuations. We shall explain how to derive
the evolution of the curvature perturbation on large scales during a period of cosmological
inflation by performing an expansion to lowest order in the slow-roll parameters. Finally,
in Section 5 we draw some concluding remarks relating our findings to the gauge-invariant
gravitational potential bispectrum which is the main physical observable which carries
information about the primordial non-Gaussianity. We also provide Appendices A and B
where the reader can find various technical details.
2. Perturbations of a flat RobertsonWalker Universe up to second-order

We first write down the perturbations on a spatially flat RobertsonWalker background
following the formalism of Refs. [6,7]. We shall first consider the fluctuations of the metric,
and then the fluctuations of the energymomentum tensor of a scalar field. Hereafter Greek
indices run from 0 to 3, while Latin indices label the spatial coordinates from 1 to 3. If not
otherwise specified we will work with conformal time , and a prime will stand for a
derivative with respect to .
2.1. The metric tensor
The components of a perturbed spatially flat RobertsonWalker metric can be written
as

g00 = a 2 ( ) 1 + 2 (1) + (2) ,

1 (2)
(1)
g0i = a 2 ( ) i + i ,
2

1
gij = a 2 ( ) 1 2 (1) (2) ij + ij(1) + ij(2) .
2
(1)
The standard splitting of the perturbations into scalar, transverse (i.e., divergence-free)
vector parts, and transverse trace-free tensor parts with respect to the 3-dimensional space
with metric ij , can be performed in the following way:
(r)
(r)
i = i (r) + i ,
(r)
ij
(r)
= Dij (r) + i j
(2)
(r)
+ j i
(r)
+ ij ,
(3)
where (r) = (1), (2) stand for the order of the perturbations, i and i are transverse
vectors ( i i(r) = i i(r) = 0), ij(r) is a symmetric transverse and trace-free tensor
( i ij(r) = 0, ii(r) = 0) and Dij = i j (1/3)ij k k is a trace-free operator.5 Here and
in the following Latin indices are raised and lowered using ij and ij , respectively.
5 Notice that our notation is different from that of Refs. [8,9] for the presence of D , while it is closer to the
ij
one used in Refs. [10,11]. As far as the first-order perturbations are concerned, the metric perturbations and E
of Refs. [8,9] are given in our notation as = (1) + (1/6)i i (1) and E = (1) /2, respectively. However, no
difference appears in the calculations when using the generalized longitudinal gauge in Eq. (28).
122
For our purposes the metric in Eq. (1) can be simplified. In fact, first-order vector
perturbations are zero in the presence of a scalar field; moreover, the tensor part gives
(1)
a negligible contribution to the bispectrum. Thus, in the following we can neglect i ,
(1)
(1)
i and ij . However the same is not true for the second-order perturbations. In the
second-order theory the second-order vector and tensor contributions can be generated by
the first-order scalar perturbations even if they are initially zero [7]. Thus we have to take
them into account and we shall use the metric

g00 = a 2 ( ) 1 + 2 (1) + (2) ,

1
1 (2)
2
(1)
(2)
g0i = a ( ) i + i + i ,
2
2

1 (2)
2
(1)
(2)
(1)
gij = a ( ) 1 2 ij + Dij +
2

1 (2)
(4)
(2)
(2)
+ i j + j i + ij
.
2
The contravariant metric tensor is obtained by requiring (up to second order) that g g =
and it is given by
2

g 00 = a 2 ( ) 1 2 (1) (2) + 4 (1) i (1) i (1) ,

1
g 0i = a 2 ( ) i (1) + i (2) + i(2) + 2 (1) (1) i (1) i (1)D i k (1) ,
2

2

1
g ij = a 2 ( ) 1 + 2 (1) + (2) + 4 (1) ij D ij (1) + (2)
2

1 i j (2)

+ j i(2) + ij (2) i (1) j (1) 4 (1) D ij (1)
2

(5)
+ D ik (1) D j k (1) .
Using g and g one can calculate the connection coefficients and the Einstein tensor
components up to second order in the metric fluctuations. Their complete expressions are
contained in Appendix A.
2.2. Energymomentum tensor of a scalar field
We shall consider a scalar field (, x i ) minimally coupled to gravity, whose energy
momentum tensor is given by

1
g + V () ,
T = g
(6)
2
where V () is the potential of the scalar field. A successful period of inflation can be
attained when the potential V () is flat enough.
The scalar field can be split into a homogeneous background 0 ( ) and a perturbation
(, x i ) as
123

1

, x i = 0 ( ) + , x i = 0 ( ) + (1) , x i + (2) , x i ,
(7)
2
where the perturbation has been expanded into a first- and a second-order part, respectively.
Using the expression (7) into Eq. (6) and calculating T = g T up to second order we
find
1 (2)
(1)
T = T (0)
+ T + T ,
2
(8)
(0)
where T corresponds to the background value, and

V (1)
1
(0)
,
T 0 0 + (1) T 0 0 = a 2 0 2 V0 + a 2 0 (1)0 (1)
2

2
1
V 2 1 (2) 2 1 (1) 2
1
a + 0
(2) T 0 0 = 2 (2) 0 (2)
a
2
2
2
2
2
2
2 V 2

1
1
k (1)k (1) (1)
a 2 (1) 0 2
2
2
2

1
+ 2 (1) (1) 0 + k (1) k (1) 0 2 ,
2

0 (0)
(1) 0
2
(1)
T i + T i = a 0 i ,

2
1
(2) T 0 i = 2 0 i (2) i (1) (1) + 20 (1) i (1) ,
a
2
(1) i
2 i (1)
T i (0)
+ 0 i (1) ,
0 + T 0 = 0

2 1
1
(2) T i 0 = 2 0 i (2) + i (2) + i(2) 0 2 + (1) i (1)
2
a 2
+ 20 (1) i (1) 20 2 (1) i (1) + 2 i (1) (1) 0
(9)
(10)
(11)
(12)
(13)

+ 20 2 (1) i (1) 0 D ij (1) j (1) 0 2 D ij (1) j (1) , (14)

1
V (1)
(0)
+ a 2 0 (1) (1)0 i j ,
T i j + (1) T i j = a 2 0 2 V0
2
(15)

1
2
1
V
1
1
2
(2) 0 (2)
a 2 (2) 0 2 + (1)
(2) T i j = 2
2
2
2
2
a

2 2 V 2
2
1
1
k (1) k (1) + 2 (1) 0 2 (1)
a
2
2
2

1
2 (1) (1) 0 k (1) k (1)0 k (1)k (1) 0 2 i j
2

(16)
+ i (1)j (1) + 0 i (1)j (1) ,
with V0 = V (0 ). A comment is in order here. As it can be seen from Eq. (5) and
Eqs. (10), (12), (14) and (16) the second-order perturbations always contain two different
124
contributions, quantities which are intrinsically of second order, and quantities which are
given by the product of two first-order perturbations. As a consequence, when considering
the Einstein equations to second order in Section 4, first-order perturbations behave as a
source for the intrinsically second-order fluctuations. This is an important issue which was
pointed out in different works on second-order perturbation theory [7,12,13] and it plays a
central role in deriving our main results on the primordial non-Gaussianity.
3. The second-order gauge-invariant comoving curvature perturbation
Let us now focus on the primordial cosmological perturbations produced during a period
of inflation driven by the inflaton . The fluctuations of the inflaton produce an adiabatic
density perturbation which is associated with a perturbation of the spatial curvature . The
density/curvature perturbation can be defined in a gauge-invariant manner as the curvature
perturbation R on slices orthogonal to comoving wordlines. At first order, the comoving
curvature perturbation is given by the gauge-invariant formula [8]
R(1) = (1) +
H (1)
,
0
(17)
where H = a /a is the Hubble rate, (1) is the gauge-dependent first-order curvature

perturbation and (1) is the first-order inflaton perturbation in that gauge. An important
feature of R(1) is that it remains constant on super-Hubble scales while approaching the
horizon entry.
We want to find the gauge-invariant comoving curvature perturbation R up to second
order. Thus we expand the curvature perturbation and the inflaton perturbation as
= (1) + 12 (2) and = (1) + 12 (2) , according to Eqs. (1) and (7). An infinitesimal
coordinate change up to second order induces a gauge transformation of the metric and the
scalar field perturbations [6,7]. In particular for a second-order shift of the time coordinate
0
(1)
+

1 0 0
0
(1) (1) (2)
2
(2) and (2) will transform as [6,7]

(2) = (2) + 2 0 (1) + 2H (1) H + 2H2 0 2
(1)
(1)
0
1 i (1)
0 0
0
i 0
H(1) (1) H(2) 2 (1) i (1) ,
3
0

(2)
0
0
0

(2)
= + (1) 0 (1) + 0 (1)
+ 2 (1) + 0 (2)
.
(18)
(19)
(20)
We find that the gauge-invariant comoving curvature perturbation R = R(1) + 12 R(2) is

provided by

(2)
1
1 ( (1) + 2H (1) + H (1) /0 )2

(1)
(2)
R=R + H +
+
2
0
2
H + 2H2 H0 /0
1
i (1) i (1).
(21)
6
125
We devote the rest of this section to illustrate how to find such a quantity.
Let us consider the first two terms in Eq. (21) and how they change under the time
coordinate shift of Eq. (18). According to our perturbative expansion

(2)
1

(1)
(2)
+H =R + H +
(22)
,
0
2
0
and using Eqs. (19) and (20) this quantity transforms as

1 0 2
H
0
2
+H
H
=
+
H
+
+
2H
(1)
0
0
2 (1)
0 0
0
1
0
2 i (1) i (1)
i (1) ,
6
where we have set T = (1) + 2H (1) + H (1) /0 . Notice that

H

0
2

T = T + (1) H + 2H + 0 ,
0
(23)
(24)

(1) = (1) H 0 and
since the usual first-order transformations for (1) and (1) are
(1)
(1) = (1) + 0 , respectively. Thus Eq. (23) can be written as

0 (1)
+H

0
0
1
1
0
= + H + T + T (1)
2 i (1) i (1)
i (1) .

0
0 2
6
(25)
0

(1) = (1) 0 to
Solving Eq. (24) for (1)
and using the first-order transformation
(1)
0 and i 0 we finally find
express i (1)
(1)
+H

1
1
T2
(1)
i (1)
+

i

0 2 H + 2H2 H0 /0 6
= +H
1
1
T2
+
i (1) i (1),

2
0 2 H + 2H H0 /0 6
(26)
which shows that Rthe combination on the r.h.s. of this equationis indeed gaugeindependent.
Notice that, by replacing with the energy density , we can find in an analogous way a
gauge-invariant expression for the curvature perturbation on uniform-density hypersurfaces
, which to first order is given by [14] (1) = (1) + H (1) /0 , where 0 is the
background energy density. Thus at second order = (1) + 12 (2) is
=
(1)

(2)
1
1 ( (1) + 2H (1) + H (1) /0 )2

(2)
+ H +
+
2
0
2
H + 2H2 H0 /0
1
i (1) i (1) .
6
(27)
126
4. Einstein equations
In this section we shall derive the behaviour on large scales of the comoving curvature
perturbation at second order R(2) introduced in Eq. (21). Our starting point is to calculate
the perturbed Einstein equations (2) G = ( 2 /2) (2) T in a generalized longitudinal
gauge. Here 2 = 8GN and T is the energymomentum tensor of the inflaton field.
From the Einstein equations in this gauge we shall pick up an equation for a single
unknown functionthe potential (2) in a way similar to the procedure used in Ref. [8]
to isolate the equation of motion for (1) in the longitudinal gauge.6 As in Ref. [8], but at
the second-order level in the perturbations, an equation linking (2) and (2) holds, so that
it is possible to obtain an explicit expression for (2) and hence for the curvature R(2) ,
once the equation for (2) has been solved. Indeed, there are many differences with respect
to the first-order case, as it will be evident to the reader from the details of the calculations
which follow. The main difficulties arise due to the fact thatcontrary to the first-order
perturbation theoryalso vector and tensor contributions are present, and to the fact that
the two scalar potential (2) and (2) differ even in the longitudinal gauge for the presence
of source terms which are quadratic in the first-order perturbations.
4.1. Einstein equations in the generalized longitudinal gauge
Up to now we have not choosen any particular gauge. Hereafter we will work in a
generalized longitudinal gauge defined as

g00 = a 2 ( ) 1 + 2 (1) + (2) ,
g0i = 0,

1 (2)
(2)
(2)
gij = a 2 ( ) 1 2 (1) (2) ij + i j + j i + ij
.
2
(28)
One can obtain the Einstein equations in this gauge either by using directly the metric
tensor in Eq. (28) or by using the more general metric of Eq. (4), where no gauge choice
has been specified yet, and reduce the equations to the longitudinal gauge only at the end.
We have performed both the computations to have a cross check for the equations obtained.
In Appendix A we give the expression for the Einstein tensor in the more general form
using the metric (4).
We shall now give the Einstein equations in the longitudinal gauge at first and second
order in the perturbations, respectively
(1) G0 0 = 2 (1) T 0 0
implies
(29)
6 We recall that the equation of motion for the potential (1) in the longitudinal gauge is [8]
(1) i i (1) + 2 H
0 (1)
+ 2 H

0
0
H (1) = 0.
0
127
2

a (1)
i (1)
2
(1) 2
(1)
(1) V 2
+6
2i = 0 0
a ,
a
(30)
(1) G0 i = 2 (1) T 0 i implies
(31)

a
2 i (1) 2i (1) = 2 0 i (1) ,
(32)
a
(1) Gi j = 2 (1) T i j implies
(33)
2

a
a
a
a
(1) + i i (1) + 4 (1) + 2 (1) i i (1) i j
2 (1) + 4 (1) 2
a
a
a
a

i
(1)
i
(1)
2
(1) 2
(1)
(1) V 2
a i j , (34)
j + j = 0 + 0
a
6
a
(2) G0 0 =
(1)
2 (2) 0
T 0
2
(35)
implies
2
2
(1) 2
2

a
a
a
a
(2) + (2) 12
3 (1)
3 (2) i i (2) +
a
a
a
a
8 (1) i i (1) 3i (1) i (1)

1
V 2 1 (1) 2 1 i (1)
1
a i (1)
= 2 (2) 0 (2)
2
2
2
2

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

a 2
0 + 2 0 ,
2
2
2 (2) i
T 0 implies
2

a
1
i (2) + i (2) + k k i(2) + 2 (1) i (1) + 8 (1) i (1)
a
4

1
= 2 0 i (2) + i (1) (1) + 20 (1) i (1) ,
2
(2) Gi 0 =
(36)
(37)
(38)
2
(2) Gi j = (2) T i j implies
(39)
2

a
a
a 2 (2) 1
a
1
k k (2) + (2) + (2) +
k k (2) + 2 (2) + (2)
2
a
a
a
2
a

2
a (1) 2
a (1) 2
a
8
+4
8 (1) (1) 3k (1) k (1)
a
a
a

1
1
2
i j i j (2) + i j (2)
4 (1) k k (1) (1)
2
2
1

1 a
+
j i(2) + i j(2) + i(2) j + j i(2) + i j(2) + i(2) j
2a
4
1
k k i(2) j + 2 i (1) j (1) + 4 (1) i j (1)
4
128

= 2
2

1 (2) 1 (2) V 2 1 (1) 2 1 (1) k (1)
0
a + k + 2 (1) 0 2
2
2
2
2

2

1
2 V 2
(40)
(1)
a 2 (1) (1) 0 i j + 2 i (1) j (1) .
2
2
In writing the perturbed Einstein equations at second order we have set throughout
(1) = (1) , since in the longitudinal gauge at first-order the two scalar potentials are
equal in the case of a scalar field, and to obtain Eqs. (36) and (40) we have made use
of the background relations a /a = H2 + H and ( 2 /2)0 2 = H2 H .
We shall now describe how to isolate the equation for the potential (2) . We use the
(0 0)-component of Einstein equations, the divergence of the (i 0)-component and
the trace of the (i j )-component, both performed with the background metric ij . Notice
that the divergence and the trace operations make the vector and the tensor modes disappear
from the equations. Thus, we are left with three equations in the three unknown functions,
(2) , (2) , and (2) .
From the divergence of the (i 0)-component of Einstein equations it is possible to
recover an expression for (2)
( (2) + H (2) + 1 ) 1
1 (2)
=
,
2
2 0
0
(41)
where
= 2 (1) i i (1) + 10i (1) i (1) + 8 (1) i i (1) ,
= i
i (1)
(1)
i (1)
(1)
+ 2
(1)
i (1)
0
(42)
+ 2i
(1) i (1)
0 ,
(43)
and 1 is the inverse of the Laplacian operator for the three spatial-coordinates. The
expression (41) and its derivative with respect to the conformal time can be used in the
trace of the (i j ) equation to obtain
1 i (2) 1 i (2)
i i
3
3
2
(1) 2
a (1) 2
a
a
7
=8
4
+ 8 (1) (1) + i (1) i (1)
a
a
a
3

8
a
a
2
+ (1) i i (1) + (1) + 1 + 2 1 2 1 2 2 1
3
a
a

2

1
1 (1) 2 1 (1) i (1)
2
2 V 2
+ 2
i + 2 (1) 0 2 (1)
a
2
6
2
2

2 (1) (1) 0 .
(44)
Thus a relation between (2) and (2) follows from Eq. (44)
(2) = (2) 1 ,
(45)
129
where stands for three times the r.h.s. of Eq. (44). Eq. (45) shows that the two scalar
potentials (2) and (2) differ for quadratic terms in the first-order perturbations, as
anticipated above.
Using Eq. (45) we are now in the position to express the other two unknown functions
(2) and (2) in terms solely of (2) . From Eq. (41) we finally obtain
1 (2)
( (2) + H (2) + 1 ) 1 1
=
2 .
2
2 0
0
0
(46)
Plugging Eqs. (45) and (46) into the (0 0) Einstein equation, the equation of motion for
(2) is derived

0 (2)
0
(2)
i (2)

i + 2 H
+ 2 H H (2)
0
0
(1) 2
(1) 2
2
(1)
i (1)
= 12H
+3
+ 8 i + 3i (1) i (1)

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

+ H 2 0 1 + 1
0

2 1
2

1
+ 2 (1) i (1) i (1) 2 (1) 0 2
2
2

2 2 V 2
1
(1) (1)
(1)
a
+
2
(47)
0 .
2
2
Eq. (47) is our master equation. Before solving it, let us stress two important points. First,
no approximation has been made up to now. In particular notice that this equation is exact
at any order in the expansion in terms of the slow-roll parameters. Secondly, the l.h.s. of
Eq. (47) is exactly the same as in the equation for (1) in the longitudinal gauge at first
order (see footnote 6). However, at second order, the key point is that Eq. (47) for (2) is
not homogeneous, but there is a source made up of terms which are quadratic in the firstorder perturbations. Notice that, as in the first-order calculation, it is not necessary to use
the perturbed KleinGordon equation to close the system of the evolution equations for the
fluctuations. Nevertheless, we also calculated the KleinGordon equation at second order
in the inflaton and metric fluctuations; this is reported in Appendix B.
4.2. The large-scale curvature perturbation R(2) in the slow-roll approximation
We shall now solve Eq. (47) in order to obtain the expression for the comoving curvature
perturbation R(2) defined in Eq. (21). First we rewrite Eq. (47) in cosmic time dt = a d ,
since it is more easy in this way to recognize the slow-roll parameters ( = H /H 2 and =
( (0 /H 0 ) [1]. During a period of inflation ( and must be 1, but only at a certain
point we will perform an expansion to lowest-order in the slow-roll parameters. Moreover,
where possible, we shall neglect some terms which give a subdominant contribution on
large scales.
130
Using cosmic time, Eq. (47) becomes

H
0 (2)
0
1
(2) 2 i i (2)
(2) + H 1 2
+ 2H 2
2
H 0
H
H 0
a

2
H
4
= 24H 2 1 + 2 (1) 24H (1) (1) 2 i (1) i (1)
H
a

0
0
2
4 1
2
1
H 1
+ 2 H 1
1
a
H 0
a
a
H 0
0
2 1
2 1 + 1
a
0

(1) 2 2 V (1) 2

(1) 2
2
2
(1) (1)
2 + 8 0
8 0 ,
2
+4
(48)
where we have replaced the term appearing in Eq. (47) with its explicit definition given
by Eqs. (44) and (45). Notice that in the source on the r.h.s. of Eq. (48) there appears always
the combination 1 ( 2 /a /a) and its derivative with respect to cosmic time. Let us
now calculate such a combination. The quantity defined in Eq. (42) can be rewritten as

= 2i i (1) (1) + 6 i (1) i (1) + (1) i i (1) .

(49)
a
Using the equation of motion (30) to express ( (1) ) , the quantity , Eq. (43), turns out to
be
1 0 i (1) 2
i + 30 i (1) i (1) + 3 0 (1) i i (1)

=
a
2 0

2
2
+ 2 k (1) k i i (1) + 2 i i (1) k k (1) (1) .
(50)
0
0
Using the equation of motion for (1) in the longitudinal gauge
(1) + H (1) =
2
0 (1) ,
2
(51)
which can be derived from Eq. (32) with (1) = (1) , we get
2

2

2 0 i (1) 2
=
i + 3H i i (1) 2i i (1) (1)
a
a
2 0

2
2
+ 2 k (1) k i i (1) + 2 i i (1) k k (1) (1) ,
a 0
a 0
(52)
and therefore

2

2 0 (1) 2
1
2
+ 3H (1) 2 (1) (1)
a
a
2 0

2
+ 2 1 k (1) k i i (1) + i i (1) k k (1) . (53)
a 0
131
The master equation (48) can be simplified if we drop those terms which are next to
leading-order in the slow-roll parameters
1
(2) + H (2) 2 i i (2)
a

2
0 (1) 2
= 12 2H (1) (1) 0 + 3 2 H
12H (1) (1) + 18H 2 (1)
0
2

(1) 2
(1) (1)
0 (1) 2
0
+4
2
2 1 + 1
+ 3H
2 0
0

12H 1 (1) k i (1)
+ 2 k i
+ i i (1)k k (1)
a 0

8
4
+ 2 1 k (1) k i i (1) + i i (1)k k (1) 2 i (1) i (1)
a 0
a
i (1) 2
2

8
8
i
2 V
(1) 2

+
+ 2 (1) i i (1)
2
2
2
2
a
a
0
8 0 (1) i (1)
i ,
a 2 0 0
(54)
where we have used Eq. (51) so that 24H 2(1 + H /H 2 )( (1) )2 24H (1) (1) =
24H ( (1) )2 12 2H (1) (1) 0 , we have used the expression for ( (1) ) from
Eq. (30), and we also explicitly written the combination (53).
Let us notice that through Eq. (51) we find

2
0 (1) 2
12H (1) (1) + 18H 2 (1)
0

2
0 (1) 2
+ 30H 2 (1)
= 18 2H (1) (1) 0 + 3 2 H
0
(1) 2
(1) 2
2
2V
= 12H
+ 6
2 2 (1) ,
12 2H (1) (1) 0 + 3 2 H
(55)
where the last passage is valid to lowest-order in the slow-roll parameters. Thus we finally
obtain

0
1
H
(2) 2 i i (2)
(2) + H (2) + 2H 2
2
H 0
H
a
2

2

(1) 2
(1) 2

0 (1) 2
+ 3H (1) 2 (1) (1)
= 12H
+ 6
+4
2 0

0 1
12H
2 + 1 + 2 1 k (1) k i i (1) + i i (1)k k (1)
0
a 0

8
4
+ 2 1 k (1) k i i (1) + i i (1)k k (1) 2 i (1) i (1)
a 0
a
i (1) 2
i
8
8 (1) i (1)
8 0 (1)
+ 2 i 2
i i (1) .
(56)
a2
a
a 0 0
2 02
132
Integrating Eq. (56) we find

2

2
2

0 (1) 2
(2) + H (2) = 12H (1) + 4
+ 3H (1) 2 (1) (1)
2 0

(1) 2
0 1
dt
+ 1 + 6
0

12H 1 (1) k i (1)
+
+ i i (1) k k (1) dt
k i
2
a 0
i (1) 2

1
8
i
(1) i (1)
4
dt
dt
i
a2
a2
2 02

8 (1) i (1)
8 0 (1) i (1)
+
(57)
dt
i dt.
i
2
a
a 2 0 0
We are now in the position to calculate the second-order comoving curvature
perturbation R(2) . From Eqs. (21), (45) and (46) we get

H
H
H
R(2) = 2 2 2 (2) + H (2) + 1

2 2 1
2 2 2 1
a
a
0
0
0
1 ( (1) + 2H (1) + H (1)/
0 )2
+ (2) 1 +
(58)
,
2
2
2 H (2 + H /H 0 /H 0 )
where the last term is the part of R(2) in Eq. (21) which is quadratic in the first-order
perturbations. Note that it is of the order of O(( 2 , 2 ) and thus we shall neglect it in the
following. In Eq. (58) there appears once more the combination 1 ( 2 /a /a). Thus
using Eqs. (53) and (57) we obtain

0 (1) 2
H
H 2 (1) 2 12H (1) 2
(2)
(1) (1)
12
6
+
dt
R = 3H
0
0
202
2 02
2 02

24H 2
1 1 (1) k i (1)
+
k i
+ i i (1)k k (1) dt
3
2
a
2 0

i (1) 2

1
i
8H
16H
(1) i (1)
dt
i
dt
a2
a2
2 02
4 04

16H
16H 0
1 (1) i (1)
1 (1)
+
i dt
i i (1) dt
2
2
3
2
2
a
a2
0
0 0
(1) 2
2
H2
2
= 3H
+ 6 2 2 (1) 6H (1)
0
0
0

2
()
12H 2
+ 3(H 2 2 +
dt
0
2 02

24H 2
1 1 (1) k i (1)
+ i i (1)i i (1) dt
+
k i
3
2
a
0
8H
2 02
+
16H
2 02

16H
1
i (1) i (1) dt
a2
4 04

16H 0
1 (1) i (1)
i dt
2
a
2 03 0
i i (1)
a2
133
2
dt
1 (1)
i i (1) dt
a2
H 0 1
1 ,
202 0
(59)
where the last passage is valid to lowest-order in the slow-roll parameters and we have used
the equation of motion (51). In Eq. (59) the terms which are not integrated in time can be
safely treated in the long-wavelength limit. To lowest order in the slow-roll parameters and
on large scales, k aH , (1) can be considered as constant and
(1) =
2 0 (1)
(1)
=(H
2 H
0
(60)
which can be derived from the equations of motion (51) and (B.4) in the longitudinal
gauge for (1) and (1) , respectively. On the other hand, from the definition of the
comoving curvature perturbation at first order, Eq. (17), and using Eq. (60) we can write
R(1) = H (1)/0 to lowest order in the slow-roll parameters and on large scales. Thus in
these approximations Eq. (60) can be rewritten as
(1) = (R(1) .
(61)
Performing various integrations by parts in expression (59), using the perturbation

equations at first-order and properly subtracting the contributions in the far ultraviolet,
we arrive at the final expression for the comoving curvature perturbation
2

R(2) = ( 3() R(1) + I,
where

2
(

4
I =

4
1 (1) i (1)
1
(1) i (1)
dt
dt
i
i
(
a2
a2
(1) 2
dt + (( )1 i R (1) i R (1) .
(62)
(63)
5. Discussion and conclusions

In this paper we have provided a complete analysis of the second-order scalar
perturbations during inflation leading to the derivation of the gauge-invariant comoving
curvature perturbation R.
The comoving curvature perturbation receives a contribution which is quadratic in
R(1) . The total curvature perturbation will then have a non-Gaussian ( 2 ) component.
Reminding that the gauge-invariant potential (which is related to Bardeens variable [10]
by = H ) and the curvature perturbation R are related by = 35 R, the following
134
simple relation in configuration space holds

(x) = Gauss (x) + d 3 y d 3 z K(y, z)Gauss (x y)Gauss(x z) + const.
(64)
which is valid on superhorizon scales and where the constant is such that (x) = 0. Here
Gauss = 35 R(1) is a Gaussian random field. The non-Gaussianity kernel in momentum
space is given by
5
K(k1 , k2 ) = ( 3() + fK (k1 , k2 ),
(65)
6
where fK (k1 , k2 ) is directly related to the function I and is first-order in the slow-roll
parameters. The gravitational potential bispectrum then reads

(k1 )(k2 )(k3 )

= (2)3 (3)(k1 + k2 + k3 ) 2K(k1 , k2 )P (k1 )P (k2 ) + cyclic ,
(66)
where P (k) is the power spectrum of the gravitational potential.
A cautionary note is in order. Our results, Eqs. (63) and (66), (67), apply strictly till
slow-roll inflation ends and as such they can be considered as initial conditions from this
epoch onwards. It is still totally unclear how to follow the evolution of the perturbations
through further processes such as reheating (or preheating) into the radiation dominated
epoch, especially when dealing with perturbations up to second-order. Note also that in
the final expression (65) we have not written possible terms of the order of O(( 2 , 2 )N
which might be sizeable for certain classes of inflationary models (N is the number
of e-foldings from the time at which a given scale crosses the horizon and the end of
inflation; for large-scale CMB anisotropies N 60). Whether these terms cancel out,
or equivalently whether the curvature perturbation remains constant on super-horizon
scales also at next-to-leading order in the slow-roll parameters remains an open issue.
However, if present, terms of order of O(( 2 , 2 )N resemble those found in Refs. [2,
15], coming from the self-interactions of the inflaton field. The primordial gauge-invariant
potential bispectrum leads to a nonzero CMB bispectrum via the SachsWolfe effect
(T /T )SW = (1/3).
Deviations from a scale-invariant spectrum can make the primordial non-Gaussianity
non-negligible. The possible presence of non-Gaussianity in primordial cosmological
perturbations is only mildly constrained by existing observations [16,17]. One way of
parametrizing the possible presence of non-Gaussianity in the primordial gravitational
2
potential is to expand it as (see, for example, [18]) = Gauss + fNL (Gauss
2
2
Gauss ) + O(fNL ), where Gauss is a zero-mean Gaussian random field and fNL is
an expansion parameter which can be observationally constrained. Recent analyses of
the angular bispectrum from 4-year COBE data [19] yield a weak upper limit, |fNL | <
1.5 103 . The analysis of the diagonal angular bispectrum of the Maxima dataset [20]
also provides a very weak constraint: |fNL | < 2330. According to Ref. [18], however,
the minimum value of |fNL | that will become detectable from the analysis of MAP and
Planck data, after properly subtracting detector noise and foreground contamination, is
about 20 and 5, respectively. These results imply that detecting non-Gaussianity at the level
135
emerging from our second-order calculation will represent a challenge for the forthcoming
satellite experiments.
Note added
While revising this paper, a related work by Maldacena, astro-ph/0210603, appeared,
where the three-point correlation function for the curvature perturbation is obtained by
computing the cubic term contributions to the Lagrangian. Our results, where comparison
is possible, agree with those in astro-ph/0210603.
Acknowledgements
We wish to thank M. Bruni for useful discussions on gauge transformations.
Appendix A. Perturbing gravity at second order

A.1. Basic notation
The number of spatial dimensions is n = 3. Greek indices (, , . . . , , , . . .) run
from 0 to 3, while latin indices (a, b, . . ., i, j, k, . . . , m, n, . . .) run from 1 to 3. The total
spacetime metric g has signature (, +, +, +). The connection coefficients are defined
as

g
g
g
1
= g
+
(A.1)
2
x
x
x
The Riemann tensor is defined as

R = ,
,
+
(A.2)
The Ricci tensor is a contraction of the Riemann tensor

R = R ,
(A.3)
and in terms of the connection coefficient it is given by
R =

+

.
(A.4)
The Ricci scalar is given by contracting the Ricci tensor

R = R .
The Einstein tensor is defined as
1
G = R g R.
2
(A.5)
(A.6)
The Einstein equations are written as G = 2 T , so that 2 = 8GN , where GN is the

usual Newtonian gravitational constant.
136
In the following expressions we have chosen a specific ordering of the terms. In the
expressions in which two spatial indices appear, such as Eq. (A.18), we have assembled
together the terms proportional to ij . The intrinsically second-order terms precede
the source terms which are quadratic in the first-order perturbations. The second-order
fluctuations have been listed in the following order as (2) , (2) , (2) , i(2) , (2) , i(2)
and ij(2) , respectively. This ordering simplifies the analogy between the first-order and the
second-order equations and allows to obtain immediately the expressions in a given gauge.
A.2. The connection coefficients
In a spatially flat RobertsonWalker background the connection coefficients are
a
a
i
,
0j
= i j ,
a
a
i
0
i
00 = 0i = j k = 0.
0
=
00
ij0 =
a
ij ,
a
(A.7)
(A.8)
The first-order perturbed connection coefficients corresponding to first-order metric

perturbations in Eq. (4) are
0
(1) 00
= (1) ,
(1) 0i0 = i (1) +
(A.9)
a
a
i (1) ,
a i (1)
+ i (1) + i (1) ,
a
a
a
a
(1) ij0 = 2 (1) ij i j (1) 2 (1) ij (1) ij Dij (1)
a
a
a
1
(1)
+ Dij ,
2
1
i
= (1) ij + Dij (1) ,
(1) 0j
2
a
(1) jik = j (1) ki k (1) ji + i (1) j k i (1) j k
a
1
1
1
+ j Dki (1) + k Dji (1) i Dj k (1) .
2
2
2
At second order we get:
i
(1) 00
=
(A.10)
(A.11)
(A.12)
(A.13)
(A.14)
1
a
0
= (2) 2 (1) (1) + k (1) k (1) + k (1)k (1) + k (1) k (1) ,
(2) 00
2
a
(A.15)

a
1
1
a
(2)
i (2) + i 2 (1) i (1) 2 (1) i (1)
(2) 0i0 = i (2) +
2
2a
a
1
(A.16)
(1) i (1) + k (1) Dik (1) ,
2
137

1

1
1 a i (2)
i
(2) 00
= i (2) +
+ i(2) + i (2) + i(2) + 2 (1) i (1)
2
2a
2

a
(1) i (1) + 2 (1) i (1) + 2 (1) i (1)
a
a
(A.17)
k (1) D ik (1) k (1) D ik (1) k (1) D ik (1) ,
a

2
a
1
a
a
a
(2) ij0 = (2) (2) (2) + 4 (1) + 2 (1) (1) + 4 (1) (1)
a
2
a
a
a

a
+ k (1) k (1) k (1) k (1) ij
a

1
1
i j (2) + Dij (2) + j i(2) + i j(2) + ij(2)
2
4
1

1 a
+
Dij (2) + i j(2) + j i(2) + ij(2) i j(2) + j i(2)
2a
4
+ 2 (1)i j (1) i (1) j (1) j (1) i (1) (1) Dij (1)
1
1
1
+ k (1) i Dkj (1) + k (1) j Dik (1) k (1) k Dij (1) , (A.18)
2
2
2

1
1

i
= (2) i j + D i j (2) + j i(2) + i j(2) + i(2) j
(2) 0j
2
4

1
a
+ j i(2) i j(2) 2 (1) (1) i j i (1) j (1) i (1) j (1)
4
a
1
+ (1) D i j (1) + (1) D i j (1) D ik (1) Dkj (1) ,
(A.19)
2

1
(2) jik = j (2) i k k (2) i j + i (2) j k
2

1
+ j D i k (2) + k D i j (2) i Dj k (2)
4
1
1
(2)
+ j k i(2) + j i(2) k + k i(2) j i j k
2
4

1 a i (2)
+ i(2) j k + 2 (1) j (1) i k k (1) i j + i (1) j k
2a
a
+ 2 (1) i (1) j k + i (1) j k (1) + (1) i (1) j k
a

+ (1) j D i k (1) + k D i j (1) i Dj k (1) + j (1) D i k (1)
a i (1)
Dj k (1)
a
a
1
1
+ m (1) D i m (1) j k i (1) Dj k (1) D im (1) j Dmk (1)
a
2
2
1 im (1)
1 im (1)
(1)
(1)
D k Dmj + D m Dj k .
(A.20)
2
2
+ k (1) D i j (1) m (1) D im (1) j k
138
A.3. The Ricci tensor components

In a spatially flat RobertsonWalker background the components of the Ricci tensor
R are given by
2
a
a
,
R0i = 0,
R00 = 3 + 3
(A.21)
a
a
2
a
a
Rij =
(A.22)
+
ij .
a
a
The first-order perturbed Ricci tensor components are
a i (1)
a
a
i + i i (1) + i i (1) + 3 (1) + 3 (1) + 3 (1) ,
a
a
a
(A.23)
2

a
a
a
1
(1) R0i = i (1) +
i (1) + 2i (1) + 2 i (1) + k D k i (1) ,
a
a
a
2
(A.24)

2

a
a
a
a
a
(1) 2 (1)
(1) Rij = (1) 5 (1) 2 (1) 2
a
a
a
a
a

2
a
a
2
(1) (1) + k k (1) k k (1) ij
a
a
2

a
a
a
1
(1)
(1)
(1)
i j
+ Dij
+ Dij +
Dij (1) + Dij (1)
a
a
a
2

a
1
+ i j (1) i j (1) 2 i j (1) + k i D k j (1)
a
2
1
1
k (1)
k
(1)
+ k j D i k Dij .
(A.25)
2
2
At second order we obtain
(1) R00 =
3 a (2) 1 i (2) 3 a (2) 3 (2) 1 a

+ i +
k k (2)
+
+
2a
2
2a
2
2a
1
a
+ k k (2) 6 (1) (1) k (1) k (1) 3 (1) (1)
2
a
a
+ 2 (1) i i (1) k (1) k (1) + 6 (1) (1) + 6 (1) (1)
a

(1) 2
a
a
+3
(1) i i (1) + k (1) k (1) + k (1) k (1)
a
a
2

a
a
a
+
k (1)k (1) k (1) k (1) + 2 (1) i i (1)
a
a
a

a
k (1) k (1) + 2 (1) i i (1) + 3 k (1) k (1) k (1) i D ik (1)
a
(2) R00 = +
139
a
a
i k (1)D ik (1) k (1)i D ik (1)
a
a
1 ik (1)
(1)
ik (1)
(1) ik (1)
k i D i k D + D Dki (1)
2

a
1 ik (1)
1
D ik (1) Dki (1) ,
+ D Dki (1) +
(A.26)
4
2a
a
1
1
1
(2)
(2)
(2) R0i = + i (2) + i (2) + k D k i (2) + i i i i i i
a
4
4
4

1 a 2 (2)
a
1 a (2)
(2)
(2)
i + i +
i + i 4 (1) i (1)
+
2 a
2 a
a

a
2 (1) i (1) + 4 (1) i (1) + 4 (1) i (1) 2 (1) i (1)
a
2

a
a
2
(1) i (1) (1) i (1) i i (1) i (1) k (1)i k (1)
a
a
a
+ k (1) i k (1) k (1) i k (1) i i (1) i (1) (1) i (1)
a
a (1)
1
5 i (1) k (1) D ik (1) + (1) k D k i (1)
a
2
1
(1)
k (1)
+ k D i k (1) Dik (1) + k (1) D k i (1)
2
a k (1)
1
1
(1)
+ Dik
+ k (1)Dik (1) k D km (1) Dmi (1)
a
2
2
1
1
D km (1) k Dmi (1) + D km (1) i Dmk (1)
2
2
1 km (1)
(1)
+ D i Dmk .
(A.27)
4
i k (1) D ik (1)
The purely spatial part of (2) R is very long, and for simplicity has been divided into
two parts, a diagonal part (2) Rijd which is proportional to ij , and a non-diagonal part Rijnd
2
2
a
1 a (2) a (2)
5 a (2)
a
(2) d
(2)
Rij =

(2)
a
2a
a
2a
a
1
a
1
1 a i (2)
i
(2) (2) + i i (2)
2
a
2
2a
2

a
a (1)2
a
a
+4
+
+ 4 (1) (1) + 10 (1) (1)
a
a
a
a
2

a
a
a (1) (1)
+ 2 (1) (1) + (1) (1) + 2 (1) (1) + 4
+

a
a
a
2

+ k (1) k (1) + (1) + k (1) k (1) + 2 (1) i i (1)
2
a
a
a
+ k (1) k (1) + 2 (1) i i (1)
k (1) k (1)
a
a
a
140
a
a
a
k (1) k (1) k (1) k (1) + 3 k (1) k (1)
a
a
a
(1) k (1)
(1)
i (1)
(1) k (1)
+ 2k + i + k
m (1) k D km (1)

a
a
k m (1) D km (1) + m k (1) Dkm (1) + k (1) m Dkm (1)
a
a

1 a mk (1)
D Dkm (1) ij ,
(A.28)
2a
1
1
a
1
(2) Rijnd = i j (2) + i j (2) i j (2) i j (2)
2
2
a
2
1 (2)
1 a (2)
(2)
(2)
i j + j i i j + j i
2a
4
2

a
a
1
Dij (2) + i j(2) + j i(2) + ij(2)
+
+
2
a
a

1

1 a
+
Dij (2) + i j(2) + j i(2) + ij(2) + k i D k j (2)
2a
2
1 i
1 i (2)
(2)
i Dij i ij
4
4

1
+ Dij (2) + i j(2) + j i(2) + ij(2) + i (1) j (1)
4
+ 2 (1)i j (1) j (1) i (1) i (1) j (1) + 3i (1) j (1)
a (1)
i j (1) + (1) i j (1) + 2 (1)i j (1)
a
a
+ i i (1) i j (1) j k (1) i k (1) 2 i (1) j (1)
a
a
2 i (1) j (1) i (1) j (1) j (1) i (1) i (1) j (1)
a
2
a
j (1) i (1) + (1) i j (1) 2
(1) Dij (1)
a
a
a
a
2 (1)Dij (1) 2 (1) Dij (1) (1) Dij (1)
a
a
a
1 (1)
1
Dij (1) (1) Dij (1) + k (1)i D k j (1)
2
2
1
1
a
+ k (1) j D k i (1) k (1) k Dij (1) 3 (1) Dij (1)
2
2
a
1 (1)
1
1
+ Dij (1) + k (1) i D k j (1) + k (1) j D k i (1)
2
2
2
3
(1) k
(1)
(1)
k
(1)
k Dij + k i D j + (1) k j D k i (1)
2
(1) k k Dij (1) + i (1) k D k j (1) + j (1) k D k i (1)
+ 2 (1) i j (1) + 4
141
1
+ k i (1) D k j (1) + k j (1) D k i (1) + k i (1) D k j (1)
2
1
1
1
(1) k (1)
k (1)
(1)
+ k j D i
k Dij
+ k (1)i Dkj (1)
2
2
2
1 k (1)
1 k (1)
(1)
k (1)
(1)
+ j Dki
k Dij
+ i Dkj (1)
2
2
1 k (1)
1
a
+ j Dki (1) k (1) k Dij (1) + k (1) i Dkj (1)
2
2
a
a k (1)
a k (1)
a
(1)
(1)
+ j Dki k Dij k k (1) Dij (1)
a
a
a
1 k (1)
1
Di Dkj (1) i Dmj (1) k D km (1)
2
2
1
1
j Dmi (1) k D km (1) + m Dij (1) k D km (1)
2
2
1
1
(1) km (1)
k i Dmj D k j Dmi (1) D km (1)
2
2
1
1
(1) km (1)
+ k m Dij D + D km (1) i j Dkm (1)
2
2
1
mk (1)
(1)
+ i D j Dmk .
(A.29)
4
A.4. Ricci scalar
At zeroth order the Ricci scalar R is given by
6 a
.
a2 a
The first-order perturbation of R is

1
a
a
(1)
R = 2 6 i i (1) 2i i (1) 2i i (1) 6 (1) 6 (1)
a
a
a

a (1)
a (1)
18
12 + 4i i (1) + k i D k i (1) .
a
a
R=
(A.30)
(A.31)
At second order we find

a (2)
a
a
6 (2) + 2i i (2) 9 (2) 3 (2)

a
a
a

2
a
1
a
(1) + 2k (1) k (1)
i i (2) 3 i i (2) + k i D ki (2) + 24
a
2
a
a (1) (1)
a
(1)
i (1)
(1) (1)
+ 4 i + 24
+ 6
+ 36 (1) (1)
a
a
(1) k (1)
(1)
i (1)
(1) (1)
+ 2k 4 i + 12
12 (1) (1)

a
a
36 (1) (1) + 6k (1) k (1) + 16 (1) i i (1) + 6 k (1) k (1)
a
a
(2) R = i i (2) 3
142
+ 12
6
a (1) i (1)
a
i + 4 (1) i i (1) + 2 (1) i i (1) 5 k (1) k (1)
a
a
a
k (1) k (1) + i i (1) i i (1) i k (1) i k (1)
a
+ 8k (1) k (1) + 2k (1) k (1) 4 (1)i i (1) 12

+ 4 (1) i i (1) + 2k (1)i D ik (1) + 2i k (1)D ik (1)
a (1) i (1)
i
a
+ 4 (1) k i D ki (1) 2k i (1) D ik (1) + 3k (1) i D k i (1)

a
+ 6 k (1) i D i k (1) + 2i (1) k D ik (1) + 2k i (1) D ik (1)
a
a
3
+ 6 k i (1) D ki (1) D ik (1) Dik (1) D ik (1) Dki (1)
a
4
a
3 D ik (1) Dik (1) 2k i Dmi (1) D km (1) + i i Dim (1) D mi (1)
a
1
k D km (1) i Dmi (1) + i D km (1) i Dmk (1) .
(A.32)
4
A.5. The Einstein tensor components
The Einstein tensor in a spatially flat RobertsonWalker background is given by

3 a 2
G0 0 = 2
(A.33)
,
a
a
2
a
a
1
i
G j = 2 2
(A.34)
i j ,
a
a
a
G0 i = Gi 0 = 0.
(A.35)
The first-order perturbations of the Einstein tensor components are

2

a
1
a
a
1
(1) G0 0 = 2 6
(1) + 6 (1) + 2 i i (1) 2i i (1) k i D k i (1) ,
a
a
a
2
a
(A.36)

a
1
1
(1) G0 i = 2 2 i (1) 2i (1) k D k i (1) ,
(A.37)
a
a
2

2
a
1
a
a
a
(1) Gi j = 2 2 (1) + 4 (1) 2
(1) + i i (1) + 4 (1) + 2 (1)
a
a
a
a
a

a
1
i i (1) + 2 i i (1) + i i (1) + k m D k m (1) i j
a
2

a
a
i j (1) + i j (1) 2 i j (1) i j (1) + D i j (1)
a
a

1
1
1 i (1) 1
i k
(1)
ik (1)
+ k D j + k j D k k D i j (1) .
+ D j
2
2
2
2
(A.38)
143
The second-order perturbed Einstein tensor components are given by

2
a
1
a
a
1
(2) + 3 (2) i i (2) + i i (2) k i D ki (2)
(2) G0 0 = 2 3
a
a
a
4
a
2
(1) 2
a
a
12
12 (1) (1) 3i (1) i (1)
a
a

2

a
a
8 (1) i i (1) + 12 (1) (1) 3 (1) + 4 (1) i i (1)
a
a
a
1 a
1
(1) k (1)
(1) k (1)
(1) i k (1)
k + i k
2 k
a
2 a
2

1
a
a
k k (1)k k (1) 2 k (1) k (1) + 4 (1) i i (1)
2
a
a
2k (1) k (1) 2 (1) i i (1) (1) i k D i k (1)
a
2 (1) k i D k i (1) + k i (1) D ki (1) 2 i k (1) D ik (1)
a
a
2 k (1) i D ik (1) k (1) i D k i (1)
a
1 i
i Dmk (1) D km (1) + m k Dik (1)D im (1)
2
1
1
+ k D km (1) i Dmi (1) i D km (1) i Dkm (1)
2
8

1 ik (1)
a ki (1)
(1)
+ D Dki
+ D Dik (1) ,
(A.39)
8
a

1 a i (2)
1
1
1

+ i (2) + k D ki (2) + i i i(2) i i i(2)
(2) Gi 0 = 2
a
a
4
4
4
2
2

a
a
a
a
i (2) i(2) + 2
i (2) + 2
i(2)
a
a
a
a
a
a
4 (1) i (1) + 4 (1) i (1) 2 (1) i (1) + 4 (1) i (1)
a
a
(1) i (1)
i (1)
+ 8
i i (1) k (1) i k (1) + i i (1) i (1)
2
a
a
+ i k (1) k (1) + 4 (1) i (1) 8
(1) i (1)
a
a
a
+ 2 (1) i (1) + i i (1) i (1) k (1) i k (1)
a
2
a
a
+ 2 (1) i (1) + 8
(1) i (1) 4 (1) i (1)
a
a

a
1
a
2 (1) i (1) k (1)D i k (1) 2 k (1) D ki (1)
a
2
a
1
k (1) D ki (1) + 2 (1) k D ki (1) + (1) k D ki (1)
2
144
1
a
k (1) D ki (1) + k (1)D i k (1) + k (1) D i k (1)
2
a
2

a
a
4
k (1) D ik (1) + 2 k (1) D ik (1)
a
a
1
1
k D km (1) D i m (1) k D i m (1) D km (1)
2
2
1 i
1 i
(1) km (1)
+ Dmk D
+ Dmk (1) D km (1)
4
2
1 ik (1)
D m D m k (1) ,
(A.40)
2

1
1
1
1
a
(2)
(2)
(2) G0 i = 2 i (2) i (2) k D k i (2) i i i + i i i
a
a
4
4
4
a
+ 8 (1) i (1) + 4 (1) i (1) + 2 (1) i (1) 4 (1) i (1)
a
a
4 (1) i (1) + i (1) i i (1) i k (1) k (1) + 8 (1) i (1)
a
2

a
a
4
(1) i (1) 2 k (1) i k (1) + i i (1) i (1)
a
a
1
+ k (1) i k (1) k (1) D k i (1) + k (1) Dik (1)
2
1
(1)
k (1)
(1) k (1)
k D i
+ k D i
(1) k D k i (1)
2
a
k (1) D k i (1) + i (1) k m D k m (1) 2 k (1) Dik (1)
a
2
a
+
k (1) Dik (1) + k (1) m i D m k (1)
a
1
1
m (1) k k Dim (1) + k D km (1)Dim (1)
2
2
1
1
+ k Dim (1) D km (1) i Dmk (1) D km (1)
2
4
1
(1) km (1)
i Dmk D
(A.41)
,
2

2
1 i (2) a (2)
1
a (2)
a
1
(2) d i
G j = 2 + i +
+2
(2) i i (2)
2
a
a
a
2
a
a
a
1
1
+ (2) + 2 (2) + i i (2) + i i (2) k i D ki (2)
a
a
2
4
2
(1) 2
a
a (1) 2
a
+4
8
8 (1) (1) k (1) k (1)
a
a
a
a
2 (1)i i (1) 4 (1) (1) 2 (1) (1) 8 (1) (1)
a
(2) Gnd j
145

2
a
2k (1) k (1) 4 (1) i i (1) + (1) + 8 (1) (1)
a
+ 4 (1) (1) + 2 (1) i i (1) (1) i i (1) 2 (1)i i (1)
a
a
3 a
k (1) k (1)
2 k (1) k (1) 4 (1) i i (1) +
a
a
2 a
2
a
a
1
k (1) k (1) + 2 k (1) k (1) i i (1) i i (1)

a
a
2
1
a
+ m k (1) m k (1) + 4 (1) i i (1) + 2 (1) i i (1)
2
a
(1) k (1)
(1)
2k
i i (1) k m (1) D km (1)
3
k (1) m D mk (1) k (1) m D mk (1) k (1) i D k i (1)
2
a
k (1) m D mk (1) k m (1) D km (1) 2 k (1) m D m k (1)
a
a
3
k (1) m
(1)
l
(1) km (1)
2 m D k + k Dml D
a
4
1 i
1
i Dml (1) D ml (1) + m k Dlk (1)D lm (1)
2
4
1
1
(1) l ml (1)
+ k Dkm D l Dkm (1) l D km (1)
2
8
1 mk (1)
3 mk (1)
(1)
+ D Dmk
+ D Dmk (1)
2
8
a mk (1)
+ D Dkm (1) i j .
(A.42)
a

1
1
1
a
1
= 2 i j (2) + i j (2) i j (2) i j (2)
2
2
a
2
a
1

1 a i (2)
(2)
j + j i(2) i j + j i(2)
2a
4
1 a i (2)
1
(2)
(2)
+
+ k i D k j (2)
+ i j + j i(2) + i j
D j
2a
2
1 i i (2) 1 i i (2)
i D j i j
4
4
1 i (2)
(2)
i (2)
+ i (1) j (1)
+ D j
+ j + j i(2) + i j
4
+ 2 (1) i j (1) 2 (1) i j (1) j (1) i (1) i (1) j (1)
a
+ 3 i (1) j (1) + 4 (1) i j (1) + 2 i (1) j (1)
a
a (1) i
(1)
(1) i
(1)
+ 4 j + j + 2 (1) i j (1)
a
a
+ i i (1) i j (1) j k (1) i k (1) 2 i (1) j (1)
a
146
a i (1)
j (1) i (1) j (1) + j (1) i (1)
a
i (1) j (1) j (1) i (1) 2 (1) i j (1) + (1) i j (1)
a
a
1
4 (1) i j (1) 2 (1) D i j (1) (1) D i j (1)
a
a
2
1
1
(1) i
(1)
(1) i k
(1)
D j
+ k D j + k (1) j D ki (1)
2
2
1
1
(1) k i
(1)
(1) ki (1)
k D j + j k D + (1) D i j (1)
2
2

a
1
+ (1) D i j (1) + 2 (1) D i j (1) + k (1) i D k j (1)
a
2
a (1) i (1)
1
(1) i
(1)
+ 2 D j
+ D j
+ k (1) j D ki (1)
a
2
3
k (1) k D i j (1) + 2 (1) k i D k j (1) + 2 (1) k j D ki (1)
2
2 (1) k k D i j (1) i i (1) D i j (1) + i (1) k D k j (1)
1
+ j (1) k D ki (1) + k i (1) D k j (1) + i (1) k D k j (1)
2
1
1
1
i (1) k
(1)
(1) ki (1)
+ k D j
+ k j D
k k (1) D i j (1)
2
2
2
1 k (1) i
1 k (1)
(1)
i
(1)
+ Dkj
+ j D k
k (1) k D i j (1)
2
2
1
1
1
+ k (1) i Dkj (1) + k (1) j D i k (1) k (1) k D i j (1)
2
2
2
a
a
+ k j (1) D ik (1) + k (1) i Dkj (1) + k (1) j D ik (1)
a
a

a k (1)
a
1
k D i j (1) + 2 k j (1) D ik (1) D ki (1) Dkj (1)
a
a
2
1 i
1
Dmj (1)k D km (1) j D i m (1) k D km (1)
2
2
1
1
i
(1)
km (1)
+ m D j k D k i Dmj (1) D km (1)
2
2
1
1
i
(1) km (1)
k j D m D + k m D i j (1) D km (1)
2
2
1
1
+ D km (1) i j Dkm (1) + i D mk (1) j Dmk (1)
2
4
a
k m D k m (1) D i j (1) Dkj (1) D ik (1)
a
1
(1) ki (1)
Dkj
D m k D m j (1) D ki (1)
2

1
m
(1) ki (1)
+ m Dkj D
(A.43)
,
2
2
147
where (2) Gd j stands for the diagonal part of (2) Gij , which is proportional to ji , and
i
(2) Gnd j is the non-diagonal contribution.

Appendix B. Perturbing the KleinGordon equation
In the homogeneous background the KleinGordon equation for the scalar field is
0 + 2
a V 2

a .
a 0
(B.1)
The perturbed KleinGordon equation at first-order is

a (1)
i i (1) (1) 0 3 (1) 0 i i (1)0
a
V
2V
.
= (1) 2 a 2 2 (1)
(1) + 2
(B.2)
At second-order we get
1
a
1
a
1

(2) (2) + i i (2) + (2) 0 + 2 (2) 0 + (2) 0
2
a
2
a
2
(1)2
3 (2) 1 i (2)
a (1) 2
+ 0 + i 0 4
0 8
0 4 (1) (1) 0
2
2
a
a
+ 2 (1) (1) + (1) (1) + 4 (1) (1) + k (1) k (1) 6 (1) (1) 0
a
+ 6 (1) (1) 0 + 3 (1) (1) k (1) k (1) + 2 (1) i i (1)
2 (1) i i (1) 0 k (1) k (1)0 k (1) k (1) 0 + 2 (1) i i (1) 0
a
+ k (1) k (1)0 + 2 k (1) k (1) 0 + k (1) k (1) 0 + 2 k (1)k (1)
a
a k (1) (1)
+ 2 k + i i (1) (1) + k (1) k (1) k (1)i D i k (1) 0
a
i k (1)D ik (1) 0 i k (1) D ik (1) k (1) i D k i (1)
1
1 2 V (2) 2 1 3 V (2) 2 2
a .
+ D ik (1)Dki (1) 0 =
a +
2
2 2
2 3
(B.3)
To obtain the KleinGordon equation in the longitudinal gauge of Eq. (28) one can simply
set (1) = (2) = 0, and (1) = (1) . Thus at first-order we find
(1) + 2
V
a (1)
2V
i i (1) 4 (1) 0 = (1) 2 a 2 2 (1)
,
a
while at second order the equation is

1
a
1
a
1
+ (2) + (2) i i (2) (2) 0 2 (2) 0 (2) 0
2
a
2
a
2
(B.4)
148
3
(2) 0 4 (1) (1) 0 4 (1) (1) 4 (1)i i (1)
2
2 3 V 2
2V
1
2V
1
= 2 (1) (1) 2 a 2 (2) 2 a 2 (1)
a ,
2
3
(B.5)
where we have used the background equation (B.1) and the first-order perturbed equation
(B.4) to simplify some terms.
References
[1] D.H. Lyth, A. Riotto, Phys. Rep. 314 (1999) 1;
A. Riotto, hep-ph/0210162.
[2] T. Falk, R. Rangarajan, M. Srednicki, Astrophys. J. 403 (1993) L1.
[3] A. Gangui, F. Lucchin, S. Matarrese, S. Mollerach, Astrophys. J. 430 (1994) 447.
[4] L. Wang, M. Kamionkowski, Phys. Rev. D 61 (2000) 063504;
A. Gangui, J. Martin, Mon. Not. R. Astron. Soc. 313 (2000) 323.
[5] D.S. Salopek, J.R. Bond, Phys. Rev. D 42 (1990) 3936.
[6] M. Bruni, S. Matarrese, S. Mollerach, S. Sonego, Class. Quantum Grav. 14 (1997) 2585.
[7] S. Matarrese, S. Mollerach, M. Bruni, Phys. Rev. D 58 (1998) 043504.
[8] V.F. Mukhanov, H.A. Feldman, R.H. Brandenberger, Phys. Rep. 215 (1992) 203.
[9] K.A. Malik, Ph.D. Thesis, astro-ph/0101563.
[10] J.M. Bardeen, Phys. Rev. D 22 (1980) 1882.
[11] H. Kodama, M. Sasaki, Prog. Theor. Phys. Suppl. 78 (1984) 1.
[12] K. Tomita, Prog. Theor. Phys. 45 (1971) 1747;
K. Tomita, Prog. Theor. Phys. 47 (1972) 416.
[13] S. Matarrese, O. Pantano, D. Sez, Phys. Rev. Lett. 72 (1994) 320;
S. Matarrese, O. Pantano, D. Sez, Mon. Not. R. Astron. Soc. 271 (1994) 513.
[14] J.M. Bardeen, P.J. Steinhardt, M.S. Turner, Phys. Rev. D 28 (1983) 679.
[15] S. Gupta, A. Berera, A.F. Heavens, S. Matarrese, Phys. Rev. D 66 (2002) 043510.
[16] L. Verde, L. Wang, A. Heavens, M. Kamionkowski, Mon. Not. R. Astron. Soc. 313 (2000) L141.
[17] L. Verde, R. Jimenez, M. Kamionkowski, S. Matarrese, Mon. Not. R. Astron. Soc. 325 (2001) 412.
[18] E. Komatsu, N. Spergel, Phys. Rev. D 63 (2001) 063002.
[19] E. Komatsu, B.D. Wandelt, D.N. Spergel, A.J. Banday, K.M. Gorski, Astrophys. J. 566 (2002) 19.
[20] M.G. Santos, et al., Phys. Rev. Lett. 88 (2002) 241302.

Cosmological constant and kinetic supersymmetry

breakdown on a moving brane
Philippe Brax a , Adam Falkowski b , Zygmunt Lalak b
a Service de Physique Thorique CEA-Saclay F-91191 Gif/Yvette, France
b Institute of Theoretical Physics, University of Warsaw, Poland
Received 28 March 2003; accepted 20 June 2003
Abstract
We consider cosmological solutions in 5d locally supersymmetric theories including boundary
actions, with either opposite tension branes for identical brane chiralities or equal tension branes for
flipped brane chiralities. We analyse the occurrence of supersymmetry breakdown in both situations.
We find that supersymmetry as seen by a brane observer is broken due to the motion of the brane
in the bulk. When the brane energymomentum tensor is dominated by the brane tension, the 4d
vacuum energy on the observable brane is positive and proportional to the inverse square of the brane
local time. We find that the mass splitting within supersymmetric multiplets living on the brane is of
the order of the inverse of the brane local time, examplifying the tight relation between the vacuum
energy scale and the supersymmetry breaking scale.
PACS: 04.64.+e; 11.30.Pb
1. Introduction
One possible interpretation of the cosmological evolution of the Universe is that it is
due to the motion of a brane embedded into a higher-dimensional ambient space. The
brane moves according to the equations of motion, and its dynamics as well as the physics
of the localized sector respond to local conditions encountered along its trajectory. In this
paper we would like to study the dynamics of supersymmetric bulk and branes, and their
mutual coupling, in terms of a local action. In such a case the cosmological evolution
may be related to the supersymmetry breakdown as seen on the brane. The aim of the
E-mail address: brax@spht.saclay.cea.fr (P. Brax).
doi:10.1016/S0550-3213(03)00548-0
150
P. Brax et al. / Nuclear Physics B 667 (2003) 149169
present work is to analyse this relationship in an explicit way. To this end we use the
brane-bulk supergravities constructed in [16], with matter sectors localized on the branes
and coupled supersymmetrically to the 5d supergravity with AdS5 bulk. We find timedependent solutions of the equations of motion that break 4d supersymmetry everywhere,
in the bulk and on the branes. The observer on the visible brane finds a positive effective
vacuum energy which changes as the inverse square of the local time. The scale of this
quintessential vacuum energy is the same as the scale of susy breaking on the brane,
hence asymptotically the observable brane becomes flat and supersymmetric. Of course,
this is only true when one neglects the contribution of the matter sector.
We investigate the effective 4d supergravity that describes the local dynamics of the
brane-bulk system in the vicinity of the brane. It turns out that the transmission of
supersymmetry breakdown to the observable sector occurs via the radion multiplet (i.e.,
the T -modulus). To make the model suitable to describe very late stages of the evolution
of the Universe one needs to enhance the local scale of brane supersymmetry breakdown
with respect to the scale of the cosmological constant. The correspondence of the 5d picture
to 4d sugra with T -modulus suggests that the relevant enhancement mechanism would
be equivalent to the tuning of the (super-)potential of the T -modulus and related to its
stabilization.
2. Supergravity with boundary branes

We discuss the cosmological evolution and the supersymmetry breaking of 5d N = 2
supergravity with a negative cosmological constant. The fifth dimension is an interval of
finite length ending with two branes located at x 5 = r0 and x 5 = r1 . These branes have
their own cosmological constants (brane tensions) and are coupled to the 5d bulk in a way
that preserves N = 1 supersymmetry [1,2,7]. We concentrate on a special subclass of these
models, in which the brane tensions are correlated with the bulk cosmological constant as
in the RandallSundrum model [8].
Let us first recall the basic features of 5d N = 2 gauged supergravity [9]. The gravity
multiplet (em , A , A ), A = 1, 2, consists of the vielbein (metric), a pair of symplectic
Majorana gravitini, and a vector field called the graviphoton. There is a global SU(2)
R-symmetry which rotates the two supercharges into each other. Making use of the
graviphoton we can gauge a U(1) subgroup of the R-symmetry group. What is important
here is that the gauging implies the presence of a negative cosmological constant in the
bulk, M 3 S5 = d 5 x ( 12 R + 6k 2 + ) where 52 = M 3 is the 5d gravitational constant.
Also, the transformation law of the gravitini is altered by the gauging. The gravitino
variation reads
1
A = D A igk A + ,
(1)
2
where g is the U (1)R gauge coupling and the dots stand for terms involving graviphoton,
which are not relevant for our discussion.
It is possible to couple branes in a supersymmetric way even when the brane actions
contain only the brane tension terms [2,7]. To see this, notice that one can reformulate the
151
theory on the interval (downstairs picture) as a theory defined on a circle (upstairs picture).
The values of a field at the two sides of a brane are related by a Z2 symmetry. We assume
that, for the brane at x5 = r0 , the Z2 symmetry acts on the gravitino as:

A

A r0 x 5 = 5 3 B B r0 + x 5 ,

A

5A r0 x 5 = 5 3 B 5B r0 + x 5 .
(2)
Similarly, the Z2 symmetry relates the values of the gravitini close to the x 5 = r1 brane as

A

A r1 x 5 = 5 3 B B r1 + x 5 ,

B
5
5A r1 x 5 = 5 QA
(3)
B 5 r1 + x .
We assume can take values +1 or 1. For = 1 each brane respects a different
chirality of the two bulk supersymmetries, so globally supersymmetry is entirely broken.
The point is that the gravitino variation has to have a jump at the branes, to be consistent
with the Z2 symmetry. This jump is a source of additional supersymmetric variation of the
action, proportional to (x 5 r0 ) and (x 5 r1 ). These can be cancelled by a suitably
chosen brane term representing a brane tension. For the case at hand the bosonic part of
the Lagrangian that locally preserves N = 1 supersymmetry is given by:

1
R + 6k 2
M 3 S = d 5 x g5
2

6 d 5 x g4 k x 5 r0 x 5 r1 .
(4)
For = 1 the above is the RandallSundrum action [10]. For = 1 the two brane
tensions are equal. We call this possibility the flipped case.
3. Branes in motion
We have described the Lagrangian coupling the N = 2 supergravity background to
boundary branes in a locally supersymmetric way. We will now consider the vacua of
such theories. In the bulk the vacua are determined by the equations of N = 2 supergravity.
For the bosonic fields this corresponds to finding solutions of the Einstein equations with a
negative bulk cosmological constant. The five-dimensional version of Birkhoffs theorem
prescribes the most general solution in the bulk [11]
ds 2 = f (r) dt 2 +
dr 2
+ r 2 dq2 ,
f (r)
(5)
where the three-dimensional subspace spanned by the coordinates i , i = 1, . . . , 3 is

respectively a spherical, flat or hyperbolic space for q = 1, 0, 1. The function f (r) is
defined as
f (r) = k 2 r 2 + q
C
.
r2
(6)
152
The parameter C is related to the mass of the black-hole located at the origin of coordinates.
The vacua are specified by the bulk metric and the motion of the two boundary branes.
This motion of a single brane follows from the Friedmann equation [11,12]
H2 =
q
C
+ 4
2
r
r
(7)
for H = r /r where the dot denotes d/d and is the proper time on the brane such that
the induced metric on the brane reads
dsB2 = d 2 + r 2 ( ) dq2 .
(8)
The first term in the Friedmann equation is the usual curvature contribution while the
second term is the so-called dark radiation. The solutions to the Friedmann equation are
easily found. For q = 0 we obtain

r( ) = 2( 0 ).
(9)
On the brane the geometry is the one of a radiation dominated FRW spacetime. The brane
either goes to infinity or is driven towards the origin at r = 0. For q = 1, 1 the solutions
are very different:

r( ) =
C ( 0 )2
q
1/2
.
(10)
For q = 1
the brane cannot escape to infinity, it may reach a maximal coordinates distance
of rmax = C before bouncing back towards the singularity. In the open case q = 1 the
brane does not return to the singularity, and may reach infinity. Notice too that as goes to
infinity, the motion of the brane becomes identical with the case where C = 0. The induced
geometry on the brane corresponds to Milne spacetime (see Appendix A for more details).
Let us now consider the motion of a two-brane system. We have to treat the unflipped
and flipped cases separately. Let us first concentrate on the unflipped case where the brane
tensions are opposite.
The bulk parameters C and q are determined by the initial positions and velocities of
the branes (the branes are specified by their position r0 and r1 , and their velocities r0 and
r1 ). The Friedmann equation implies that
C = r02 r12
r12 r02
r02 r12
(11)
As soon as the initial velocities of the branes are different, the black-hole mass C has to be
non-zero. The initial velocities are constrained by
q=
r12 r12 r02 r02

r02 r12
(12)
Notice that q = 0 and C = 0 is only compatible with static branes. In all other cases either
q or C must be different from zero.
153
4. Bulk supersymmetry breakdown

We have seen that the vacua are characterized in the bulk by two parameters, the blackhole mass C and the curvature q. As soon as the branes move, one of these parameters is
non-vanishing. We will now show that this implies that supersymmetry is broken by the
motion of the branes. In this section we show that in the background with non-zero q or C
it is impossible to define Killing spinors.
4.1. Killing spinors
Let us first define Gaussian normal coordinates x5

2
2
r+ + r
l
x5 =
,
ln
2
4l 2
(13)
where
2 + r2
2
r 2 r
r
+
cosh + +
,
2
2
such that the metric becomes
r2 =
ds 2 = dx52

+
2 + r2
r
+
4l 2
2
r
2
+ r+
(14)
sinh2 ( )
2 +r 2
r
+
2
cosh( ) +
2 r 2
r+
dt 2
2 r2
r+
cosh( ) +
dq2 .
2
(15)
Let us identify this form of the metric with ds 2 = dx52 f 2 (x5 ) dt 2 + g 2 (x5 ) dq2 . We can
now read the spin connection
t t5 = f ,
k ij = k ij ,
k i5 = g eik
,
(16)
where hatted indices are local frame indices, = d/dx5 and the tilded symbols refer to the
symmetric space of curvature q.
Let us now analyse the Killing spinors in the bulk. They satisfy
k A
Da A a 3 B B = 0.
2
This leads to
k A
5 A = 5 3 B B .
2
Define now two spinors
A
A B

= 5 3 B .
(19)
This implies that

k
x5 , x = e 2 x5 x .
(20)
(17)
(18)
154
Now the other components of the Killing spinor equations lead to

A
A B
k
g

Di
i 3 B = 0,
+
2g 2
i is the covariant derivative on the symmetric space of curvature q, and
where D

A
A B
k
f
+
0
0 3 B = 0.
2f
2
(21)
(22)
The integrability conditions in the maximally symmetric curved space

j ] = R

i , D
[D
i aj
b
a b
(23)
with
abcd = q(gac gbd gad gcd )
R
(24)
give
q
ij =
g2
g
k
2 2g
2
ij .
(25)
This equation leads to two different possibilities. If q = 0 and C = 0 then g /2g = k/2
implying that = 0 and + is left undetermined. Coming back to the Killing equations
we find that + is independent of x . This is the usual BPS situation where 1/2 of
supersymmetry is preserved by the vacuum. In all the other cases, i.e., as soon as the
branes start moving the only solution is = 0 leading to a complete breakdown of
supersymmetry.
Hence we conclude that the motion of the branes is responsible for the breakdown
of supersymmetry. This breakdown is spontaneous as realized due to the lack of
supersymmetry of the vacuum solution, despite the supersymmetric invariance of the
action.
4.2. KK-modes of gravitini
Let us consider the gravitino equation of motion in the bulk,
3 A
D A k 3 B B = 0.
(26)
2
The last term comes from the gravitino mass term in the bulk. We work in the gauge where
5 = 0. Let us focus on the A = 1 component, the A = 2 component being equivalent
due to the symplectic-Majorana condition imposed on five-dimensional spinors. Let us
decompose
1

1
0 L = + (x5 ) 0+ L x ,
(27)
0 R = (x5 ) 0 R x ,
and
i1

L

g +
(x5 ) i+ L x ,
f
1

g
i R = (x5 ) i R x ,
f
(28)
155
where are four-dimensional Majorana spinors. Notice the explicit breaking of Lorentz
invariance when f = g.
We are going to solve the gravitino equation separating the variables and decomposing
the wave function into a product of the five-dimensional and four-dimensional factors. We
shall denote the separation constant by m, and call it a mass term, although strictly speaking
we do not identify this parameter with the physcial mass. The precise correspondence
between these two can be made explicit, e.g., in the case of AdS4 geometry in 4d, see
[6], but here we consider more complicated 4d geometries. Whenever we use the term
four-dimensional mass we have in mind mass parameters in the sense of a general
four-dimensional supergravity Lagrangian. Hence, we take the massive four-dimensional
gravitini to satisfy
D4 = m ,
(29)
where is the Dirac matrix on the four-dimensional spacetime with spatial curvature q.
Notice the necessary exchange of chiralities in the Weyl basis.
The gravitino equation (26) can be treated in two steps. First of all, putting = 0 leads
to the evolution equations
3
g
m
5 + + k + + =
2
g
g
(30)
3
g
m
5 + k + = + ,
2
g
g
(31)
and
where 5 acts as 1 on left-handed spinors The = k = 1, 2, 3 equation leads to new

constraints on the spinors. We find that

3 g f + ki +
1
1
k0i
(32)
0 i R +
i L = 0.
gf 2 g 2 f
2 f
g
This is consistent provided we impose the irreducibility constraint
i i = 0,
i i+ = 0.
(33)
In conclusion we have found that the spectrum of gravitini corresponds to massive gravitini
of the 4d hyperbolic space together with an irreducibility constraint on spinors on the 3d
hyperbolic space.
We can now examine the nature of the spectrum for both equal and opposite tension
cases. We concentrate on the zero modes (m = 0) which are determined by
+ = e3kx5/2 g 1 0+ ,
= e3kx5 /2 g 1 0 .
(34)
Notice now that g does not vanish on the whole x5 axis. In the opposite tension brane, one
can choose 0 = 0 and then one obtains a single massless gravitino in four dimensions.
The remaining chirality is the same as the chirality of the Killing spinor. When the
chiralities are flipped, this implies that 0 = 0 therefore all the massless gravitini are
projected out. This is the expected result for flipped supersymmetry breaking.
156
5. Brane supersymmetry breaking

Let us concentrate on opposite tension branes moving away to infinity. This is the
moving brane version of the static RandallSundrum setting. Asymptotically the branes
are located at
x5(2) = t,
x5(1) = t R,
(35)
where R is the radion field. The far away metric becomes now
ds52 = dx52 + e2kx5 g dx dx ,
(36)
where the background metric g = g is the metric on the hyperbolic space

ds02 = dt 2 + l 2 d 2 .
(37)
We will now perform the dimensional reduction and express the Lagrangian in terms of
the metric g . Let us start with the vacuum energy due to the bulk and brane contributions.
The brane contributions comprise the brane action and the GibbonsHawking terms, [13].
It is straightforward to show that the contributions of bulk and brane cancel so that the
hard 4d cosmological constant vanishes altogether.
The EinsteinHilbert action from the bulk becomes the four-dimensional gravitational
action for g with a gravitational coupling
t
2kx5
1
t R e
=
(38)
2
52
implying that in the large time regime, the gravitational coupling becomes
a 2 (t) K/3
1
=
e
,
2 (4 )2
(39)
where a(t) = ekt and

1
1
= 2.
2
(4 )
k5
(40)
We have identified
2kR = T + T
(41)
as the real part of a scalar field which belongs to a chiral superfield T . Such a chiral
multiplet contains also the fifth component of the graviphoton, which plays the role of an
axion, and the fifth component of the gravitino, called the radino. The Khler potential of
the radion superfield has been identified as [14]

K = 3 ln 1 e(T +T ) .
(42)
As x5 l the form of the metric simplifies and f = g = e2x5 / l . In that case, there is
an asymptotic restoration of supersymmetry implying that the massless gravitini coincide
with the ones of the static supersymmetric RandallSundrum model. Moreover the Killing
157
spinors are now non-vanishing corresponding to a local, i.e., occurring in the vicinity of
the moving brane, restoration of supersymmetry. We can identify the zero mode for the
gravitini
= a 1/2 (x5 ) (x)
(43)
where is a four-dimensional massless Majorana spinor. Similarly one can solve for the
Killing spinor with the result
= a 1/2 (x5 ) (x),
(44)
reflecting the fact that locally only one combination of supercharges is preserved in the
bulk/brane system. We also include the contribution coming from
5 (x) = a 1/2(x5 ) 5 (x).
(45)
the covariant derivative in four dimensions with respect to g , we

Denoting by D
identify the low energy Lagrangian

1
+ 5 D
5 + 5 5 D
,
d 4 x g 2 R(g) + D
2
(46)
where are the 4d gamma matrices with respect to g . To remove the kinetic mixing
between 5 and we redefine the fields
1
5 5 ,
2
which leads to the action

1
4
+ 3 5 D
5 .
d x g 2 g R(g) + D
2
2
(47)
(48)
Let us analyse the supersymmetry variations of the gravitini. We find that

+ k g0 ,
D
2
(49)
i.e., = 0. The first contribution is the usual 4d supergravity variation, the other
one comes from the time dependence of the normalization factor of gravitini on the brane.
This second term is the signal of supersymmetry breaking. This is yet another example of
supersymmetry breakdown due to the boundary conditions in the brane system, this time
along the time direction while in the ScherkSchwarz case the new direction is spacelike. We shall call the present mechanism a kinetic supersymmetry breaking. Of course,
the kinetic breaking is 4d Lorentz non-covariant, hence it shall manifest itself somewhat
differently than the usual breaking through the F-terms and D-terms. One would like to
obtain the local interpretation of this susy breakdown given by a local observer bound
to the brane. Locally, near the brane (here we choose the positive tension brane to be
the observable one) it should be possible to understand the situation in terms of a local
supergravity defined in the vicinity of the brane.
158
A crucial sign of supersymmetry breaking is the presence of a time-dependent

E = a 2 (t)e K/3 g . Using
cosmological constant in the Einstein frame defined by g

d 4 x a 2 (t)eK(T )/3 g R(g) d 4 x gE R(gE ) 6gE ln a ln a ,

(50)
we find that in the Einstein frame the initial time dependence of the gravitational constant
becomes a time dependent vacuum energy

3 d ln a 2
= 2
(51)
d
4
where is the brane local time.1 Using a(t) = / l this leads to
=
3
.
(4 )2 2
(52)
This is a positive time dependent cosmological constant which decays as

corresponding to the asymptotic restoration of supersymmetry.
One can define canonically normalized gravitini and radino in the Einstein frame
3 1/2
1/2
K/12
a
(x),
5 (x) =
(t)eK/12 5 (x).
(x) = a (t)e
(53)
2
Now in the Einstein frame the gravitino variation becomes
eK/6 0 E
, ,
DE
8la(t)
(54)
where we have defined = a 1/2(t)eK/12 and D E is the covariant derivative with respect
to the Einstein metric. The matrix 0 is the local frame gamma matrix. We have used the
fact that the induced metric leads to an extra spin connection
0

eK/6 E
e ,
la(t)
(55)
E is the veilbein of the metric g E . Now
where e
= 0 due to the generator
associated with a Lorentz boost.

The variation of the gravitino (the one propagating on the brane) in the effective 4d
theory is
= 2D + ieK/2W + ,
(56)
where D = + 1/4mn [m , n ]. Comparing this to (54) we can see that the Lorentz
structure of the variation induced by the brane motion is not that of the term associated
with the expectation value of the superpotential W , but rather should be seen as a timedependent background connection.
1 We reserve the label for the brane local time, while t denotes the coordinate time.
159
Usually, in spontaneously broken supergravity a positive cosmological constant is

associated with non-vanishing F terms. Here the subtlety lies in the fact, that the vacuum
solution we have at our disposal is time-dependent, and it is determined in the full theory
containing both branes. It is not obvious that, for instance, in the flipped (twisted) case
of [4] the four-dimensional description of the full system in terms of the standard 4d
supergravity is possible. On the other hand, for an observer localized on the travelling
brane, who has at his disposal local supercharges operating on the brane, it should be
possible to describe the local dynamics in terms of sugra Lagrangian containing degrees of
freedom that are allowed to fluctuate on the brane. We find such an approximate description
possible, but the price to pay is the explicit dependence on time of the supersymmetry
variations. In the case discussed in detail in this section it is natural to keep in the effective
brane theory the degree of freedom corresponding to the relative motion of the branes, the
radion field. The overall motion of the system with respect to the bulk is encoded in the
explicit time dependence of the gravitino variations.
Let us analyse first the type of geometry seen by a low energy observer on the brane.
Taking the ansatz
E
= d 2 + a 2 ( ) dx i dxi
g
(57)
where spatial sections are flat now. The Friedmann equation reads
42
1
(58)
= 2
3
implying that a( ) = . Notice that the Hubble rate as seen from the effective action
coincides with Hubble rate obtained from the induced metric on the brane.
Let us now consider the coupling to supersymmetric matter on the brane. It is crucial to
B = a 2 (t)g
notice that matter couples to the induced metric g
for far away branes. The
kinetic terms are, therefore,
Skin = d 4 x a 2 (t) g g .
(59)
H2
One can deduce the complete Khler potential including the radion superfield T and chiral
matter on the brane [14]

K(T , ) = 3 log 1 e(T +T ) ||2 .
(60)
Let us now go to the Jordan frame where matter on the brane couples directly to the
B = a 2 (t)g . The action contains a term
induced metric g
d 4 x a 2 (t)eK(T ,)/3 g R(g)

d 4 x gB eK(T ,)/3 R(gB ) 6gB ln a ln a .

(61)
Using the explicit form of the Khler potential and time dependence of ln a, this leads to a
soft breaking mass term for the chiral superfields living on the brane
<m2 =
3
,
2
(62)
160
and this is equal, up to the powers of the 4d Planck scale to the cosmological-constant.
Notice that the splitting within the matter multiplets is also related to the Hubble rate on
the brane
<m2 = 3H 2 ,
(63)
showing the intrinsic link between the kinetic breaking of supersymmetry and the
cosmological evolution of the brane. As expected the mass splitting dies off as the brane
recedes away towards infinity.
One may give a simple explanation for the appearance of a direct link between the
mass splitting and the Hubble rate. This relationship comes from the kinetic term for the
modulus T , the only bulk modulus that is left in the effective local sugra. The necessary
ingredient is the presence of the Khler potential mixing between T and chiral multiplets
. This mixing does exist indeed, as seen in (60), and results in

T
Lkin = gT T
2

= 3 T

2
(T +T)

e
1 + ||2 + ,

(64)
hence the (unnormalized) mass terms are

T
m2 = 3
2
(T +T)
e
.

(65)
In this frame the mass terms are induced by the non-trivial time evolution of the modulus
T when we identify the kinetic energy of the modulus T with the Hubble rate as given by
the five-dimensional dynamics, i.e., the Friedmann equation

T
3H 2 = gT T
2

= 3 T

2
(T +T)
e
.

(66)
This again leads to a mass splitting proportional to the cosmological scale H (63).
The result obtained in this section has also another nice interpretation in terms of
temperature supersymmetry breaking on the brane. Focusing on one of the moving brane,
it is easy to see that the asymptotic geometry is of the Milne type (see Appendix A), i.e.,
a flat brane. The motion of the brane in the bulk implies that an observer on the brane
coupled to the bulk vacuum detects a thermal spectrum of temperature [16]
TB =
1
.
2
(67)
Now in flat Minkowski space, a globally supersymmetric theory such as the MSSM is
broken by temperature effects. The mass splitting of the supersymmetric multiplet is of
order O(TB ). This is not surprising, since also the temperature supersymmetry breakdown
can be understood as the result of non-trivial boundary conditions imposed on bosons and
fermions along the Euclidean time direction.
161
6. Flipped supersymmetry breaking

6.1. Branes beyond the horizon
Let us now come back to the flipped case. The two branes have positive tensions
preventing the embedding of the two branes on the same side of the EinsteinRosen bridge
at U = V = 0 (see Appendix A for details on the global structure of spacetime). However,
one can embed the branes on both sides of the EinsteinRosen bridge. In that case the two
branes are separated by a wormhole. The initial condition problem can be analysed and
leads to the same conclusions as for the opposite tension case. Here the two branes are in
causally disconnected regions.
For far away branes we focus on the AdS wormhole q = 1, C = 0. The global structure
of spacetime consists of two copies of the same r l space glued at r = l by a wormhole
whose topology is the one of 1 . To see that one can use the isotropic coordinates defined
by

1
ln +
r = lcosh ln tan
(68)
2 l
4
such that the metric becomes

1
2
2
ln +
ds = sinh ln tan
dt 2
2 l
4

l 2 cosh2 ln tan 12 ln l + 4 2
d + 2 d 2 .
+
2
(69)
Notice that this parametrization realizes explicitly r l. Moreover, the relation between
r and is two to one, i.e., the coordinate [0, ) covers the line r l twice. This is
the origin of the two sides of the horizon. A more precise description can be obtained by
noticing that the two sides of the horizon are in fact exchanged by the isometry
=
l2
.
(70)
The fixed point of this transformation

c = l
(71)
is precisely located at the horizon.

On both sides of the horizon, the branes follow motions identical to the one obtained
in the opposite tension case. The main difference comes from the fact that the horizon is
now visible for both branes. This leads to a greatly modified behaviour of gravity due to
the strong Lorentz violation in the region close to the horizon. However, the contribution
to the four-dimensional Planck scale coming from the near-horizon region is finite and
necessarily much smaller (see the next subsection) than the contributions from the nearbrane regions, because the warp factor grows towards the branes. Hence the effective
supergravity/matter model on the positive tension brane in the asymptotic far-away region
of the spacetime is the same as in the moving RS case discussed earlier.
162
6.2. Gravity and Lorentz violation

We will now consider the graviton in the AdS wormhole with q = 1, C = 0
corresponding to a moving brane receding away towards infinity. We work in the traceless transverse gauge h = 0 and Da hab = 0. The zero modes can be written in the form
hij = eiu ejv huv ,
(72)
where (i, j ) are indices of the symmetric space of curvature q. The graviton depends
on the vielbein eiu on the symmetric space. The matrix huv is symmetric and traceless,
corresponding to five degrees of freedom. Gauge invariance allows to fix three degrees
of freedom, leaving two independent polarizations. Zero modes correspond to timeindependent matrices huv .
The graviton equation is easily obtained by noticing that the AdS wormhole (q =
1, C = 0) is locally isometric to AdS spacetime and therefore [17]
Da D a hij + 2k 2 hij = 0.
(73)
Using the connections

0
=
05
f
,
f
ji5 =
g i
,
g j
5
00
=
f
f
(74)
and
ij5 =
g
gij
g
we find that the graviton equation reads

g
f
g
g g f

2
hij +
+
hij 2
2 +
k hij = 0.
f
g
g
g
g
f
(75)
(76)
Close to the horizon we have f /f 1/x5 , g /g 0 implying that

hij +
1
h + 2k 2 hij = 0.
x5 ij
1/2
Putting hij = x5
Hij +
(77)
Hij we obtain that
1
Hij = 0
4x52
(78)
and close to the horizon

hij = h0ij + ln |x5 |h1ij ,
0,1
are constant tensors in terms of x5 .
where hij
Let us now evaluate the kinetic terms of the graviton

1
2,
dx5 d 4 xfg 3 2 ( h)
252
(79)
(80)
163
where hij = (x5 )h ij and the contractions are made in flat space. Integrating out the extra
dimension between branes yields an effective Planck scale
1
1
=
242 252
dx5 fg 3 2 .
(81)
This is well defined everywhere except at the horizon where the integral behaves like

dx5 x5 ln2 (x5 ) < .
(82)
Hence, the effective Planck mass is finite.

Notice that close to the horizon, Lorentz invariance is broken implying that one cannot
write a gravity theory in the usual Einstein Hilbert form. It is only at spatial infinity where
f = g that Lorentz invariance is restored, allowing to consider more familiar looking
effective actions.
7. Summary
In this paper we have discussed the physics on a non-static brane of locally
supersymmetric five-dimensional supergravity which moves with respect to the bulk
according to the classical equations of motion. We have found that in such a case the
cosmological evolution is related to the dynamics of the supersymmetry breakdown seen on
the brane. Within the brane-bulk supergravity of [2,4] there exist time-dependent solutions
that break 4d supersymmetry everywhere, in the bulk and on the branes. The observer on
the visible brane finds a positive effective vacuum energy which changes as the inverse
square of the local time. Having obtained a decaying cosmological constant, one may
wonder if this is not the realization of quintessence. In fact one finds that such a time
dependence is marginally consistent with the supernovae data [15], however at very late
times the behaviour of the system needs to be modified. The scale of this quintessential
vacuum energy is the same as the scale of susy breaking on the brane, as can be seen
from the explicit computation of the soft masses on the brane, hence asymptotically the
observable brane becomes flat and supersymmetric. Of course, this is strictly true in the
approximation when one neglects the contribution of the matter sector. We have discussed
the effective 4d supergravity that describes local dynamics of the brane-bulk system in the
vicinity of the brane. It turns out that the supersymmetry breakdown in our model is seen on
the brane as the kinetic breaking mediated to the observable sector by the radion multiplet,
via the matterT-modulus mixing in the Khler function. To make the model suitable to
describe realistically very late stages of the evolution of the Universe one needs to enhance
the local scale of the brane supersymmetry breakdown with respect to the scale of the
cosmological constant. The correspondence of the 5d picture to 4d sugra with T -modulus
suggests that the relevant enhancement mechanism would be equivalent to the tuning of
the (super-)potential of the T -modulus and related to its stabilization.
164
Acknowledgements
We thank John Ellis and Patrizia Bucci for very interesting discussions. Z.L. thanks
Theory Division at CERN for hospitality. This work was partially supported by the EC
Contracts HPRN-CT-2000-00152 and HPRN-CT-2000-00148 for years 20002004, by the
Polish State Committee for Scientific Research grants, and by POLONIUM 2003.
Appendix A. Global structure of spacetime

A.1. Global structure
Let us analyse the global structure of spacetime for any value of q and C. To do that
we will define Kruskal coordinates covering spacetime completely. It is convenient to
introduce tortoise coordinates
dr = f (r) dr,
(A.1)
leading to
r =

r r+
r
l 2 r+
r
ln
+
.
arctan
2
2
2 r+ + r
r + r+
r+
r
We have defined r as the roots of

2

2
2
r 4 + ql 2 r 2 Cl 2 = r 2 + r
r r+
.
(A.2)
(A.3)
More explicitly this leads to
ql 2 + l 4 + C 2 l 2
ql 2 + l 4 + C 2 l 2
2
2
r+ =
(A.4)
,
r =
.
2
2
Notice that r+ = l, r = 0 when q = 1 and C = 0. The metric is originally defined for
r r+ . We will extend it to the origin. Let us introduce the surface gravity
+ =
2 + r2
r+
.
l 2 r+
(A.5)
Now the EddingtonFinkelstein coordinates,

u = t r ,
v = t + r ,
(A.6)
correspond to null directions. The Kruskal coordinates are

U = e+ u ,
V = e+ v
in such a way that

r r+ 2 2 rr arctan rr
.
UV =
e +
r + r+
(A.7)
(A.8)
The metric reads now

r
r

1
2 2 r+ arctan r
ds 2 = 2 (r + r+ )2 r 2 + r
dU dV + r 2 dq2 .
e
2
+ r
165
(A.9)
As can be seen in the last expression the metric can be extended to 0 < r r+ without
any coordinate singularity. After doing so the product U V goes from negative to positive
values. The horizon r = r+ is mapped to the two lines U = 0 and V = 0. In particular one
should notice that the regions on both sides of the EinsteinRosen bridge at U = V = 0 are
isometric and exchanged under the isometry U U , V V .
A.2. Milne branes
Focusing on the case of branes receding away towards infinity, the motion is well
approximated by the Friedmann equation
1
,
r2
where H = r /r with the proper time defined by
2

r
1
2 1 t2 + 2
r 2 = 1.
r
l
1
l2
H2 =
This implies that

2

l
t = ln 2 1 .
2
l
(A.10)
(A.11)
(A.12)
This asymptotic situation corresponds to the case q = 1, C = 0. The geometry on the

brane is dictated by the Milne metric
2
dsB2 = d 2 + 2 d1
,
(A.13)
where
d 2 = d 2 + sinh2 d 2
(A.14)
d 2
and
is the metric on the unit two-sphere. Locally Milne space is isometric to
Minkowski space upon using the following reparametrization t = cosh , r = sinh,
leading to
dsB2 = dt 2 + dr 2 + r 2 d 2 .
(A.15)
Therefore the brane is in fact a flat brane embedded in AdS5 .

Appendix B. Brane temperature
Let us consider a free scalar field living on a Milne brane
ds 2 = dt 2 + t 2 d 2 .
(B.1)
166
The KleinGordon equation can be easily reduced to a differential equation by separation

of variables
= Jk ()k (t),
(B.2)
where the Jk s are hyperbolic harmonics

(sinh2 Jk )
2
sinh

= k 2 l 2 + 1 Jk
(B.3)
and hyperbolic angular momenta k are measured in units of 1/ l. This leads to the Klein
Gordon equation
3
k2l2 + 1
k = 0.
k + k +
t
t2
The modes are given by
1ikl
t
.
k =
l
(B.4)
(B.5)
This defines the positive and negative frequency modes in Milne space. One can quantize
the fields by expanding

= dk k ak + +k ak ,
(B.6)
where the operators ak and ak are annihilation and creation operators for the vacuum
| vacB on the brane.
The same quantization procedure for a static observer, r = r0 = const, in the bulk whose
position is instantaneously coincident with the brane leads to the differential equation
d 2 k k 2 l 2 + 1
+
k = 0,
dto2
l2
where
to =
r02 l 2
t.
r
This leads to the static observers modes
k = ei t ,
as in flat Minkowski space where
k2l2 + 1
=
l
and
r0
=
.
r02 l 2
(B.7)
(B.8)
(B.9)
(B.10)
(B.11)
167
We then identify with the energy of the particles with hyperbolic angular momentum k.
At high energy 1/ l and large distances from the horizon r0 l we find that the energy
coincides with k and .
We can then decompose fields as

= d ei t a + +ei t a ,
(B.12)
where the operators a and a are annihilation and creation operators for the static bulk
vacuum |vac.

The two Hilbert spaces can be seen not to be unitarily equivalent. Indeed, one can
decompose

k= dp Ak ei t + Bk ei t
(B.13)
from which we get

ak .
a = dk Ak ak + Bk
(B.14)
The Bogoliubov coefficients depend on the overlap between the eigenstates in the two
Hilbert spaces
Ak =
(B.15)
k eit dt.
4
Similarly we have that
Bk = Ak, .
Using the explicit expression of
(B.16)
k
one finds [18]
Bk = ekl Ak .
Using the normalization conditions for the Bogoliubov coefficients

AA kp BB kp = kp ,
(B.17)
(B.18)
we find that

BB kp =
kp
(B.19)
e2kl 1
as a function of the hyperbolic angular momentum k. From this we conclude that the
number of particles on the brane with momentum k is given by
Nk =
,
1
i.e., a thermal spectrum with a temperature 1/2l.
It is more transparent to reexpress this result as a function of the energy
E=
e2lk
dt
,
d
(B.20)
(B.21)
168
corresponding to the energy measured by an observer whose clock ticks at the same rate as
the clock on the moving brane. This leads to
NE =
e2
E 2 (r 2 l 2 )1
(B.22)
corresponding to a thermal spectrum at high energy on the brane with temperature

1

TB =
2l
r2
l2
(B.23)
At large distances, when the brane is far away from the horizon we find that the
temperature is
TB =
1
,
2
(B.24)
where is the proper time on the brane.
References
[1] R. Altendorfer, J. Bagger, D. Nemeschansky, Supersymmetric RandallSundrum scenario, Phys. Rev. D 63
(2001) 125025, hep-th/0003117.
[2] A. Falkowski, Z. Lalak, S. Pokorski, Supersymmetrizing branes with bulk in five-dimensional supergravity,
Phys. Lett. B 491 (2000) 172, hep-th/0004093.
[3] E. Bergshoeff, R. Kallosh, A. Van Proeyen, Supersymmetry in singular spaces, JHEP 0010 (2000) 033,
hep-th/0007044.
[4] P. Brax, A. Falkowski, Z. Lalak, Non-BPS branes of supersymmetric brane worlds, Phys. Lett. B 521 (2001)
105, hep-th/0107257.
[5] P. Brax, Z. Lalak, Brane world supersymmetry, detuning, flipping and orbifolding, Acta Phys. Pol. B 33
(2002) 2399, hep-th/0207102.
[6] Z. Lalak, R. Matyszkiewicz, On ScherkSchwarz mechanism in gauged five-dimensional supergravity and
on its relation to bigravity, Nucl. Phys. B 649 (2003) 389, hep-th/0210053.
[7] T. Gherghetta, A. Pomarol, Bulk fields and supersymmetry in a slice of AdS, Nucl. Phys. B 586 (2000) 141,
hep-ph/0003129.
[8] L. Randall, R. Sundrum, An alternative to compactification, Phys. Rev. Lett. 83 (1999) 4690, hepth/9906064.
[9] A. Ceresole, G. DallAgata, General matter coupled N = 2, D = 5 gauged supergravity, Nucl. Phys. B 585
(2000) 143, hep-th/0004111.
[10] L. Randall, R. Sundrum, A large mass hierarchy from a small extra dimension, Phys. Rev. Lett. 83 (1999)
3370, hep-ph/9905221.
[11] P. Kraus, Dynamics of anti-de Sitter domain walls, JHEP 9912 (1999) 011, hep-th/9910149.
[12] P. Bowcock, C. Charmoussis, R. Gregory, General brane cosmologies and their global spacetime structure,
Class. Quantum Grav. 17 (2000) 4745, hep-th/0007177.
[13] Z. Lalak, R. Matyszkiewicz, Boundary terms in brane worlds, JHEP 0111 (2001) 027, hep-th/0110141.
[14] A. Falkowski, Z. Lalak, S. Pokorski, Four-dimensional supergravities from five-dimensional brane worlds,
Nucl. Phys. B 613 (2001) 189, hep-th/0102145.
[15] J.R. Ellis, N.E. Mavromatos, D.V. Nanopoulos, Time-dependent vacuum energy induced by D-particle
recoil, Gen. Relativ. Gravit. 32 (2000) 943, gr-qc/9810086.
169
[16] S. Das, A. Zelnikov, Unruh radiation, holography and boundary cosmology, Phys. Rev. D 64 (2001) 104001,
hep-th/0104198.
[17] C. Csaki, J. Erlich, C. Grojean, Gravitational Lorentz violation and adjustment of the cosmological constant
in asymmetrically warped spacetimes, Nuc. Phys. B 604 (2001) 312, hep-th/0012143.
[18] P. Townsend, Black holes, gr-qc/9707012.

Exact results in non-supersymmetric large N

orientifold field theories
A. Armoni a , M. Shifman a,b , G. Veneziano a
a Theory Division, CERN, CH-1211 Geneva 23, Switzerland
b William I. Fine Theoretical Physics Institute, University of Minnesota, Minneapolis, MN 55455, USA 1
Received 7 March 2003; received in revised form 27 May 2003; accepted 17 June 2003
Abstract
We consider non-supersymmetric large N orientifold field theories. Specifically, we discuss a
gauge theory with a Dirac fermion in the antisymmetric tensor representation. We argue that, at
large N and in a large part of its bosonic sector, this theory is nonperturbatively equivalent to N = 1
SYM, so that exact results established in the latter (parent) theory also hold in the daughter orientifold
theory. In particular, the non-supersymmetric theory has an exactly calculable bifermion condensate,
exactly degenerate parity doublets, and a vanishing cosmological constant (all this to leading order
in 1/N).
PACS: 11.15.Pg; 11.25.-w; 11.30.Pb
1. Introduction
Gauge field theories at strong coupling are of great importance in particle physics. Exact
results in gauge theories at strong coupling have a special weight. In the last few years
supersymmetry (SUSY) proved to be a guiding principle (see, e.g., [1]) providing deep
insights in gauge dynamics. Recently it was suggested that a large number of symmetry
relations valid in supersymmetric theories remain valid in the large N limit in nonsupersymmetric daughter theories obtained from the parent one through orbifoldization [2].
The most popular (non-SUSY) orbifolds are Z2 , or Zk in general [35]. In particular, in
E-mail addresses: adi.armoni@cern.ch (A. Armoni), michael.shifman@cern.ch (M. Shifman),
gabriele.veneziano@cern.ch (G. Veneziano).
1 Permanent address.
doi:10.1016/S0550-3213(03)00538-8
A. Armoni et al. / Nuclear Physics B 667 (2003) 170182
171
the last paper it was argued that the superpotential of SUSY gluodynamics [6],
W (S) = N(S log S/3 S),
S = + ,
(1)
can be in a sense extended to its Z2 orbifold. This is equivalent to the statement that the
daughter theory has N vacua labeled by a bifermion condensate

2k
3
.
k = N exp i
(2)
N
While the planar equivalence is certainly true in perturbation theory [7,8] its nonperturbative status is being debated; a number of arguments pro and con can be found in the
literature [35].
The purpose of this work is to present a non-supersymmetric daughter (different from
the Zk orbifolds) for which the large N equivalence between the parent SUSY theory
and the daughter non-supersymmetric one at nonperturbative level rests on a more solid
ground than in the case of the Zk orbifolds. This orientifold field theory was suggested
in Refs. [9,10] in a somewhat different context. One of its advantages over more popular
Zk orbifolds is the absence in it of the twisted sector.
The field content of the orientifold gauge field theory differs from the one of its parent
theory, U(N ) SUSY gluodynamics, in that the gluinos are replaced by massless Dirac
fermions in the rank-two antisymmetric tensor representation of U(N ) (denoted by + ).
The total number of (say) left handed fermions is thus N(N 1) in the daughter theory
and N 2 in the parent theory and agrees to leading order in 1/N . Similarly, one can discuss
+
). As we will see,
a theory with a Dirac fermion in the symmetric representation (
this theory lives on a brane configuration of type 0A string theory [9] which consists of NS5
branes, D4 branes and an orientifold planehence the name orientifold field theory. The
daughter theory in our case is a much closer cousin of N = 1 SYM than the Z2 orbifold.
Indeed, the gauge groups in the parent and daughter theories are the same, and no rescaling
of the gauge couplings is needed.
Assuming that both theories are in the confining regime,2 we will show that in the
large N limit many results and symmetry relations that were obtained for N = 1 SYM
hold also for the above orientifold field theory. One specific quantity is the bifermion
condensate (2). It labels distinct vacua. The number of vacua turns out to be the same in
the parent and daughter theories, N . Another result is the vanishing of the vacuum energy
density in the daughter theory.
We would like to emphasize that the spectrum of the orientifold field theory does not
coincide with that of N = 1 SYM in the large N limit. In particular, there is no SUSY.
While the composite color-singlet hadrons of N = 1 are FermiBose degenerate, the
composite color-singlet hadrons of the orientifold field theory are purely bosonic.
The organization of this paper is as follows. In Section 2 we present our main result
the perturbative and nonperturbative equivalence. In Section 3 we briefly present the string
theory realization of the theory in its conjectured relation to M-theory. In Section 4 we
compare our analysis and results with those referring to Zk orbifold field theories and
summarize conclusions.
2 In the Higgs regime the 1/N expansion becomes more subtle, see a discussion below.
172
2. Orientifold field theory and N = 1 SYM

In this section we will argue that in the N limit there is a sector in the orientifold
theory exactly identical to N = 1 SYM and, therefore, exact results on the IR behavior of
this theory can be obtained.
The parent theory is N = 1 SUSY gluodynamics with the gauge group U(N ). In the
large N limit the U(1) factor is irrelevant. The daughter theory has the same gauge group,
and the same gauge coupling.3 The gluino field ij is replaced by two Weyl spinors [ij ]
and [ij ] . We can combine the Weyl spinors into one Dirac spinor, either [ij ] or [ij ] .
Note that the number of fermionic degrees of freedom is N 2 N , as in the parent theory
in the large N limit.
The hadronic (color-singlet) sectors of the theories are different. In the parent theory
composite fermions with mass scaling as N 0 exist, and moreover, they are degenerate with
their bosonic SUSY counterparts. In the daughter theory any interpolating color-singlet
current with the fermion quantum numbers contains a number of constituents growing
with N . Hence at N = the spectrum contains only bosons.
Classically, the parent theory has just an R symmetry corresponding to chiral rotations
of the gluino field. Instantons break this symmetry down to Z2N . The daughter theory has,
on top, a conserved anomaly free current
.
(3)
In terms of the Dirac spinor this is the vector current . If the corresponding charge
is denoted by Q, in the color-singlet bosonic sector Q = 0, with necessity. Then the only
global symmetry which remains in both theories is Z2N spontaneously broken down to Z2
by the respective bifermion condensates. This explains the existence of N vacua in both
cases. We will compare the bosonic sectors of the parent and daughter theories. Note that
in the daughter theory the part of the bosonic sector probed by the operators of the type (3),
which have no analogs in the parent theory, is inaccessible.
Let us start from perturbative consideration. The general argument for any orbifold/orientifold field theory is given in [7,8]. Let us see how it works in our orientifold
field theory. The Feynman rules of the planar theory are shown in Fig. 1. The difference
between the orientifold theory and N = 1 is that the arrows on the fermionic lines point
in the same direction, since the fermion is in the antisymmetric representation, in contrast
to the supersymmetric theory where the gaugino is in the adjoint representation and the
arrows point in the opposite directions. This difference between the two theories does not
affect planar graphs provided each gaugino line is replaced by the sum of [..] and [..] .
There is a one-to-one correspondence between the planar graphs of the two theories.
Diagrammatically this works as follows (see, for example, Fig. 2). Consider any planar
diagram of the daughter theory: by definition of planarity it can be drawn on a sphere.
3 To be more precise, the gauge groups are almost the same. The U(1) factor completely decouples in the
parent theory, while it does not decouple in the daughter one. Moreover, in the former theory the Zk center of the
gauge group acts trivially while in the latter one it is only Z2 that acts trivially. These distinctions are unimportant
in the large N limit.
173
Fig. 1. (a) The fermionic propagator and the vertex. (b) N = 1 SYM. (c) The non-SUSY theory.
Fig. 2. (a) A typical planar contribution to the vacuum polarization. (b) For N = 1 SYM. (c) For the non-SUSY
theory.
The fermionic propagators form closed, non-intersecting loops that divide the sphere into
regions. Each time we cross a fermionic line the orientation of color-indices loops (each
one producing a factor N ) changes from clock- to anti-clock-wise, and vice-versa, as easily
seen in Fig. 2(c). Thus, the fermionic loops allow to attribute to each one of the above
regions a binary label (say 1) according to whether the color loops go clock- or anticlock-wise in that region. Now imagine that the orientation of color loops in all regions
with the 1 label is reversed. We will get a planar diagram of the SYM theory in which
all color loops go, by convention, clock-wise. The number associated with both diagrams
will be the same since the diagrams inside each region always contain an even number of
powers of g so that the relative minus signs of Fig. 1 do not matter.
Let us illustrate how this works, say, for the inside part of the graph in Fig. 2(b). In
the parent theory we have the color factor Tr(T a T b T c )f abc while in the daughter one
Tr(T a T b T c )f abc where T = T and the tilde marks the transposed matrix.4 Using the
fact that

T a T b = if abc T c ,
a b
T T = if abc T c ,
we immediately come to the conclusion that the above two expressions coincide.
4 To make the following expressions concise we use a shorthand, T a = T a and T a = T a .

174
Thus, all perturbative results that we know of in N = 1 SYM apply for the orientifold
model as well. For example, the function of the orientifold field theory is

3N 2
1
1
1+O
,
=
(4)
2 1 (N)/(2)
N
2
g
where = 4
. In the large N limit it coincides with the N = 1 SYM result [11]. Note that
the corrections are 1/N rather than 1/N 2 . For instance, the exact first coefficient of the
function is 3N 4/3 versus 3N in the parent theory.
Now let us argue that the perturbative argument can be elevated to nonperturbative level
in the case at hand. A heuristic argument in favor of the nonperturbative equivalence is that
the coincidence of all planar graphs of the two theories implies that the relevant Casimir
operators of the two representations are equivalent in the large N limit. The partition
functions of the two theories depend on the Casimir operators and therefore must coincide
as well.
A more formal line of reasoning is as follows. It is essential that the fermion fields
enter bilinearly in the action, and that for any given gauge field configuration in the parent
theory there is exactly the same configuration in the daughter one. (The latter feature is
absent in the Zk orbifolds.) Our idea is to integrate out fermion fields for any fixed gluon
field configuration, which yields respective determinants, and then compare them.
Consider the partition function of N = 1 SYM,

Z0 = DA D exp iS[A, , J ] ,
(5)
where J is any source coupled to color-singlet gluon operators (we will discuss colorsinglet fermion bilinears later).
For any given gluon field, upon integrating out the gaugino field, we obtain

a
,
Z0 = DA exp iS[A, J ] det / + A
(6)
/ a Tadj
a
where Tadj
is a generator of the adjoint representation.
If one integrates out the fermion fields of the non-supersymmetric orientifold theory, at
fixed A, one arrives at a similar expression, but with the generators of the antisymmetric
a
Tasa .
representation instead of the adjoint, Tadj
To compare the fermion determinants in the parent and daughter theories (assuming
that the gauge field configuration Aa (x), a = 1, . . . , N 2 , is the same and fixed) we must
cast both fermion operators in similar forms. To this end we will extend both theories. In
the parent one we introduce the second adjoint Weyl fermion ji and combine two Weyl
fermions into one Dirac adjoint fermion ji . The determinant in this extended theory is the
square of the original one; we assume that taking the square root at the end is harmless.
In the daughter theory instead of [ij ] , we will work with the reducible representation,
combining both symmetric and antisymmetric, ij (no (anti)symmetrization over the color
indices). Then the number of the fermion Dirac fields is the same as in the extended parent.
Again, as in the previous case, the determinant in the extended daughter theory is the
square of the original one. For infrared regularization we will introduce small mass terms,
assuming that the vanishing mass limit is smooth.
175
Next, we will make use of the fact that

a
Ta
Tadj
= Ta 1 + 1 T a ,
(7)
and
Tda = Ta 1 + 1 Ta ,
(8)
Tda = T a 1 + 1 T a .
(9)
or
The subscript d stands for daughter. Let us now introduce, as an auxiliary object,

/ a Ta 1 + B/ a 1 T a ,
F (A, B) = tr ln / + A
(10)
and notice that (at large N ) F is invariant under separate gauge transformations
acting on
Aa and B a . Using the fact that the Wilson loop operators WC (A) = tr P exp(i C A) form
a complete set of gauge-invariant operators, we can write

exp F (A, B) =
(11)
C ,C WC (A)WC (B).
C ,C
The partition function of N = 1 SYM can be written, at large N , as

1
Z0 = DA DB (A B) exp S[A] + S[B]
2

C ,C WC (A)WC (B).
(12)
C ,C
Now let us turn to the daughter theory. The partition function of the orientifold theory,
at large N , can be written as (12), but with the orientations of the Wilson loops W(B)
reversed, since we replace the fundamental fermions by antifundamental fermions. This
gives
WC WC = {WC } .
Moreover, WC is real, and this is why reversing the orientation should not change the value
of the partition function.
Let us present now a more detailed derivation which relies on the fact that at large N
the two kinds of gluons do not interact with each other. The partition function (12) can be
written as

Z0 = DA exp S[A] + F (A, B = A) .
(13)
Let us expand exp(S)
n

(1)n
S[A] .
exp S[A] =
n!
n
(14)
176
At N , the vertices or propagators coming from each factor of S cannot connect A

with B. Thus, we have
S
n
Ck S k [A]S nk [B],
(15)
k=0
where the sign means can be replaced by and Ck is a combinatorial factor

corresponding to the various choices of picking k out of n,
Ck =
n!
.
k!(n k)!
Then
(1)n

k
nk
n!
S[A] S[B]
k!(n
k)!
n
k

= exp S[A] S[B] .
exp(S)
n!
(16)
Thus, we conclude that exp(S) acts as exp(S[A] S[B]) and, at large N , we can think
of the partition function as if we actually have two gauge fields. Thus, at large N , Eq. (12)
becomes

Z0 =
(17)
C ,C WC WC .
C ,C
In order to pass to the daughter theory one has to replace B = B a T a by B = B a Ta .

But B is a dummy variable and, hence, this substitution will not change the value of
the partition function. Though we used a perturbative intuition in order to arrive to the
factorization of either the Wilson loops in (12) or the action in Eq. (16) we believe that
this is an exact property of the partition function. Note also that the factorization in the
partition function follows immediately from the existence of a master field [12]. Thus,
given the existence of a large N master field, the equivalence of the parent and daughter
theories is proven.
Another line of reasoning supporting nonperturbative planar equivalence of the parent
and orientifold theories is based on lattice formulations of both theories. If one examines
strong coupling diagrams for, say, a Wilson loop on the lattice, one readily concludes that
the large N diagrams of the two theories can be put in a one-to-one correspondence and
agree to leading order in 1/N .
For all these reasons we believe that the full equivalence of these two theories takes
place in the large N limit, for quantities which do not involve fermion external legs.
The equivalence is nonperturbative. An immediate result is that the vacuum energy of the
orientifold theory is zero in the large N limit (or at finite N it is 1/N relatively to a natural
N 2 behavior). To be more precise, the expected dependence of the vacuum energy on the
UV cut-off and N is generically

c2
c1
4
2
+ 2 + .
= UV N c0 +
(18)
N
N
177
In the present theory c0 = 0 while, generically, c1 , c2 , . . . are nonzero. Namely, the vacuum
energy is zero only in the planar theory. If one keeps the combination 4UV N 2 fixed then
the limiting theory has a zero vacuum energy. While this is an almost trivial statement
in the UV, simply because of the equivalence of the planar graphs, it is highly nontrivial
from the IR point of view. The daughter theory hadronic spectrum consists of bosons only,
since it is impossible to form light color-singlet fermions. Nonetheless, one should remark
that contributions to the cosmological constant from color-singlet loops only enter at the
level of the genus-1 (torus) diagrams. These are already O(1/N 2 ) down with respect to the
leading contribution to the cosmological constant (which vanishes) and O(1/N) relative
to the presumed leading nonvanishing contribution in the daughter theory.
In order to go beyond the equivalence of vacuum diagrams and correlators with external
gluonic sources we have to understand how the above argument can be extended if we add
fermionic bilinear sources.
It looks quite obvious that sources coupled to in the parent theory can be
mapped into sources coupled to (1 5 ) in the daughter theory. This will be enough
to prove the equality of condensates in the two theories and the x-independence of certain
correlators, hence the parity-doublet structure of the spectrum in both theories (see below).
In particular, a mass terms can be added to both theories without spoiling their large N
equivalence. Although we have not made a systematic study of this problem it looks that
many other fermionic bilinears (FB)involving the gluon field as wellcan be mapped
in the two theories so that
i
i

j
j
, JFB = Wori Jglue
, JFB 1 + O(1/N)
WSYM Jglue
(19)
j
j
with an explicit dictionary relating JFB and JFB . An example of such a pair is
(G )ij ( )ki ( )k (G )ij ( )[ik] ( )[kj ] ,

j
(20)
where (G )ij is the gluon field strength tensor.

As was mentioned, we can derive the bifermion condensate = 0 in the daughter
theory starting from = 0 in the parent one, hence
2k
k = N3 exp i
,
N
as in Eq. (2).
Another property which is inherited by the daughter theory is the constancy of the chiral
correlator
(x1 ) (x2 ) (xn ) = const.
(21)
Physically it is related to the mass degeneracy of scalar and pseudoscalar mesons.

It should be stressed that in Zk orbifolds some background gauge field configurations
are present in the parent theory and absent in the daughter one, so that comparison of the
fermion determinants does not prove nonperturbative equivalence.
To summarize, we can relate many correlators which involve even number of fermions
and/or gluon operators (hence, the corresponding hadron spectra). However, the parent
and daughter theories differ from each other in the sector of odd number of fermions. For
178
example, there is a gauge invariant three-fermions state in N = 1 SYM. Such a state does
not exist in the orientifold theory. Remember also that the sector of the daughter theory
probed by the operators of the type (3) is not accessible for predictions.
3. The relation with type 0A string theory and M-theory

In this section we show that our orientifold field theory has a simple realization in
type 0A string theory. Type 0A/B are bosonic closed string theories with a low-energy
spectrum which consist of the universal NSNS sector, a tachyon and a doubled set of
RR forms with respect to type IIA/B string theories. Type 0B string theory has three
kinds of orientifolds [13,14]. We will be mostly interested in the nontachyonic one [13,14].
Consider the action (1)fR , namely the world-sheet parity combined with the worldsheet fermion number. The NSNS vacuum is odd with respect to this action and, therefore,
it is removed from the spectrum. Thus, the bulk theory is tachyon free. In addition this
orientifold removes half of the RR fields so that the theory now has only one set of RR
fields, as in the type IIB theory. In order to remove RR tadpoles one has to introduce 32
D9 branes. The field theory on the D9 branes is a 10-dimensional U(32) gauge theory with
an antisymmetric fermion.
One can perform a sequence of T dualities to obtain a system of D4 branes and O4
plane. Moreover, a brane configuration which consists of NS5 branes and an O4 plane as
in Fig. 3, leads [10] to our orientifold field theory.
We would like to use the relation of type 0A string theory to M-theory to study the
strong coupling regime of the orientifold theory at large N . This argument is certainly not
a proof and we present it just to indicate that our results in Section 2 are in agreement with
the string theory picture.
Similarly to type IIA string theory, type 0A can be obtained from M-theory by a
compactification of the eleventh dimension on a ScherkSchwarz circle [15]. By using
this conjecture we can lift the brane configuration (Fig. 3) to M-theory, similarly to the
lift of the analogous type IIA brane configuration [1618]. Note that the presence of the
orientifold plane can be neglected in the large N limit, since its RR charge is negligible
with respect to the RR charge of the branes (N ). Similarly to the type IIA situation we
will obtain a smooth M5 brane and the resulting curve (the shape of the M5) will be the
Fig. 3. The type 0A brane configuration. The solid lines denote D4 branes. The dashed line denotes the orientifold
4-plane.
179
same as the curve of N = 1 SYM [18]

S N = 1.
(22)
The meaning of this curve is that there are N vacua with the bifermion condensate (2) as
the order parameter, in agreement with our field theory results.
4. Discussion and conclusions

One of the goals of this section is to compare in more detail the results that are obtained
in this work for orientifold field theories with the previous results for orbifold field theories.
In order to be concrete let us discuss, as an example, the Z2 orbifold theory (see Table 1).
This theory as well has a realization in type 0A theory [9]. It lives on a brane configuration
of type 0A which consists of electric and magnetic D-branes, hence the labels e and
m in Table 1.
Let us divide the operators in the Z2 orbifold theory to operators that are invariant
(even) under the exchange of the labels e and m and operators which are odd under
this exchange. The first are called untwisted operators, the second twisted operators.
2 + tr F 2 is an untwisted operator whereas O =
For example, the operator O+ = tr Fee
mm
2
2
tr Fee tr Fmm is a twisted operator.
The perturbative relation between the orbifold theory and its supersymmetric parent
concerns only the untwisted sector [7,8]. The parent theory does not carry information
about the twisted sector of the daughter theory. It is always assumed that the vacuum of the
daughter theory is Z2 -invariant. However, the Z2 -symmetry might be broken dynamically,5
due to a an expectation value of O .
A possible sign of the Z2 instability comes from perturbation theory. Indeed, let us
assume for a moment that at some UV scale where perturbation theory is applicable the
gauge couplings of the two U(N) factors in the orbifold theory are slightly different. We
will denote 2/e = ze and 2/m = zm where the subscripts e, m refer to the first and
second U(N) factors, respectively. It is not difficult to find the renormalization group flow
of z towards the IR domain. As long as e,m we have
3N 2
d(z)
= 2 (z) + higher orders.
d ln
z
(23)
Neglecting weak logarithmic dependence of we get

z() = z(0 )
0
3N 2 2 /(4 2 )
.
(24)
If z is small in UV, it grows towards the IR domain, an indication on a destabilization

tendency. A similar analysis for the conformal daughter theory of extended SUSY, N = 4,
was carried out in Ref. [19].
5 A.A. thanks Y. Shadmi for suggesting this scenario.
180
Table 1
The Z2 orbifold theory
Ue (N ) Um (N )
Aee
Amm
em
me
adj.
1

1
adj.

The advantage of the orientifold theory over the orbifold theory is the absence of the
twisted sector. Moreover, the gauge groups are the same in the parent and daughter theories,
and so are the patterns of the spontaneous breaking of the global symmetry and the numbers
of vacua. This is also seen from the string theory standpoint. The orbifold field theory
originates from the type 0A string theory which is obtained by a Z2 orbifold of type IIA.
The result is a bosonic string theory with a tachyon in the twisted sector. In contrast, the
orientifold field theory originates from a configuration with an orientifold which removes
the twisted sector (the result is very similar to the bosonic part of the type-I string).
If the equivalence between N = 1 SUSY gluodynamics and orientifold theory does
hold nonperturbatively, this must have a strong consequence for the symmetry of the IR
theory. Indeed, the degeneracy of the meson masses inherited from the parent theory would
imply that in the daughter theory there is a tensorial operator (other than the energy
momentum tensor) which is conserved in the large N limit. This is in no contradiction with
the ColemanMandula theorem [21] since in the large N limit all scattering amplitudes
vanish, and the S matrix tends to unity. Note that we discuss here only the large N theory,
namely, the theory of planar graphs. In this limit the theory is expected to be a free theory
of color singlet glueballs. Glueball couplings are all suppressed by powers of 1/N .
Since the perturbative planar equivalence does not depend on the geometry of space
time, one can compactify one or more dimensions, with a small compactification size, to
make the theory weakly coupled. One can then compare the parent and the orbifold theories. This was done, in particular, in Ref. [3] where it was shown that the toron contributions
in the parent and Z2 orbifold theories do not match (T 4 compactification is implied).
Compactifying one or more extra dimensions one should be very cautious, however. In
doing so we get a theory with scalar moduli (flat directions at the classical level). It may
happen (and in fact happens) that one or several components of the moduli fields develop
vacuum expectation values which scale as N 1/2 . This breaks (a part of) the gauge symmetry
leading to the Higgs regime. Simultaneously, the 1/N expansion is broken too. Indeed, in
this expansion we assume that the smallness of g 2 is compensated not by a large value of
the gluon field (the gluon propagator is O(N 0 )) but by a large number of components of
the gluon fields circulating in loops. Upon compactification we get scalar fields, just a few
components of which may condense, compensating the smallness of g 2 .
An example was given by Tong [4], who considered R 3 S compactification at one
loop. On R 3 S the third spatial component of the gluon field becomes a scalar field
with a flat (vanishing) potential at the classical level, (A3 )ij ji . (Alternatively, one
may speak of the Polyakov line in the x3 direction.) In the parent theory the flatness
is (perturbatively) maintained to all orders by supersymmetry. Nonperturbatively, the
flatness is lifted by R 3 S instantons (monopoles) which generate a superpotential.
181
It turns out that in supersymmetric vacua the non-Abelian gauge symmetry is completely
broken by ji = vi ji , down to U(1)2N , sothat the theory is in the Coulomb phase [20].
The expectation values vi scale as 1/g N . This explains why the generated masses
gvi N 0 .
Tong showed that in the daughter theory a potential emerges at one loop, making the
point of the broken gauge symmetry unstable. Shifting from this point, for a trial, one finds
oneself in the Z2 -noninvariant (or twisted) sector, for which no planar equivalence exists,
and which proves to be energetically favored in this case. In the true vacuum the energy
density is negative rather than zero and the full gauge symmetry is restored. The daughter
theory is in the confining phase. Obviously, then there is no equivalence.
The above remark implies that in considering equivalence between N = 2 (or N = 4)
theories where scalar moduli are abundant, with the corresponding orbifold/orientifold
theories, one should be sure to be in the confining rather than Higgs regime.
Another indication that the presence of the twisted sector in Zk orbifolds may have
a negative impact on the untwisted sector came from consideration of the low-energy
theorems. In particular, topological susceptibilities in the parent and Z2 daughter theories
were analyzed in [3].
The topological susceptibility reflects dependence of the vacuum energy on the vacuum
angle . For massless fermions such dependence is absent and the topological susceptibility
vanishes. Only if m = 0, the topological susceptibility does not vanish and can be readily
derived to leading order in m. Thus it is necessary to deform the parent/daughter theories
by fermion mass terms. This deformation does not affect perturbative planar equivalence.
It was shown [3] that under certain reasonable assumptions the topological susceptibilities do not match, the discrepancy being a factor of 2. This factor can be traced back to
the fact that the number of vacua in U(2N ) SUSY parent is 2N while its Ue (N) Um (N)
daughter (Z2 orbifold) has only N vacua.
Needless to say, in our case of the orientifold daughter the topological susceptibilities
are identical, as so are the gauge groups, gauge couplings and the number of vacua.
In conclusion, let us formulate a question which naturally comes to ones mind at the
end of this paper: What is the symmetry of the daughter theory, weaker than SUSY, which
nevertheless implies infinite number of degeneracies in the spectrum?
Acknowledgements
A.A. would like to thank O. Aharony, C. Angelantonj, O. Bergman, Y. Frishman,
A. Hanany, R. Rabadan, E. Rabinovici, and Y. Shadmi for useful discussions. M.S. is
grateful to A. Gorsky and D. Tong for numerous communications and discussions of the Zk
orbifold theories. The work of M.S. is supported in part by DOE grant DE-FG02-94ER408.
References
[1] N. Seiberg, in: K.C. Wali (Ed.), Proceedings 4th Int. Symposium on Particles, Strings, and Cosmology
(PASCOS 94), Syracuse, New York, May 1994, World Scientific, Singapore, 1995, p. 183, hep-th/9408013;
N. Seiberg, Int. J. Mod. Phys. A 12 (1997) 5171, hep-th/9506077.
182
[2] M.J. Strassler, On methods for extracting exact nonperturbative results in non-supersymmetric gauge
theories, hep-th/0104032.
[3] A. Gorsky, M. Shifman, Phys. Rev. D 67 (2003) 022003, hep-th/0208073.
[4] D. Tong, Comments on condensates in non-supersymmetric orbifold field theories, hep-th/0212235.
[5] R. Dijkgraaf, A. Neitzke, C. Vafa, Large N strong coupling dynamics in non-supersymmetric orbifold field
theories, hep-th/0211194.
[6] G. Veneziano, S. Yankielowicz, Phys. Lett. B 113 (1982) 231.
[7] M. Bershadsky, Z. Kakushadze, C. Vafa, Nucl. Phys. B 523 (1998) 59, hep-th/9803076.
[8] M. Bershadsky, A. Johansen, Nucl. Phys. B 536 (1998) 141, hep-th/9803249.
[9] A. Armoni, B. Kol, JHEP 9907 (1999) 011, hep-th/9906081.
[10] C. Angelantonj, A. Armoni, Nucl. Phys. B 578 (2000) 239, hep-th/9912257.
[11] V.A. Novikov, M.A. Shifman, A.I. Vainshtein, V.I. Zakharov, Nucl. Phys. B 229 (1983) 381;
V.A. Novikov, M.A. Shifman, A.I. Vainshtein, V.I. Zakharov, Phys. Lett. B 166 (1986) 329.
[12] E. Witten, Recent developments in gauge theories, in: G. t Hooft, et al. (Eds.), Proceedings of the 1979
Cargese Summer Institute, Plenum, New York, 1980.
[13] A. Sagnotti, Some properties of open string theories, hep-th/9509080.
[14] A. Sagnotti, Nucl. Phys. B (Proc. Suppl.) 56 (1997) 332, hep-th/9702093.
[15] O. Bergman, M.R. Gaberdiel, JHEP 9907 (1999) 022, hep-th/9906055.
[16] K. Hori, H. Ooguri, Y. Oz, Adv. Theor. Math. Phys. 1 (1998) 1, hep-th/9706082.
[17] E. Witten, Nucl. Phys. B 507 (1997) 658, hep-th/9706109.
[18] A. Brandhuber, N. Itzhaki, V. Kaplunovsky, J. Sonnenschein, S. Yankielowicz, Phys. Lett. B 410 (1997) 27,
hep-th/9706127.
[19] I.R. Klebanov, Phys. Lett. B 466 (1999) 166, hep-th/9906220.
[20] N.M. Davies, T.J. Hollowood, V.V. Khoze, M.P. Mattis, Nucl. Phys. B 559 (1999) 123, hep-th/9905015.
[21] S.R. Coleman, J. Mandula, Phys. Rev. 159 (1967) 1251.

The exact superconformal R-symmetry maximizes a

Ken Intriligator, Brian Wecht
Department of Physics, University of California, San Diego, La Jolla, CA 92093-0354, USA
Received 2 May 2003; accepted 21 May 2003
Abstract
An exact and general solution is presented for a previously open problem. We show that the
superconformal R-symmetry of any 4d SCFT is exactly and uniquely determined by a maximization
principle: it is the R-symmetry, among all possibilities, which (locally) maximizes the combination
of t Hooft anomalies atrial (R) (9 Tr R 3 3 Tr R)/32. The maximal value of atrial is then, by
a result of Anselmi et al. the central charge a of the SCFT. Our atrial maximization principle
almost immediately ensures that the central charge a decreases upon any RG flow, since relevant
deformations force atrial to be maximized over a subset of the previously possible R-symmetries.
Using atrial maximization, we find the exact superconformal R-symmetry (and thus the exact
anomalous dimensions of all chiral operators) in a variety of previously mysterious 4d N = 1 SCFTs.
As a check, we verify that our exact results reproduce the perturbative anomalous dimensions in all
perturbatively accessible RG fixed points. Our result implies that N = 1 SCFTs are algebraic: the
exact scaling dimensions of all chiral primary operators, and the central charges a and c, are always
algebraic numbers.
PACS: 11.25.Hf; 11.40.-q; 11.30.Pb
1. Introduction
The 4d N = 1 superconformal algebra is SU(2, 2|1), whose bosonic part is SO(4, 2)
U (1)R . Thus every N = 1 superconformal field theory (SCFT) must have a conserved
U (1)R symmetry, whose current is in the same superconformal multiplet as the stressenergy tensor. There might be additional global flavor symmetries F ; the full symmetry
group of the N = 1 SCFT is then SU(2, 2|1) F . The additional global symmetry F acts
as a non-R-symmetry (i.e., the supercharges are invariant). For example, N = 1 SQCD
E-mail addresses: keni@bohr.ucsd.edu (K. Intriligator), bwecht@physics.ucsd.edu (B. Wecht).
doi:10.1016/S0550-3213(03)00459-0
184
K. Intriligator, B. Wecht / Nuclear Physics B 667 (2003) 183200
is believed to flow to an interacting SCFT for Nf in the range 3Nc > Nf > 32 Nc [1], and
the additional global symmetry of the SCFT is F = SU(Nf ) SU(Nf ) U (1)B .
The U (1)R symmetry residing in SU(2, 2|1) yields important and exact results for
SCFTs. For example, all operators O have dimension (O) which satisfy
3
3
(O) R(O) and (O) R(O),
(1.1)
2
2
with R(O) the U (1)R charge of O. The first inequality is saturated for pure chiral primary
operators, and the second for pure antichiral operators. Since the R charge of composite
operators is simply additive, so are the anomalous dimensions of composite chiral primary
operators; this is the statement that they form a chiral ring, with non-singular mutual OPE.
The condition that the U (1)R global symmetry be free of ABJ type anomalies, i.e., that
R-charge conservation must not be violated in any gauge field instanton backgrounds, is
precisely the condition that the NSVZ exact beta functions [2] vanish for all gauge groups.
Another remarkable utility of the superconformal U (1)R symmetry was found by
Anselmi et al. [3,4]: the U (1)R t Hooft anomalies completely determine the a and c central
charges of the superconformal field theory:

3
1
3 Tr R 3 Tr R ,
9 Tr R 3 5 Tr R .
(1.2)
c=
32
32
Because of t Hooft anomaly matching, this means that these central charges can
be computed simply in terms of the weakly coupled UV spectrum, even for highly
interacting IR fixed points. It is believed that the central charge a obeys the 4d analog
of Zamolodchikovs c-theorem [5]: under any renormalization group flow, perturbing
away from any UV fixed point and flowing to a new IR fixed point reduces the central
charge: aIR < aUV . This was verified to be the case in many supersymmetric examples by
using (1.2), as in [3,4]. See [6] for some more recent developments.
So the superconformal R-symmetry is extremely useful. . . provided that it can be found!
The symmetry constraints generally do not uniquely determine U (1)R whenever F is nontrivial. This is because if R0 is some valid U (1)R symmetry, then so is

Rt = R0 +
(1.3)
sI FI ,
a=
where FI are all of the non-R flavor charges in the global symmetry group F 1 and sI
are arbitrary real parameters. The superconformal U (1)R SU(2, 2|1) corresponds then
to some particular choice sI of these parameters, and we need something beyond just
symmetry considerations to determine what these are.
There are some physical requirements which sometimes help. For example, the
superconformal U (1)R should commute with the non-Abelian flavor symmetries, so
we can restrict the linear combination in (1.3) only to those Abelian flavor generators
1 We emphasize that all currents in (1.3) are bona fide symmetries, which are anomaly free and respected by
all superpotential terms. In particular, we are not considering the situation discussed in, e.g., [3,4] of the RG flow
of the R current in the stress tensor supermultiplet between its weak coupling expression and that of the IR fixed
point SCFT. We are not considering RG flows here, only aspects of the interacting SCFT RG fixed points.
185
which commute with all non-Abelian elements of F . For the example of SQCD, this
uniquely fixes U (1)R once we put in the additional physical requirement that baryons and
antibaryons have the same superconformal R-charge.
Another physical requirement is unitarity, which requires that all gauge invariant chiral
operators must have U (1)R charge R 2/3, with R = 2/3 iff it is a free field. But this
condition actually does not constrain the superconformal R-charge assignments of the
fields in the Lagrangian. Indeed, there are examples where the R-charge assignments of
the fields are such that some gauge invariant operators appear to violate this unitarity
condition. When the R 2/3 condition appears to be violated, it actually just means that
the corresponding operator is actually a free field, and there is then some accidental extra
symmetry which mixes with U (1)R to make R = 2/3 for that operator (with the R-charge
of the other gauge invariant operators unaffected).
In general, however, the superconformal R-symmetry is not uniquely determined on
symmetry grounds or the above considerations. One well-known example is N = 1 SQCD
with an added adjoint and zero superpotential. Because no other condition to determine the
superconformal R-symmetry had been previously known (as far as we are aware), it had
not been possible to apply the above powerful constraints of superconformal invariance to
generic N = 1 SCFTs.
We will here present and explore a simple prescription for uniquely and exactly
determining the exact superconformal U (1)R for any 4d SCFT. The idea is to parametrize
the most general possible R-symmetry as in (1.3). The subscript t is for trial. The
superconformal U (1)R SU(2, 2|1) then corresponds to some particular values, sI ,

sI FI ,
R = R0 +
(1.4)
I
our goal then is to determine the values of the sI . What we show is that the sI can be
uniquely determined by imposing the following conditions on the t Hooft anomalies:

9 Tr R 2 FI = Tr FI ,
(1.5)
where R is the U (1)R SU(2, 2|1) and FI are all flavor charges in F , and also
Tr RFI FK < 0.
(1.6)
Specifically, this matrix has all negative eigenvalues for all flavor symmetries. We will
prove (1.5) and (1.6) using general properties of 4d N = 1 SCFTs. Plugging (1.4) into (1.5)
leads to a quadratic equation for each of the sI . Then (1.6) uniquely determines which are
the correct roots of the quadratic equations.
The identities (1.5) and (1.6) have a nice interpretation. Introduce

3
3 Tr Rt3 Tr Rt ,
(1.7)
32
for the general trial Rt symmetry (1.3) as a function of the parameters sI . When the sI are
the special values sI , where Rt becomes the U (1)R SU(2, 2|1), the value of atrial is the
central charge a of the SCFT, as in (1.2). The conditions (1.5) and (1.6) can be equivalently
stated: the U (1)R SU(2, 2|1) is precisely that which (locally) maximizes atrial(s)! This
atrial(s)
186
is simply because

a(s)
3
=
9 Tr Rt2 FI Tr FI
sI
32
and
2 a(s)
27
=
Tr Rt FI FK .
sI sK
16
(1.8)
So (1.5) and (1.6) together imply that, among all possible R-symmetries, the U (1)R
SU(2, 2|1) is that which (locally) maximizes a.
As an example of how atrial maximization determines the superconformal U (1)R ,
consider the case of a free theory of |G| vector multiplets and M chiral multiplets i ,
i = 1 M, with trial charges Rt (i ) = ri . We then have

3
3
atrial =
(1.9)
3(ri 1) (ri 1) .
2|G| +
32
i
If we now extremize with respect to the ri , we get

1
(ri 1)2 = ,
(1.10)
9
with ri = 2/3 the root which is a local maximum and ri = 4/3 the root which is a local
minimum. Our general identities (1.5) and (1.6) imply that the correct superconformal
R-symmetry in this case is to take ri = 2/3 as the charge of all chiral superfields. This is
indeed the correct result for the free theory and, at this local maximum,
1
3
|G| + M.
(1.11)
16
48
Note that, because atrial is a cubic function of the ri , there is no global maximum or
minimum: atrial if we take ri .
Several comments:
a = afree =
(1) If a flavor symmetry FI has vanishing t Hooft anomaly, Tr FI = 0, then the

condition (1.5) becomes Tr R 2 FI = 0. For non-Abelian flavor symmetries (which satisfy
Tr FI = 0 simply by tracelessness of the generators) and for baryon number, the solution
of Tr R 2 FI = 0 is to take R to commute with FI . This is as expected on physical grounds.
The condition (1.6) then generally imposes a non-trivial condition on SCFTs: since the
non-Abelian flavor generators have positive Tr FI FJ , (1.6) implies that the U (1)R charge
of all matter fields transforming in non-Abelian flavor representations must satisfy R < 1,
so the fermion component has negative R-charge.
(2) The atrial maximization principle almost immediately ensures the a theorem: aIR <
aUV for any RG flow between UV and IR fixed points. The reason is that generally
FIR FUV , since the relevant deformations of the UV theory break some of the flavor
symmetries.2 Since at the IR fixed point atrial is maximized over a subspace of the
parameter space of UV fixed point, the maximal value will be smaller, showing that
aIR < aUV . The almost is because of two potential inadequacies in this argument. The
first is that sometimes there are additional, accidental flavor symmetries of the IR fixed
2 This also applies to Higgsings, if we interpret F as also containing the global component of the gauge group,
which mixes with the flavor symmetries upon Higgsing. The previous comment ensures that R commutes with
the global component of the gauge group too.
187
point, so sometimes FIR is not a subset of FUV . The second caveat is that, since the atrial
maximum is only a local maximum, it is possible for the maximal value on a restricted
subspace to actually exceed that of the larger parameter space, e.g., one could get a larger
value than (1.11) if some added interaction restricts some of the R-charges to a subspace
where they are sufficiently large. So the a theorem requires that those relevant deformations
which can drive the theory to a new RG fixed point cannot possibly restrict the possible
R charges to a subspace where some are extremely large. This fits with the constraint
mentioned in comment (1), and with the intuition that interactions generally reduce the
dimensions of chiral primary operators.
(3) The superconformal U (1)R satisfying (1.5) and (1.6) must give a real R-charge to
all chiral fields.
(4) Our approach relies on being able to identify the full symmetry group of the RG
fixed point, e.g., via analyzing the UV Lagrangian away from the RG fixed point. But
strongly interacting RG fixed points can also have enhanced symmetries, which are not
visible in any weakly coupled Lagrangian description. In particular, the superconformal
U (1)R of a SCFT could be such a symmetry. This is the case, for example, in the N = 1
and N = 2 SCFTs presented in [7,8]. Though our t Hooft anomaly identities should be
applicable also at such RG fixed points, it remains to be seen whether or not they can be
used to determine the superconformal U (1)R in such cases.
(5) The superconformal U (1)R charges determined by our procedure outlined above
will always be algebraic numbers, i.e., rationals and roots of rationals. This is because
they are found by solving quadratic equations with rational coefficients (which are the
t Hooft anomalies). Thus, for any SCFT, the exact anomalous dimensions of all chiral
primary operators, and the exact central charges a and c, will always be algebraic numbers.
(SCFTs of the type discussed in the previous comment could be exceptions to this general
statement, though the known examples of this type actually have rational R-charges.)
The outline of this paper is as follows. In Section 2 we will discuss some general
aspects of currents, anomalies, and supersymmetry. After presenting some background
material, we will argue for the result (1.5) by using a result due to Osborn [9]: the 3-point
function of two stress tensor supermultiplets and one flavor current supermultiplet is of a
form completely determined by the superconformal symmetry with only a single overall
multiplicative coefficient to be determined. The condition (1.6) will also be obtained,
by relating the t Hooft anomalies to currentcurrent correlators and using unitarity,
as in [4]. In Section 3 we verify, in complete generality, that our atrial maximization
precisely reproduces the known, leading order, anomalous dimensions in all perturbatively
accessible RG fixed points of the type conjectured by [10,11].
In Section 4 we use atrial maximization to obtain the exact R-charge for some previously
mysterious examples. In particular, we consider in complete generality theories with two
different matter field representations and zero superpotential. A special case of this is
SU(N) with an adjoint chiral superfield and Nf fundamental flavors, with W = 0. This
theory was argued to have a non-trivial RG fixed point in [12], and was further explored
in [13], but the exact R-charges could not be determined before the present paper. We also
discuss some chiral quiver N = 1 SCFT examples, both with and without superpotential,
which gives another check of the a-theorem.
188
2. Currents, anomalies, and supersymmetry

In this section we argue for the result (1.5) by showing that a result due to Osborn [9]
implies that Tr R 2 FI and Tr FI are necessarily proportional to each other. As we also
discuss, following [4], the Tr RFI FJ t Hooft anomalies are proportional to the JI JJ
current correlator, whose sign is constrained by unitarity. The reader need not get too
bogged down with the numerical coefficients in the following section, since we only need
to establish general proportionality relations. The constants of proportionality can then
always be determined by considering the particular case of a free field theory.
2.1. Review of currents and anomalies
Let us first review some basics of currents and anomalies in theories which are not
necessarily supersymmetric. We will call the gauge group G and suppose, for illustrative
purposes, that the flavor group is a product U (1)1 U (1)2 , with left-handed chiral currents
JI for I = 1, 2. (The generalization to non-Abelian flavor symmetries is straightforward,
with the most interesting aspects for our discussions in the U (1) factors anyway.)
The currents JI must not have ABJ anomalies, i.e., the triangle diagram with a current
insertion at one vertex and G gauge fields at the other two must vanish. Suppose that there
are ni chiral fermions i , all in G representation ri , with U (1)I flavor charge qiI , which
all run in the loop. The vanishing ABJ anomaly condition is

(2.1)
qiI ni (ri ) = 0,
i
where the (ri ) are the quadratic Casimirs from the coupling to the two G gauge fields,

Trri TGA TGB = (ri ) AB ,
(2.2)
2h (Ad),
and with TGA the G generators in representation ri . The anomaly free condition (2.1) can
equivalently be stated as the condition that the fermion zero modes of the G instantons
t Hooft vertex must be flavor neutral. The G instanton has ni (ri ) of the ,i fermion
zero modes. (Our normalization is, e.g., () = 1 for SU(N).)
We now consider current correlators. In any conformal field theory, the current twopoint functions are completely determined by conformal invariance, up to the overall
coefficient:

I
1

IK
J (x)JK (0) =
(2.3)
.

16 4
x4
In any unitary theory, I K should be a matrix with all positive definite eigenvalues.
The form of the current 3-point functions are also highly constrained. The aspect of
interest to us here is the anomalous violation of the current conservation in contact terms.
For example,

I

kI I I
J (x)JI (y)JI (z) =
. (x z)(y z),
z
x y
48 2
(2.4)
where the coefficient kI I I is the Tr U (1)3I t Hooft anomaly

3
kI I I =
ni |ri | qiI .
189
(2.5)
Here |ri | is the dimension of the representation ri .

Similarly, there are current conservation violating contact terms in current correlators
involving mixtures of the two currents, such as JL (x)JL (y)JI (z) with I = L, which has
an anomalous contact term proportional to the mixed t Hooft anomaly

2
kLLI Tr U (1)2L U (1)I =
(2.6)
ni |ri |qiI qiL .
i
Another anomalous contact term, violating current conservation, occurs in the 3-point
functions involving the current at one vertex of the triangle diagram and stress-energy
tensors at the other two, e.g., T (x)T (y)J I (z), which has anomalous contact terms
proportional to the t Hooft anomaly

ni |ri |qiI .
kI Tr U (1)I =
(2.7)
i
The contact terms in the above mentioned three point functions can be conveniently
expressed in terms of a lack of current conservation when the currents JI are coupled
to general background gauge fields AI , and the stress tensor T is coupled to a general
metric g . For example, we have
kI I I
I
I + kI I L FI F
L + kLLI FL F
L + kI R R,
J (z) =
FI F
2
2
2
z
48
16
16
384 2
(2.8)
= 1 . F F , and R R
= 1 . R R . The first term comes

where L = I , F F
2
2
from the descent formalism on 3!1 (F /2i)3 . On the other hand, the third term comes
directly from the index theorem, as in the ABJ anomaly, involving 2!1 (F /2i)2 ; this is
one way to understand the relative factor of three between these two terms (one can also
see it directly from the symmetry factors with the two corresponding triangle diagrams).
The last term in (2.8) is the Pontrjagin density; it can be written in terms of the Weyl
tensor (the Riemann tensor minus all non-zero contractions of indices), so it vanishes in
any conformally flat background.
2.2. Supersymmetric theories
Consider a general 4d N = 1 supersymmetric gauge theory, with gauge group G, which
we will take to be simple to streamline the present discussion (the generalization to product
gauge groups is easy). The theory has chiral
superfields in the G representation (which,
of course, must be gauge anomaly free): si=1 ni ri . If there is no added superpotential, the
full symmetry group of anomaly free global symmetries is
U (1)R U (1)
s1
s

i=1
SU(ns ).
(2.9)
190
(E.g., for SQCD we have s = 2, with r1 = , r2 = , and n1 = n2 = Nf .)

In the flavor group (2.9), we have already eliminated one U (1) classical global
symmetry by the vanishing ABJ anomaly condition (2.1). The chiral U (1)R symmetry
in (2.9) assigns charge 1 to the gauginos, charge Ri to the scalar components of the chiral
superfields i , and charge Ri 1 to the fermion components of the chiral superfields; thus
the general condition (2.1) of vanishing ABJ anomaly becomes

ni (Ri 1)(ri ) = 0.
2h +
(2.10)
i
The remaining flavor symmetries in (2.9) are not R-symmetries, so the gauginos are neutral
and all components of the chiral superfield i , in representation ri , carry the same charge
qi . According to (2.1), these must satisfy

ni qi (ri ) = 0.
(2.11)
i
The non-Abelian part of the currents in (2.9) is always anomaly free since their generators
are traceless, so (2.11) only constrains the overall U (1) flavor symmetries in (2.9), which
commute with the non-Abelian flavor symmetries. Any superpotential terms will further
constrain the above flavor symmetries and charges, which could be easily incorporated
into this discussion.
The anomalous contact terms in current three point functions or equivalently the lack of
current conservation when the global symmetries are coupled to non-trivial backgrounds
are as described in the previous subsection, with the t Hooft anomalies

ni (Ri 1)3 |ri |,
Tr R 3 kRRR = |G| +
Tr R kR = |G| +
ni (Ri 1)|ri |,
Tr R 2 F kRRF =
ni (Ri 1)2 qi |ri |,
Tr F kF F F =
3
Tr F kF =
ni qi3 |ri |,
ni qi |ri |,
Tr RF 2 kRF F =
ni (Ri 1)qi2|ri |.
(2.12)
G is the dimension of the gauge group, and |ri | is that of ri .

The flavor currents and their anomalies in general backgrounds can be expressed in
whose = 0
terms of current superfields. One is the superstress tensor T (x, , ),
component is the superconformal U (1)R symmetry, components linear in and are the
supersymmetry currents, and terms quadratic in the and are the stress-energy tensor.
The other, non-R, flavor currents reside in current superfields

1
I , with current component J = , JI . (2.13)
JI (x, ) = T
I

=0
4
191
are the chiral and antichiral matter fields and TI is the appropriate flavor
Here and
generator, labeled by I . The superstress tensor T couples to the metric in terms of the
metric superfield H , while the non-R current superfields (2.13) couple to background
superfield vector multiplets VI . See [14,15] for background material and references.
The superstress tensors anomaly can be expressed in the general form [14,15]
T = LT ,
(2.14)
where LT is the trace anomaly, which can be written in terms of the variation of the action
with respect to the chiral compensator superfield of supergravity. For example, in a theory
with non-vanishing beta function we would have LT 1 tr Wg2 , giving T 1 Tr Fg2 ,

with Fg the gauge field strength and Wg its chiral superfield. Were interested in conformal
field theories, so = 0 and LT = 0 when in flat space and trivial background gauge fields.
When coupled to non-trivial backgrounds, however, we have
c
a
1
W2
I K Tr WI WK .
LT =
(2.15)
2
2
24
24
96 2
IK
The coefficients c and a are the central charges, W 2 12 W W is the square of the
is the chirally projected super2 + R)(G2 + 2R R)
super-Weyl tensor, and W 2 + (
Euler density; see Appendix A of [3] for a discussion of these terms. Taking components of
the first two terms in (2.15) is how [3] obtained the relations (1.2). The remaining terms in
(2.15) are the contributions to the scaling anomaly proportional to the superfield strengths
WI of the background gauge fields VI , which couple to the flavor supercurrents JI . The
coefficients I K are the same as those in (2.3), following the discussion in [4].
2.3. The flavor superanomaly
We now turn to the anomaly of the non-R flavor currents JI in general backgrounds. The
2 J , so the super-anomaly can
divergence of the current in (2.13) is proportional to D 2 J D
2

be written as a non-zero chiral contribution to D J . We were not able to find a discussion
of this in the literature or supersymmetry textbooks so, as far as we are aware, our result
2 J is a chiral
2 J in a general background is new. It is clear how to proceed: since D
of D
object, its non-zero anomaly must involve chiral field strengths of the background fields.
Further, the current anomaly component of the result must reproduce the terms in (2.8).
The result is

kI I L
kI
kI LL
2 JI = kI I I WI2 +
+
D
(2.16)
W
W
+
W
W
W 2.
I
L
L
L
2
2
2
48
16
16
384 2
L=I

2 + R in (2.16). We will continue
2
2
(In a curved background, we replace [14] D
2
; the implicit curvature term does not contribute to our final result.)
to just call this D
The crucial aspect of (2.16) for our purposes is that the gravitational contribution only
involves W 2 , with no additional term proportional to the chiral Euler/Pontrjagin density .
This differs from the superstress tensor anomaly (2.15), which involves both chiral field
strengths. To motivate the absence of an additional term proportional to in (2.16), we
note that has a component which is the Euler density, which is non-vanishing even
192
for a conformally flat metric. On the other hand, W 2 does vanish in a conformally flat
background. The absence of a term proportional to in (2.16) is related to the fact that
supercurrents should be conserved in a conformally flat background, since there is no
unavoidable charge violation by particle production in conformally flat backgrounds. For
example, in conformally flat AdS space, with unbroken supersymmetry, the J component
of J is conserved because the Pontrjagin density vanishes. Supersymmetry then implies
2 J identically vanishes.
that the holomorphic quantity D
The gravitational part of (2.16) can equivalently be described in terms of an anomalous
contribution to the current superfield 3-point function

2 T (1)T (2)JI (3) = contact terms,
D
(2.17)
3
where (1) etc., label the superspace coordinates at the three points. The fact that there is
only one independent gravitational term in (2.16), as opposed to the two terms in (2.15), is
related to a result of Osborn [9] for the relevant 3-point function:

T (1)T (2)JI (3) = KI I ,
(2.18)
(1, 2, 3).
Here I ,
(1, 2, 3) is a completely determined function on superspace [9], and the only
2 of (2.18) gives the
dependence on the theory is in a single overall coefficient KI . Taking D
3
3
contact terms indicated in (2.17). This would then lead to (2.16), with the single overall
normalization constant KI , needed to fix the three-point function (2.18), thus proportional
to the Tr U (1)I t Hooft anomaly kI .
In contrast, it was also shown in [9] that the T T T 3-point function depends on two
overall coefficients, which is related to the fact that both W 2 and appear in (2.15).
The fact that the 3-point function involving two stress tensors and one flavor current
T (1)T (2)JI, (3) is completely fixed up to a single overall normalization coefficient
also holds in non-supersymmetric conformal field theories, as shown by [16]. There, too,
we can say that the single undetermined overall coefficient must be proportional to the
Tr U (1)I t Hooft anomaly.

2.4. Why a is maximized for the correct superconformal R-symmetry
In this section we prove (1.5) and (1.6), from which atrial maximization follows.
Consider (2.16), with the background gauge fields WI set to zero and only the non-trivial
backgrounds those in W 2 , namely, the metric and the U (1)R SU(2, 2|1) background
gauge field strength, which we denote by FR . The divergence of the U (1)I flavor current is

i
ikI 2 2 2 2
2 JI
=
W
JI = 2 ,
W

4
384 2
=0
=0

8
kI
+ FR F
R .
RR
=
(2.19)
384 2
3
3 Rather than directly showing that the contact terms indeed lead to (2.16), we note that, because the function
2
2
I,
(1, 2, 3) in (2.18) is completely determined, the possible gravitational contributions to D J , namely, W
and , must appear with a fixed ratio. Then a free-field calculation would suffice to show that the coefficient of
is actually zero, as in (2.16).
193
Comparing with (2.8), the first term is the expected result involving the anomaly
proportional to the kI = Tr U (1)I t Hooft anomaly multiplied by the Pontrjagin density.
The second term in (2.19) is the anomaly which, as in (2.8), should be proportional to
the kI RR = Tr U (1)I U (1)2R t Hooft anomaly, but we see in (2.19) that it is instead also
proportional to kI . Taking into account the coefficients in (2.19), as compared with (2.8),
we have thus shown the result (1.5). As a check of Eqs. (2.16) and (1.5), we note that (2.18)
suffices to show that the ratio kI /kI RR is some fixed number for all currents, independent
of the theory; so it can be evaluated for the case of free fields, where the matter fermions
have R = 1/3, showing that the ratio is 9. We can write the result (1.5) as

ni |ri |qi 9(Ri 1)2 1 = 0.
(2.20)
i
To prove (1.6), suppose that we take trivial metric and FR backgrounds but turn on
background gauge fields coupled to the currents JI . Then (2.14) and (2.15) imply, as in [4],
that the U (1)R symmetry has a divergence given by
I J
J , and hence kRI J Tr RFI FJ = I J ,

R =
(2.21)
FI F
48 2
3
where I J are the same coefficients as in (2.3). Since I J must be a matrix with all positive
definite eigenvalues in any unitary theory, we see that supersymmetry, along with unitarity,
requires Tr RFI FJ to be negative, as in (1.6). It is easy to see this in, for example, a free
field theory, where the R-charge of any matter fermion field is 1/3.
3. A check: comparing with perturbative RG fixed points

Let us write the beta function as:
(g) = 1 g 3 + 2 g 5 + .
(3.1)
As pointed out in [10,11], there can be a RG fixed point in the perturbative regime if the
one loop beta function is negative and the two loop beta function positive, so 1 and 2
in (3.1) are both positive, and the coupling g2 1 /2 where they cancel is sufficiently
small. The expectation is that, in this case, higher order corrections might shift the fixed
point value g a bit, but not wipe out the qualitative feature of a RG fixed point. This is
the case when 1 is very small, so the theory is just barely asymptotically free. In terms
of a large Nc expansion, where 1 is order Nc and 2 is order Nc2 (corresponding to the
expansion in t Hooft coupling g 2 Nc ) we get a small 1 by adding matter flavors with total
quadratic index T proportional to Nc , to make 1 order Nc0 . The 2 is still order Nc2 , so
we get g2 Nc 1/Nc : the t Hooft coupling is parametrically small, so the existence of the
RG fixed point is on fairly solid ground.
We will be completely general,
letting the gauge group be G and there be matter chiral
superfields in representations i ni ri . To be in the perturbative RG fixed point regime, we
want to have
1
1 = 3h T h., with 0 < . 1,
(3.2)
2
194

and T i ni (ri ) the total of the matter fields quadratic Casimirs (2.2). We first
compute the superconformal U (1)R symmetry via atrial maximization, and then compare
the result to a direct perturbative computation of the anomalous dimensions at the RG fixed
point.
We assign the fields R-charge Ri subject to the vanishing anomaly constraint (2.10); to
order . we do this as

2
2
(1)
(1)
Ri = + Ri . + O . 2 , where
(3.3)
ni (ri )Ri = h.
3
3
i
We now compute
9 Tr R 2 F Tr F = 6.

(1)
Ri ni |ri |qi + O . 2 ;
(3.4)
this must vanish for the superconformal U (1)R . This vanishing must hold for all possible
choices of flavor charges qi which satisfy the anomaly free condition (2.11). This requires
that Ri(1) = (ri )/|ri |, and we can fix the overall normalization via (3.3) to obtain

1

2h
(ri )
2
2
1
+ O .2 .
Ri = .
nj (rj ) |rj |
3
3
|ri |
(3.5)
Of course, we could just as well obtain the exact answer via atrial maximization, but the
result will be complicated, and we are only interested here in comparing the order . term
to a perturbative computation.
We now check that the result (3.5) agrees with an explicit perturbative computation of
the anomalous dimension, using

2
2
1
Ri = i =
(3.6)
1 + i (g ) ,
3
3
2
with i (g ) the anomalous dimensions, evaluated at the RG fixed point coupling g where
(g ) = 0. Working to one-loop, which corresponds to order ., we have
i (g) =

g 2 |G|(ri )
+ O g4 .
2
8
|ri |
(3.7)
This can be seen by considering the matter field propagator, with a single gluon (plus
gluino) loop. The group theory factor comes from the sum over gluons coupling to the
matter field in representation ri :
|G |

a=1
Trai Trai =
|G|(ri )
1|ri ||ri | ,
|ri |
(3.8)
with Trai the G generators in representation ri , with a the adjoint index. The group theory
factors in (3.8) are seen by comparing the trace of (3.8) with (2.2).
The RG fixed point coupling g is determined by requiring that the NSVZ beta function
vanish, which is equivalent to the condition that the U (1)R symmetry (3.6) satisfy the
anomaly free condition (2.10), i.e.,

g2
1
ni (ri ) 2 |G|(ri )|ri |1 + O g4 = 0,
2h +
3 16
195
(3.9)
which gives

1
g2 |G|
2
1
= (6h T )
ni (rj ) |rj |
.
8 2
(3.10)
Recalling that 6h T 2h., we see that (3.6), together with (3.7) and (3.10) indeed
agree with (3.5). It would be interesting to compare higher orders in ., where our exact
results make predictions about the higher loop anomalous dimensions.
4. Some previously mysterious examples

4.1. General case where U (1)R mixes with a single flavor U (1)
In this section, we will explicitly determine the superconformal U (1)R for cases where
there is a single flavor current J which can mix non-trivially with U (1)R . This is the
case, for example, for theories with two types of representations (e.g., SU(N) with Nf
fundamental flavors and an adjoint chiral superfield) and W = 0.
a maximization leads to the U (1)R SU(2, 2|1) as given by
R = R0 + s J,
(4.1)
where using (1.5) we have that s is determined by

1
(4.2)
Tr J = 0.
9
This quadratic equation can be solved, with the correct root for maximizing a, i.e.,
satisfying (1.6), given by

Tr R0 J 2 (Tr R0 J 2 )2 Tr J 3 (Tr R02 J 19 Tr J )
s =
(4.3)
.
Tr J 3
must be positive.
Of course s must be real, so the quantity in the
Let us apply this to a general theory with two kinds of matter representations and
W = 0. Consider a theory with gauge group G, n1 matter fields in representation r1 and
n2 matter fields in representation r2 . We can take our initial R0 to be one under which all
chiral superfields have the same charge, with value determined from (2.10), and choose our
= R0 + s J is
current J to be one satisfying (2.11). Then R
s2 Tr J 3 + 2s Tr R0 J 2 + Tr R02 J
s
1 ) = 1 2h +
,
R(r
T
n1 (r1 )
s
2 ) = 1 2h
R(r
,
T
n2 (r2 )
(4.4)
196
where T
ni (ri ). The value of s is found by just plugging into (4.3) with

|r1 |
2h
|r2 |
2
Tr R0 J =
+
,
T n1 (r1 )2 n2 (r2 )2
2

1
1
4h
|r1 |
|r2 |
,
Tr R02 J Tr J =
9
9
(r1 ) (r2 )
2T
|r1 |
|r2 |
.
Tr J 3 = 2
n1 (r1 )3 n22 (r2 )3
(4.5)
Asymptotic freedom, T < 6h, is sufficient to ensure that s , as given by (4.3), is indeed
real.
This general discussion can be applied, e.g., to the case of SU(Nc ) with Nf fundamental
flavors and an adjoint chiral superfield. We just plug in n1 = 2Nf (counting fundamentals
and antifundamentals together, since they have the same R-charge, as in comment (1) in
the introduction), |r1 | = Nc , (r1 ) = 1, n2 = 1, |r2 | = Nc2 1, (r2 ) = 2Nc . This yields
=
R(Q) = R(Q)
Nf
s
+
,
Nc + Nf
2Nf
R() =
Nf
s
,
Nc + Nf
2Nc
where s is determined by (4.3) with

Nc
Nc
Nc2 1
2
,
Tr R0 J =
+
Nc + Nf 2Nf
4Nc2

2

Nc2
Nc + 1
1
1
Tr R02 J Tr J =
,
9
2Nc
(Nc + Nf )2 9
Tr J 3 =
Nc
4Nf2
Nc2 1
.
8Nc3
As an example, we consider the case

of Nc 1, with fixed Nf /Nc . 1. We then get,
to leading order in ., s 2Nf (3 5)/3, which gives
5 Nf
3 5,
R(Q) = R(Q)
(4.6)
R()
.
3
3Nc
These are positive, and R(Q) satisfies the R < 1 constraint needed for fields transforming
non-trivially under non-Abelian flavor symmetries, as mentioned in comment (1) of the
and Tr 2 which apparently

introduction. There are gauge invariant operators, such as QQ
violate the R 2/3 constraint but, as mentioned in the introduction, this just means
that these particular operators are actually free, and their R-charge gets corrected to 2/3
by additional accidental symmetries. (A similar situation occurs when these theories are
deformed by a Tr k+1 superpotential [13].) Note that (4.6) implies that the Tr k+1
superpotential
of [13] is a relevant deformation of the W = 0 RG fixed point for k + 1 <
6Nc / 5 Nf in the above Nc 1, Nf Nc limit. This fits with the qualitative discussion
of [13] on how this superpotential can affect the IR physics, even when k 2, despite the
is a
fact that it naively appears to be irrelevant. Also, as expected, we find that Tr QQ
relevant superpotential deformation of the W = 0 fixed point.
197
4.2. A quiver example and connection to AdS/CFT

As an interesting example of a theory with more than one flavor current, we consider the
theory given by the quiver in Fig. 1 with gauge group U (N)4 . This quiver arises naturally in
string theory as the IR worldvolume theory of N coincident D3-branes placed at the tip of a
non-compact CalabiYau X6 , where X6 is a complex cone over the first del Pezzo surface
dP1 . This string theory construction leads to a non-zero superpotential, and the large N
limit of that theory is an N = 1 SCFT which is dual to IIB string theory on AdS5 H5
with H5 the horizon of the complex cone over dP 1 . This H5 is a U (1) fibration over dP 1 .
We will discuss this SCFT, for all N , shortly.
First, let us consider the theory with the quiver diagram of Fig. 1 and with no
superpotential, W = 0. This theory has no known string theory construction and is hence
more mysterious than the theory with non-zero W . We expect that the W = 0 theory also
flows to an interacting N = 1 SCFT in the IR, and we use the atrial maximization to
determine the exact U (1)R charges and hence the exact anomalous dimensions and central
charges at this new RG fixed point.
The IR SCFT has symmetry group SU(2, 2|1) F , with flavor group F = U (1)1
U (1)2 SU(2) SU(2) SU(3). The non-Abelian symmetries rotate the multiple bifundamental flavors and classically there is an additional flavor U (1)6 , one for each of
the six legs of the quiver. Enforcing the anomaly free condition for each of the four
gauge groups reduces this U (1)6 down to U (1)2 . As discussed in the introduction, atrial
maximization implies that only the U (1)2 can mix non-trivially with U (1)R . Without loss
of generality, we can assign U (1) charges as:
X21
X14
X43
X32
X13
X42
J1
J2
R0
2/3 1/3 3/4
1
0
1/2
2
1
1/4
1
0
1/2
0
1
3/4
0
1
3/4.
Fig. 1. The U (N )4 quiver diagram.
(4.7)
198
These charges form a basis of the solutions of the appropriate anomaly-free conditions
at each node. (Of course, any non-degenerate linear combinations of these would be an
equally valid basis choice.) For completeness we include all relevant t Hooft anomalies
here (omitting an overall factor of N 2 for each): Tr R0 = 0, Tr R03 = 3, Tr J1 = Tr J2 = 0,
Tr J13 = 44/9, Tr J23 = 8/9, Tr J12 J2 = 40/9, Tr J1 J22 = 20/9, Tr R02 J1 = 1/4,
Tr R02 J2 = 1/2, Tr R0 J12 = 16/3, Tr R0 J22 = 4/3, Tr R0 J1 J2 = 5/3.
We write the exact superconformal U (1)R symmetry as R = R0 + s1 J1 + s2 J2 , and
determine the values of sI by imposing 9 Tr R 2 JI = Tr JI , for I = 1, 2, and use (1.6)to
determine the correct roots of the resulting quadratic equations; the result is s1 = 2+6 5
and s2 =
7+2 5
.
12
We thus obtain for the exact R-charges
7 5
1+ 5
,
R(X14 ) = R(X32 ) =
,
R(X21 ) =
6
6
3 5
1+ 5
,
R(X13 ) = R(X42 ) =
.
R(X43 ) =
(4.8)
2
6
The R charges of all gauge invariant chiral operators, such as X21 X14 X42 , satisfy the
condition R > 2/3.
The value of the central charge a, found by plugging (4.8) into (1.2) is
3+5 5 2
N 0.886 N 2.
a=
(4.9)
16
One final comment on our proposed RG fixed point for the W = 0 theory. Some of the
nodes of the quiver have Nf = 3Nc , and are thus not asymptotically free (the one-loop
beta function vanishes, and the two loop beta function is positive). Their gauge couplings
would flow to zero in the IR, in the absence of other interactions. Our claim is that, upon
including the gauge interactions of the other nodes, the theory flows to an IR fixed point
SCFT, where all gauge groups are interacting. The NSVZ beta function for each gauge
group vanishes, since (4.8) is anomaly free. We have not proven that the fixed point exists,
since we have not proven that the dynamics can actually realize the R-charges (4.8), but
the various consistency checks give us confidence that the fixed point SCFT exists.
Let us now consider the theory with the non-trivial superpotential, as obtained via D3branes at the singularity of a local CY which is a complex cone over dP 1:
W = X21 X14 X42 + X21 X13 X32 + X21 X14 X43 X32 .
(4.10)
Since we only care about the symmetries, we have suppressed the coefficients and flavor
indices; see [17] for the precise superpotential. We can think of the resulting SCFT as the
IR limit of a RG flow, where the UV limit of the RG flow is the above described W = 0
SCFT, deformed by the relevant perturbation (4.10). We now obtain the exact R-charges
of this IR SCFT.
The superpotential (4.10) reduces the U (1)2 flavor symmetry of the W = 0 theory
to U (1): the superpotential respects the R0 symmetry in (4.7) (since all terms in the
superpotential have R-charge 2) but is only neutral under J = 2J1 + J2 . So rather than
maximizing atrial with respect to s1 and s2 independently, as before, we now maximize
subject to the constraint that s1 = 2s2 s. We can impose (1.5) by simply plugging into
199
(4.3), which gives s = 0. Thus the R0 charges in (4.7) are the exact R-charges for the theory
with superpotential (4.10). Using these charges, one finds a = 27N 2 /32 0.844 N 2 . This
is less than the value (4.9) which we found in the W = 0 case, which is consistent with the
a-theorem.
String theory gives another way to determine the exact central charge a, and also the
exact R-charges, in terms of the H5 geometry. In particular, the baryonic operators such
as B12 = det X12 correspond to particles in AdS5 , which arise from D3-branes wrapped
on 3-cycles of H5 . The R-charges of the baryons, and hence the bi-fundamentals, are then
related to the volume of the H5 3-cycles which the D3 wraps. This can be computed and
shown to agree perfectly with the above results [18].
Acknowledgements
We thank T. Banks, M. Gross, J. Kumar, D. Kutasov, H. Osborn, M. Rocek, I. Rothstein,
and C. Vafa for discussions and correspondence. This work was supported by DOE-FG0397ER40546.
References
[1] N. Seiberg, Electricmagnetic duality in supersymmetric non-Abelian gauge theories, Nucl. Phys. B 435
(1995) 129, hep-th/9411149.
[2] V.A. Novikov, M.A. Shifman, A.I. Vainshtein, V.I. Zakharov, Exact Gell-MannLow function of supersymmetric YangMills theories from instanton calculus, Nucl. Phys. B 229 (1983) 381.
[3] D. Anselmi, D.Z. Freedman, M.T. Grisaru, A.A. Johansen, Nonperturbative formulas for central functions
of supersymmetric gauge theories, Nucl. Phys. B 526 (1998) 543, hep-th/9708042.
[4] D. Anselmi, J. Erlich, D.Z. Freedman, A.A. Johansen, Positivity constraints on anomalies in supersymmetric
gauge theories, Phys. Rev. D 57 (1998) 7570, hep-th/9711035.
[5] A.B. Zamolodchikov, Irreversibility of the flux of the renormalization group in a 2D field theory, JETP
Lett. 43 (1986) 730, Pisma Zh. Eksp. Teor. Fiz. 43 (1986) 565.
[6] D. Anselmi, Inequalities for trace anomalies, length of the RG flow, distance between the fixed points and
irreversibility, hep-th/0210124.
[7] P.C. Argyres, M.R. Douglas, New phenomena in SU(3) supersymmetric gauge theory, Nucl. Phys. B 448
(1995) 93, hep-th/9505062.
[8] P.C. Argyres, M. Ronen Plesser, N. Seiberg, E. Witten, New N = 2 superconformal field theories in four
dimensions, Nucl. Phys. B 461 (1996) 71, hep-th/9511154.
[9] H. Osborn, N = 1 superconformal symmetry in four-dimensional quantum field theory, Ann. Phys. 272
(1999) 243, hep-th/9808041.
[10] D.J. Gross, F. Wilczek, Asymptotically free gauge theories. 2, Phys. Rev. D 9 (1974) 980.
[11] T. Banks, A. Zaks, On the phase structure of vector-like gauge theories with massless fermions, Nucl. Phys.
B 196 (1982) 189.
[12] K.A. Intriligator, N. Seiberg, Phases of N = 1 supersymmetric gauge theories in four-dimensions, Nucl.
Phys. B 431 (1994) 551, hep-th/9408155.
[13] D. Kutasov, A. Schwimmer, N. Seiberg, Chiral rings, singularity theory and electricmagnetic duality, Nucl.
Phys. B 459 (1996) 455, hep-th/9510222.
[14] S.J. Gates, M.T. Grisaru, M. Rocek, W. Siegel, Superspace, or one thousand and one lessons in
supersymmetry, Front. Phys. 58 (1983) 1, hep-th/0108200.
[15] I.L. Buchbinder, S.M. Kuzenko, Ideas and Methods of Supersymmetry and Supergravity: or a Walk through
Superspace, IOP Publishers, Bristol, UK, 1998.
200
[16] J. Erdmenger, Gravitational axial anomaly for four-dimensional conformal field theories, Nucl. Phys. B 562
(1999) 315, hep-th/9905176.
[17] B. Feng, S. Franco, A. Hanany, Y.H. He, Symmetries of toric duality, JHEP 0212 (2002) 076, hepth/0205144.
[18] K. Intriligator, B. Wecht, Baryon charges in 4D superconformal field theories and their AdS duals, hepth/0305046.

Universality of T-odd effects in single spin

and azimuthal asymmetries
D. Boer, P.J. Mulders, F. Pijlman
Department of Physics and Astronomy, Vrije Universiteit De Boelelaan 1081,
NL-1081 HV Amsterdam, The Netherlands
Abstract
We analyze the transverse-momentum-dependent distribution and fragmentation functions in
space-like and time-like hard processes involving at least two hadrons, in particular, 1-particle
inclusive leptoproduction, the DrellYan process and two-particle inclusive hadron production in
electronpositron annihilation. As is well known, transverse momentum dependence allows for
the appearance of unsuppressed single spin azimuthal asymmetries, such as Sivers and Collins
asymmetries. Recently, Belitsky, Ji and Yuan obtained fully color-gauge-invariant expressions for
the relevant matrix elements appearing in these asymmetries at leading order in an expansion in the
inverse hard scale. We rederive these results and extend them to observables at the next order in
this expansion. We observe that at leading order one retains a probability interpretation, contrary
to a claim in the literature and show the direct relation between the Sivers effect in single spin
asymmetries and the QiuSterman mechanism. We also study fragmentation functions, where the
process-dependent gauge link structure of the correlators is not the only source of T-odd observables
and discuss the implications for universality.
PACS: 13.60.Hb; 13.87.Fh; 13.88.+e
1. Introduction
The study of polarization and transverse-momentum-dependent distribution functions
was initiated by Ralston and Soper [1]. Their study of the DrellYan process was performed
at tree level and did not address the color gauge (non-)invariance of the distribution
functions. At leading order (tree level) no single spin asymmetries were obtained in the
E-mail address: dboer@nat.vu.nl (D. Boer).
doi:10.1016/S0550-3213(03)00527-3
202
D. Boer et al. / Nuclear Physics B 667 (2003) 201241
DrellYan process (see also [2]). Sivers [3] proposed a specific non-trivial correlation
involving polarization and transverse momentum, that would lead to unsuppressed single
spin azimuthal asymmetries. For distribution functions such a correlation seemed to entail
a violation of time-reversal invariance. Collins [4] showed that this was not the case
for similar correlations in the fragmentation process. Nevertheless, phenomenological
studies of the consequences of the Sivers effect were performed [5,6]. Recently, Brodsky,
Hwang and Schmidt [7] (BHS) demonstrated in an explicit model calculation that the
Sivers asymmetry can in principle arise, after which Collins [8] demonstrated that it is
the presence of a path-ordered exponential in the definition of transverse-momentumdependent distribution functions that allows for the Sivers effect without a violation of
time-reversal invariance.
This generated renewed interest in the proper gauge-invariant definition of transversemomentum-dependent correlators. The definitions of transverse-momentum-dependent
parton densities of Refs. [9,10] did contain path-ordered exponentials (links) to ensure
color gauge invariance, but these were not closed paths (each quark field has a straight link
to infinity attached to it, but pieces at infinity are missing). If one includes such links by
hand, it is no problem to consider also closed paths, but Efremov and Radyushkin [11] had
demonstrated that the path of the link can be derived in transverse-momentum-integrated
parton densities and this can also be done when the transverse momentum is not integrated
over. In this way different processes can yield different paths [12], but no physical
observable effects were expected from such links. However, until recently these derivations
were incomplete, since the obtained paths were not closed. The missing piece would have
to involve transverse gluon fields at lightcone infinity, which were thought not to affect
physical matrix elements or at the very least lead to contributions suppressed compared to
the leading order. Recently, the derivation of fully color-gauge-invariant matrix elements,
with paths closed at light-cone infinity, was completed by Belitsky, Ji and Yuan [13,14].
They observed that transverse gluon fields do not always lead to suppression, contrary to
common belief, formalizing the model results of BHS. The resulting fully color-gaugeinvariant matrix elements strengthen the observation of Collins [8] that the presence of the
link invalidated the earlier proof of the absence of the Sivers function due to time-reversal
invariance.
With all these technical details clarified, the justification of the phenomenological
studies of the Sivers (and similar) effects was provided. Next, however, the question of
observable process-dependence arose. Collins [8] demonstrated that the Sivers asymmetry
in (semi-inclusive) deep-inelastic scattering (DIS) and the DrellYan process must occur
with opposite signs. This has been confirmed in the BHS model calculation [15], but still
awaits experimental verification. It would be the first demonstration of an observable effect
due to the presence of a path-ordered exponential in the hadron correlators and thereby
would show the intrinsic non-locality of the operators occurring in these semi-inclusive
processes.
Given this process dependence it is relevant to study which color-gauge-invariant
distribution and fragmentation functions appear in different processes. In this paper
we study color gauge invariance of transverse-momentum-dependent distribution and
fragmentation functions appearing in hard processes, in particular in semi-inclusive deepinelastic leptoproduction (SIDIS), the DrellYan process (DY) and e+ e annihilation.
(a)
203
(b)
Fig. 1. Diagrams contributing in inclusive deep-inelastic scattering.
We will employ a field theoretical approach to these hard processes and follow the
notation and derivation of Ref. [12], now taking into account the additional contributions
uncovered by Belitsky, Ji and Yuan [14]. Our analysis is different at several points, but
we confirm their results. In addition, we obtain new results for the first sub-leading order
results in an expansion in inverse powers of the hard scale. For instance, we demonstrate for
the first time a direct relation between the Sivers effect in single spin asymmetries and the
QiuSterman mechanism. Also, we study transverse-momentum-dependent fragmentation
functions, where the process dependence of the gauge link structure of the correlators is not
simply an overall sign. Rather one finds that two different (but universal) matrix elements
enter in different combinations.
In this paper leading and sub-leading always refer to the expansion in inverse powers
of the hard scale. Perturbative QCD corrections beyond tree level (next-to-leading order
in s ) will need to be taken into account as well in further studies (see Refs. [4,10,16] for
discussions of additional complications beyond tree level). In this paper we present a new
way of isolating the leading and first sub-leading order parts of the cross sections in terms
of correlators including the proper gauge links before evaluating them explicitly. These
separate color-gauge-invariant expressions for each order have not been presented before.
They facilitate the evaluation of asymmetries arising at a given order.
Now we will outline the more technical steps to be followed in this paper. In the field
theoretical approach, expressions for the structure functions of inclusive deep-inelastic
scattering (DIS) are obtained from diagrams as shown in Fig. 1 [17,18]. In these diagrams
soft parts appear that represent matrix elements of the fields corresponding to the quark
and gluon legs connecting the hard and soft parts of a diagram. The expressions for
SIDIS structure functions are obtained from diagrams as shown in Fig. 2. Again soft parts
represent specific matrix elements. In this paper we will only consider tree-level results,
which means that if gluons appear, they are in essence legs of the soft parts. In other words,
their (soft) couplings to the hard scattering part are included in the definition of the matrix
elements. This approach requires a careful treatment of the diagrams involving quark
quarkgluon matrix elements such as those in Fig. 1(b) or Fig. 2(b)(f). This is important in
order to arrive at color-gauge-invariant matrix elements that form the universal quantities,
the distribution and fragmentation functions, appearing in cross sections.
In Section 2 we outline the diagrammatic approach in a number of steps, using two
complementary lightcone directions n+ and n , which in the presence of a hard scale (in
DIS or SIDIS, the photon momentum) are fixed by the hadron momenta. The simplest
204
(a)
(b)
(c)
(d)
(e)
(f)
Fig. 2. Diagrams contributing in 1-particle inclusive deep-inelastic scattering.
(handbag) diagrams in Figs. 1(a) and 2(a) only involve quarkquark matrix elements. In
DIS the hadron momentum defines the lightcone direction n+ and the nonlocality in the
matrix elements is restricted along the lightcone direction n (for which n+ n = 1). As
is well known, diagrams as in Fig. 1(b) with any number of A+ = A n gluons yield
the necessary gauge link connecting the two quark fields [11]. The nonlocal quarkquark
operator combination with a gauge link can be expanded into a tower of local twist-two
operators with different spins. Their matrix elements appear in the cross section as leading
terms in an expansion in inverse powers of the hard scale. Diagrams with (transverse)
AT gluons or with A gluons appear in matrix elements of higher twist operators, which
appear in the cross section in terms suppressed by inverse powers of the hard scale.
The situation in SIDIS (Fig. 2), discussed in Section 3, differs in a subtle way from
that of DIS, because the nonlocality in the operator combinations is not restricted to
the lightcone, but involves also transverse separations. The kinematics only constrain the
nonlocality to the light front. In our analysis we first consider the A+ gluon legs in diagrams
205
as in Fig. 2(b) and (f). These diagrams, as in DIS, will give rise to gauge links, but in this
case connecting a quark field along the n direction to (cf. [10,12]), where the sign
depends on the type of process. By including diagrams of the type in Fig. 2(c) as well, one
can absorb all A+ gluons into the lower blob. Diagrams like Fig. 2(d) and (e) allow one
to absorb all A gluons into the upper blob, resulting again in gauge links running along
n+ to infinity. Effectively one then considers only Fig. 2(a), but now with a and that
contain the gauge links.
Diagrams with transverse AT -legs, lead to quarkquarkgluon matrix elements, which
will turn out to be suppressed, except for the boundary terms at lightcone infinity, recently
discussed by Belitsky et al. [14]. We outline an alternative for this procedure and show that
the latter appear when one expresses these fields in the appropriate field strength tensor
G+ . The boundary terms that arise in this way combine into the transverse piece that
completes the gauge link connecting the two quark fields (running via lightcone infinity).
Upon integration over transverse momenta the result reduces to the correct gauge-invariant
operator of ordinary inclusive DIS.
In Sections 46 a comparison is made between different processes involving at least
two hadrons, in particular between 1-particle inclusive leptoproduction, the DrellYan
process and two-particle inclusive hadron production in electronpositron annihilation.
For instance, in DrellYan the links run in opposite directions along the n direction
compared to SIDIS (as noticed in [8,12]). Because the two situations can be connected
via a time-reversal operation, one can define T-even and T-odd functions that appear
in the parametrization of the color-gauge-invariant matrix elements. For these functions
factorization in principle should hold, although they appear with different signs in
SIDIS and DY [8]. The T-odd functions appear in single spin asymmetries in these
processes [3,4,6,19] or they appear in pairs in unpolarized azimuthal asymmetries [2022].
In Section 7 we study the time-reversal properties of distribution and fragmentation
functions and present explicit parametrizations.
2. Hadron tensor and correlators in SIDIS

The hadron tensor for 1-particle inclusive leptoproduction is given by

1
d3 PX
(lH )
(2)4 4 (q + P PX Ph )
2MW (q; P , S; Ph , Sh ) =
(2)4 (2)3 2PX0
lH
H
(PX ; P S; Ph Sh ),
(1)
with H being the product of current expectation values

(lH )
H
(PX ; P , S; Ph , Sh ) = P , S|J (0)|PX ; Ph Sh PX ; Ph Sh |J (0)|P , S.
(2)
Due to the fact that the summation and integration over final states is not complete,
prohibiting the formal use of the operator product expansion, we proceed along the lines
of the diagrammatic approach of Refs. [1,17], based on nonlocal operators. The quark and
gluon lines connected to the soft parts represent matrix elements of (nonlocal) quark and
gluon operators.
206
The hadron tensor is calculated for current fragmentation in deep-inelastic scattering.

In that case the exchanged momentum q 2 Q2 is large and one has for the target
momentum P and the produced hadron momentum Ph the conditions that P q, Ph q
and P Ph are large, of O(Q2 ). One is able to make a systematic expansion in orders
of 1/Q, of which we will only consider the first two terms, (1/Q)0 and (1/Q)1 . In this
situation one uses the scaling variables
Ph q
Q2
,
2P q
Ph P
2Ph q
P Ph
zh =
,
P q
Q2
xB =
(3)
(4)
where the approximate sign indicates equalities up to 1/Q2 (mass) corrections. It is

convenient to introduce lightlike vectors n+ and n satisfying n+ n = 1 along the hadron
momenta writing
M2
Q
n + n+ ,
2
Q 2
Mh2
Q
n+ ,
Ph = n +
2
Q 2
P =
(5)
(6)
Q
Q
q = n n+ + qT ,
(7)
2
2
with qT2 Q2T and Q 2 = Q2 + Q2T . These equations define the lightcone coordinates
{ }
a a n and the transverse projector gT = g n+ n . In our treatment of the
1-particle inclusive process we will consider Q2T Q2 , hence Q 2 Q2 , while xB
and zh up to mass corrections of order 1/Q2 . The vector qT q + x P
Ph /z determines the off-collinearity in the process. In principle, mass corrections can

straightforwardly be incorporated. Important to note is that the lightlike directions n =
n (P , Ph ) are determined by the hadron momenta P and Ph .
From the diagrammatic expansion (see Fig. 2(a)(e)) one obtains up to O(g),
2MW (q; P , S; Ph , Sh )

=
d4 p d4 k 4 (p + q k) Tr (p) (k)

k/ p
/1 + m
d4 p1 Tr
(p,
p
p
)
(k)
A
1
(k p1 )2 m2 + i)

k/ p
/1 + m
(k)
(p
p
,
p)
d4 p1 Tr
A
1
(k p1 )2 m2 i)

p
/ k/1 + m
(p)
(k
k
,
k)
d4 k1 Tr
A
1
(p k1 )2 m2 + i)

p
/ k/1 + m
d4 k1 Tr
(k,
k
k
)
(p)
A
1
(p k1 )2 m2 i)
[term 1]
[term 2]
[term 3]
[term 4],
(8)
(a)
(b)
(c)
(d)
207
Fig. 3. Soft parts representing the quarkquark and quarkquarkgluon matrix elements used in Eqs. (9)(12).
where
d4 ip
e P , S| j (0)i ( )|P , S,
(9)
(2)4

d4 ik
e 0|i ( )|Ph , X Ph , X| j (0)|0,
ij (k; Ph , Sh ) =
(10)
(2)4
X

d4 d4 ip ip1 ( )
Aij
(p, p p1 ; P , S) =
e e
(2)4 (2)4
(11)
P , S| j (0)gA ()i ( )|P , S,
d4 d4
Aij (k, k k1 ; Ph , Sh ) =
eik eik1 ( ) 0|i ( )gA ()|Ph , X
(2)4 (2)4
ij (p; P , S) =
Ph , X| j (0)|0,
(12)
illustrated in Fig. 3. In the above expression we have omitted the contributions with the
opposite direction on the fermion line. It adds to the result in Eq. (8) terms with q q
and . In cross sections it will always lead to extending a sum over quarks and
antiquarks.
The aim of the calculation is an expansion in powers of 1/Q. For this a number
of considerations are important. First, the matrix elements represented by blobs in the
diagrammatic expansion should vanish fast enough when any of the products of momenta
involved becomes large, e.g., the virtualities of the quarks or gluons. To be precise, in
Fig. 3 the products p2 p12 p p1 p P p1 P P 2 = M 2 Q2 . With the choice
of parametrization in Eqs. (5)(7), this implies that for the momenta in Fig. 3(a) and (c)
one has for the plus-components p+ , p1+ , P + Q, while the minus-components p , p1 ,
P 1/Q. For the fragmentation parts (Fig. 3(b) and (d)) one has minus-components k ,
k1 , Ph Q, while for the plus-components k + , k1+ , Ph+ 1/Q. The transverse momenta
are of O(M). Introducing momentum fractions x = p+ /P + and z = Ph /k , writing
p = p n + xP + n+ + pT ,
(13)
208
k =
Ph
n + k + n+ + kT ,
z
(14)
one finds, when neglecting O(1/Q2 ) contributions, that 4 (p + q k) (x xB )(z

zh ) 2 (pT + qT kT ), thus identifying the scaling variables and momentum fractions,
x = xB and z = zh .
Thus, in a calculation up to O(1/Q2 ), the integration over the minus-components of
momenta in the matrix elements and A can be performed, restricting them to the
lightfront,

ij (x, pT ) = dp ij (p; P , S)

d d2 T ip
=
(15)
e P , S|j (0)i ( )|P , S
,
3
(2)
+ =0
while A (p+ , pT , p1+ , p1T ) = dp dp1 A (p, p p1 ; P , S) involves two integrations

over the minus-components of the parton momenta. We will occasionally also use the
variable x1 defined via p1+ = x1 P + . The integrations over minus-components are sufficient
to render the time-ordering in these matrix elements superfluous, which can be proven
completely analogous to the proof for the matrix elements in which also the integration over
transverse momenta is performed, in that case restricting them to the lightcone [23,24]. In
the matrix elements of the types and A the integrations over the plus-components of
the quark and gluon momenta can be performed, leading to lightfront correlation functions
(z, kT ),

ij (z, kT ) = dk + ij (k; Ph )

d + d2 T

ik
=
e 0|i ( )|Ph , X Ph , X|j (0)|0
,
3
(2)
=0
X
(16)
and A (k , kT , k1 , k1T ) = dk + dk1+ A (k, k k1 ; Ph ).

Matrix elements like (x, pT ) have a particular Dirac structure, Lorentz structure,
and canonical dimension, which must be visible in the parametrization of through
the dependence on non-integrated parton momenta and the
hadron momentum and spin

vectors. One deduces immediately the Dirac structure dp n/ + = giving a
leading contribution and the Dirac structure involving the unit matrix in Dirac space
requiring in addition a factor P in the parametrization, which will
lead in the calculation
of Eq. (8) to a suppression factor 1/Q. The leading structure of dp
dp1 A matrix
elements involving
integrations over minus components gives
dp dp1 A+
two
for

, similar to dp , but for a transverse gluon one gets dp dp1 AT P

leading to a 1/Q suppression in the calculation of Eq. (8) (apart from the subtlety with
the boundary terms, where the role of P is taken over by (p1+ ), to be elaborated upon
below). Of course, in a parametrization of the latter matrix element also the transverse
index must appear, e.g., a non-integrated parton transverse momentum pT or the spin
vector ST in case of a transversely polarized hadron, but these are not relevant for an
209
expansion in powers of 1/Q. The matrix element dp dp1 A will always appear
suppressed by at least (P )2 1/Q2 .
the fragmentation parts

one has after integration over plus-components dk +
For
+
dk + dk1+ AT Ph+ + , becoming suppressed by 1/Q
dk + dk1+
A , while
(again apart from the subtlety with boundary terms). The matrix element dk + dk1+ +
A
will always appear suppressed by at least (Ph+ )2 1/Q2 . The explicit parametrizations
for the matrix elements in terms of distribution and fragmentation functions have been
extensively discussed in many papers [2,6,19,20,25] and will be summarized in Section 7.
In order to find the leading contributions, we need in the calculation of Eq. (8) not
only the first term (diagram in Fig. 2(a)), but also the terms involving A+ and
A
++
(Fig. 2(b)(e)) and even multiple-gluon matrix elements of the form AA
(Fig. 2(f)), etc.
Such a resumming of multiple-gluon matrix elements can be easily performed in DIS,
where the integration over transverse momenta of partons can always be performed in
addition to the minus-integration.
The resummation leads to a modified first term in Eq. (8)
with in the (x) = dp d2 pT (p; P , S) matrix element the inclusion of a gauge link

U[a,
] = P exp ig

a

(ig)
N=0

d A+ ( )

+ = + =a + ,T =T =aT
d1 A+ (1 )
N1

dN A+ (N )

, (17)
i+ = + =a + ,iT =T =aT
connecting the quark fields, rendering the object color-gauge-invariant [11]. We will
discuss the full procedure to obtain a color-gauge-invariant object in 1-particle inclusive
leptoproduction in the next section, following in part Ref. [12] and recent work by Belitsky,
Ji and Yuan [14].
3. Color gauge-invariance in SIDIS

In this section we will discuss the resummation of contributions in SIDIS coming from
diagrams in Fig. 2(b)(f).
At orders (1/Q)0 (leading) and (1/Q)1 (first sub-leading) the integrations over p , p1 ,
and k + in the corresponding soft parts can be performed. The result of the first term of four
quarkquarkgluon contributions in Eq. (8) is

[term 1] =
d4 p d4 k 4 (p + q k)

k/ p
/1 + m
d4 p1 Tr
(p,
p
p
)
(k)
A
1
(k p1 )2 m2 + i)
210

=
d2 pT d2 kT 2 (pT + qT kT ) dp1+ d2 p1T

d d2 T d d2 T ip ip1 ( )
e e
(2)3
(2)3
k/ p
/1 + m
P , S| (0) (z, kT )
(k p1 )2 m2 + i)

gA ()( )|P , S
,
+
+
(18)
= =0
where A is made explicit (Eq. (11)) and the minus- and plus-integrations are performed.
In the expression after the second equal sign, it is understood that in the integrand
p = p1 = k + = 0 while p+ = xP + and k = Ph /z.
Next, we will split off from the quark propagator those parts that are relevant at leading
and first sub-leading order in 1/Q. These parts depend on whether the index of the
gluon field is plus, transverse or minus. For the 1/Q order, we will restrict ourselves
to obtaining a color-gauge-invariant expression for the hadron tensor integrated over the
transverse momentum qT of the photon. For this result one first needs to consider the
leading order term unintegrated over qT . The end results for the hadron tensor in several
cases are summarized in the next section.
One finds for the quark propagator explicitly (with k q = Q/ 2 )

(/k + m) n
/ + p1+ p
/ 1T
k/ p
/1 + m
.
(k p1 )2 m2 + i) p1+ Q 2 + (kT p1T )2 m2 + i)
(19)
/ term of the quark propagator,

Obvious contributions at leading order are, the k n
which in combination with A+ gluons leads to the link operator in the direction. Less
obvious are the contributions from fields that are independent of which, as can be
seen from Eq. (18), lead to a delta-function (p1+ ). In that case other leading contributions
appear. In particular, a contribution coming from the last (transverse) term will lead to
(leading) link contributions in the transverse direction.
Contributions at order 1/Q are coming from the n
/ + term of the quark propagator in
combination with the transverse gluons, and the transverse part of the quark propagator,
p
/ 1T , in combination with the A+ gluons. These contributions can be combined into a
color-gauge-invariant matrix element containing the field strength tensor. A summary is
given in Table 1.
The leading contribution in Eq. (18) comes from A+ . We use that (/k + m) =
2k (/k m) , the fact that (k)(/k m) A 1/Q (QCD equations of motion)

Table 1
The color-gauge-covariant objects into which the gluon fields in SIDIS
are combined depending on the Dirac structure of specific terms in the
hard quark propagator to which the gluon couples
A+
AT
/n+
/n
G+
G+
UT
211
to obtain
k/ p
/1 + m
A+ ()
2
(k p1 ) m2 + i)
k/ p
/1 + m +
(k) +
A ()
2k (p1 i))
(k)
(k)
(k)
p
(k)(/k m) A+ ()
/ 1 A+ ()
A+ ()
+
(k)
+
p1+ i)
2k (p1+ i))
2k (p1+ i))
A+ ()
/ 1T A+ ()
p
+ (k)
+
p1 i)
Q 2 (p1+ i))
(20)
with omitted parts being of O(1/Q2 ). The first term inserted in Eq. (18) gives a leading
contribution,

[term 1.1] = d2 pT d2 kT 2 (pT + qT kT ) dp1+ d2 p1T

e e
(2)3
(2)3

gA+ ()
P , S| (0) (z, kT ) +
( )|P , S
p1 i)
+ =+ =0

d d2 T ip
e
= d2 pT d2 kT 2 (pT + qT kT )
(2)3
P , S| (0) (z, kT ) (ig)

+
(21)
.
d A ()( )|P , S
+ +
= =0;T =T
This is precisely the O(g) term in the expansion of U[,

] multiplying ( ). The result of
+
the diagram in Fig. 2(f) with two A -gluons gives the O(g 2 ) term, etc. From the second
term in Eq. (18) (diagram in Fig. 2(c)) one obtains the O(g) term in the expansion of
following (0).
U[0,]
The A -gluons in the other diagrams in Fig. 2(d) and (e) and corresponding higher
orders can all be absorbed into link operators in modified soft parts of the form

d d2 T ip
j (0)U U i ( )|P , S
ij (x, pT )
e
P
,
S|
[0,] [, ]
+ , (22)
3
(2)
=0
d + d2 T
+
ij (z, kT )
eik 0|U[,
] i ( )|Ph , X
(2)3
X

+
(23)
,
Ph , X| j (0)U[0,] |0
=0
+
where U[,
] indicates a link along the lightcone plus-direction running from
+
to . These quantities, however, are not color-gauge-invariant, although we note that upon
212
integration over pT and kT one obtains color-gauge-invariant lightcone correlators (x)

and (z), in which the two links merge into one connecting the lightlike separated points 0
and . These are, e.g., important in qT -integrated SIDIS cross sections at leading order. For
the transverse-momentum-dependent functions, however, we are still missing a transverse
piece that leads to color-gauge-invariant definitions. It has to come from transverse gluons,
which are next to be investigated.
Since the dominant part of dk + is proportional to + , one finds (naively) for the
transverse gluons in term 1 (Eq. (18)),
(k)
p1+
k/ p
/1 + m
A
()
(k)
A ()
(k p1 )2 m2 + i) T
2k (p1+ i)) T
(k) AT ().
Q 2
(24)
The (remaining) second term in Eq. (20) and the result of Eq. (24) give as O(1/Q)
contribution in term 1,
d2 pT d2 kT 2 (pT + qT kT ) dp1+ d2 p1T

e e
(2)3
(2)3
1
P , S| (0) (z, kT )
Q 2

p
/ 1T A+ ()
AT () +
( )|P , S| + =+ =0
(p1 i))

d d2 T ip
e
= d2 pT d2 kT 2 (pT + qT kT ) d2 p1T
(2)3
1
P , S| (0) (z, kT )
Q 2
+
AT ( ) d T A ( ) ( )|P , S
.
+
[term 1.2] =
(25)
=0
Including in addition all diagrams with longitudinal A+ gluon fields, all colored fields
become linked along the minus-direction, with the same link directions for and A .
Using the relation between G+ and AT , outlined in Appendix A including all minus
links, we find (suppressing the links U )

AT ( )
T A+ ( ) =
d G+ () + AT ( ),
(26)
213
with the points = ( , + , T ) and = (, + , T ). The part of term 1.2 containing

G+ is

[term 1.2a] =

d2 pT d2 kT 2 (pT + qT kT )
d d2 T ip
e
(2)3
1
P , S| (0) (z, kT )
Q 2

d G+ () ( )|P , S
+ +
(27)
= =0;T =T
Upon integrating the hadron tensor over transverse momenta qT , the above convolution
factorizes and produces a color-gauge-invariant O(1/Q) term,

d2 qT [term 1.2a]

= Tr

Q 2
dp1+

+ +
i
+
p , p p1 (z) ,
p1+ i) G
(28)
where (including link operators)

+ +

d d ip ip1 ( )
Gij
e e
p , p p1+ =
2 2

P , S| j (0)U[0,]
gG+ ()U[,
(
)|P
,
S
] i

(29)
LC
(x, x x ) with x = p+ /P + , and with LC denoting { + = + = =

also denoted G
1
1
T
1
T = 0}.
We are left with a boundary term containing AT ( ), which needs special care. The
argument of the transverse field in the boundary term is fixed by the link direction in U .
The consequence is that the dependence disappears. We note that the integration over
thus can simply be performed, showing that one deals with a matrix element that is
proportional to (p1+ ) in momentum space [26],

p1+ A()ij
(p, p p1 )
d4 d4 ip ip1 ( )
e e
(2)4 (2)4
P , S| j (0)gAT (, + , T )i ( )|P , S. (30)
Because (p1+ ) 1/Q one finds that dp dp1 A()

+ , i.e., it is not suppressed.
This means we have to revisit the approximations made to the fermion propagator for the
boundary term. Going back to the starting point in Eq. (18) we obtain for the boundary
214
contribution after integration over ,

[term 1.2b] = d2 pT d2 kT 2 (pT + qT kT ) dp1+ d2 p1T
d d2 T d2 T ip ip1T (T T ) +
e e
p1
(2)3 (2)2
P , S| (0) (z, kT )

k/ p
/1 + m
(
)|P
,
S
gA
,
T
T
+ + .
(k p1 )2 m2 + i)
= =0
(31)
After a substitution for AT ,

T
gAT () = i (ig)

dT AT , + , T ,
(32)
we do a partial integration. In the matrix element, we then encounter the following part,
which we need to consider in the soft gluon limit (p1+ = 0) in which the denominator of
the quark propagator can no longer be approximated as in Eqs. (20) or (24). Realizing that
p1+ Q and p1 1/Q we obtain

k/ p
/1 + m
p1+
2
2
(k p1 ) m + i)

k/ p
/1 + m
p1+
(k)/
p1
(k p1 )2 m2 + i)

k/ p
/1 + m
= (k)(/k + p
/ 1 + m)
p1+
(k p1 )2 m2 + i)

= (k) p1+ .
(k)/
p1T
(33)
The result after integration over T , p1+ and p1T is a term

[term 1.2b] =
d pT d kT (pT + qT kT )
2
d d2 T ip
e
(2)3
P , S| (0) (z, kT ) (ig)

T

dT AT (, 0, T )( )|P , S
+ =0
(34)
which gives precisely the first term of the transverse link that is needed to modify Eq. (22)
into a fully color-gauge-invariant matrix element. Note that we did not assume a specific
pure gauge expression for the AT field at = . Furthermore, we did not neglect the
quark masses in the quark propagator. For N transverse gluons we have to work out the
boundary terms with more transverse gluons, for which we need the following relation that
215
Fig. 4. Link structure for [+] (x, pT ) (a) and (b) and for (x) (c) after integration over transverse momenta.
also holds for nonabelian fields:

(ig)
dMT AT (M ) gAT ()
M1

= i
d1T AT (1 )

(ig)

d1T AT (1 )
M+1
dM+1T AT (M+1 ) .
(35)
This indeed produces nicely the nested integrations needed in the path-ordered exponential
and we find that the terms 1.1 and 1.2b, are taken care of by using in the O(g 0 ) part of
Eq. (8) the color-gauge-invariant matrix element [14]

d d2 T ip
ij[+] (x, pT ) =
e
(2)3

T
T

U[0
U
U
(
)|P
,
S
,
P , S| j (0)U[0,]

T ,T ] [T ,T ] [, ] i
+ =0
(36)
with

T

=
P
exp
ig
d
A
(
)
U[a,
T
T
]

=

N=0
+ = + =a + , = =a
T

d1T AT (1 )
(ig)N
aT
N1
dNT

AT (N )

.
i+ = + =a + ,i = =a
(37)
T
U[0
-link
T ,T ]
The
of course comes from diagrams with transverse gluons like in Fig. 2(c)
and higher orders. The link structure for the soft part describing the distribution of quarks
in a hadron probed by a spacelike photon is illustrated in Fig. 4. The direction of the link,
running to + along the minus direction is indicated via the superscript [+] in Eq. (36).
In other processes one will find that the link can also run in the opposite direction to
along the minus direction. This will be indicated with a superscript [].
Eq. (36) is an important expression, since in its full generality it allows for certain
distribution functions, usually referred to as T-odd functions, that would be absent in case
one ignores the gauge links (cf., e.g., [3,4,6]). Without the transverse gauge links, it may
therefore seem that a choice of A+ = 0 gauge would demonstrate the absence of such
216
T-odd functions. In the derivation of Eq. (36) no gauge was assumed (one can actually
arrive at this result by first considering the A = 0 gauge as done in Ref. [12]), hence it
should not be viewed as one out of many ways to gauge-invariantize the matrix element
[27]. Up to the order we consider here, the result is derived rather than assumed.
We note that in leading results the color-gauge-invariant object [+] (x, pT ) contracted
with + still is a semi-positive definite matrix in Dirac space, which is the basis for deriving
positivity conditions, such as the Soffer bound [28] and many more [29]. Hence, we
disagree with the statement Structure functions are not parton probabilities by Brodsky
et al. [30]. To be precise, the distribution functions containing the transverse link are still
probability densities.
As mentioned already, when one considers the qT -integrated results one obtains the
lightcone quarkquark correlations with the link in Eq. (17). The transverse link does not
affect that result and one has [+] (x) = [] (x) = (x) (see Fig. 4(c)). If one looks at
azimuthal asymmetries or weighted cross sections one needs to consider matrix elements
weighted with transverse momentum. In those cases one explicitly needs to take into
account the transverse part of the link. We define transverse moments [] (x),

[] (x) d2 pT pT [] (x, pT ),
(38)
that in a straightforward way can be related to color-gauge-invariant quarkquarkgluon
and , the latter involving the covariant derivative,
matrix elements G
D
[]
(x)
ij

d d2 ip
2
= d pT
e
(2)3
T
T
P , S| j (0)U[0,]
U[0
i U[
U
( )|P , S| + =0
T ,T ]
T ,T ] [, ] i

d ip

e
=
P , S| j (0)U[0,
] iDT i ( )|P , S LC
(2)
P , S| j (0)U[0,]

U[,]
gG+ ()U[,
] i ( )|P , S
, (39)
LC
or
[] (x) = D
(x)
where
Dij
(x) =

=
dp1+
p1+
+ +

i
p , p p1+ ,
G
i)
(40)
+ +
p , p p1+
dp1+ Dij

d ip
e P , S|j (0)U[0, ] iD ( )i ( )|P , S .

2
LC
(41)
217
The important observation we want to make here is that the difference between correlation
functions with links running to , respectively, is related to a quarkgluon correlator,
(x, x),
[+] (x) [] (x) = 2G
(42)
the latter being given the name gluonic-pole matrix element since it corresponds to the softgluon point p1+ = 0. Its consequences have been studied for several processes [26,3135]
and it is viewed as one of the possible mechanism to generate single spin asymmetries. We
will comment on this further below, but already mention that the above relation between
(x, x) implies that the Sivers effect [3] is directly related to the Qiu
[] (x) and G
Sterman mechanism (the gluonic-pole matrix element), i.e., if one is nonzero, then the
other also is. We will make this relation more specific below.
We further define

1
(x) [+] (x) + [] (x) ,
(43)
2
1
A (x) = PV dx1 G
(44)
(x, x x1 ),
x1
where we use A (x) to distinguish the function from the non-gauge-invariant A (x).
These definitions imply
(x, x),
[] (x) = (x) G
(x) = D (x) A (x).
(45)
(46)
The relations in Eqs. (44)(46) are relations connecting color-gauge-invariant quantities.

We will return to the above functions and their properties in Section 7.
We end this section by giving the remaining contributions in the qT -integrated hadron
tensor at order 1/Q. We give a systematic summary of the SIDIS hadron tensor for several
cases in the next section. We can use Eq. (40) to rewrite the twist-3 contribution in Eq. (28)
obtained after qT -integration into the form

[+]
2
d qT [term 1.2a] = Tr
(47)
D (x) (x) (z) .
Q 2
The transverse gluons in the second term of Eq. (18) (diagram in Fig. 2(c)) produces
T
, already absorbed into Eq. (36), also a twist-three
besides the transverse link U[0
T ,T ]
piece. For this one needs matrix elements with interchanged arguments such as A (p
p1 , p). The resulting twist-3 term after integration over transverse momenta is

+

i
+
+
2
+
dp1 +
p p1 , p (z)
p1 + i) G
Q 2

[+]
= Tr 0 D
(48)
(x)0 (x) (z) ,
Q 2
where we have used that [+] (x) = 0 [+] (x)0 . The third and fourth O(g) terms in
Eq. (18) (diagrams in Fig. 2(d) and (e)) need the (direct and conjugate) quark propagators
218
(a) spacelike distribution [+]
(b) timelike distribution []
(c) timelike fragmentation [+]
(d) spacelike fragmentation []
Fig. 5. Link structure for (x, pT ) (a) and (b) and (z, kT ) (c) and (d).
(with p+ q + = Q/ 2)
(/
p + m) n
/ k1 k/1T
p
/ k/1 + m
(p k1 )2 m2 i) k1 Q 2 + (pT k1T )2 m2 i)
(49)
+
+
The calculation yields the links U[,
] and U[0,] running along the plus-direction
in the matrix elements (z, kT ) in Eq. (23) for the fragmentation part. Including the
transverse gluons we get the fully color-gauge-invariant matrix element indicated as the
spacelike fragmentation [] (z, kT ) with the link as indicated in Fig. 5. Furthermore one
obtains twist-3 contributions containing G (k k1 , k) and G (k, k k1 ), which after
integration over qT yield
2
dk1
+

i

(x)
k
k
,
k
1
k1 i) G
Q 2

+
[]
= Tr 0 D (z)0 (z) (x) ,

(50)
Q 2

i
+

2

dk1
k , k k1 (x)

k1 + i) G
Q 2

= Tr

(z)
(x)
.
D (z) []
Q 2
(51)
In deriving these results we stepped over some subtleties involving the inclusion of
diagrams with crossed gluon lines and the use of the full QCD equations of motion, but for
this we refer to Ref. [12].
219
4. SIDIS and DIS cross sections

The basic expression for W (q; P , S; Ph , Sh ) contains a convolution of the transversemomentum-dependent functions. We have seen that upon integration over qT ,

d2 qT d2 pT d2 kT 2 (pT + qT kT ) = d2 pT d2 kT ,
(52)
the integral can be deconvoluted. This is also true for azimuthal asymmetries constructed
by weighting with qT ,

d2 qT qT d2 pT d2 kT 2 (pT + qT kT ) = d2 pT d2 kT kT pT .
(53)
If one calculates the O(1/Q) result, one has to be careful, however. One cannot simply
perform the integration over qT in the hadron tensor, since the lepton tensor involves q.
To proceed, one starts with a (Cartesian) set of vectors, starting with q defining in SIDIS a
spacelike direction while the other external vectors are used to define orthogonal directions.
In particular, it is convenient to start with q and P = P (P q/q 2)q,
q
= q ,
Q
q + 2xB P
P
t =
,
Q
P 2
z
(54)
(55)
with z 2 = 1 and t2 = 1. These two vectors can be used both for inclusive and semiinclusive leptoproduction. The lepton tensor can be expressed in the vectors t and z and
a perpendicular vector (equal for the lepton in initial and final state) which defines the
azimuthal angle of the lepton
scattering plane. From the Cartesian vectors two new lightlike
vectors n = (t z )/ 2 = n (P , q) can be constructed as well as a perpendicular
{ }
tensor g g + q q t t = g n+ n . Since the lightlike directions n are
determined by P and q, instead of P and Ph , the momentum of the produced hadron Ph
will have in general a nonvanishing perpendicular component enabling us to define a vector
X Ph /zh , which defines the azimuthal angle of the hadron production plane.
It is straightforward to see that up to O(1/Q2 ) corrections, the previously defined set
{n+ , n , qT } is related to the set {t, z , X} or {n+ , n , X} via
X

n n 2 ,
Q 2
n+ n+ ,
(56)
qT X
(57)
Q2T
Q
n+ ,
(58)
and X2 Q2T . We note that the leptonic tensor is independent of X. The 1/Q term
appearing on the righthandside of Eq. (58) is irrelevant in our calculations. Hence for
experimental cross sections
we can use
for integrated and
weighted cross
up to that order
sections the replacements d2 X d2 qT and d2 X X d2 qT qT . We
220
now consider separately integrated SIDIS, DIS and azimuthal asymmetries in SIDIS. We
will also consider a few special cases. If one measures in the final state the jet-direction,
this can be considered as a measurement of pT , i.e., qT = pT . This is referred to as JET
SIDIS.
4.1. Integrated SIDIS cross section at leading order
We have seen that inclusion of appropriate quarkgluon matrix elements make the
O(g 0 ) result in Eq. (8) color-gauge-invariant,
(0)
2MW
(q; P , S; Ph , Sh )

= d2 pT d2 kT 2 (pT + qT kT ) Tr [+] (x, pT ) [] (z, kT ) .
(59)
At leading order the integration over transverse momenta of the produced hadrons simply
can be performed and one obtains the basic result

(0)
d2 X 2MW
(60)
(q; P , S; Ph , Sh ) = Tr (x) (z) n n + O(1/Q).
4.2. Azimuthal asymmetries in SIDIS at leading order

We consider here cross sections obtained after integration over X and explicit weighting
with X . In practice, this means measurement of the azimuthal angle of the produced
hadron and compare it with other azimuthal angles, such as that of the lepton scattering
plane, the (transverse) spin of the target hadron or the
(transverse) spin of the produced

hadron. For our purposes it implies calculation of d2 X X 2MW , which at leading
order gives

(0)
(q; P , S; Ph , Sh )
d2 X X 2MW

[+]
= Tr (x) []
(z)
Tr
(x)
(z)

+ O(1/Q).
n n
n n
(61)
In these cross sections one finds for instance the Collins and Sivers effects [5,6,36,37].
4.3. Integrated SIDIS cross section at O(1/Q)
As outlined one must be careful in integrating over transverse momenta. At O(1/Q)
the differences between using n (P , Ph ) or n (P , q) matter. In particular, the correlator
n
/ will lead to terms proportional to q/ T /Q 2. To find the n
/ dependence in a Dirac
/ n
/ /2. We use that the leading
space correlator, we use the projectors P = /2 = n
term in satisfies P = P+ = P P+ to write

1
/ n
/ + (z, kT )/
n+n
/
(z, kT ) = n
4

1
/ + (z, kT ) + (z, kT )/
n+q/ T ,
(z, kT ) n n q/ T n
(62)
Q 2
221
and we obtain the 1/Q contribution coming from Eq. (59),

(0)
d2 X 2MW
(q; P , S; Ph , Sh )
= O(1) result (Eq. (60)

1

d2 qT d2 pT d2 kT 2 (pT + qT kT )
Q 2

Tr [+] (x, pT )q/ T n
/ + [] (z, kT )

n+q/ T
+ Tr [+] (x, pT ) [] (z, kT )/
= O(1) result (Eq. (60))

+ Tr [+] (x) (z) Tr (x) []
(z)
Q 2
Q 2

[+]
[]
+ Tr (x) (z) Tr (x) (z) .
Q 2
Q 2
(63)
Including these 1/Q-terms and the four contributions from quarkgluon correlators with
transverse gluons, one obtains the full integrated SIDIS cross section up to O(1/Q),

(0+1)
(q; P , S; Ph , Sh )
d2 X 2MW
= O(1) result (Eq. (60))

[]
[+]
+ Tr (x) (z) Tr (x) (z)
Q 2
Q 2

+ Tr [+] (x) (z) Tr (x) []
(z)
Q 2
Q 2

+ Tr
D (x) [+] (x) (z)
Q 2

(x)0 [+] (x) (z)

+ Tr 0 D
Q 2

+
[]
+ Tr [0 D (z)0 (z)] (x)

Q 2
+

[D (z) []
+ Tr
(z)]
(x)
Q 2
= O(1) result (Eq. (60))

+

(x) (z) + Tr
+ Tr
D
D (z) (x)
Q 2
Q 2
+

[]
[]
Tr
(z) (x) Tr (z) (x)
Q 2
Q 2
+ ( ) ,
(64)
222
where the hermiticity properties of the various matrix elements have been used (see
Section 7). We note that the 1/Q part in Eq. (63) is by itself not electromagnetically gaugeinvariant, but together with the parts arising from quarkgluon matrix elements, leading to
the result in Eq. (64), it is gauge-invariant.
4.4. Integrated DIS cross section at O(1/Q)
The hadron tensor for this case is obtained by integrating the SIDIS result over z
and using for the fragmentation part the free (massless) quark quark result, (z) =
n
/ (1 z) and []
(z) = D (z) = 0. This gives the well-known result [17],
2MW (q; P , S)

= Tr (x) +

+
+ Tr
+ Tr
.
D (x)
D (x)
Q 2
Q 2
(65)
4.5. Azimuthal asymmetries in JET SIDIS at leading order

Azimuthal asymmetries in JET DIS are obtained at measured qT = Pjet and with in
addition still fixed P and q. We simply replace (z, kT ) = n
/ (1 z) 2 (kT ). Hence, we
start with

(0)
2MW
(q; P , S; qT ) = Tr [+] (x, qT ) + .
(66)
In this situation one will have to be careful with Sudakov effects [16], but the following
weighted azimuthal asymmetry is free of these effects,

(0)
(67)
d2 qT qT 2MW
(q; P , S; qT ) = Tr [+] (x) + .
This result is also a direct consequence of Eq. (61), taking for [] (z, kT ) the quark
quark limit.
4.6. Other subleading cross sections
Azimuthal asymmetries at O(1/Q), azimuthal asymmetries involving higher weighting
than with one power of the momentum X or SIDIS cross sections at O(1/Q2 ), require
considerably more theoretical efforts than the one presented above. Moreover, it might
be impossible to unambiguously disentangle hadrons within a jet from hadrons in other
jets, since the separation of jets is merely an O(Q2 ) effect, meaning Pjet 1 Pjet 2 Q2 .
This also implies that a factorization proof most likely cannot be given, just like the failure
of factorization in unpolarized processes at 1/Q4 [38]. As is well known, these difficulties
do not appear for the inclusive DIS cross section involving just one (target) hadron which
allows for a rigorous treatment at any order in powers of 1/Q.
(a)
223
(b)
Fig. 6. Quarkquark (a) and one of the quarkquarkgluon (b) correlators in tree-level diagrams for DrellYan
scattering.
5. The DrellYan cross sections

For DrellYan, one has a similar treatment as for leptoproduction. The calculation
involves now two soft distribution parts and annihilation of a quarkantiquark pair into
a gauge boson (we will only discuss the vector coupling here). The handbag diagram is
given in Fig. 6(a) and an example of a diagram with an additional gluon in Fig. 6(b).
A full calculation at tree level including quarkgluon matrix elements as discussed for
leptoproduction gives in this case
2MW (q; PA , SA ; PB , SB )

= d4 p d4 k 4 (p + k q) Tr((p) (k)
)

/k p
/1 + m
4
A (p, p p1 ) (k)
d p1 Tr
(k + p1 )2 m2 + i)

/k p
/1 + m
4
d p1 Tr
(k) A (p p1 , p)
(k + p1 )2 m2 i)

p
/ + k/1 + m
4
(p) A (k k1 , k)
d k1 Tr
(p + k1 )2 m2 + i)

p
/ + k/1 + m
4
d k1 Tr
A (k, k k1 ) (p) + , (68)
(p + k1 )2 m2 i)
where (p) and A (p, p p1 ) are the same as in leptoproduction, but the role of and
A is taken over by

d4 ik
ij (k; PB , SB ) =
(69)
e
PB , SB |i ( ) j (0)|PB , SB ,
(2)4

d4 d4 ik ik1 ( )
Aij
(k, k k1 ; PB , SB ) =
e
e
(2)4 (2)4
PB , SB |i ( )gA () j (0)|PB , SB
(70)
kT )).
(note that this implies (x, kT ) = k (x,
224
The important difference between DY and SIDIS turns out to be the direction of the
links. The result for the quark propagator in a quarkgluon diagram as in Fig. 2(b), but
then in the case of DY with a timelike outgoing photon (cf. Fig. 6(a)) yields a propagator
(/k m) n
/ + p1+ p
/ 1T
/k p
/1 + m
(71)
+
2
2
(k + p1 ) m + i) p1 Q 2 + (kT + p1T )2 m2 + i)
(since k q = Q/ 2 ). The difference with Eq. (19) is the sign with which p1+ appears
in the denominator. For the A+ -gluons this produces a link running along the minus
direction to , i.e., one finds transverse-momentum-dependent distribution functions
[] (xA , pT ), where xA p+ /PA+ q + /PA+ . Also in the antiquark matrix element the
link runs to (along the plus direction with our choice of lightlike vectors), i.e., the A
gluons lead to the matrix element [] (xB , kT ), where xB k /PB q /PB .
As in leptoproduction the tree-level calculation is most conveniently done with lightlike
directions n (PA , PB ) defined via the hadron momenta. In order to perform the transverse
integration at the level of the hadron tensor one introduces [2,39,40] a Cartesian set
{t, z , X} starting with t = q/Q. A symmetric choice for
z is z = (xA PA xB PB )/Q that
defines the CollinsSoper frame. Using n = (t z )/ 2, one has
qT
,
Q 2
qT

n+ n+
,
Q 2
n n
(72)
(73)
while as in leptoproduction one finds that for the calculation of the cross section the
orthogonal direction X differs from qT only by the irrelevant O(1/Q) terms multiplying
n , i.e., for our purposes X qT q xA PA xB PB .
5.1. Integrated DY cross section at leading order
As indicated, the inclusion of appropriate quarkgluon matrix elements make the O(g 0 )
result in Eq. (8) color-gauge-invariant leading to
(0)
(q; PA , SA ; PB , SB )
W

= d2 pT d2 kT 2 (pT + kT qT ) Tr [] (xA , pT ) [] (xB , kT ) .
(74)
This is analogous to the expression employed by Ralston and Soper [1], except that now
the correlation functions are fully color-gauge-invariant. At leading order the integration
over transverse momenta of the produced hadrons simply can be done and one obtains the
basic result for the DY process,

(0)
(q; PA , SA ; PB , SB )
d2 X W

B )
= Tr (xA ) (x
(75)
+ O(1/Q).

n n
225
5.2. Azimuthal asymmetries in DY at leading order

We consider here cross sections obtained after integration over X and explicit weighting
with X . In practice this means measurement of the azimuthal angle of the leptonic plane
with respect to that of the hadron plane or
relative to (transverse) spin azimuthal angles.

For our purposes it implies calculation of d2 X X W , which at leading order gives

(0)
(q; PA , SA ; PB , SB )
d2 X X W

[]
= Tr (xA ) [] (xB )
+
Tr
(x
)
)
(x
A
B

n n
+ O(1/Q).
n n
(76)
5.3. Integrated DY cross section at O(1/Q)

As outlined, above, one must be careful in the DY process in integrating over transverse
/ will lead to terms proportional
momenta.In this case both correlators n
/ + and n
/ dependence in a Dirac space correlator, we again use the
to q/ T /Q 2. To find the n
projectors P . We now get

1
(xA , pT ) = n
/ (xA , pT )/
nn
/+
/ +n
4

(xA , pT )

n n

1
/ (xA , pT ) + (xA , pT )/
nq/ T ,
q/ T n
2Q 2

1
B , kT )/
B , kT ) = n
/ n
/ + (x
n+n
/
(x
4

B , k T )
(x

(77)
n n

1
B , kT )/
B , kT ) + (x
/ + (x
n+q/ T .
(78)
q/ T n
2Q 2
Combining this 1/Q contribution coming from Eq. (74) with the parts from the quark
gluon diagrams one obtains

(0+1)
(q; PA , SA ; PB , SB )
d2 X W
= O(1) result (Eq. (75))

+

Tr
D (xA ) (xB ) + Tr
D (xB ) (xA )
Q 2
Q 2

+
1
1
B ) Tr [] (xA ) (x
B)
+ Tr
[] (xA ) (x
2
2
Q 2
Q 2

+

1
1

[]
[]
(xB ) (xA ) + Tr (xB ) (xA )

Tr
2
2
Q 2
Q 2
+ ( ) ,
(79)
226
(a)
(b)
Fig. 7. Quarkquark (a) and one of the quarkquarkgluon (b) correlators in tree-level diagrams for back-to-back
jet production in electronpositron annihilation.
Section 7).
6. Back-to-back jet production in electronpositron annihilation

Also for 2-particle inclusive electronpositron annihilation we have a quite similar
procedure. The calculation involves two soft fragmentation parts and the creation of a
quarkantiquark pair. We will discuss only the case of creation from a (timelike) photon.
The handbag diagram is given in Fig. 7(a) and an example of a diagram involving an
additional gluon in Fig. 7(b).
The calculation of this tensor in a diagrammatic expansion proceeds as in the case of
leptoproduction and gives
W (q; P1 , S1 ; P2 , S2 )

= d4 p d4 k 4 (p + k q) Tr (p)
(k)

k/ + p
/1 + m
A (p, p p1 ) (k)
(k + p1 )2 m2 + i)

k/ + p
/1 + m
d4 p1 Tr
(k)
(p
p
,
p)
A
1
(k + p1 )2 m2 i)

/
p k/1 + m
(p)
(k
k
,
k)
d4 k1 Tr
1
A
(p + k1 )2 m2 + i)

/
p k/1 + m
4
A (k, k k1 ) (p) + , (80)

d k1 Tr
(p + k1 )2 m2 i)

d p1 Tr
4
227
where

d4 ip
e
0| j (0)|P2 , X P2 , X|i ( )|0,
(2)4
X
d4 d4
eip eip1 ( )
Aij (p, p p1 ; P , S) =
(2)4 (2)4
ij (p; P , S) =
(81)
0| j (0)gA ()|P2 , X P2 , X|i ( )|0
(82)
pT )).
(note that (z, pT ) = pT (z,
It turns out that the direction of the links in the fragmentation functions changes
in going from SIDIS to the annihilation process, just as the direction of the links in
distribution functions changed in going from SIDIS to DY. The A -gluons connected to
hadron 1, produce a link running along the plus direction to +, i.e., one finds transversemomentum-dependent fragmentation functions [+] (z1 , kT ), where z1 P1 /k
P1 /q . Also in the antiquark matrix element the link runs to + (along minus direction
with our choice of lightlike vectors), i.e., the inclusion of A+ gluons lead to the matrix
element [+] (z2 , pT ), where z2 P2+ /p+ P2+ /q + .
Again, to perform transverse integrations at the levelof the hadron tensor one switches
from n (P1 , P2 ) directions to directions n = (t z )/ 2 with t = q/Q. In order to treat
both 1-particle and 2-particle inclusive annihilation it is convenient [20] to fix z via one
hadron momentum, for which we will choose P2 , i.e., z = (q 2P2 /z2 )/Q. The relation
between n and n , then is the same as in leptoproduction. The orthogonal direction
determining the azimuthal angle of the hadron production plane, then is X = P1 /z1
which for our purposes equals X qT q P1 /z1 P2 /z2 .
6.1. Integrated annihilation cross section at leading order
The inclusion of appropriate quarkgluon matrix elements make the O(g 0 ) result in
Eq. (80) color-gauge-invariant,
(0)
W
(q; P1, S1 ; P2 , S2 )

= d2 pT d2 kT 2 (pT + kT qT ) Tr [+] (z2 , pT ) [+] (z1 , kT ) .
(83)
At leading order the integration over transverse momenta of the produced hadrons can be
performed and one obtains the basic result

(0)
2 ) (z1 )
d2 X W
(84)
(q; P1 , S1 ; P2 , S2 ) = Tr (z
+ O(1/Q).

n n
2) = n
By taking the free (massless) quark result (z
/ + (1 z2 ) we obtain the 1-particle
inclusive result (qT = 0),

(0)
W
(85)
(q; Ph , Sh ) = Tr (zh ) + O(1/Q).
228
6.2. Azimuthal asymmetries in the annihilation process at leading order

We consider
here cross sections obtained after integration over X and explicit weighting
with X , i.e., d2 X X W , which at leading order gives

(0)
d2 X X W
(q; P1 , S1 ; P2 , S2 )

[+]
2 ) [+] (z1 )
= Tr (z
Tr
(z
)
(z
)

2

+ O(1/Q).
n n
n n
(86)
2) = n
By taking (z
/ + (1 z2 ) and [+]
(z2 ) = 0, we get the weighted e+ e
h(Ph ) + X hadron tensor

(0)
d2 X X W
(q; Ph , Sh ; Pj ) = Tr [+]
(zh )
n n
jet(Pj )+
+ O(1/Q).
(87)
6.3. Integrated annihilation cross section at O(1/Q)

As before, one must be careful in integrating over transverse momenta
at O(1/Q). In
particular the correlator n
/ will lead to terms proportional to q/ T /Q 2. We obtain

1
/ + (z1 , kT )/
n+n
/
(z1 , kT ) = n
/ n
4

1
qT n
/ + z1 , kT ) + (z1 , kT )/
n+q/ T , (88)
(z1 , kT ) n n (/
Q 2
which in analogy to leptoproduction leads to the full, integrated annihilation cross section
up to O(1/Q),

(0+1)
(q; P1 , S1 ; P2 , S2 )
d2 X W
= O(1) result (Eq. (84))

+

2)
D (z2 ) (z1 ) Tr
D (z1 ) (z
+ Tr
Q 2
Q 2
+

[+]
[+]
+ Tr
(z1 ) (z2 ) Tr (z1 ) (z2 )
Q 2
Q 2
+ ( ) ,
(89)
Section 7).
With the choice of t and z , the 1-particle inclusive result is found by taking (z1 ) =
(z1 ) = D (z1 ) = 0. After a (for ease of comparison) change of plus
n
/ (1 z1 ), [+]
minus and taking the crossed (particle antiparticle, ) term we obtain

W (q; Ph , Sh ) = O(1) result (Eq. (85))
+

+

D (zh ) Tr
D (zh ) .
Tr
Q 2
Q 2
(90)
229
7. Time-reversal properties
In order to discuss the possible relations between the expressions for the various
processes given in the previous three sections, we first discuss the time-reversal properties
of the correlation functions involved. After that we will discuss the parametrizations, such
that we are able to address the Sivers effect, the Collins effect and the QiuSterman
mechanism (gluonic poles), explicitly.
7.1. Distribution functions
Hermiticity, parity and time-reversal invariance yield conditions for the correlator ,
that constrain its parametrization,
(p; P , S) = 0 (p; P , S)0
[Hermiticity],
(91)
0
(p, P , S) = 0 (p;
P , S)
[Parity],
(92)
(p; P , S) = (i5C)(p;
P , S)(i
5 C)
[Time reversal],
(93)
where C = i 2 0 , i5C = i 1 3 and p = (p0 , p). Similar conditions arise for

the fragmentation matrix elements. Including link operators and for quarkgluon matrix
elements slightly different conditions apply. For the gauge link one has
U[a,
] = U[,a] ,
PU[a, ] P = U[a,
] ,
T U[a, ] T = U[a,
] ,
(94)
for which we used A = A , PA ( )P = A ( ) and T A ( )T = A ( ). This means

that the space-reversed (time-reversed) correlation function has a different link structure,
namely, a link running from a (a),
respectively. However, if the common point is defined
with respect to the two fields in the matrix element, no problem arises. For example, the
straight line link with path z (s) = (1 s)0 + s gives a path z after applying parity,
but after a change of variables one ends up with the same path; similarly for time reversal.
For the transverse-momentum-dependent functions with links running along the minus
direction connected at infinity, the situation is different. The point = is defined
by n+ = , which after parity transforms into the point n + = , but after timereversal transforms into the point n + = . As a consequence, one finds that for the
p -integrated functions
[+] (x, pT ) = 0 [+] (x, pT )0 ,
[+]
(x, pT ) = 0
[+]
(x, pT )0 ,
[+] (x, pT ) = (i5C) [] (x, pT )(i5 C),
(95)
(96)
(97)
where [] is defined with the link running via = , referred to as timelike

distribution in Fig. 5. Concluding, in the parametrization of [+] (x, pT ) time reversal
does not pose constraints. Application of this operation transforms [] into [+] and
vice versa. T-odd quantities will be defined as the ones that vanish when [] = [+] .
230
Accounting for the transformation in Dirac space and the sign change of pT , we define
2 [T-even] (x, pT ) [+] (x, pT ) + (i5 C) [+] (x, pT )(i5 C)
= [+] (x, pT ) + [] (x, pT ),
2
[T-odd]
(x, pT )
[+]
[+]
(x, pT ) (i5 C)
(x, pT )
[]
[+]
(98)
(x, pT )(i5 C)
(x, pT ).
(99)
Note that the name T-odd does not imply a violation of time-reversal invariance. For
the integrated distributions the different links ([]) merge into one, (x) = [+] (x) =
[] (x) and (x) is T-even. For transverse-momentum-dependent correlations the sum
of [+] and [] , i.e., (x), is T-even, while the difference, related to G

(x, x) (see
Eq. (42)), is T-odd. Summarizing, we have
[] (x, pT ) = [T-even] (x, pT ) [T-odd] (x, pT ),
(100)
and for the integrated and weighted distribution correlators

(x) = [T-even] (x),
[] (x) = [T-even] (x) [T-odd] (x, x).
(101)
(102)
These results imply that T-odd distribution functions, e.g., the Sivers effect appearing
in single spin azimuthal asymmetries in leptoproduction ([+] ) and in DrellYan
scattering ([] ) are opposite in sign [8]. As shown in Ref. [29] the T-odd functions
can be considered as imaginary parts in a helicity matrix representation, leading to the
representation in Fig. 8(a). The behavior under time reversal of gluonic matrix elements,
like A , D and G , can also be studied separately. It turns out that the quantity
(x, x) is T-odd. The latter can also be seen in A+ = 0 gauge.
D (x) is T-even, while G
(x, x) =
Using relation (A.10) in the matrix elements in Eqs. (29) and (30) yields 2G
A() (x, x) A() (x, x), i.e., the gluonic pole matrix element is the difference of the
boundary terms that transform into each other under time reversal.
7.2. Fragmentation functions
Constraints on the correlator come from hermiticity, parity and time-reversal
invariance. The essential difference with distribution functions is that time-reversal
transforms the out-states in the definition of fragmentation matrix elements into in-states.
Taking this into account and explicitly adding subscripts in and out, one obtains the
conditions,
[]
[]
out (z, kT ) = 0 out (z, kT )0 ,
[]
[]
out (z, kT ) = 0 out (z, kT )0 ,
[+]
[]
out (z, kT ) = (i5 C)in (z, kT )(i5 C),
(103)
(104)
(105)
where [] are the spacelike/timelike fragmentation functions, illustrated in Fig. 5.

Defining O and FSI as sum and differences of matrix elements with out and in-states
231
Table 2
Summary of time-reversal allowed (yes) and forbidden (no) parts in integrated or weighted distribution and
fragmentation correlators
D
A
G
T-even
T-odd
yes
yes
yes
yes
no
no
no
no
no
yes
D
A
G
T-even
T-odd
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
O
O
D O
A O
G O
T-even
T-odd
yes
yes
yes
yes
no
no
no
no
no
yes
FSI
FSI
D FSI
A FSI
G FSI
T-even
T-odd
no
no
no
no
yes
yes
yes
yes
yes
no
respectively, one has

-even] (z, k ) = [+] (z, k ) + [] (z, k ) + [+] (z, k ) + [] (z, k ), (106)
4[T
T
T
T
T
T
out
out
O
in
in
[T-odd]
[+]
[]
[+]
[]
4O
(z, kT ) = out (z, kT ) out (z, kT ) + in (z, kT ) in (z, kT ), (107)
[T-odd]
4
(z, k ) = [+] (z, k ) + [] (z, k ) [+] (z, k ) [] (z, k ), (108)
T
T
T
T
T
out
out
FSI
in
in
[T-even]
[+]
[]
[+]
[]
(z, kT ) = out (z, kT ) out (z, kT ) in (z, kT ) + in (z, kT ),
4FSI
(109)
which implies
[T-even]
-odd] (z, k )
(z, kT ) + [T
[]
T
out (z, kT ) = O
FSI
[T-odd]

[T-even]
O
(z, kT ) + FSI
(z, kT ) ,
and for the integrated and weighted fragmentation functions
-even] (z) + [T-odd] (z),
out (z) = [T
O
FSI

[T-even]
-odd] (z)
[]
(z)
=
(z)
+
[T
out
O
FSI

(110)
(111)
out (z)
-odd] (z, z) + [T-even] (z, z) .

[T
GO
GFSI

(112)
Gout (z,z)
The essential difference with the distribution functions is that for the fragmentation
functions the differences between in and out states become relevant. These constitute final
state interactions within the soft part (here labelled by FSI), decoupled from the quark
and gluon operators that make the connection to the hard scattering part. The effects arising
from the difference between [+] and [] are labelled by a subscript O. Note that in the
literature [7,15,30] this is also referred to as initial or final state interactions, depending on
the process under consideration. The behavior of the various correlators is given in Table 2.
Therefore, in contrast to the distribution functions, and G contain T-even and T-odd
parts. Also the correlators D and A contain T-even and T-odd parts.
At first sight Eqs. (110) and (112) seem to imply a breaking of universality. Particular
transverse-momentum-dependent functions obtained as Dirac projections of []
out (z, kT )
give unrelated T-odd (and also T-even) results for the two possible link configurations,
indicated with superscripts [+] or []. In []

out there arise T-odd parts from FSI and
from GO , the sign in front of the latter coupled to the link structure. Thus, the T-odd
232
(a)
(b)
(c)
Fig. 8. Illustration of the role of gauge links ([+] versus []) and the role of final states ([in] versus [out]) in
T-odd effects for distribution and fragmentation functions.
(Collins) effects in pion leptoproduction and in electronpositron annihilation are a priori

not identical. The fact that a certain azimuthal asymmetry arises from a combination of
correlation functions ( and G ), however, need not imply a breaking of universality. It
is quite similar to the qT -integrated structure functions of different processes which involve
different flavor weights. It seems possible that a factorization proof can be established for
the correlation functions and G separately.
In Fig. 8 we illustrate the differences between [] and those between the in/out . For
the former one can argue that forgetting about the effect from in- and out-states one has
the same situation as for distribution functions where the T-odd parts can be considered
as imaginary parts of helicity amplitudes (shown as the projections along the vertical axis
in Fig. 8(a)). For in/out one can use the fact that in- and out-states can be obtained by
different Mller operators, allowing a connection of in and out via a unitary operation,
illustrated in Fig. 8(b). The combined effect is illustrated in Fig. 8(c). It shows that the
[]
+
T-odd parts of [+]
out in e e and out in SIDIS are in general not equal in magnitude.
Actually, not only the T-odd effects acquire contributions from both terms in G ,
but also the T-even effects, hence affecting all comparisons of azimuthal asymmetries in
leptoproduction and electronpositron annihilation.
7.3. Parametrization of quark and quarkgluon correlation functions
Here we will discuss the parametrizations for transverse-momentum-dependent distribution functions and fragmentation functions. The parametrizations of (x, pT ) and
(z, kT ) consistent with the conditions imposed by hermiticity and parity, including the
parts proportional to (M/P + )0 and (M/P + )1 are given in Appendix B [6,19]. In principle, the functions could differ depending on the gauge link structure for both distribution
and fragmentation functions and also on the O/FSI characterization for fragmentation functions.
We will first discuss the parametrizations for the transverse momentum integrated and
then the once-weighted functions (transverse moments). As explained in Section 4.6, we
do not address twice-weighted functions (relevant for kT broadening and the average kT2 in
jets) in view of potential problems with factorization.
The result for the correlator after integration is

1
S T , n/ + ]
5 [/
(x) =
f1 (x)/
n+ + SL g1 (x)5n
/ + + h1 (x)
2
2

M
5 [/
n+ , n/ ]
+
e(x) + gT (x)5S/ T + SL hL (x)
.
2P +
2
233
(113)
For the distribution functions the T-odd part vanishes, i.e., eL (x) = fT (x) = h(x) = 0. For
the parametrization of fragmentation functions one has both T-even and T-odd functions,
i.e., the functions EL (z), DT (z) and H (z) appear. Explicitly,
S hT , n/ ]
5 [/
n + ShL zG1 (z)5n
/ + + zH1 (z)
out (z) = zD1 (z)/
2

n , n/ + ]
5 [/
Mh
+ zE(z) + zGT (z)5S/ hT + ShL zHL (z)
2
Ph

i[/
n , n/ + ]
.
2
(114)
[T-even]
The T-even functions are O-type, the T-odd functions are FSI-type, since out
(z) =
[T-even]
[T-odd]
[T-odd]
(z) and
(x) =
(z), while the link structure does not play a role.
ShL zEL (z)i5 + zDT (z))T ShT + zH (z)
out
FSI
T-odd functions only appear in sub-leading parts of the cross sections.

Next we consider the pT -weighted results referred to as transverse moments. Restricting
ourselves to the leading (M/P + )0 part, we write for the correlators [] (Eq. (38))

M (1)[]
5 [ , n/ + ]
(1)[]
/ + SL h1L
(x)
[] (x) =
g1T (x)ST 5n
2
2

i[ , n/ + ]
(1)[]
,
(x)) n+ ST h(1)[]
(x)
f1T
(115)
1
2
where we have defined p2T /2M 2 -moments (transverse moments) as

p2
(1)
g1T (x) = d2 pT T 2 g1T (x, p2T ),
2M
(116)
and similarly for the other functions.

For the transverse moments, the average of the correlators in Eq. (43) is T-even, i.e.,

M (1)
5[ , n/ + ]
(1)
/ + SL h1L (x)
g (x)ST 5n
,
(x) =
(117)
2 1T
2
(x, x) in Eq. (42) is T-odd. Writing
while the gluonic pole contribution G

M
i[ , n/ + ]
(1)
f1T
,
(x, x) =
(x)) n+ ST h (1)
(x)
G
1
2
2
(118)
we have the relations

(1)[]
(1)
g1T
(x) = g1T
(x),
(119)
234

(1)[]
h1L
(1)
(x) = h1L (x),
(120)
(1)[]
(1)
f1T
(x) = f1T (x),
(1)[]
(1)
h1
(x) = h 1 (x).
(121)
(122)
These results show that azimuthal spin asymmetries involving the distribution functions
(1)
(1)
(1)
(1)
g1T and h1L are process-independent, while those involving the functions f1T and h1
change sign. Eq. (121) represents the explicit connection between the Sivers effect (l.h.s.)
and the QiuSterman effect (r.h.s., cf. Ref. [26]); Eq. (122) is its chiral-odd counterpart.
(1)
We note that due to the presence of gluonic pole effects the evolution equations of f1T
(1)
and h1 [41] need to be reconsidered. These functions originate solely from gluonic pole
effects.
contains both
For the fragmentation functions, not only the parametrization for []
T-even and T-odd parts, but also the average contains T-even and T-odd parts, i.e.,

5 [ , n/ ]
(1)
(1)
(z) = Mh zG1T (z)ST 5n

/ ShL zH1L (z)
2

i[ , n/ ]
(1)
(1)

.
zD1T (z)) n ShT zH1 (z)
(123)
2
For the gluonic pole contribution one obtains
(1)(z) 5 [ , n/ + ]
(1)(z)S 5n
G (z, z) = Mh zG
/
S
z
H
hL
hT
1T
1L
2

i[ , n/ ]
(1)
,
zD 1T
(z)) n ShT zH 1(1) (z)
2
(124)
and we obtain for the transverse moment in []

,
(1)[]
G1T
(z),
(z) = G1T (z) G
1T
(1)[]
H1L
(1)
(1)
(1)
(1)
(z) = H1L (z) H 1L (z),
(1)[]
(1)
(1)
D1T
(z) = D1T (z) D 1T (z),
(1)[]
(1)
(1)
H1
(z) = H1 (z) H 1 (z).
(125)
(126)
(127)
(128)
The occurrence of out-states in the fragmentation matrix elements is responsible for

(1) and H (1) and the T-odd functions D (1)
the appearance of the T-even functions G
1T
1L
1T
(1)
and H1 . We note that these results differ from Ref. [19]. For example, Eq. (128)
shows the explicit forms of the Collins function in e+ e (plus sign) and SIDIS (minus
sign), respectively. Furthermore, the above results imply that the evolution equations of
(1)
(1)
(1)
G(1)
[41] also need to be reconsidered.
1T , H1L , D1T and H1
For functions weighted twice with a transverse momentum in , one needs higher
(2)
transverse moments of the functions, such as h1T (x). Relations for these functions
involve not only twist-three, but also twist-four parts in the correlators. This includes the
simple pT2 average in 2 (x). As mentioned above, for these functions we do not expect a
235
simple process dependence as given for the once-weighted results, but this requires further
investigation.
In Refs. [26,42] parametrizations have been given for two-argument quarkgluon
correlations. As shown in Eq. (41) one finds after integration correlation functions with
D (x) at the same point, for which the QCD equations of motion can be used. These
relate D (x) (see Eq. (41)) and also 0 D (x)0 to (x); explicitly,

M
m
m
5 [ , n/ + ]
D
(x) =
/ + + SL xhL (x) g1 (x)
xgT (x) h1 (x) ST 5n
2
M
M
4

[ , n/ + ]
m
xe(x) f1 (x)
(129)
,
M
4
which contains only T-even functions. For fragmentation functions one also finds T-odd
functions,

m
zH1 (z) ShT

5 n
/
D (z) = Mh GT (z)
Mh

m
5 [ , n/ ]
+ ShL HL (z)
zG1 (z)
Mh
4

[ , n/ ]
m
+ DT (z)) n ShT
zD1 (z)
E(z)
Mh
4

i[ , n/ ]
i5 [ , n/ ]
+ H (z)
(130)
+ EL (z)
.
4
4
Besides the relations resulting from the equations of motion, also Lorentz invariance
may lead to relations between correlators [19,41,43]. As can be seen from the explicit
treatment in Ref. [44], these relations are derived from the Lorentz structure of nonintegrated quarkquark correlators as in Eq. (9). At present it is not clear how matrix
elements of AT fields at infinity and hence the link structure play a role in these relations
(see also Ref. [45]).
8. Summary and conclusions

In this paper we have analyzed transverse-momentum-dependent distribution and
fragmentation functions appearing in several hard processes in which at least two hadrons
are involved in initial and/or final state. In these processes one has besides a hard scale
Q, a non-collinearity qT which is characterized by a hadronic scale QT and an azimuthal
angle. We have shown explicitly, using the results of Belitsky et al. [14] how quarkquark
gluon matrix elements lead to fully color-gauge-invariant definitions of the correlation
functions that appear in leading and first sub-leading order in 1/Q in the hadron tensor of
hard processes. The gluon fields appear in the gauge link connecting the quark fields. The
transverse gluons needed in the gauge link for transverse-momentum-dependent functions
involve gluon fields at lightlike infinity. The fact that the gluonic effects can be cast
into (conjugate) links attached to the two (conjugate) quark fields still allows for an
interpretation of these functions as probability densities.
236
The structure of the gauge links in hard processes is not always the same. In particular
the gauge links in distribution functions in SIDIS and the DY process run in opposite
lightlike directions indicated with indices . The two different correlators are connected
via a time-reversal operation. Similarly, the gauge links in fragmentation functions in
SIDIS and electronpositron annihilation run in opposite lightlike directions. At leading
order, the difference between correlators vanishes upon integration over transverse
momenta. In qT -weighted cross sections, one finds correlators weighted with transverse
momentum (transverse moments), which are dependent on the () link structure. The
same quantities appear in subleading integrated cross sections. The difference between
transverse moments with different () link structure corresponds to a (color-gaugeinvariant) gluonic pole matrix element, the word pole referring to the fact that one deals
with a zero-momentum gluon field. This matrix element appears in different processes as
the QiuSterman effect.
Considering the behavior under time reversal for distribution functions, it turns out
that the gluonic pole contributions coincide with T-odd effects, leading to single spin
asymmetries, like the Sivers effect. This establishes the direct connection between the
Sivers effect and the QiuSterman effect. Since for distribution functions gluonic poles
are the only source of T-odd effects, one finds that these effects have opposite signs in
SIDIS and the DY process. For fragmentation functions, T-odd effects leading to single
spin asymmetries, like the Collins effect in pion leptoproduction, arise not only from the
gluonic pole contribution, but also from final state interactions. The latter are purely soft
interactions in the fragmentation part and has the same sign in different processes. Hence
the T-odd effects in SIDIS and electronpositron annihilation are not connected by a simple
sign relation. This is not a breaking of universality, but rather the appearance of different
combinations of fragmentation functions in different processes. The T-odd effects will
not only appear in Collins asymmetries in SIDIS or electronpositron annihilation, but
likely also in other processes. We emphasize the importance of an analysis of the link
structure in processes like pp X. We note that also for distribution functions Todd contributions with the same sign in different processes could arise if time reversal is
realized in a nonstandard way [46]. This would spoil the simple sign switches in Eqs. (121)
and (122).
Gluonic pole contributions appear in the azimuthal asymmetries, in the processes that
we have considered, with a particular sign. This affects not only T-odd, but (in case
of fragmentation) also T-even azimuthal asymmetries. The transverse moments with a
different link structure differ by effective twist-three functions. The consequences of
these gluonic pole contributions on the evolution of the transverse moments have not been
considered. Given the known operator structure of the contributions, however, such a study
ought to be doable.
Although in trying to model distribution or fragmentation functions [44,47] one is
always stuck with the problem of evolution, modelling has been proven useful to illustrate
[]
several effects [7,15,48,49], such as the sign change in the Sivers functions f1T
and the
[]
sign behavior for the Collins function H1 . As seen from Eq. (128), the latter can have
two contributions with different signs. It is not clear whether both contributions are present
in the models studied.
237
Our general analysis of the various correlators in leading and subleading order single
spin and azimuthal asymmetries in hard processes may help to analyze the results of
the first generation of experiments that presently are being performed or planned by, for
instance, HERMES (DESY), COMPASS (CERN), BELLE (KEK) and the results that are
obtained by looking at existing LEP data. We hope to see the emergence of a coherent
picture of these asymmetries.
Acknowledgements
We acknowledge discussions with A. Bacchetta, A. Belitsky, S.J. Brodsky, J.C. Collins,
D.S. Hwang, P. Hoyer, X. Ji, A. Metz, and J.G. Milhano. The research of D.B. has been
made possible by financial support from the Royal Netherlands Academy of Arts and
Sciences.
Appendix A. Transverse gluon fields and the field strength tensor

In lightcone gauge A+ = 0, the relation between AT and G+ becomes
+
x , AT (x) = G+ (x),
(A.1)
AT
which can be inverted to yield

in terms of a boundary term and an integral along the
minus direction with G+ in the integrand. Without gauge choice we have

igG+ (x) = iD + (x), iD (x) = iD + (x), gAT (x) ig x , A+ (x)
(A.2)
as our starting point. Next we multiply from left and right with link operators U[a,x]
and
U[x,a] , respectively, built from A+ fields, running along the minus direction. They are
denoted by

x

U[a,x]
= P exp ig d A+ , x + , xT ,

(A.3)
and satisfy
x+ U[a,x]
= U[a,x]
D + (x).
(A.4)
We then obtain
G+ (x)U[x,a]
igU[a,x]

igU[a,x]
x , A+ (x) U[x,a]
= U[a,x] iD + (x)U[x,a]
, U[a,x]
gAT (x)U[x,a]

igU[a,x]
x , A+ (x) U[x,a]
= ix+ , U[a,x]
gAT (x)U[x,a]
.
(A.5)
Thus we find

+

+
= U[a,x]
G (x) + x , A+ (x) U[x,a]
,
, U[a,x] AT (x)U[x,a]
(A.6)
238
which is the relation needed to express the transverse gluon fields in terms of the field
strength. In particular, one has
U[,
] AT ( )U[,] AT ( )

=
+

G () + T , A+ ( ) U[,]
d U[,]
,
(A.7)
or in A+ = 0 gauge
AT ( ) AT ( ) =
d G+ ().
(A.8)
By considering Eq. (A.7) for = , one arrives at
U[,]
AT ( ) AT ( )U[,]

=
+

G () + T , A+ ( ) U[,]
d U[,]
,
(A.9)
or in A+ = 0 gauge
AT ( ) AT ( ) =
d G+ ().
(A.10)
Appendix B. Parametrizations of transverse-momentum-dependent functions

The parametrization of (x, pT ) for a spin 0 or spin 1/2 target, consistent with the
conditions imposed by hermiticity and parity, including the parts proportional to (M/P + )0
and (M/P + )1 is given by
(x, pT ) =

1
/ + + g1s (x, pT )5n
/+
f1 x, p2T n
2

S T , n/ + ]
pT , n/ + ]
5 [/
5 [/
+ h
+ h1T x, p2T
1s (x, p T )
2
2M

n p S

pT , n/ + ]
+ T T
2 i[/
+ h
+ f1T
x, p2T
x,
p
T
1
M
2M

p

M
/T
+ gT x, p2T 5S/ T
e x, p2T + f x, p2T
+
2P
M

ST ,p
/T
/T ]
5p
2 5 [/
+ h
+ gs (x, pT )
T x, p T
M
2M

n+ , n/ ]
5 [/
fT x, p2T )T ST
+ hs (x, pT )
2

) pT
es (x, pT )i5
SL fL x, p2T T
M

i[/
n+ , n/ ]
.
+ h x, p2T
2
Here the spin vector is defined

+

P
P
S = SL
n+
n + ST
M
M
239
(B.1)
(B.2)
(S = 0 for spin 0), and we have used the shorthand notation

(pT S T )
,
g1s (x, pT ) SL g1L x, p2T + g1T x, p2T
M
(B.3)
and similarly for h

1s , gs and hs . We note that all noncontracted pT -dependence (including
appearance of dot products like pT ST ) is treated explicitly, leaving functions depending
on pT2 , which is important in distinguishing T-even and T-odd behavior. The structures
multiplying the twist-two functions f1 , g1s , h1T , h

1s or the twist-three functions e, f ,

gT , gs , hT , hs are T-even (satisfying Eq. (97)). The structures multiplying the twist-two
, h or the twist-three functions f , f , e , h are T-odd (satisfying Eq. (97)
functions f1T
T
s
L
1
with an additional minus sign).
As notation in the parametrizations for fragmentation we employ capital letters, to be
precise
n + zG1s (z, zk T )5n

/
(z, kT ) = zD1 (z, zk T )/
S hT , n/ ]
5 [/
5 [/kT , n/ ]
+ zH1s
+ zH1T (z, zk T )
(z, zk T )
2
2Mh
n k S
)
i[/kT , n/ ]
T hT
+ zD1T
(z, zk T )
+ zH1 (z, zk T )
Mh
2Mh

k
/
Mh
T
+ zGT (z, zk T )5S/ hT
+ zE(z, zk T ) + zD (z, zk T )
Mh
Ph
S hT , k/T ]
5k/T
5 [/
+ zG
+ zHT (z, zk T )
s (z, zk T )
Mh
2Mh
n , n/ + ]
5 [/
+ zDT (z, zk T ))T ShT

+ zHs (z, zk T )
2
) k T
+ ShL zDL (z, zk T ) T
zEs (z, zk T )i5
Mh

i[/
n , n/ + ]
.
+ zH (z, zkT )
(B.4)
2
240
Here the spin vector is defined

P+
P
Sh = ShL h n h n+ + ShT
Mh
M
(B.5)
(Sh = 0 for spin 0), and we have used the shorthand notation G1s , etc.,
G1s (z, zk T ) = ShL G1L (z, zk T ) + G1T (z, zk T )
(k T S hT )
.
Mh
(B.6)
The second argument of the fragmentation functions is chosen to be kT = zkT , which is
the transverse momentum of the produced hadron with respect to the quark in a frame in
which the quark does not have transverse momentum. In fact, the functions only depend
on kT 2 . As for the distribution functions a division can be made into T-even functions (D1 ,
, E, D , G , G , H , H ) and T-odd functions (D , H , D , D , E ,
G1s , H1T , H1s
s
T
s
s
T
T
L
1T
1
H ).
References
[1] J.P. Ralston, D.E. Soper, Nucl. Phys. B 152 (1979) 109.
[2] R.D. Tangerman, P.J. Mulders, Phys. Rev. D 51 (1995) 3357.
[3] D. Sivers, Phys. Rev. D 41 (1990) 83;
D. Sivers, Phys. Rev. D 43 (1991) 261.
[4] J.C. Collins, Nucl. Phys. B 396 (1993) 161.
[5] M. Anselmino, M. Boglione, F. Murgia, Phys. Lett. B 362 (1995) 165;
M. Anselmino, F. Murgia, Phys. Lett. B 442 (1998) 470;
M. Anselmino, U. DAlesio, F. Murgia, Phys. Rev. D 67 (2003) 074010.
[6] D. Boer, P.J. Mulders, Phys. Rev. D 57 (1998) 5780.
[7] S.J. Brodsky, D.S. Hwang, I. Schmidt, Phys. Lett. B 530 (2002) 99.
[8] J.C. Collins, Phys. Lett. B 536 (2002) 43.
[9] J.C. Collins, D.E. Soper, Nucl. Phys. B 194 (1982) 445.
[10] J.C. Collins, D.E. Soper, G. Sterman, Phys. Lett. B 109 (1982) 388;
J.C. Collins, D.E. Soper, G. Sterman, Nucl. Phys. B 223 (1983) 381.
[11] A.V. Efremov, A.V. Radyushkin, Theor. Math. Phys. 44 (1981) 774.
[12] D. Boer, P.J. Mulders, Nucl. Phys. B 569 (2000) 505.
[13] X. Ji, F. Yuan, Phys. Lett. B 543 (2002) 66.
[14] A.V. Belitsky, X. Ji, F. Yuan, Nucl. Phys. B 656 (2003) 165.
[15] S.J. Brodsky, D.S. Hwang, I. Schmidt, Nucl. Phys. B 642 (2002) 344.
[16] D. Boer, Nucl. Phys. B 603 (2001) 195.
[17] R.K. Ellis, W. Furmanski, R. Petronzio, Nucl. Phys. B 212 (1983) 29;
R.K. Ellis, W. Furmanski, R. Petronzio, Nucl. Phys. B 207 (1982) 1.
[18] A.V. Efremov, O.V. Teryaev, Sov. J. Nucl. Phys. 39 (1984) 962.
[19] P.J. Mulders, R.D. Tangerman, Nucl. Phys. B 461 (1996) 197;
P.J. Mulders, R.D. Tangerman, Nucl. Phys. B 484 (1997) 538, Erratum.
[20] D. Boer, R. Jakob, P.J. Mulders, Nucl. Phys. B 504 (1997) 345;
D. Boer, R. Jakob, P.J. Mulders, Phys. Lett. B 424 (1998) 143.
[21] D. Boer, S.J. Brodsky, D.S. Hwang, Phys. Rev. D 67 (2003) 054003.
[22] L.P. Gamberg, G.R. Goldstein, K.A. Oganessyan, Phys. Rev. D 67 (2003) 071504.
[23] R.L. Jaffe, Nucl. Phys. B 229 (1983) 205.
[24] M. Diehl, T. Gousset, Phys. Lett. B 428 (1998) 359.
[25] J. Levelt, P.J. Mulders, Phys. Lett. B 338 (1994) 357.
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
D. Boer, P.J. Mulders, O.V. Teryaev, Phys. Rev. D 57 (1998) 3057.

X. Ji, J.P. Ma, F. Yuan, Nucl. Phys. B 652 (2003) 383.
J. Soffer, Phys. Rev. Lett. 74 (1995) 1292.
A. Bacchetta, M. Boglione, A. Henneman, P.J. Mulders, Phys. Rev. Lett. 85 (2000) 712.
S.J. Brodsky, P. Hoyer, N. Marchal, S. Peigne, F. Sannino, Phys. Rev. D 65 (2002) 114025.
J. Qiu, G. Sterman, Phys. Rev. Lett. 67 (1991) 2264;
J. Qiu, G. Sterman, Nucl. Phys. B 378 (1992) 52.
N. Hammon, O. Teryaev, A. Schfer, Phys. Lett. B 390 (1997) 409.
J. Qiu, G. Sterman, Phys. Rev. D 59 (1999) 014004.
Y. Kanazawa, Y. Koike, Phys. Rev. D 64 (2001) 034019.
D. Boer, J. Qiu, Phys. Rev. D 65 (2002) 034008.
A. Kotzinian, Nucl. Phys. B 441 (1995) 234;
A.M. Kotzinian, P.J. Mulders, Phys. Lett. B 406 (1997) 373.
M. Boglione, P.J. Mulders, Phys. Rev. D 60 (1999) 054007.
R. Basu, A.J. Ramalho, G. Sterman, Nucl. Phys. B 244 (1984) 221.
J.C. Collins, D.E. Soper, Phys. Rev. D 16 (1977) 2219.
R. Meng, F.I. Olness, D.E. Soper, Nucl. Phys. B 371 (1992) 79.
A.A. Henneman, D. Boer, P.J. Mulders, Nucl. Phys. B 620 (2002) 331.
R.L. Jaffe, X. Ji, Nucl. Phys. B 375 (1992) 527.
A.P. Bukhvostov, E.A. Kuraev, L.N. Lipatov, Sov. Phys. JETP 60 (1984) 22.
R. Jakob, P.J. Mulders, J. Rodrigues, Nucl. Phys. A 626 (1997) 937.
K. Goeke, A. Metz, P.V. Pobylitsa, M.V. Polyakov, hep-ph/0302028.
M. Anselmino, V. Barone, A. Drago, F. Murgia, hep-ph/0209073.
A.V. Efremov, K. Goeke, P.V. Pobylitsa, Phys. Lett. B 488 (2000) 182;
P. Schweitzer, et al., Phys. Rev. D 64 (2001) 034013.
A. Bacchetta, R. Kundu, A. Metz, P.J. Mulders, Phys. Rev. D 65 (2002) 094021.
A. Metz, Phys. Lett. B 549 (2002) 139.
241

Three loop anomalous dimension of the second

moment of the transversity operator in the MS and
RI schemes
J.A. Gracey
Theoretical Physics Division, Department of Mathematical Sciences, University of Liverpool,
PO Box 147, Liverpool L69 3BX, UK
Received 12 May 2003; accepted 18 June 2003
Abstract
We compute the anomalous dimension of the second moment of the transversity operator,
D , at three loops in both the MS and RI schemes. As a check on the result we
also determine the O(1/Nf ) critical exponent of the nth moment of the transversity operator
in d dimensions in the large Nf expansion and determine leading order information on the n
dependence of the anomalous dimension at three and four loops in MS. In addition the RI anomalous
D , is also determined.
dimension of the non-singlet twist-2 operator,
PACS: 11.10.Gh; 12.38.Bx; 11.15.Pg
1. Introduction
Recently there has been renewed interest in the transversity distribution of the partons
in the nucleons. Introduced originally in [13] the transversity measures the difference
in probabilities of finding a quark in a transversely polarized nucleon which is polarized
parallel to that of the nucleon with that in the antiparallel polarization. Although there
has not been as large an experimental activity in this area compared with the conventional
distribution functions of deep inelastic scattering, it is possible RHIC may take transversity
data in the near future, [4]. (For recent reviews see, for instance, [5,6].) Therefore,
to improve our interpretation of the results it is necessary to extend our theoretical
E-mail address: jag@amtp.liv.ac.uk (J.A. Gracey).
doi:10.1016/S0550-3213(03)00543-1
J.A. Gracey / Nuclear Physics B 667 (2003) 242260
243
understanding to a similar level as that achieved in deep inelastic scattering. For instance,
there has been a large activity in determining the three loop anomalous dimensions of the
twist-2 unpolarized and polarized flavour non-singlet and singlet operators as a function
of the operator moment, n, [711]. These three loop results are necessary for the full two
loop renormalization group evolution of the structure functions. Exact analytic results as
a function of n are required since the Mellin transform with respect to n determine the
parton splitting functions as a function of the momentum fraction, x. Such three loop
calculations, however, require a huge increase in resources compared with the earlier one
and two loop results of [1219]. For instance, due to the presence of a large number
of Feynman diagrams with and without subgraphs, one encounters nested or harmonic
sums whose mathematics has had to be studied and developed in, for example, [20].
Moreover, this has then to be encoded in the symbolic manipulation language F ORM,
[21], in order to handle the incredibly large amounts of algebra. Whilst results for nonsinglet operators are expected soon it is clear that the same machinery can be applied to
determining the anomalous dimensions of the moments of the transversity operator as they
differ only in their -algebra structure. For instance, the transversity operator involves
the antisymmetric tensor = 12 [ , ] instead of one -matrix. Whilst an analytic
result for transversity at three loops is some way off, one can turn to the original three
loop approach of [7] and determine information on the low moments of the anomalous
dimension of the operator. Indeed in [22] the anomalous dimension of the first moment
is available at three loops in QCD, building on the two loop result for this tensor current,
is also available in QED at four
[23]. Moreover, the anomalous dimension of
loops in MS in the quenched approximation, [24]. In this article we will extend [22] by
D at three loops where D is the usual
computing the anomalous dimension of
covariant derivative. Although one motivation for the result lies in the provision of an
independent calculation which will be useful to check against the result as a function of n
when it is determined, there are several other reasons for concentrating on this one operator.
These lie in other theoretical techniques to determine information on the associated matrix
element. Using a lattice regularization one can deduce numerical estimates for matrix
elements. However, such results are necessarily in a lattice regularization scheme and
need to be matched to the standard MS renormalization scheme. Therefore, we will
not only determine the anomalous dimension at three loops in MS but also in the RI
scheme which denotes the modified regularization invariant scheme, [25]. This scheme
was introduced in [25] and developed for the problem of improving estimates for quark
masses in [26,27]. For applications to similar problems in applying the lattice to construct
matrix elements relevant in deep inelastic scattering see, for instance, [28,29]. It is also
worth drawing attention to an alternative lattice approach to constructing moments of the
parton distribution functions, [30]. This uses the Schrdinger functional technique of [31]
and is non-perturbative. More recently the RI scheme has been examined in detail in [32]
where QCD has first been renormalized at three loops before reproducing the results of
[27] at three loops for the scalar current. Hence, the RI expression for the anomalous
dimension of the tensor current was determined to the same order, [32]. By contrast to MS
the anomalous dimensions of these currents cease to be independent of the covariant gauge
parameter. However, only results are required in the Landau gauge for the lattice matching.
Nevertheless from the point of view of internal consistency in multiloop computations
244
we will perform our calculations in an arbitrary covariant gauge before deducing the
requisite conversion functions in the Landau gauge. The disadvantage of an arbitrary
gauge calculation is that the computation is algebraically more intense which is further
complicated by the fact that the presence of a covariant derivative, itself, introduces another
level of integration by parts, also slowing the procedure. Therefore, we have chosen to only
focus on the second moment. Choosing the Feynman gauge would speed the computer
algebra manipulations but would not be useful for the necessary matching which is gauge
dependent. Further, matrix elements for higher moment operators from the lattice are
currently much harder to extract.
To be more specific about the results we will report here, we will renormalize both the
D at three loops in MS and RI .
D and
(flavour non-singlet) operators
Although the anomalous dimension of the former operator is known in MS, [7,1219], we
will use it as a check on the computer algebra programmes we have written in the symbolic
manipulation language F ORM, [21], before extracting the RI anomalous dimension. Then
the Feynman rule for the operator insertion in a quark two-point function will be replaced
with that for the transversity operator. One check on both the MS results will be that
the anomalous dimensions of both operators will be independent of the covariant gauge
parameter, . The method of calculation is to apply the M INCER algorithm, [33], written
in F ORM, [34], which determines the poles in in dimensional regularization, with d =
4 2, as well as the finite part of massless three loop two-point functions. Each operator
will be inserted at zero external momentum to form a two-point function. Moreover, lattice
calculations are for the operator inserted at zero momentum so the renormalization scheme
conversion functions are required for this configuration. We have used Q GRAF, [35], to
generate the Feynman diagrams to three loops and the output has been converted to the
format used by the F ORM version of M INCER.
The paper is organised as follows. In Section 2 we introduce the method we will
use by considering the twist-2 flavour non-singlet operator and reproduce the three loop
anomalous dimension for the second moment in MS before deriving the RI scheme result.
This approach is extended in Section 3 to the case of the second moment of the transversity
operator. The conversion functions which allow one to translate from one renormalization
scheme to another for each of the operators we consider are discussed in Section 4 and
provide a partial check on our results. As an independent check on our three loop result
for the transversity we compute the large Nf leading order critical exponent for the nth
moment of the operator using the large Nf critical point technique in Section 5. Finally,
we conclude with discussion in Section 6.
2. Second moment of non-singlet twist-2 operator

As we are considering operator insertions similar to the tensor current which was studied
in [22,32], we briefly recall the RI renormalization scheme definitions. To determine the
correct finite part of the Greens function with the operator inserted we must project out
all possible components consistent with Lorentz symmetry. For instance, for the flavour
D we have in momentum space,

non-singlet operator

G
D (p) = (p) S D (0)(p)

2
(1)

=
p
/
(p)
p
+
D
d

2
p
(2)

/
+
/ p
D (p) p p p
d
245
(2.1)
where p is the momentum flowing through the external quark legs and S denotes both the
symmetrization with respect to and and that the operator is traceless.1 We could have
introduced the null vector, , to achieve this but it is not necessary here as we will in fact
D . To be more specific
reproduce the known anomalous dimension for
2
D .

d
The various components are determined through

1
2
(1)
/ G
D (p) =
(p)
tr p + p p
D
8(d 1)
d

p2
2 tr p pp
/ G
/ p
,
D (p)
d

1
2
tr
p
/
G
(p)
=
p
+
(p)
(2)

D
D
4(d 1)
d

p2
/ G
/ p
.
(d + 2) tr p pp
D (p)
d
D +
D
S D =
(2.2)
(2.3)
To renormalize (p)[S D ](0)(p)

in our symbolic manipulation approach we
follow the method of [36]. There the bare Greens function is computed in terms of the bare
coupling constant and covariant gauge parameter. Renormalized variables are introduced
through the usual renormalization constant definitions such as g0 = Zg g where g is
the coupling constant appearing in the covariant derivative D = + igT a Aa , is the
renormalization scale introduced to ensure the coupling constant remains dimensionless
in d dimensions and T a are the generators of the colour group whose structure functions
are f abc . The renormalization constant associated with the operator, ZO , is deduced by
ensuring that the Greens function is finite after multiplying by ZO Z where the quark
wave function renormalization constant, Z , arises from the external quark legs. For the
RI since this renormalization
RI scheme it is the first component which determines ZO
constant is defined by, [25,32],
RI

(1)

lim ZRI Z
(2.4)
D D (p) p 2 =2 = 1.
0
In other words, unlike the MS scheme in addition to the poles in the O(1) piece
is absorbed into the renormalization constant. Though it is important to note that this
1 The symmetrization and traceless symbol, S, is understood in all our subscripts.
246
renormalization constant will also render the other components finite but not necessarily
zero or unity.
Having summarized our method of calculation, we now record our results for the
D . We find
renormalization of
MS
Z
D

8
4
1
= 1 + CF a + (4TF Nf + 8CF 11CA ) 2
3
9

4
1
+ (47CA 14CF 16TF Nf ) CF a 2
27

8
121CA2 132CA CF 88CA TF Nf + 32CF2

+
81
1
8
+ 48CF TF Nf + 16TF2 Nf2 3 +

718CA CF 823CA2 + 544CA TF Nf
243

1
168CF2 140CF TF Nf 64TF2 Nf2 2

2
+
648 (3) + 2615 CA 1944(3) + 1066 CA CF
729

1296 (3) + 782 CA TF Nf + 1296(3) 70 CF2

1
CF a 3 + O a 4
+ 1296 (3) 853 CF TF Nf 112TF2 Nf2
(2.5)

which is clearly gauge independent and agrees exactly with the result of [7]. Throughout
we will present each renormalization constant as an aid to interested readers who wish
to perform the check that the double and triple poles in can be derived from the simple
poles of the previous loop order which follows from the fact that MS is a mass independent
renormalization scheme. Further, in (2.5) (n) is the Riemann zeta function and we have set
a = g 2 /(16 2 ), T a T a = CF I , f acd f bcd = CA ab and Tr(T a T b ) = TF ab . Consequently,
having checked that our programming reproduces known results it is elementary to change
the renormalization scheme and deduce

1
8
RI
+ (9 + 31) CF a
Z
D = 1 +
3 9

1
4
4
+ (4TF Nf + 8CF 11CA ) 2 +
47CA + (48 + 18)CF
9
27

1
1
+
16TF Nf
81 3 + 486 2 324(3) + 2439
324

3564 (3) + 12808 CA + 162 2 + 630 + 2592(3) 456 CF

(720 + 5336)TF Nf a 2

8
121CA2 132CA CF 88CA TF Nf + 32CF2 + 48CF TF Nf

+
81
247
1
4
(413 297)CA CF 1646CA2

+
3

243
+ 1088CA TF Nf + (216 + 408)CF2 + (108 + 92)CF TF Nf
1
2
243 3 + 1458 2 972(3) + 9855

128TF2 Nf2 2 +

729

18468 (3) + 42902 CA CF + 2592(3) + 10460 CA2 5184(3)

+ 3182 CA TF Nf + 486 2 + 1134 4252 + 12960(3) CF2
1

448TF2 Nf2 + 5184(3) 22396 3024 CF TF Nf

1
5
4
3
8748 + 83106 17496(3) + 493776 3
+
69984
+ 1659204 2 96228(3) 2 2523312(3) + 174960(5)
+ 16TF2 Nf2
+ 7952229 11944044(3) + 746496(4) + 524880(5)

+ 38226589 CA2 + 17496 4 + 95256 3 + 151632(3) 2

+ 416016 2 + 346032(3) + 466560(5) + 921564

4914432 (3) 2239488(4) + 8864640(5) + 3993332 CA CF
77760 3 + 466560 2 124416(3) + 4091040

369792 (3) + 1492992(4) + 24752896 CA TF Nf
155520 2 622080(3) + 1937088 3234816(3)

1492992 (4) + 9980032 CF TF Nf

+ 345600 + 221184(3) + 3391744 TF2 Nf2
+ 17496 3 + 289656 2 373248(3) 2 715392(3)

+ 879336 + 10737792(3) + 1492992(4)

9331200 (5) 3848760 CF2 CF a 3 + O a 4 .
(2.6)
It is worth commenting on the status of the variables in the renormalization constants in

both schemes. We adopt the convention that the scheme is indicated on the renormalization
constant itself and therefore the coupling constant and covariant gauge parameter are the
variables of the same scheme. However, the three loop renormalization of QCD in the
RI scheme has been given in [32] and the relationship between the parameters has been
determined, [32], as
5
aRI = aMS + O aMS
(2.7)
and
RI

a

2
= 1 + 9MS
18MS 97 CA + 80TF Nf MS
36
4
3
2
+ 18MS 18MS + 190MS 576(3)MS + 463MS + 864(3)
248

2
7143 CA2 + 320MS
320MS + 2304(3) + 4248 CA TF Nf
a2

+ 4608 (3) + 5280 CF TF Nf MS
288
6
5
4
4
4
5670(5)MS
18792MS
+ 486MS + 1944MS + 4212(3)MS
3
3
3
2
+ 48276 (3)MS
6480(5)MS
75951MS
52164(3)MS
2
2
2
+ 2916 (4)MS
+ 124740(5)MS
+ 92505MS
1303668(3)MS
+ 11664 (4)MS + 447120(5)MS + 354807MS + 2007504(3)

4
3
+ 8748 (4) + 1138050(5) 10221367 CA3 + 12960MS
8640MS
2
2
129600 (3)MS
147288MS
+ 698112(3)MS 312336MS

+ 1505088 (3) 279936(4) 1658880(5) + 9236488 CA2 TF Nf
2
2
+ 248832 (3)MS
285120MS
+ 248832(3)MS 285120MS

5156352 (3) + 373248(4) 2488320(5) + 9293664 CA CF TF Nf
2
+ 38400MS
221184(3)MS + 55296MS 884736(3)

1343872 CA TF2 Nf2 + 3068928(3) + 4976640(5)

3
aMS
988416 CF2 TF Nf + 2101248(3) 2842368 CF TF2 Nf2

31104 MS
4
+ O aMS
(2.8)
where the scheme of the variable is indicated as a subscript and we note also that the RI
and MS -functions are formally equivalent at four loops, [32].
From these renormalization constants we can deduce the anomalous dimensions from
MS
D (a) = (a)
MS
ln Z
D
MS (a)
MS
ln Z
D
(2.9)
and
RI
D (a) = (a)
RI
ln Z
D

RI (a)
RI
ln Z
D
(2.10)
We find, in our conventions,

8
8CF
MS
[47CA 14CF 16TF Nf ]a 2
D (a) = CF a +
3
27

8CF
648(3) + 2615 CA2 1944(3) + 1066 CA CF

+
243

1296 (3) + 782 CA TF Nf + 1296(3) 70 CF2

+ 1296 (3) 853 CF TF Nf 112TF2 Nf2 a 3 + O a 4
(2.11)
249
which agrees with [7] and

8
CF
RI
27 2 + 81 + 1434 CA 224CF 504TF Nf a 2
D (a) = CF a +
3
54
CF
486 4 + 4131 3 972(3) 2 + 19899 2 10044(3)

+
1944

+ 64098 115344(3) + 806800 CA2 1458 2 + 15066

+ 10368 (3) + 172856 CA CF 4320 2 2592(3)

+ 22248 + 25920(3) + 534400 CA TF Nf

+ 3888 + 41472(3) 43328 CF TF Nf

+ 82944 (3) 4480 CF2 + 78208TF2 Nf2 a 3 + O a 4
(2.12)
in four dimensions. We adopt the same convention for the renormalization group functions
as for the renormalization constants in that the scheme the variables correspond to is
indicated on the left-hand side. Restricting to SU(3) and the Landau gauge we have
SU(3) 32
4
RI

a
[378Nf 6005]a 2
D (a) =0 =
9
243
8
10998Nf2 6318(3)Nf 467148Nf
+
6561

524313(3) + 3691019 a 3 + O a 4 .
(2.13)
As an aid to the lattice matching procedure we note that the finite parts of the various
components in the two schemes after renormalization are

finite
(1)MS
D (p) p 2 =2

1
31
9487 5
2 8(3) + 2(3) CF CA
= 1 CF a
9
324
3
8

8195
2101
1 2 3
TF Nf CF +

+ 8(3) CF2 a 2
162
2
2
648

452579 976
4
64
635
(3) (3) (4) TF Nf CF CA
216
2187
81
9
3

61322 272
64
64
85
+
(3) (3) + (4) TF Nf CF2
+
9
2187
9
9
3

63602 256
+
(3) TF2 Nf2 CF
+
2187
81

953
75
21026833
1
135 5 2
+ 2 + 3
+ (5)
+
69984
1728
64
48
4
4

1 3
167
3230
719
3
3
+
+ 2 2 +
(3) +
2 (4) CF CA2
3
36
81
48
8
16

605431
440 40
5 3 41 2 979

+
(5)
+
8
72
27
4374
3
3
250

1
439 53
2 + 3 (3) 38(4) CF2 CA
9
2
6

16
400
64
7964 164
2
2 + 3 (3) + (4)
(5)
+
81
9
3
3
3
3

79
2935 821
1
+
+ 2 3 CF3 a 3 + O a 4 ,
(2.14)
5832
36
36
4

finite
(2)MS
D (p) p 2 =2

2
284 29
+ 2 CF a
+ + 2 2 6(3) 2(3) CF CA
=
3
9
3

74
20
35
+ 16(3) CF2 a 2
TF Nf CF + 2
3
9
3

136
2267
87980
184
(3) +
(3)
TF Nf CF CA
3
9
54
243

4913 1088
1399
9680 2 2
+
(3) TF Nf CF2 +
T N CF
+
81
27
9
243 F f

431 2 37 3
2762093 28855
445
+
+
+ +
+ 5 (5)
+
3888
216
16
8
3

23 2 1165
1617
(3) CF CA2
+
+
4
18
4

198059
1480 40
55 2 4790
3
(5)
+ +
36
81
324
3
3

7450 131
4232
1760
2
2
+
+
5 (3) CF CA +
(5)
+ 8 (3)
9
9
3
9

79
29483 3281
1
+
+ 2 3 CF3 a 3 + O a 4
(2.15)
972
324
18
2
and

(2)RI finite
D (p) 2 2
p =

2
284 271
1
+ 2 CF a
+
+ 3 2 + 3 6(3) 2(3) CF CA
=
3
9
18
2

20 40
604
+ TF Nf CF + 2 + 3
+ 16(3) CF2 a 2
3
9
27

32
20 3 40 2 9470
87980
184
(3) + (3)
TF Nf CF CA
3
9
9
3
81
243

40 2
36536 1088
+ 52
+
(3) 32(3) TF Nf CF2
9
243
9

569 2 127 3
9680 800
2762093 289991
+
TF2 Nf2 CF +
+
+
+
+
243
81
3888
1296
12
9
251

1617
19 4 1 5
1 3 11 2 685
+
+ +
+
(3)
8
4
2
4
9
4

469555
1 4 5 3 107 2 1429
445
+ 5 (5) CF CA2 +
+ +
+
+
3
2
2
9
36
972

1480 40
7216
7 2 (3)
(5) CF2 CA
+
9
3
3

3688
56287 35
31 2 1 3 3 3
1760
(5)
8 (3)
+ + CF a
+
3
9
486
6
6
2
4
+O a .
(2.16)
+
These imply
SU(3),=0

(1)MS finite
D (p) 2 2
p =

68993 160
2101
124
a
(3)
Nf a 2
=1
27
729
9
243

8959
4955
451293899 1105768
(3) +
(4) +
(5)
157464
2187
324
81

640
8636998
224
(3)
(4)
+
Nf
81
27
19683

63602 256
+
(3) Nf2 a 3 + O a 4
+
6561
243
(2.17)
and
SU(3),=0

finite
(2)MS
(p)
2 2
D
p =

2224 40
40
8
136281133 376841
2
+ (3) Nf a
(3)
= a
9
27
9
9
26244
243

43700
1232
15184
9680 2 3
+
(5) +
(3)
Nf a + O a 4 . (2.18)
Nf +
81
81
27
729
3. Second moment of transversity operator

D is similar to that for
The renormalization of the transversity operator
D aside from the decomposition of the corresponding Greens function. The

specific operator we are interested in is
2
D

(d 1)

D +
D

D +
D
S D =
+
1
(d 1)
(3.1)
252
which is symmetric with respect to the indices and and satisfies the traceless
conditions, [37],
D = 0.
S D = S
(3.2)
The Lorentz decomposition of the particular Greens function we will renormalize is
G
D (p)

= (p) S D (0)(p)

(d + 2)
(1)

p
p
p
+
p
= D (p) p + p
p2

+ (2)
D (p) p + p

(d 1)(d + 2)

+
p
p
p
(d
+
1)
p2

(3)
+ D (p) p p p + p p p + d p p p p p2
(3.3)
where the components are determined through
(1)
D (p)

1
(d + 2)
tr p + p
p p p
8(d 1)(d 2)
p2

+ p G
(p)
tr p p p
D
2
8(d 1)(d 2)p

+ p p p + d p p p p p2 G
D (p) ,
(2)
D (p)

(d 2 d + 2)
=
tr p + p (d + 1) p
8(d 1)(d 2 1)(d 2 4)

(d 1)(d + 2)
+
p p p G
D (p)
p2
1
+
tr p p p + p p p
2
8(d + 1)(d 2)p

+ d p p p p p2 G
D (p) ,
(3)
D (p)

(d + 2)
1
p p p
tr p + p
8(d 1)(d 2)
p2

+ p G
(p)
D

1
tr p + p (d + 1) p
8(d + 1)(d 2)

(d 1)(d + 2)
G
+
p
p
p
(p)

D
p2
(d + 2)
tr p p p + p p p
2
8(d + 1)(d 2)p

+ d p p p p p2 G
D (p) .
253
(3.4)
D but inUsing the same programmes which determined the renormalization of

stead including the Feynman rule for the transversity operator, we find after renormalizing
the Greens function in the MS scheme that
MS
Z
D
3
= 1 + CF a

1
1
+ (4TF Nf + 9CF 11CA ) 2 + (35CA 9CF 12TF Nf )
CF a 2
2
4

+ 242CA2 297CA CF 176CA TF Nf + 81CF2 + 108CF TF Nf

1
+ 1143CA CF 1178CA2 + 784CA TF Nf

3
18
1
243CF2 252CF TF Nf 96TF2 Nf2
36 2

2
+ 12553CA 7479CACF 5184(3) + 4168 CA TF Nf + 1782CF2

1

+ 5184 (3) 3240 CF TF Nf 368TF2 Nf2
CF a 3 + O a 4
324
+ 32TF2 Nf2
and with the RI scheme definition

RI

(1)

lim ZRI Z
D
D (p) p 2 =2 = 1
0
we have
RI
Z
D
(3.5)
(3.6)

3 1
1
=1+
+ (3 + 7) CF a + (4TF Nf + 9CF 11CA ) 2
2
2
1
+ 35CA + (33 + 18)CF 12TF Nf

4

1
3
9 + 54 2 36(3) + 271 324(3) + 1189 CA
+
24

+ 36 2 + 108 + 288(3) 165 CF (80 + 468)TF Nf a 2

+ 242CA2 297CA CF 176CA TF Nf + 81CF2 + 108CF TF Nf
254
+ (450 297)CA CF 1178CA2 + 784CA TF Nf

3
18
1
+ (243 + 324)CF2 + 108CF TF Nf 96TF2 Nf2
36 2
3
2
+ 729 + 4374 2916(3) + 30456 26244(3)

+ 101196 CA CF + 25106CA2 10368(3) + 8336 CA TF Nf

+ 2916 2 + 6561 14904 + 23328(3) CF2 736TF2 Nf2
1

+ 10368 (3) 51192 9396 CF TF Nf
648
1
972 5 + 9234 4 1944(3) 3 + 54864 3 10692(3) 2

+
5184
+ 184356 2 282312(3) + 19440(5) + 874185

1644228 (3) + 447120(5) + 3930425 CA2 + 3888 4 + 24300 3

+ 32TF2 Nf2
+ 3888 (3) 2 + 116748 2 34992(3) + 51840(5) + 375732

+ 1260576 (3) 466560(5) 511188 CA CF
8640 3 + 51840 2 13824(3) + 454560

124992 (3) + 124416(4) + 2438256 CA TF Nf

+ 38400 + 18432(3) + 323200 TF2 Nf2
34560 2 82944(3) + 295920 82944(3) 124416(4)

+ 620592 CF TF Nf + 7776 3 31104(3) 2 + 54432 2 + 32076

+ 51840 (3) + 207360(5) 266868 CF2 CF a 3 + O a 4 .
(3.7)
Consequently, the renormalization group function
CF
[35CA 9CF 12TF Nf ]a 2
2

CF
12553CA2 7479CA CF 5184(3) + 4168 CA TF Nf
+
108

+ 1782CF2 + 5184(3) 3240 CF TF Nf 368TF2 Nf2 a 3 + O a 4
MS
D (a) = 3CF a +
(3.8)
has emerged as being independent of as expected. The two loop part of (3.8) agrees with
the known results, [3741], when they are restricted to the second moment. A second check
on the result is that the double and triple poles in at three loops in the renormalization
constant have correctly emerged. Also, in four dimensions

CF
2
9 + 27 + 364 CA 54CF 128TF Nf a 2
12
CF
162 4 + 1377 3 324(3) 2 + 6633 2 3348(3)

+
432
RI
D (a) = 3CF a +
255
+ 21366 42768(3) + 224296 CA2 + 162 3 + 324 2 1674

+ 38016 (3) 71100 CA CF 1440 2 864(3) + 7416

+ 5184 (3) + 145600 CA TF Nf + 432 + 6912(3)

4032 CF TF Nf + 7128CF2 + 20992TF2 Nf2 a 3 + O a 4 .
(3.9)
Thus, for SU(3) and the Landau gauge we have
SU(3)

MS
= 4a 2[2Nf 31]a 2
D (a)
=0

1
92Nf2 + 4320(3)Nf + 8412Nf 86229 a 3 + O a 4
81
(3.10)
and
SU(3)
4

RI
= 4a [16Nf 255]a 2
D (a)
=0
9

2
656Nf2 396(3) + 27636 Nf 29106(3)
+
81

+ 218367 a 3 + O a 4
(3.11)
which are the main results of this article. To assist with the lattice matching procedure
we record that the finite parts of the components in both the MS and RI schemes after
renormalization are

(1)MS finite
D (p) 2 2
p =

3
3
7
943 3
21
+
+ + 2 (3) + (3) CF CA
=1
CF a
2 2
24
4
8
2
2

3
37 5
+ 2 12(3) CF2 a 2
16TF Nf CF
2
2
4

478 47
+ 8 (3) + 24(4) TF Nf CF2
27
3

1160 32
61 10
2 2
+ (3) TF Nf CF +
+ (3)
+
27
9
9
3

104843 68
253 2
2656369 18379
24 (4)
TF Nf CF CA +
+
+
324
9
5184
576
32

9
113 3
26839 131
1

+
2 3 (3)
+
96
144
12
16
3

3 2
5
69 3
5
(4) + 45 + 2 (5) CF CA2
+
16 8
16
4
4

3 2
1699 131
+
3 (3) (70 + 10)(5)
+
6
4
4
256

5
158969 2657
7
2 + 3 CF2 CA
6 (4)
432
48
8
8

2
+ 40 (5) 74 + 24 + 6 2 3 (3)
3

33 2 3 3 3 3
521 59
+ + CF a + O a 4 ,
+
12
2
8
8

finite
(2)MS
(p)

D
and

(3)MS finite
D (p)
p 2 =2

1

finite
= (1)MS
(p)
2 2 + O a4
D
p =
3
p 2 =2

(3)RI finite
=
(p)

D
p 2 =2

= O a4
(3.12)
(3.13)
(3.14)
to three loops. Hence,

SU(3),=0

(1)MS finite
D (p) 2 2
p =

2237 62
32
14
1852993 97391
2
(3) Nf a
(3)
=1 a
3
18
3
3
432
108

79
7060
122
80
306881
+ (4) +
Nf
(5) +
(3) (4)
4
27
9
3
486

1160 32
+ (3) Nf2 a 3 + O a 4 .
+
(3.15)
81
27
4. Conversion functions
As in [32] we have checked our expressions by also constructing the conversion
functions CO (a, ) for each of the operators explicitly. These functions allow one to move
from one scheme to another and are defined by, [42],
CO (a, ) =
RI
ZO
MS
ZO
(4.1)
The renormalization group functions are then related by

RI
MS
O
(aRI ) = O
(aMS ) (aMS )
MS MS (aMS )
ln CO (aMS , MS )
aMS
ln CO (aMS , MS )
MS
(4.2)
where we have indicated the scheme of the variables explicitly. Being careful to first
convert the variables to the same reference scheme, which we will take to be the MS
257
scheme, we have from the various renormalization constants,

C D (a, )
CF a
= 1 + (9 + 31)
9

+ 162 2 162 (3) + 783 1782(3) + 6404 CA

CF a 2
+ 81 2 + 315 + 1296(3) 228 CF 2668TF Nf
162
3
2
2
+ 161838 201204(3) + 942597 2124792(3) + 174960(5)

+ 4796982 11944044(3) + 746496(4) + 524880(5) + 38226589 CA2
+ 26244 3 + 151632 (3) 2 + 159408 2 + 346032(3) + 466560(5)

+ 554904 4914432(3) 2239488(4) + 8864640(5) + 3993332 CA CF
+ 528768 (3) 1469016 + 369792(3) 1492992(4)

24752896 CA TF Nf + 17496 3 373248(3) 2 + 289656 2

715392 (3) + 879336 + 10737792(3) + 1492992(4) 9331200(5)

3848760 CF2 497664(3) + 351648 3234816(3) 1492992(4)

CF a 3
+ O a4
+ 9980032 CF TF Nf + 221184(3) + 3391744 TF2 Nf2
69984
(4.3)
and
C D (a, )

CF a
+ 36 2 36(3) + 174 324(3) + 1189 CA

2
CF a 2

+ 36 2 + 108 + 288(3) 165 CF 486TF Nf
24
3
2
2
+ 17982 22356(3) + 104733 238032(3) + 19440(5)

+ 523602 1644228(3) + 447120(5) + 3930425 CA2
+ 10692 3 + 3888 (3) 2 + 63180 2 34992(3) + 51840(5)

+ 312876 + 1260576(3) 466560(5) 511188 CA CF

+ 58752 (3) 163224 + 124992(3) 124416(4) 2438256 CA TF Nf
+ 7776 3 31104 (3) 2 + 54432 2 + 32076 + 51840(3) + 207360(5)

266868 CF2 41472(3) + 101520 82944(3) 124416(4)
= 1 + (3 + 7)

CF a 3
+ O a4 .
+ 620592 CF TF Nf + 18432(3) + 323200 TF2 Nf2
5184
(4.4)
258
With these expressions given in terms of the MS coupling constant and covariant gauge
parameter, we have checked that the same RI scheme anomalous dimensions emerge for
each operator and for non-zero .
5. Large Nf critical exponent for transversity operator

In order to provide another independent check on our transversity results for MS we
have extended the leading order large Nf calculation of the critical exponent associated
with the flavour non-singlet twist-2 operator anomalous dimension as a function of d =
2 and arbitrary moment n, [43], to the transversity case. In other words we consider the
1 D 2 D n1 which is symmetric in the indices {i } and satisfies a more
operator
general traceless condition to that for n = 2, [37]. As the large Nf critical point technique
has been widely documented in [4345] we refer interested readers to these sources and
quote the main result of our computations. Following the procedure of [43], we found the
d-dimensional critical exponent for the transversity operator is
(n)

2( 1)CF 10
( 2)2
=
( 1)
(2 1)( 2)TF Nf
( + n 1)( + n 2)

1
,
+ 2 ( + n 1) () + O
Nf2
(5.1)
where
10 =
(2 1)( 2)+(2)
4+ 2 ()+( + 1)+(2 )
(5.2)
and (x) = d ln +(x)/dx and +(x) is the Euler gamma function. If one expands (5.1) in
powers of , = 2 , then there is a one-to-one correspondence with the leading large
Nf coefficients of the associated anomalous dimension to all orders in the (perturbative)
coupling constant. Moreover, since we have a result for arbitrary n we can deduce an
MS
analytic expression for the leading large Nf part of the three loop term of
D (a)
and check that it agrees with our result for n = 2. In particular if we formally write

r1

j
j
(n)MS
2
(5.3)
brj TF Nf a r ,
D (a) = CF b1 a + (b21 TF Nf + b20 )a +
r=3 j =0
where the coefficients, {bij }, are functions of n and b1 , b20 and b21 are known, [3741],
then we find the leading order three loop coefficient is

4
3(17n2 + 17n 8)
b32 =
(5.4)
48S3 (n) 80S2 (n) 16S1 (n) +
,
27
n(n + 1)

where Sl (n) = ni=1 1/i l . Setting n = 2 we find exact agreement with the corresponding
term of (3.8). Moreover, we can expand the critical exponent to four loops and deduce,
using the four loop -function, [46],

8
96S4 (n) 160S3 (n) 32S2 (n) + 192(3)S1 (n) 32S1 (n)
b43 =
81

(131n4 + 262n3 + 211n2 16n 48)
144 (3) +
.
n2 (n + 1)2
259
(5.5)
6. Discussion
To conclude we have derived expressions for the three loop anomalous dimension of the
second moment of the transversity operator in both the MS and RI schemes. The results
for the latter will be important in the extraction of lattice estimates of the associated matrix
elements. Moreover, we have provided the same information for the usual twist-2 flavour
non-singlet operator. Although we have had to consider various projections of the Greens
function with the operator inserted in a quark two-point function it ought to be possible to
extend the present calculations to derive the three loop anomalous dimensions for higher
moments of the transversity operator in the MS scheme. In this case since one is only
interested in the divergent part and not the finite part of the Greens function, it would
not require the same use of the projections considered here. This is important since the
algebraic manipulations are slowed for a large number of projections and the inclusion
of more covariant derivatives in the operator itself. However, since those anomalous
dimensions are independent of the gauge parameter, the algebraic computations would
be speeded by considering the Feynman gauge.
Acknowledgements
The author thanks Dr. P.E.L. Rakow and Dr. C. McNeile for valuable discussions.
References
[1] J.P. Ralston, D.E. Soper, Nucl. Phys. B 152 (1979) 109.
[2] R.L. Jaffe, X. Ji, Phys. Rev. Lett. 67 (1991) 552;
R.L. Jaffe, X. Ji, Nucl. Phys. B 375 (1992) 527.
[3] J.I. Cortes, B. Pire, J.P. Ralston, Z. Phys. C 55 (1992) 409.
[4] M. Grosse Perdekamp, A. Ogawa, K. Hasuko, S. Lange, V. Siegle, Nucl. Phys. A 711 (2002) 69c.
[5] V. Barone, A. Drago, P.G. Ratcliffe, Phys. Rep. 359 (2002) 1;
P.G. Ratcliffe, hep-ph/0211222.
[6] P.G. Ratcliffe, hep-ph/0211232.
[7] S.A. Larin, T. van Ritbergen, J.A.M. Vermaseren, Nucl. Phys. B 427 (1994) 41.
[8] S.A. Larin, P. Nogueira, T. van Ritbergen, J.A.M. Vermaseren, Nucl. Phys. B 492 (1997) 338.
[9] S. Moch, J.A.M. Vermaseren, Acta Phys. Polon. B 33 (2002) 3019.
[10] S. Moch, J.A.M. Vermaseren, A. Vogt, Nucl. Phys. B 646 (2002) 181.
[11] J.A.M. Vermaseren, S. Moch, A. Vogt, hep-ph/0211296.
[12] D.J. Gross, F.J. Wilczek, Phys. Rev. D 9 (1974) 980.
260
[13] E.G. Floratos, D.A. Ross, C.T. Sachrajda, Nucl. Phys. B 129 (1977) 66;
E.G. Floratos, D.A. Ross, C.T. Sachrajda, Nucl. Phys. B 139 (1978) 545, Erratum.
[14] A. Gonzlez-Arroyo, C. Lpez, F.J. Yndurin, Nucl. Phys. B 153 (1979) 161.
[15] G. Curci, W. Furmanski, R. Petronzio, Nucl. Phys. B 175 (1980) 27.
[16] E.G. Floratos, D.A. Ross, C.T. Sachrajda, Nucl. Phys. B 152 (1979) 493.
[17] A. Gonzlez-Arroyo, C. Lpez, Nucl. Phys. B 166 (1980) 429.
[18] C. Lpez, F.J. Yndurin, Nucl. Phys. B 183 (1981) 157.
[19] W. Furmanski, R. Petronzio, Phys. Lett. B 97 (1980) 437.
[20] E. Remiddi, J.A.M. Vermaseren, Int. J. Mod. Phys. A 15 (2000) 725.
[21] J.A.M. Vermaseren, math-ph/0010025.
[22] J.A. Gracey, Phys. Lett. B 488 (2000) 175.
[23] D.J. Broadhurst, A.G. Grozin, Phys. Rev. D 52 (1995) 4082.
[24] D.J. Broadhurst, Phys. Lett. B 466 (1999) 319.
[25] G. Martinelli, C. Pittori, C.T. Sachrajda, M. Testa, A. Vladikas, Nucl. Phys. B 445 (1995) 81.
[26] E. Franco, V. Lubicz, Nucl. Phys. B 531 (1998) 641.
[27] K.G. Chetyrkin, A. Rtey, Nucl. Phys. B 583 (2000) 3.
[28] M. Gckeler, R. Horsley, D. Pleiter, P.E.L. Rakow, A. Schfer, G. Schierholz, hep-lat/0209160.
[29] M. Gckeler, R. Horsley, H. Oelrich, H. Perlt, D. Petters, P.E.L. Rakow, A. Schfer, G. Schierholz,
A. Schiller, Nucl. Phys. B 544 (1999) 699;
S. Capitani, M. Gckeler, R. Horsley, H. Perlt, P.E.L. Rakow, G. Schierholz, A. Schiller, Nucl. Phys. B 593
(2001) 183.
[30] M. Guagnelli, K. Jansen, F. Palombi, R. Petronzio, A. Shindler, L. Wetzorke, hep-lat/0303012.
[31] M. Lscher, R. Narayanan, P. Weisz, U. Wolff, Nucl. Phys. B 384 (1992) 168.
[32] J.A. Gracey, hep-ph/0304113.
[33] S.G. Gorishny, S.A. Larin, L.R. Surguladze, F.K. Tkachov, Comput. Phys. Commun. 55 (1989) 381.
[34] S.A. Larin, F.V. Tkachov, J.A.M. Vermaseren, The Form version of Mincer, NIKHEF-H-91-18.
[35] P. Nogueira, J. Comput. Phys. 105 (1993) 279.
[36] S.A. Larin, J.A.M. Vermaseren, Phys. Lett. B 303 (1993) 334.
[37] A. Hayashigaki, Y. Kanazawa, Y. Koike, Phys. Rev. D 56 (1997) 7350.
[38] F. Baldracchini, N.S. Craigie, V. Roberto, M. Socolovsky, Fortschr. Phys. 30 (1981) 505.
[39] X. Artru, M. Mekhfi, Z. Phys. C 45 (1990) 669.
[40] W. Vogelsang, Phys. Rev. D 57 (1998) 1886.
[41] J. Blmlein, Eur. Phys. J. C 20 (2001) 683.
[42] J.C. Collins, Renormalization, Cambridge Univ. Press, Cambridge, 1984.
[43] J.A. Gracey, Phys. Lett. B 322 (1994) 141.
[44] A.N. Vasilev, Yu.M. Pismak, J.R. Honkonen, Theor. Math. Phys. 46 (1981) 157;
A.N. Vasilev, Yu.M. Pismak, J.R. Honkonen, Theor. Math. Phys. 47 (1981) 291.
[45] A.N. Vasilev, M.Yu. Nalimov, Theor. Math. Phys. 56 (1982) 643.
[46] T. van Ritbergen, J.A.M. Vermaseren, S.A. Larin, Phys. Lett. B 400 (1997) 379.

Magnetized four-dimensional Z2 Z2 orientifolds

M. Larosa, G. Pradisi
Dipartimento di Fisica, Universit di Roma Tor Vergata, INFN, Sezione di Roma Tor Vergata, Via della
Ricerca Scientifica 1, I-00133 Roma, Italy
Abstract
We study deformations of Z2 Z2 (shift-)orientifolds in four dimensions in the presence of both
uniform Abelian internal magnetic fields and quantized NSNS Bab backgrounds, that are shown
to be equivalent to asymmetric shift-orbifold projections. These models are related by T-duality to
orientifolds with D-branes intersecting at angles. As in corresponding six-dimensional examples,
D9-branes magnetized along two internal directions acquire a charge with respect to the RR sixform, contributing to the tadpole of the orthogonal D5-branes (brane transmutation). The resulting
models exhibit rank reduction of the gauge group and multiple matter families, due both to the
quantized Bab and to the background magnetic fields. Moreover, the low-energy spectra are chiral
and anomaly free if additional D5-branes longitudinal to the magnetized directions are present, and
if there are no RamondRamond tadpoles in the corresponding twisted sectors of the undeformed
models.
PACS: 11.25.-w; 11.25.Hf; 11.25.Mj; 11.25.Db; 12.60.-i
1. Introduction
Perturbative type-I vacua [13] (for reviews see [4,5]) are a small corner in the moduli
space of the unified underlying and yet poorly understood eleven-dimensional M-theory
[6]. Nowadays, however, they are the most promising perturbative models to test possible
stringy effects at the next generation of accelerators [7]. Indeed, while in the usual
heterotic SUSY-GUT scenarios [9] the string scale is directly tied to the Planck scale,
making it hard to conceive probes of low-energy effects, the type-I string scale is basically
an independent parameter, that can be lowered down to a few TeVs [10]. In this setting,
E-mail address: gianfranco.pradisi@roma2.infn.it (G. Pradisi).
doi:10.1016/S0550-3213(03)00551-0
262
M. Larosa, G. Pradisi / Nuclear Physics B 667 (2003) 261309
and more generally in the context of brane-world scenarios [7,11], the gauge degrees of
freedom are confined to some (stacks of) branes while the gravitational interactions invade
the whole higher-dimensional spacetime. In order to respect the experimental limits on
gravitational interactions, the extra dimensions orthogonal to the branes could be up to
sub-millimeter size [12], while the extra dimensions longitudinal to the branes should be
quite tiny (at least of TeV scale) but still testable in future experiments. In this context,
however, several aspects of the conventional Standard Model picture like, for instance,
the problem of scale hierarchies and the unification of the running coupling constants at a
scale of order 1016 GeV in the MSSM desert hypothesis, have to be reconsidered [5,7,8].
Some other issues, like supersymmetry breaking, find instead new possibilities in type-I
perturbative vacua.
There are basically four known ways to break supersymmetry within String Theory.
The first is to combine left and right moving modes in a non-supersymmetric fashion,
like, for instance, in the type-0 models [13] and in the corresponding lower-dimensional
compactifications and orientifolds [2,1416]. Some type-I-like instances, the type 0 B and
its compactifications [14], are also free of the tachyons that typically plague this kind
of models. The second is the generalization to String Theory of the ScherkSchwarz
mechanism [17], available also in the heterotic case, in which the breaking is due to a
generalized KaluzaKlein compactification that involves different periodicities for bosons
and fermions, thereby not respecting supersymmetry [18]. In this framework, the scale of
supersymmetry breaking is inversely proportional to the volume of the internal manifold
and some residual global supersymmetries may be left at tree level on some branes (brane
supersymmetry) [19,20]. The third possibility is related to models with supersymmetry
non-linearly realized on some branes, as a result of the simultaneous presence of branes and
anti-branes of the same [2123] or of different types [22,24,25], or of the introduction of
exotic orientifold planes [26]. These are referred to as brane supersymmetry breaking
models, and the corresponding scale of supersymmetry breaking can be tied generically
to the string scale. Finally, supersymmetry can be broken [2729] by the introduction
of internal magnetic fields [30] in the open sectors [28,3133]. In Refs. [31,32], sixdimensional (T 2 T 2 )/Z2 orientifold models with a pair of internal magnetic fields turned
on inside the internal tori have been discussed in detail.
In this paper we extend the construction of [31,32] to four-dimensional models obtained
as magnetic deformations of Z2 Z2 orientifolds, possibly combined with momentum or
winding shifts along some of the internal directions [20,22]. Some preliminary results have
already appeared in Ref. [34]. As in the six-dimensional models of [31,32], for generic
values of the magnetic fields NielsenOlesen instabilities [35] manifest themselves in the
presence of tachyonic excitations, and supersymmetry is broken due to the unpairing of
states of different spins. However, the compactness of the internal space allows self-dual or
anti-self-dual Abelian field configurations with non-vanishing instanton number, that can
compensate the RR charge excess, eliminating the tachyons and retrieving supersymmetry
if the BPS bound is saturated, or giving rise to brane supersymmetry breaking models
if the magnetized D9-branes transmute into anti-D5-branes [31,32]. The resulting models
exhibit several interesting features, to wit ChanPaton gauge groups of reduced rank and
several families of matter multiplets, linked in a natural way to the degeneracy of the
Landau levels. Moreover, in the presence of D-branes longitudinal to the directions along
263
which the magnetic fields are turned on, the four-dimensional models can also be chiral. It
should be stressed that these models with internal background (open) magnetic fields are
connected by T-duality to orientifolds with D-branes intersecting at angles [3644], that
have received much attention in the last few years in attempts to recover (extensions of)
the Standard Model as low-energy limits of String Theory or M-theory. A byproduct of
this analysis is a precise link between a quantized NSNS Bab and shift-orbifolds.
This paper is organized as follows. In Section 2, after a brief discussion of the relation
between shifts and the quantized NSNS two-form Bab , we shortly review the basic facts
characterizing Z2 Z2 (shift-)orientifolds in its presence. In Section 3 we discuss general
aspects of the inclusion of uniform internal Abelian magnetic fields in the open-string
sector and shortly review the models in [31]. In Section 4 we describe four-dimensional
examples based on magnetic deformations of the Z2 Z2 (shift-)orientifolds previously
introduced. Section 5 is devoted to one brane supersymmetry breaking example and
finally Section 6 contains our conclusions. Notations, conventions and tables are displayed
in the appendices. In particular, Appendix A collects the relevant lattice sums that enter
the one-loop partition functions. The Z2 Z2 characters are defined in Appendix B, while
Appendix C is a collection of tables that summarize massless spectra, gauge groups and
tadpole cancellation conditions for all the models considered in this paper.
2. Z2 Z2 orientifolds
In this section we review some four-dimensional type-I vacua obtained as orientifolds
of Z2 Z2 orbifolds, or of freely acting Z2 Z2 shift-orbifolds. The starting point is an
orbifold of the type-IIB superstring compactified on an internal six-torus that, without any
loss of generality for our purposes, can be chosen to be a product of three two-tori T 45 ,
T 67 and T 89 along the three complex directions Z 1 = X4 + iX5 , Z 2 = X6 + iX7 and
Z 3 = X8 + iX9 . Each two-torus can be equipped with a NSNS background two-form Bi
of rank ri (with ri = 0 or 2) that, if the orientifold projection is induced by the world-sheet
parity operator , is a discrete modulus and may thus take only quantized values [45,46].
The orbifold group will be taken to be the combination of the Z2 Z2 generated by the
elements
g: (+, , ) and h: (, , +),
(2.1)
where the minus signs indicate the two-dimensional Z2 inversion of the corresponding
coordinates (Z i Z i ), with momenta and/or winding shifts along the real part of (some
of) the three complex directions.
The conventional Z2 Z2 orbifolds, allowing for the introduction of a discrete
torsion[47], give rise to supersymmetric orientifolds [39,48,49] as well as to orientifolds
with brane supersymmetry breaking [22]. Moreover, the freely acting Z2 Z2 (shift-)
orbifolds produce ten classes of orientifolds with different amounts of supersymmetry
(models with brane supersymmetry [19,20]), together with a huge number of variants
with braneanti-brane pairs and brane supersymmetry breaking [22].
264
2.1. NSNS Bab and shifts

Before moving to the detailed analysis of the models, it is worth clarifying the
relation between a non-vanishing discretized NSNS two-form Bab on orientifolds [45,46]
and momentum and winding shifts. In particular, we want to show that the presence
of a quantized NSNS two-form field Bab on a two-torus is exactly equivalent to an
asymmetric shift-orbifold in which a momentum shift along the first direction of the torus
is accompanied by a winding shift along the other direction.1 For simplicity, let us take for
the two-torus a product of circles both of radius R. Parametrizing the discretized two-form
as [45]

0 1
,
B=
(2.2)
2 1 0
the generalized momenta of Eqs. (A.1) and (A.2) reduce to the expressions
p(L,R)a = ma
1
gab nb ,
(2.3)
where ma = ma 12 ab nb . The presence of Bab makes the ma s integers or half-integers
depending on the oddness or evenness of the integer na s. As a result, omitting the prime
on the dummy m-variables for the rest of this section, the torus partition function of Eq.
(A.3) can be decomposed in the form

1
(B) = (m1 , m2 , 2n1 , 2n2 ) + m1 + , m2 , 2n1 , 2n2 + 1
2

1
+ m1 , m2 + , 2n1 + 1, 2n2
2

1
1
+ m1 + , m2 + , 2n1 + 1, 2n2 + 1 ,
(2.4)
2
2
where (m1 , m2 , n1 , n2 ) denotes the two-dimensional lattice sum over momenta ma /R
and windings na R.
The same partition function can be obtained as an asymmetric shift-orbifold. Indeed,
projecting the conventional (B = 0) (m1 , m2 , n1 , n2 ) under the action of a p1 w2 shift
and completing the modular-invariant with the addition of twisted sectors, the resulting
partition function is

1
(p1 w2 ) = (m1 , m2 , n1 , n2 ) + (1)m1 +n2 (m1 , m2 , n1 , n2 )
2

1
1
+ m1 , m2 + , n1 + , n2
2
2

1
1
+ (1)m1 +n2 m1 , m2 + , n1 + , n2 .
(2.5)
2
2
1 This argument was partly developed in collaboration with E. Dudas and J. Mourad. For related observations
in the context of duality chains, see [50].
265
This expression is exactly the one in Eq. (2.4) after doubling the radius (R 2R) along
the first direction. Thus, it should not come as a surprise that in some cases the effect
of the shifts can be compensated by the presence of a quantized Bab . However, all the
models in this paper display a reduction of the rank of the ChanPaton group when a nonvanishing quantized Bab is turned on. The reason is that the shifts we consider affect only
one real coordinate of the tori, rather than two as in the previous asymmetric shift-orbifold
construction. Nonetheless, in some cases, the shifts make the multiplicities of the matter
multiplets independent of the rank of Bab . By T-duality, other allowed discrete moduli
[5153] like, for instance, the off-diagonal components of the metric in orientifolds of the
type-IIA superstring, can also be related to suitable shift-orbifolds.
A nice geometric interpretation of the rank reduction of the ChanPaton groups can
also be given resorting to the asymmetric shift-orbifold description of Bab . As we shall
extensively see in the next sections, a momentum shift orthogonal to D-branes splits
them into multiple images. After a T-duality along the second direction of the two-torus,
the p1 w2 becomes a p1 p2 shift-orbifold, that admits orientifold projections containing
D1-branes parallel to p1 and orthogonal to p2 . The corresponding annulus amplitude can
be written
1
1/2
1/2
W2 + W2 ,
A = N 2 P1 + P1
(2.6)
2
where Pi and Wi are the usual one-dimensional momentum and winding lattice sums [45],
1/2
1/2
respectively, while Pi and Wi are the corresponding shifted ones, and the consistent
Mbius amplitude, describing the unoriented projection, is

1
1/2 W
1/2 .
M= N P
(2.7)
1 W2 + P
1
2
2
Eqs. (2.6) and (2.7) neatly display the expected doublet structure of the D1-brane
configuration, and the analysis of the tadpole cancellation conditions reveals the related
rank reduction of the ChanPaton group. For instance, an equivalent eight-dimensional
p1 p2 shift-orientifold compactification of the type-IIB superstring, would yield type-I
models with an SO(16) gauge group, thus providing a rank reduction by a factor of two.
2.2. Z2 Z2 models and discrete torsion
Let us now turn to the analysis of the Z2 Z2 models. Aside from the identity, the
Z2 Z2 elements can be grouped together in the matrix

+
0 = + ,
(2.8)
+
whose rows represent the action of g, f = g h and h on the three internal torus coordinates
Z i . The one-loop closed partition function can be obtained supplementing the Z2 Z2
projections of the toroidal amplitude with the inclusion of three twisted sectors, located
at the three fixed tori, to complete the modular invariant. There are actually two options,
related to the freedom of introducing a discrete torsion [47], i.e., a relative sign between
266
two disconnected orbits of the modular group. The result is
1
|Too |2 1 (B1 )2 (B2 )3 (B3 )
T =
4

42 2

+ |Tog | 1 (B1 ) + |Tof | 2 (B2 ) + |Toh | 3 (B3 ) 2
2
2 2

4
+ |Tgo |2 1 (B1 ) + |Tf o |2 2 (B2 ) + |Tho |2 3 (B3 ) 2
4
2 2

4
+ |Tgg |2 1 (B1 ) + |Tff |2 2 (B2 ) + |Thh |2 3 (B3 ) 2
3

83
+ |Tgh |2 + |Tgf |2 + |Tfg |2 + |Tf h |2 + |Thg |2 + |Thf |2

2 3 4
2

,

(2.9)
where the i s are the two-dimensional Narain lattice sums for the three internal tori (see
Appendix A), that depend on the two-dimensional blocks (Bi ) of the NSNS two-form
Bab , and = 1 is the sign associated to the discrete torsion. We have expressed the torus
amplitude in terms of the 16 quantities (i = o, g, h, f )
Tio = io + ig + ih + if ,
Tig = io + ig ih if ,
Tih = io ig + ih if ,
Tif = io ig ih + if ,
(2.10)
where the 16 Z2 Z2 characters il [48], combinations of products of level-one SO(2)

characters, are displayed in Appendix B. The geometric model, related to the charge
conjugation modular-invariant, corresponds to the choice = 1, as can be deduced
from the massless spectra reported in Table 2 (see Appendix C). It is a compactification
on (a singular limit of) a CalabiYau threefold with Hodge numbers (h11 = 51, h21 = 3),
while the = 1 choice, linked in this context to the T-dual compactification, leads to (a
singular limit of) the mirror symmetric CalabiYau threefold, with h11 = 3, h21 = 51.
The starting point for the orientifold construction are the Klein-bottle amplitudes
1
K=
P1 P2 P3 + 24 P1 W2 (B2 )W3 (B3 ) + 24 W1 (B1 )P2 W3 (B3 )
8

+ 24 W1 (B1 )W2 (B2 )P3 Too
r2 r3

+ 2 16 2 2 2 1 P1 + 22 W1 (B1 ) Tgo
r1 r3

+ 2 2 2 2 P2 + 22 W2 (B2 ) Tf o

r1 r2

2
(2.11)
+ 2 2 2 3 P3 + 22 W3 (B3 ) Tho
,
4
that project the oriented closed spectra into unoriented ones. The signs i are linked to the
discrete torsion through the crosscap constraint [54] by the relation [22]
1 2 3 = .
(2.12)
267
The transverse channel amplitude, obtained performing an S modular transformation, is

5

v1
= 2
K
W e P e (B2 )P3e (B3 )
v1 v2 v3 W1e W2e W3e + 24
8
v2 v3 1 2

4 v2
e e
e
4 v3
e
e
e
+2
P W (B2 )P3 (B3 ) + 2
P (B1 )P2 (B2 )W3 Too
v1 v3 1 2
v1 v2 1

r2 r3
P e (B1 )
+ 2 2 2 2 1 v1 W1e + 22 1
Tog
v1

r1 r3
P e (B2 )
+ 2 2 2 2 v2 W2e + 22 2
Tof
v2

2
e
r
r
2
(2.13)
21 22
e
2 P3 (B3 )
+2
3 v3 W3 + 2
Toh
,
v3
2
where the superscript e denotes the usual restriction of the sums to even subsets and the
vi denote the volumes of the three internal tori. At the origin of the lattices, the reflection
coefficients are perfect squares,

5
r
r
r
r
v1
v2
2
22 23
21 23

K0 =
v1 v2 v3 + 2
1
+2
2
8
v2 v3
v1 v3
2

r
r
v3
21 22
+2
3
oo
v1 v2

r2 r3
r1 r3
v1
v2
+
v1 v2 v3 + 2 2 2 1
2 2 2 2
v2 v3
v1 v3
2

r1 r2
v
3
2 2 2 3
og
v1 v2

r2 r3
r1 r3
v1
v2
+
v1 v2 v3 2 2 2 1
+ 2 2 2 2
v2 v3
v1 v3
2

r1 r2
v
3
2 2 2 3
of
v1 v2

r2 r3
r1 r3
v1
v2
v1 v2 v3 2 2 2 1
2 2 2 2
+
v2 v3
v1 v3
2

r1 r2
v3
+ 2 2 2 3
oh ,
(2.14)
v1 v2
and encode the presence, together with the conventional orientifold 9-planes (O9+ -planes
from now on), of three kinds of O5-planes, that we shall denote O51 , O52 and O53 .
These (non-dynamical) planes are fixed under the combined action of and the inversion
along the directions orthogonal to them, namely, g for the O51 , f for the O52 and h for
the O53 , and the index reflects their RR charge. We shall use the + sign to indicate
268
O-planes with tension and RR charge opposite to the corresponding quantities for the
D-branes, and the sign for the exotic orientifold planes with reverted tension and
RR charge. As is evident from Eq. (2.14), the i are proportional to the RR charges of the
O5i . While manifestly compatible with the usual positivity requirements, the eight different
choices reported, for the case with Bab = 0, in Table 3, affect the tadpole conditions. In
particular, the presence of exotic O5i requires the introduction of anti-branes in order to
globally neutralize the RR charge of the vacuum configuration. In this respect, according
to [24], = 1 implies the reversal of at least one of the O5-plane charges, producing
type-I vacua with brane supersymmetry breaking [22]. Moreover, the presence of the
NSNS two-form blocks Bi affects the reflection coefficients in front of a crosscap by the
familiar powers of two, responsible for the rank reduction of the ChanPaton gauge groups
[45,46].
In this section, we shall limit ourselves to the discussion of the orientifolds of the unique
supersymmetric model with i = +1, leaving to Section 5 some examples with brane
supersymmetry breaking. The unoriented closed spectra are reported in Table 4, while the
annulus amplitude can be written as
1 2 r6
N 2 P1 (B1 )P2 (B2 )P3 (B3 ) + D12 2r1 2 P1 (B1 )W2 W3
A=
8

+ D22 2r2 2 W1 P2 (B2 )W3 + D32 2r3 2 W1 W2 P3 (B3 ) Too
r2 r3

+ 2 2 + 2 2ND1 2r1 2 P1 (B1 ) + 2D2 D3 W1 Tgo
r1 r3

+ 2 2 + 2 2ND2 2r2 2 P2 (B2 ) + 2D1 D3 W2 Tf o

r1 r2
2
+2
r3 2
2
(2.15)
2ND3 2
+2
P3 (B3 ) + 2D1 D2 W3 Tho
,
4
where r = r1 + r2 + r3 is the total B-rank. Aside from the standard NN open-strings, there
are the three types of open-strings with Dirichlet boundary conditions along two of the
three internal directions, as well as mixed ND open strings. The corresponding vacuumchannel amplitude displays four independent squared reflection coefficients, related to the
ubiquitous D9-branes on which the NN strings end, and to three types of D5-branes. In
particular, we call D5i -branes those with world-volume that invades the four-dimensional
spacetime and the internal Z i coordinate. Again, the presence of the NSNS two-form
reflects itself in the generic appearance of additional matter multiplets whose multiplicities
depend on the rank of the Bi -blocks along the directions orthogonal to the fixed tori.
N and Di in Eq. (2.15) indicate the traces of the ChanPaton matrices, or ChanPaton
multiplicities, corresponding to the D9- and D5-branes, respectively.
Standard methods [14] determine the direct-channel Mbius amplitude
1 r6
2 2 NP1 (B1 , 1 )P2 (B2 , 2 )P3 (B3 , 3 )
M=
8
+2
+2
+2
r1 6
2
r2 6
2
r3 6
2
D1 P1 (B1 , 1 )W2 (B2 , 2 )W3 (B3 , 3 )

D2 W1 (B1 , 1 )P2 (B2 , 2 )W3 (B3 , 3 )

oo
D3 W1 (B1 , 1 )W2 (B2 , 2 )P3 (B3 , 3 ) T
269
2
2

2
(N + D1 )P1 (B1 , 1 ) + 2 (D2 + D3 )W1 (B1 , 1 ) Tog
2

2
r2 2

of 2
2 2 (N + D2 )P2 (B2 , 2 ) + 21 (D1 + D3 )W2 (B2 , 2 ) T
2
r3 2

2 2 (N + D3 )P3 (B3 , 3 ) + 21 (D1 + D2 )W3 (B3 , 3 )
2
2

,
Toh
(2.16)
2

r1 2
2
where the hatted version of the blocks in Eq. (2.10) is linked, as usual, to the choice of a
real basis of characters. A proper particle interpretation of the annulus and Mbius strip
amplitudes requires a rescaling of the charges in such a way that N = 2n and Di = 2di .
The (untwisted) tadpole conditions reported in Table 5 emphasize the usual rank reduction
due to the presence of quantized values of Bab and demand that the signs and satisfy
the conditions

i = 2,
i = 2(2ri )/2 .
(2.17)
i =0,1
i =0,1Ker(B)
There are several solutions for the allowed gauge groups, that depend on the additional
signs i and i defined by

(2.18)
i = 2i ,
i = 2(2ri )/2 i .
i =0,1
i =0,1Ker(B)
As shown in Table 5, they are products of four factors, chosen to be USp or SO depending
on the values of i and i . The massless unoriented open spectra, encoded in the annulus
and Mbius amplitudes at the lattice origin,

1 2
n + d12 + d22 + d32 (oo + og + oh + of )
A0 =
2
r2
r3
r1
r3
+ 2 2 + 2 (2nd1 + 2d2d3 )(go + gg + gh + gf )

+ 2 2 + 2 (2nd2 + 2d1d3 )(f o + fg + f h + ff )

r1 r2
+ 2 2 + 2 (2nd3 + 2d1d2 )(ho + hg + hh + hf )
and

1
d1
n
M0 = oo (1 2 3 1 2 3 ) + (1 2 3 1 2 3 )
2
2
2

d2
d3
+ (1 2 3 1 2 3 ) + (1 2 3 1 2 3 )
2
2

n
d1
+ og (1 2 3 1 + 2 + 3 ) + (1 2 3 1 + 2 + 3 )
2
2

d3
d2
+ (1 2 3 1 + 2 + 3 ) + (1 2 3 1 + 2 + 3 )
2
2
(2.19)
270
d1
n
(1 2 3 + 1 2 + 3 ) + (1 2 3 + 1 2 + 3 )
2
2

d2
d3
+ (1 2 3 + 1 2 + 3 ) + (1 2 3 + 1 2 + 3 )
2
2

d1
n
+ oh (1 2 3 + 1 + 2 3 ) + (1 2 3 + 1 + 2 3 )
2
2
d2
+ (1 2 3 + 1 + 2 3 )
2

d3
+ (1 2 3 + 1 + 2 3 ) ,
(2.20)
2
+ of
are reported in Table 6. Being non-chiral, these models are clearly free of anomalies.
2.3. Z2 Z2 shift-orientifolds and brane supersymmetry
In this section we review how (L , R ) shifts can be combined with Z2 Z2 orbifold
operations in the open descendants of type-IIB compactifications. As in [20,22], we
shall distinguish between symmetric momentum shifts (p) = (, ) and anti-symmetric
winding shifts (w) = (, ), since the two have very different effects on the resulting
spectra. These orbifolds correspond to singular limits of CalabiYau manifolds with Hodge
numbers (19, 19), (11, 11) and (3, 3) in the cases of one, two and three shifts, respectively,
as shown in Table 15. Let us begin by introducing a convenient notation to specify the
orbifold action Zi (Zi ) on the complex coordinates of the three internal tori. There
are several ways to combine the three operations g, f and h of the matrix (2.8) with
shifts consistently with the Z2 Z2 group structure. However, up to T-dualities and
corresponding redefinitions of the projection, all non-trivial possibilities are captured
by [20]

1 2 1
2 3 ,
1 (1 , 2 , 3 ) = 1
1 1
3

1 1
1
2 3 ,
2 (1 , 2 , 3 ) = 1
(2.21)
1 2 3
where the three lines refer to the new operations, that we shall continue to denote by g,
f and h, and where i indicates the combination of a shift in the real part of the ith
coordinate with the orbifold inversion. Notice that when a line of the table contains p
or w, the corresponding D5-branes are eliminated. One thus obtains the ten different
classes of models reported in Table 1, with the w2 p3 model now linked to the 1 action,
correcting a misprint in Ref. [20]. In listing these models, we have followed the choices of
axes made in Ref. [20], so that when a single set of D5-branes is present, this is always the
first, D51 , and when two sets are present, these are always D51 and D52 . All these freely
acting orientifolds have N = 1 supersymmetry in the closed sector, but exhibit interesting
instances of brane supersymmetry in the open part: additional supersymmetries are
present for their massless modes, that in some cases extend also to the massive ones
271
Table 1
Shifts and brane supersymmetry for the various models
Models
Shift
p3
w 2 p3
w 1 w 2 p3
p2 p3
w 1 p2
w 1 p2 p3
w 1 p2 w 3
p1 p2 p3
p1 w 2 w 3
w1 w2 w3
1
1
2
2
2
2
1
1
2
1
D9 susy
N
N
N
N
N
N
N
N
N
N
=1
=2
=4
=1
=2
=2
=4
=1
=2
=4
D51 susy
N
N
N
N
N
N
N
=2
=2
=4
=2
=4
=4
=4
D52 susy
N =2
N =4
N =4
[19] confined to some branes. Table 1 also collects the number of supersymmetries of the
massless modes for the various branes present in each model. The unoriented truncations
and the open spectra are generically affected by the shifts, that lift in mass some treelevel closed string terms eliminating the corresponding tadpoles, and determine the brane
content of the models, related to the presence of the projectors
1 1 + (1)1 +2 + (1)2 +3 + (1)1 +3 ,
(2.22)
2 1 + (1)1 + (1)2 +3 + (1)1 +2 +3 ,
(2.23)
for the 1 and 2 tables, respectively, along the tube.

The open-string spectra of the models in Table 1 are shown in Table 41. They correspond
to peculiar and interesting brane configurations, related to the fact that some projections
are absent in the NN or D9D9 string contributions, as well as in the DD or D5D5 string
contributions. These features admit a nice geometrical interpretation: they are linked to the
presence of multiplets of branes, associated with multiplets of tori fixed by some Z2 Z2
elements and interchanged by the action of the remaining ones. Only the projections
introduced by the former elements are thus present, since in these sectors the physical
states are combinations of multiplets localized on the image branes. If one attempts to
insert all branes at a fixed point, the other operations inevitably move them, giving rise to
multiple images. Equivalently, as already discussed in Section 2.1, brane multiplets may be
traced to the presence of momentum shifts orthogonal to the branes [19]. As a consequence,
the massless modes (or at times the full spectrum) exhibit enhanced supersymmetry. Fig. 1
displays the brane configuration of the w2 p3 model. The three axes refer to the three twotori T 45 , T 67 and T 89 , and the D51 -branes, drawn as wavy blue lines, occupy a pair of
fixed tori, while the D52 -branes, the solid red lines, occupy all the four fixed tori along the
Z 2 direction. The D51 D51 configurations correspond to doublets of branes, associated
to the pair of tori fixed by g and interchanged by f and h. As expected and as shown in
Table 1, there is an N = 2 supersymmetry associated to the D51 -brane doublets together
with an N = 4 supersymmetry associated to the quartets of D52 -branes.
272
Fig. 1. D51 - and D52 -brane configurations for the w2 p3 model.
3. Magnetic deformations and brane transmutation

A uniform background magnetic field Hi on the ith torus is a monopole field, a U (1)
bundle with non-trivial transition functions gluing two local charts, whose consistency with
particle dynamics requires a Dirac quantization condition
ki
,
(3.1)
vi
where qi is the U (1)-charge, vi is the volume of the torus and ki is an integer defining the
(quantized) number of elementary fluxons or, equivalently, the Landau-level degeneracy.
The background magnetic field couples solely to the open string ends through a boundary
term in the action [30]. Thus, denoting with qL and qR the U (1)-charges at the two ends of
the open strings and defining the total charge Q = qL + qR , one has to distinguish between
neutral, Q = 0, sectors and charged, Q = 0, ones. Charged open strings with Q = 1
or Q = 2 are characterized by oscillators whose frequencies are shifted according to
2 qi Hi =

1
arctg(2 qL Hi ) + arctg(2 qR Hi ) .
(3.2)
Moreover, there is a Landau-level degeneracy that exactly amounts to the ki factor in

Eq. (3.1) and reflects the fact that the zero-modes along the two directions of the torus
do not commute. Moreover, on a ZN -orbifold the unit cell has a volume reduced by a
factor N with respect to the torus. As a result, the degeneracies of Landau levels should
be multiples of N . For instance, ki are typically even on the Z2 -orbifold. However, a nonvanishing quantized NSNS Bab could make some orbifold projections ineffective, thus
allowing more choices for the ki [32]. On the other hand, neutral dipole strings with
qL = qR (Q = 0), have integer-mode frequencies, but the structure of their zero-modes
zi =
273
is rather peculiar. Indeed, both momenta and windings are now allowed, but the effect
of the magnetic field on this sector is simply to introduce rescalings in the momentum
and winding lattices entering the one-loop annulus partition function. An analysis of the
resulting projector along the tube shows that in the open one-loop channel the lattice sums
over momenta and windings are indeed subjected to the complex boosts

mi
,
ni ni 1 + (2 qi Hi )2 .
mi
(3.3)
1 + (2 qi Hi )2
Let us stress that a uniform Abelian magnetic field can be given a dual interpretation in
terms of rotated branes [36]. In particular, after a T-duality along one of the two directions
of one torus, the modified boundary conditions can be rephrased in terms of rotations of
the corresponding D-branes by angles i such that
tan i = 2 qi Hi .
The Dirac quantization condition in Eq. (3.1) can then be interpreted as the requirement that
the D-branes wrap exactly ki -times the tori. Notice that we always normalize the electric
charge of the open-strings in such a way that it corresponds to the elementary quantum.
This is the reason why we can describe all spectra using just one integer ki for each twotorus or, in other words, setting to one the electric winding number of the corresponding
D-branes.
Let us briefly review how these configurations manifest themselves in the orientifold of
type-IIB on a magnetized (T 2 T 2 )/Z2 [31]. This is a deformation of the six-dimensional
(T 2 T 2 )/Z2 [2,55] model, whose massless oriented closed spectrum is reported in
Table 7 and comprises N = (2, 0) supergravity coupled to 21 tensor multiplets, as expected
for a singular limit of a K3 compactification. The unoriented spectra, not affected by the
constant magnetic backgrounds, are reported in Table 8 and exhibit at zero mass N = (1, 0)
supergravity modes coupled to hypermultiplets and tensor multiplets whose numbers
depend upon the total rank r of Bab . In the open sector, the two magnetic fields H1 and
H2 turned on inside the two T 2 s are aligned along the same U (1) subgroup of the Chan
Paton gauge group. Generic fluxes of H1 and H2 produce the breaking of supersymmetry
and the presence of NielsenOlesen instabilities, signaled by the appearance of tachyonic
states. Absence of tachyons and supersymmetry can be attained choosing self-dual field
configurations, that at the same time allow to compensate the non-vanishing instanton
density with additional lower-dimensional branes. As a result, the magnetized D9-branes
mimic the behaviour of the D5-branes and couple to the RR six-form, contributing
to the tadpole cancellation conditions. This is the brane transmutation phenomenon,
first described, in this context, in Ref. [31] and linked there to the peculiar Wess
Zumino couplings in the D-brane actions [56]. There are several solutions, reported in
Tables 9, 10, 11, 12 (see Appendix C for notations and conventions used in the tables),
depending on the O-plane configurations, or equivalently on the signs, analogous to the
ones in Eq. (2.18), contained in the Mbius-strip amplitudes. In particular, the RR tadpole
cancellation conditions and the resulting ChanPaton gauge groups are shown in Table 9
for the models with complex charges and in Table 11 for the models with real charges.
Moreover, the untwisted NSNS tadpoles are related to the derivatives of the BornInfeld
action with respect to the untwisted moduli, while the twisted ones are associated with
274
corresponding couplings in the effective Lagrangian, as in [31,32]. The resulting open

spectra, reported in Table 10 for complex charges and in Table 12 for real charges, can be
easily recognized as deformations of the models without background magnetic fields [2,
55]. Several interesting facts, however, emerge from the analysis of Tables 10 and 12. First,
there is an unusual rank reduction of the ChanPaton gauge group. Second, some matter
multiplets appear in multiple families, because of the degeneracy introduced by the Landau
levels. Third, as already stressed the magnetized D9-branes behave exactly like D5-branes.
This is particularly evident for the models without D5-branes, related, for instance, to
the choice r = 0, k2 = k3 = 2, d = 0, n = 12 and m = 4 in Table 10, and corresponds
[57] to the fact that the D5-branes can be interpreted as instantons of vanishing size.
So, in the presence of self-dual configurations of the internal magnetic fields, a stack of
r
2 2 |k2 k3 | D5-branes is replaced by a fat instanton that invades the whole ten-dimensional
spacetime, in a transition related by T-duality [39] to the inverse small-instanton transition
discussed in [57]. Finally, introducing anti-self-dual configurations for the magnetic field,
the magnetized D9-branes mimic anti-D5-branes [32]. Tachyons are again eliminated, but
supersymmetry is broken at tree level in the open sector at the string scale or, better, is
non-linearly realized in the anti-D5-brane sector, as discussed in Refs. [58,59].
4. Magnetized four-dimensional models

The four-dimensional models developed in Refs. [20,22] display D9 and D5 branes
in their perturbative spectra, and are thus a natural arena to build consistent magnetized
models sharing the qualitative features of the six-dimensional T 4 /Z2 model. As we shall
see, their spectra exhibit rank reductions of the gauge groups and multiple matter families.
In addition, as in the six-dimensional examples, there is the option of introducing pairs
of magnetic fields aligned along the same U (1) subgroup. This is allowed only if the
undeformed model contains corresponding O5-planes orthogonal to the two magnetized
directions, or, equivalently, if there are sources that add to the magnetized D9-branes in
such a way as to compensate the RR charge excess. We shall always introduce uniform
Abelian background magnetic fields (H2 , H3 ) along the (Z 2 , Z 3 ) directions, thus requiring
just the presence of O51 -planes to balance the RR charge excess of magnetized D9-branes
and D51 -branes.
If D5-branes whose world-volume invades coordinates longitudinal to the magnetized
directions are also present, one obtains, as a bonus, chiral fermions. Chirality is connected
on the one hand to the intersection of two sets of orthogonal D5-branes, and on the other to
the chiral asymmetry in the pure magnetic sector. Moreover, the phenomenon of brane
transmutation acquires in this setting its full-fledged form. Indeed, as stressed in Section 2,
shift-orientifolds are characterized by the presence of multiplets of defocalized D5-branes.
This implies, as mentioned before, that some of the D5-branes cannot be put on the same
fixed tori but have to be distributed among the images interchanged by the action of some
orbifold group elements. The magnetized D9-branes remember the localized distribution
of the D5-branes they are mimicking. Indeed, although they invade the whole internal
space, the centers of the corresponding classical Landau orbits organize themselves in
multiplets that reflect the structure of the D5-branes.
275
4.1. Magnetized Z2 Z2 orientifolds

Let us begin the discussion of magnetic deformations by considering the plain
Z2 Z2 model with i = 1 (for related examples in the language of intersecting D-brane
models, see [39]). As stated in Section 2, the remaining models in Table 3 give rise to
orientifolds with brane supersymmetry breaking, that, for brevity, will not be explicitly
discussed here. However, in Section 5 we shall describe one class of orientifolds with
brane supersymmetry breaking related to a class of w2 p3 shift-orbifold models.
The closed string amplitudes, not affected by the introduction of constant background
magnetic fields H2 and H3 along the Z 2 and Z 3 directions, as well as the O-plane content,
are described in Section 2.2. In addition, the closed unoriented spectra are collected in
Table 3. The annulus amplitude can be obtained using the techniques of [31], reviewed in
Section 3, as a deformation of the annulus amplitudes of Eq. (2.15). The result can be cast
into the sum of the following three contributions (for notations and conventions see the
appendices)
1 2 r6
N 2 P1 (B1 )P2 (B2 )P3 (B3 ) + 2mm2
r6 P1 (B1 )P2 (B2 )P3 (B3 )
A(Q=0) =
8
+ D12 2r1 2 P1 (B1 )W2 W3 + D22 2r2 2 W1 P2 (B2 )W3

+ D32 2r3 2 W1 W2 P3 (B3 ) Too (0; 0; 0)
2

r2 r3

+ 2 2 + 2 Tgo (0; 0; 0) 2ND1 2r1 2 P1 (B1 ) + 2D2 D3 W1

4 (0)
2

r1 r3

+
r
2
2
+ 2 2 2 Tf o (0; 0; 0) 2ND2 2
P2 (B2 ) + 2D1 D3 W2
4 (0)
r1 r2

+
r
2
3
+ 2 2 2 Tho (0; 0; 0) 2ND3 2
P3 (B3 ) + 2D1 D2 W3
2
(4.1)
4 (0)
for the zero-charge sectors,
k3
1
k2
2r2 2mNToo (0; z2 ; z3 )P1 (B1 )
A(Q=1) =
8
1 (z2 ) 1 (z3 )
k2
k3
2r2 2mNT
oo (0; z2 ; z3 )P1 (B1 )

1 (z2 ) 1 (z3 )
r2 r3

+ 2 2 + 2 Tgo (0; z2 ; z3 ) 2mD1 2r1 2 P1 (B1 )

4 (z2 ) 4 (z3 )
r2 r3

+ 2 2 + 2 Tgo (0; z2 ; z3 ) 2mD

1 2r1 2 P1 (B1 )
4 (z2 ) 4 (z3 )
r+r2
k2
+ 2 2 Tf o (0; z2 ; z3 )[2imD2]
4 (0) 1 (z2 ) 4 (z3 )
r+r2
k2
+ 2 2 Tf o (0; z2 ; z3 )[2i mD
2]
4 (0) 1 (z2 ) 4 (z3 )
276
k3
4 (0) 4 (z2 ) 1 (z3 )

r+r3
k3
2
+2
Tho (0; z2 ; z3 )[2i mD
3]
4 (0) 4 (z2 ) 1 (z3 )
+2
r+r3
2
Tho (0; z2 ; z3 )[2imD3 ]
for the charge-one sectors, and
2k3
1
2k2
2r2 m2 Too (0; 2z2 ; 2z3 )P1 (B1 )
A(Q=2) =
8
1 (2z2 ) 1 (z3 )

2k2
2k3
r2 2
Too (0; 2z2 ; 2z3 )P1 (B1 )
2 m
1 (2z2 ) 1 (2z3 )
(4.2)
(4.3)
for the total charge-two sectors. The Mbius amplitude can be obtained in a similar way
from the undeformed case, distinguishing only the uncharged (Q = 0) sector from the
charged (Q = 2) one, since the Q = 1 sector is absent in M because of the oriented nature
of the corresponding open-strings. Thus
1 r6
2 2 NP1 (B1 , 1 )P2 (B2 , 2 )P3 (B3 , 3 )
M(Q=0) =
8
r1 6
+ 2 2 D1 P1 (B1 , 1 )W2 (B2 , 2 )W3 (B3 , 3 )
+2
r2 6
2
D2 W1 (B1 , 1 )P2 (B2 , 2 )W3 (B3 , 3 )

D3 W1 (B1 , 1 )W2 (B2 , 2 )P3 (B3 , 3 ) Too (0; 0; 0)
r1 2

2 2 (N + D1 )P1 (B1 , 1 ) + 21 (D2 + D3 )W1 (B1 , 1 )
2
2

Tog (0; 0; 0)
2
r2 2

2 2 (N + D2 )P2 (B2 , 2 ) + 21 (D1 + D3 )W2 (B2 , 2 )
2
2
Tof (0; 0; 0)
2

r3 2
2 2 (N + D3 )P3 (B3 , 3 ) + 21 (D1 + D2 )W3 (B3 , 3 )
2
2

Toh (0; 0; 0)
,
(4.4)
2
+2
and
r3 6
2
r2
2k3
1
2k2
M(Q=2) = 2 2 mToo (0; 2z2 ; 2z3 )P1 (B1 , 1 )
8
1 (2z2 ) 1 (2z3 )
r2
2k2
2k3
2 2 m
Too (0; 2z2 ; 2z3 )P1 (B1 , 1 )
1 (2z2 ) 1 (2z3 )
2
r1 2
2
mTog (0; 2z2 ; 2z3 )P1 (B1 , 1 )
2 (2z2 ) 2 (2z3 )
r1 2
2
m
Tog (0; 2z2 ; 2z3 )P1 (B1 , 1 )
277
2 (2z2 ) 2 (2z3 )
2k2
2
2
+ im2 Tof (0; 2z2 ; 2z3 )
2 (0) 1 (2z2 ) 2 (2z3 )
r2
2
2k2
2
i m2
2 Tof (0; 2z2 ; 2z3 )
2 (0) 1 (2z2 ) 2 (2z3 )

r2
2
r3
2
2k3
2
+ im2 2 Toh (0; 2z2 ; 2z3 )
2 (0) 2 (2z2 ) 1 (2z3 )

r3
2
2
2k3

.
i m2
2 Toh (0; 2z2 ; 2z3 )
2 (0) 2 (2z2 ) 1 (2z3 )
(4.5)
These amplitudes describe the couplings of conventional D9- and D5i -branes with an
additional set of (magnetized) D9-branes. This natural interpretation is encoded into the
four o, g, f and h projections, present in the (Q = 2) sector of the Mbius amplitudes. The
tadpole cancellation conditions may be extracted combining the transverse (tree) channel
Klein-bottle amplitude of Eq. (2.13) with the transverse annulus amplitudes

v1 r1 2 2
25
2r6 v1 v2 v3 N 2 W1 (B1 )W2 (B2 )W3 (B3 ) +

2
D1 W1 (B1 )P2 P3
A=
8
v2 v3
v2 r2 2 2
v3 r3 2 2
+
2
D2 P1 W2 (B2 )P3 +
2
D3 P1 P2 W3 (B3 )
v1 v3
v1 v2

1 + q 2 H22 1 + q 2 H32
+ 2r6 v1 v2 v3 2mm

W1 (B1 )W2 (B2 )W3 (B3 ) Too (0; 0; 0)

k2 k3
oo (0; z2; z3 ) v1 W1 (B1 )
+ 2r2 8N mToo (0; z2; z3 ) + mT
1 (z2 ) 1 (z3 )

2 Too (0; 2z2; 2z3)
+ 2r2 4 m2 Too (0; 2z2; 2z3) + m
2k2 2k3
v1 W1 (B1 )
1 (2z2 ) 1 (2z3 )

r2 r3
P1 2 2
+ 2 2 + 2 Tog (0; 0; 0) 2r1 2 2ND1 v1 W1 (B1 ) + 2D2 D3
v1 2 (0) 2 (0)

r1 r3
P2 2 2
+ 2 2 + 2 Tof (0; 0; 0) 2r2 2 2ND2 v2 W2 (B2 ) + 2D1 D3
v2 2 (0) 2 (0)

r1 r2
P3 2 2
+ 2 2 + 2 Toh (0; 0; 0) 2r3 2 2ND3 v3 W3 (B3 ) + 2D1 D2
v3 2 (0) 2 (0)
r2 r3

2
2
+ 2 2 + 2 Tog (0; z2 ; z3) 2r1 2 2mD1 v1 W1 (B1 )
2 (z2 ) 2 (z3 )
r2 r3

2
2
1 v1 W1 (B1 )
+ 2 2 + 2 Tog (0; z2; z3 ) 2r1 2 2mD
2 (z2 ) 2 (z3 )
r1 r3
2 2k2 2
+ 2 2 + 2 Tof (0; z2; z3 )2mD2
2 (0) 1 (z2 ) 2 (z3 )
278
2k2
2
2
2 (0) 1 (z2 ) 2 (z3 )
r1 r2
2 2k3
2
+ 2 2 + 2 Toh (0; z2; z3 )2mD3
2 (0) 2 (z2 ) 1 (z3 )

r1 r2
2
2
2k3
+2
2
Toh (0; z2; z3 )2mD
3
+2
,
2 (0) 2 (z2 ) 1 (z3 )
r1
r3
+ 2 2 + 2 Tof (0; z2; z3 )2mD

2
(4.6)
and with the transverse Mbius amplitudes

2 r6

oo (0; 0; 0)
2 2 Nv1 v2 v3 W1 (B1 , 1 )W2 (B2 , 2 )W3 (B3 , 3 ) T
M=
8

r1 6 v
1
D1 W1 (B1 , 1 )P2 (B2 , 2 )P3 (B3 , 3 ) Too (0; 0; 0)
+ 2 2
v2 v3
r2 6 v

2
+ 2 2
D2 P1 (B1 , 1 )W2 (B2 , 2 )P3 (B3 , 3 ) Too (0; 0; 0)
v1 v3
r3 6 v

3
+ 2 2
D3 P1 (B1 , 1 )P2 (B2 , 2 )W3 (B3 , 3 ) Too (0; 0; 0)
v1 v2
r1

+r2 +r3 1
2
4 mToo (0; z2 ; z3 ) + m
Too (0; z2; z3 )
+2
k2
k3
v1 W1 (B1 , 1 )
1 (z2 ) 1 (z3 )

r1
+ Tog (0; 0; 0) 2 2 1 (N + D1 )v1 W1 (B1 , 1 )

21
2
2
+
(D2 + D3 )P1 (B1 , 1 )
v1
2 (0) 2 (0)

r2
+ Tof (0; 0; 0) 2 2 1 (N + D2 )v2 W2 (B2 , 2 )

21
2
2
+
(D1 + D3 )P2 (B2 , 2 )
v2
2 (0) 2 (0)

r3
+ Toh (0; 0; 0) 2 2 1 (N + D3 )v3 W3 (B3 , 3 )

21
2
2
+
(D1 + D2 )P2 (B2 , 2 )
v3
2 (0) 2 (0)
r1

1

mTog (0; z2 ; z3 ) + m
Tog (0; z2; z3 ) v1 W1 (B1 , 1 )
+2 2
2
2
2 (z2 ) 2 (z3 )

2 2k2 2
+ mTof (0; z2; z3 ) m
Tof (0; z2 ; z3)
2 (0) 1 (z2 ) 2 (z3 )

2
2 2k3

+ mToh (0; z2 ; z3 ) m
.
Toh (0; z2; z3 )
2 (0) 2 (z2 ) 1 (z3 )
(4.7)
Apart from the m = m

condition, automatic for the unitary gauge group selected by
the magnetic background, all RR tadpole cancellation conditions directly tied to four-
279
dimensional non-Abelian anomalies arise from the untwisted sector. After the charges are
rescaled and parametrized in such a way that N = 2n, Di = 2di and m 2m, the resulting
RR conditions are as in Table 13, provided the signs and satisfy the same identities
(2.17) of the undeformed case. The NSNS tadpoles, canceled only at the supersymmetric
H2 = H3 point, are related to the derivatives of the BornInfeld action with respect to
the moduli, exactly as in the six-dimensional case [31]. Several choices of gauge group are
again allowed by the additional signs i and i , in Eq. (2.18). Introducing the combinations
2,o = a1 a2 a3 a1 a2 a3 ,
2,g = a1 a2 a3 a1 + a2 + a3 ,
2,f = a1 a2 a3 + a1 a2 + a3 ,
2,h = a1 a2 a3 + a1 + a2 a3 ,
(4.8)
where ai = i if = n while ai = i but ak = k , k = i if = di , the massless spectra are

encoded in

n(n no ) d1 (d1 d1 o ) d2 (d2 d2 o )
+
+
A0 + M0 = oo (0)
2
2
2
d3 (d3 d3 o )
+ mm
+
2

n(n ng ) d1 (d1 d1 g ) d2 (d2 d2 g )
+ og (0)
+
+
2
2
2
d3 (d3 d3 g )
+ mm
+
2

n(n nh ) d1 (d1 d1 h ) d2 (d2 d2 h )
+ oh (0)
+
+
2
2
2
d3 (d3 d3 h )
+ mm
+
2

n(n nf ) d1 (d1 d1 f ) d2 (d2 d2 f )
+ of (0)
+
+
2
2
2
d3 (d3 d3 f )
+ mm
+
2

r2 r3
+ gh (0) + gf (0) 2 2 + 2 (nd1 + d2 d3 )

r1 r3
+ fg (0) + f h (0) 2 2 + 2 (nd2 + d1 d3 )

r1 r2
+ hg (0) + hf (0) 2 2 + 2 (nd3 + d1 d2 )

+ oh (+) + of (+) 2r2 +r3 |k2 k3 |nm

+ oh () + of () 2r2 +r3 |k2 k3 |nm

r2 + r3

r2 r3
+ gh (+) + gf (+) 2 2 2 md1 + gh () + gf () 2 2 + 2 md
1

r1 + r3

r1 + r3
+ f h (+) 2 2 2 md2 |k2 | + fg () 2 2 2 md
2 |k2 |
280

r1 r2

r1 r2
+ hg (+) 2 2 + 2 md3 |k3 | + hf () 2 2 + 2 md
3 |k3 |
m(m 1)

+ oh (2+)
2

r2 +r3
r2
r3
r2 +r3
2|k2 k3 | + 2 2 1 |k2 k3 | + 1 + 2 2 |k2 | 2 2 |k3 |
2

m(m + 1)
+ oh (2+)
2

r2 +r3
r3
r2
r2 +r3
2
2|k2 k3 | 2 2 1 |k2 k3 | 1 2 2 |k2 | + 2 2 |k3 |

m(m 1)
+ of (2+)
2

r2 +r3
r3
r2
r2 +r3
2|k2 k3 | + 2 2 1 |k2 k3 | + 1 2 2 |k2 | + 2 2 |k3 |
2
m(m + 1)

+ of (2+)
2

r2 +r3
r3
r2
r2 +r3
2|k2 k3 | 2 2 1 |k2 k3 | 1 + 2 2 |k2 | 2 2 |k3 |
2
m(

m
1)
+ of (2)
2

r2 +r3
r2
r3
r2 +r3
2|k2 k3 | + 2 2 1 |k2 k3 | + 1 + 2 2 |k2 | 2 2 |k3 |
2

m(
m
+ 1)
+ of (2)
2

r2 +r3
r3
r2
r2 +r3
2
2|k2 k3 | 2 2 1 |k2 k3 | 1 2 2 |k2 | + 2 2 |k3 |

m(
m
1)
+ oh (2)
2

r2 +r3
r3
r2
r2 +r3
2|k2 k3 | + 2 2 1 |k2 k3 | + 1 2 2 |k2 | + 2 2 |k3 |
2
m(

m
+ 1)
+ oh (2)
2

r2 +r3
r3
r2
r2 +r3
2|k2 k3 | 2 2 1 |k2 k3 | 1 + 2 2 |k2 | 2 2 |k3 | ,
2
(4.9)
where (0), () and (2) are shorthand notations for the arguments (0, 0, 0), (0; z2 ;
z3 ) and (0; 2z2 ; 2z3 ), respectively, and the characters with non-vanishing
arguments actually denote restrictions to their massless parts.
The resulting gauge groups are reported in Table 13, while the open unoriented spectra
are displayed in Table 14. As expected from the previous discussion, chirality originates
from two different sources. The first is the chiral asymmetry in the pure magnetic sector,
due to the misalignment introduced by the combined action of magnetic backgrounds
and orbifold projections on the Mbius amplitudes. The second is the coupling between
magnetized D9-branes and D5-branes longitudinal to the magnetized complex directions,
familiar from the T-dual picture, where chiral fermions live in a natural way at brane
281
intersections [36]. Whenever potentially anomalous U (1)s are present, they call for a
generalization of the DineSeibergWitten mechanism [61], an option that, as in six
dimensions [62], requires generalized GreenSchwarz couplings in the RamondRamond
sectors [63].
4.2. Magnetized Z2 Z2 shift-orientifolds
In this section we describe the magnetized versions of the Z2 Z2 shift-orientifolds
introduced in [20] and reviewed in Section 2.3. As in the plain Z2 Z2 models of
Section 4.1, chiral matter can be obtained if open strings stretched between magnetized
D9-branes and D5-branes longitudinal to the Z 2 and/or Z 3 directions are present. An
inspection of Table 1 shows that the p3 , w2 p3 and w1 w2 p3 models are potentially chiral,
while the remaining models are not.
The p3 model requires a separate discussion, since in its undeformed version [20] it
exhibits an f -twisted RR tadpole condition, corresponding to the action of the Z2 Z2
element that fixes the T 67 -torus, one of the two along which we turn on background
magnetic fluxes (see Table 16 for the unoriented closed spectra and Table 41 for the
unoriented open spectra with complex ChanPaton charges). This tadpole condition can
no longer be satisfied if the background magnetic field is present, since some states are
lifted in mass by a term depending solely on the field strength H2 along T 67 , rather than on
the difference H2 H3 as in the case for the tadpole conditions coming from the untwisted
or from the g-twisted sectors. The resulting models are thus anomalous as string vacua,
because the magnetic deformations are, in the aforementioned sense, incompatible with the
p3 shift. The natural geometric interpretation of this phenomenon is as follows: the p3 shift
is introducing a net number of fractional branes [66] that, differently from what happens in
the remaining models, are partly longitudinal and partly orthogonal to the magnetic fields
carrying a non-vanishing twisted RR charge, whose excess can be canceled only turning
off the background magnetic field.
In the following we shall analyze the chiral and non-chiral examples with selfdual
configurations of the magnetic field, i.e., at supersymmetric points, leaving to Section 5
the discussion of models with brane supersymmetry breaking.
4.2.1. w2 p3 models
Let us first analyze in detail the w2 p3 class of orientifolds, that captures all interesting
features of the models discussed in this paper. In the presence of the NSNS two-form
Bab , the unoriented truncation of the closed spectrum is obtained adding to the halved
torus amplitude the Klein-bottle amplitude
1
P1 P2 P3 + 24 P1 W2 (B2 )W3 (B3 ) + 24 W1 (B1 )P2 W3 (B3 )
K=
8

+ 24 W1 (B1 )(1)n2 W2 (B2 )(1)m3 P3 Too
r2 r3
r1 r3
1/2
+ 2 16 2 2 2 P1 Tgo + 2 2 2 P2 Tf o
2
r
r
21 22 2 1/2
(4.10)
+2
2 W3 (B3 )Tho
.
4
282
The resulting massless unoriented closed spectra are reported in Table 17. The transversechannel amplitude
2
v2
v1 v2 v3 + 2
oo
v1 v3
2

r
r
r
r
v1
v2
22 23
21 23
+
v1 v2 v3 + 2
2
og
v2 v3
v1 v3
2

r
r
r
r
v1
v2
22 23
21 23
+
v1 v2 v3 2
2
oh
v2 v3
v1 v3

2

r
r
r
r
v1
v2
22 23
21 23
+
v1 v2 v3 2
+2
of
v2 v3
v1 v3
5
0 = 2
K
8

r2 r3
2 2
r1 r3
v1
+ 2 2 2
v2 v3
(4.11)
displays very neatly the presence of one conventional O9-plane and of the O51 - and O52 planes, while the O53 -plane of the Z2 Z2 -models in Eq. (2.14) is no longer a fixed
manifold of the combined orbifold and shifts, and is thus eliminated.
The annulus amplitude is more subtle. In the presence of Bab , H2 along T 67 and H3
along T 89 , it is again the sum of three contributions:

N 2 r6
1/2
2 P1 (B1 ) P2 (B2 ) + P2 (B2 ) P3 (B3 )
2

2mm
r6
1/2
3 (B3 )
2 P1 (B1 ) P2 (B2 ) + P2 (B2 ) P
+
2

D2
1/2
+ 1 2r1 2 P1 (B1 )W2 W3 + W3
2

D22 r2 2
1/2
1/2
+
2
W1 P2 (B2 ) + P2 (B2 ) W3 + W3
Too (0; 0; 0)
4

2

G21 2mm
2 2
r1 2 G
+2
+
+
Tog (0; 0; 0)P1(B1 )
2
2
2
2 (0)

2
r1 r3
+ 2 2 + 2 Tgo (0; 0; 0)2ND12r1 2 P1 (B1 )

4 (0)

2
r1 r3
+ 2 2 + 2 Tgg (0; 0; 0)2GG12r1 2 P1 (B1 )

3 (0)
2

r1 r3

1/4
3/4
+ 2 2 + 2 Tf o (0; 0; 0)ND22r2 2 P2 (B2 ) + P2 (B2 )
4 (0)
2

r1 r2
1/4
3/4
+ 2 2 + 2 Tho (0; 0; 0)2r32 D1 D2 W3 + W3
(4.12)
4 (0)
1
A(Q=0) =
8
283
for the Q = 0 sectors,
k3
1
k2
A(Q=1) =
2r2 2mNToo (0; z2 ; z3 )P1 (B1 )
8
1 (z2 ) 1 (z3 )
k3
k2
1 (z2 ) 1 (z3 )

2
2
+ Tog (0; z2 ; z3 ) mG mG2r12 P1 (B1 )
2 (z2 ) 2 (z3 )

2
2
+ Tog (0; z2 ; z3 ) mG mG2
r1 2 P1 (B1 )
2 (z2 ) 2 (z3 )
r2 r3

+ 2 2 + 2 Tgo (0; z2 ; z3 ) 2mD1 2r1 2 P1 (B1 )

4 (z2 ) 4 (z3 )
r2 r3

+ 2 2 + 2 Tgo (0; z2 ; z3 ) 2mD

1 2r1 2 P1 (B1 )
4 (z2 ) 4 (z3 )
r2 r3

+ 2 2 + 2 Tgg (0; z2 ; z3 ) mG1 2mG1 2r1 2 P1 (B1 )

3 (z2 ) 3 (z3 )
r2 r3

1 2r1 2 P1 (B1 )
+ 2 2 + 2 Tgg (0; z2 ; z3 ) mG1 2mG
3 (z2 ) 3 (z3 )
r r2
k2
+ 2 2 + 2 Tf o (0; z2 ; z3 )[2imD2 ]
4 (0) 1 (z2 ) 4 (z3 )
2r2 2mNT
oo (0; z2 ; z3 )P1 (B1 )
r2
+ 2 2 + 2 Tf o (0; z2 ; z3 )[2i mD
2]

k2
,
4 (0) 1 (z2 ) 4 (z3 )
(4.13)
for the Q = 1 sectors, and

A(Q=2) =
k3
1
2k2
2r2 m2 Too (0; 2z2 ; 2z3 )P1 (B1 )
8
1 (2z2 ) 21 (z3 )
2r2 m
2 Too (0; 2z2 ; 2z3 )P1 (B1 )
+ 2r1 2 m2
k2
k3
1 (2z2 ) 1 (2z3 )
2
m2
2
Tog (0; 2z2 ; 2z3 )P1 (B1 )
2
2 (2z2 ) 2 (2z3 )
m
2
Tog (0; 2z2 ; 2z3 )P1 (B1 )
2

2
2
,
2 (2z2 ) 2 (2z3 )
+ 2r1 2 m2
(4.14)
for the Q = 2 sectors, where the magnetized ChanPaton charge multiplicity is denoted
by m. The coefficients mG , mG , mG1 , mG1 , m2 and m2 must be chosen in such a
284
way that the annulus amplitudes become real. The Mbius amplitude can be deduced in a
similar way from the undeformed case, adding to the uncharged (Q = 0) contributions the
charged (Q = 2) ones. The result is
M(Q=0) =
1 r6
2 2 NP1 (B1 , 1 )P2 (B2 , 2 )P3 (B3 , 3 )
8
r1 6
+ 2 2 D1 P1 (B1 , 1 )W2 (B2 , 2 )W3 (B3 , 3 )
+2
r2 6
2

D2 W1 (B1 , 1 )P2 (B2 , 2 )W3 (B3 , 3 ) Too (0; 0; 0)
2
r1 2

og (0; 0; 0) 2
2 2 (N + D1 )P1 (B1 , 1 ) T
2
2
r2 2

2
1/2

2 2 (N + D2 )P2 (B2 , 2 ) Tof (0; 0; 0)
2
2

oh (0; 0; 0) 2
,
21 (D1 + D2 )W3 (B3 , 3 ) T
2
(4.15)
and
r2
2k3
1
2k2
M(Q=2) = 2 2 mToo (0; 2z2 ; 2z3 )P1 (B1 , 1 )
8
1 (2z2 ) 1 (2z3 )
2
2
2
r2
2
m
Too (0; 2z2 ; 2z3 )P1 (B1 , 1 )
2k2
2k3
1 (2z2 ) 1 (2z3 )
mTog (0; 2z2 ; 2z3 )P1 (B1 , 1 )
r1 2
2
m
Tog (0; 2z2 ; 2z3 )P1 (B1 , 1 )
r2
2 2 imTof (0; 2z2 ; 2z3 )
r1 2
2
2 (2z2 ) 2 (2z3 )
2k2
2 (2z2 ) 2 (2z3 )
2
2 (0) 1 (2z2 ) 2 (2z3 )
r2
2
Tof (0; 2z2 ; 2z3 )

2 (i)m

2
2k2
2
.
2 (0) 1 (2z2 ) 2 (2z3 )
(4.16)
In order to analyze in some detail the tadpole cancellation conditions, it is worth displaying
the transverse (tree) channel amplitudes. The annulus part comprises untwisted and twisted
terms
A = A U + AT ,
285
where
25
AU =
8

N2
W1 (B1 ) W2 (B2 ) + (1)n2 W2 (B2 ) W3 (B3 )
2

v1 r1 2 D12
W1 (B1 )P2 P3 + (1)m3 P3
+
2
v2 v3
2

v2 r2 2 D22
P1 W2 (B2 ) + (1)n2 W2 (B2 ) P3 + (1)m3 P3
+
2
v1 v3
4

2mm

2 (B2 )
+ 2r6 v1 v2 v3
1 + q 2 H22 1 + q 2 H32 W1 (B1 ) W
2

3 (B3 ) Too (0; 0; 0)
2 (B2 ) W
+ (1)n2 W
2r6 v1 v2 v3
r2 r3

2 2
+ 2 2 + 2 2r1 2 2ND1 v1 W1 (B1 ) Tog (0; 0; 0)
2(0) 2 (0)
r1 r3
n
+
r
2
+ 2 2 2 2 2 ND2 v2 (i) 2 W2 (B2 )

2 2
+ (i)n2 W2 (B2 ) Tof (0; 0; 0)
2(0) 2 (0)

r1 r2

2 2
1 m3
+2
m3
2
(i) P3 + (i) P3 Toh (0; 0; 0)
+2
D1 D2
v3
2(0) 2 (0)

k2 k3
oo (0; z2; z3 )
+ 2r2 8Nv1 W1 (B1 ) mToo (0; z2 ; z3 ) + mT
1 (z2 ) 1 (z3 )

r r1
+
2
og (0; z2; z3 )
+ 2 2 2 2D1 v1 W1 (B1 ) mTog (0; z2; z3 ) + mT
2
2
2 (z2 ) 2 (z3 )
r2 r3
2
2
+ 2 2 + 2 [2r1 2 2mD
1 v1 W1 (B1 )]Tog (0; z2; z3 )
2 (z2 ) 2 (z3 )
r r2
2k
2
2
2
+ 2 2 + 2 2mD2 Tof (0; z2 ; z3 )
2 (0) 1 (z2 ) 2 (z3 )
r r2
2k2
2
2
2 2 + 2 2mD
2 Tof (0; z2 ; z3)
2 (0) 1 (z2 ) 2 (z3 )
2

r2
2 Too (0; 2z2; 2z3 )
+ 2 4v1 W1 (B1 ) m Too (0; 2z2; 2z3 ) + m

2k2 2k3
,
(4.17)
1 (2z2 ) 1 (2z3 )
and

2
G2 1 2mm
25 r1 2
G
A =
2
+
Tgo (0; 0; 0)
16v1 W1 (B1 )
8
2
2
2
4(0) 4 (0)
r r1
+
2
2 2 2 v1 W1 (B1 )8GG1 Tgg (0; 0; 0)
3(0) 3 (0)
286

+ 2r1 2 16Gv1 W1 (B1 ) mG mTgo (0; z2 ; z3 ) + mG mT
go (0; z2; z3 )
4 (z2 ) 4 (z3 )

r r1
2 2 + 2 2 8G1 v1 W1 (B1 ) mG1 mTgg (0; z2 ; z3 )

+ mG1 mT
gg (0; z2 ; z3)
3 (z2 ) 3 (z3 )

r1 2
2
+2
8v1 W1 (B1 ) m2 m Tgo (0; 2z2; 2z3)

2
.
Tgo (0; 2z2; 2z3)
+ m2 m
(4.18)
4 (2z2 ) 4 (2z3 )
This is to be contrasted with the transverse Mbius amplitude, that contains only untwisted
contributions
2 r6

2 2 Nv1 v2 v3 W1e (B1 , 1 )W2e (B2 , 2 )W3e (B3 , 3 )
M=
8
r1 6 v1
+2 2
D1 W1e (B1 , 1 )P2e (B2 , 2 )P3e (B3 , 3 )
v2 v3

r2 6 v2
+2 2
D2 P1e (B1 , 1 )W2e (B2 , 2 )P3e (B3 , 3 ) Too (0; 0; 0)
v1 v3
r1
2
2
+ 2 2 1 (N + D1 )v1 W1e (B1 , 1 )Tog (0; 0; 0)
2(0) 2 (0)
r2
+ 2 2 1 (N + D2 )v2 W2e (B2 , 2 )Tof (0; 0; 0)
2
2
2(0) 2 (0)
21
2
2
(D1 + D2 ) B3 P3e (B3 , 3 )Toh (0; 0; 0)
v3
2(0) 2 (0)

r
+ 2 2 1 4v1 W1e (B1 , 1 ) mToo (0; z2; z3 ) + m
Too (0; z2; z3)
k3
k2
1 (z2 ) 1 (z3 )
r1

Tog (0; z2; z3 )
+ 2 2 1 v1 W1e (B1 , 1 ) mTog (0; z2 ; z3 ) + m
r2
2
2
2 (z2 ) 2 (z3 )
+ 2 2 mTof (0; z2 ; z3 )
2 2k2 2
2 (0) 1 (z2 ) 2 (z3 )

2k2
2
2

,
2 m
Tof (0; z2; z3 )
2 (0) 1 (z2 ) 2 (z3 )
r2
2
(4.19)
where B3 is a suitable phase that depends on the rank of Bab , not directly relevant for our
discussion. The untwisted tadpole cancellation conditions are related to the superposition
A
and M,
and can be obtained as follows. The residues corresponding to the RR
of K,
part of the 0 character are

r

1 4 2 2 q 2 H2 H3
v1 v2 v3 2 2 N + (m + m)

2 2 2
+ (m m)8i
q (H2 + H3 ) 32

r2 +r3
v1 r1
+ 1
2 2 D1 32 2 2
v2 v3

r1 +r3
v2 r2
2 2 D2 32 2 2
= 0,
+ 2
v1 v3
287
(4.20)
together with the complex conjugates, where 1 is +1 for = o, g and is 1 for = h, f

while 2 is +1 for = o, f and is 1 for = g, h. In order to obtain Eq. (4.20), the
conditions in (2.17) for the signs and must be used, and, in order to ensure the
vanishing of the imaginary part the numerical constraint, m = m
must also be enforced.
Using the Dirac quantization condition in Eq. (3.1), it is interesting to notice that the
magnetized D9-branes contribute not only to the tadpole of the RR ten-form, but also
to the tadpole of the RR six-form. As in the six-dimensional examples, this signals the
phenomenon of brane transmutation. In particular, disentangling the diverse contributions,
one obtains
r

v1 v2 v3 2 2 (N + m + m)
= v1 v2 v3 32
(4.21)
for the D9-brane sector,

v
r2 +r3
r
v1 r1
1
2
2
=
2 D1 + 2 |k2 k3 |(m + m)
32 2 2
v2 v3
v2 v3
for the D51 -brane sector and

r1 +r3
v2 r2
v2
2
2 D2 =
32 2 2
v1 v3
v1 v3
(4.22)
(4.23)
for the D52 -brane sector.

The twisted tadpole conditions determine the nature of the allowed ChanPaton charges.
There are two options, that result in complex or real ChanPaton charges. In the complex
case, one must choose
mG = mG = mG1 = mG1 = i,
m2 = m2 = 1,
and the tadpole cancellation condition can be written

rr1

2

8 2rr1 +1 [G + im i m]
= 0,
2 + 2 2G1 + 2 2 (G + im i m)
(4.24)
while the real charges are determined by the choice

mG = mG = mG1 = mG1 = m2 = m2 = 1,
and the corresponding tadpole cancellation condition can be written in the form
rr1

2

8 2rr1 +1 [G + m + m]
= 0.
2 + 2 2G1 + 2 2 (G + m + m)
(4.25)
288
The analysis of the open spectra is very similar to the one in Section 2, and therefore we
shall not repeat it here. With complex charges, after the choice of signs in Eq. (2.18), one
has to fix
2 3 = 2 3 = 1,
(4.26)
while 1 and 1 are free signs. In order to obtain amplitudes with a proper particle
interpretation, the magnetic charges must be rescaled by a factor of two, and a suitable
parametrization of the ChanPaton multiplicities is
N = 2(n + n),
D1 = 2(d + d),
G = 2i(n n),
G1 = 2i(d d),
D2 = 4d2.
(4.27)
With this choice, the tadpole cancellation conditions are reported in Table 18, together
with corresponding options for the ChanPaton gauge groups. The resulting open spectra
at the supersymmetric point are reported in Table 19, and chirality emerges again both at
brane intersections and due to the chiral asymmetry present in the pure magnetic sector. It
should be noticed that, as in [31], the Mbius-strip amplitudes must be suitably interpreted,
since naively they are not compatible with the corresponding annulus amplitudes. As is
familiar from rational models, some missing parts must be identified with differences of
pairs of identical terms, one symmetrized and the other anti-symmetrized by the action
of the open twist [2,4,54]. The real-charge solutions, present only if the B-rank is nonvanishing, correspond to the choice
2 3 = 2 3 = 1,
(4.28)
with 1 and 1 again free signs. In this case after rescaling the magnetic charge m by a
factor of two, a suitable parametrization for the ChanPaton multiplicities is
N = 2(n1 + n2 ),
G = 2(n1 n2 ),
D1 = 2(d1 + d2 ),
G1 = 2(d1 d2 ),
D2 = 4d3 .
(4.29)
Table 20 displays the tadpole cancellation conditions and the allowed ChanPaton gauge
groups, while the resulting chiral open spectra are exhibited in Table 21.
To conclude, let us mention that at the supersymmetric point, i.e., for self-dual
configurations of the background magnetic fields, the tadpoles originating from the NSNS
sectors are also automatically canceled. As in Ref. [31], they can be traced to corresponding
derivatives of the BornInfeld-type action for the untwisted sectors (the dilaton tadpole, for
instance, is one of them). Moreover, the twisted NSNS tadpoles are subtle: they are not
perfect squares because of the behaviour of the magnetic field under time reversal [4,31].
Still, they introduce additional couplings in the twisted NSNS sectors that are proportional
to H2 H3 and are thus canceled at the (self-dual) supersymmetric point.
4.2.2. w1 w2 p3 models
Another interesting class of chiral orientifolds can be derived from deformations of the
w1 w2 p3 models. Since this is very similar to the w2 p3 case, we shall not perform a detailed
description of all the amplitudes as in Section 4.2.1, but we shall just quote the results. The
289
Klein bottle amplitude
1
P1 P2 P3 + 24 P1 W2 (B2 )W3 (B3 ) + 24 W1 (B1 )P2 W3 (B3 )
K=
8

+ 24 (1)n1 W1 (B1 )(1)n2 W2 (B2 )(1)m3 P3 Too
r2 r3
r1 r3
1/2
1/2
+ 2 16 2 2 2 P1 Tgo + 2 2 2 P2 Tf o
2
r1 r2
1/2
+ 2 2 2 22 W3 (B3 )Tho
,
4
(4.30)
produces the unoriented closed spectra, whose massless part is reported in Table 22. It
should be noticed that the result does not depend on the rank of Bab , in agreement with
the considerations made in Section 2.1 relating quantized values of Bab to shifts. An S
transformation of (4.30) yields the transverse channel amplitude, that at the origin of
the lattice sums is identical to Eq. (4.11). As a result, the models contain O9+ , O51+
and O52+ planes. In order to neutralize the RR charge, (magnetized) D9-branes, D51 branes and D52 -branes are introduced. Due to the triple shifts, the tadpole cancellation
conditions derive only from the untwisted sectors, and their analysis is very similar to the
one in Section 4.2.1. After a proper normalization, the ChanPaton charge multiplicities
are displayed in Table 23, where the allowed ChanPaton gauge groups are also reported.
Apart from the m charges, all others are real, as emerges from the open unoriented chiral
spectra shown in Table 24.
4.2.3. Non-chiral models
In this section we discuss the remaining models in Table 1 that admit magnetic
deformations, namely those containing D5-branes along T 45 that can absorb the RR
charge flux of the magnetized D9-branes. It is easy to see that the p2 p3 , w1 p2 , w1 p2 p3
and w1 p2 w3 models do admit magnetic deformations, while the p1 p2 p3 , p1 w2 w3 and
w1 w2 w3 do not. The four models are quite different, but inherit an effective world-sheet
parity projection that allows one to express the Klein bottle amplitude in the following
form:
1
P1 P2 P3 + 24 P1 W2 (B2 )W3 (B3 ) + 24 W1 (B1 )(1)2 P2 (1)3 W3 (B3 )
K=
8

+ 24 (1)1 W1 (B1 )(1)2 W2 (B2 )(1)3 P3 Too
+ 2 16 2
r2 r3
2 2

1
P1 Tgo
2
,
(4.31)
where 1 is 0 if 1 = p1 and 12 if 1 = w1 , while obviously the shifts affect the sums only if
they are present in the corresponding table and are of the same type as the lattice sums.
The transverse channel gives the O-plane content in the four cases, that is expected to be
290
the same for the four classes of models. Indeed, at the origin of the lattices
2

5
r2 r3
v
2
0 =
v1 v2 v3 + 2 2 2
(oo + og )
K
8
v2 v3
2

r
r
v1
22 23
v1 v2 v3 + 2
(oh + of ) ,
+
v2 v3
(4.32)
so that in all four classes of models only O9+ and O51+ are present. On the other hand,
both the unoriented closed spectra and the open sectors are quite distinct, as can be deduced
from the diverse D-brane multiplet configurations of the undeformed models in Table 41.
Of course, only (magnetized) D9-branes and D51 -branes are needed, with a consequent
lack of chirality. The unoriented closed spectra of the four classes of models can be found
in Tables 25, 26 and 27. Only the p2 p3 models show a dependence on the rank of Bab ,
while the two models with three shifts have identical massless closed spectra.
There are two different p2 p3 unoriented partition functions, that differ in the openstring sectors, depending on the sign freedom for the Mbius projections. With complex
and properly normalized charges, untwisted and twisted tadpole cancellation conditions
are summarized in Table 28, where the resulting ChanPaton gauge groups are also
reported. The open spectra can be read from Table 29, where R stands for the symmetric
representation if 1 = +1, or for the anti-symmetric representation if 1 = 1. The second
solution is linked to a real parametrization of the ChanPaton charges that results into
tadpole cancellation conditions and gauge groups as in Table 30. It should be noticed that
in this case group factors, other than U (m), must be all orthogonal or all symplectic. The
massless open spectra can be found in Table 31.
The w1 p2 models and the w1 p2 p3 models are very similar and, independently of the
presence of Hi , differ solely in their massive excitations. In other words, the analysis of
the massless excitations is not sufficient to distinguish these two classes of models. Their
unoriented closed spectra are different, as emerges from Tables 26 and 27, but they have
identical open spectra. The tadpole cancellation conditions and the resulting ChanPaton
groups are reported in Table 32 for complex charges and in Table 34 for real charges. The
non-chiral and coincident open spectra are reported in Table 33 for the complex charge
cases, and in Table 35 for the real charge cases.
Finally, the w1 p2 w3 models exhibit unoriented bulk spectra identical to the one of the
w1 p2 p3 models in Table 27, but with tadpole conditions and ChanPaton groups as in
Table 36. The resulting non-chiral massless open spectra are contained in Table 37.
5. Brane supersymmetry breaking

In this section we discuss one significant class of magnetized orientifolds in which
the field configurations are chosen so that magnetized D9-branes mimic anti-D5-branes
rather than D5-branes, thus breaking supersymmetry in the open-string sector (brane
supersymmetry breaking). We analyze in some detail a variant of the w2 p3 class of models
extensively discussed at the supersymmetric point in Section 4.2.1. For simplicity, we shall
291
confine ourselves to the Bab = 0 case. The oriented closed spectrum is always the one
contained in Table 15, but the Klein-bottle projection, described by

1
P1 P2 P3 + P1 W2 W3 + W1 P2 W3 + W1 (1)n2 W2 (1)m3 P3 Too
K=
8

2

1/2
2 16 Tgo P1 Tf o P2 1/2 + W3 Tho
(5.1)
,
4
is now different, due to the inversion of some signs, and produces the massless unoriented
closed spectra of Table 38. The transverse channel amplitude at the lattice origin,
2

25
v1
v2

v1 v2 v3
oo
K0 =
8
v2 v3
v1 v3
2

v1
v2
+
v1 v2 v3
+
og
v2 v3
v1 v3
2

v1
v2
+
v1 v2 v3 +
+
oh
v2 v3
v1 v3

2

v1
v2
+
v1 v2 v3 +
of ,
(5.2)
v2 v3
v1 v3
displays very neatly the presence of one O9+ -plane and of the two exotic O51 and O52
planes, that require the introduction of anti-D51 -branes and anti-D52 -branes, together with
the (magnetized) D9-branes. In order to neutralize the global RR charge, one has to
sit at the anti-self-dual background field configuration, corresponding to H2 = H3 in our
conventions. Only the RR tadpole cancellation conditions can be imposed, while the NS
NS tadpoles survive, signaling, as customary, the need for a non-Minkowskian vacuum
[65]. The results for the ChanPaton gauge groups are displayed in Table 39, while the open
and unoriented massless spectra are displayed in Table 40. As usual, in the models with
brane supersymmetry breaking supersymmetry is exact at tree level on the D9-branes but
it is only non-linearly realized on the anti-D5-branes [58,59]. This can be foreseen from
Table 40, where modes originally in a given supermultiplet are assigned to different gauge
group representations.
6. Conclusions
In this paper we have analyzed in detail four-dimensional orientifolds originating from
Z2 Z2 toroidal orbifolds and from freely acting Z2 Z2 shift-orbifolds of the type-IIB
superstring, in the presence of uniform background magnetic fluxes along four of the six
internal directions and of a quantized NSNS Bab , that has been shown to be equivalent
to an asymmetric shift-orbifold projection. These models are connected by T-duality to
models with branes intersecting at angles and contain magnetized D9-branes charged also
with respect to the RR six-form, thus exhibiting several interesting novelties. In particular,
292
for suitable self-dual configurations of the internal backgrounds, that in the T-dual picture
correspond to suitable angles between the branes, it is possible to obtain non-tachyonic
four-dimensional supersymmetric models with spectra containing in a natural way several
families of matter fields whose numbers are related to the multiplicities of the Landau
levels. Moreover, the instanton-like behaviour of the magnetized D9-branes that mimic
localized D5-branes produces an interesting rank reduction of the ChanPaton gauge
groups. As a bonus, if D5-branes longitudinal to the directions of the internal magnetic
fields are present, the models can acquire chiral spectra, due to the unpairing of fermions at
the intersections and to the chiral asymmetry in the pure magnetic sectors. Geometrically,
chirality is related to configurations in which D-branes are not parallel to the corresponding
O-planes, differently from the models of Ref. [60], where the D9-branes are parallel to the
O9-planes, but the orbifold projection produces only left-handed fermions. Introducing
anti-self-dual background fields, it is also possible to obtain non-tachyonic models with
brane supersymmetry breaking, for which supersymmetry is exact at tree level in the
bulk and on the D9-branes, but is non-linearly realized and thus effectively broken at the
string scale on the anti-D5-branes and on the equivalent magnetized D9-branes. The chiral
four-dimensional models can be used to build realistic extensions of the Standard Model
in a brane-world-like scenario, introducing braneanti-brane pairs or Wilson lines. This is
a very interesting and widely pursued effort, but the dynamical stability of all these vacua
is still an open question. It would be also interesting to analyze in some detail mechanisms
to reduce the number of moduli, for instance, introducing background fluxes, as recently
discussed in Ref. [44].
Acknowledgements
It is a pleasure to acknowledge C. Angelantonj, P. Bantay, M Berg, M. Bianchi,
R. Blumenhagen, E. Dudas, L. Genovese, M. Haack, J. Mourad, Ya.S. Stanev and
A.M. Uranga for stimulating discussions and especially A. Sagnotti for the collaboration
at the early stages of this work and for several illuminating discussions. G.P. is grateful to
the Theoretical Physics Department of the Etvs Lornd University of Budapest for the
kind hospitality while this work was being completed. This work was supported in part by
I.N.F.N., by the E.C. RTN programs HPRN-CT-2000-00122 and HPRN-CT-2000-00148,
by the INTAS contract 99-1-590, by the MIUR-COFIN contract 2001-025492 and by the
NATO contract PST.CLG.978785.
Appendix A. Lattice sums in the presence of a quantized Bab

In this appendix we collect the relevant lattice sums that enter the one-loop partition
functions. We follow mainly the notation of [4], and display only the sums modified by the
presence of an anti-symmetric tensor Bab . Since each surface of vanishing Euler number
has a different double cover, the sums also differ in their proper time dependence. We will
denote with the loop channel modulus of each surface and with ; the modulus of the
doubly covering tori. Let us begin by recalling that, in presence of a Bab background, the
293
generalized d-dimensional momenta pL and pR are [64]:

1
(gab Bab )nb ,

1
pR,a = ma (gab + Bab )nb .
The corresponding lattice sums on the torus take the form

pL,a = ma +
(B) =
(A.1)
(A.2)
q 4 pLT g 1 pL q 4 pRT g 1 pR
|( )|2d
m,n
(A.3)
as in Ref. [64]. For the direct-channel Klein-bottle amplitudes, only the winding sums are
modified and become
1
W (B) =
2i T
B
q 2 n
=0,1 n
T gn
e n
2 (2i )
(A.4)
while in the transverse channel the momentum sums are

P (B) =
(e2; )
m+ 1 B T g 1 m+ 1 B

2 (i;)
=0,1 m
(A.5)
In the annulus amplitudes the situation is reverted, and modified momentum sums
P (B) =
q
T 1

1
m+ 1 B
2 m+ B g
2 (i/2)
=0,1 m
(A.6)
appear in the direct channel, while modified winding sums

(e2; ) 4 n gn e
2 (i;)
n
1
W (B) =
2i T
n B
(A.7)
=0,1
appear in the transverse channel. The direct Mbius amplitudes involve

P (B, ) =
q
T 1

1
m+ 1 B
2 m+ B g
2 i2
=0,1 m
1
2
(A.8)
and
q 2 nT gn e nT B
W (B, ) =
,

2 i2 + 12
=0,1 n
1
2i
(A.9)
while the transverse Mbius amplitudes involve

(e2; ) 4 n gn e
2 (i;)
n
1
W (B, ) =
=0,1
2i T
n B
(A.10)
294
(e2; ) (m+ B)T g 1 (m+ B)

1
P (B, ) =
=0,1 m
2 (i;)
(A.11)
All sums displayed in this appendix are two-dimensional, since for simplicity the sixdimensional internal torus is chosen to be factorized as a product of two-dimensional
tori, while the corresponding anti-symmetric two-tensor is also chosen, for simplicity, in a
block-diagonal form of two-by-two matrices.
Appendix B. Characters for the T 6 /Z2 Z2 orbifolds
In this appendix we list the Z2 Z2 characters that enter the one-loop amplitudes. Using
the conventions of Ref. [4], they may be expressed as ordered products of the four SO(2)
level-one characters, O2 , V2 , S2 and C2 , as follows:
oo = V2 O2 O2 O2 + O2 V2 V2 V2 S2 S2 S2 S2 C2 C2 C2 C2 ,
og = O2 V2 O2 O2 + V2 O2 V2 V2 C2 C2 S2 S2 S2 S2 C2 C2 ,
oh = O2 O2 O2 V2 + V2 V2 V2 O2 C2 S2 S2 C2 S2 C2 C2 S2 ,
of = O2 O2 V2 O2 + V2 V2 O2 V2 C2 S2 C2 S2 S2 C2 S2 C2 ,
go = V2 O2 S2 C2 + O2 V2 C2 S2 S2 S2 V2 O2 C2 C2 O2 V2 ,
gg = O2 V2 S2 C2 + V2 O2 C2 S2 S2 S2 O2 V2 C2 C2 V2 O2 ,
gh = O2 O2 S2 S2 + V2 V2 C2 C2 C2 S2 V2 V2 S2 C2 O2 O2 ,
gf = O2 O2 C2 C2 + V2 V2 S2 S2 S2 C2 V2 V2 C2 S2 O2 O2 ,
ho = V2 S2 C2 O2 + O2 C2 S2 V2 C2 O2 V2 C2 S2 V2 O2 S2 ,
hg = O2 C2 C2 O2 + V2 S2 S2 V2 C2 O2 O2 S2 S2 V2 V2 C2 ,
hh = O2 S2 C2 V2 + V2 C2 S2 O2 S2 O2 V2 S2 C2 V2 O2 C2 ,
hf = O2 S2 S2 O2 + V2 C2 C2 V2 C2 V2 V2 S2 S2 O2 O2 C2 ,
f o = V2 S2 O2 C2 + O2 C2 V2 S2 S2 V2 S2 O2 C2 O2 C2 V2 ,
fg = O2 C2 O2 C2 + V2 S2 V2 S2 C2 O2 S2 O2 S2 V2 C2 V2 ,
f h = O2 S2 O2 S2 + V2 C2 V2 C2 C2 V2 S2 V2 S2 O2 C2 O2 ,
ff = O2 S2 V2 C2 + V2 C2 O2 S2 C2 V2 C2 O2 S2 O2 S2 V2 .
(B.1)
Appendix C. Massless spectra

This appendix collects the massless spectra of all the models in the paper. N indicates
the number of supersymmetries and H , V , C and CL,R denote hypermultiplets, vector
multiplets and chiral multiplets with a Majorana or a Weyl (left or right) spinor,
respectively. CY(h11 , h12 ) is referred to the fact that the related orbifold is a singular limit
of a CalabiYau compactification with Hodge numbers (h11 , h12 ), is the discrete torsion
295
while i are signs that satisfy 1 2 3 = (Section 2). ki are the integer multiplicities of
the Landau level degeneracies, rj is the rank of the two-by-two j th block of the Bab NS
NS anti-symmetric tensor and r = r1 + r2 + r3 is the total B rank. The i and i , introduced
in Eq. (2.18) can be 1, and both choices are allowed if their values are not specified.
For what concerns the ChanPaton gauge groups, F , S, A and Adj denote, respectively,
the fundamental, symmetric, anti-symmetric and adjoint representations. When two Chan
Paton groups or two multiplets are within brackets, one of the two factors can be chosen
independently. Finally, the notation related to the ChanPaton charges or to the number of
branes reserves the ns to the uncharged D9-branes, the ms to the magnetized D9-branes
and the ds to the D5i -branes.
C.1. Closed spectra of the Z2 Z2 orbifolds
Table 2
Oriented closed spectra of Z2 Z2 orbifolds
Untwisted
SUGRA
Untwisted
H
Untwisted
V
Twisted
H
Twisted
V
+1
1
N =2
N =2
1+3
1+3
3
3
16 + 16 + 16
0
0
16 + 16 + 16
CY(3, 51)
CY(51, 3)
Table 3
Unoriented closed spectra of the Z2 Z2 orbifolds
Untwisted
SUGRA
(1 , 2 , 3 )
(+, +, +)
(+, , )
(, +, )
(, , +)
+
+
+
+
N
N
N
N
(, , )
(+, +, )
(+, , +)
(, +, +)
N
N
N
N
Untwisted
C
Twisted
C
Twisted
V
=1
=1
=1
=1
1+3+3
1+3+3
1+3+3
1+3+3
16 + 16 + 16
16
16
16
0
16 + 16
16 + 16
16 + 16
=1
=1
=1
=1
1+3+3
1+3+3
1+3+3
1+3+3
16 + 16 + 16
16 + 16 + 16
16 + 16 + 16
16 + 16 + 16
0
0
0
0
Table 4
Unoriented closed spectra of the Z2 Z2 orbifolds with i = +1
B rank
Untwisted
Untwisted
Twisted
Twisted
V
r1
r2
r3
SUGRA
N =1
1+3+3
16 + 16 + 16
2
0
0
0
2
0
0
0
2
N =1
N =1
N =1
1+3+3
1+3+3
1+3+3
16 + 12 + 12
12 + 16 + 12
12 + 12 + 16
0+4+4
4+0+4
4+4+0
2
0
2
2
2
0
0
2
2
N =1
N =1
N =1
1+3+3
1+3+3
1+3+3
12 + 12 + 10
10 + 12 + 12
12 + 10 + 12
4+4+6
6+4+4
4+6+4
N =1
1+3+3
10 + 10 + 10
6+6+6
296
C.2. Open spectra of the Z2 Z2 orientifolds with i = +1

Table 5
ChanPaton groups and tadpole conditions for the [T 2 T 2 T 2 ]/
Z2 Z2 models

USp(d2 )
USp(d3 )
USp(n)
USp(d1 )
SO(n)
SO(d1 )
SO(d2 )
SO(d3 )
n = d1 = d2 = d3 = 16 2r/2
Table 6
Open spectra of the Z2 Z2 orientifold with = 1
Multiplets
C
C
C
C
Number

3
1
2
0
or
if USp,
or
if SO
0
2
1
3

3
1
2
0
or
if USp,
or
if SO
0
2
1
3

3
1
2
0
or
if USp,
or
if SO
0
2
1
3

3
1
2
0
or
if USp,
or
if SO
0
2
1
3

r2 +r3
2
r1 +r3
2 2
r1 +r1
2 2
(A, 1, 1, 1)
(S, 1, 1, 1)
(1, A, 1, 1)
(1, S, 1, 1)
(1, 1, A, 1)
(1, 1, S, 1)
(1, 1, 1, A)
(1, 1, 1, S)

(F, F, 1, 1), (1, 1, F, F )
Rep.
(F, 1, F, 1), (1, F, 1, F )

(F, 1, 1, F ), (1, F, F, 1)
C.3. Spectra of the [T 2 (H2 ) T 2 (H3 )]/Z2 orientifolds

Table 7
Oriented closed spectra of the [T 2 T 2 ]/Z2 orbifolds
Untwisted
SUGRA
Untwisted
T
Twisted
T
N = (2, 0)
1+4
16
Table 8
Unoriented closed spectra of the [T 2 T 2 ]/Z2 orientifolds
B rank
r
Untwisted
SUGRA
Untwisted
H
Untwisted
T
Twisted
H
Twisted
T
0
2
4
N = (1, 0)
N = (1, 0)
N = (1, 0)
4
4
4
1
1
1
16
14
10
0
2
6
Table 9
ChanPaton groups and tadpole conditions for the [T 2 T 2 ]/Z2
models (complex charges)
U (n) U (d) U (m)
n + n + m + m
= 32 2r/2
r/2
d + d + 2 |k2 k3 |(m + m)
= 32 2r/2
n = n,
d = d, m = m
Table 10
Open spectra of the [T 2 (H2 ) T 2 (H3 )]/Z2 orientifolds (complex charges)
Multiplets
Number
Rep.
H
H
H
H
1, 1)
(A + A,
1)
(1, A + A,
(F, 1, F )
(F , 1, F )
1
1
2r |k2 k3 | 4
2r |k2 k3 | + 4
r

2 + 2r/2 |k2 k3 | 2
r

2 2r/2 |k2 k3 |
H
H
2r/2
2r/2
(F, F , 1)
(1, F , F )
(1, 1, A)
(1, 1, S)
Table 11
ChanPaton groups and tadpole conditions for the [T 2 T 2 ]/Z2
models (real charges)
USp(n1 ) USp(n2 ) USp(d1 ) USp(d2 ) U (m)
n1 + n2 + m + m
= 32 2r/2
d1 + d2 + 2r/2 |k2 k3 |(m + m)
= 32 2r/2
n2 = n1 + m + m,
d1 = d2 , m = m
Table 12
Open spectra of the [T 2 (H2 ) T 2 (H3 )]/Z2 orientifolds (real charges)
Multiplets
Number
Rep.
H
H
(F, F, 1, 1, 1)
(1, 1, F, F, 1)
H
H
1
1

r
2 + 2r/2 |k2 k3 | 2
r

2 2r/2 |k2 k3 |
2r 2|k2 k3 | + 2
(1, 1, 1, 1, S)
(F, 1, 1, 1, F )
2r 2|k2 k3 | 2
(1, F, 1, 1, F )
2 2 1
2 2 1
(1, F, 1, F, 1)
22
(1, 1, F, 1, F )
r
r
(1, 1, 1, 1, A)
(F, 1, F, 1, 1)
C.4. Open spectra of the magnetized Z2 Z2 orientifolds with i = +1

Table 13
ChanPaton groups and tadpole conditions for the magnetized [T 2 T 2
T 2 ]/Z2 Z2 models

USp(d2 )
USp(d3 )
USp(n)
USp(d1 )
U (m)
SO(n)
SO(d1 )
SO(d2 )
SO(d3 )
r
n+m+m
= 16 2 2
r2 r3
r
d1 + 2 2 + 2 |k2 k3 |(m + m)
= 16 2 2
r
r
d2 = 16 2 2 , d3 = 16 2 2 , m = m
297
298
Table 14
Open spectra of the magnetized Z2 Z2 orientifolds with = 1
Multiplets
Number

3
1
2
0
or
if USp,
or
if SO
0
2
1
3

3
1
2
0
or
if USp,
or
if SO
0
2
1
3

3
1
2
0
or
if USp,
or
if SO
0
2
1
3

3
1
2
0
or
if USp,
or
if SO
0
2
1
3
3
C
C
C
C
C
r2 +r3
2
r1 +r3
2 2
r1 +r2
2 2
r2 +r3
2 2
C
C
C
CL
CL
CL
(F, F, 1, 1, 1), (1, 1, F, F, 1)
CL
Rep.

(A, 1, 1, 1, 1)
(S, 1, 1, 1, 1)

(1, A, 1, 1, 1)
(1, S, 1, 1, 1)

(1, 1, A, 1, 1)
(1, 1, S, 1, 1)

(1, 1, 1, A, 1)
(1, 1, 1, S, 1)
(1, 1, 1, 1, Adj)
(F, 1, F, 1, 1), (1, F, 1, F, 1)

(F, 1, 1, F, 1), (1, F, F, 1, 1)
(1, F, 1, 1, F + F )
(F, 1, 1, 1, F + F )
2r2 +r3 |k2 k3 |

r2 +r3
2
r2 +r3
r
+r
+1
2 2 3 |k2 k3 | 2 2
r2 +r3
2r2 +r3 +1 |k2 k3 | + 2 2
r2 +r3
2r2 +r3 +1 |k2 k3 | 2 2
2r2 +r3 +1 |k2 k3 | + 2
r2
2
r2
1 |k2 k3 | 1 2 2
r2
1 |k2 k3 | + 1 2 2
r2
1 |k2 k3 | 1 + 2 2
1 |k2 k3 | + 1 + 2
r+r2
2 2
r+r3
2 2
CL
CR
r3
2
r3
|k2 | + 2 2
r3
|k2 | + 2 2
r3
|k2 | 2 2
|k2 | 2
|k3 |
(1, 1, 1, 1, A)
|k3 |
(1, 1, 1, 1, S)
|k3 |
(1, 1, 1, 1, A)
|k3 |
(1, 1, 1, 1, S)
|k2 |
(1, 1, F, 1, F )
|k3 |
(1, 1, 1, F, F )
C.5. Oriented closed spectra of the Z2 Z2 shift-orientifolds
Table 15
Oriented closed spectra of the Z2 Z2 shift-orbifolds
Model
Untwisted
SUGRA
Untwisted
H
Untwisted
V
Twisted
H
Twisted
V
p3
N =2
1+3
16
16
CY(19, 19)
p2 p3
w 2 p3
w 1 p2
N =2
N =2
N =2
1+3
1+3
1+3
3
3
3
8
8
8
8
8
8
CY(11, 11)
CY(11, 11)
CY(11, 11)
=2
=2
=2
=2
=2
=2
1+3
1+3
1+3
1+3
1+3
1+3
3
3
3
3
3
3
0
0
0
0
0
0
0
0
0
0
0
0
CY(3, 3)
CY(3, 3)
CY(3, 3)
CY(3, 3)
CY(3, 3)
CY(3, 3)
p1 p2 p3
p1 w 2 w 3
w 1 p2 p3
w 1 p2 w 3
w 1 w 2 p3
w1 w2 w3
N
N
N
N
N
N
299
C.6. Unoriented closed spectra of the p3 models

Table 16
Unoriented closed spectra of the p3 models
B rank
r1
r2
Untwisted
Untwisted
Twisted
Twisted
r3
SUGRA
N =1
1+3+3
16 + 16
2
0
0
0
2
0
0
0
2
N =1
N =1
N =1
1+3+3
1+3+3
1+3+3
14 + 14
14 + 14
12 + 12
2+2
2+2
4+4
2
0
2
2
2
0
0
2
2
N =1
N =1
N =1
1+3+3
1+3+3
1+3+3
12 + 12
11 + 11
11 + 11
4+4
5+5
5+5
N =1
1+3+3
10 + 10
6+6
C.7. Orientifolds of the w2 p3 models

Table 17
Unoriented closed spectra of the w2 p3 models
B rank
r2 + r3
Untwisted
SUGRA
Untwisted
C
Twisted
C
Twisted
V
N =1
1+3+3
8+8
2
4
N =1
N =1
1+3+3
1+3+3
6+6
5+5
2+2
3+3
Table 18
ChanPaton groups and tadpole conditions for the
w2 p3 models (complex charges)

USp(d2 )
U (m)
U (n) U (d1 )
SO(d2 )
r
n + n + m + m
= 16 2 2
r2 r3
r
2 + 2 |k2 k3 |(m + m)
= 16 2 2
r
d2 = 8 2 2 , n = n,
d1 = d1 , m = m
d1 + d1 + 2
Table 19
Open spectra of the w2 p3 models (complex charges)
Multiplets
Number
Rep.
(Adj, 1, 1, 1), (1, Adj, 1, 1)

(1, 1, 1, Adj)

1, 1), (S + S,
1, 1, 1)
(1, S + S,
(A + A, 1, 1, 1), (1, A + A, 1, 1)
(1, 1, A, 1) or (1, 1, S, 1)
(F, 1, 1, F ), (F , 1, 1, F )
(F , 1, 1, F ), (F, 1, 1, F )
(continued on next page)
C
C
C
C

2
0
if USp,
if SO
0
2
3
2r2 +r3 |k2 k3 | + 2
2r2 +r3 |k2 k3 | 2
300
Table 19 (Continued)
Multiplets
Number
C
C
CL
CL
CL
CL
CL
Rep.
r2 +r3
22 2
r2 +r3
22 2
r2 +r3
2r2 +r3 2|k2 k3 | + 1 + 2 2
r2 +r3
2r2 +r3 2|k2 k3 | + 1 2 2
r2 +r3
2r2 +r3 2|k2 k3 | + 1 + 2 2
r2 +r3
2r2 +r3 2|k2 k3 | + 1 2 2
(F, F, 1, 1), (F , F , 1, 1)
(1, F, 1, F ), (1, F , 1, F )
1 |k2 k3 | + 1
1 |k2 k3 | 1
1 |k2 k3 | + 1
1 |k2 k3 | 1
r2
+2 2
r2
2 2
r2
2 2
r2
+2 2
|k2 |
(1, 1, 1, A)
|k2 |
(1, 1, 1, S)
|k2 |
(1, 1, 1, A)
|k2 |
(1, 1, 1, S)
r1 +r3
2 2 +r2 2|k2 |
(1, 1, F, F )
Table 20
ChanPaton groups and tadpole conditions for the w2 p3 models (real charges)

USp(d3 )
USp(n1 ) USp(n2 ) USp(d1 ) USp(d2 )
U (m)
SO(n1 ) SO(n2 ) SO(d1 ) SO(d2 )
SO(d3 )
r
n1 + n2 + m + m
= 16 2 2
r2 r3
r
d1 + d2 + 2 2 + 2 |k2 k3 |(m + m)
= 16 2 2
r
d3 = 8 2 2 , n2 = n1 + m + m,
d1 = d1 , m = m
Table 21
Open spectra of the w2 p3 models (real charges)
Multiplets
Number
C
C
1
2

1
0
if USp,
if SO
0
1

1
0
if USp,
if SO
0
1

3
0
if USp,
if SO
0
3
r
+r
2 2 3 |k2 k3 | + 2
C
C
C
C
C
C
C
CL
CL
CL
CL
CL
Rep.
(1, 1, 1, 1, 1, Adj)
(F, F, 1, 1, 1, 1), (1, 1, F, F, 1, 1)

(S, 1, 1, 1, 1, 1), (1, S, 1, 1, 1, 1)
(A, 1, 1, 1, 1, 1), (1, A, 1, 1, 1, 1)

(1, 1, S, 1, 1, 1), (1, 1, 1, S, 1, 1)
(1, 1, A, 1, 1, 1), (1, 1, 1, A, 1, 1)

(1, 1, 1, 1, 1, S)
(1, 1, 1, 1, 1, A)
(1, F, 1, 1, 1, F + F )
(F, 1, 1, 1, 1, F + F )
2r2 +r3 |k2 k3 | 2

r2 +r3
2
r2 +r3
22 2
22
r2 +r3
2r2 +r3 2|k2 k3 | 1 + 2 2
r2 +r3
2r2 +r3 2|k2 k3 | 1 2 2
r2 +r3
2r2 +r3 2|k2 k3 | 1 + 2 2
r2 +r3
2r2 +r3 2|k2 k3 | 1 2 2
1 |k2 k3 | + 1
1 |k2 k3 | 1
1 |k2 k3 | + 1
1 |k2 k3 | 1
r1 +r3
2 2 +r2 2|k2 |
(F, 1, 1, F, 1, 1), (1, F, F, 1, 1, 1)

(1, 1, 1, F, 1, F + F )
r2
+2 2
r2
2 2
r2
2 2
r2
+2 2
|k2 |
(1, 1, 1, 1, 1, A)
|k2 |
(1, 1, 1, 1, 1, S)
|k2 |
(1, 1, 1, 1, 1, A)
|k2 |
(1, 1, 1, 1, 1, S)
(1, 1, 1, 1, F, F )
301
C.8. Orientifolds of the w1 w2 p3 models

Table 22
Unoriented closed spectra of the w1 w2 p3 models
Untwisted
SUGRA
Untwisted
C
Twisted
C
Twisted
V
N =1
1+3+3
Table 23
ChanPaton groups and tadpole conditions for the w1 w2 p3 models

USp(n)
USp(d2 )
USp(d1 )
U (m)
SO(d1 )
SO(d2 )
SO(n)
r
n+m+m
= 8 2 2
r2 r3
d1 + 2 2 2
|k2 k3 |(m + m)
= 8 2 2
r
d2 = 8 2 2 ,
m=m
Table 24
Open spectra of the w1 w2 p3 models
Multiplets
Number
Rep.
(Adj, 1, 1, 1), (1, Adj, 1, 1)

(1, 1, Adj, 1), (1, 1, 1, Adj)
2r2 +r3 2|k2 k3 |
CL
CL
CL
CL
r +r
2 3
2r2 +r3 4|k2 k3 | + 2 2
r2 +r3
2r2 +r3 4|k2 k3 | 2 2
r2 +r3
2r2 +r3 4|k2 k3 | + 2 2
r2 +r3
2r2 +r3 4|k2 k3 | 2 2
r2
1 2|k2 k3 | + 2 2
r2
1 2|k2 k3 | 2 2
r2
1 2|k2 k3 | 2 2
r2
1 2|k2 k3 | + 2 2
(F, 1, 1, F + F )
2|k2 |
(1, 1, 1, A)
2|k2 |
(1, 1, 1, S)
2|k2 |
(1, 1, 1, A)
2|k2 |
(1, 1, 1, S)
r1 +r3
2 2 +r2 4|k2 |
CL
(1, 1, F, F )
C.9. Non-chiral orientifolds

Table 25
Unoriented closed spectra of the p2 p3 models
B rank
r2 + r3
Untwisted
SUGRA
Untwisted
C
Twisted
C
Twisted
V
N =1
1+3+3
8+8
2
4
N =1
N =1
1+3+3
1+3+3
6+6
5+5
2+2
3+3
Table 26
Unoriented closed spectra of the w1 p2 models
Untwisted
SUGRA
Untwisted
C
Twisted
C
Twisted
V
N =1
1+3+3
302
Table 27
Unoriented closed spectra of the w1 p2 p3 and w1 p2 w3 models
Untwisted
SUGRA
Untwisted
C
Twisted
C
Twisted
V
N =1
1+3+3
Table 28
p2 p3 models (complex charges)
U (n1 ) U (n2 ) U (d) U (m)
r
n1 + n 1 + n2 + n 2 + m + m
= 32 2 2
r
r2
r
3
d + d + 2 2 + 2 |k2 k3 |(m + m)
= 32 2 2
n1 = n 1 , n2 = n 2 , d = d,
m=m
Table 29
Open spectra of the magnetized p2 p3 models (complex charges)
Multiplets
Number
|k2 k3 | r2 +r3
+1
4 2
|k2 k3 | r2 +r3
1
4 2
C
C
C
Rep.
(F, F , 1, 1), (F , F, 1, 1)
(1, 1, Adj, 1)
(F, F, 1, 1), (F , F , 1, 1)
1)
(1, 1, R + R,
1, 1, 1), (1, R + R,
1, 1)
(R + R,
1)
(1, 1, R + R,
(F, 1, 1, F ), (1, F, 1, F )
(F , 1, 1, F ), (1, F , 1, F )
(F , 1, 1, F ), (1, F , 1, F )
(F, 1, 1, F ), (1, F, 1, F )
r2 +r3
2
(1, 1, F, F ), (1, 1, F , F )
2 3
|k2 k3 | r2 +r3
2
+1
+ 1 2 2
2
r2 +r3

|k2 k3 | r2 +r3
1 2 2
2
2
r +r
(1, 1, 1, A + A)
(1, 1, 1, S + S)
Table 30
ChanPaton groups and tadpole conditions for the p2 p3 models (real charges)

USp(n1 ) USp(n2 ) USp(n3 ) USp(n4 ) USp(d1 ) USp(d2 )
U (m)
SO(n1 ) SO(n2 ) SO(n3 ) SO(n4 ) SO(d1 ) SO(d2 )
r
n1 + n2 + n3 + n4 + m + m
= 32 2 2
r2
r3
d1 + d2 + 2 2 + 2 |k2 k3 |(m + m)
= 32 2 2
n3 + n4 = n1 + n2 + m + m,
d1 = d2 , m = m
303
Table 31
Open spectra of the magnetized p2 p3 models (real charges)
Multiplets
Number
Rep.
(F, F, 1, 1, 1, 1, 1), (1, 1, F, F, 1, 1, 1)

(1, 1, 1, 1, 1, Adj, 1), (1, 1, 1, 1, Adj, 1, 1)
(F, 1, F, 1, 1, 1, 1), (1, F, 1, F, 1, 1, 1)
(1, 1, 1, 1, F, F, 1)
(F, 1, 1, F, 1, 1, 1), (1, F, F, 1, 1, 1, 1)
(1, 1, 1, 1, F, F, 1)
r2 +r3
2
(F, 1, 1, 1, 1, F, 1), (1, F, 1, 1, 1, F, 1)

(1, 1, F, 1, F, 1, 1), (1, 1, 1, F, F, 1, 1)
r2 +r3
C
C
(1, 1, 1, 1, 1, F, F + F )
(F, 1, 1, 1, 1, 1, F + F ), (1, F, 1, 1, 1, 1, F + F )
2 2
|k2 k3 | r2 +r3
+1
4 2
|k k |
2 3 2r2 +r3 1
4
r2 +r3
|k2 k3 | r2 +r3
+ 1 2 2
2
+1
2
r2 +r3
|k2 k3 | r2 +r3
2
1 2
2
2
C
C
C
(1, 1, F, 1, 1, 1, F + F ), (1, 1, 1, F, 1, 1, F + F )
(1, 1, 1, 1, 1, 1, A + A)
(1, 1, 1, 1, 1, 1, S + S)
Table 32
w1 p2 and w1 p2 p3 models (complex charges)

USp(d)
U (n)
U (m)
SO(d)
r
n + n + m + m
= 16 2 2
r2 r3
2d1 + 2 2 + 2
|k2 k3 |(m + m)
= 16 2 2
n = n,
m=m
Table 33
Open spectra of the magnetized w1 p2 and w1 p2 p3 models (complex charges)
Multiplets
Number
Rep.
(Adj, 1, 1), (1, Adj, 1)

(1, 1, Adj)
(1, Adj, 1)
1, 1) or (S + S,
1, 1)
(A + A,
(F + F , 1, F + F )
C
C
C
|k k |
2 3 2r2 +r3
2
r2 +r3
r +r
2|k2 k3 | 2 2 3 + 1 2 2
r2 +r3

2|k2 k3 | 2r2 +r3 1 2 2
(1, 1, A + A)
(1, 1, S + S)
304
Table 34
w1 p2 and the w1 p2 p3 models (real charges)

USp(d1 )
USp(n1 ) USp(n2 )
U (m)
SO(n1 ) SO(n2 )
SO(d1 )
r
n1 + n2 + m + m
= 16 2 2
2d1 + 2
r2 r3
2 + 2
|k2 k3 |(m + m)
= 16 2 2
m=m
Table 35
Open spectra of the magnetized w1 p2 and w1 p2 p3 models (real charges)
Multiplets
Number
Rep.
(Adj, 1, 1, 1), (1, Adj, 1, 1)

(1, 1, Adj, 1), (1, 1, 1, Adj)
(1, 1, Adj, 1), (F, F, 1, 1)
(F, 1, 1, F + F ), (1, F, 1, F + F )
C
C
|k2 k3 | r2 +r3
2 2
r2 +r3
r +r
2
3 + 2 2
2|k2 k3 | 2
1
r2 +r3

2|k2 k3 | 2r2 +r3 1 2 2
C
C
(1, 1, 1, A + A)
(1, 1, 1, S + S)
Table 36
ChanPaton groups of the w1 p2 w3 models

USp(n)
USp(d)
U (m)
SO(n)
SO(d)
r
n+m+m
= 8 2 2
r2 r3
+2 2 + 2
|k2 k3 |(m + m)
= 8 2 2
m=m
Table 37
Open spectra of the magnetized w1 p2 w3 models
Multiplets
Number
Rep.
2|k2 k3 |2r2 +r3

r2 +r3
r +r
2|k2 k3 | 22 2 3 + 1 2 2
r2 +r3

2|k2 k3 | 22r2 +r3 1 2 2
(Adj, 1, 1), (1, Adj, 1)

(1, 1, Adj)
(F, 1, F + F )
C
C
(1, 1, A + A)
(1, 1, S + S)
305
C.10. w2 p3 models with brane supersymmetry breaking
Table 38
Unoriented closed spectra of the w2 p3 models with brane supersymmetry breaking
Model
Untwisted
SUGRA
Untwisted
C
Twisted
C
Twisted
V
w 2 p3
N =1
1+3+3
Table 39
ChanPaton groups of the w2 p3 models with brane supersymmetry breaking
SO(n1 ) SO(n2 ) Usp(d1 ) U sp(d2 ) Usp(d3 ) U (m)
n1 + n2 + m + m
= 16
= 16
d1 + d2 + |k2 k3 |(m + m)
m=m
d3 = 8, n2 = n1 + m + m,
Table 40
Open spectra of the w2 p3 models with brane supersymmetry breaking
States
Number
Rep.
Scalars
Scalars
C
Scalars
C
C
C
Scalars
Scalars
C
Scalars
Scalars
Scalars
Scalars
Scalars
4
2
4
2
|k2 k3 |/2 + 2
|k2 k3 |/2 2
|k2 k3 | 4
|k2 k3 | + 4
2
4
3|k2 k3 | 2 |k2 |
|k2 k3 | + |k2 |
3|k2 k3 | 2 + |k2 |
|k2 k3 | |k2 |
(Adj, 1, 1, 1, 1, 1), (1, Adj, 1, 1, 1, 1), (1, 1, Adj, 1, 1, 1)

(1, 1, 1, Adj, 1, 1), (1, 1, 1, 1, Adj, 1), (1, 1, 1, 1, 1, Adj)
(Adj, 1, 1, 1, 1, 1), (1, Adj, 1, 1, 1, 1), (1, 1, A, 1, 1, 1)
(1, 1, 1, A, 1, 1), (1, 1, 1, 1, A, 1), (1, 1, 1, 1, 1, Adj)
(F, F, 1, 1, 1, 1), (1, 1, F, F, 1, 1), (1, 1, 1, 1, Adj, 1)
(F, F, 1, 1, 1, 1), (1, 1, F, F, 1, 1), (1, 1, 1, 1, Adj, 1)
(F, 1, F, 1, 1, 1), (1, F, 1, F, 1, 1)
(1, F, F, 1, 1, 1), (F, 1, 1, F, 1, 1)
(F, 1, 1, 1, 1, F + F )
(1, F, 1, 1, 1, F + F )
(F, 1, 1, 1, 1, F + F )
(1, F, 1, 1, 1, F + F )
(1, 1, 1, F, 1, F + F )
(1, 1, F, 1, 1, F + F )
(1, 1, 1, 1, 1, A + A)
(1, 1, 1, 1, 1, S + S)
(1, 1, 1, 1, 1, A + A)
(1, 1, 1, 1, 1, S + S)
CL
CL
CR
CR
3|k2 k3 | + 2 + |k2 |
|k2 k3 | |k2 |
3|k2 k3 | + 2 |k2 |
|k2 k3 | + |k2 |
(1, 1, 1, 1, 1, A)
(1, 1, 1, 1, 1, S)
(1, 1, 1, 1, 1, A)
(1, 1, 1, 1, 1, S)
CL
|k2 |
(1, 1, 1, 1, F, F )
Scalars
|k2 |
(1, 1, 1, 1, F, F + F )
306
C.11. Open spectra of the undeformed Z2 Z2 shift-orientifolds

Table 41
Open spectra of the undeformed Z2 Z2 shift-orientifolds
Model
CP group
[U (n1 ) U (n2 )]9
U (d1 )51 U (d2 )52
p3
p23
[U (n1 ) U (n2 )]9 U (d)51
SO(no ) SO(ng )
SO(nh ) SO(nf )
p123
Constraints
SUSY
Chiral multiplets
n1 + n2 = 16
d1 = d2 = 8
N =1
n1 + n2 = 16
d =8
N =1
1, 1, 1)99 , (1, A + A,
1, 1)99
(A + A,
(F + F , F + F , 1, 1)99
(1, 1, A, 1)55 , (1, 1, 1, A)55
(F, 1, F, 1)59 , (F, 1, 1, F )59
(1, F , 1, F )59 , (1, F, F, 1)59
1, 1)99 , (1, A + A,
1)99
(A + A,
55
(F + F , F + F , 1)99 , (1, 1, A + A)
(F, 1, F )59 , (F , 1, F )59
(1, F, F )59 , (1, F , F )59
(F, F, 1, 1)99 , (F, 1, F, 1)99
(F, 1, 1, F )99 , (1, F, F, 1)99
(1, F, 1, F )99 , (1, 1, F, F )99
i ni = 32
N =1
n=8
d1 = d2 = 8
n = d1 = 8
n = d1 = 8
n=8
d1 = d2 = 8
n = d1 = 8
n=8
n=8
N =2
Hypermultiplets
w 2 p3
U (n)9 U (d1 )51 SO(d2 )52
w 1 p2
w 1 p2 p3
w 1 w 2 p3
U (n)9 SO(d1 )51

U (n)9 SO(d1 )51
SO(n)9 SO(d1 )51 SO(d2 )52
w 1 p2 w 3
p1 w 2 w 3
w1 w2 w3
SO(n)9 SO(d1 )51

U (n)9
SO(n)9
N =2
N =2
N =4
N =4
N =4
N =4
2(A, 1, 1)99 , 2(1, A, 1, )51 51

2, (F, F, 1)951
2(A, 1)99
2(A, 1)99
References
[1] A. Sagnotti, in: G. Mack, et al. (Eds.), Cargese 87, Non-Perturbative Quantum Field Theory, Pergamon
Press, Oxford, 1988, p. 521, hep-th/0208020.
[2] M. Bianchi, A. Sagnotti, Phys. Lett. B 247 (1990) 517;
M. Bianchi, A. Sagnotti, Nucl. Phys. B 361 (1991) 519.
[3] G. Pradisi, A. Sagnotti, Phys. Lett. B 216 (1989) 59.
[4] C. Angelantonj, A. Sagnotti, Phys. Rep. 371 (2002) 1, hep-th/0204089.
[5] E. Dudas, Class. Quantum Grav. 17 (2000) R41, hep-ph/0006190.
[6] C.M. Hull, P.K. Townsend, Nucl. Phys. B 438 (1995) 109, hep-th/9410167;
P.K. Townsend, Phys. Lett. B 350 (1995) 184, hep-th/9501068;
E. Witten, Nucl. Phys. B 443 (1995) 85, hep-th/9503124;
For a review, see, e.g., M.J. Duff, Int. J. Mod. Phys. A 11 (1996) 5623, hep-th/9608117;
M. Li, hep-th/9811019.
[7] I. Antoniadis, N. Arkani-Hamed, S. Dimopoulos, G.R. Dvali, Phys. Lett. B 436 (1998) 257, hep-ph/9804398;
For reviews, see, e.g., [5,8].
[8] I. Antoniadis, hep-th/0102202.
[9] For a review, see: K.R. Dienes, Phys. Rep. 287 (1997) 447, hep-th/9602045.
[10] J.D. Lykken, Phys. Rev. D 54 (1996) 3693, hep-th/9603133;
K.R. Dienes, E. Dudas, T. Gherghetta, Nucl. Phys. B 537 (1999) 47, hep-ph/9806292.
[11] N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 429 (1998) 263, hep-ph/9803315;
Z. Kakushadze, S.H. Tye, Nucl. Phys. B 548 (1999) 180, hep-th/9809147;
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
307
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;
For a review, see: R. Dick, Class. Quantum Grav. 18 (2001) R1, hep-th/0105320.
J.C. Long, J.C. Price, hep-ph/0303057.
L.J. Dixon, J.A. Harvey, Nucl. Phys. B 274 (1986) 93;
N. Seiberg, E. Witten, Nucl. Phys. B 276 (1986) 272.
A. Sagnotti, hep-th/9509080;
A. Sagnotti, Nucl. Phys. B (Proc. Suppl.) 56 (1997) 332, hep-th/9702093;
G. Pradisi, Nuovo Cimento B 112 (1997) 467, hep-th/9603104.
C. Angelantonj, Phys. Lett. B 444 (1998) 309, hep-th/9810214;
C. Angelantonj, A. Armoni, Nucl. Phys. B 578 (2000) 239, hep-th/9912257;
R. Blumenhagen, A. Font, D. Lust, Nucl. Phys. B 558 (1999) 159, hep-th/9904069;
R. Blumenhagen, A. Kumar, Phys. Lett. B 464 (1999) 46, hep-th/9906234;
K. Forger, Phys. Lett. B 469 (1999) 113, hep-th/9909010.
E. Dudas, J. Mourad, A. Sagnotti, Nucl. Phys. B 620 (2002) 109, hep-th/0107081.
J. Scherk, J.H. Schwarz, Nucl. Phys. B 153 (1979) 61.
R. Rohm, Nucl. Phys. B 237 (1984) 553;
S. Ferrara, C. Kounnas, M. Porrati, Phys. Lett. B 181 (1986) 263;
S. Ferrara, C. Kounnas, M. Porrati, Nucl. Phys. B 304 (1988) 500;
S. Ferrara, C. Kounnas, M. Porrati, Phys. Lett. B 206 (1988) 25;
C. Kounnas, M. Porrati, Nucl. Phys. B 310 (1988) 355;
I. Antoniadis, C. Bachas, D.C. Lewellen, T.N. Tomaras, Phys. Lett. B 207 (1988) 441;
S. Ferrara, C. Kounnas, M. Porrati, F. Zwirner, Nucl. Phys. B 318 (1989) 75;
C. Kounnas, B. Rostand, Nucl. Phys. B 341 (1990) 641;
I. Antoniadis, Phys. Lett. B 246 (1990) 377;
I. Antoniadis, C. Kounnas, Phys. Lett. B 261 (1991) 369;
E. Kiritsis, C. Kounnas, Nucl. Phys. B 503 (1997) 117, hep-th/9703059.
I. Antoniadis, E. Dudas, A. Sagnotti, Nucl. Phys. B 544 (1999) 469, hep-th/9807011;
I. Antoniadis, G. DAppollonio, E. Dudas, A. Sagnotti, Nucl. Phys. B 553 (1999) 133, hep-th/9812118;
R. Blumenhagen, L. Gorlich, Nucl. Phys. B 551 (1999) 601, hep-th/9812158;
C. Angelantonj, I. Antoniadis, K. Forger, Nucl. Phys. B 555 (1999) 116, hep-th/9904092;
A.L. Cotrone, Mod. Phys. Lett. A 14 (1999) 2487, hep-th/9909116.
I. Antoniadis, G. DAppollonio, E. Dudas, A. Sagnotti, Nucl. Phys. B 565 (2000) 123, hep-th/9907184.
G. Aldazabal, A.M. Uranga, JHEP 9910 (1999) 024, hep-th/9908072;
G. Aldazabal, L.E. Ibanez, F. Quevedo, A.M. Uranga, JHEP 0008 (2000) 002, hep-th/0005067.
C. Angelantonj, I. Antoniadis, G. DAppollonio, E. Dudas, A. Sagnotti, Nucl. Phys. B 572 (2000) 36, hepth/9911081.
C. Angelantonj, R. Blumenhagen, M.R. Gaberdiel, Nucl. Phys. B 589 (2000) 545, hep-th/0006033.
I. Antoniadis, E. Dudas, A. Sagnotti, Phys. Lett. B 464 (1999) 38, hep-th/9908023.
M. Bianchi, J.F. Morales, G. Pradisi, Nucl. Phys. B 573 (2000) 314, hep-th/9910228.
S. Sugimoto, Prog. Theor. Phys. 102 (1999) 685, hep-th/9905159.
E. Witten, Phys. Lett. B 149 (1984) 351.
C. Bachas, A way to break supersymmetry, hep-th/9503030.
M. Bianchi, Y.S. Stanev, Nucl. Phys. B 523 (1998) 193, hep-th/9711069.
E.S. Fradkin, A.A. Tseytlin, Phys. Lett. B 158 (1985) 316;
A. Abouelsaood, C.G. Callan, C.R. Nappi, S.A. Yost, Nucl. Phys. B 280 (1987) 599.
C. Angelantonj, I. Antoniadis, E. Dudas, A. Sagnotti, Phys. Lett. B 489 (2000) 223, hep-th/0007090.
C. Angelantonj, A. Sagnotti, hep-th/0010279.
R. Blumenhagen, L. Goerlich, B. Kors, D. Lust, JHEP 0010 (2000) 006, hep-th/0007024;
R. Blumenhagen, L. Goerlich, B. Kors, D. Lust, Fortschr. Phys. 49 (2001) 591, hep-th/0010198;
R. Blumenhagen, B. Kors, D. Lust, JHEP 0102 (2001) 030, hep-th/0012156.
M. Larosa, hep-th/0111187;
M. Larosa, hep-th/0212109;
G. Pradisi, hep-th/0210088.
308
[35] N.K. Nielsen, P. Olesen, Nucl. Phys. B 144 (1978) 376;

J. Ambjorn, N.K. Nielsen, P. Olesen, Nucl. Phys. B 152 (1979) 75;
H.B. Nielsen, M. Ninomiya, Nucl. Phys. B 169 (1980) 309.
[36] M. Berkooz, M.R. Douglas, R.G. Leigh, Nucl. Phys. B 480 (1996) 265, hep-th/9606139;
V. Balasubramanian, R.G. Leigh, Phys. Rev. D 55 (1997) 6415, hep-th/9611165.
[37] R. Blumenhagen, B. Kors, D. Lust, T. Ott, Nucl. Phys. B 616 (2001) 3, hep-th/0107138;
R. Blumenhagen, V. Braun, B. Kors, D. Lust, JHEP 0207 (2002) 026, hep-th/0206038;
R. Blumenhagen, L. Gorlich, T. Ott, JHEP 0301 (2003) 021, hep-th/0211059;
D. Lust, S. Stieberger, hep-th/0302221.
[38] G. Aldazabal, S. Franco, L.E. Ibanez, R. Rabadan, A.M. Uranga, J. Math. Phys. 42 (2001) 3103, hepth/0011073;
G. Aldazabal, S. Franco, L.E. Ibanez, R. Rabadan, A.M. Uranga, JHEP 0102 (2001) 047, hep-ph/0011132;
L.E. Ibanez, F. Marchesano, R. Rabadan, JHEP 0111 (2001) 002, hep-th/0105155;
D. Cremades, L.E. Ibanez, F. Marchesano, JHEP 0207 (2002) 022, hep-th/0203160;
D. Cremades, L.E. Ibanez, F. Marchesano, hep-th/0205074;
A.M. Uranga, hep-th/0208014.
[39] M. Cvetic, G. Shiu, A.M. Uranga, Nucl. Phys. B 615 (2001) 3, hep-th/0107166.
[40] M. Cvetic, G. Shiu, A.M. Uranga, Phys. Rev. Lett. 87 (2001) 201801, hep-th/0107143;
M. Cvetic, P. Langacker, G. Shiu, Phys. Rev. D 66 (2002) 066004, hep-ph/0205252;
M. Cvetic, P. Langacker, G. Shiu, Nucl. Phys. B 642 (2002) 139, hep-th/0206115;
M. Cvetic, I. Papadimitriou, G. Shiu, hep-th/0212177;
M. Cvetic, I. Papadimitriou, hep-th/0303083;
M. Cvetic, I. Papadimitriou, hep-th/0303197.
[41] C. Kokorelis, JHEP 0208 (2002) 036, hep-th/0206108;
C. Kokorelis, hep-th/0207234;
C. Kokorelis, hep-th/0212281.
[42] S. Forste, G. Honecker, R. Schreyer, Nucl. Phys. B 593 (2001) 127, hep-th/0008250;
S. Forste, G. Honecker, R. Schreyer, JHEP 0106 (2001) 004, hep-th/0105208.
[43] D. Bailin, G.V. Kraniotis, A. Love, hep-th/0108127;
D. Bailin, G.V. Kraniotis, A. Love, Phys. Lett. B 530 (2002) 202, hep-th/0108131;
D. Bailin, G.V. Kraniotis, A. Love, hep-th/0208103;
D. Bailin, G.V. Kraniotis, A. Love, JHEP 0302 (2003) 052, hep-th/0212112.
[44] R. Blumenhagen, D. Lust, T.R. Taylor, hep-th/0303016;
J.F. Cascales, A.M. Uranga, hep-th/0303024.
[45] M. Bianchi, G. Pradisi, A. Sagnotti, Nucl. Phys. B 376 (1992) 365;
M. Bianchi, Nucl. Phys. B 528 (1998) 73, hep-th/9711201;
E. Witten, JHEP 9802 (1998) 006, hep-th/9712028.
[46] Z. Kakushadze, G. Shiu, S.H. Tye, Phys. Rev. D 58 (1998) 086001, hep-th/9803141;
C. Angelantonj, Nucl. Phys. B 566 (2000) 126, hep-th/9908064.
[47] C. Vafa, Nucl. Phys. B 273 (1986) 592;
A. Font, L.E. Ibanez, F. Quevedo, Phys. Lett. B 217 (1989) 272;
A. Font, L.E. Ibanez, F. Quevedo, A. Sierra, Nucl. Phys. B 337 (1990) 119;
C. Vafa, E. Witten, J. Geom. Phys. 15 (1995) 189, hep-th/9409188.
[48] M. Bianchi, PhD Thesis, preprint ROM2F-92/13;
A. Sagnotti, Anomaly cancellations and open string theories, in: From Superstrings to Supergravity, Erice
1992, Proceedings, hep-th/9302099.
[49] M. Berkooz, R.G. Leigh, Nucl. Phys. B 483 (1997) 187, hep-th/9605049.
[50] J. de Boer, R. Dijkgraaf, K. Hori, A. Keurentjes, J. Morgan, D.R. Morrison, S. Sethi, Adv. Theor. Math.
Phys. 4 (2002) 995, hep-th/0103170.
[51] C. Angelantonj, R. Blumenhagen, Phys. Lett. B 473 (2000) 86, hep-th/9911190.
[52] R. Blumenhagen, L. Gorlich, B. Kors, Nucl. Phys. B 569 (2000) 209, hep-th/9908130;
R. Blumenhagen, L. Gorlich, B. Kors, JHEP 0001 (2000) 040, hep-th/9912204.
309
[53] G. Pradisi, Nucl. Phys. B 575 (2000) 134, hep-th/9912218.

[54] D. Fioravanti, G. Pradisi, A. Sagnotti, Phys. Lett. B 321 (1994) 349, hep-th/9311183;
G. Pradisi, A. Sagnotti, Y.S. Stanev, Phys. Lett. B 354 (1995) 279, hep-th/9503207;
Related reviews are: G. Pradisi, in Ref. [14];
A. Sagnotti, Y.S. Stanev, Fortschr. Phys. 44 (1996) 585;
A. Sagnotti, Y.S. Stanev, Nucl. Phys. B (Proc. Suppl.) 55 (1997) 200, hep-th/9605042;
Y.S. Stanev, hep-th/0112222.
[55] E.G. Gimon, J. Polchinski, Phys. Rev. D 54 (1996) 1667, hep-th/9601038.
[56] M. Li, Nucl. Phys. B 460 (1996) 351, hep-th/9510161;
M.R. Douglas, hep-th/9512077;
C. Schmidhuber, Nucl. Phys. B 467 (1996) 146, hep-th/9601003;
M.B. Green, J.A. Harvey, G.W. Moore, Class. Quantum Grav. 14 (1997) 47, hep-th/9605033;
J. Mourad, Nucl. Phys. B 512 (1998) 199, hep-th/9709012;
Y.K. Cheung, Z. Yin, Nucl. Phys. B 517 (1998) 69, hep-th/9710206;
R. Minasian, G.W. Moore, JHEP 9711 (1997) 002, hep-th/9710230.
[57] E. Witten, Nucl. Phys. B 460 (1996) 541, hep-th/9511030.
[58] E. Dudas, J. Mourad, Phys. Lett. B 514 (2001) 173, hep-th/0012071.
[59] G. Pradisi, F. Riccioni, Nucl. Phys. B 615 (2001) 33, hep-th/0107090.
[60] C. Angelantonj, M. Bianchi, G. Pradisi, A. Sagnotti, Y.S. Stanev, Phys. Lett. B 385 (1996) 96, hepth/9606169.
[61] M. Dine, N. Seiberg, E. Witten, Nucl. Phys. B 289 (1987) 589.
[62] A. Sagnotti, Phys. Lett. B 294 (1992) 196, hep-th/9210127.
[63] L.E. Ibanez, R. Rabadan, A.M. Uranga, Nucl. Phys. B 542 (1999) 112, hep-th/9808139;
G. Aldazabal, D. Badagnani, L.E. Ibanez, A.M. Uranga, JHEP 9906 (1999) 031, hep-th/9904071.
[64] K.S. Narain, Phys. Lett. B 169 (1986) 41;
K.S. Narain, M.H. Sarmadi, E. Witten, Nucl. Phys. B 279 (1987) 369.
[65] E. Dudas, J. Mourad, Phys. Lett. B 486 (2000) 172, hep-th/0004165;
R. Blumenhagen, A. Font, Nucl. Phys. B 599 (2001) 241, hep-th/0011269.
[66] M.R. Douglas, JHEP 9707 (1997) 004, hep-th/9612126;
M.R. Douglas, B.R. Greene, D.R. Morrison, Nucl. Phys. B 506 (1997) 84, hep-th/9704151;
D.E. Diaconescu, M.R. Douglas, J. Gomis, JHEP 9802 (1998) 013, hep-th/9712230.

Localized tachyon mass and a g-theorem analogue

Sang-Jin Sin a,b
a Stanford Linear Accelerator Center, Stanford University, Stanford, CA 94305, USA
b Department of Physics, Hanyang University, Seoul 133-791, South Korea
Received 19 March 2003; received in revised form 6 June 2003; accepted 18 June 2003
Abstract
We study the localized tachyon condensation (LTC) of non-supersymmetric orbifold backgrounds
in their mirror LandauGinzburg picture. Using he existence of four copies of (2, 2) worldsheet
supersymmetry, we show at the CFT level, that the minimal tachyon mass in twisted sectors shows
2 | satisfies m(M)
somewhat analogous properties of c- or g-function. Namely, m := | Mmin
m(D1 D2 ) = max{m(D1 ), m(D2 )}. c-, g-, m-functions share the common property f (M)
f (D1 D2 ) for f = c, g, m, although they have different behavior in detail.
PACS: 11.25.-w
1. Introduction
The study of open string tachyon condensation [1] has led to many interesting
consequences including classification of the D-brane charge by K-theory. While the closed
string tachyon condensation involve the change of the background spacetime and much
more difficult, if we consider the case where tachyons can be localized at the singularity,
one may expect the maximal analogy with the open string case. Along this direction, the
study of localized tachyon condensation (LTC) was considered first in [2] using the brane
probe and renormalization group flow and by many others [37,9]. The basic picture is that
tachyon condensation induces cascade of decays of the orbifolds to less singular ones until
the spacetime supersymmetry is restored. Therefore the localized tachyon condensation
has geometric description as the resolution of the spacetime singularities.
Work supported partially by the Department of Energy under contract number DE-AC03-76SF005515.
E-mail address: sjs@hepth.hanyang.ac.kr (S.-J. Sin).
doi:10.1016/S0550-3213(03)00541-8
S.-J. Sin / Nuclear Physics B 667 (2003) 310320
311
Soon after, Vafa [3] considered the problem in the LandauGinzburg (LG) formulation
using the Mirror symmetry and confirmed the result of [2]. In [4], Harvey, Kutasov,
Martinec and Moore studied the same problem using the RG flow as deformation of chiral
ring and in term of toric geometry. In both papers, the worldsheet N = 2 supersymmetry
was utilized in essential ways.
The tachyon condensation process can be regarded as a RG-flow, along which there
is a decreasing quantity, c-function, [10] for unitary conformal field theories. For the
localized tachyon condensation in non-compact space, however, c is constant [2,4] and
cannot measure any dynamical process. Therefore it would be very interesting to have
a quantity which has a property of monotonicity along the RG-flow. Along this line, the
authors of [4] suggested a quantity, gcl , which is a closed string analogue of the ground state
degeneracy in open string theory [11]. On the other hand, Dabholkar and Vafa [5] suggested
the maximal R-charge of Ramond sector (see [17,18] for earlier study on this quantity),
as the height of tachyon potential in the twisted sector describing the localized tachyon
condensation. Although both suggestions are well motivated by physical intuitions, the
prediction of two quantity are slightly different [7]. In [7], it was also suggested that the
lowest twisted tachyon mass increases along RG flow. Using the spectral flow of N = 2
CFT and CTP invariance of the Ramond sector together with the charge-mass relation
for chiral primaries, one can easily see that the proposal of [7] is equivalent to the GSO
projected version of the one given by [5].
The monotonically increasing property of R-charge is related to a theorem in
singularity theory called semi-continuity of spectrum [13] in singularity theory, which was
conjectured by Arnold [12] and proved later by Varchenko [14] and Steenbrink [15]. These
mathematical result can be applied [17] to the N = 2 supersymmetric LandauGinzburg
theory due to non-renormalization theorem. In our case, the LG theory that is mirror to
the orbifold Cr /Zn is not an ordinary LG theory but an orbifolded LG [16] model whose
chiral ring structure is very different from the ordinary LG and hence the theorem cannot
be applied directly. Although it is easy to see the monotonicity of R-charge for C/Zn case,
it is non-trivial for C2 /Zn or higher-dimensional cases.
The main goal of this paper is to prove that the lowest tachyon mass or equivalently
the minimal R-charge in orbifold CFT and type 0 theories increases when we compare
those in UV and IR fixed points, which is conjectured in [5,7]. As shown in [3,4], the
result of decay of C2 /Zn is a sum of two disconnected theory and this fact introduces extra
care when we compare with initial and final theories. When we denote the process of a
mother theory decaying into two disconnected daughter theories by M D1 D2 , the
c-function satisfies c(M) c(D1 ) and c(M) c(D2 ), while g-function seems to have
g(M) g(D1 ) + g(D2 ) [4]. It turns out that for minimal tachyon mass, if we define
2 |, it satisfies m(M) m(D D ) = max{m(D ), m(D )}, which is again
m = | Mmin
1
2
1
2
different from both c- and g-function. Although c-, g-, m-functions have different behavior
in detail they share the common property
f (M) f (D1 D2 ),
f = c, g, m.
(1.1)
In order to avoid possible confusion and also to emphasize the difference with c-theorem,
we call this monotonicity property of minimal tachyon mass as m-theorem.
312
Interestingly, our method applies only to the orbifolded LandauGinzburg theory and
does not apply to the generic LG theories. In this sense, our method is complementary to
the method used in mathematical literature.
We use the mirror LG picture of Vafa and the existence of the many copies of (2, 2)
worldsheet supersymmetries for C2 /Zn . In CFT of C2 /Zn , there are 22 extended chiral
ring structures according to the choice of complex structures of each C factor. We call them
as cc, ca, ac and aa rings. For string spectrum, we need to put spectrum of all 4 sectors
together. When we consider the behavior of cc ring elements under the condensation of a
tachyon in cc ring, we can establish an explicit mapping between spectrum of initial and
final orbifold conformal field theories. We can show that individual R-charge of tachyons
increases under the process. This is possible since we have control over the RG-process
due to the world sheet (2, 2) supersymmetry off the criticality, which provide the nonrenormalization theorem. However, what happen to the R-charges of operators in ca or
other rings when a tachyon operator in cc ring condensate? The answer is that we lose
control, since we lose all supersymmetry off the criticality hence we do not have nonrenormalization theorem.
What saves us from this difficulty is the presence of the enhanced 2r copies of (2, 2)
worldsheet SUSY in orbifold CFTs. This is because its presence allows us to choose the
supersymmetry generators G
1/2 and complex structure such that the condensing tachyon
belongs to cc-ring. We can then determine the generators of the daughter theories. Since
we know that the final products of the decay are again orbifold theories [24], knowing
the fate of the cc-ring element is enough to establish the fate of entire spectrum. Using
this special property of orbifold theories, we will be able to establish linear mappings for
each of 4 chiral rings.1 We can also show that the linear mapping has the property such
that the R-charges of their images are bigger than the R-charges of the originals. The mere
existence of such mappings will enable us to show our main goal: the minimal charge
increases under TC.
In Section 2, we give a brief summary of mirror LG model. In Section 3, we give the
proof of the statement and we conclude in Section 4.
2. Mirror symmetry and orbifolds

In this subsection we give a summary of Vafas work [3] on localized tachyon
condensation. The orbifold Cr /Zn is defined by the Zn action given by equivalence relation

(X1 , . . . , Xr ) k1 X1 , . . . , kr Xr ,
(2.1)
= e2i/n .
We call (k1 , . . . , kr ) as the generator of the Zn action. The orbifold can be imbedded
into the gauged linear sigma model (GLSM) [19]. The vacuum manifold of the latter is
described by the D-term constraints

ki |Xi |2 = t.
n|X0 |2 +
(2.2)
i
1 Some of these mapping does not describe tachyon condensation process of individual R-charges. They just
connect between the spectrum of mother and daughter theories.
313
Its t limit corresponds to the orbifold and the t limit is the O(n) bundle
over the weighted projected space W Pk1 ,...,kr . X0 direction corresponds to the non-compact
fiber of this bundle and t plays role of size of the W Pk1 ,...,kr .
By dualizing this GLSM, we get a LG model with a superpotential [20]
W=
r
exp(Yi ),
(2.3)
i=0
where twisted chiral fields Yi are periodic Yi Yi + 2i and related to Xi by Re[Yi ] =

Yi /n , the D-term constraint is expressed as
|Xi |2 . Introducing
k the variable ui := e
Y
t
/n
2i/n u which
i
e 0 =e
i
i u . The periodicity of Yi imposes the identification: ui e
necessitate modding out each ui by Zn . The result is usually described by

r

n
t /n
ki
(Zn )r1
W=
(2.4)
ui + e
u
i=1
which describe the mirror LandauGinzburg model of the linear sigma model. As a
t limit, mirror of the orbifold is

r

uni
(Zn )r1 .
W=
(2.5)
i=1
Since it is not ordinary LandauGinzburg theory but an orbifolded version, the chiral ring
structure of the theory is very different from that of LG model. For example, the dimension
of the local ring of the superpotential is always n 1, regardless of r.
We list some properties of orbifolded LG theory for later use.
The true variable of the theory are Yi not ui related by ui = eYi /n . As a consequence,
monomial basis of the chiral ring is given by

p1 p2
u1 u2 |(p1 , p2 ) = n{j k1 /n}, n{j k2/n} , j = 1, . . . , n 1 ,
(2.6)
p
and u1 1 u2 2 has weight (p1 , p2 ) and charge (p1 /n, p2 /n).

It has been known (see, for example, [8], for C2 /Zn , any worldsheet fermion generated
tachyon can be constructed as a BPS state, i.e., a member of a chiral ring as a
consequence of existence of 4 copies of (2, 2) worldsheet SUSY for this special theory;
It is convenient to consider the weight of a state as sum of contribution from each
1 na2
complex plane. For example, the weight of una
1 u2 can be considered as sum of a1
from u1 and a2 from u2 . (a1 , a2 ) form a point in the weight space. As we vary j in
ai = {j ki /n}, the trajectory of the point in weight space will give us a parametric plot
in the plane;
The weight space is a lattice in torus of size n n. The identification of weights by
modulo n corresponds to shifting string modes. However, periodicity of the generator
(k1 , k2 ) is 2n and (k1 , k2 ) and (k1 , k2 + n) do not generate the same theory in general.
We choose the standard range of ki between n + 1 to n 1. This is because the
GSO projection depends not only the R-charge vector ({j k1/n}, {j k2 /n}) but also the
G-parity number G = [j k1 /n] + [j k2 /n]. We will comeback to this when we discuss
the GSO projection;
314
When n and ki are not relatively prime, we have a chiral primary whose R-charge
vector is (p/n, 0). We call this as the reducible case and eliminate from our interests.
This is a spectrum that is not completely localized at the tip of the orbifold. Sometimes,
even in the case we started from non-reducible theory, a tachyon condensation leads
us to the reducible case.
3. M-theorem
The conformal weight in the NS-sector is related to the mass by [2123]
1 2
1
M = .
4
2
(3.1)
2 = 2(1 1/n) is proportional to the

For C1 /Zn model, min = 1/2n so that Mmin
deficit angle of the cone. The maximal R-charge in NS sector and the minimal tachyon
mass is simply related. Let us imbed the orbifold into 8-dimensional transverse target space
of lightcone string theory. Then the transverse spacetime is Cr /Zn R 82r . The ground
states of the twisted sectors in NS sectors are chiral or anti-chiral primary and the charge
is simply given by the weight q = 2. By the spectral flow, qR = qNS c/2,
we have
1 2
1
c
1
M
=
(q
+
)
and
4
2 R
2
2
2
= Q5min + c 2,
Mmin
(3.2)
where 2 in
= 2qR,min comes from summing left and right R-charges. Using the CPT
symmetry on the Ramond sector, we have qRmin = qRmax . Therefore, above statement can
also be written as

max M 2 = Q5max + 2 c := m.
(3.3)
Q5min
The main goal of this paper is to show that m has a property that is a reminiscent of a
c-function:
m(UV) m(IR).
(3.4)
We call this as a m-theorem to prevent possible confusion with c-theorem or g-theorem.

3.1. Proof of the theorem
This work is heavily dependent on the result of companion paper [8], where details
of transition of spectrum is discussed. Our interest is how the spectrum flows under the
RG-flow. In lack of control of off-shell theory, it is in general difficult question to address.
However, spectrum of UV and IR theories are readily available since both are conformal
field theory. Since any tachyon generated by worldsheet fermion can be thought as a chiral
ring element in one choice of (2, 2) SUSY, we can assume, without loss of generality, that
the condensing tachyon with weight p = (p1 , p2 ) is an element of cc-ring. Then consider
other element in the same cc-ring whose weight is q = (q1 , q2 ). For definiteness, let us
say q in the notation of [8]. The R-charge of it is Rq = (q1 + q2 )/n. Now after the
condensation of p = (p1 , p2 ), q is moved to q = Tp (q) [8] where,

p2 /n p1 /n
Tp =
,
0
1
315
(3.5)
whose R-charge is
Rq = (q2 p q/n)/p2 .
(3.6)
If p represents a tachyonic state, it must be below the diagonal. Namely,

p1 + p2 n.
(3.7)
Therefore, the difference in the R-charge between before and after the process is given by
Rq Rq = (n p1 p2 )q2 /np2 0.
(3.8)
The same inequality holds for q + . For p, q ac ring, we can apply the same argument
by replacing
p p = (p1 , n p2 ),
q q = (q1 , n q2 ).
(3.9)
Therefore, we arrive at the result:

Lemma. The R-charge of a relevant chiral primary operator increases under condensation
of tachyon in the same ring.
Eq. (3.8) also shows that under the condensation of marginal operator, there is no change
in R-charge of any operator. Due to the mass-charge relation discussed before, we can
make the same statements for the tachyon mass. The above statement shows that any of
the spectrum is a candidate of the c-function of the twisted sector. However, this statement
does not exclude the possibility of level crossing. That is, the ordering of the R-charge can
be changed during the process.
What happen to the R-charges of operators in ca ring when a tachyon in cc ring
condensate? The answer is that we lose control, since we lose the world sheet (2, 2)
supersymmetry off the criticality and we lose non-renormalization theorem. In fact if one
naively apply the mapping Tp to the ca-ring elements, we get non-integer power of ui s.
Similarly, when we condense a ca ring element, we lose control over the cc ring spectrum.
However, when an element in ca ring turn on, we have control over other ca ring
elements instead. It is holomorphic and protected by worldsheet SUSY G+
ca . Since we
have choice of selecting complex structure in each plane independently, we can choose
any combination of complex structure to define the holomorphic co-ordinate of C2 . We
can call u1 , u 2 as the holomorphic co-ordinates just as we can call u1 , u2 as a holomorphic
coordinate. As far as other combination does not enter in the theory, operators are protected
by the worldsheet supersymmetry.
As a warm up, we first consider the case where the minimal charge of initial and final
theories belong to cc-ring, so that we do not need to consider other chiral rings. Let q0
denote the state of minimal R-charge, namely,
R[q0 ] R[q],
for all q.
(3.10)
316
We want to compare the minimal charge of the initial charge and that of the final state.
Let q be a minimal charge of a final theory. There are two theories in the final states
and one choose any of it, say up-theory. Then q should come from a q + such that
q = Tp+ (q). Due to the monotonicity of R-charge, we have R[q ] > R[q]. On the other
hand, R[q] cannot be smaller than R[q0 ], by definition of q0 . Therefore we have inequality
initial
final
R qmin
(3.11)
< R qmin
.
The same inequality holds for the down-theory as well.
Some of the relevant operators, which are precisely those in the triangle BP A, will be
pushed out to irrelevant operator after P is condensed. One may worry about the converse
possibility that some irrelevant operators of the initial theories flow to the relevant operator.
Following lemma tells us that it does not happen.
Lemma. Relevant chiral primary states of final theory comes only from the relevant ones
in the initial theory.
Proof. Let q /p2 be the charge of a relevant operator in the down-theory and q be its
pre-image in the original theory, i.e., q = Tp (q). Our question is whether q1 + q2 < p2
implies q1 + q2 < n or not. This can be answered simply by calculating the inverse of Tp .

n 1 p1 /n
q1
(nq1 + p1 q2 )/p2
=
.
q = (Tp )1 (q ) =
(3.12)
q2
q2
p2 0 p2 /n
Now,

q1 + q2 = nq1 + q2 (p1 + p2 ) /p2 n(q1 + q2 )/p2 < n.
(3.13)
Following is an easy consequence: Minimal R-charge of the cc ring increases under

condensation of tachyon in cc-ring. More explicitly,
n1

min {lk1 /n} + {lk2 /n}
l=1
(3.14)
increases under tachyon condensation.

So far is story of cc-ring. We need to consider all chiral rings together. So we are
interested in the behavior of the R-charge which is smallest in the union of cc- and caring. To do this, we reconsider the problem of the fate of ca-ring under the condensation
of tachyon in cc-ring. Although we do not have any control over the flow of the ca ring
spectrum, we know what is the final theory and its total set of the spectrum. We ask whether
any tachyon mass of the final theory can be considered as an image of some mapping with
the property of R-charge increasing. To do this we want to show that there is a map that
takes the some of chiral ring of the mother theory to ca or ac ring of the daughter theories.
Notice that, in general, the ca ring of n(k1 , k2 ) is cc ring of n(k1 , k2 ) and the daughter
theory has structure p1 (k1 , p k/n) p2 (p k/n, k2 ). First, the ca ring of the daughter
theory p1 (k1 , p k/n) is cc-ring of p1 (k1 , p k/n) which is expected to be the image
of the cc ring of n(k1 , k2 ) under some mapping Fp+ , which is not necessarily associated
317
+
with physical process. It turns out that Fp+ can be chosen as T(p
:
1 ,p2 )
Fp+ (q)
:= Tp+ (q)
= (q1 , p1 p q/n),
where q = (q1 , n q2 ) ca ring and p = (p1 , p2 ). One can check that

R Fp+ (q)
> R[q]
if q ca ring.
(3.15)
(3.16)
Similarly, the ac ring of the daughter theory p2 (p k/n, k2 ) is cc ring of p2 (p

k/n, k2 ), which can be considered as the image of the cc ring of n(k1 , k2 ) by the map
Fp defined by
Fp (q)
:= Tp
(n q1 , q2 ) = (p2 + p q/n, q2 ),
where q = (n q1 , q2 ) ac ring. It can be also shown that

R Fp (q)
> R[q]
if q ac ring.
(3.17)
(3.18)
q0
Now let be the tachyon with lowest mass in the daughter theory. Let it belong to the
up-theory. Then we can without loss of generality assume that it belongs to ca ring due to
the equivalence ca and ac ring in their spectrum. (If it belongs to cc ring, we have shown
for some q,
which has bigger charge
already what we want to show.) Then q = Fp+ (q)
than the minimal charge of initial theory. Using the property of Eq. (3.16),
R[q ] R[q] R[q0 ]
(3.19)
q
as desired. Similarly, if belongs to down theory, we can assume that it belongs to the
ac-ring of the down theory. Then q = Fp (q)
for some q,
which has bigger charge than
the minimal charge of initial theory q0 . Using the property of Eq. (3.18),
R[q ] R[q]
R[q0 ]
(3.20)
as desired.
Therefore we proved following: in the conformal field theory of orbifolds, the minimal
R charge of the final theory is bigger than that of the initial theory under the condensation
of a tachyon generated by a world sheet fermion
n1

min min {lk1 /n} + {lk2 /n} 1, {lk1 /n} {lk2 /n}
l=1
(3.21)
increases when we compare its value in the initial and final theories. Equivalently, in terms
of absolute values of tachyon mass,
m(IR) m(UV).
(3.22)
A few remarks are in order:

These theorems are world sheet fact. The same statement conjectured in [7] is the GSO
projected version for which we need to take into account the GSO projection. However,
tachyons in NSNS sector is not projected out by the type 0 GSO projection, so the
above conclusion is true in type 0 string theory level. For type II we will discuss in
detail in next subsection;
318
There are two independent theories in the final stage of tachyon condensation. Each
theory will have its own minimal charges. We should take the smaller of the two, since
the minimal mass of final theory is the minimal over all final spectra. Namely, the
minimal R-charges of two theories in final stage are not to be added to compare with
the initial one, contrary to the treatment of gcl in [4];
From the discussion of previous subsection, this is the proof of the conjecture stated
by Dabholkar and Vafa in [5]. The R-charge here is that in NS sector. The R-charge
of Ramond sector is related to that of NS sector by spectral flow. Since there are CPT
invariance in Ramond sector, the statement that minimal charge increases is equivalent
to statement that maximal charge decreases.
4. Discussion and conclusion

In this paper, we have studied the localized condensation in non-supersymmetric
orbifold using the (2, 2) world sheet SUSY in mirror LG picture. We study the localized
tachyon condensation in Mirror LandauGinzburg picture as well as the toric geometry
picture of non-supersymmetric orbifold backgrounds. Due to the two copies of (2, 2)
worldsheet supersymmetry, any worldsheet fermion generated tachyon can be considered
as a BPS state. Utilizing this fact, we showed that the R-charge of chiral primaries increases
under the process of localized tachyon condensation. The minimal tachyon mass in twisted
sectors increases in CFT and type 0 string and plays similar role of the c- or g-function in
the twisted sectors. By working out how the individual chiral primaries are mapped under
the tachyon condensation, we have proved that R-charges of chiral primaries increase under
tachyon condensation. We studied the GSO projection and found that in many aspects, the
separate notion of type II and type 0 is not preserved under the tachyon condensation.
We emphasized similarity of c-, g- and m-functions in the introduction. Here we want to
draw the readers attention to one more object; the orbifold Witten index = n, counting
the number of (un)twisted sector. Under the transition n(1, k) p1 (1, s) p2 (s, k),
obviously satisfy the relation
(M) (D1 ) + (D2 )
(4.1)
from the very fact that (p1 , p2 ) is a tachyon. This property is similar to the g-function.
Notice, however, for marginal case (M) = (D1 ) = (D2 ) does not hold for Witten
index unlike the c-function. This is just parallel to the case of minimal R-charge. Inspite of
its simplicity, we did not use this as an semi-c-object to study orbifold decay for following
reasons;
It does not depends on the generator k. Too many theories have the same Witten index.
Therefore, it is not a sharp indicator;
It cannot be defined off-shell. It is true topological index that is independent of finite
deformation of the theory. Therefore, it cannot be used for full study of RG-flow. The
minimal tachyon mass, on the other hand, can be defined off-shell along the line of
[17,18].
319
We now discuss the limitation and related future works. First of all, our work is confined
to orbifold fixed points before and after the tachyon condensation. It would be interesting
to work out the detail of the off-shell. One may ask what is the geometry for the finite
condensation coefficient of LG in terms of gauged linear sigma model? A work related to
this question has appeared [24]. Another related work is [25], where the Bondi energy [26]
as a c-function was discussed based on the earlier work by Tseytlin [27]. The use of this
work, however, seems to be rather limited for studying orbifold transitions, since it requires
that the asymptotic regions be invariant under the transition, which cannot be fulfilled by
the orbifolds.
Secondly, we considered the case where only one tachyon generated by a worldsheet
fermion and the method of this paper does not work if we consider simultaneous
condensation of two tachyons in the different chiral rings.
Thirdly, our work is mostly about CFT and type 0 theory rather than type II theory. For
type II theory, there is only one way by which the theorem can be broken, namely, if the
minimal charge of mother theory is projected out by the GSO and the minimal charge of the
daughter theory is kept, then it may happen that the minimal charge of the daughter theory
is smaller than that of the mother theory. It is very plausible that such possibility happens.
Surprisingly, however, in all example we considered, the tachyon condensation that cause
such possibility is forbidden due to GSO projection in all the examples we looked at so far.
We wish to come back to this issue in later publication.
Acknowledgements
I would like to thank Lance Dixon, Michael Gutperle, Shamit Kachru, Amir KashaniPoor, Matthias Klein and Michael Peskin for discussions, and especially to Eva Silverstein
for her support and interests, and to Allan Adams for his collaboration in the initial stage.
This work is partially supported by the Korea Research Foundation Grant (KRF-2002-013D00030).
References
[1] A. Sen, Non-BPS states and branes in string theory, hep-th/9904207, and references therein;
For review, see: A. Sen, Stable non-BPS states in string theory, JHEP 9806 (1998) 007, hep-th/9803194, and
references therein.
[2] A. Adams, J. Polchinski, E. Silverstein, Do not panic! Closed string tachyons in ALE spacetimes,
JHEP 0110 (2001) 029, hep-th/0108075.
[3] C. Vafa, Mirror symmetry and closed string tachyon condensation, hep-th/0111051.
[4] J.A. Harvey, D. Kutasov, E.J. Martinec, G. Moore, Localized tachyons and RG flows, hep-th/0111154.
[5] A. Dabholkar, C. Vafa, tt* geometry and closed string tachyon potential, hep-th/0111155.
[6] A. Dabholkar, Tachyon condensation and black hole entropy, Phys. Rev. Lett. 88 (2002) 091301, hepth/0111004.
[7] S.-J. Sin, Tachyon mass, c-function and counting localized degrees of freedom, Nucl. Phys. B 637 (2002)
395, hep-th/0202097.
[8] S.-J. Sin, Comments on the fate of localized tachyons.
[9] T. Takayanagi, T. Uesugi, Orbifolds as Melvin geometry, hep-th/0110099;
320
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
Y. Michishita, P. Yi, D-brane probe and closed string tachyons, Phys. Rev. D 65 (2002) 086006, hepth/0111199;
S. Nam, S.-J. Sin, Condensation of localized tachyons and spacetime supersymmetry, hep-th/0201132;
R. Rabadan, J. Simon, M-theory lift of braneantibrane systems and localised closed string tachyons,
JHEP 0205 (2002) 045, hep-th/0203243;
A.M. Uranga, Localized instabilities at conifolds, hep-th/0204079;
T. Sarkar, Brane probes, toric geometry, and closed string tachyons, Nucl. Phys. B 648 (2003) 497, hepth/0206109;
T. Suyama, Closed string tachyons and RG flows, JHEP 0210 (2002) 051, hep-th/0210054;
J.L. Barbon, E. Rabinovici, Remarks on black hole instabilities and closed string tachyons, hep-th/0211212;
Y.H. He, Closed string tachyons, non-supersymmetric orbifolds and generalised McKay correspondence,
hep-th/0301162.
A.B. Zamolodchikov, Irreversibility of the flux of the renormalization group in a 2-D field theory, JETP
Lett. 43 (1986) 730, Pisma Zh. Eksp. Teor. Fiz. 43 (1986) 565 (in Russian).
I. Affleck, A.W. Ludwig, Universal noninteger ground state degeneracy in critical quantum systems, Phys.
Rev. Lett. 67 (1991) 161.
V.I. Arnold, On some problems in singularity theory, in: Geometry and Analysis, Papers dedicated to
V.K. Patodi, 1981, pp. 19.
V.I. Arnold, in: Sinularity Theory (Selected Papers), Cambridge Univ. Press, Cambridge, 1981;
V.I. Arnold (Ed.), Dynamical Systems VI, Springer.
A.N. Varchenko, Semi-continuity of the spectrum and an upper bound for the number of singular points of
a projective hypersurface, Sov. Math. Dokl. 27 (1983) 735739.
J.H.M. Steenbrink, Semicontinuity of the singularity spectrum, Invent. Math. 79 (1985) 557566.
C. Vafa, String vacua and orbifoldized LG models, Mod. Phys. Lett. A 4 (1989) 1169;
K.A. Intriligator, C. Vafa, LandauGinzburg orbifolds, Nucl. Phys. B 339 (1990) 95.
S. Cecotti, C. Vafa, Topological antitopological fusion, Nucl. Phys. B 367 (1991) 359.
S. Cecotti, P. Fendley, K.A. Intriligator, C. Vafa, A New supersymmetric index, Nucl. Phys. B 386 (1992)
405, hep-th/9204102.
E. Witten, Phases of N = 2 theories in two dimensions, Nucl. Phys. B 403 (1993) 159, hep-th/9301042.
K. Hori, C. Vafa, Mirror symmetry, hep-th/0002222.
L.J. Dixon, D. Friedan, E.J. Martinec, S.H. Shenker, The conformal field theory of orbifolds, Nucl. Phys.
B 282 (1987) 13.
A. Dabholkar, Strings on a cone and black hole entropy, Nucl. Phys. B 439 (1995) 650, hep-th/9408098.
D.A. Lowe, A. Strominger, Strings near a Rindler or black hole horizon, Phys. Rev. D 51 (1995) 1793,
hep-th/9410215.
E.J. Martinec, G. Moore, On decay of K-theory, hep-th/0212059;
E.J. Martinec, Defects, decay, and dissipated states, hep-th/0210231.
M. Gutperle, M. Headrick, S. Minwalla, V. Schomerus, Spacetime energy decreases under world-sheet RG
flow, JHEP 0301 (2003) 073, hep-th/0211063.
H. Bondi, M. van der Burg, A. Metzner, Gravitational waves in general relativity VII. Waves from axisymmetric isolated systems, Proc. Roy. Soc. London Ser. A 269 (1962) 21.
A.A. Tseytlin, Conditions of Weyl invariance of two-dimensional sigma model from equations of stationarity
of central charge action, Phys. Lett. B 194 (1987) 63.

Supersymmetric effects in deep inelastic

neutrinonucleus scattering
A. Kurylov a , M.J. Ramsey-Musolf a,b,c , S. Su a
a California Institute of Technology, Pasadena, CA 91125, USA
b Department of Physics, University of Connecticut, Storrs, CT 06269, USA
c Institute for Nuclear Theory, University of Washington, Seattle, WA 98195, USA
Received 29 January 2003; received in revised form 19 May 2003; accepted 11 June 2003
Abstract
We compute the supersymmetric (SUSY) contributions to ( )-nucleus deep inelastic scattering
in the Minimal Supersymmetric Standard Model (MSSM). We consider the ratio of neutral current
to charged current cross sections, R and R , and compare with the deviations of these quantities
from the Standard Model predictions implied by the recent NuTeV measurement. After performing
a model-independent analysis, we find that SUSY loop corrections generally have the opposite sign
from the NuTeV anomaly. We also study for R-parity-violating (RPV) contributions. Although RPV
effects could, in principle, reproduce the NuTeV anomaly, such a possibility is also ruled out by other
precision electroweak measurements.
PACS: 12.60.Jv; 13.15.+g; 12.15.LK
1. Introduction
Neutrino scattering experiments have played a key role in elucidating the structure of
the Standard Model (SM). Recently, the NuTeV Collaboration has performed a precise
determination of the ratio R (R ) of neutral current (NC) and charged current (CC) deepinelastic ( )-nucleus cross sections [1], which can be expressed as
2
2
R( ) = gLeff + r (1) gReff ,
E-mail address: shufang@theory.caltech.edu (S. Su).
doi:10.1016/S0550-3213(03)00528-5
(1)
322
A. Kurylov et al. / Nuclear Physics B 667 (2003) 321348
CC
eff 2
where r = CC
N /N and (gL,R ) are effective hadronic couplings (defined below).
eff )2 with the values obtained by the NuTeV
Comparing the SM predictions [2,3] for (gL,R
exp
SM
Collaboration yields deviations R( ) = R( ) R(
)
R = 0.0029 0.0015,
R = 0.0015 0.0026.
(2)
Within the SM, these results may be interpreted as a test of the scale-dependence of the
eff 2
sin2 W since the (gL,R
) depend on the weak mixing angle. While the SM prediction for
2
sin W at = MZ has been confirmed with high precision at LEP and SLC, the predicted
running of this parameter to lower scales has yet to be studiedsystematically. The results
from the NuTeV measurement imply a +3 deviation at 20 GeV, while the current
value of the cesium weak charge, extracted from atomic parity-violation (PV), implies
agreement with the predicted SM running at a much lower scale [4]. Measurements of the
PV asymmetries in polarized ee and ep scattering will provide further tests of this running
at 0.2 GeV [5,6].
This interpretation of the NuTeV results has been the subject of some debate.
Unaccounted for QCD effects, such as charge symmetry-breaking in parton distributions
or nuclear shadowing [79], have been proposed as possible remedies for the anomaly.
Alternatively, one may consider physics beyond the SM, as reviewed in Ref. [7]. In what
follows, we focus on one new physics scenario, namely, supersymmetry (SUSY). While a
brief discussion of SUSY is given in Ref. [7], an extensive, detailed treatment has yet to
appear in the literature. Because SUSY is one of the most strongly-motivated extensions
of the SM, undertaking such an analysis is a timely endeavor. The goal of the present study
is to provide this comprehensive treatment.
In performing our analysis, we work within the framework of the Minimal Supersymmetric Standard Model (MSSM), which remains the standard baseline for considering
SUSY effects in precision observables. Typically, MSSM studies adopt one or more models
of SUSY-breaking mediation, thereby vastly simplifying the task of analyzing the MSSM
parameter space. Here, however, we carry out a model-independent treatment, avoiding
the choice of a specific mechanism for SUSY-breaking mediation. Since generic features
of the superpartner spectrum implied by the most widely adopted SUSY-breaking models
may not be consistent with precision data [10], we wish to determine whether there exist
any choices for MSSM parameters that could account for the NuTeV result, even if such
choices lie outside the purview of standard SUSY-breaking models.
We find that it is difficult to choose MSSM parameters so as to improve agreement with
the NuTeV result. When R-parity is conserved and SUSY effects only arise via radiative
corrections, the magnitude of their contribution is generally too small for the NuTeV
anomaly to generate significant constraints. Moreover, for nearly all parameter choices,
the sign of the SUSY corrections to R and R is opposite to that of the NuTeV anomaly.
We also consider possible tree-level contributions from R-parity-violating (RPV)
interactions, whose presence would render the lightest supersymmetric particle (LSP)
unstable. We observe that purely leptonic RPV effects, which arise via the definition of
sin2 W , could in principle also generate a right sign effect to account for the NuTeV
anomaly. However, precision data also limits the magnitude of this contribution to be
323
considerably smaller than necessary. On the other hand, the NuTeV result does yield new
constraints on possible semileptonic RPV interactions.
In short, our qualitative conclusions agree with those of Ref. [7], though we believe we
have carried out a more exhaustive analysis. Thus, one would have to look to more exotic
new physics possibilities, such as a Z boson coupling only to second generation fermions,
if SM QCD effects are ultimately unable to explain the NuTeV anomaly. We note that the
situation here contrasts with that for the PV electron scattering asymmetries, where the
SM contribution is fortuitously suppressed and where SUSY radiative corrections could,
in principle, produce observable contributions [11].
The analysis leading to these conclusions is organized in the remainder of the paper
as follows. In Section 2, we discuss general features of MSSM contributions to ( )nucleus scattering, including both SUSY loop corrections and RPV effects. Section 3 gives
details of the loop computation as well as the scan over the MSSM parameter space. In
Section 4, we give the RPV analysis, including results of a fit to other precision data.
Section 5 contains a summary of our results. Explicit formulae for loop corrections with
relevant Feynman rules are given in Appendices A and B.
2. N ( N ) scattering in the MSSM: general considerations
For momentum transfers q satisfying |q 2 |
MZ2 , the neutrinoquark interactions can
be represented with sufficient accuracy by an effective four fermion Lagrangian:
NC

q

G N
q
(1 5 )
LNC
=
q
2L PL + 2R PR q,
q
2
q
(3)
CC
G N
(1 5 ) u
=
(1 5 )d + h.c.,
LCC
q
2
(4)
where PL = (1 5 )/2, PR = (1 + 5 )/2 and

q
L = IL3 Qq sin2 W + L ,
(5)
q
R
(6)
= Qq sin
q
W + R .
q
NC
CC
= N
= = 1 and L,R = 0 at tree-level in the SM. These
The parameters N
quantities differ from their tree-level values when O() corrections in the SM or MSSM
are included or when other new physics contributions arise. The precise values of these
quantities individually are renormalization scheme-dependent. When computing MSSM
loop contributions, we use the modified dimensional reduction (DR) scheme, in which
the spatial dimension of momenta are continued into d = 4 2 dimensions while the
Dirac matrices remain four-dimensional, as required by SUSY invariance. Quantities
renormalized in the DR scheme will be indicated by a hat. Note also that for the neutrino
NC
CC
reactions of interest here, N
, N
, and are universal (independent of quark flavor),
q
while the L,R are flavor-dependent.
The NC to CC cross section ratios R and R can be expressed in terms of the above
eff )2 appearing in Eq. (1) in a straightforward
parameters via the effective couplings (gL,R
324
Fig. 1. One loop contributions to the neutrinoquark neutral current amplitude. The blob denotes the one loop
irreducible diagram or the counter term. The Feynman diagrams for the external leg corrections are not shown
explicitly.
Fig. 2. One loop contributions to the neutrinoquark charge current amplitude. The blob denotes the one loop
irreducible diagram or the counter term. The Feynman diagrams for the external leg corrections are not shown
explicitly.
way:
eff
gL,R
2
2 2 2
NC 2
q 2
MW q 2 2 N
MZ
L,R .
2
2 q 2
M
M
CC
W
(7)
The SM values for these quantities are [2,3] (gLeff )2 = 0.3038 0.0003 and (gReff )2 =
0.03011 while the NuTeV results imply (gLeff )2 = 0.3005 0.0014 and (gReff )2 = 0.0310
0.0011.
q
NC,CC
, , L,R , R ,
In what follows, we concentrate on the MSSM contributions to N
and R . When the R-parity quantum number (1)3(BL)+2S is conserved, the MSSM
contributes to these quantities only via loop effects. The relevant diagrams are shown in
Figs. 1, 2 and Appendices A and B. Note that all gauge boson self energy corrections,
as well as leptonic vertex and external leg corrections, contribute only to the universal
NC,CC
and . For the NC amplitudes, non-universal box diagrams and quark
parameters N
L,R;q
q
vertex and external leg corrections (V B ) are contained in the L,R . All CC box graphs
CC . The CC box
as well as hadronic vertex and external leg contributions (VCCB ) appear in N
graphs also generate amplitudes involving products of scalar and pseudoscalar currents.
However, the corresponding O() corrections to R and R are suppressed by lepton
1 The error in (g eff )2 is smaller than 0.00005, so it was rounded down to zero in the Particle Data Group
R
listing [2,3].
325
and quark masses, so we neglect them here. In terms of these various corrections, the
renormalized parameters are
W W (q 2 )
W W (0)
+ VCCB V B ,
2
2
2
MW q
MW
ZZ (q 2 )
W W (0)
NC
N
=1+ 2
+ V V B ,
2
2
MZ q
MW
Z (q 2 ) sin2 W

c
A, q 2 ,
+
4c2 F
= 1 +
2
2
s
q
sin W
CC
N
=1+
L,R = V B
q
L,R;q
(8)
(9)
(10)
(11)
V V (q 2 ) are the gauge boson self-energies renormalized in the DR scheme at a

where
scale = MZ ; V B denote vertex, external leg, and box graph corrections entering the
muon-decay amplitude; and V indicates the neutral current neutrino vertex and lepton
external leg correction. Note that the muon decay corrections enter the semileptonic
amplitudes since the Fermi constant G is taken from the muon lifetime. The correction
A, arises from the charge radius.2 Superpartner loop contributions to
Z (q 2 = 0)
F
2
2

are zero, so that Z (q )/q gives a finite contribution to at the photon point. The shift
in sin2 W arises from its definition in terms of , G , and MZ
,
s2 c2 =
2
2 MZ G [1 +r (MZ )]
(12)
s 2 = 1 c2 = sin2 W (MZ ) and [12]

(0) + 2
+r =
ZZ (M 2 )
Z (0)
W W (0)
s
+
+ V B
2
2
2
c MZ
MZ
MW
(13)
(q 2 ) =
(q 2 )/q 2 . Writing +r = +r SM + +r SUSY one has
with
2
sSUSY
c2
(14)
=
+r SUSY .
s 2
c2 s2
In Section 3, we discuss our computation of the MSSM loop contributions to these
parameters in detail; explicit expression for the various loop amplitudes appears in
Appendices A and B.
When R-parity is not conserved, however, new tree-level contributions appear. The
latter are generated by the B L violating superpotential
1
k + ij k Li Qj D
k + 1 ij k U
i D
j D
k ,
WRPV = ij k Li Lj E
(15)
2
2
where Li and Qi denote lepton and quark SU(2)L doublet superfields, Ei , Ui , and
Di are singlet superfields and the ij k etc. are a priori unknown couplings. We have
2 The charge radius is equivalent to its anapole moment in the MSSM.
326
Fig. 3. Tree level RPV contributions to muon decay (plot (a)), q charged current (plot (b)) and q
neutral current (plots (c) and (d)) amplitudes.
neglected additional leptonHiggs mixing term in Eq. (15) for simplicity. In order to avoid
unacceptably large contributions to the proton decay rate, we set the +B = 0 couplings
ij k to zero. The purely leptonic terms (12k ) contribute to neutrino scattering amplitudes
via the normalization of CC and NC amplitudes to G and through the definition of sin2 W
[13]. The remaining semileptonic, +L = 1 interactions (ij k ) give direct contributions
to the neutrino scattering amplitudes. The latter may be obtained computing the Feynman
amplitudes in Fig. 3 and performing a Fierz reordering. For neutrinoquark scattering, one
obtains the effective Lagrangian
LEFF
RPV =
|2k1 |2
2M 2 k
dR dR L L +
dL
|21k |2
2M 2 k
dL dL L L
dR
|21k |2
u L dL L L
2M 2 k
dR

+ h.c. ,
(16)
where we have taken |q 2 |

M 2 and have retained only the semileptonic terms relevant to
f
q scattering.
In terms of these parameters, the shifts induced in the effective ( )N parameters
q
CC,NC
and L,R are:
N
k
NC
,
= 12k eR
N
(17)
k
CC
NC

= + 21k d ,
(18)
N
= 21k
k
1
+ x 12k eR
,
3
1
k
,
Rd = 2k1 dLk + x 12k eR
3
k
2
,
Lu = Ru = x 12k eR
3
where
Ld
dRk
|ij k |2
ij k (f) =
4 2 G M 2
f
(19)
(20)
(21)
(22)
327
with a similar definition for the primed quantities, and [13]

s 2 c2
0.35.
c2 s 2
The corresponding shifts in R( ) are

k
SM
k
4 u 2 d
d
R( ) = x L + L 1 + r (1) 12k eR
2 R(
) + L 21k dR
3
3

+ 2r (1) Rd 2k1 dLk

k
SM
k
0.25 1 + r (1) 12k eR

2 R(
) 0.43 21k dR

+ 1.6 r (1)2k1 dLk .
x
(23)
(24)
k ) and (d k ) are constrained by other precision

As we discuss in Section 4, 12k (eR
21k R
electroweak data, while 2k1 (dLk ) is relatively unconstrained. In Eq. (24), the coefficients
k
) is negative. Since
of 21k (dRk ) and 2k1 (dLk ) are positive, while the coefficient of 12k (eR
k ) and rather small

the ij k are non-negative, we would require sizable value of 12k (eR
values of (dk ) and (dk ) to account for the negative shifts in R and R implied
21k
2k1
k ), however, are fairly stringent,

by the NuTeV analysis. The present constraints on 12k (eR
ruling out sizable values for the semileptonic corrections with fairly high confidence.3
3. SUSY loop contributions

Instead of working in a specific SUSY breaking scenario, we perform a modelindependent analysis by varying all the possible soft SUSY-breaking parameters [14].
Although such an approach is insensitive to the effects of any particular SUSY breaking
parameter, it does allow us to obtain the size of SUSY contributions in the most general
way. In our analysis, we set the momentum transfer q 2 = 0.
We first compute loop effects only (R-parity being conserved) by scanning over the
MSSM parameters in the ranges shown in Table 1. Here, tan = vu /vd is the ratio of the
up and down type Higgs vacuum expectation values; is bilinear Higgs coupling in the
supersymmetric Lagrangian, which gives the mass to the higgsino; M1 , M2 and Mg are the
masses for the U(1)Y , SU(2)L and SU(3)c gauginos, respectively; M i are the diagonal
fL,R
mass parameters for the left- and right-handed squarks and sleptons of generation i, while
(M i )2 are the leftright mixing parameters. In order to avoid unacceptably large flavorfLR
changing neutral currents, we do not allow for flavor mixing between squark generations
but do allow for superpartners of different generations to have different masses.
3 The authors of Ref. [7] considered the implications of a general set of four-fermion, semileptonic
interactions. Their analysis differs from our analysis of RPV SUSY effects, however, which also includes the
k ) corrections). Inclusion of the latter are required since
impact of purely leptonic interactions (i.e., the 12k (eR
they affect both the normalization of CC and NC amplitudes in terms of G and the value of sin2 W . For a more
general discussion, see Ref. [13].
328
Table 1
denotes any of ||, M1,2,g , or the diagonal sfermion mass
Ranges of SUSY parameters scanned. Here, M
parameters M i . The parameter can take either sign. The generation index i runs from 1 to 3
fL,R
Parameter
Min
Max
tan

M
i
M
1.4
60
50 GeV
106 GeV2
1000 GeV
106 GeV2
2
fLR
In randomly choosing values for these parameters, we follow the conventional practice
of using a linear distribution for all parameters except tan , for which we use a logarithmic
distribution. We discard any points that yield SUSY particle masses below present collider
lower bounds. In addition, we impose constraints from the Z-pole electroweak precision
measurements, which are embodied in terms of the three oblique parameters S, T and U
NC and (Eqs. (9) and (10)) in terms
[15]. To that end, we express the SUSY shifts in N
of these quantities:
NC
N
= T
V B + V ,

2
Z (M 2 )
SUSY
c
c2
Z (q )
Z
+
= 2
S
VB
c s2
4s 2 c2
s
q2
MZ2

(MZ2 ) + SUSY
c2
SUSY 2
A,
(25)
q ,
+ 2
+
4c2 F
2
2
c s
MZ
where remaining terms proportional to q 2 are negligible and have been dropped [recall
that includes the effect of changes in s 2 due to SUSY loops as per Eq. (14)]. Here
+ is the SUSY contribution to the difference between the fine structure constant and the
electromagnetic coupling renormalized at = MZ : + = [(M
Z ) ]SUSY .
Note that only S and T enter these expressions. Since these quantities are correlated, we
use the 95% C.L. S T constraints [2] and retain only those parameter choices consistent
with these constraints. We observe that this procedure is not entirely self-consistent, since
we have not taken into account non-oblique corrections to Z-pole observables in deriving
the oblique parameter constraints. Moreover, as discussed in Ref. [16], non-oblique
SUSY loop contributions to some Z-pole observables can be important. Nevertheless, the
essential, qualitative implications of precision Z-pole data for SUSY loop effects on the
q parameters are unaffected by this inconsistency. We also omit parameter choices
generating too large a SUSY contribution to the muon anomalous magnetic moment [17].
Doing so limits leftright mixing for second generation sleptons to be fairly small.
Before presenting our results, we also comment on the inputs used in Ref. [7]. In that
study, the authors only investigated the slepton contribution with four MSSM parameters:
M1 , , tan and Ml, assuming the GUT relation for the gaugino masses and the slepton
mass degeneracy. Squark contributions were neglected because they would inevitably
NC ) of the wrong sign. In our analysis, we include the
induce a contribution to T (or N
329
Fig. 4. Plot (a) gives the MSSM contribution to R and R with MSSM parameters chosen randomly from range
shown in Table 1. Plot (b) shows the dependence of R on the oblique parameter T with the random MSSM
parameter set.
contribution from both the squark and slepton sectors, taking into account sfermion left
right mixing, and allowing for non-universality between generations.
With a sample of about 3000 randomly selected parameter sets, we calculated the
MSSM contributions to R and R . The numerical results are shown in Fig. 4(a). We
observe that R and R are highly correlated. This correlation arises because MSSM
contributions to the r (1) (gReff )2 term in Eq. (1) are small, while the contribution to the
(gLeff )2 terms are the same for both R and R . We also find that the magnitude of
R, is dominated by the SUSY contributions to T (see Figs. 4(b) and 5), which is
sensitive to isospin-breaking in the SUSY spectrum. It is bounded above by other precision
330
Fig. 5. Plot (a) shows the contributions to R from NC CC (dashed line), (dotted line) and L,R
(dash-dotted line). The solid line is the sum of all the contributions to R . Plot (b) shows the size of various
CC
components of NC CC (solid line): T (dashed line), V B (dotted line) and VNC
B V B (dash-dotted line).
The x-axis is the common first generation squark mass and first and second generation slepton mass. The other
MSSM parameters are chosen to be tan = 10, 2M1 = M2 = = 200 GeV. The mass for the second and third
generation squarks and third generation slepton are taken to be 1000 GeV. Note that neutralinos and charginos
are light, which leads to a non-decoupling effect in R even when squarks and sleptons are heavy.
electroweak data, thereby limiting the size of possible SUSY loop contributions to R and
R to be considerably smaller than the deviations in Eq. (2). A breakdown of the various
classes contributions is given in Fig. 5. More significantly, the sign of the SUSY loop
corrections is nearly always positive, in contrast to the sign of the NuTeV anomaly.
331
We do, however, find one corner of the parameter space which admits a negative loop
contribution. This scenario involves gluino loops, whose effect can become negative when
the first generation up-type squark and down-type squarks are nearly degenerate and left
right mixing is close to maximal. In this particular region of the MSSM parameter space,
there is no gluino loop correction to the quark charged and neutral vector currents, while
the axial currents receive large corrections. When M 3 = 0 and the third generation
fLR
soft parameters are sufficiently heavy4 (corresponding to a small value for T ), the gluino
contribution could give rise to a negative correction to R, . However, equal and large left
right mixing for both up- and down-type squarks is inconsistent with the other precision
electroweak inputs, such as MW and charged current universality [10]. Moreover, a gluino
loop effect large enough to account for the NuTeV anomaly would require a large splitting
between the two up-type (down-type) squark mass eigenvalues, which in turn requires large
leftright mixing M 1,2
in the first two generation off-diagonal squark mass matrices. Such
fLR
large values for M 1,2

entail breaking the color and charge neutrality of the vacuum [18].
fLR
It is interesting to note that if these other considerations did not rule out a sizable gluino
loop effect, it could in principle affect the value of sin2 W obtained from the NuTeV
analysis. The latter effectively relies on a modified version of the PaschosWolfenstein
relation [19]
R

1
R rR
= 1 2 sin2 W + ,
1r
2
(26)
where we have shown the SM prediction for R with the + indicating higher-order
corrections. In practice, the extraction of sin2 W from the NuTeV data relies on a fit to
data that is approximately equivalent to using a slightly different combination of R and
R in the numerator,
R R
R
(27)
1r
with differing from r in order to account for charm quark mass uncertainties in the
sin2 W extraction [20,21]. It is straightforward to show that gluino loop contributions to
are proportional to r. When | r| is small, gluino loop effects on sin2 W are
R
suppressed. Specifically, we find that for maximal leftright mixing u1 = 1 = /4,
d

1
= s ( r) 1 + r s 2 1 5 s 2 2V2 Mg , M 1
R
q heavy , Mq light
3
r(1 r)
9
1
1

V2 Mg , Mq heavy, Mq heavy V2 Mg , Mq1 light, Mq1 light ,
(28)
where s is the strong coupling constant and V2 (M, m1 , m2 ) is defined in Eq. (B.11). In
the analysis of NuTeV data, two different fitting procedures were used, equivalent to two
different values of [21]. The 0C fit left mc as a free parameter,5 and was equivalent
4 The second generation squark mass parameters are chosen to be the same as for the first generation to avoid
conflicts with low-energy constraints.
5 For this fit, the discrepancy with the SM prediction for sin2 is 2.5 .
W
332
to = 0.453. For this case, gluino loop effects on the extracted value of sin2 W are
suppressed. The 1C fit included an additional constraint on mc from dimuon production
[22] and corresponded to approximately = 0.249. In this case, the gluino loop effect
could be more substantial were it not constrained by other considerations.
4. RPV contributions
As in the case of SUSY loop corrections, the effects of RPV contributions to R and
R are correlated with similar effects on other precision electroweak observables [13]. The
relevant correlations are indicated in Table 2, where we list the RPV contribution to four
relevant precision observables: superallowed nuclear -decay that constrains |Vud | [23],
atomic PV measurements of the cesium weak charge QCs
W [4], the e/ ratio Re/ in l2
decays [24], and a comparison of the Fermi constant G with the appropriate combination
of , MZ , and sin2 W [25]. The values of the experimental constraints on those quantities
are given in the last column.
Given the experimental constraints on the first four quantities in Table 2, we obtain the
k
) and 21k (dRk ) shown in Fig. 6(a) by the solid ellipse.6
one allowed region for 12k (eR
() 2
Since the RPV corrections ()
ij k |ij k | 0, the physically allowed regionindicated by
the shaded region in Fig. 6(a)corresponds to all of the ()
ij k in Table 2 being non-negative.
k
k

Taking the values of 12k (eR ) and 21k (dR ) from this region, we obtain the allowed shifts
in R and R shown in Fig. 6(b), dashed ellipse. We also show the corresponding 95% C.L.
region [solid line in Fig. 6 (b)]. For simplicity, we have set 2k1 (dLk ) = 0 since a non-zero
value would yield only a positive contribution to these quantities. Even so, the possible
effects on R and R are by and large positive. While small negative corrections are also
possible, they are numerically too small to be interesting.
Table 2
Cs
RPV contributions to |Vud |2 /|Vud |2 , QCs
W /QW , Re/ , G /G , R and R . Columns give the
coefficients of the various corrections ij k and 12k to the different quantities. The last column gives the
experimentally measured value of the corresponding quantity
k)
11k (dR
|Vud
|2 /|V
Cs
QCs
W /QW
Re/
ud
|2
2
4.82
2
k)
1k1 (qL
0
5.41
0
G /G
R
R
0
0
0
0
k)
12k (eR
2
0.05
0
1
0.21
0.077
k)
21k (dR
k)
2k1 (dL
Quantity
0.0029 0.0014
0
2
0
0
0.0040 0.0066
0.0042 0.0033
0.22
0.132
0.08
0.32
6 In performing the fit, we allow the signs of , to be unconstrained.

ij k
ij k
0.00025 0.001875
0.0029 0.0015
0.0015 0.0026
333
k ) (d k ) plane, with the best fit value denoted by the

Fig. 6. Plot (a) shows the 1 allowed region in 12k (eR
21k R
cross. The solid (dashed) ellipse is the fit excluding (including) the NuTeV R, results. Shading indicates the
physically allowed region, corresponding to ij k > 0 and ij k > 0, for the fit excluding NuTeV results. Plot (b)
shows the prediction for R and R , using the 95% C.L. (solid line) or 1 (dashed line) allowed values for
k ) and (d k ) from fitting to |V |2 /|V |2 , QCs /QCs , R
k

12k (eR
ud
ud
e/ and G /G with 2k1 (dL ) set
W
W
21k R
to zero.
We also performed a fit to the five RPV parameters including in addition the NuTeV
k ) and (d k ) is given by
results for R and R . The one allowed region for 12k (eR
21k R
()
the dashed ellipse in Fig. 6(a). At 95% C.L., at least one of the ij k must be negative in
this fit. Thus, inclusion of the NuTeV results appears to exclude the RPV SUSY effects
summarized in Table 2 with fairly high confidence.
334
5. Conclusion
The NuTeV anomaly remains in need of explanation. If one is ultimately unable to
account for it with conventional, SM effects (e.g., small effects in parton distribution
functions), then one would require an explanation lying outside the SM. Here, we have
shown that such an explanation would be hard to come by in the MSSM alone. In general,
SUSY loop corrections to R and R generally have too small a magnitude and the wrong
sign to account for the effect.
R-parity-violating effects also appear at odds with the NuTeV anomaly. Inclusion
of these effects could resolve an apparent conflict between tests of charged current
universality and implications of SUSY-breaking models for the MSSM spectrum [10].
However they would also render the LSP unstable and, therefore, rule out SUSY dark
matter. At face value, the NuTeV results appear to disfavor this resolution of the charged
current universality problem. In short, we conclude that the MSSMwith or without
R-parity conservationis likely not responsible for the NuTeV anomaly. The culprit,
apparently, is to be found elsewhere.
Acknowledgements
We thank H. Schellman and J. Erler for useful discussions. We thank Petr Vogel and
Mark Wise for careful reading of the manuscript. This work is supported in part under
US Department of Energy contract #DE-FG03-02ER41215 (A.K. and M.J.R.-M.), #DEFG03-00ER41132 (M.J.R.-M.), and #DE-FG03-92-ER-40701 (S.S.). A.K. and M.J.R.-M.
are supported by the National Science Foundation under award PHY00-71856. S.S. is
supported by the John A. McCone Fellowship.
Appendix A. Essential Feynman rules

The complete set of Feynman rules for the MSSM can be found in [14,26]. Here, we
give only a brief list of relevant vertices to be self-contained.
The fermionsfermiongaugino vertices are:
where e is the absolute value of the electron charge, gS is the SU(3)c coupling constant,
and a are Gell-Mann matrices [26], normalized according to tr[a b ] = 1/2 ab . We
use the capitalized letters I and J to denote the family index for quarks and leptons
(I, J = 1, . . . , 3), small letters i and j to denote the index for squarks and sleptons
(i, j = 1, . . . , 6 except for sneutrino, when i, j = 1, . . . , 3), and small letters p and n to
335
denote the index for the neutralinos (p, n = 1, . . . , 4) and charginos (p, n = 1, 2). The
index a is reserved for the gluino index, a = 1, . . . , 8.
and
The first vertex represents the coupling of the fermion FJ to the sfermion F
i
+
+c
chargino p (or its charge conjugation field p if F = D, L). The chargino coupling
F ip
J
for the quark sector are as follows (the repeated index is summed over):
constants gL,R

1
mDI
U ip
I +3,i
Ii
gL J = VIJ ZD
ZD
Up1
Up2 ,
s
2 MW cos
mUI
1
U ip
Ii
ZD
Up2 ,
gR J = VIJ
s
2 MW sin

1
mUI
DJ ip
I +3,i
I i
gL
= VI J ZU Vp1
Vp2 ,
ZU
s
2 MW sin
mDI
1
D ip
I i
ZU
Vp2 ,
gR J = VI J
(A.1)
s
2 MW cos
where s (c) is the sine (cosine) of Weinberg angle, VI J is the usual CKM matrix, mUI , mDI
is the mass of up or down quark for generation I, tan = vu /vd is the ratio of the
expectation values of the Higgs scalars, ZU,D are the unitary 6 6 mixing matrices for up
and down squarks, respectively, that diagonalize full sfermion mass matrices, and U and V
are the unitary mixing matrices that diagonalize the chargino mass matrix [14]. In practical
calculations masses of the first and second generation quarks can be neglected as they are
much smaller than the mass of the W boson. Similarly, for lepton sector,

1
mL J
J ip
J +3,i
Ji
gL = ZL Up1
Up2 ,
ZL
s
2 MW cos
ip
gRJ
= 0,
= ZJ i Vp1
,
s
mL J
1
L ip
gR J =
Z J i Vp2 .
s 2 MW cos
L ip
gL J
(A.2)
Note that the mixing matrix for sneutrinos Z is 3 3 because in the MSSM neutrinos are
purely left-handed.
The second vertex represents the coupling of a fermion to a sfermion and a neutralino.
FJ ip
The corresponding coupling constants g0L,0R
for the fermionsfermionneutralino vertex
are:

1
1
mUJ
UJ ip
J +3,i
J i
Z
Np4 ,
g0L =
ZU Np2 c + Np1 s +
3
MZ sin U
2 sc

1
mUJ
4 J +3,i
U ip
J i
Np1
Np4 ,
ZU
ZU
g0RJ =
MZ sin
2c 3

1
1
mDJ
D ip
J +3,i
Ji
ZD
Np2
g0LJ =
ZD
c Np1
s
Np3 ,
3
MZ cos
2 sc
336

1
mDJ
2 J +3,i
Ji
ZD Np1 +
ZD
=
Np3 ,
MZ cos
2c 3

1
J ip
ZJ i Np2
=
c Np1
s ,
g0L
2 sc
D ip
g0RJ
ip
J
g0R
= 0,

1
mL J
ZLJ +3,i Np3

,
=
c + Np1
s
ZLJ i Np2
MZ cos
2 sc

1
mL J
L ip
g0RJ =
ZLJ i Np3 ,
2ZLJ +3,i Np1 +
MZ cos
2c
L ip
g0LJ
(A.3)
where N is a 4 4 mixing matrix that diagonalizes neutralino mass matrix [14].

FJ i
are:
Finally, quarksquarkgluino (the third vertex) couplings gGL,GR
J i
UJ i
= 2 ZU
,
gGL
DJ i
Ji
gGL
= 2 ZD
,
J +3,i
UJ i
gGR
= 2 ZU
,
DJ i
J +3,i
gGR
= 2 ZD
.
(A.4)
The gauge bosongauginogaugino couplings are given as follows:
where

1
1
Opn = Np2 Un1 + Np3 Un2 ,

= Np2 Vn1 Np4 Vn2 ,
2
2

1
L
Opn
= Vp1 Vn1
Vp2 Vn2
+ pn s 2 ,
2

1
R
2
Opn = Up1 Un1 Up2 Un2 + pn s ,
2

1
1
L
R
L
Opn
Opn = Np3 Nn3 + Np4 Nn4 ,
= Opn
.
2
2
L
Opn
(A.5)
In addition, for the photoncharginochargino vertex A n+ p+ is given by

ie pn .
337
The gauge bosonsfermionsfermion vertices are given by
For the W -coupling:

1
I i Jj
ZU ,
gW D
VJ I Z D
j =
i U
2s
1
gW
ZLI i ZJj
Li j =
2s
(A.6)
and for the Z and couplings:

gZ Fi Fj =

1 F I i Ij
I Z Z QF s 2 ij ,
sc 3 F F
g Fi Fj = QF ij .
(A.7)
, respectively.
Here, I3F and QF are the isospin and the electric charge of the sfermions F
The two scalar-two gauge boson vertices necessary for the calculation of the gauge
boson self energies are (see Ref. [26]):
where
1 Ki Kj
Z Z g ,
2s 2 F F

1
Kj
gZZ Fi Fj = 2 2 2I3F I3F 2QF s 2 ZFKi ZF + 2Q2F s 4 ij g ,
s c

1
Kj
gZ Fi Fj = QF 2I3F ZFKi ZF 2QF s 2 ij g ,
sc
g Fi Fj = 2Q2F ij g .
gW W Fi Fj =
(A.8)
Appendix B. Expressions for the relevant Feynman diagrams

Radiative corrections in MSSM have been studied extensively in the literature [27
29]. To be self-contained, we list in this section analytical expressions for the MSSM
contributions to the gauge boson self-energies, the fermion wave function renormalization,
the gauge bosonfermionfermion vertex correction, and the box diagrams relevant to
the neutrinonucleus scatterings. Modified dimensional reduction (DR) scheme is used in
computing MSSM loop contributions, although in Appendices A and B, we have neglected
the hat in all relevant variables.
338
B.1. Gauge boson self-energies

We first define the following class of two-point integration functions:
1
Fn (m1 , m2 , m3 ) =

x n ln xm21 + (1 x)m22 x(1 x)m23 /2 ,
(B.1)
F12 (m1 , m2 , m3 ) = F1 (m1 , m2 , m3 ) F2 (m1 , m2 , m3 ),
(B.2)
m23
will be replaced with either

where is the renormalization scale. In our analysis,
the external fermion mass squared or one of the Mandelstaam variables. For the first two
generation quarks and the first two generation leptons the fermion mass can be set to zero.
Focusing on the case when all Mandelstaam variables are small compared to m21,2 , we
obtain:
F0 (m1 , m2 , 0) = ln m21 1 +
m22
m22 m21
ln
m22
m21
ln 2 ,

2m22
2m42
m22
1
2
2
F1 (m1 , m2 , 0) = 2 ln m1 1 + 2
+
ln
2 ln .
4
m1 m22 (m22 m21 )2 m21
(B.3)
B.1.1. Gaugino loops
The Feynman diagrams of this type are shown in Fig. 7(a). The contribution of gaugino
loops to W and Z self-energies is given by

gL gR + gL gR mn mp F0 (mn , mp , q)
V V q 2 =
2 p,n

+ |gL |2 + |gR |2 2q 2F12 (mn , mp , q)

m2n F1 (mn , mp , q) m2p F1 (mp , mn , q) ,
(B.4)
where q is the momentum carried by gauge boson V , and q 2 = q q . The couplings gL
and gR are listed in Table 3 for W W and ZZ .
The Z mixing tensor and photon self-energy from chargino loops is given by

+
V V q 2 = q 2
(B.5)
gV gV F12 (mp+ , mp+ , q),
p
with the couplings gV and gV listed in Table 4.
Fig. 7. Feynman diagrams of one loop SUSY contributions to the gauge boson self energy.
339
Table 3
V V
0 +
W W
+ +
ZZ
0 0
ZZ
gL
gR
p0
n+
L /s
Opn
R /s
Opn
p+
n+
L /sc
Opn
R /sc
Opn
p0
n0
L /sc
Opn
R /sc
Opn
Comment
multiply 1/2
Table 4
V V
gV
gV
+
Z
L + O R )/sc
(Opp
pp
Comment
multiply 2
Table 5
(V , V )
i , F
j )
(F
i , U
j )
(
Li , j ), (D
(W, W )
(Z, )
i , U
j ), (D
i , D
j )
Li ,
Lj ), (U
(i , j ), (

(Li , Lj ), (Ui , Uj ), (Di , Dj )
( , )
i , U
j ), (D
i , D
j )
Lj ), (U
(
Li ,
(Z, Z)
B.1.2. Scalar loops

Diagrams of this type are shown in Fig. 7(b), (c). The contribution of Fig. 7(b) to W W ,
ZZ , Z and is given by

q2

VFVF q 2 =
gV Fi Fj gV F F m2F + m2F
i j
i
j
4
3
i,j
2
2 mF F1 (mFi , mFj , q) + m2F F1 (mFj , mFi , q)
i
j

2
q F12 (mFi , mFj , q) .
(B.6)
j ) running in the loop for V V is listed in Table 5. For squark
i , F
The sfermion pair (F
contributions, an additional color factor of Nc enters the right-hand side of Eq. (B.6).
The contribution of Fig. 7(c) to the self-energies is given by

VFV q 2 =
(B.7)
gV V Fi Fi m2F 1 ln m2F + ln 2 ,
i
i
4
i
= ,
and D
except for V = , for which does not contribute.
where F
L, U
2
When q 0, Z (q 2 )/q 2 and (q 2 )/q 2 from sfermion loops reduce to

V V (q 2 )

=
gV Fi Fi gV F F ln m2F ln 2 .
i i
i
q2
12
i
(B.8)
340
Fig. 8. Feynman diagrams for one loop SUSY contribution to the fermion wave function renormalization.
Fig. 9. Feynman diagrams for one loop SUSY contribution to the gauge coupling vertex.
B.2. Field strength renormalization for fermions

The diagrams contributing to the field strength renormalization for fermions all have
the form shown in Fig. 8. Therefore, the same formula can be applied to contributions
with charginos and neutralinos in the loop, provided that the appropriate couplings and
masses are used. The gluino contributions are multiplied by an extra factor of Casimir
factor C2 (N) = 4/3 for SU(3)c . The field strength renormalization for the left-handed
quark FJ is
FJ ip 2
FJ ip 2
F
gL
g0L
F1 (mF , mp+ , mFJ ) +
F1 (mFi , mp0 , mFJ )
ZLJ =
i
4
4
i,p
4 S FJ i 2
gGL F1 (mFi , mg , mFJ ),
+
3 4
i,p
(B.9)
where mg is the gluino mass and F stands for the isopartner of the fermion F (e.g., if F is
the up-type quark, then F is the down-type quark). By replacing L R one easily obtains
the field strength renormalization for the right-handed fermions. The same formulae apply
to the sleptons provided that the appropriate couplings and masses are used. Naturally,
corrections involving the strong coupling are absent in that case.
B.3. Vertex corrections
One-loop SUSY corrections to the V f f vertex are shown in Fig. 9. There are two
types of corrections: loops with the vector boson coupling to the scalar particles, Fig. 9(a),
and loops with the vector bosons coupling to the gauginos, Fig. 9(b). The complete set of
Feynman diagrams for each vertex, such as W du, can contain more than one diagram of
each type. However, since these diagrams differ only in the specific values of the masses
and couplings needed to obtain the numerical answer we do not show all graphs explicitly.
Below, we use the superscripts (a) and (b) to distinguish between the two types of loop
graphs. To distinguish the graphs of the same type, we supplement the superscript with a
number.
341
We first define the three-point integration functions needed for the evaluation of the
triangle diagrams:
1
V1 (M, m1 , m2 ) =
1
dx
dy
V2 (M, m1 , m2 ) =
dx
0
y
,
D3 (M, m1 , m2 )

dy y ln D3 (M, m1 , m2 )/2 ,

D3 (M, m1 , m2 ) = (1 y)M 2 + y (1 x)m21 + xm22 .
(B.10)
Explicitly:
m2
m21 ln M12
m2
m22 ln M22
+
,
(M 2 m21 )(m22 m21 ) (M 2 m22 )(m21 m22 )

m21
2m41
1
ln
V2 (M, m1 , m2 ) = 2 ln M 2 3 +
4
(M 2 m21 )(m22 m21 ) M 2
m22
2m42
2
ln
.
+
2
ln
(M 2 m22 )(m21 m22 ) M 2

V1 (M, m1 , m2 ) =
(B.11)
Defining the matrix element for the vertex to be

J PL FI + VR F F F
PR FI V .
M = ie VL FI FJ F
I J J
(B.12)
Diagram (a) gives

L;(a)
V FI FJ =

I
J
gV Fi Fj gL
gL
V2 (mp , mFi , mFj ),
4
(B.13)

I
J
gV Fi Fj gR
gR
V2 (mp , mFi , mFj ).
4
(B.14)
p,i,j
R;(a)
V FI FJ =
p,i,j
Diagram (b) gives

L;(b)
V FI FJ =

L
I
J
I
J
gV gL
gL
mp mn V1 gVR gL
gL
(1/2 + V2 ) ,
4
(B.15)

R
I
J
I
J
gV gR
gR
mp mn V1 gVL gR
gR
(1/2 + V2 ) ,
4
(B.16)
p,n,i
R;(b)
V FI FJ =
p,n,i
with the argument for functions V s in Eqs. (B.15), (B.16) being V1,2 (mFi , mp , mn ). The
I,J
I,J
explicit expression for gV Fi Fj , gVL,R
, gL and gR for each individual vertex diagrams
will be given below.
342
Table 6
(a)
V F F
I J
i F
j g
p F
VF F
i j
(b)
gI L,R gJ L,R V F F
I J
i
p n F
(a1)
+
0
i U
j g g DI ip g UJ jp (b1)
W D U p0 D
W Di Uj
W DI UJ p n Di
0L,R
0L,R
I J
gVL
gVR
L /s
Opn
R /s
Opn
gI L,R gJ L,R
D ip
U in
I
J
g0L,R
gL,R
(b2)
R /s O L /s g DI ip g UJ in
i Onp
W D U p+ n0 U
np
0L,R
L,R
I J
Table 7
(a)
V F F
I J
(b)
J
i F
j g g I
p F
V Fi Fj
L,R g L,R V FI FJ
i
p n F
(a1)
0
0
i D
j g g DI ip g DJ jp (b1)
ZD D p0 D
Z Di Dj
0L,R 0L,R
ZDI DJ p n Di
I J
gVL
L /sc
Onp
gVR
gI L,R gJ L,R
D ip
D in
R /sc g I
J
Onp
0L,R g0L,R
DJ in
DI ip
(a2)
+
+
R
L
i U
j g g DI ip g DJ jp (b2)
ZD D p+ U
Z Ui Uj
L,R
L,R
ZDI DJ p n Ui Opn /sc Opn /sc gL,R gL,R
I J
Table 8
(a)
V F F
I J
(a1)
ZU U
I J
i F
j
p F
(b1)
L /sc O R /sc g UI ip g UJ in

i Onp
p0 n0 U
np
0L,R
0L,R
i j
gI L,R gJ L,R V F F
I J
i U
j g
p0 U
ZU U
J
I
g0L,R
g0L,R
ZU U
I J
i j
(a2)
i D
j g
ZU U p+ D
Z Di Dj
I J
U ip
U ip
I
gL,R
U jp
U jp
J
gL,R
i
n F
(b)
gV F F
gVL
gVR
gI L,R gJ L,R
(b2)
L /sc O R /sc g UI ip g UJ in
i Onp
ZU U p+ n+ D
np
L,R
L,R
I J
B.3.1. Charged current vertex

The vertex correction to W DI UJ is presented in Table 6.
(a1)
The vertex correction due to gluino exchange is similar to W
DI UJ , with the substitution
of
S ,
p0 g,
D ip
DI i
I
g0L,R
gGL,R
,
U jp
U j
J
J
g0L,R
gGL,R
,
(B.17)
and multiplication of the whole expression by 4/3. This substitution rule also applies for
the neutral current vertex listed in the next section.
Similar to the case of the field strengths, the corresponding vertex corrections involving
leptons V LI J are obtained from by using the appropriate masses and couplings. The
gluino loop corrections must be omitted in that case.
B.3.2. Neutral current vertex
The ZDI DJ vertex is presented in Table 7.
The ZUI UJ vertex is presented in Table 8.
The radiative corrections to the lepton neutral current vertex are directly obtained from
the above expressions by replacing the up (s)quark with the (s)neutrino and the down
(s)quark with the (s)electron.
343
B.4. Anapole moment corrections

In the presence of parity-violating interactions, higher-order contributions can generate
the photonfermionfermion coupling of the form (see, e.g., Ref. [30]):
LPV
ff = e

FA,f (q 2 ) 2
f q q/ q 5 f A .
2
MZ
(B.18)
The contributions from gauginosfermion loop at q 2 0 are:

(a1) 2
(a2) 2
(b) 2
FA,f q 2 = FA,f
q + FA,f
q + FA,f
q ,
(a1) 2
FA,f
q = Qf

MZ2 f ip 2 f ip 2
x 3 dx
g0L g0R
,
48
(1 x)m2 0 + xm2
i,p
1
MZ2 f ip 2 f ip 2

(a2)
gL
FA,f q 2 = Qf
gR
48
i,p
1
0
2

f MZ
(b)
g f ip 2 g f ip 2
FA,f q 2 = 2I3
L
R
48
i,p
1
0
fi
x 3 dx
,
(1 x)m2 + + xm2
fi
x 2 (x 3) dx
,
(1 x)m2 + xm2 +
fi
(B.19)
f
where
stands for the isopartner of the fermion f (e.g., if f is the neutrino, then f
is the electron). Notice that for quark anapole moment, an additional gluino contribution
should be added, when parity is broken in the presence of a non-zero leftright mixing
in the squark sector and non-equal diagonal left and right squark masses. The gluino
(a1)
contribution can be obtained from FA,f
using the substitution rule give in Eq. (B.17). It
should be remembered that the anapole moment of an elementary particle is not a physical
observable [30]. Here, we separate it from the other one-loop contributions purely for
clarity of presentation.
B.5. Box graphs
Let us introduce the following notation:
L(f , f ) = f (p ) (1 5 )f (p),
R(f , f ) = f (p ) (1 + 5 )f (p),
(B.20)
where p and p are the momenta of the incoming particle f and outgoing f , respectively.
The four-point integration functions are defined as
1
B1 (M1 , M2 , m1 , m2 ) =
1
dx
1
dy
dz
0
z(1 z)
,
D4 (M1 , M2 , m1 , m2 )
344
B2 (M1 , M2 , m1 , m2 ) =
dx
0
1
dy
dz
0
z(1 z)
D42 (M1 , M2 , m1 , m2 )

D4 (M1 , M2 , m1 , m2 ) = z (1 x)M12 + xM22 + (1 z) ym21 + (1 y)m22 .
(B.21)
The explicit formulae for B1 and B2 are
B1 (M1 , M2 , m1 , m2 )
m21
M22
+
=
2(m21 M12 )(m21 m22 )(m21 M22 ) 2(m22
M2
M14 ln 12
M2
+
,
2
2
2
2(M1 m1 )(M1 m22 )(M12 M22 )
m41 ln
m22
M22
m21 )(m22 M12 )(m22
m42 ln
M22 )
B2 (M1 , M2 , m1 , m2 )
M22
m21
+
= 2
2
2
(m1 M1 )(m1 m22 )(m21 M22 ) (m22
M2
M12 ln 22
M1
+
.
(M12 m21 )(M12 m22 )(M12 M22 )
m21 ln
M22
m22
2
2
m1 )(m2 M12 )(m22
m22 ln
M22 )
(B.22)
B.5.1. Charged current box

We define the matrix element of the box diagram as
G CC
MCC
box = i B L(, ) L(u, d) +
2
(B.23)
to factor out the overall Fermi-constant. The dots denote other Dirac structures appearing
in the amplitude that make negligible contributions (suppressed by m ) to the total crosssection.
For the graphs in Fig. 10 we have: BCC = BCC;a + BCC;b + BCC;c + BCC;d ,
BCC;a =
2 s2
MW
ip in djp uj n
g0L
gL g0L gL mp0 mn+ B2 (m i , mD
j , mp0 , mn+ ),
4
p,n,i,j
BCC;b =
BCC;c =
BCC;d
2 s2
MW
2 s2
MW
ip in djp uj n
gL g0L gL g0L mp+ mn0 B2 (m

j , mp+ , mn0 ),
L i , mU
p,n,i,j
in ip djp uj n
gL g0L B1 (m i , mUj , mp+ , mn0 ),
g0L
gL
p,n,i,j
2 s2
MW
in ip djp uj n
=
gL g0L g0L gL B1 (m
j , mp0 , mn+ ).
L i , mD
4
p,n,i,j
(B.24)
345
Fig. 10. Feynman diagrams for one loop SUSY contribution to neutrino charged current box diagram.
Fig. 11. Feynman diagrams for one loop SUSY contribution to u neutrino neutral current box diagram.
Fig. 12. Feynman diagrams for one loop SUSY contribution to d neutrino neutral current box diagram.
Similarly, the box diagram contributing to the muon decay e e can be obtained
from BCC using the substitution:
BCC B ,
djp
d e,
ejp
gL,R,0L,0R gL,R,0L,0R ,
u e ,
uj n
j
j j ,
D
Lj , U
jn
e
gL,R,0L,0R gL,R,0L,0R
.
(B.25)
B.5.2. Neutral current box

We define the matrix element of the neutral current box diagram as

G L;q
R;q
MNC
box = i B L(, ) L(q, q) + B L(, ) R(q, q) .
2
The explicit expressions for the q box graphs are given below.
(B.26)
346

L;(u,a)
Up-quark box diagrams (Fig. 11): BL;u = B

BR;(u,b) ,
L;(u,a)
(NC) =
L;(u,b)
+ B
L;(u,c)
+ B
R;(u,a)
, BR;u = B
2 s2
MW
ip in ujp uj n
g0L
g0L g0L g0L mp0 mn0
4
p,n,i,j
B2 (m i , mUj , mp0 , mn0 ),
R;(u,a)
(NC) =
2 s2
MW
ip in ujp uj n
g0L
g0L g0R g0R B1 (m i , mUj , mp0 , mn0 ),
4
p,n,i,j
L;(u,b)
R;(u,b)
(NC) =
(NC) =
2 s2
MW
4
2 s2
MW
in ip ujp uj n
g0L
g0L g0L g0L B1 (m i , mUj , mp0 , mn0 ),
p,n,i,j
in ip ujp uj n
g0L
g0L g0R g0R mp0 mn0
p,n,i,j
B2 (m i , mUj , mp0 , mn0 ),
BL;(u,c)(NC) =
2 s2
MW
in ip ujp uj n
gL gL gL gL
4
p,n,i,j
B1 (m
j , mp+ , mn+ ).
L i , mD
(B.27)
Down-quark box diagrams (Fig. 12): BL;d = BL;(d,a) + BL;(d,b) + BL;(d,c) , BR;d =
R;(d,a)
R;(d,b)
B
+ B
,
BL;(d,a)(NC) =
2 s2
MW
ip in djp dj n
g0L
g0L g0L g0L mp0 mn0
4
p,n,i,j
B2 (m i , mD
j , mp0 , mn0 ),
R;(d,a)
(NC) =
2 s2
MW
ip in djp dj n
g0L
g0L g0R g0R B1 (m i , mD
j , mp0 , mn0 ),
4
p,n,i,j
L;(d,b)
(NC) =
BR;(d,b)(NC) =
2 s2
MW
4
2 s2
MW
in ip djp dj n
g0L
g0L g0L g0L B1 (m i , mD
j , mp0 , mn0 ),
p,n,i,j
in ip djp dj n
g0L
g0L g0R g0R mp0 mn0
p,n,i,j
B2 (m i , mD
j , mp0 , mn0 ),
L;(d,c)
(NC) =
2 s2
MW
ip in djp dj n
gL gL gL gL mp+ mn+
4
p,n,i,j
B2 (m
j , mp+ , mn+ ).
L i , mU
(B.28)
347
Appendix C. Radiative correction to neutrinonucleus interactions

FI
Given the expressions for the fermion field strength renormalization ZL,R
, vertex
L,R;q
L,R
CC
correction VL,R
in
FI FJ and box diagrams B , V B in Eq. (8), V in Eq. (9) and V B
Eq. (11) can be expressed as
L

1
L
V B = ZL + ZL + ZLe + ZLe 2 s W
(C.1)
e e + W + B ,
2
L

1
L
CC
VCCB = ZL + ZL + ZLu + ZLd 2 s W
(C.2)
+ W du + B ,
2
L
V = ZL 2scZ
(C.3)
,

1 q
L;q
L
(C.4)
= ZL scZqq
+ B ,
2
1 q
R;q=u,d
R;q
R
V B
(C.5)
= ZR scZqq
+ B .
2
The neutrino anapole moment contribution to the neutral current neutrinonucleus
scattering is shown in Fig. 1(b), which is absorbed into as

= 4c2 FA, q 2 .
(C.6)
L;q=u,d
V B
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
NuTeV Collaboration, G.P. Zeller, et al., Phys. Rev. Lett. 88 (2002) 091802.
Review of particle properties, Phys. Rev. D 66 (2002) 010001.
J. Erler, private communication.
S.C. Bennett, C.E. Wieman, Phys. Rev. Lett. 82 (1999) 2484;
V.A. Dzuba, V.V. Flambaum, J.S.M. Ginges, Phys. Rev. D 66 (2002) 076013;
A.I. Milstein, O.P. Sushkov, I.S. Terekhov, hep-ph/0212072.
SLAC Experiment E-158, E.W. Hughes, K. Kumar, P.A. Souder, spokespersons.
Jefferson Lab Experiment E-02-020, R. Carlini, J.M. Finn, S. Kowalski, S. Page, spokespersons.
S. Davidson, et al., JHEP 02 (2002) 037.
G.A. Miller, A.W. Thomas, hep-ex/0204007;
See also K.S. McFarland, hep-ex/021001.
S.A. Kulagin, hep-ph/0301045;
J.T. Londergan, A.W. Thomas, hep-ph/0301147;
P. Gambino, hep-ph/0211009.
A. Kurylov, M.J. Ramsey-Musolf, Phys. Rev. Lett. 88 (2002) 071804.
A. Kurylov, M.J. Ramsey-Musolf, S. Su, hep-ph/0205183, Phys. Rev. D, in press.
See, e.g., D.M. Pierce, hep-ph/9805497.
M.J. Ramsey-Musolf, Phys. Rev. D 62 (2000) 056009.
H.E. Haber, G.L. Kane, Phys. Rep. 117 (1985) 74.
M.E. Peskin, T. Takeuchi, Phys. Rev. Lett. 65 (1990) 964;
M.E. Peskin, T. Takeuchi, Phys. Rev. D 46 (1992) 381.
J. Erler, D.M. Pierce, Nucl. Phys. B 526 (1998) 53.
Muon g 2 Collaboration, G.W. Bennett, et al., Phys. Rev. Lett. 89 (2002) 101804.
A. Casas, in: G.L. Kane (Ed.), Perspectives on Supersymmetry, World Scientific, Singapore, 1998, p. 378.
E.A. Paschos, L. Wolfenstein, Phys. Rev. D 7 (1973) 91.
H. Schellman, private communication.
348
[21] G.P. Zeller, PhD thesis, Chapter 8 for the description of determining sin2 W .
[22] M. Goncharov, et al., Phys. Rev. D 64 (2001) 112006.
[23] I.S. Towner, J.C. Hardy, in: P. Herczeg, et al. (Eds.), Proceedings of the Fifth International WEIN
Symposium: Physics Beyond the Standart Model, World Scientific, Singapore, 1999, p. 338.
[24] G. Czapek, et al., Phys. Rev. Lett. 70 (1993) 17;
D.I. Britton, et al., Phys. Rev. Lett. 68 (1992) 3000.
[25] W.J. Marciano, Phys. Rev. D 60 (1999) 093006.
[26] J. Rosiek, Phys. Rev. D 41 (1990) 3464.
[27] D.M. Pierce, J.A. Bagger, K.T. Matchev, R. Zhang, Nucl. Phys. B 491 (1997) 3.
[28] D.M. Pierce, hep-ph/9805497.
[29] J.A. Grifols, J. Sola, Phys. Lett. B 137 (1984) 257;
J.A. Grifols, J. Sola, Nucl. Phys. B 253 (1985) 47;
P.H. Chankowski, S. Pokorski, J. Rosiek, Nucl. Phys. B 423 (1994) 437;
S. Martin, M. Vaughn, Phys. Lett. B 318 (1993) 331;
S. Martin, M. Vaughn, Phys. Rev. D 50 (1994) 2282;
D. Pierce, D. Papadopoulos, Nucl. Phys. B 430 (1994) 278;
N.V. Krasnikov, Phys. Lett. B 345 (1995) 25;
A. Dinini, Nucl. Phys. B 467 (1996) 3;
P.H. Chankowski, A. Dabelstein, W. Hollik, W.M. Mosle, S. Pokorski, J. Rosiek, Nucl. Phys. B 417 (1994)
101;
P.H. Chankowski, Z. Pluciennik, S. Pokorski, Nucl. Phys. B 439 (1995) 23;
A. Dabelstein, Z. Phys. C 67 (1995) 495;
M. Dress, K. Hagiwara, A. Yamada, Phys. Rev. D 45 (1992) 1725.
[30] M.J. Musolf, B.R. Holstein, Phys. Rev. D 43 (1991) 2956.

Top-pions and single top production at the HERA

and THERA colliders
Chongxing Yue a , Hongjie Zong b , Wei Wang a
a Department of Physics, Liaoning Normal University, Dalian 116029, China
b College of Physics and Information Engineering, Henan Normal University, Henan 453002, China
Abstract
The presence of physical top-pions in low-energy spectrum is an inevitable feature of topcolor
scenario. We consider the contributions of the physical top-pions predicted by topcolor-assisted
technicolor (TC2) models to the single top production via the t-channel process eq et at the
HERA and THERA colliders. We find that the neutral top-pion can generate large contributions to
the process ec et. In most of the parameter space, the production cross section is in the range
of 16 pb. The signals and observability of the neutral top-pion can be studied in the HERA and
THERA colliders.
PACS: 14.80.-j; 14.65.Ha; 12.60.Nz
1. Introduction
The top quark, with a mass of the order of the electroweak symmetry breaking (EWSB)
scale, is singled out to play a key role in probing the new physics beyond the standard
model (SM). The properties of the top quark could reveal information regarding flavor
physics, EWSB mechanism, as well as new physics beyond the SM [1]. In particular, the
anomalous top couplings, which affect top production and decay at high energy colliders
as well as precisely measured quantities with virtual top contributions, offer a unique place
for testing the SM flavor structure.
In the SM, the anomalous top quark couplings tqV (q = u- or c-quarks and V = Z,
or g gauge bosons), which are arised from the flavor changing (FC) interactions, vanish at
E-mail address: cxyue@lnnu.edu.cn (C. Yue).
doi:10.1016/S0550-3213(03)00553-4
350
C. Yue et al. / Nuclear Physics B 667 (2003) 349358
tree level but can be generated at the one-loop level. However, they are very suppressed
by the GIM mechanism, which cannot be detected in the present and near future high
energy experiments [2]. Hence, any signal indicating these types of couplings is evidence
of new physics beyond the SM and will shed more light on flavor physics in the top quark
sector. It has been suggested that the couplings of these types of couplings may be large in
some extensions to the SM, such as supersymmetry or other models with multiple Higgs
doublets [2,3] and dynamics EWSB models with new strong interactions of the top quark
[4].
Single top production is very sensitive to the anomalous top quark couplings tqV . These
couplings effectively appear in supersymmetry or in the scenario where new dynamics take
place in the fermion mass generation. Thus, studying the contributions of the anomalous
top quark couplings tqV to the single production of the top quark is of special interest. It
will be helpful to test the SM flavor structure and new physics beyond the SM. This fact
has lead to many studies involving single top production given by the tqV couplings in
lepton colliders [5,6] and hadron colliders [7].
The HERA and THERA colliders [8] with the center-of-mass energy s = 320 GeV
and 1000 GeV, respectively are the experimental facility where high energy electron
proton and positronproton interactions can be studied. It is well known that in the SM,
single top quark cannot be produced at an observable in these high energy colliders [9].
However, it has been shown that the HERA collider and THERA collider can provide a
very good sensitivity on the anomalous top quark couplings tqV via single top production
[10]. Single top production mediated by the FC interactions via the anomalous top quark
couplings tqV (q = u- or c-quarks and V = or Z gauge bosons) is a t-channel process
involving a massive final state top quark at the HERA and THERA collider experiments.
Several model-independent studies of this type of single top production have appeared
in the literature [11], which have shown that single top production is detectable, and the
HERA and THERA colliders are powerful tools for searching for the anomalous top quarks
couplings tq and tqZ.
To completely avoid the problem arising from the elementary Higgs field in the SM,
various kinds of dynamical EWSB model have been proposed, and among which the
topcolor scenario is attractive because it explains the large top quark mass and provides
possible EWSB mechanism. Topcolor-assisted technicolor (TC2) models [12], flavoruniversal TC2 models [13], top see-saw models [14], and the top flavor see-saw models
[15] are four of such examples. All of these models propose that the scale of the gauge
groups should be flavor non-universal. For example, SU(3) gauge group is flavor nonuniversal in TC2 models and U(1) gauge group is flavor non-universal in the flavoruniversal TC2 models. When one writes the non-universal interactions in the mass eigenbasis, it can induce the tree-level FC couplings. Thus the new particles, such as top-pions
(t , t0 ) predicted by TC2 models, have the tree-level FC couplings to the ordinary
particles and may generate large contributions to the single top production in ep collisions
at the HERA and THERA colliders.
In this paper, we study the contributions of the top-pions to the single production of the
top quarks in ep collisions at the HERA and THERA collider experiments in the context
of TC2 models. Our aim is to investigate whether the t-channel process eq et mediated
by the FC couplings via the anomalous top quars vertices tq and tqZ can be used to
351
detect the top-pions (t0 , t ) and further probe new physics beyond the SM. We find that
the contributions of the charged top-pions t to the single top production is very small.
The neutral top-pion t0 can generate significant contributions to the single top production
via the process ec et. In most of the parameter space, the production cross section is in
the range of 16 pb. The virtual effects of the neutral top-pion t0 on single top production
can be detected in the HERA and THERA collider experiments.
The paper is organized as follows: in Section 2 we give and discuss the anomalous top
quark couplings tqV (V = or Z) arised from the top-pions t , t0 . Their contributions
to the single top production process ec et are calculated in Section 3. The conclusions
and discussions are given in Section 4.
2. Top-pions and the anomalous top quark couplings tqV

In TC2 models [12], the TC interactions play a main role in breaking the electroweak
gauge symmetry. The ETC interactions give rise to the masses of the ordinary fermions
including a very small portion of the top quark mass, namely mt with a model-dependent
parameter 1. The main part of the top quark mass is dynamically generated by topcolor
interactions at a scale of 1 TeV, which also make small contributions to EWSB. This
means that the associates top-pions t,0 are not the longitudinal bosons W and Z, but
are separately physically observable objects. The presence of physics top-pions in the lowenergy spectrum is an inevitable feature of topcolor scenario that purports to avoid finetuning[16]. The flavor diagonal couplings of top-poins to fermions can be written as [4,12,
16]

2
2

mt (1 ) W Ft 5 0
i t tt + 2 tR bL t+ + 2 bL tR t ,
(1)
W
2 Ft
where W = / 2 = 174 GeV and Ft 50 GeV is the top-pion decay constant, which can
be estimated from the PagelsStokar formula.
For TC2 models, the underlying interactions, topcolor interactions, are non-universal
and therefore do not possess the GIM mechanism. This is an essential feature of these
kinds of models due to the need to single out top quark for condensate. The non-universal
gauge interactions result in the FC coupling vertices when one writes the interactions in
the mass eigen-basis. Thus, the top-pions have large Yukawa couplings to the third family
quarks and can induce the tree-level FC scalar couplings [17,18]

2 F2
W
t tc tt
mt
t c t t
0
ikU R kU L tL cR t0 + ikU
L kU R cL tR t
2Ft
W
t c bb
t c bb

+
+ 2 kU
(2)
R kDL cR bL t + 2kU R kDL bL cR t ,
where kU L(R) and kDL(R) are rotation matrices that diagonalize the up-quark and down+
+
dia
dia
quark mass matrices MU and MD , i.e., kU
L MU kU R = MU and kDL MD kDR = MD ,
+
for which the CKM matrix is defined as V = kU L kDL . To yield a realistic form of the
CKM V , it has been shown [17] that the top-pions mainly couple to the right-handed top
352
(tR ) or charm (cR ) quark and thus the values of the coupling parameters can be taken as:

tt
bb
tc
tc
2
kU
(3)
kU
kU
R kDL 1,
L 0,
R 2 .
tc
In the following calculation, we will assume kU R = 2 2 and take as a free
parameter, which is assumed to be in the range of 0.030.1 [4,12].
The relevant Feynman diagrams for the contributions of the neutral top-pion t0 to the
anomalous top quark couplings tc and tcZ via the tree-level FC couplings are shown
in Fig. 1. Using Eqs. (1)(3) and other relevant Feynman rules, we can give the effective
forms of the anomalous top quark coupling vertices Ztc and tc:

Z tc = ie F1Z + F2Z 5 + Pt F3Z + F4Z 5 + Pc F5Z + F6Z 5 ,
(4)

,
Fi = FiZ 2
.
=
(5)
tc
Z tc FiZ Fi
Vt = 3 ,at =0
The form factors FiV are expressed in terms of two- and three-point standard Feynman
integrals [19]. The expressions of FiV are given in Appendix A.
The neutral top-pion t0 can also generate the anomalous top quark couplings tu and
tuZ via the tree-level FC coupling t0 tu. However, it has been argued that the maximum
flavor mixing occur between the third generation fermions and the second generation
fermions, and the FC coupling t0 tu is very small which can be neglected [18]. Hence
we will ignore the anomalous top quark couplings tu and tuZ, and only calculate the
contributions of the neutral top-pions t0 to the production cross section of the t-channel
process ec et via the anomalous top quark couplings tc and tcZ in the following.
The charged top-pions t can contribute to the anomalous top quark couplings tc
and tcZ via the Feynman diagrams as depicted in Fig. 2. These diagrams are mediated
Fig. 1. Feynman diagrams for the contributions of the neutral top-pion t0 to the anomalous top quark couplings
tc and tcZ.
Fig. 2. Feynman diagrams for the contributions of the charged top-pions t to the anomalous top quark couplings
tc and tcZ.
353
by the FC couplings t bc. The internal fermion line must be bottom quark. Similar to
calculation about Fig. 1, we can give the effective vertices tc and Z tc arising from
Fig. 2. However, compared to tc and Z tc given by Fig. 1, these effective vertices are
very small and can be safely ignored [6]. So, in the following calculation, we will not
consider the contributions of the charged top-pions t to the single top production via the
process ec et at the HERA and THERA colliders.
3. The neutral top-pion t0 and single top production
From above discussion, we can see that the neutral top-pion t0 can induce large
anomalous top quark couplings tc and tcZ. Hence, it is possible that t0 generate
significant contributions to the single top production at the HERA and THERA colliders
via the t-channel process ec et with the relevant Feynman diagrams shown in Fig. 3.
Using the effective vertices Z tc and tc given by Eqs. (4) and (5), we can obtain the
cross section (s ) of the subprocess ec et which can be written as

1
1
d
2
2
=
|
+
|M
|
+
2
Re
M
M
(6)
|M
Z
dt
64 s (m2c s)
with
MZ =
1
m2Z

u t Z tc uc u e (ve ae 5 ) ue ,
1
M = u t tc uc u e ue .
t
Here
ve =

e
2
1 + 4SW
,
4SW CW
(7)
(8)
ae =
e
,
4SW CW
(9)
where s is the center-of-mass energy of the subprocess ec et in ep collisions. The

total cross section (s) of the single top production at the HERA and THERA colliders
can be obtained by folding the cross section with the charm-quark distribution function
fc (x) in the proton:
1
=
xmin
t+
fc (x) dx
d
dt
dt
(10)
Fig. 3. Feynman diagrams for the single top production at the HERA and THERA colliders, due to the anomalous
top quark couplings tcV (V = or Z).
354
Fig. 4. The production cross section (s) contributed by t0 as a function of mt for = 0.02 (solid line),
= 0.05 (dashed line), and = 0.08 (dotted line) at the HERA collider.
m2 +m2
(m2 s )(s m2 )
c
with s = xs, xmin = t s e , t = t s
and t = tcut = 0.001 (GeV)2 . The
parton distribution function fc (x) of the charm quark runs with the energy scale. In our
calculation, we take CTEQ5 parton distribution [20] for fc (x).
The cross sections (s) of the single top production at the HERA and THERA colliders
are plotted in Figs. 4 and 5 respectively, as a function of the mass mt of the neutral top
pion t0 for the parameter = 0.02 (solid line), 0.05 (dashed line) and 0.08 (dotted line).
1
2 = 0.2315, m = 175 GeV, m = 1.2 GeV,
In our calculation, we have taken e = 128.8
, SW
t
c
mZ = 91.18 GeV and Z = 2.49 GeV [21].
We can see from these two figures that the cross sections decrease with mt increasing
and decreasing. In all of the parameter space of TC2 models, the cross section
of single top production at the THERA collider is large than that of at the HERA
collider. For 0.02 0.08 and 200 mt 350 GeV, the cross sections are in
the ranges of 0.53.7 pb and 0.866.4 pb for the HERA collider and THERA collider,
respectively.
In order to estimatethe number of the events, we consider two ep collider scenarios: the
1
HERA collider with s = 320 GeV
and a yearly integrated luminosity of L = 160 pb
and the THERA collider with s = 1000 GeV and a yearly integrated luminosity of
L = 470 pb1 [8]. The yearly production events of the single top production at the HERA
and THERA colliders can be easily calculated. In most of the parameter space, there may
be several hundreds of the single top events at the HERA collider and hundreds and up
to thousands single top events at the THERA collider to be generated. For example, for
mt = 300 GeV and = 0.05, the HERA collider can generate 240 single top events and
the THERA collider can generate 1250 single top events. Thus, the contributions of the
355
Fig. 5. The production cross section (s) contributed by t0 as a function of mt for = 0.02 (solid line),
= 0.05 (dashed line), and = 0.08 (dotted line) at the THERA collider.
neutral top pion t0 to the single top production may be detected at the HERA collider or
the THERA collider. The signals of t0 can be studied at these colliders via the process
ec et.
4. Conclusions and discussions

The anomalous top quark couplings tqV (q = u- or c-quarks and V = Z, or gauge
bosons), which are arised from the FC interactions, offer an ideal place to search for
new physics beyond the SM as they are very small in the SM. Single top production is
very sensitive to the tqV couplings in the HERA and THERA colliders. Studying the
contributions of the tqV couplings to the single production of the top quark will be helpful
to test the SM flavor structure and new physics beyond the SM. In this paper, we study the
contributions of the top-pions predicted by TC2 models to the single top production in ep
collisions at the HERA and THERA colliders. Our results show.
(1) Although the contributions of the up-quark distribution in the proton to the single
top production is large than those of the charm-quark distribution, the contributions of the
anomalous top quark couplings tuV given by the TC2 models to the single top production
in ep collisions are much smaller than those of the tcV couplings and can be safely
neglected.
(2) The charged top-pions t can generate the anomalous top quark couplings tcV
via the FC couplings t bc. However, compared with the tcV couplings generated by the
356
neutral top-pion t0 via the FC coupling t0 tc, their values are negligibly small. The t
cannot give significantly contributions to the production cross section of the single top
quark at the HERA and THERA colliders.
(3) The contributions of the neutral top-pion t0 to the production cross section
of the t-channel process ec et increases with s increasing and mt decreasing. For

200 mt 400 GeV and = 0.05, the value of varies from 0.99 to 2.35 pb and from
1.7 to 4.1 pb at the HERA and THERA colliders, respectively. Thus, there may be several
hundreds and up to thousands of the single top production events to be generated at these
two colliders. We can study the signature and observability of t0 via the process ec et
at the HERA collider or the THERA collider.
TC2 models also predict the existence of the neutral cp-even state, called top-Higgs
boson h0t , which is a t t bound and analogous to the particle in low energy QCD. Its
mass can be estimated in the NambuJona-Lasinio model in the large Nc approximation
and is found to be of the order mht mt [18]. The main difference between t0 and h0t
is that h0t can couple to gauge boson pairs W W and ZZ at tree level, which is similar to
that of the SM Higgs boson H 0 . Thus, the contributions of the top-Higgs h0t to the single
top production at the HERA and THERA colliders are similar to those of t0 . We find that,
in most of the parameter space of TC2 models, the cross section of single top production
given by the top-Higgs boson h0t varies in the range of 0.86.2 pb. Thus the top-Higgs
boson h0t can also be detected at the HERA collider or the THERA collider via the process
ec et.
The key feature of TC2 models is that a large part of the top quark mass is dynamically
generated by topcolor interactions at a scale of order 1 TeV, which is flavor non-universal.
To ensure that the top quark condenses and receives a large mass while the bottom quark
does not, the topcolor gauge group is usually taken to be a strongly coupled SU(3) U (1).
The U (1) provides the difference that cause only top quark to condensate. Thus, TC2
models predict the existence of topcolor gauge bosons BA and an extra U (1) gauge boson
Z . Tree-level FCNC for the topcolor gauge bosons BA and Z are generated when quarks
fields are rotated to the mass eigenstate basis. The couplings of Z to ordinary fermions
are non-universal and stronger for the third generation fermions, yielding potentially large
top-charm couplings. Thus, the topcolor gauge bosons can generate large anomalous top
couplings tcV (V = or Z), which may produce significant contributions to the single
top production at the HERA and THERA collider via the process ec et. However, the
limits on the masses of the topcolor gauge bosons BA and Z can be obtained via studying
their effects on various experimental observables [4]. For example, Ref. [22] considered the
bound placed by the electroweak measurement data on the gauge boson Z . They find that
Z predicted by the TC2 models and the flavor universal TC2 models must be heavier than
1 TeV. If we assume MZ = MB 1 TeV, then we find that the production cross section
is smaller than 1 102 pb in most of the parameter space of the TC2 models. Thus, it
is very difficult to detect the topcolor gauge bosons via the single top production at the
HERA and THERA colliders.
357
Acknowledgements
We thank C.-P. Yuan for pointing out that we should use the evolved parton distribution
function of the charm quark to calculate the production cross section. This work was
supported in part by the National Natural Science Foundation of China (90203005).
Appendix A. The form factors in the effective vertices Z tc and tc for the neutral
top-pion t0

F1Z = g B0 + m2t C0 2C24 + m2t (C11 C12 ) B0 B1 vt ,

F2Z = g B0 + m2t C0 2C24 m2t (C11 C12 ) + B0 + B1 4C24 at ,
F3Z = 2mt g(C21 + C22 2C23 )vt ,
F4Z = 2mt g(C21 C22 + 2C23 2C22 + C12 + C0 )at ,
F5Z = 2mt g(C22 C23 + C12 )vt ,
F6Z = 2mt g(C22 + C23 + C12 2C22 + 3C12 + 2C23 2C11 C0 )at .
Here

mt
1
g=
2
16
2 Ft

1
vt =
1
4SW CW
w2 Ft2 2
w

8 2
SW ,
3
KUt cR KUt t L ,
at =
1
.
4SW CW
The expressions of two- and three-point scalar integrals Bn and Cij in this paper are [19]:
Bn = Bn ( t, mt , mt ),
Bn = Bn (pc , mt , mt ),
Bn = Bn (pt , mt , mt ),
Cij = Cij (pt , t, mt , mt , mt ),
C0 = C0 (pt , t, mt , mt , mt ),
Cij = Cij (pc , t, mt , mt , mt ),
C0 = C0 (pc , t, mt , mt , mt ).
References
[1] For reviews see: M. Beneke, et al., A. Ahmadov, et al., Top quark physics, hep-ph/0003033;
E.H. Simmons, Top physics, hep-ph/0011244.
[2] G. Eilam, J.L. Hewett, A. Soni, Phys. Rev. D 44 (1991) 1473.
358
[3] B. Grzadkowski, J.F. Gunion, P. Krawczyk, Phys. Lett. B 268 (1991) 106;
M. Luke, M.J. Savage, Phys. Lett. B 307 (1993) 387;
L.J. Hall, S. Weinberg, Phys. Rev. D 48 (1993) 979;
G. Couture, C. Hamzaoui, H. Knig, Phys. Rev. D 52 (1995) 1713;
D. Atwood, L. Reina, A. Soni, Phys. Rev. D 53 (1996) 1199;
J.L. Lopez, D.V. Nanopoulos, R. Rangarajan, Phys. Rev. D 56 (1997) 3100.
[4] K. Lane, Technicolor 2000, in: Nuclear, Subnuclear and Astroparticle Physics, Frascati, 2000, pp. 235280;
C.T. Hill, E.H. Simmons, Phys. Rep. 381 (2003) 235.
[5] K.J. Abraham, K. Whisnant, B.-L. Young, Phys. Lett. B 419 (1998) 381;
V.F. Obraztsov, S.R. Slabospitsky, O.P. Yushchenko, Phys. Lett. B 426 (1998) 393;
B.A. Arbruzov, M.Yu. Osipov, hep-ph/9802392;
T. Han, J.L. Hewett, Phys. Rev. D 60 (1999) 074051;
S. Bar-Shalom, J. Wudka, Phys. Rev. D 60 (1999) 094916;
J.A. Aguilar-Saavedra, Phys. Lett. B 502 (2001) 115;
J.A. Aguilar-Saavedra, T. Riemann, hep-ph/0102197.
[6] C. Yue, et al., Phys. Lett. B 496 (2000) 93;
C. Yue, et al., Phys. Lett. B 525 (2002) 301.
[7] E. Malkawi, T. Tait, Phys. Rev. D 54 (1996) 5758;
T. Han, et al., Phys. Lett. B 385 (1996) 311;
T. Tait, C.-P. Yuan, Phys. Lett. B 55 (1997) 7300;
M. Hosch, K. Whisnant, B.-L. Young, Phys. Rev. D 56 (1997) 5725;
T. Han, et al., Phys. Rev. D 58 (1998) 073008;
T. Tait, C.-P. Yuan, Phys. Rev. D 63 (2000) 014018;
F. del Aguila, J.A. Aguilar-Saavedra, Phys. Lett. B 462 (1999) 310;
F. del Aguila, J.A. Aguilar-Saavedra, Nucl. Phys. B 576 (2000) 56;
J. Cao, et al., hep-ph/0212114.
[8] A.K. Cifici, S. SuHansoy, . Yavas, in: Proc. of EPAC2000, 2000, p. 388;
P.J. Bussey, Int. J. Mod. Phys. A 17 (2002) 1065.
[9] U. Baur, J.J. van der Bij, Nucl. Phys. B 304 (1988) 451.
[10] A. Belyaev, N. Kidonakis, Phys. Rev. D 65 (2002) 037501.
[11] H. Fritzsch, D. Holtmannsptter, Phys. Lett. B 457 (1999) 186;
O. Cakir, S. Sultansoy, M. Yilmaz, hep-ph/0105130;
A.T. Alan, A. Senol, Europhys. Lett. 57 (2002) 669.
[12] C.T. Hill, Phys. Lett. B 345 (1995) 483;
K. Lane, T. Eichten, Phys. Lett. B 352 (1995) 382;
K. Lane, Phys. Lett. B 433 (1998) 96;
G. Cvetic, Rev. Mod. Phys. 71 (1999) 513.
[13] M.B. Popovic, E.H. Simmons, Phys. Rev. D 58 (1998) 095007.
[14] B.A. Dobrescu, C.T. Hill, Phys. Rev. Lett. 81 (1998) 2634;
R.S. Chivukula, B.A. Dobrescu, H. Georgi, C.T. Hill, Phys. Rev. D 59 (1999) 075003;
H.-J. He, C.T. Hill, Phys. Rev. D 65 (2002) 055006.
[15] H.J. He, T.M.P. Tait, C.P. Yuan, Phys. Rev. D 62 (2000) 011702.
[16] G. Burdman, D. Kominis, Phys. Lett. B 403 (1997) 101.
[17] H.-J. He, C.-P. Yuan, Phys. Rev. Lett. 83 (1999) 28.
[18] G. Burdman, Phys. Rev. Lett. 83 (1999) 2888.
[19] G. Passarino, M. Veltman, Nucl. Phys. B 160 (1979) 151;
A. Axelrod, Nucl. Phys. B 209 (1982) 349;
M. Clements, et al., Phys. Rev. D 27 (1983) 570.
[20] CTEQ Collaboration, H.L. Lai, et al., Eur. Phys. J. C 12 (2000) 375;
J. Pumplin, D.R. Stump, J. Huston, H.L. Lai, P. Nadolsky, W.K. Tung, JHEP 0207 (2002) 012, hepph/0201195.
[21] Particle Data Group, D.E. Groom, et al., Eur. Phys. J. C 15 (2000) 1.
[22] R.S. Chivukula, E.H. Simmons, Phys. Rev. D 66 (2002) 015006.

Towards an effective-action approach to

fermion-loop corrections
W. Beenakker a , A.P. Chapovsky b , A. Kanaki c ,
C.G. Papadopoulos c,d , R. Pittau e,1
a Theoretical Physics, Univ. of Nijmegen, NL-6500 GL Nijmegen, The Netherlands
b Institut fr Theoretische Physik E, RWTH Aachen, D-52056 Aachen, Germany
c Institute of Nuclear Physics, NCSR Demokritos, 15310 Athens, Greece
d Theory Division, CERN, CH-1211 Geneva, Switzerland
e Dipartimento di Fisica Teorica, Universit di Torino, and INFN, Sezione di Torino, Italy
Abstract
We present a study of the effective action approach to incorporate higher-order effects in
e+ e n fermions. In its minimal version, the effective action approach is found to exhibit
problems with unitarity and high-energy behaviour. We identify the origin of these problems by
investigating the zero-mode solutions of the Ward identities. A numerical analysis of the importance
of the zero-mode solutions is presented for four-fermion production processes.
PACS: 12.15.Lk; 11.10.Gh; 11.10.Lm; 13.10.+q; 14.70.-e
1. Introduction
Multi-fermion production processes constitute one of the most important classes
of reactions at electronpositron colliders [1]. Through high-precision studies of these
reactions valuable information is gained on the electroweak parameters, on the interactions
between the electroweak gauge bosons and on the mechanism of electroweak symmetry
breaking. High-precision studies of this kind demand a precise description of the physics
Work supported by the European Union under contract HPRN-CT-2000-00149.

E-mail address: costas.papadopoulos@cern.ch (C.G. Papadopoulos).
1 Financially supported by MIUR under contract 2001023713-006.
doi:10.1016/S0550-3213(03)00545-5
360
W. Beenakker et al. / Nuclear Physics B 667 (2003) 359393
of the unstable gauge bosons that occur during the intermediate stages of the reactions. One
problematic, though crucial ingredient for achieving such a description is the incorporation
of the associated finite-width effects. To this end one has to resum the relevant gaugeboson self-energies, which results in a mixing of different orders of perturbation theory
and thereby jeopardizes gauge invariance. Since the high precision of the experiments has
to be matched by the precision of the theoretical predictions, both an adequate treatment of
the finite-width effects and a sufficiently accurate perturbative expansion are required. The
clash between resummation and perturbative expansion can therefore not be ignored.
A procedure to overcome this dilemma has been proposed several years ago and is
known under the name of fermion loop (FL) scheme [25]. In this scheme a resummation
of all one-loop fermionic corrections to gauge-boson self-energies is performed. In order
to account for a consistent and gauge-invariant treatment, the one-particle-irreducible (1PI)
fermionic one-loop corrections to the other n-point gauge-boson functions (with n 3) are
f
included as well. The FL scheme essentially involves the closed subset of all O([NC /]n )
f
contributions to a given physical process, with NC denoting the colour degeneracy of
fermion f , and as such it is manifestly consistent. The reason for singling out the fermionic
one-loop corrections lies in the fact that the unstable gauge bosons decay exclusively into
fermions at lowest order. The FL scheme has proven particularly successful in dealing with
four-fermion production processes. Although in the beginning it merely served the purpose
of a consistent scheme for including the width of the W boson [2], which is closely related
to the imaginary part of the W -boson self-energy, very soon people realized that it can also
accommodate the resummation of the real parts of the gauge-boson self-energies [3,4],
which are responsible for the running of the couplings with energy.
Unfortunately there are several limitations related to the FL scheme. First of all, it
is clearly a partial answer to the problem of resumming higher-order corrections. It is
restricted to closed fermion loops, which means that bosonic contributions are ignored.
Several methods have been proposed to overcome this limitation. The most efficient one
is the so-called pole-scheme [6], which amounts to a systematic expansion of the matrix
elements around the complex poles in the unstable-particle propagators. In leading order
of this expansion the radiative corrections involve the full set of one-loop corrections to
on-shell gauge-boson production and decay (factorizable corrections) [7,8], as well as
soft-photonic corrections that take into account the fact that the production and decay
stages of the reaction do not proceed independently (non-factorizable corrections) [9].
However, in reactions with several intermediate unstable gauge bosons, like, e.g., sixfermion production, it becomes rather awkward to perform the complete pole-scheme
expansion [7]. Secondly, even though the FL scheme is conceptually straightforward, it
becomes more and more involved computationally once one goes beyond the four-fermion
production processes. For instance, for general multi-fermion production processes one
has to consider the complete set of fermionic one-loop corrections to the 1PI four-point
gauge-boson functions, five-point gauge-boson functions, and so on.
In the meantime a novel proposal has emerged, as described in the paper by Beenakker,
Berends and Chapovsky [10], abbreviated as BBC from now on. Their proposal consists
essentially in a rearrangement of the expansion of the effective action of the theory, which
is usually performed in terms of the 1PI Feynman amplitudes, in such a way that the new
expansion is manifestly gauge invariant. Restricting ourselves, for simplicity, to a pure
361
SU(N) gauge theory, the expansion looks like

SNL = d 4 x d 4 y G2 (x, y) Tr U (y, x)F (x)U (x, y)F (y)

+ d 4 x d 4 y d 4 z G3 (x, y, z)

Tr U (z, x)F (x)U (x, y)F (y)U (y, z)F (z)
+ .
(1)
Here the trace has to be taken in group space and F gi [D , D ] is the SU(N) fieldstrength tensor, expressed in terms of the covariant derivative D and the gauge coupling g.
The operator U (x, y) is a path-ordered exponential, which carries the gauge transformation
from one spacetime point to the other (see Section 2 for a more detailed definition). In
Eq. (1) each gauge-invariant non-local operator is multiplied by an appropriate spacetime
function Gi that can, in principle, be computed within perturbation theory. In the context of
fermionic loop effects, the various terms in Eq. (1) can be viewed as the result of integrating
out the fermions in the functional integral, resulting in a kind of non-local Lagrangian for
gauge-boson interactions. The minimum number of gauge bosons that participate in the
effective interaction is two for the first term of Eq. (1), three for the second term, and
so on. Note, however, that each term will also generate all higher n-point interactions,
through the expansion of the path-ordered exponentials (see Section 2). These higher npoint interactions are essential for achieving gauge invariance for the individual terms
of Eq. (1). Since all ingredients for the resummation of the gauge-boson self-energies
are contained in the first (self-energy-like) term of Eq. (1), it was proposed in the BBC
approach to truncate the series at this first term. In this way an economic gauge-invariant
framework for resumming self-energies is obtained, leading to matrix elements that satisfy
all relevant Ward identities. Two questions remain open at this point: How should one
match the spacetime function G2 with the actual fermion-loop corrections? and Is
gauge invariance sufficient for obtaining well-behaved matrix elements?.
In this paper we undertake the effort to confront the BBC idea with actual calculations,
addressing in this way the two outstanding questions. In Section 2 we consider the
matching aspect. We introduce the set of gauge-invariant operators that is relevant for
an exact description of fermion-loop corrections in the two-point gauge-boson sector of
the Standard Model (SM), involving both electroweak gauge bosons and Higgs fields.
In Section 3 we identify and analyse a problem with the high-energy behaviour of the
matrix element for the reaction e+ e W + W . This problem is related to the nonunitary character of the truncation in the BBC approach. We pin-point the source of the
problem to be in the zero-mode solutions of the Ward identities, like the second term of
Eq. (1), which are absent in the BBC approach. In Section 4 the set-up of the calculations
as well as the numerical results are presented and discussed. Particular emphasis is put on
an investigation of the numerical importance of the zero-mode solutions. Finally, the paper
is concluded with Appendices AD, where all relevant information pertaining to the nonlocal Feynman rules, renormalization schemes and the unitarity problem in the reaction
e+ e W + W can be found.
362
2. The effective-action approach

2.1. Notation and conventions
Before turning our attention to the non-local Lagrangian, we first introduce the notation
and conventions that will be used throughout the remainder of this paper. In the SM there
are four gauge fields, the SU(2)L (isospin) gauge fields Wa (a = 1, 2, 3) and the U (1)Y
(hypercharge) gauge field B . The corresponding field-strength tensors are given by
F = W W ig2 [W , W ],
B = B B ,
(2)
using the shorthand notations

a
,
F T a F
W T a Wa .
(3)
The SU(2)L and U (1)Y gauge couplings are indicated by g2 and g1 , respectively, and the
SU(2)L generators T a can be expressed in terms of the standard Pauli spin matrices a
(a = 1, 2, 3) according to T a = a /2. These generators obey the commutation relation
[T a , T b ] = i# abc T c , with the SU(2) structure constant # abc given by
+1 if (a, b, c) = even permutation of (1, 2, 3),

abc
# = 1 if (a, b, c) = odd permutation of (1, 2, 3),
(4)
0
else.
The physically observable gauge-boson states are given by

1
W = W1 iW2 ,
2
A = cW B sW W3 ,
Z = cW W3 + sW B ,
(5)

= g2 / g12 + g22 and sW =
for the W bosons, Z boson and photon, respectively. Here cW

2 are the cosine and sine of the weak mixing angle. The electromagnetic coupling
1 cW

constant can be obtained from g1 and g2 according to e = 4 = g1 g2 / g12 + g22 .

Since we want to discuss the entire gauge-boson sector, we also need to introduce the
would-be Goldstone bosons and that are intimately linked to the longitudinal degrees
of freedom of the massive W and Z gauge bosons. To this end we introduce the (Y = 1)
Higgs doublet

+ (x)

(x) =
(6)
v + H (x) + i(x) / 2
and the corresponding covariant derivative
D = ig2 W + i
g1
B .
2
(7)
Here v/ 2 is the non-zero vacuum expectation value of the Higgs field, yielding MW =
vg2 /2 and MZ = MW /cW for the masses of the W and Z bosons in this convention.
363
A few more definitions are needed for the description of the fermionic corrections to
the various self-energies in the gauge-boson sector of the SM. A generic SM fermion
will be indicated by f and its isospin partner by f . The SU(3)C colour factor, mass,
f
electromagnetic charge and isospin of the fermion f are denoted by NC , mf , eQf and
3
If , respectively. Finally, the hypercharge of the left-handed and right-handed fermions is
denoted by YfL and YfR , respectively.
2.2. The non-local Lagrangian
Following Ref. [10] we introduce an effective action that includes all relevant two-point
interactions in the gauge-boson sector, involving both gauge-boson and Higgs fields. This
non-local Lagrangian can be written as

1
SNL =
d 4 x d 4 y 1 (x y)B (x)B (y)
4

1
d 4 x d 4 y 2 (x y) Tr U2 (y, x)F (x)U2 (x, y)F (y)

2

2 g1
2
d 4 x d 4 y 3 (x y) (x)F (x)(x) B (y)
v g2

4
4 d 4 x d 4 y 4 (x y) (x)F (x)(x) (y)F (y)(y)
v

+ d 4 x d 4 y 5 (x y) D (x) U2 (x, y)U1(x, y)D (y)

2
+ 2 d 4 x d 4 y 6 (x y) (x)D (x) (y)D (y) .
(8)
v
A few comments and definitions are in order here. First of all, the arguments of the nonlocal coefficients 1 (x y), . . . , 6 (x y) follow directly from translational invariance.
Furthermore, the trace appearing in the 2nd term has to be taken in SU(2)L group space.
Finally, the path-ordered exponentials for the SU(2)L and U (1)Y gauge groups are defined
according to

y
U2 (x, y) = P exp ig2

W () d ,
x
Y
U1 (x, y) = P exp +ig1
2
y

B () d ,
(9)
where Y = 1 for the Higgs doublet and d is the element of integration along some path
(x, y) that connects the points x and y. According to Ref. [10] the path is defined in
such a way that it does not involve closed loops, i.e., the null path (x, x) always has
zero length. Moreover, the choice of path should be such that it gives rise to path-ordered
exponentials with specific properties under differentiation.
Let us repeat the main points of the BBC approach.
364
The BBC effective action in Eq. (8) is gauge invariant by construction.

Through the expansion of the path-ordered exponentials, the effective action incorporates a set of higher 3- ,4- , . . . , n-point functions that automatically satisfy the Ward
Identities of the theory. A complete set of three-point Feynman rules based on Eq. (8)
is given in Appendix A.
A set of unknown coefficients 1 , . . . , 6 is introduced.
There are several ways to determine the unknown coefficients. Some simplified
expressions, corresponding to existing ad hoc approximations for incorporating finitewidth effects, have already been presented in Ref. [10]. These expressions involve only
a partial resummation of the fermionic corrections, in contrast to the full 1PI resummation
that is performed in the FL scheme. In this paper we investigate how the unknown BBC
coefficients can be matched with the well-established two-point fermion-loop contributions
in the SM. By doing so, we obtain an exact correspondence between the SM and the
effective BBC action for all reactions that involve at most two-point interactions among the
gauge bosons. For reactions that involve interactions among three gauge bosons or more,
the effective BBC approach provides us with a minimal set of contributions that is required
for satisfying all relevant Ward identities. Although this approximation cannot be identical
to that of the FL scheme, one might anticipate that it provides a much more economic
approach to multi-fermion production processes. After all, in the FL scheme one has to
perform a complete calculation of the SM n-point functions with three or more external
gauge bosons, which constitutes a rather intensive and costly procedure. On the other
hand, by truncating the non-local action at two-point order several parts of the higherorder corrections are neglected. It is therefore important to understand to what extent one
can trust such an approximation.
2.3. The matching procedure
In order to set up the framework of our studies, we present in this subsection the
matching procedure, i.e., the determination of the non-local coefficients 1 , . . . , 6 . Using
the knowledge of all two-point functions in the FL scheme (see Appendix B), we can
perform the first level of matching: mapping the unrenormalized self-energies directly onto
the non-local coefficients. The second level of matching, between the so-obtained non-local
matrix elements/cross sections and the explicit experimental observables, should take care
of any necessary redefinition (renormalization) of couplings and masses.
As can be seen from Appendix B, we indeed need all six non-local operators in
Eq. (8) in order to match the six independent gauge-boson self-energies (after tadpole
renormalization). The 1st, 2nd and 5th operators in Eq. (8) are non-local extensions of
terms in the local SM Lagrangian. They take care of all UV-divergent terms present in the
fermionic one-loop self-energies. The remaining three operators are higher-dimensional
(dim > 4). The corresponding coefficients are finite, as expected for a renormalizable
theory, and can be viewed as non-local versions of the oblique S-, T - and U -parameters of
Peskin and Takeuchi [11]. These operators are required for achieving an explicit breaking
of the global isospin symmetry among the SU(2) gauge bosons, usually referred to as
custodial SU(2) symmetry [12]. After all, also the loop effects in the SM explicitly break
365
this global symmetry as a result of hypercharge interactions and specific fermion-mass

effects.
1 , . . . ,
6 ,
Below we list the results for the first level of matching of the coefficients
which represent the Fourier transforms of the non-local coefficients 1 , . . . , 6 appearing
in Eq. (8). These results will be expressed in terms of the transverse and longitudinal
self-energy functions TV1 V2 (s) and LV1 V2 (s), where V1,2 = , Z, W and s represents the
square of the momentum at which the self-energies are evaluated. The explicit expressions
for these functions can be found in Appendix B in Eqs. (B.1)(B.5).
1 can be obtained through the relation
The transverse, pure hypercharge coefficient

ZZ
1 2
Z
2
T (s) LZZ (s)
cW T (s) + 2sW cW T (s) + sW
s

L 2
R 2
Yf
Yf
1 f
=
NC
+
f (s),
2
2cW
2cW
1 (s) =
(10)
where the vacuum-polarization function f (s) is defined in Eq. (B.2). Since f (s) is
1 (s). Note also that this first coefficient is
UV-divergent, the same must be true for
proportional to g12 .
3 reads
The mixed hypercharge-isospin coefficient

Z
ZZ
1 2
cW 2
2
2
s cW
T (s) cW
T (s) LZZ (s)
c (s) +
s W T
sW W
3
L
Yf
1 cW f If
=
NC
f (s),
2 sW
sW
2cW
3 (s) =
(11)

f 3 L
which is finite because of the quantum-number identity
f NC If Yf = 0. With our
definition, involving the extra factor g1 /g2 in Eq. (8), this coefficient is proportional to
g22 .
2 and
4 are given by
The remaining two transverse, pure isospin coefficients
f
If3 2

4 (s)
2 (s) = 1 TW W (s) LW W (s) = 1
NC
f (s)
s
2
sW
(12)
and
2
ZZ

Z
2
4 (s) = 1 sW
T (s) LZZ (s)
T (s) 2sW cW T (s) + cW
s

TW W (s) LW W (s)

2m2f

f
B0 (s, mf , mf ) B0 (s, mf , mf )
=
1
+
N
C
2
s
24sW
f

4m2f (m2f m2f )

B0 (s, mf , mf ) B0 (0, mf , mf ) ,
s2
(13)
366
where the scalar two-point functions B0 are defined in the usual way [14]. The first
coefficient is clearly UV-divergent and the second one clearly finite. Both coefficients are
again proportional to g22 .
3 and
4 vanish at high energies (s m2 ) and in the absence
The coefficients
f
of doublet splitting (mf = mf ). The former reflects the fact that there are only two
independent self-energies in the unbroken SM, whereas the latter indicates that there is no
fermion-mass-induced custodial SU(2) breaking if the fermions within an SU(2) doublet
4 (s) = 0, which implies that we can
3 (s) = s
have the same mass. For s = 0 we obtain s
match the transverse gauge-boson sector without the explicit need for finite shifts of the
gauge-boson masses. Such finite shifts will occur only in the longitudinal/scalar sector, as
they should.
In the longitudinal/scalar sector we have two coefficients to match:

LW W (s)
1 f 2
6 (s),
5 (s) =
=
NC mf B0 (s, mf , mf )
(14)

2
8 2 v 2
MW
no tadpole
f
6 (s) =
LW W (s)
2
MW
LZZ (s)
MZ2

1 f 2
N
m
C f B0 (s, mf , mf ) B0 (s, mf , mf )
8 2 v 2
f
m2f
m2f
s
B0 (s, mf , mf ) B0 (0, mf , mf )

,
(15)
where again the first coefficient is clearly UV-divergent and the second one clearly finite.
As was to be expected, both coefficients are proportional to 1/v 2 . The finite shifts of the
gauge-boson masses at s = 0 have been absorbed into the non-local T -parameter (or parameter)
6 (0) =
LW W (0)
2
MW
LZZ (0)
MZ2
2

f
m2f
mf
1
2
=
NC mf 1 2
log
.
2
2
2
16 v
mf mf
m2f
f
(16)
4 , also
6 vanishes at high energies (s m2 ) and in the absence of doublet
3 and
Like
f
splitting (mf = mf ).
The explicit expressions for the resummed gauge-boson propagators in the covariant R
gauge can be found in Appendix A for both the transverse and longitudinal/scalar sectors.
2.4. Running couplings
In the sequel of this section we show explicitly how the introduction of running
couplings leads to an effective description, where in analogy to the FL scheme one just
has to replace bare with running couplings in tree-order matrix elements in order to
properly take into account the resummed fermionic corrections. As we have seen from
367
the explicit expressions for the various non-local coefficients, all six non-local coefficients
are proportional to just one type of bare coupling. In order to make the discussion of the
running couplings easier, we therefore extract these couplings from the coefficients:
1 (s) = g12
S1 (s),
2 (s) = g22
S2 (s),
1
5 (s) =
S5 (s)d,
v2
3 (s) = g22
S3 (s),
1
6 (s) =
S6 (s),
v2
4 (s) = g22
S4 (s),
(17)
with similar relations in coordinate space between i (x y) and Si (x y) for i = 1, . . . , 6.

Upon closer investigation of Eq. (8) we notice that SNL contains exclusively the
combinations /v, g1 B or g2 W . This means that the couplings can be absorbed into the
field-definition. For the corresponding local SM action we have

1
1
SL = 2 d 4 x g1 B (x)g1 B (x) 2 d 4 x Tr g2 F (x)g2 F (x)
4g1
2g2

1
1
D (x).
+ v 2 d 4 x D (x)
(18)
v
v
The UV-divergences contained in the non-local coefficients Si (x y) have the simple form
SiUV (4)(x y) for i = 1, 2, 5. Combining this with the local SM action, we end up with the
minimal renormalization requirement that 1/g12 + S1UV , 1/g22 + S2UV and v 2 + S5UV should
become finite.
Having this in mind, we perform the re-diagonalization procedure in the transverse
sector by introducing the running couplings

1
1
sW Z
1
s
+
(s)
+
(s)
= 2 +
S1 (s) +
S3 (s),
T
cW T
g12 (s) sg12
g1

1
cW Z
1
1
(s)
(s)
= 2 +
S2 (s) +
S3 (s) +
S4 (s),
s
+
T
T
2
2
sW
g2 (s) sg2
g2
v 2 (s) v 2 +
S5 (s),
(19)
the finiteness of which is consistent with the minimal renormalization requirement given
above. From this a few more (finite) running quantities can be derived:
1
e2 (s)
2
(s)
sW
1
1
1
1
2
+ 2
= 2 +
S1 (s) +
S2 (s) + 2
S3 (s) +
S4 (s),
4(s) g1 (s) g2 (s) e
e2 (s)
,
g22 (s)
2
2
cW
(s) 1 sW
(s) =
e2 (s)
,
g12 (s)
1
2
2
(s) cW
(s)MZ2 (s) v 2 (s)g22 (s).
MW
4
(20)
368
At this point the correspondence with the low-energy S-, T - and U -parameters of Peskin
and Takeuchi [11] can be made more explicit (see, e.g., Ref. [16]):

2 c2
2 s2
cW
sW
1
Z
W
W
ZZ
ZZ
T (s) T (s)
S = 16 2 lim
T (s) L (s) +
e s0 s
sW cW
= 16
S3 (0),

WW
L (0) LZZ (0)
v2
1

T=
=
S (0),
2
2
2 (0) 6
(0)v 2 (0)
(0)v
MW
MZ
2

ZZ
sW
1
2
lim TW W (s) LW W (s) cW
T (s) LZZ (s)
2
e s0 s

Z
2
+ 2sW cW T (s) sW
T (s)
= 16
S4 (0).
U = 16
(21)
Next we want to verify that indeed all couplings have become running ones and that the
propagator matrix in the transverse neutral sector has become diagonal. The easiest way to
do this is by realizing that the complete matrix element for a given reaction can be written
in terms of subsets of matrix elements, each with a particular configuration of intermediate
gauge bosons and associated would-be Goldstone bosons. For the discussion of a particular
intermediate gauge/Goldstone boson that carries a particular momentum, the relevant set
of matrix elements can be represented by a propagator function that multiplies two distinct
gauge/Goldstone-boson currents. In analogy to what was done above, the trick is now to
explicitly pull out the coupling strength in these currents. For the B- and W a -currents this
amounts to JB = g1 jB and JW a = g2 jW a , respectively. Similarly a factor 1/v is pulled out
in the would-be Goldstone-boson currents: J = j /v and J = j /v. Finally, we have to
proof that the combination of propagator functions and pulled-out coupling factors gives
rise to running couplings and diagonal propagators.
Let us start with the transverse neutral sector, where we have to switch to the physical
mass eigenstates (see Eq. (5)):
J = cW JB sW JW 3 = e(jB jW 3 ),
JZ = sW JB + cW JW 3 = sW g1 jB + cW g2 jW 3 .
(22)
The generic amplitude structure for intermediate transverse neutral gauge bosons then
reads

Z

PT , (q) PT , (q)
J
J JZ
Z
JZ
PT , (q) PTZZ
, (q)

PT , (q) PT , (q)
e sW g1
= jB jW 3
Z
e cW g2
P
(q) PTZZ
, (q)

T ,
jB
e
e
(23)
.

sW g1 cW g2
jW
3
369
Using the propagator functions listed in Appendix A and the definitions of the running
couplings in Eqs. (19) and (20), one can rewrite this product of matrices according to

PT , (q) PT , (q)
e sW g1
e
e
Z
sW g1 cW g2
e cW g2
P
(q) P ZZ (q)
T ,
T ,

0
P T , (q)
e(s) sW (s)g1 (s)
=
e(s) cW (s)g2 (s)
0
PTZZ
, (q)

e(s)
e(s)
.
sW (s)g1 (s) cW (s)g2 (s)
(24)
The diagonal transverse propagators are given by

i
q q

P T , (q) = g 2 ,
s
q
PTZZ
, (q) =
2 (s)
g22 cW
P ZZ (q)
2 T ,
g22 (s)cW

g 2 (s)
M 2 (s) 1
q q
i

S3 (s) Z
,
=
1 22
g 2
s
q
s(s)
cW (s)
(25)
using the non-local -parameter

(s) =
S5 (s)
v2 +
v 2 (s)
=
.
v2 +
S5 (s) +
S6 (s) v 2 (s) +
S6 (s)
(26)
For the W -boson the generic amplitude structure reads
W

W

(q)g2 jW
= jW g2 (s)PTW,
(q)g2 (s)jW
,
jW g2 PTW,
(27)
with
W
(q) =
PTW,
g22
P W W (q)
2
g2 (s) T ,

M 2 (s) 1
q q
i
S3 (s) g22 (s)
S4 (s) W
.
1 g22 (s)
g 2
s
q
s
(28)
So, indeed all couplings have been transformed into (finite) running couplings and the
effective propagators are diagonal and finite. The complex poles of the diagonalized
transverse gauge-boson propagators can be obtained by solving the equations
s = 0,
s = Z =
s = W =
MZ2 (Z )/(Z )
1
g22 (Z )
2 ( ) S3 (Z )
cW
Z
2 ( )
MW
W
1 g22 (W )
S3 (W ) g22 (W )
S4 (W )
(29)
370
for the photon, Z-boson and W -boson, respectively.

In the longitudinal/scalar sector we get

0
PL, (q, )
e sW g1
e
jB jW 3
ZZ (q, )
s
e cW g2
g1
W
0
PL,
Z
g
2
g2
= jZ
P ZZ (q, Z )
j
2cW L,
2cW Z

iq q
g2
g2
= jZ
j .
DLZZ (s, Z )
2cW
q2
2cW Z
e
cW g2
jB

jW
3
(30)
In order to achieve this simplification, the Z-boson interactions were split into an
electromagnetic and isospin piece according to
JZ = sW g1 jB + cW g2 jW 3
2 c2
sW
g2
W
e(jB jW 3 ) +
(jB + jW 3 )
2sW cW
2cW
2
s 2 cW
g2
W
J +
jZ .
2sW cW
2cW
(31)
The electromagnetic Ward identity q J = 0 then takes care of all scalar electromagnetic
interactions, leaving behind a pure isospin piece that has to be combined with the would-be
Goldstone boson . In the next step we combine the left-over Z-boson amplitude with the
corresponding -amplitudes:

Z
ZZ

PL, (q, Z ) P (q, Z )
MZ /v
jZ
MZ /v
0
0
jZ j
Z
0
1/v
0
1/v
j
P (q, Z ) P (s, Z )

M2
1
1
= j i Z DLZZ (s, Z ) 2iMZ P Z (s, Z ) + P (s, Z ) j
v
s
v

1

1
1 i
1 i
1
S6 (s)
(s)
j = j
j .
= j
(32)
1+ 2
v(s) s
v(s)
v(s) s
v (s)
v(s)
Here we have used the propagator functions listed in Appendix A, the running couplings
as defined in Eqs. (19) and (20), and the two neutral-current Ward identities q J = 0 and
q JZ = iMZ J for an incoming momentum q. So, again we obtain running couplings. In a
similar way we can combine the W -boson amplitudes in the longitudinal/scalar sector with
the corresponding -amplitudes, yielding the generic amplitude structure j i/[sv 2 (s)]j .
If we would now use the unitary gauge in the massive gauge-boson sector, W/Z ,
all propagators involving would-be Goldstone bosons would vanish and the dressed W boson and Z-boson propagators would become
1 g q q /W(s)
g22 (s)
,
1 g22 (s)
S3 (s) g22 (s)
S4 (s)
2
s W(s)
g2
1

g 2 (s)c2
g 2 (s)
g q q /Z(s)
Z
ZZ

P
(33)
,
S3 (s)
(q, Z ) i 22 2 W 1 22
s Z(s)
g2 cW (s)
cW (s)
W
WW
P
(q, W ) i
371
where
W(s) =
2 (s)
MW
,
1 g 2 (s)
S3 (s) g 2 (s)
S4 (s)
2
Z(s) =
MZ2 (s)/(s)
1
g22 (s)
2 (s) S3 (s)
cW
(34)
This of course leads to a huge reduction of the effective number of Feynman rules.2 These
resummed expressions in the unitary gauge are a suitable starting point for the second level
of matching, i.e., the renormalization, which is performed explicitly in Appendix C.
3. High-energy behaviour and the zero-mode solutions

Although Eq. (8) describes a gauge-invariant action, there are other properties of
the local theory (SM) that are not shared by the truncated effective action. The most
pronounced one is certainly unitarity and the related high-energy behaviour of the
matrix elements. To exemplify this point, we have computed the matrix elements
M[e+ (p1 )e (p2 ) W + (p+ )W (p )] analytically. In Appendix D it is shown that the
matrix element for transversely polarized W bosons and left-handed electrons exhibits an
incorrect high-energy behaviour as a result of the presence of the factor
2
2 )
2 (p
2 (p+ )
(p+ p )
2 p2
p+
in the non-local triple gauge-boson interaction. This factor clearly diverges for large
2 is a constant. Furthermore, as we will see in the next section, there is a
energies, unless
rather important numerical discrepancy in the calculation of the cross section (e+ e
in the extreme forward region, which is dominated by the exchange of nearly
e e ud)
on-shell space-like photons.
It is obvious that any difference between the BBC approach and the calculations in the
FL scheme must originate from the different treatment of the three-point vertices, since the
two-point functions are identical in both schemes. In order to understand the discrepancies
in a more explicit way, let us define by the generic symbol B the difference between a
three-point vertex as computed in the FL scheme (FL ) and the one in the BBC approach
(BBC ). In the case of the photon, for instance, we obtain
B W + W (q, p+ , p ) = BBC, W + W FL, W + W ,
B W + (q, p+ , p ) = BBC, W + FL, W + ,

2 If all fermions would be massless or if there would be no doublet splitting (m = m ), then
S3 (s) =
f
f
2 (s) = c2 (s)Z(s). In that case we reproduce the propagators of the so-called

S4 (s) =
S6 (s) = 0 and W(s) = MW
W
massive fermion-loop (MFL) scheme for a massless internal world [4].
372
B + W (q, p+ , p ) = BBC, + W FL, + W ,
B + (q, p+ , p ) = BBC, + FL, + .

The momenta and Lorentz indices of the incoming gauge bosons are denoted by (q, )
for the photon, (p+ , ) for the W + -boson and (p , ) for the W -boson, respectively.
Similarly, the momenta of the incoming would-be Goldstone bosons are given by p .
Since all two-point functions are identical in the FL and BBC schemes, the above vertex
quantities should satisfy a number of equations, namely Ward identities with all two-point
functions switched off. These so-called zero-mode equations can be written as
q B W + W (q, p+ , p ) = 0,
p+ B W + W (q, p+ , p ) MW B + W (q, p+ , p ) = 0,
p B W + W (q, p+ , p ) + MW B W + (q, p+ , p ) = 0,
q B W + (q, p+ , p ) = q B + W (q, p+ , p ) = 0,
p+ B W + (q, p+ , p ) MW B + (q, p+ , p ) = 0,
p B + W (q, p+ , p ) + MW B + (q, p+ , p ) = 0,
q B + (q, p+ , p ) = 0.
(35)
For the Z-boson one obtains a similar set of zero-mode equations. In that case, however,
also the would-be Goldstone boson will feature explicitly in the expressions.
In order to study the zero modes in detail, we introduce the following general form of
the triple gauge-boson vertex (excluding #-tensor contributions3)
V W + W (q, p+ , p )

= ig W W x1 p+ g + x2 p g + x3 p+
g + x4 p
g + x5 p+
g + x6 p
g

+ x7 p+ p
p + x8 p p+
p+ + x9 p+ p+
p + x10p p+
p + x11 p+ p
p+

+ x12 p p p+ + x13p+ p+ p+ + x14 p p p ,
(36)
where g W W = e. The coefficients xi are scalar functions that depend on the squared
momenta and masses. As a result of CP-invariance, there is a general symmetry of this
vertex under the simultaneous transformations
p+ p
and ,
(37)
which turn incoming W bosons into outgoing W bosons with the same momenta and
Lorentz indices. This results in the relation

2
2
2
2
xs(i) q 2 , p
,
, p
, p+
xi q 2 , p+
(38)
where s(i) = {2, 1, 6, 5, 4, 3, 8, 7, 10, 9, 12, 11, 14, 13} for i = {1, . . . , 14}.
3 It is well known that these terms satisfy the Ward identities on their own, without involving the two-point
functions.
373
A similar Lorentz-covariant parametrization can be made for the other three-point

vertices
V W + (q, p+ , p )

= ig W W y1 g + y2 p+ p+
+ y3 p+ p
+ y4 p p+
+ y5 p p
,
V + W (q, p+ , p )

+ z3 p+ p
+ z4 p p+
+ z5 p p
= ig W W z1 g + z2 p+ p+
(39)

V + (q, p+ , p ) = ig W W w1 (q p )p+ (q p+ )p ,
(40)
and
where in the latter case the relevant Ward Identity has been taken into account. As a result
of CP-invariance we may relate the coefficients yi to the coefficients zi in Eq. (39).
If we now demand that all three-point vertices satisfy Eq. (35) we end up with 25
coefficients satisfying a system of 21 equations. This can be solved algebraically in terms
of 9 coefficients, the number of which can be reduced to 5 independent functions if the
symmetry relations are exploited.
In order to keep the discussion of Eq. (35) as simple as possible, we choose to neglect
all contributions from vertices involving would-be Goldstone bosons by considering
2
exclusively massless fermions. This can be done without loss of generality, since both
and the ensuing unitarity problem for transverse W bosons are also present in the unbroken
theory. In that case Eq. (36) is also valid for the ZW W vertex, provided that g W W is
replaced by gZW W = ecW /sW . It is not difficult to verify that the reduced system of
zero-mode equations has always a solution and that one can express all coefficients xi in
2 , b = p2 and c = (p p ), a
terms of four independent ones. For instance, using a = p+
+
solution may be represented by

b+c
(ax11 + bx12 ax13 bx14),
ab

c(a + c)
c(b + c)
b
x12 x14 ,
(x11 x13) +
x3 =
ab
ab
a
a(a + c)
b+c
(x11 + x13 ) +
(ax12 + bx14),
x4 =
ab
ab

a+c
b+c
a
x11 +
x7 =
x12 x13 x14 ,
ab
ab
b
c
x9 = x13.
b
x1 =
(41)
The rest of the coefficients are determined by using the symmetry relations (38). Notice
that, although we have expressed the solution algebraically in terms of four coefficients,
this number can be reduced to two independent functions by means of Eq. (38).
The four algebraically independent Lorentz structures to be used in the zero-mode
solution BV W + W (for V = , Z) may be represented as follows in momentum space.
374
The simplest structure corresponds to (x11 , x12 , x13 , x14) = (b + c, a c, 0, 0) and reads

p+ .
V1 = (qp )p+ (qp+ )p (p+ p )g p
(42)
The second one corresponds to the solution (1, 1, 0, 0):

2
V2 = (qp )p+ (qp+ )p g + g (p+ p )p+
p+
p

p
p+ p+ p
q + p q p+
.
g (p+ p )p
(43)
Note that this vertex originates from the operator

OF F F = Tr U2 (z, x)F
(x)U2 (x, y)F (y)U2 (y, z)F (z) ,
as was predicted in Eq. (1). The third structure corresponds to (0, 0, b, a):

2

2
(qp+ )p+
g p p
p+
q
V3 = p
2

2
g p+ p+
p
q .
p+
(qp )p
(44)
Finally the fourth structure corresponds to (0, 0, b(b + c), a(a + c)):

V4 = (qp )p+ (qp+ )p

2 2
2
2

p g p+
p p p
p+ p+ + (p+ p )p+
p .
p+
(45)
The triple gauge-boson vertex in the FL scheme, as presented in Ref. [3], can now be
expressed in terms of the vertex in the BBC approach plus a linear combination of all
four zero modes of Eqs. (42)(45) [17]. It is exactly this difference between the BBC
approach and the FL scheme, i.e., the zero-mode solution BV W + W , that we are after.
For our purposes, however, it would be enough to just determine the zero-mode solutions
that apply to either the q 2 0 or q 2 limits, since in those limits the BBC approach
starts to deviate.
There are several ways to attack the problem, but we think that the most economical
one would be to reduce as much as possible the information on the exact three-point vertex
FL,V W + W . This is motivated by the fact that we have future applications in mind where
vertices with more than three gauge bosons are needed, such as six-fermion processes or
four-fermion processes with an additional photon. In those cases one would like to avoid
a complete fermion-loop computation as much as possible. In fact, we may further reduce
the problem by taking into account the fact that, at least for four-fermion processes, we are
dealing with conserved external currents. These conserved external currents are the result
of either having massless fermions in the final state or having massive fermions that couple
can be neglected, leading

and p
to photons. This means that terms proportional to q , p+
to the following simpler form for Eq. (36):
VV W + W (q, p+ , p )

x1 x2
x11 x12

(p+ p ) g +
(p+ p ) p
p+
= igV W W
2
2

+ x4 p
g + x5 p+
g
.
(46)
375
The idea is now to use the information from the triple gauge-boson vertex in the FL scheme
and keep only those terms that are proportional to the four tensor structures appearing
in Eq. (46). The algebra of the vertex corrections has been performed with the help of
Form [13], resulting in an expression in terms of tensor coefficients [14]. Subsequently,
FeynCalc [15] has been used to reduce these tensor coefficients to scalar one-loop integrals
according to the PassarinoVeltman decomposition. The results obtained in this way fully
agree with the ones published in Ref. [3]. In the next step, all terms proportional to the
scalar three-point functions are discarded, since in the non-local approach we consider
only corrections based on two-point functions. A complete set of such three-point terms
obviously satisfies the zero-mode equations, but it cannot compensate any incorrect highenergy behaviour originating from the two-point sector. The remaining expressions consist
2 , 0, 0) as
of terms proportional to the scalar two-point functions B0 (q 2 , 0, 0) and B0 (p
well as rational terms that come from the tensor reduction and the four-dimensional limit.
Since our final goal is to provide a correction term to the BBC description, it is more
convenient to re-express these two-point functions in terms of the non-local coefficients
2 ) using the results of the previous section. Subsequently, the fermion2 (q 2 ) and
2 (p
2 , which will allow us to take the zero-virtuality limit.

mass dependence is restored in
Let us first discuss the final results for the zero-mode solution in the limit q 2 0. These
results can be represented in the following way:

x1 x2
x1 x2
= 0,
1
2
2
BBC
FL

x11 x12
x11 x12
2
2
2
BBC
FL
s2 + s3
s2 + s3

2 (s1 )
= 16gBBC
(s2 s3 )2
s2 s3
2s 3 7s2 s32 + 4s22 s3 s23
2s23 7s22 s3 + 4s2 s32 s33
2 (s2 ) 3
2 (s3 ),
3
s2 (s2 s3 )
(s2 s3 )3 s3
3 (x4 )BBC (x4 )FL
+
s2
s2
s (2s2 + 3s3 )
s (s 2s3 )2
2 (s1 ) + 2
2 (s2 ) 2 2
2 (s3 ),
+
s2 s3 s3
(s2 s3 )2
(s2 s3 )2 s3
4 (x5 )BBC (x5 )FL = 3 (s2 s3 ),
(47)
= 16gBBC
2 , s = p2 as well
where we have introduced the shorthand notations s1 = q 2 , s2 = p+
3
2
2
as gBBC = g2 /(64 ). The next step is to translate these four quantities into the basic
coefficients x11 , x12 , x13 and x14 with the help of Eq. (41):
2 (s1 s2 + s3 ) (3 + 4 )
,
s1

s2 s3 s1 s32 (s2 s1 )2
3 + 4
x13 = 1
+
2
s1 s2
2s1 s2
s3 s2 s1

2
s2 + s3 s1
+
+ 4 ,
3
s3 s2 s1
2s2
x11 =
(48)
376
with x12 and x14 determined by means of Eq. (38). These four coefficients can be inserted
in Eq. (41) in order to reconstruct a complete zero-mode solution that can be subtracted
safely from the BBC vertex. Note that the translation between 1 , . . . , 4 and x11, x13 ,
given in Eq. (48), has by itself already an important part of the information encoded. For
instance, a finite difference between the BBC and FL vertex corrections in the limit s1 0
is equivalent with the conditions
3 + 4
,
s3 s2
which is in full agreement with the explicit expressions in Eq. (47). These conditions
guarantee that the coefficients x11 , x12, x13 , x14 are finite, which in turn guarantees that
all x1 , . . . , x14 are finite, since no factors 1/s1 = 1/(a + b + 2c) are present in Eq. (41).
The same exercise can be performed for the limit s1 . In that case we find
1 = 0 and 2 =
(s ) s3
2 (s3 )
2 (s3 )
2 (s2 )
s1
s
2 (s1 ) + 2 2 2
+ 8gBBC
,
2
s2 s3
s2 s3
2 (s2 )
2 (s1 )
2 (s3 ) 2
2 (s2 )
2 (s3 ) 16gBBC
+
,
2 =
s2 s3
s1
3 = 4 = 0,
1 =
(49)
where we have kept all terms that can give rise to contributions to the amplitude that are
not suppressed by inverse powers of s1 . It is not difficult to see that the leading terms in
Eq. (49) can in fact be absorbed completely into the simplest zero-mode structure V1 of
Eq. (42), if multiplied by

2 (s2 )
2 (s3 )
igV W W
.
(p+ p )
s2 s3
This completes the explicit construction of the zero-mode solutions that should contain
the bulk of the differences between the BBC approach and the SM in the limits s1 0
and s1 . It is worthwhile to underline that the investigation performed in this section
does not, by any means, address the problem of unitarization of the effective BBC action
in general. The objective is to identify the differences, through the zero-mode solutions,
between the BBC and the FL approach, which is manifestly unitary, in order to assess their
physical significance. A numerical analysis of the importance of the zero-mode solutions
is the subject of the next section.
4. Results
In this section we present, as an illustrative example of our approach, numerical results
based on four-fermion production processes that involve interactions among three gauge
bosons: the so-called CC20 and CC10 families. We focus our studies on three particular
kinematical configurations:
(1) the small-angle (or single-W ) regime, using the process e+ e e e ud with a cut
on the angle of the outgoing electron;
377
Table 1
Input parameters versus computed quantities
Input parameters
mW = 80.35 GeV
Output values

Re(W ) = 80.3235 GeV
mZ = 91.1867 GeV
Re[ (5) (m2Z )1 ] = 128.89
mu = md = 0.0475188 GeV
(0)1 = 137.03599976
GF = 1.16639 105 GeV2
Im(W )
= 2.0575 GeV
Re(W )
mt = 146.966 GeV
(2) the configuration without angular cuts, using the total cross section for the process
e+ e e e ud (which only involves technical cuts related to the use of massless
fermions);
(3) the high-energy regime, using the process e+ e ud.

Our numerical analysis is based on NEXTCALIBUR [18]. The matrix-element computations are performed with the help of a new version of HELAC [19] that includes all relevant
vertices coming from the non-local effective action of Eq. (8), as described in Appendix A.
The gauge invariance of this implementation has been checked extensively by comparing
the results for the t HooftFeynman and unitary gauges. Particular attention has been paid
i in all possible ranges covto the numerical convergence of the non-local coefficients
2
ered by both q and the fermion masses. Finally, the computation of all necessary one-loop
three-point tensor coefficient functions is based on the numerical programme FF [20].
The subtraction of the zero-mode solutions has been limited to the two ranges q 2 0
and q 2 , where q 2 is the virtuality of the relevant exchanged photon or Z boson. In the
former limit q 2 t = (pe pe )2 , with pe and pe denoting the momenta of the incoming
and the outgoing electrons. In the latter limit q 2 s, where s represents the centre-of-mass
energy squared of the process.
In practice, one has to decide on the intervals of q 2 in which the zero-mode corrections
are switched on. In the present calculation we have selected the range 1 q 2 0 GeV2
for the first kind of zero-mode correction.4 For the high-energy regime we have applied the
zero-mode corrections in the full q 2 0 range, since our processes are anyway dominated
by double-resonant and single-resonant W -boson contributions.
In Table 1 we summarize the input parameters of our renormalization scheme and give
the resulting output values for the computed quantities. Typically, three bare quantities
the electromagnetic constant e, the weak coupling g2 and the Higgs vacuum expectation
value vhave to be fixed by three experimental data points. On the other hand, there
are several well-measured experimental quantities. Therefore, in order to add part of the
missing higher-order contributions and improve the predictive power of our computation,
we have decided to work with five experimental data points instead of three. This means
that, besides e, g2 and v, two more parameters get fixed. The first parameter is the
4 We have checked that our results remain the same when varying the lower cut-off value between 0.04 and
25 GeV2 .
378
Fig. 1. The evolution of the squared electromagnetic (e2 ) and weak (g22 ) couplings as a function of the scale |q|
in GeV. The solid (dashed) line represents the evolution for positive (negative) values of q 2 . The values for e2
and g22 predicted by the FW scheme are given by 0.09523 and 0.4260, respectively.
top-quark mass mt , which allows an effective description of the missing non-fermionic

corrections at high mass scales [2]. The second parameter is the common light-quark mass,
m = mu = md , which allows us to take into account the electromagnetic constant at zero
virtuality. For the other fermionic masses we use their PDG values [21]. The resulting
running of the renormalized electromagnetic and weak couplings are presented in Fig. 1.
More details on our renormalization procedure are given in Appendix C.
Since it is our aim to compare different schemes, we now present a few different
approaches. The first one is the widely used Fixed Width (FW) scheme, where a fixed
W -boson width is implemented in all W -boson propagators and where the GF -scheme is
applied for evaluating the weak parameters. We recall that the latter is defined by using
mW , mZ and GF as input parameters, together with the two relations
2
=1
sW
m2W
m2Z
2
2
GF m2W sW
.
In addition we introduce two hybrid schemes, where the real (imaginary) part is fixed
by the FL (BBC ) scheme and vice versa. This we do in order to investigate possible
differences between the real and imaginary parts of the corrections that are missing in the
BBC approach. Finally, we denote by BBCN the scheme that subtracts the relevant zeromode solutions of the Ward identities.
379
Table 2
using the cuts | cos(e )| > 0.997 and M(ud)
> 45 GeV.
Cross sections (in fb) for the process e+ e e e ud,
In each second row of FL-scheme entries we give the results taken from Ref. [22], which differ slightly from our
results owing to a different treatment of the hadronic part of the photonic vacuum polarization
s [GeV]
FW
BBC
Re(BBC) + Im(FL) Re(FL) + Im(BBC)
BBCN
FL
183
89.17(26)
80.00(32)
81.80(32)
82.23(35)
189
99.80(24)
89.38(34)
92.19(35)
92.02(35)
200
120.98(31)
108.41(42)
111.50(43)
111.52(43)
500
1000
897.1(3.2)
2064(12)
814.8(4.6)
1931(16)
837.2(4.7)
1968(29)
833.6(5.6)
2042(55)
84.83(32)
84.38(33)
83.28(6)
95.13(36)
94.60(36)
93.79(7)
114.69(44) 114.61(44)
113.67(8)
856.3(4.8)
856.3(4.8)
1937(16)
1964(16)
In order to study the small-angle behaviour of the various approximations, we focus on

the reaction e+ e e e ud with the following two cuts

cos(e ) > 0.997,
> 45 GeV.
M(ud)
The first cut, on the angle between the outgoing electron and the electron beam, ensures
sensitivity to contributions that are mediated by t-channel graphs. The second cut, on the
invariant mass of the ud system, is added mainly to comply with earlier calculations. The
corresponding results are presented in Table 2, from which we deduce that
the FW scheme overestimates the cross sections by up to 6%. This is mainly due to the
use of non-running couplings, especially in the electromagnetic sector;
the BBC approach underestimates the cross sections by up to 6%. This, in contrast
to the previous case, is due to differences in the treatment of the triple gauge-boson
vertex. This fact reflects the importance of subtracting zero-mode contributions;
the BBCN approach reproduces the results obtained in the FL scheme within MC
accuracy.
In Table 3 the predicted cross section is shown for different angular regions of the
outgoing electron. The four rows correspond to the FW, BBC, BBCN and FL schemes,
respectively. The previous observations, which were deduced for the total cross section
with angular cut | cos(e )| > 0.997, i.e., e < 4.44 , are more or less reproduced uniformly
in the extreme-forward angular distribution between 0.0 and 0.4 .
In Table 4 we present results without angular cuts. The discrepancy is now reduced
substantially, reflecting the fact that an important component of the total cross section,
namely the contribution of double resonant graphs, is equally well described by the
different schemes. This fact ceases to be true at energies above 500 GeV where singleresonant and multi-peripheral contributions take over again. Nevertheless the BBCN
scheme still follows the FL results within MC accuracy.
Finally, as already stated, the BBC approach violates unitarity and as such gives rise
to a bad high-energy behaviour. In order to better reveal this property, one has to go to
The results
rather high energies. To this end, we consider the process e+ e ud.
for the corresponding total cross section are presented in Table 5, which shows rather
380
Table 3
using the cut M(ud)
> 45 GeV. The results are presented
for different energies s and for different bins of the angle e between the outgoing electron and the electron
beam. The four rows correspond to the FW, BBC, BBCN and FL schemes, respectively
e
183 GeV
189 GeV
200 GeV
0.0 0.1
49.01(17)
42.98(23)
46.17(24)
45.56(24)
55.25(19)
48.55(26)
52.47(28)
51.53(27)
67.81(24)
59.35(32)
63.73(34)
63.17(34)
0.1 0.2
7.03(7)
6.16(9)
6.77(9)
6.60(9)
7.87(7)
6.94(10)
7.63(11)
7.46(10)
9.36(9)
8.35(12)
8.93(13)
8.84(13)
0.2 0.3
4.21(5)
3.59(7)
3.92(7)
3.87(7)
4.55(6)
4.21(8)
4.34(8)
4.26(8)
5.40(7)
4.84(9)
5.17(10)
5.13(10)
0.3 0.4
2.80(4)
2.61(6)
2.85(6)
2.81(6)
3.23(5)
2.81(6)
3.08(7)
3.17(7)
3.87(6)
3.55(8)
3.80(8)
3.76(8)
Table 4
using only the cut M(ud)
> 45 GeV, i.e., no angular cuts
are imposed
s [GeV]
FW
BBC
BBCN
FL
183
189
200
500
1000
766.6(1.0)
808.7(1.1)
851.4(1.2)
1377(4)
2555(17)
770.3(2.7)
807.4(2.7)
846.9(2.9)
1299(6)
2387(16)
775.3(2.7)
813.3(3.0)
857.1(3.0)
1344(8)
2463(28)
780.1(3.1)
810.9(2.9)
854.5(2.9)
1347(8)
2471(30)
773.9(2.7)
815.1(2.7)
859.8(3.3)
1341(6)
2414(18)
Table 5
using the cut M(ud)
> 45 GeV
Cross sections (in fb) for the process e+ e ud,
s [GeV]
FW
BBC
BBCN
200
500
1000
2000
5000
10000
686.86(81)
270.71(45)
103.67(19)
36.107(75)
8.067(25)
2.445(11)
702.8(2.5)
275.5(1.2)
107.38(46)
43.36(18)
53.05(22)
187.53(62)
704.6(2.4)
276.4(1.2)
106.91(47)
40.05(18)
30.35(13)
94.66(39)
704.6(2.4)
276.5(1.2)
106.87(47)
40.10(19)
30.81(19)
95.31(44)
777.7(3.1)
814.2(3.0)
860.9(3.0)
1345(8)
2463(27)
FL
702.3(2.5)
704.9(2.4)
275.0(1.2)
276.3(1.2)
105.84(46) 106.10(47)
36.45(17)
36.89(19)
8.486(61)
8.225(45)
4.227(26)
2.548(27)
clearly that the BBC approach and its hybrids start to diverge above 1 TeV. In contrast,
the BBCN scheme exhibits a good high-energy behaviour. However, in comparison with
the FW and FL schemes, the BBCN approach exhibits a substantial discrepancy above
7 TeV. In trying to analyse this point, we found that the BBCN and FL schemes agree
381
very well for massless fermions: e.g. at 10 TeV the FL scheme gives = 1.209(77) fb
whereas the BBCN approach yields = 1.207(77) fb. If we would use the nominal values
for the masses of the light fermions, but reduce the top-quark mass to mt = 10 GeV, these
numbers would change to 1.226(78) and 1.233(78), respectively. Recall that in the highenergy regime the BBCN approach is defined by assuming massless fermions. The explicit
fermion-mass dependence is re-introduced only through the non-local coefficients. The
lesson to be learned here is that fermion masses, and more in particular the top-quark mass,
play a rather important role in the triple gauge-boson vertex and cannot be accommodated
2 non-local coefficient alone.
by the
5. Conclusions
In this paper we have presented an analysis that represents a first step towards an
effective-action description of fermion-loop corrections to multi-fermion reactions like
e+ e n fermions. The study is based on the proposal formulated in Ref. [10]. It relies
on a re-organization of the expansion of the effective action of the full theory. This reorganization is performed in terms of gauge-invariant operators, involving an arbitrary
number of gauge-boson and Higgs fields, multiplied by non-local coefficients. After a
truncation of this expansion at the two-point-function level, a non-local effective theory
is obtained that is consistent with all Ward identities for arbitrary n-point functions.
The next important step was the identification (matching) of the non-local coefficients
with the fermionic one-loop self-energy contributions predicted within the Standard Model
of electroweak interactions. Applied to physical processes that do not involve interactions
among more than two gauge bosons, the so-obtained effective theory and the Standard
Model lead to identical results. After renormalization, a complete description in terms
of running couplings is established and the correct (fermion-loop) scale dependence is
obtained, both in the high- and low-scale regimes.
Although the truncation of the expansion at the level of the two-point functions
is gauge-invariant, it introduces nevertheless a bad high-energy behaviour by violating
unitarity. This shows up, for instance, in the amplitudes for the production of transversely
polarized massive gauge bosons, like e+ e WT+ WT . We have identified explicitly the
zero modes of the Ward identities that are responsible for such behaviour. Based on the
appropriate high- and low-scale limits, we have reconstructed the two simplest zero-mode
solutions that, if subtracted from the non-local triple gauge-boson vertex, would restore
the agreement between the effective theory and the Standard Model. More specifically,
we have studied the numerical effect of the zero-mode solutions for several four-fermion
production processes, both for CC20 and CC10 families. We have observed the following:
the effect of the zero modes is essential in restoring the good high-energy behaviour
above 500 GeV;
the zero modes also account for the substantial discrepancy between the effective
theory and the Standard Model in the extreme forward region if electrons/positrons
are present in the final state;
382
the contribution of the zero modes to the intermediate-energy regime may be

neglected safely.
It remains an open question how to construct in a more systematic way the relevant zeromode solutions that will restore, to a given accuracy, the agreement between the effective
theory and the Standard Model. Moreover, in order to extend the scheme to processes
that involve interactions among four or more gauge bosons, like for instance six-fermion
production in e+ e collisions, special care should be devoted to a unitarity-preserving
reformulation of the non-local effective action.
Acknowledgements
We would like to thank Frits Berends for his support during the earlier stages of this
work.
Appendix A. Non-local Feynman rules

In this appendix we list the non-local contributions to the various two-point and
three-point interactions (see Ref. [10] for more details). We start off with the two-point
interactions. In order to calculate the non-local propagators we add the local gauge-fixing
Lagrangian corresponding to the covariant R gauge:

2
2
1 1
1
LR (x) =
A (x) +
Z (x) Z MZ (x)
2
Z

2 +
+
W (x) iW MW (x) W (x) + iW MW (x) ,

+
W
(A.1)
where , Z and W are gauge parameters. Taking into account all local and nonlocal bilinear interactions we find the following (dressed) gauge-boson propagators after
inversion (V = , Z, W ):
VV
V
VV
P
(q, V ) = PTV,
(q) + PL,
(q, V )

q q
q q
,
iDLV V q 2 , V
= iDTV V q 2 g 2
q
q2

q q
Z
Z
Z
Z
P
(q) = P
(q) = PT , (q) = iDT q 2 g 2 .
q
Writing q 2 = s, the gauge-boson propagator functions DT and DL are given by

2
5 (s) 1 ,
2 (s) MW
DTW W (s) = s 1 +
1+
5 (s)] W M 2
s[1 +
W
,
DLW W (s, W ) = W
5 (s)](s W M 2 )2
[1 +
W
(A.2)
383

2
2
3 (s) + cW
4 (s)
1 (s) 2
2 (s) +
DT (s) = s 1 + sW
D(s)

2
5 (s) +
6 (s) D(s),
MZ 1 +
1
DL (s, ) = ,
s

ZZ
2
2
3 (s) +
4 (s) + cW
2 (s) + 2
DT (s) = s 1 + sW
1 (s) D(s),
5 (s) +
6 (s)] Z M 2
s[1 +
Z
,
5 (s) +
6 (s)](s Z M 2 )2
[1 +
Z

s2
Z
2 (s)
3 (s)
4 (s) + W
1 (s)
3 (s)
D(s), (A.3)
DT (s) = ssW cW
2
cW
DLZZ (s, Z ) = Z
with

s2

1 (s) 1 +
2 (s) +
4 (s) W
3 (s) 2
1+
2
cW

5 (s) +
6 (s)
sMZ2 1 +

2
2
3 (s) +
4 (s) + cW
1 + sW
1 (s) .
2 (s) + 2
D(s) = s 2
(A.4)
Note that DTW W (0) = DLW W (0) and DTZZ (0) = DLZZ (0) as a result of analyticity
requirements.
The dressed propagators involving the would-be Goldstone bosons are relatively simple:
PW
(q, W ) = P
q P W
(q, W ) = iq
5 (s)
W MW
5 (s)](s W M 2 )2
[1 +
W
(s, W ),
2 [1 +
5 (s)]
s W MW
,
P (s, W ) = i
2 )2

[1 + 5 (s)](s W MW
PZ (q, Z ) = PZ (q, Z ) = q
5 (s) +
6 (s)]
Z MZ [
,

[1 + 5 (s) + 6 (s)](s Z MZ2 )2
q P Z (s, Z ),
P (s, Z ) = i
5 (s) +
6 (s)]
s Z MZ2 [1 +
.
5 (s) +
6 (s)](s Z M 2 )2
[1 +
Z
(A.5)
Now we come to the three-point vertices. For a compact notation we first introduce the
following three tensor structures:
T 1 2 (q1 , q2 ) = (q1 q2 )g 1 2 q1 2 q2 1 ,
A1 ,2 3 (q) = g 1 2 q 3 g 1 3 q 2 ,
5 (q 2 )
5 (q 2 )
2
1
(q2 q1 ) .
A5 (q1 , q2 ) =
q22 q12
(A.6)
384
Table 6
V1 V2 V3
W +W
ZW + W
A2
A31
A41
Non-zero coefficients Aj kl
sW
cW
sW
2 /c
sW
W
sW
cW
A231 = 2cW sW
2 c2
A123 = A132 = 1, A231 = sW
W
In terms of these tensor structures the non-local three-point interactions read
:

(2qj + ql )l
1 j ,k l
j kl
2
j k
ig2 A2
# 2 (qj )
T
(qj , qk ) + A
(qj )
2
(qj + ql )2 qj2
perm

3 q12 + A41
4 q12
+ A1 ,2 3 (q1 ) A31

2
MW
j k l
Aj kl g
A5 (qj , qk ) .
+
2cW perm
(A.7)
The summation includes all possible permutations (j, k, l) of (1, 2, 3) and the various
couplings are given in Table 6.
:

2
2
2
2
sW
ig2
1 2

C31 3 q1 + C32 3 q2 + C41 4 q1 + C42 4 q2
(q1 , q2 )
T
MW
cW

M2
5 qj2 + 2g j k C6j k
6 qj2 .
C5j k q j A5 k (qj , q) + g j k
W
2cW perm
(A.8)
The summation includes all possible permutations (j, k) of (1, 2) and the various couplings
are given in Table 7.
ie
2
1 2
E1 (q2 q3 )A5 1 (q2 , q3 ) + q2 1
5 q2 q3
5 q3
2cW sW

6 q12 + E3 q 1
6 q32 .
+ E23 (q2 q3 )1
3
(A.9)
385
Table 7
SV1 V2
C31
C32
C41
C42
C512
C521
C612
C621
H ZZ
sW cW
sW cW
2
cW
2
cW
1/cW
1/cW
1/cW
1/cW
H Z
2
sW
2
cW
sW cW
sW cW
sW cW
sW cW
2
sW
H W +W
2
sW
cW
cW
W +W
icW
icW
ZW
sW
cW
cW
sW
2 c2
sW
W
2sW cW
Table 8
V1 S2 S3
E1
2
2
sW cW
E23
E3
2cW sW
2i
W H
cW
icW
2icW
Z +
+
ZH
Table 9
S1 S2 S3
Non-zero coefficients Fj kl
Fj kl = 1 for all permutations of (1, 2, 3)
HHH
H
F123 = F132 = F231 = F321 = F213 = F312 = 1
H +
F123 = F132 = 1
F123 = F132 = i
The various couplings are given in Table 8.

i
6 qj2 .
(qj ql )Fj kl
v perm
(A.10)
The summation includes all possible permutations (j, k, l) of (1, 2, 3) and the various
couplings are given in Table 9.
Appendix B. Two-point functions in the FL scheme

In this appendix we list the various unrenormalized fermion-loop self-energies in the
SM, which will be needed for determining the six non-local coefficients.
386
The gauge-boson self-energies can be written as

:

V1 V2
(q) = iTV1 V2 q 2
i
q q
q q
iLV1 V2 q 2
,
g 2
q
q2
(B.1)
where T and L are the transverse and longitudinal gauge-boson self-energies,

respectively. Writing q 2 = s, we find for the transverse self-energies

f 2
1
+ s B0 (s, mf , mf ) B0 (s, 0, 0)
T (s) =
NC Qf s B0 (s, 0, 0)
3
3
f

2
+ 2mf B0 (s, mf , mf ) B0 (0, mf , mf )
f
s
(B.2)
NC Q2f f (s)
f
and

|Qf |
sW 2
Qf f (s),
4sW cW
cW
f

f (c2 s 2 )|Qf | s 2
TZ (s)
W
W
W
ZZ
2
NC
+ 2 Qf f (s) + 2 ,
T (s) = s
2
2
4sW cW
cW
cW
f
Z
T (s) = s
TW W (s) = s
NC
NC
|Qf |
2
4sW
f (s) + TW (s).
(B.3)
The scalar two-point functions B0 are defined in the usual way [14] and
f 2|Qf | 1
f 2
2
TZ (s) = 2MW
T
N
m
B
(s,
m
,
m
)
s
NC
f (s),
0
f
f
C f
2
2
8sW
8sW
f
f

f
s m2f B0 (s, mf , mf ) B0 (s, mf , mf )
TW (s) = TZ (s) +
N
C
2
24sW f
m2f (m2f m2f )

s
B0 (s, mf , mf ) B0 (0, mf , mf )

(B.4)
The constant T represents the universal tadpole contribution, which does not need to be
specified in view of the fact that we will perform tadpole renormalization anyhow (see
later). In a similar way the longitudinal gauge-boson self-energies are given by
L (s) = 0,
Z
L (s) = 0,
LZZ (s) = 2MZ2 T
f
NC m2f B0 (s, mf , mf ),
2 c2
8sW
W f

2
T+
LW W (s) = 2MW
2
8sW
387

f
NC m2f B0 (s, mf , mf )

.
m2f
m2f

iV1 S2 (q) = iq V1 S2 q 2
(B.6)
i S1 S2 (q 2 ).
(B.7)
B0 (s, mf , mf ) B0 (0, mf , mf )
(B.5)
s
In the longitudinal/scalar sector there are a few more self-energies to be considered:
These self-energy functions can be expressed in terms of the longitudinal gauge-boson

self-energies given above:

i ZZ
L (s) MZ2 T ,
Z (s) = Z (s) =
MZ

s ZZ
(s) = 2 L (s) 2MZ2 T ,

MZ

1 WW

2
W (s) = W (s) =
L (s) MW
T ,
MW

s WW
2
T .
(s) = 2 L (s) 2MW
(B.8)
MW
For completeness we also give the fermion-loop self-energy of the physical Higgs boson:

1 f 2 2
2
H H (s) = 3MH
4m
T
N
m
s
B0 (s, mf , mf )
f
f
C
8 2 v 2
f

2
B
+ 2mf 0 (0, mf , mf ) + 1 .
(B.9)
The above-given longitudinal/scalar self-energies satisfy the following Ward identities:
MZ2
(B.10)
(s) = 0,
s
M2

LW W (s) 2MW W (s) W (s) = 0.
(B.11)
s
As a next step we perform tadpole renormalization. This involves shifting the bare
vacuum in such a way that at one-loop level it coincides with the true vacuum of the
Higgs potential. Or in other words, the one-point counterterm generated by the finite shift
of the bare vacuum v completely compensates the tadpole self-energy terms T . This is
equivalent to the following effective procedure: keep the bare vacuum as it is, but remove
the terms T from the W W, ZZ, W , Z , and H H self-energies. The and selfenergies receive both one-point and two-point counterterms, which exactly cancel each
other. So, the tadpole contributions to these self-energies should be kept and are therefore
merged with the rest of the fermion-loop corrections. This is a trivial exercise, since the
and self-energies have an internal cancellation of all terms T (see the expressions
above).
LZZ (s) 2iMZ Z (s)
388
Appendix C. Renormalization conditions

An essential ingredient of the matching procedure is the renormalization of the nonlocal coefficients, which takes the form of matching the non-local matrix elements and
cross sections with explicit experimental observables. Various options are open, each with
their own merits. Let us go through the most popular renormalization/matching conditions,
bearing in mind that we only have to fix the running couplings v 2 (s), 1/g22 (s) and 1/e2 (s).
C.1. Muon decay
One of the often applied matching conditions is based on the charged-current muondecay process e e . In the unitary gauge we obtain the matrix element

g22 W W
P (qW , W ) u (p ) u (p )
2

u e (pe ) ve (p e ) ,
M1 =
(C.1)
with qW = pe + p e and = (1 5 )/2. Upon neglecting me and m with respect to

MW , the expression simplifies to
M1 =

i
u (p ) u (p ) u e (pe ) ve (p e )
2
V1 (qW )
in terms of the inverse amplitude

2
2
1 2
1
2
2

.
V1 (qW ) = 2qW 2 2 S3 qW S4 qW v 2 qW
2
g2 (qW )
(C.2)
(C.3)
This final step was obtained with the help of Eqs. (20), (33) and (34). In the muon-decay
2 = O(m2 ) M 2 . In that case the (lowprocess we can go one step further, since qW
W
2
energy) inverse amplitude reads V1 (0) = v (0)/2. The actual matching condition links
this inverse amplitude to the experimentally-determined coefficient of the effective (lowenergy) charged-current V A Lagrangian

Leff = 2 2 GF e e + .
(C.4)
In this way we obtain
1
S5 (0) =
v 2 (0) = v 2 +
2 GF
1
v 2 (s) =
+
S5 (s)
S5 (0).
2 GF
(C.5)
C.2. The W -boson mass

A second, optional matching condition involves the mass of the W bosons. For the
2 ). The most commonly
definition we can again make use of the inverse amplitude V1 (qW
used procedure is the on-shell condition

Re V1 m2W = 0

Re v 2 (m2W )
1
=
Re
S2 m2W ,
2
2
g2
4mW
(C.6)
389
where mW is the experimentally-determined W -boson mass (based on an on-shell

analysis). This results in

Re v 2 (m2W )
1
=
S2 (s) +
Re
S2 m2W +
S3 (s) +
S4 (s).
2
2
g2 (s)
4mW
(C.7)
The on-shell procedure breaks down if one includes two-loop corrections [23], therefore
it is sometimes better to use the complex W -boson pole W in the matching procedure:
V1 (W ) = 0
1
v 2 (W )
=
S2 (W ).
4W
g22
(C.8)
The real part of this complex pole can now be identified with the experimentallydetermined W -boson mass, provided the same complex procedure is adopted in the data
analysis.
C.3. The Z-boson mass
A very precisely known experimental observable is the Z-boson mass, so it is a natural
candidate for performing the matching. A defining process for the Z-boson mass is
the reaction e e , which leads in the unitary gauge to the following (inverse)
amplitude:

i
M2 =
u (p ) v (p ) ve (p e ) ue (pe ) ,
2
V2 (qZ )
2 2

2
2

2 cW (qZ )

S6 qZ2 ,
V2 qZ = 4qZ 2 2 S3 qZ v 2 qZ2
(C.9)
g2 (qZ )
with qZ = pe + p e . This expression was again obtained with the help of Eqs. (20), (33)
and (34). Using Eqs. (19) and (20), the experimentally-determined value of the Z-boson
mass in the on-shell approach (mZ ) and the on-shell condition Re V2 (m2Z ) = 0, one arrives
at the following quadratic equation:
Re A Re B
+ 2 + Re C,
g24
g2
2
2
A = e mZ ,
2

B = 2e2 m2Z
S2 mZ +
S3 m2Z +
S4 m2Z 1,
0=
C=
S6 (m2Z )
v 2 (m2Z ) +
2

2
S2 mZ +
S3 m2Z +
S4 m2Z
+ e2 m2Z
4m2Z
2

S2 mZ
S4 m2Z .
The relevant solution is given by

1
1
=
Re B (Re B)2 4 Re A Re C .
2
2 Re A
g2
(C.10)
(C.11)
For a matching procedure based on the complex Z-boson pole, Z , one merely has to
replace Re A, Re B, Re C and m2Z by A, B, C and Z .
390
C.4. The electromagnetic coupling

The matching condition for the electromagnetic coupling has to be addressed with care,
in view of its far-reaching consequences for the low- and high-scale behaviour of the
cross sections. The complication is caused by the hadronic part of the photonic vacuum
polarization, which is sensitive to non-perturbative strong-interaction effects through the
exchange of gluons with low momentum transfer.
We can either match at a high scale, like q2 = m2Z , or at a low scale, like the Thomson
limit q2 = 0. In the former case we have to evolve the running coupling down to low
scales in order to deal with phenomena that involve nearly on-shell photons (cf. single
W production). In the latter case we have to evolve the running coupling up to high
scales in order to properly describe high-scale reactions. From this it should be clear
that preferably we want to match the complete running of the electromagnetic coupling,
instead of matching it in a single point. To this end we have to exploit the explicit
parametric dependence of the non-local coefficients, i.e., the dependence on the fermion
masses mf . It has no use fiddling around with the lepton masses, since these masses are
experimentally well-known and the leptonic evolution of the electromagnetic coupling
is free of ambiguities. The same does not apply to the hadronic part of the photonic
vacuum polarization, in view of the various light-quark bound states that contribute. So,
the light-quark masses are prime candidates for tuning the evolution of the electromagnetic
coupling.
We start with the electromagnetic coupling at the LEP1 Z peak. Usually this coupling is
presented for five active quark flavours and without imaginary part, i.e., Re[ (5) (m2Z )1 ].
The relation between (5) (s) and (s) is fixed by the requirement of top-quark decoupling
at s = 0:
1 f 2 f (s)
1
(s) t (0)
1
= +
= (5)
.
NC Qf
+ NCt Q2t t
(s)
(s)
(C.12)
By fixing the value of Re[ (5) (m2Z )1 ] Z1 we obtain

2
f
f (mz )
1
1
1
2
t
2 t (0)
=
N
=
N
Q
Re
Q
.
f
t
C
C
e2 4 4Z
e2
e2
(C.13)
f =t
The leptonic as well as top-quark contributions can now be calculated perturbatively.

Subsequently we tune the evolution to lower scales by using a set of effective lightquark masses. In the standard procedure this set of light-quark masses represents a
perturbative fit to the once-subtracted dispersion integral of the experimental observable
R (s) = 43s2 (0) (e+ e hadrons):

f =q, f =t
f
NC Q2f
f (s) f (0)
4
s
=
12 2

4m2
ds
R (s )
,
s (s s i)
(C.14)
which automatically covers all non-perturbative hadronic contributions. Note that in this
way the quality of (0) is linked to the quality of the effective-mass parametrization at the
391
matching scale s = m2Z and to the quality of the perturbative calculation used in the fit.
This can be circumvented by adopting an alternative procedure, which uses the value of
(0) 0 as additional matching condition, leading to
f
f (0)
1
1
2
=
N
Q
.
f
C
e2 40
e2
(C.15)
This second matching condition can be implemented by tuning the effective u- and d-quark
masses.
Based on the matching conditions discussed above, various matching procedures are
possible.
(1) The LEP1 procedure. This procedure combines the measured top-quark mass at
the Tevatron with three of the above-mentioned matching conditions, i.e. the muondecay condition and the two on-shell LEP1 conditions for the Z-boson mass and the
electromagnetic coupling. So, with this procedure the input parameters are GF , mZ ,
Re[ (5) (m2Z )1 ] Z1 , mt , the lepton masses and the standard set of effective light-quark
masses.
(2) The LEP2 procedure. This procedure adds the on-shell W -boson mass mW to the
matching conditions of the LEP1 procedure. Because 1/g22 is now matched twice, we end
up with a consistency relation. The dominant ingredient in this relation is the top-quark
contribution. Therefore we can determine the top-quark mass by solving the consistency
relation iteratively. Since we exclusively take into account fermion loops, the resulting topquark mass will be an effective parameter that mimics bosonic loop effects. Its value will
come out appreciably lower than the experimental measurement (see Ref. [3]).
(3) Our procedure. In the present analysis we have added the electromagnetic coupling
in the Thomson limit, (0) 0 , to the matching conditions of the LEP2 procedure. This
results in an alternative set of effective light-quark masses, with adjusted values for the
u- and d-quark masses. Moreover, we have replaced the on-shell conditions for the W and Z-boson masses by their complex counterparts. So, we use Eqs. (C.5), (C.8), (C.13)
and (C.15), as well as the complex version of Eq. (C.11). This system of equations is
solved iteratively, resulting in the determination of the complex gauge-boson poles W,Z ,
the effective top-quark mass mt and the effective common light-quark mass mu = md .
Appendix D. e+ e WT+ WT and high-energy unitarity
In this appendix we have a closer look at the process of on-shell transverse W -pair
production, e+ (p1 )e (p2 ) WT+ (p+ )WT (p ), in the high-energy limit. Neglecting the
electron mass, the momenta of the particles are defined as follows in the initial-state centreof-mass frame
s
s
(1, 0, 0, 1),
p+ =
(1, 0, sin , cos ),
p1 =
2
2
s
s
p2 =
(1, 0, 0, 1),
p =
(1, 0, sin , cos ),
2
2
392

2 /s is the velocity of the W bosons. In
where is the scattering angle and = 1 4MW
this frame the polarization vectors for transversely polarized W bosons are given by
1
# (p+ ) = (0, i, cos , sin ),

2
1
# (p ) = (0, i, cos , sin ),

2
where the subscripts indicate the transverse helicities 1 of the considered W boson.
Note that these polarization vectors are orthogonal with respect to both p+ and p . Finally,
the left- and right-handed electronpositron currents can be written as
s(0, i, 1, 0),
JR = ve (p1 ) + ue (p2 ) = s(0, i, 1, 0),
JL = ve (p1 ) ue (p2 ) =
which is orthogonal with respect to the total initial-state momentum p1 + p2 = p+ + p .

Exploiting the various properties of the polarization vectors and electronpositron
currents, we end up with the following leading non-local contributions to the right- and
left-handed transverse amplitudes at high energies:

s 2 s
2
MR (ie) sin 2 MW
2
D(s)

s
3 (s) + MZ2 1 +
6 (s) ,
5 (s) +
2
cW

s 2 s
2

ML (ie) sin 2 MW
2
D(s)

2

sW
s
2

2 1 + 1 (s) 2 3 (s) + MZ 1 + 5 (s) + 6 (s) .
(D.1)
2sW
cW
A few remarks are in order here. First of all, the matrix elements MR,L given in Eq. (D.1)
are valid only if both W bosons have the same transverse helicity. The matrix elements
(M 2 )/2 is the leading
vanish if the W bosons have opposite helicity. Second, the factor s
2
W
high-energy component of the non-local triple gauge-boson vertex of Eq. (A.7), which has
to be contrasted with the corresponding factor 1 for the local triple gauge-boson vertex.
(M 2 ) originates from the typical non-local expression
The derivative
W
2
2)
2)
2 (p
2 (p+
2
p
2
p+
2
2 MW
.

2 M 2
p
W
(D.2)
Since D(s) s 2 at high energies (see Eq. (A.4)), ML will have an incorrect highenergy behaviour, growing with s as a result of the non-local triple gauge-boson factor
(M 2 )/2. (Note that this is not the case for MR in view of the fact that
3 (s) vanishes
s
2
W
at high energies.) On the basis of this observation we conclude that the BBC approach has
a problem with high-energy unitarity.
393
References
[1] M.W. Grnewald, et al., hep-ph/0005309;
ECFA/DESY LC Physics Working Group Collaboration, E. Accomando, et al., Phys. Rep. 299 (1998) 1,
hep-ph/9705442.
[2] E.N. Argyres, et al., Phys. Lett. B 358 (1995) 339, hep-ph/9507216.
[3] W. Beenakker, et al., Nucl. Phys. B 500 (1997) 255, hep-ph/9612260.
[4] G. Passarino, Nucl. Phys. B 574 (2000) 451, hep-ph/9911482.
[5] E. Accomando, A. Ballestrero, E. Maina, Phys. Lett. B 479 (2000) 209, hep-ph/9911489.
[6] R.G. Stuart, Phys. Lett. B 262 (1991) 113;
A. Aeppli, G.J. van Oldenborgh, D. Wyler, Nucl. Phys. B 428 (1994) 126, hep-ph/9312212.
[7] W. Beenakker, F.A. Berends, A.P. Chapovsky, Nucl. Phys. B 548 (1999) 3, hep-ph/9811481.
[8] A. Denner, S. Dittmaier, M. Roth, D. Wackeroth, Nucl. Phys. B 587 (2000) 67, hep-ph/0006307.
[9] K. Melnikov, O.I. Yakovlev, Nucl. Phys. B 471 (1996) 90, hep-ph/9501358;
W. Beenakker, A.P. Chapovsky, F.A. Berends, Phys. Lett. B 411 (1997) 203, hep-ph/9706339;
W. Beenakker, A.P. Chapovsky, F.A. Berends, Nucl. Phys. B 508 (1997) 17, hep-ph/9707326;
A. Denner, S. Dittmaier, M. Roth, Nucl. Phys. B 519 (1998) 39, hep-ph/9710521;
A. Denner, S. Dittmaier, M. Roth, Phys. Lett. B 429 (1998) 145, hep-ph/9803306.
[10] W. Beenakker, F.A. Berends, A.P. Chapovsky, Nucl. Phys. B 573 (2000) 503, hep-ph/9909472.
[11] M.E. Peskin, T. Takeuchi, Phys. Rev. D 46 (1992) 381.
[12] P. Sikivie, L. Susskind, M.B. Voloshin, V.I. Zakharov, Nucl. Phys. B 173 (1980) 189.
[13] J.A. Vermaseren, The symbolic manipulation program FORM, KEK-TH-326, http://www.nikhef.nl/form/
FORMdistribution/index.html.
[14] A. Denner, Fortschr. Phys. 41 (1993) 307.
[15] R. Mertig, M. Bohm, A. Denner, Comput. Phys. Commun. 64 (1991) 345.
[16] J.L. Hewett, hep-ph/9810316.
[17] A. Kanaki, PhD thesis, in preparation.
[18] F.A. Berends, C.G. Papadopoulos, R. Pittau, Comput. Phys. Commun. 136 (2001) 148, hep-ph/0011031.
[19] A. Kanaki, C.G. Papadopoulos, Comput. Phys. Commun. 132 (2000) 306, hep-ph/0002082;
A. Kanaki, C.G. Papadopoulos, HELAC-PHEGAS: automatic computation of helicity amplitudes and cross
sections, hep-ph/0012004.
[20] G.J. van Oldenborgh, Comput. Phys. Commun. 66 (1991) 1.
[21] K. Hagiwara, et al., Phys. Rev. D 66 (2002) 010001.
[22] G. Passarino, Nucl. Phys. B 578 (2000) 3, hep-ph/0001212.
[23] A. Sirlin, Phys. Rev. Lett. 67 (1991) 2127;
A. Sirlin, Phys. Lett. B 267 (1991) 240.

On a possible measurement of S from BB

correlations in Z 0 decay
A. Brandenburg a , P. Nason b , C. Oleari c
a DESY-Theorie, 22603 Hamburg, Germany
b INFN, Sezione di Milano, Piazza della Scienza 3, 20126 Milan, Italy
c Department of Physics, IPPP, South Road, Durham DH1 3LE, UK
Abstract
Motivated by recent preliminary results from the SLD Collaboration on the measurement of angledependent BB energy correlations in Z 0 bb events, we propose a class of observables that can
be computed as a power expansion in the strong coupling constant S , of order S at the Born level
and that can be used for a precision measurement of S (MZ ). We compute their next-to-leading
order O(S2 ) corrections in the strong coupling constant, including exactly quark-mass effects. We
show that, in the theoretical evaluation of these quantities, large logarithms of the ratio of the mass of
the final quark over the centre-of-mass energy cancel out. Thus, these variables have a well-behaved
perturbative expansion in S (MZ ). We study the theoretical uncertainties due to the renormalizationscale dependence and the quark-mass scheme and we address the question of which mass scheme is
more appropriate for these variables.
PACS: 12.38.Bx; 14.65.Ha
1. Introduction
Hadronic final states in e+ e annihilation have been intensively studied in recent years,
both at LEP and SLC, in order to perform QCD tests and to determine S [1]. Interesting
studies on bottom flavoured hadronic events have also been performed, in order to test the
flavour independence of the strong coupling constant and to search for running mass
effects [27]. However, up to now, heavy flavoured jets have been studied with the same
E-mail address: oleari@pheno.physics.wisc.edu (C. Oleari).
doi:10.1016/S0550-3213(03)00530-3
A. Brandenburg et al. / Nuclear Physics B 667 (2003) 394410
395
tools used to study light flavoured jets. For example, the same shape variables used for
light-quark jets (like thrust, cluster multiplicities, etc.) have been applied to heavy flavoured
events. On the other hand, heavy flavoured events may have some advantages, because
the kinematic distributions of the produced heavy quark can be computed theoretically
with much better accuracy than the corresponding light-quark quantities. Furthermore, the
kinematic of the final-state heavy flavoured hadron is more closely related to that of the
final-state quark, since hadronization effects involve energy scales that are much smaller
than the quark mass.
Recently, the SLD Collaboration [8] has published preliminary data on the double
inclusive BB production, and studied quantities (proposed in Ref. [9]) that carry
information about the angular dependence of the BB energy correlations.
The main purpose of the present work is to demonstrate that a slight variant of the
quantities introduced in Ref. [9] has suitable properties to perform a novel, heavy-flavourspecific determination of the strong coupling constant. We will demonstrate the following
facts.
The observables we are considering have a well-defined, finite expansion in the strong
coupling constant.
Non-perturbative corrections of order QCD /m, where m is the heavy-quark mass,
cancel in these variables. The only remaining non-perturbative corrections are of order
QCD /ECM , where ECM is the centre-of-mass energy.
Potentially large terms, enhanced by powers of log(ECM /m), cancel at all orders in
perturbation theory, provided S is evaluated at a scale near ECM . This cancellation
holds as long as one can neglect multiple heavy-flavour pair production in our process,
which is certainly the case for b production in Z 0 decays.
The quantities we are considering have a well-defined massless limit, provided one
can neglect multiple heavy-flavour pair production. This limitation is what makes these
quantities heavy-flavour specific. In fact, only for heavy flavours multiple pair production
is truly negligible. Thus, although these quantities have formally a well-defined massless
limit, they cannot be used in a straightforward way to study light-flavour production.
Tools to compute heavy-flavour production at the next-to-leading-order (NLO) level
have been available in the literature for the past few years [1014]. We have used our
calculations [1214] in order to compute the NLO corrections to the observables we
propose.
In Section 2, we define the variables we are considering and show that their perturbative
expansion in terms of the strong coupling constant S (MZ ) is well behaved, and free from
large logarithms of the ratio of the quark mass over the centre-of-mass energy. The SLD
Collaboration has provided us with (preliminary) data for these new variables [15], and
thus we have been able to check our NLO predictions against the data.
In Section 3 we investigate the massless limit of these variables, and show explicitly
that potentially large logarithms cancel out.
In Section 4 we study theoretical uncertainties of our results. We analyze the dependence
of the NLO results on the renormalization scale and compare the results obtained with the
pole-mass definition to those obtained using the MS mass, addressing the question of which
396
of the two mass definitions is more suitable for this calculation. Finally, in Section 5 we
give our conclusions.
In Appendix A, we collect some leading order (LO) formulae used in our analysis, and
in Appendix B we comment on the renormalization scheme used in this paper.
2. Double-inclusive heavy-quark cross section

Heavy flavoured hadron production proceeds through the process
e+ e Z 0 / Q + Q + X,
(where Q is a heavy quark of mass m) followed by the hadronization of the final heavy
quarks. In the following we will focus upon B meson production.
Defining the B hadron scaled energy xB = 2EB /ECM , we consider the following
quantities1

dB
1
Di =
dxB ,
(2.1)
xBi1
B
dxB

1
d 3 B
j 1
dxB1 dxB2 ,
x B2
xBi1
Dij () =
(2.2)
1
B
dxB1 dxB2 d cos
Dij ()
Gij () =
(2.3)
,
Di Dj
where dB /dxB is the differential cross section for the final-state hadron production,
is the angle between the two hadrons, B the total cross section and xB1/2 are the scaled
energies of the two final B hadrons.
The quantities studied by the authors of [9] are related to the ratios Gij /G11 by a
calculable perturbative factor (Pi Pj /P12 in the notation of [9]). In Ref. [8] these ratios
were compared to experimental results. However, these quantities have a few drawbacks.
First of all, they depend upon a theoretical factor. It would be preferable to deal with
quantities that have an experimental definition, free of theoretical assumptions. Second
and most important, they have a perturbative expansion that starts at leading order with a
constant. Therefore, even when corrected at the NLO level, they would allow only for a
LO determination of S . Conversely, they can be used to test the production mechanism,
since in the ratios S cancels at first approximation.
In this work, we claim that the quantities Gij , defined in Eq. (2.3), should instead be
studied in order to perform precision QCD tests. We will in fact show in the following that
these quantities have a perturbative expansion in S (ECM ) with coefficients that remain
finite even in the limit of large ECM /m ratios.
We write the hadronic differential cross sections as

dB
db B
(2.4)
=
D (x)(xB xxb ) dxb dx,
dxB
dxb NP
1 The quantities D and D were also considered in [9] and the ratios G in [16].
i
ij
ij
d 3 B
=
dxB1 dxB2 d cos
397
d 3 b
B
DB (x1 )DNP
(x2 )
dxb1 dxb2 d cos NP
(xB1 x1 xb1 )(xB2 x2 xb2 ) dxb1 dxb2 dx1 dx2 ,
(2.5)
where db , xb , xb1/2 are the quark differential cross section and the quark scaled energies.
The quark differential cross sections db are calculable order-by-order in perturbative
QCD, since the heavy-quark mass m acts as a cut-off for final-state collinear singularities.
However, in order to go from the QCD partonic cross section db /dxb to the hadronic one
dB /dxB , we have to take into consideration non-perturbative effects, of order QCD /m,
that are present in the hadronic cross section. We assume that all these effects are described
B .
by a non-perturbative fragmentation function DNP
The quark differential cross sections, although calculable, contain logarithmically
enhanced terms of the form Sk logl (m/ECM ), so that their perturbative expansion is not
well behaved for ECM m.
In the quantities Gij () these large logarithms cancel, together with the dependence
upon the non-perturbative fragmentation function. This is a consequence of the factorization theorem, if one makes the additional assumption that secondary production of heavyflavour pairs is negligible, which is in fact the case at the Z 0 peak.
The factorization theorem states that mass singularities in inclusive cross sections can
be absorbed into process independent, universal factors. In our case, mass singularities are
precisely the log(ECM /m) terms. Thus, for ECM m we have

db
d
(2.6)
=
Db (z)(xb zxb ) d x b dz,
dxb
d x b

d 3
d 3 b
=
Db (z1 )Db (z2 )
dxb1 dxb2 d cos
d xb1 d xb2 d cos
(xb1 z1 xb1 )(xb2 z2 xb2 ) d xb1 d xb2 dz1 dz2 ,
(2.7)
where the hatted cross sections do not depend upon m, and all terms that are large in
the m 0 limit (i.e., terms proportional to powers of log(ECM /m)) are absorbed in the
fragmentation functions Db (z).2 Inserting Eqs. (2.6) and (2.7) into Eqs. (2.4) and (2.5) and
considering the definitions in Eqs. (2.1) and (2.2), we get

d
1
i1 B
i1
Di =
(2.8)
x DNP (x) dx
z Db (z) dz
d x b ,
xbi1
B
d xb

j 1 B
B
Dij () =
x1i1 DNP
(x1 ) dx1
(x2 ) dx2
x2 DNP

j 1
i1
z2 Db (z2 ) dz2
z1 Db (z1 ) dz1
1
B
j 1
xb2
xbi1
1
d 3
d xb1 d x b2 ,
d xb1 d x b2 d cos
(2.9)
2 Notice that if we had not assumed a negligible secondary heavy-flavour pair production, we should have
introduced also the gluon and light-quark component of the heavy-quark fragmentation function.
398
and we see that, in the ratio defining Gij () (see Eq. (2.3)), the Db factors (that contain
mass singularities) cancel out, leaving a finite expression.3
Since non-perturbative effects cancel in the definition of Gij , from now on we will
make no distinction between the quantities Di , Dij defined at the quark and at the hadron
level. With the available codes for the calculation of NLO corrections in the production
of heavy quarks [13,14] at e+ e colliders,4 we can compute the contributions up to order
S2 to the quantities Di , Dij () and Gij (). We can write the perturbative expansions for
Di and Dij () for massive final-state quarks, renormalized in the pole-mass scheme, in
the region away from the two-jet final state (the two parton final state gives a contribution
proportional to ( )) as

S ()
+ O S2 ,
2

2

S ()
S () 2
+ Cij () + 2b0 Bij () log
Dij () = Bij ()
2
2
2
ECM
3
+ O S , for
= ,
Di = 1 ai
(2.10)
(2.11)
so that the expression for Gij becomes

S ()
Gij () = Bij ()
2

2

S () 2
+ Cij () + 2b0 log
+ ai + aj Bij ()
2
2
ECM
3
+ O S ,
(2.12)
where
11CA 4nf TF
,
(2.13)
12
is the renormalization scale, CA = Nc = 3 for an SU(3) gauge theory, TF = 1/2 and nf
is the number of active flavours.
We observe that expanding the ratio in powers of S is essential to implement the
cancellation of large logarithms in Gij order-by-order in the perturbative series. If we had
not done so, the numerator and denominator of Gij would contain large logarithms at some
fixed order, and their ratio would be misleading (i.e., it would be sensitive to higher order
terms with higher powers of logarithms).
In Fig. 1 we have plotted Gij at leading and next-to-leading order in S , for i, j =
1, 2, 3, at a renormalization scale = ECM = 91.2 GeV, using S (ECM ) = 0.12, with
nf = 5 active flavours. The NLO curves are in good agreement with the SLD preliminary
data.
b0 =
3 Of course, this holds if we neglect terms that are suppressed by powers of m. Terms suppressed by powers of
m and enhanced by powers of log(ECM /m) may still (and in fact are) present in Gij . Furthermore, incalculable
effects of order QCD /ECM may still be present.
4 A version of the program described in [13] was used which implements the dipole subtraction method for
massive partons [17] to combine real and virtual corrections.
399
Fig. 1. Leading order (dashed) and next-to-leading order (solid) curves for Gij , i, j = 1, 2, 3, at a renormalization
scale = ECM = 91.2 GeV, using S (ECM ) = 0.12. The error bars in the preliminary data from the SLD
Collaboration [15] are the quadratic sum of systematic and statistical errors.
Conversely, the SLD data can be used to extract the value of S . In fact, from the
knowledge of the numerical values of the coefficients of the S and S2 contributions in
Eq. (2.12), we can fit Gij () to the data, keeping S as a free parameter.
We have computed tables for the S and S2 coefficients of Eq. (2.12), that can be used in
a more detailed analysis. These tables (and the programs to generate them) can be obtained
from the authors.
2.1. Secondary b-quark and the running coupling
We should mention here that, in the calculations of Ref. [14], the b-quark is treated as a
heavy particle, and thus does not contribute to the running of the coupling constant. Thus,
in principle, one should use this calculation in conjunction with the 4-flavour coupling
constant S(4) . Since we are dealing with Z 0 decays, however, secondary b-quark pairs
400
Fig. 2. Secondary b-quark production.
may in fact be produced (from gluon splitting), and when this happens, they behave, in
some sense, as light flavours, contributing terms that effectively modify the running of S .
In Fig. 2, we show the Feynman graphs for secondary heavy-flavour production, giving
rise to four massive-quark final state. The graphs in Fig. 2 also appear with the two identical
quarks and/or the two identical antiquarks exchanged. When taking the amplitude squared,
this gives rise to two interference terms. Their contribution is known to be extremely small,
and will be neglected in the following.
In addition, virtual b-quark loops arise only in the self-energy insertion of the emitted
(on-shell) gluon. Their contribution is zero in the renormalization scheme we use, since,
in this scheme, the heavy flavour decouples for small momenta (see Appendix B for more
details). Thus, the only effect of the heavy flavour is given by the graphs of Fig. 2.
In the quantities we are considering, the possibility of detecting secondary (instead
of primary) B mesons has little impact, since secondary Bs are very soft. If we make
the approximation of neglecting effects that are suppressed by powers of the mass in the
secondary b line, the contribution of the graphs of Fig. 2 (without interference) is easily
included using the following prescription:
compute the cross section with the number of light flavours nlf equal to 5;
(5)
use the 5 flavour coupling S ;
do not include any other graph with secondary heavy flavours.
In essence, the prescription amounts to including the secondary heavy flavour by simply
increasing nlf by one. When doing this, one in fact is also changing the renormalization
scheme for the heavy flavour, by treating it as if it were light. Thus, the correct coupling
(5)
constant to use in this framework is S . In Appendix B we discuss this change of scheme
in more detail and from a slightly different point of view.
Secondary b-quark production may also arise from primary light quarks, connected
with the Z 0 / vertex. We neglect altogether this mechanism, since it generates only soft b
quarks.
3. The m 0 limit
As explained in Section 2, Gij () are free from potentially dangerous logarithms of the
ratio ECM /m. In order to investigate the stability of our expressions in the m 0 limit,
we introduce the integral of Gij over and we define (we drop the B/b subscript for ease
401
of notation)
1
rij =
j 1
Gij () d cos =
where
x1i1 x2
j 1
x1i1 x2
(3.1)

x1i1 = x2i1 Di ,
j 1
x1i1 x2
(3.2)
1
Dij () d cos .
(3.3)
To evaluate Eq. (3.1) up to order S2 and for i, j = 2, 3, we rewrite rij as follows

j 1
rij = 1 +
=1+
x1i1 x2
j 1
x1i1 x2
j 1
x1i1 x2
j 1
j 1
(1 x1i1 )(1 x2 ) 1 x1i1 1 x2
,
j 1
x1i1 x2
(3.4)
and we expand in S

S ()
+ O S2 ,
1 x1i1 = 1 x2i1 = ai
2

j 1
1 x1i1 1 x2

2

S ()
S () 2
+ cij + 2b0 bij log
= bij
+ O S3 ,
2
2
2
ECM
to obtain

2
S ()
+ cij + (ai + aj )bij ai aj + 2b0 bij log

2
2
ECM

S () 2
+ O S3 .
2
(3.5)
(3.6)
rij = 1 + bij
j 1
(3.7)
Note that, in Eq. (3.4), the expectation value of x1i1 and x2 should be known only
j 1
up to order S , and that, in the correlation term (1 x1i1 )(1 x2 ), the two-jet final
state gives zero contribution, since it is proportional to (1 x1 )(1 x2 ), at all orders
in S . This is the reason why we do not need to know the S2 two-loop virtual term for
e+ e QQ.
The coefficients ai and bij can be computed analytically. We have collected their
expressions in Appendix A. In Table 1, we give the numerical values for the coefficients ai ,
bij and cij , for a down-type quark with mass m = 0.01, 0.1, 1, 3, 5 and 10 GeV, at a centerof-mass energy ECM = = 91.2 GeV. Observe that, while the individual coefficients ai
and cij can be quite large, due to the presence of logarithmic terms, the ratio rij is well
402
Table 1
Results for the coefficients ai , bij and cij appearing in the definition of rij (see Eqs. (3.5)(3.7)). The mass m is
in GeV and = ECM = 91.2 GeV
a2
a3
b22
b32
b33
c22
c32
c33
m = 0.01
m = 0.1
m=1
m=3
m=5
m = 10
29.1611
46.6475
0.666666
1.13333
1.91111
820.1(3)
1290.5(4)
2027.5(5)
20.9741
33.8553
0.666589
1.13321
1.91093
420.51(4)
663.99(7)
1046.5(1)
12.7866
21.0574
0.661895
1.12569
1.89945
155.14(1)
247.19(2)
393.09(3)
8.87675
14.9209
0.636326
1.08448
1.83563
75.787(7)
122.15(1)
196.60(2)
7.05412
12.0317
0.598452
1.02307
1.73915
49.176(3)
80.053(6)
130.23(1)
4.56726
8.01007
0.478637
0.826778
1.42415
22.943(1)
38.268(3)
63.871(5)
Table 2
Results for rij . The mass m is in GeV and = ECM = 91.2 GeV
r22 1
r32 1
r33 1
0.01
0.106103 S + 0.215(4) S2
0.180375 S + 0.403(6) S2
0.304162 S + 0.750(9) S2
0.1
0.106091 S + 0.217(1) S2
0.180356 S + 0.407(2) S2
0.304133 S + 0.755(3) S2
0.105344 S + 0.2175(4) S2
0.179159 S + 0.4066(7) S2
0.302307 S + 0.7524(1) S2
0.101274 S + 0.2100(1) S2
0.292149 S + 0.7287(5) S2
0.0952466 S + 0.19905(9) S2
0.0761774 S + 0.16351(4) S2
0.172600 S + 0.3930(3) S2
0.162827 S + 0.3725(2) S2
0.131586 S + 0.30605(7) S2
0.226661 S + 0.5706(1) S2
10
0.276794 S + 0.6919(3) S2
behaved, as illustrated in Table 2. This is due to the fact that the large logarithms present
in the single coefficients cancel out in the ratio rij .
4. Theoretical uncertainties
In this section, we investigate the theoretical uncertainties related to the renormalization
scale and mass scheme used in the calculation.
The renormalization-scale dependence of rij is illustrated in Table 3 for the case
m = 5 GeV, using the two-loop evolution of S with S = 0.12 for = ECM = 91.2 GeV.
For ECM /2 < < 2ECM the variation of (rij 1) around the central value = ECM is
roughly 10% at leading order. It is reduced to 5% at NLO.
Table 3
Results for rij for different renormalization scales with S (ECM ) = 0.12, ECM = 91.2 GeV and m = 5 GeV
= ECM /2
r22 1
r32 1
r33 1
= ECM
= 2ECM
LO
NLO
LO
NLO
LO
NLO
0.0127974
0.0218776
0.0371903
0.014937(2)
0.026116(3)
0.045455(5)
0.0114296
0.0195393
0.0332153
0.014296(1)
0.024903(2)
0.043179(4)
0.0103320
0.0176629
0.0300257
0.013622(1)
0.023667(2)
0.040922(3)
403
In order to investigate the mass-scheme dependence, we now explicitly distinguish

between the pole mass m and the MS mass m() (also called running mass). Using the
relation between m and m()

2
2
m()
3
S ()
CF 2 log
+
O
m = m()
1+
(4.1)
S ,
2
2
2
and defining

m 2
,
z=
ECM

z () =
m()
ECM
2
,
(4.2)
we have

rij (z) rij z () + ,z()

S () 2
= rij (z()) +
,bij z () + O S3 ,
2
with
(4.3)

E2
dbij
3
,bij (z()) = 4CF z () 1 ln z () CM
z () .
2
4
dz
(4.4)
We have evaluated Eq. (4.3) for m(E

CM ) = 3 GeV using the analytic expressions for
CM ) = 3 GeV corresponds
the coefficients bij collected in Appendix A. The value m(E
roughly to what one gets by converting m at the b-mass scale to m,
and then evolving m
up
to = ECM . We have further studied the dependence of Eq. (4.3) on the renormalization
scale . We have used the 2-loop evolution of m(),
which gives m(E

CM /2) = 3.20 GeV
rij 1) are listed in Table 4. Using the
and m(2E
CM ) = 2.83 GeV. The results for (

MS definition for the mass, the theoretical uncertainties due to the scale ambiguity are
practically of the same size as in the pole-mass scheme.
Comparing the results for rij obtained with an on-shell mass m = 5 GeV with those
obtained with an MS mass m(E
CM ) = 3 GeV, we found that, at leading order, the
differences between (rij 1) using the two different mass definitions amount to 56%.
This theoretical uncertainty is reduced to less than 4% at NLO.
An interesting question can be raised about the use of the pole mass versus the MS
mass, that is to say, which one is more natural in this context. In practice, this is a very
difficult question to answer. In inclusive processes, in general, the MS mass is better, since
Table 4
CM ) =
Results for rij for different renormalization scales with S (ECM ) = 0.12, ECM = 91.2 GeV and m(E
3 GeV
= ECM /2
r22 1
r32 1
r33 1
= ECM
= 2ECM
LO
NLO
LO
NLO
LO
NLO
0.0135360
0.0230755
0.0390735
0.015444(2)
0.026986(4)
0.046904(8)
0.0121529
0.0207120
0.0350579
0.014884(2)
0.025897(4)
0.044811(7)
0.0110323
0.0187981
0.0318083
0.014265(2)
0.024743(3)
0.042675(5)
404
Fig. 3. The quantities Xij , defined in Eq. (4.5), plotted as a function of m, both in the pole-mass scheme and in
the MS-mass scheme, with ECM = = 91.2 GeV.
it accounts for large powers of logarithms of m/ECM multiplying powers of the mass at
all orders in perturbation theory. The case we are considering, however, is not an inclusive
process, since we are weighting the cross section with final-state kinematic variables. We
are thus unable to give a full answer to this question. However, we can certainly ask
whether, at the order we are considering, there is some numerical indication that using
the MS mass accounts at least for a large part of the logarithmic terms in our expression.
(2)
The mass dependence of the coefficients of rij is reported in Table 2. Calling rij the
coefficient of S2 in rij , we consider the expression
Xij (z) =
rij(2) (z) rij(2) (0)
(4.5)
as a function of log z. The same expression in the MS scheme is given by

(2)
X ij (z) =
(2)
rij (z) rij (0)

z

dbij
4
3
log
z
(z),
C
1
F
2
(2)
4
dz
(4.6)
where we use = ECM , and z = z (). If the slope of the second expression versus log m
is smaller than the slope of the first one, we can take this as an indication of the fact that it
is more appropriate to use the MS mass instead of the pole mass. The results are displayed
405
(2)
in Fig. 3. The value rij (0) has been taken to correspond to m = 0.1 GeV in Table 2 (the
point m = 0.01 GeV has been excluded, because of the large errors). The points at m = 1,
3, 5 and 10 GeV are then plotted. It is quite clear from the figure that the logarithmic slope
is smaller for the pole-mass scheme. Furthermore, the radiative correction to the mass term
is always smaller for the pole-mass scheme than for the MS scheme. We thus conclude that
the quantities rij seem to be more naturally described in the pole-mass scheme rather than
in the MS-mass scheme.
5. Conclusions
In this paper we have investigated new measurable quantities (related to the quantities
introduced in Ref. [9]) defined in e+ e annihilation, that carry information about the
angle-dependent BB energy correlation. We have shown that these variables have a wellbehaved expansion in terms of the strong coupling constant S , since large logarithms in
the ratio of the mass of the final quarks over the centre-of-mass-energy cancel out order by
order, together with non-perturbative effects of order QCD /m.
We have compared the computed NLO results with the available preliminary data from
the SLD Collaboration [15], and found good agreement. In addition, we have generated
tables5 that can be used in a fit to the data, in order to extract the strong coupling constant
S at next-to-leading order.
We have investigated the renormalization-scale and mass-scheme dependence of our
results, and found that, varying the renormalization scale in the range ECM /2 < <
2ECM , the scale dependence of (rij 1) is reduced from roughly 10% at leading order
to 5% at next-to-leading order. The theoretical uncertainties related to the mass definition
are less than 4% at NLO. We have shown that these quantities seem to be more naturally
described in the pole-mass rather than in the MS-mass scheme.
From the discussion given in this paper, one may wonder if similar quantities may be
defined for massless quarks, given the fact that mass singularities cancel in the variables
we are considering, and thus they remain well defined in the massless limit. This is not
quite the case. In fact, it is crucial for our discussion that the secondary heavy-flavour
production may be neglected. If this were not the case, evolution of the fragmentation
function would become multi-dimensional, involving several flavour components and a
singlet contribution. Thus, the advantage of having a heavy flavour in the final state lies in
the fact that the evolution becomes essentially non-singlet, so that we are left with a single
component of the fragmentation function, that simply cancels in the ratios we propose.
Acknowledgements
We would like to thank P.N. Burrows and G. Nesom for numerous discussions and for
making available to us experimental data on the BB correlations. Thanks also to P. Uwer
5 The programs and the tables can be obtained upon request from the authors.
406
for discussions and comments on the manuscript. The research of A.B. was supported by
a Heisenberg grant of the DFG and C.O. thanks the UK Particle Physics and Astronomy
Research Council (PPARC) for supporting his research.
Appendix A. Analytic results for a2 , a3 , b22 , b32, b33

In this appendix, we collect the analytic results for the coefficients ai and bij , i, j = 2, 3,
defined in Eqs. (3.5) and (3.6). Calling the generic coefficient F (z), where z = m2 /s and
2 , we write
s = ECM
F (z) =
where =
given by

2CF
cV F V (z) + cA F A (z) ,
[cV (1 + 2z) + cA (1 4z)]
(A.1)
1 4z and = (1 )/(1 + ). The electroweak couplings cV and cA are

2

cV = Q2b f + 2gVb Qb Re (s) f Z + gVb2 (s) f ZZ ,

2
b2
cA = gA
(s) f ZZ ,
(A.2)
with
f = 1 + ,

e2
e
2( + )gVe gA
f ZZ = (1 + ) gVe2 + gA
,
e
.
f Z = (1 + )gVe + ( + )gA
(A.3)
f
Here, Qb = 1/3 is the electric charge of the bottom quark, and gA/V denote the axial and
e = 1 for an
vector couplings of the fermion f . In particular, gVe = 12 + 2 sin2 W , gA
2
1
2
1
b
2
b
electron, and gV = 2 + 3 sin W , gA = 2 for a bottom quark, where W is the weak
mixing angle. The function (s) reads
(s) =
s
1
,
4 sin2 W cos2 W s MZ2 + iMZ Z
(A.4)
where MZ and Z stand for the mass and the width of the Z boson. Finally, (+ )
denotes the longitudinal polarization of the electron (positron) beam.
gives
A simple calculation using the Born differential cross section for e+ e b bg

14 3
2
11 7
V
2
a2 (z) = + z z ln() + + z + 7z
(A.5)
,
3
3
3
6
3

a2A (z) a2V (z)
20
8
=
10z + 12z2 + z3 ln()
z
3
3

35 59
(A.6)
z 10z2
,
+
3
3
3

5
25 10
32
11
a3V (z) = + z + z2 z3 z4 ln()
24
3
2
3
12

779 2
433 137
+
z+
z + 11z3
,
+
12
2
6
24

a3A (z) a3V (z)
55
40 3 15 4
2
=
20z + 21z + z + z ln()
z
12
3
2

823 845
175 2
3
+
z
z 15z
,
36
18
6
4

1
1 19
V
+ z z2 z3 ,
b22
(z) = 2z 1 + z3 ln() +
4
6
6
A
V

b22 (z) b22(z)
= 2z 2 + 3z 5z3 ln()
z

5 2
3 17
3
+ z + z + 5z ,
4
6
6

19 1
17 3 1 4
V
2
+ z 2z + z + z ln()
b32 (z) = z
6
3
3
3

47 2 103 3 1 4
17 1721
+
z+ z
z z
,
+
20
180
45
18
3
2
407
(A.7)
(A.8)
(A.9)
(A.10)
(A.11)
A (z) b V (z)

b32
32
= z 7 + 10z 10z3 14z4 ln()
z

16 2 37 3
27 287
4
(A.12)
z + z + z + 7z ,
+
20
60
15
6

14 13
32
16
V
z 8z2 + z3 + z4 z5 ln()
(z) = z
b33
3
6
3
3

221 2 1037 3 31 4
43 533
5
+
z
z
z z +z
,
+
(A.13)
30
45
36
90
6
2

A (z) b V (z)
b33
4
70
33
= z 12 + 17z + z2 12z3 28z4 z5 ln()
z
3
3

53 2 157 3 287 4
109 661
z+ z +
z +
z + 35z5
.
+
15
30
20
6
6
3
(A.14)
In the m 0 limit, we have 1, z = m2 /s 0, so that

11
2
a2 = 2CF ln(z)
,
3
9

433
25
a3 = 2CF ln(z)
,
24
288
1
b22 = 2CF ,
4
408
17
,
40
43
b33 = 2CF .
(A.15)
60
This behavior is well illustrated from the collections of values in Table 1: while the bij are
finite in the m 0 limit, the coefficients ai diverge as log(m/ECM ).
b32 = 2CF
Appendix B. Renormalization schemes

Our calculation was carried out in the mixed renormalization scheme of Ref. [18],
in which the light flavours nlf are subtracted in the MS scheme, while the heavy-flavour
loops are subtracted at zero momentum. In this scheme the heavy flavour decouples at
low energy. In fact, convergent heavy-flavour loops are suppressed by powers of the mass
of the heavy flavour. The only unsuppressed contributions come from divergent graphs.
But those are subtracted at zero external momenta, so their contribution is removed by
renormalization for small momenta. In the mixed scheme, charge renormalization is given
by the prescription

2 7
2
TF
1 (nlf )
b
27
+ O S ,
S = S () 1 S () b0
(B.1)
7
m2 3
where Sb is the bare coupling constant, and S () is the mixed-scheme coupling constant
at the scale . We have defined
(nlf )
b0
11CA 4TF nlf

,
12
(B.2)
and
1 1
= + log(4) E ,
(B.3)
7 7
where dimensional regularization is used, with the number of spacetime dimensions equal
to 4 27.
In the MS scheme the renormalization prescription is

1 (nf )
b
27
S = S () 1 S () b0
(B.4)
,
7
where nf = nlf + 1, and S () is the MS coupling constant at the scale . The
couplings in the two schemes obey a 4-flavour and 5-flavour renormalization-group
equation respectively
d
(n )
S () = b0 lf S2 (),
2
d log
d
(n )
S () = b0 f S2 (),
d log 2
(B.5)
409
that can be easily derived by imposing the constancy of the bare coupling Sb under a
renormalization group transformation. Combining Eqs. (B.1) and (B.4) we have
2

TF
log
S2 () + O S3 ,
S () = S ()
(B.6)
2
3
m
which is the standard MS matching condition for flavour crossing.
In the heavy flavour production calculation we are considering, if we express the result
(usually given in terms of S ) in terms of S , an extra term arises, equal to
2
TF
Born S ()
(B.7)
log
.
3
m2
The only other diagrams where secondary heavy flavours appear are the graphs with a
gluon splitting into a heavy-flavour pair. It can be shown that, if we neglect powers of
m/ECM , the term in Eq. (B.7) is exactly the term one needs to turn the calculation of
the heavy-flavour splitting, which is regulated in the collinear limit by the heavy-flavour
mass, into the corresponding MS subtracted calculation, provided that interference terms
of the heavy flavours coming from gluon splitting with the primary heavy flavours are fully
neglected.
Eq. (B.7) follows immediately from the theory of heavy-flavour fragmentation functions. According to Ref. [19], Eq. (3.16), the cross section for the inclusive production of a
heavy quark via gluon splitting, with a fraction x of the energy of the gluon, is (at leading
order and neglecting powers of m/ECM )
2

d
S TF 2
d
= 0
+ O S2 ,
+
(B.8)
x + (1 x)2 log
2
dx
2
dx
m
where 0 is the cross section for the production of the gluon and stands for the massless
limit, MS subtracted cross section for the gluon splitting process. Integrating both sides in
x one finds
2
S TF
= 0
(B.9)
log
.
3
m2
The second term on the right-hand side is precisely the contribution of Eq. (B.7).
References
[1] I. Hinchliffe, Quantum chromodynamics, Phys. Rev. D 66 (2002) 010001.
[2] ALEPH Collaboration, R. Barate, et al., A measurement of the b-quark mass from hadronic Z 0 decays, Eur.
Phys. J. C 18 (2000) 113, hep-ex/0008013.
[3] DELPHI Collaboration, P. Abreu, et al., mb at MZ , Phys. Lett. B 418 (1998) 430442.
[4] OPAL Collaboration, G. Abbiendi, et al., Test of the flavour independence of s using next-to-leading order
calculations for heavy quarks, Eur. Phys. J. C 11 (1999) 643659, hep-ex/9904013.
[5] OPAL Collaboration, G. Abbiendi, et al., Determination of the b-quark mass at the Z mass scale, Eur. Phys.
J. C 21 (2001) 411422, hep-ex/0105046.
[6] SLD Collaboration, K. Abe, et al., An improved test of the flavor independence of strong interactions, Phys.
Rev. D 59 (1999) 012002, hep-ex/9805023.
410
[7] A. Brandenburg, P.N. Burrows, D. Muller, N. Oishi, P. Uwer, Measurement of the running b-quark mass
events, Phys. Lett. B 468 (1999) 168177, hep-ph/9905495.
using e+ e bbg
[8] SLD Collaboration, K. Abe, et al., Measurement of the double-inclusive bb quark fragmentation function
in Z 0 decays and first measurement of angle dependent BB energy correlations, Contributed to 31st
International Conference on High Energy Physics (ICHEP 2002), Amsterdam, The Netherlands, 2431 July
2002.
[9] P.N. Burrows, V. Del Duca, P. Hoyer, Heavy-quark correlations in e+ e annihilations, Z. Phys. C 53 (1992)
149156.
[10] G. Rodrigo, A. Santamaria, M.S. Bilenky, Do the quark masses run? Extracting m
b (MZ ) from LEP data,
Phys. Rev. Lett. 79 (1997) 193196, hep-ph/9703358.
[11] G. Rodrigo, M.S. Bilenky, A. Santamaria, Quark-mass effects for jet production in e+ e collisions at the
next-to-leading order: results and applications, Nucl. Phys. B 554 (1999) 257297, hep-ph/9905276.
[12] W. Bernreuther, A. Brandenburg, P. Uwer, Next-to-leading order QCD corrections to three-jet cross sections
with massive quarks, Phys. Rev. Lett. 79 (1997) 189192, hep-ph/9703305.
[13] A. Brandenburg, P. Uwer, Next-to-leading order QCD corrections and massive quarks in e+ e 3 jets,
Nucl. Phys. B 515 (1998) 279320, hep-ph/9708350.
[14] P. Nason, C. Oleari, Next-to-leading-order corrections to the production of heavy-flavour jets in e+ e
collisions, Nucl. Phys. B 521 (1998) 237273, hep-ph/9709360.
[15] P.N. Burrows, G. Nesom, private communication.
[16] P.N. Burrows, private communication.
[17] S. Catani, S. Dittmaier, M.H. Seymour, Z. Trocsanyi, The dipole formalism for next-to-leading order QCD
calculations with massive partons, Nucl. Phys. B 627 (2002) 189265, hep-ph/0201036.
[18] J.C. Collins, F. Wilczek, A. Zee, Low-energy manifestations of heavy particles: application to the neutral
current, Phys. Rev. D 18 (1978) 242.
[19] B. Mele, P. Nason, The fragmentation function for heavy quarks in QCD, Nucl. Phys. B 361 (1991) 626644.
Nuclear Physics B 667 [PM] (2003) 413434

The group approach to AdS space propagators

Thorsten Leonhardt, Ruben Manvelyan 1 , Werner Rhl
Department of Physics, Erwin Schrdinger Strae, University of Kaiserslautern, Postfach 3049,
67653 Kaiserslautern, Germany
Received 6 June 2003; accepted 15 July 2003
Abstract
We show that AdS two-point functions can be obtained by connecting two points in the interior of
AdS space with one point on its boundary by a dual pair of Dobrevs boundary-to-bulk intertwiners
and integrating over the boundary point.
PACS: 11.10.-z
1. Introduction
Though AdS field theory is a classical subject in field theory the appearance of the
AdS/CFT correspondence [1] has revived interest in this subject considerably. In most
investigations of AdS fields it is tacitly assumed that such a theory is based on a Lagrangian
with interactions which are treated by a perturbative expansion. In a seminal paper
Fronsdal [2] showed that massless higher spin (tensor) fields can be defined and possess
a vanishing double trace. In a long series of articles M.A. Vasiliev [3] has studied an
interacting theory of infinitely many tensor fields of all ranks, which is invariant under a
generalized gauge symmetry and perturbative with respect to a small curvature parameter.
These interactions, including gravity, are only possible on AdS space.
The AdS/CFT correspondence maps an AdS field theory with higher spin gauge fields
holographically on a conformal field theory on Minkowski space containing a tensor
E-mail addresses: tleon@physik.uni-kl.de (T. Leonhardt), manvel@physik.uni-kl.de,

manvel@moon.yerphi.am (R. Manvelyan), ruehl@physik.uni-kl.de (W. Rhl).
1 On leave from Yerevan Physics Institute.
doi:10.1016/j.nuclphysb.2003.07.007
414
T. Leonhardt et al. / Nuclear Physics B 667 [PM] (2003) 413434
current source for each gauge field. Several such conformal field theories are known: the
critical O(N) sigma model at large N in d spacetime dimensions, where 2 < d < 4, or the
N = 4 supersymmetric YangMills theory in 4 dimensions and small t Hooft coupling .
The first class of models has been proposed as a candidate for such a correspondence by
Klebanov and Polyakov [4]. In these models all rank l currents have been constructed
by operator product expansion and their anomalous dimensions were calculated in [5,6],
with the result that they are all nonzero for tensor rank l = 2. In addition, these models
possess a scalar field of approximate conformal dimension 2, independent of d, which
is termed auxiliary or Lagrangian multiplier field. Its 3- and 4-point functions are
known [5,6] and serve as a source of detailed information such as the coupling constants
between two scalar fields and a current [7]. As a whole they offer themselves as a test
object for the AdS/CFT correspondence as proposed by Klebanov and Polyakov. To
construct an AdS field candidate for this purpose and to test the correspondence one has
to compute the respective 3- and 4-point functions. As a first step in this direction we have
developed an algorithm for the derivation of two-point functions in the AdS theory (bulkto-bulk propagators) for all traceless symmetric tensor fields. In a Lagrangian setting they
represent only one irreducible component of a tensor field. However, whether the AdS
field theory is Lagrangian or, such as conformal field theory, based on a set of fundamental
fields, a skeleton expansion and representation theory need not be answered at the start
of the investigation. With this article we hope to put representation theory in the right
position.
Propagators of symmetric tensor fields in AdS are known for the ranks l = 0 [8],
l = 1 [9,10] and l = 2 [1012]. They have been derived from field equations and the
requirement of a specific asymptotic behaviour on the boundary of AdS space. The l = 2
field has a nonvanishing trace. Our approach is based on representation theory and the
use of intertwiners constructed by Dobrev [13], which are bulk-to-boundary propagators.
It applies to all kinds of tensor fields, but for comparison we have treated the cases of
symmetric traceless tensors of ranks l {0, 1, 2} explicitly.
The basic notions of representation theory are presented in Section 2. In Section 3 we
evaluate the convolution of two scalar bulk-to-boundary propagators by integration over
the boundary. It is the crucial constructive element for any bulk-to-bulk propagator. These
integrals are first presented in the form of Appells F4 functions of two variables. However,
as proved by Allen and Jacobson [9], there must be a representation of these functions
in terms of the geodesic distance alone, thereby replacing the F4 function by functions
of a single variable. This is proved to be possible indeed, the resulting functions are
Legendre functions of the second kind, which are two-parameter Gaussian hypergeometric
functions. The propagators are then expressed in terms of rank l monomials of basic
bitensors. In Appendix A the technical problem of extracting traces from these monomials
of bitensors is studied. Due to the small number of such bitensors the extraction leads
to an overdetermined system of linear equations (for l 4), but the excessive equations
can be shown to be linearly dependent on the others, thereby resulting in a unique
solution.
415
2. The setting
2.1. Euclidean conformal field theory
We consider a Euclidean conformal field theory in d space(time) dimensions. The
isometry group is then given by the conformal group G = SO(d + 1, 1). The fields in
this theory are characterized by their transformation behaviour given by a representation
of G, which is induced from the subgroups of Euclidean rotations M = SO(d) and the
behaviour under dilatations, i.e.,
(
x ) r (r x ),
r R+ ,
(
x ) D (m)(m
x ),
m M.
(1)
D (m)
is the representation matrix

Here, is the conformal dimension of the field and
of m M. This determines the representation of G, thus we will write = [, ].
Moreover, let us denote by = [ , d ] the quantum numbers of the shadow field,
which is obtained by exchanging the M representation by its mirror image and the
conformal weight by d . Since the dimension of the shadow field appears very
often, it is convenient to introduce the notation
:= d .
(2)
For generic conformal dimension , the representations and are equivalent, i.e., there
exists an invertible operator
G : C C ,
(3)
which maps the representation spaces on each other and which commutes with the action
of G. Such an operator is called intertwining operator for C and C . It is well known
that the two-point function of a conformal field with quantum numbers agrees with this
intertwiner G . The functional form of a two-point function with quantum numbers
itself is fixed by conformal covariance to

x) ,
G (
x ) = 2 D r(
(
x )

2xi xj
r(
x) =
(4)
ij , i, j = 1, . . . , d.
x 2
In this formula is a normalization constant, whose value is not important for our
purposes. For later use, let us write down the explicit form of a propagator of a symmetric
traceless tensor field t (l) of conformal dimension . In this case we can write down a
generating function:
1

2 l (l)
(l)
t (
x )t (0) a l bl = 1 2 a , r(

x )b + 2 a 2 b2
x
( )
l0
( )
( ) ( )
( )( )

a , r(
x )b)
l C 1 (
,
l |
a ||b|
l

(|
a ||b|)
l0
(5)
416
1
where a , b Rd and Cl
is the Gegenbauer polynomial of degree l with parameter
1. Here we use the convenient abbreviation = d/2, which will be applied throughout
this work.
2.2. AdS/CFT correspondence as representation equivalence

On the other hand, we consider a field theory on a (d + 1)-dimensional Anti-de-Sitter
space, which is given by a classical action. The AdS/CFT correspondence connects this
field theory with a theory living on the boundary of the AdS space. The isometry group of
(d + 1)-dimensional AdS space is G = SO(d + 1, 1), thus matching the isometry groups of
both theories implies that the boundary theory must be conformal. By an identification of
generating functionals we obtain a prescription for constructing d-dimensional correlation
functions, which in turn define a conformal field theory. The procedure may be sketched as
follows: we take a set of vertices in the AdS theory, connect some of them by bulk-to-bulk
propagators and the remaining vertices with the boundary by bulk-to-boundary propagators
[14]. These two kinds of propagators can be obtained by solving the free equations of
motion derived from the action of the AdS theory. Let us write down the actual form of
the scalar bulk-to-boundary propagator, which constructs an AdS scalar field of mass m
from a conformal scalar of dimension . We skip the normalization constants, which can
be found in [15]. The mass m is linked to by m2 = ( d):

x10
,
K (x1 , x2 ) =
(6)
2
x12
where x1 = (x10 , x1 ) R+ Rd , x2 Rd , and we introduced the notation

2

2
2
2
= x1 x2 = x10
+ x1 x2 .
x12
(7)
Moreover, we note the vectorial bulk-to-boundary propagator, which transfers a ddimensional vector field of conformal dimension to a (d + 1)-dimensional AdS-vector
field of mass m

x12 )
x 1 2a, x12(b,
(1)
x1 , x2 ) = 10

(a, b;
(
a
,
b)
K
(8)
2 )
2
(x12
x12
with the relation
m2 = ( d) + (d 1).
(9)
In (8) we contracted the external indices of the (d + 1)- and d-dimensional vectors
with a Rd+1 and b Rd , respectively, via the (d + 1)- and d-dimensional standard
scalar products. The construction of the bulk-to-boundary propagator, which connects a
symmetric traceless second rank tensor field of conformal dimension with a symmetric
traceless second rank tensor field on (d + 1)-dimensional AdS is simple and can easily be
generalized to any rank:
2

(2)
x1 , x2 ) = K (1) (a, b;
x1, x2 ) 2 b trace with respect to b
K (a, b;
/2
d
2 2
2
2

x10
x12 , b)
x12
b 2a, x12
2a, x12(

= 2
(10)
(
a , b)
a
,
2
2
d
(x12)
x12
x12
417
where we contracted the external indices with a Rd+1 and b Rd , as in (8). After a quick
computation one agrees that the trace of this bulk-to-boundary propagator with respect to
a vanishes, too.
As shown by Dobrev [13], these bulk-to-boundary propagators can be given an
interpretation as intertwining operators, and they can even be constructed by this property.
They map the conformal representations C to representations C induced from the
maximal compact subgroup K = SO(d + 1) G, where is an irreducible representation
of K containing the irrep from = [, ] of M. Now C is neither uniquely determined
(there are infinitely many different irreps of K containing ) nor irreducible. The lack
of uniqueness can be cured by the choice of some minimal irrep of K. To obtain
irreducibility, one has to impose a constraint on the behaviour of the functions in C for
x0 0. The resulting representation C turns out to be irreducible for generic , and the
bulk-to-boundary propagators are then the integral kernels of the intertwiners
K : C C .
(11)
Dobrevs group-theoretical arguments apply for all elementary irreps of G and all irreps
induced from K, therefore, they do not depend on any actions or field equations.
After all these preliminaries, let us describe the idea of constructing AdS two-point
functions, where we restrict on symmetric tensor representations of M and K, so that
the respective propagators are labeled by their tensor rank l and the parameter . It is
simple to convolute the conformal intertwiner G(l)
with the bulk-to-boundary propagator
(l)
K
and check that we obtain a bulk-to-boundary propagator for a conformal field of
dimension d .2 This begs the following question: if we take a further bulk-to-boundary
(l)
propagator K of dimension and convolute it with the remaining d-dimensional leg
(l)
, do we obtain a scalar bulk-to-bulk propagator W(l) ? We show by
of the resulting Kd
explicit construction that this guess is almost correct, except that a certain doubling occurs.
(l)
Each two-point function W in AdS obtained this way comes along with its shadow
(l)
. This seems natural from the conformal point of view, since in any CFT
partner Wd
exchange the shadow of any exchanged field appears, because the shadow representations
are equivalent to the original ones.
2.3. Geometric properties of AdS propagators
Let us mention the following geometric aspects of the two-point functions. The twopoint function t (l) (x)t (l) (x ) transforms under a coordinate transformation as a tensor
of rank l at the points x and x , i.e., the two-point function is a bitensor. Thanks to
the maximal symmetry of the (Euclidean) AdS space we have the theorem [9] that every
bitensor can be expressed in terms of the metric and three fundamental bitensors, which
are obtained by differentiating the geodesic distance (x, x ) of two points x, x AdS.
The first two fundamental bitensors are the two tangent unit vectors along the geodesic
n = D (x, x )
and n = D (x, x ),
2 This is already proved in general in [13].
(12)
418
where the unprimed/primed indices denote tangent space indices at x and x , respectively,
and D denotes covariant derivatives. The third fundamental bitensor is the parallel
transporter along the geodesic connecting x and x . We choose to work in Poincar
coordinates x = (x0 , x1 , . . . , xd ) = (x0 , x ) R+ Rd for the AdS space, in which the
geodesic distance and the fundamental bitensors are expressed as
x 2 + x0 2 + (

x x )2
,
= cosh (x, x ) = 0

2x0 x0

.
n =
,
g, = +
+1
2 1
(13)
We note that the tangent vectors are proportional to the first derivative of and the parallel
transporter is essentially given by two derivatives of , one with respect to each variable.
Moreover, it turns out to be convenient to contract these bitensors with tangent vectors
a Tx AdS, c Tx AdS; we denote this by the (d + 1)-dimensional scalar product .
Then we arrive at the following algebraic basis of maximally symmetric bitensors
I1 := a, c, ,
I2c := c, ,
I2a := a, ,
and a 2 =
d

a2 ,
c2 =
=0
d

c2
I2 := I2a I2b ,
for l 2.
(14)
=0
In our construction we have to perform the splitting (a0 , a1 , . . . , ad ) = (a0 , a ) and the
same for c. The invariants then acquire the form
(
a , c) a0 c0
c0
a0
a, c, ,
x0 x0
x0 x0
x0
x0

1

a, = (
a , ) + a0
,
x
x0
0
1
c, = (
c, ) + c0
.
x0 x0
a, c, =
(15)
(16)
(17)
We are interested in propagators for symmetric traceless tensors of rank l. In order to

subtract traces we need further bitensors
I3 :=
a2 2
c2 2
I
+
I2a ,
2c
x02
x0 2
I4 :=
a 2 c2
x02 x0 2
(18)
Since taking traces on products of invariant bitensors (14) results in products of invariants
with at least one factor I3 or I4 , we call the latter ones trace terms. For a symmetric
traceless tensor of rank l we obtain a basis labelled by the pairs {l1 , l2 } with l1 + l2 = l:

(l1 ,l2 )
Am
I m1 I2m2 I3n1 I4n2 ,
I1l1 I2l2
(19)
1 m2 n1 n2 1
where the sum is restricted to
m1 + m2 + 2(n1 + n2 ) = l,
n1 + n2 > 0.
(20)
419
(l ,l )
The coefficients Am11 m22 n1 n2 can be determined by requiring (19) to be harmonic with

respect to the (naive) Laplacian a =
a a . The solution of this removing of traces
will be presented in the appendix.
3. Convolution integrals as hypergeometric functions

Now we do the actual computations. First we calculate the integral

1
2

x10
x30
d
A1 ,2 (x1 , x3 ) := d x2
,
2 + |
2 + |
x10
x12|2
x30
x32|2
(21)
where 1 , 2 are two real parameters, which are chosen in such a way to ensure
convergence of the integral but are independent otherwise. After introducing a Feynman
parameter for the two denominators the resulting d-dimensional integral is easy and we get
A1 ,2 (x1 , x3 )
0(1 + 2 ) 1 2
=
x x
0(1 )0(2 ) 10 30
1
2
2
2 (1 +2 )
x13
dt t 1 1 (1 t)2 1 t (1 t)
+ tx10
+ (1 t)x30
.
(22)
0
2 + (1 t)x 2 out of [ ] in (22) and choose the coordinates x , x in
Now we extract tx10
1 3
30
such a way that the resulting expression in [ ] can be expanded in a binomial series.
The case of arbitrary values of the coordinates are afterwards obtained by analytical
continuation. Then the Feynman parameter can be integrated and we obtain
A1 ,2 (x1 , x3 ) =
0(1 + 2 ) 1 d212 (1 + 2 )k
x10 x30
0(1 )0(2 )
k!
k0
0(1 + k)0(2 + k) k
F
0(1 + 2 + 2k)
1 +2 +k,1 +k
1 +2 +2k
;1

,
(23)
where we introduced the abbreviations

:=
2
x13
2
x10
:=
2
x30
2
x10
(24)
Next we apply an analytical continuation formula for the Gaussian hypergeometric

function (formula 9.132,1 of [16]) to transform the argument 1 1 to and obtain
as a result two hypergeometric functions:
d
A1 ,2 (x1 , x3 ) =
x30 2
1
0(1 )0(2 ) x10
420
k
k0
k!
2 0( 2 )0(1 + 2 + k)
0(2 + k) 1 +2 +k,2 +k
F
;
0( + k)
2 +1
+k,+k
1
; .
+ 0(2 )0(1 + k)F
1+2
(25)
The first term will be called direct term and the second will be called shadow term.
In the sequel all appearing propagators can be expressed as linear combinations of
A1 ,2 and derivatives thereof. In these applications the two parameters 1 and 2 are
not independent but fulfill an equation
1 + 2 = d + q,
(26)
where q is an integer. One can check that it suffices to use only one of the two summands
in A1 ,2 , because the other one is obtained by substituting the parameter by the shadow
parameter . Therefore, we insert
1 = r,
2 = s,
with + = d and r, s Z,
(27)
project onto the (say) first term in (25) and thus define

r
s

x10
x30
d

r,s (x1 , x3 ) := d x2

2
2
2
2
x10 + |
x12 |
x30 + |
x32 |
direct term
d(s)
0( + s) x30
r
0( r)0( s) x10
k
k0
k!
0( r s + k)
0( s + k) rs+k,s+k
F
; .
0( + k)
s+1
(28)
The series of hypergeometric functions may be summed up in terms of Appells
F4 -function (see 9.18 of [16] and references therein):
r,s (x1 , x3 ) =
0( + s) s 0( r s)0( s)
2
0( r)0( s)
0()
F4 ( s, r s, s + 1, ; , ).
(29)
Let us note the action of a d-dimensional Laplacian on r,s :

d(s)
x
)
r,s = 0( + s) 30
(,
r
0( r)0( s) x10

4
2
x10
k
k0
0( s + k + 1)
0( r s + k + 1)
0( + k)
rs+k+1,s+k+1
F
; ,
k!
s+1
(30)
where =
x1,i ,
421
i = 1, . . . , d. This can be written again in terms of r,s :
)
r,s = 4 ( s)( s + 1)r,s1 4( s)2 r,s2 .
(,
x30
(31)
We introduce Legendre functions of the second kind and write them in terms of
Gaussian hypergeometric functions:

(s)/2,(s+1)/2
s,t ( ) := 0( s)2(s) (s) F
(32)
; 2 ,
s+t +1
where s, t Z and the AdS invariant variable is defined by (13) with x = x1 and x = x3 .
We note the derivative of r,s with respect to :

(s+1)/2,(s+2)/2
d
s,t ( ) = ( s)0( s)2(s) (s+1) F
; 2
s+t +1
d
= 2s1,t 1( ).
(33)
)

Using this equation we get for the action of the d-dimensional Laplacian (,

) 2s1,t 1( ) (,
).

)
s,t ( ) = 4s2,t 2( )(,
(,
(34)
Moreover, we note the following two identities, which are simple consequences of
Eqs. 9.137, 6 and 12 in [16], respectively:
1
s2,t ( ),
( s + t + 1)2
s
1
s,t ( ) +
s2,t 1( ).
s1,t 1( ) =
2
s+t +1
s,t ( ) = s,t +1 ( ) +
(35)
Now we want to express certain linear combinations and derivatives of the functions
r,s in terms of the s,t . This can be established with the help of the following two
formulae. They both hold in the case r + s = m:
m

( r)j
m
1
rj,s+j (x1 , x3 )
mj
j
j ( s j )j x
j =0
30 x10
=
0( + s)
s+m,m ( )
0( r)0( s)
(36)
and
)
m r,s (x1 , x3 )
(,
m

0( + s) 4 m m
(1)j j/2
=
smj,m ( ). (37)
0( r)0( s) x30
j ( s + 1)j
j =0
The proof of (36) proceeds by induction, so let m = 0, to formulate the start of the
induction. We have r = s, and in this case the F4 -function of (29) can be summed up
in terms of a Gaussian hypergeometric function (Eq. 9.182,8 of [16]), which after some
422
algebra with the coordinates reads:

0( + s) s
2
0( r)0( s)
0( s)F4 ( s, , s + 1, ; , )
0( + s)
s,0 ( ),
=
0( r)0( s)
r,s (x1 , x3 ) =
(38)
thus showing the start of the induction of the proof of our formula.
step,

For
mthe
inductive
m
we take a look at the left-hand side of (36) for m + 1 and use m+1
=
+
to
obtain
j
j 1
j
m+1

( r)j
1
m
m
rj,s+j (x1 , x3 )
+
( s j )j x m+1j x j
j 1
j
j =0
30
10
m
j

( r)j
m
(1)
=
j (s + 1)j x mj x j
j =0
30
10

1
r +j
1

rj 1,s+j +1 (x1 , x3 ) +
rj,s+j (x1 , x3 ) .
s + j + 1 x10
x30
(39)
With formula 9.137,17 of [16], the first summand in [ ] of (39) can be written as
+s+j +1
0( + s + j + 1)
r + j x30
sj 1
s + j + 1 x r+j +2
0( r + j + 1)0( s j 1)
10
0( r s + k)0( s j 1 + k) rs+k,sj 1+k
F
k
;
sj
k! 0( + k)
k0
+s+j 1
0( + s + j ) x30
sj
0( r + j )0( s j ) x r+j
10
0( r s + k)
0( s j + k)
k
k! 0( + k)
k0

rs+k,sj +k
F
;
sj +1
rs+k,sj 1+k
1+k
F
; .
+
s j 1+k
sj +1
(40)
The first term in { } in (40) together with the prefactors and the sum cancels the second
term in [ ] in (39), therefore we obtain
m

m ( r )j (1)j

LHS of (36) m+1 =

mj
j

sj
j (s )j x
j =0
30 x10

r j,s+j (x1 , x3 )
(41)
,
, ,
423
where we have set = 1, = 1, = 1 and r = r 1, and the function

is to be understood with the unprimed parameters replaced by the primed ones. Now we
recognize that r + s = m, thus the sum can by done by induction hypothesis, giving

0( + s)

(
)
LHS of (36) m+1 =
s+m,m

s 0( r )0( s)
, ,
=
0( + s)
s+m+1,m+1 ( ),
0( r)0( s)
(42)
where we used the definition (32) of s,t . This proves (36).

Using (31), the proof of (37) can also be done by induction, where the start of the
induction is the same as for (36), but in this case the direct proof is even simpler. To this
)
on the series in (28)
end we apply m times the Laplacian (,

n
x +s
)
m r,s (x1 , x3 ) = 0( + s) 30 s 4
(,
r
2
0( r)0( r) x10
x10
0( s + k + n) +k,s+k+n
F
k
; .
(43)
s+1
k!
k0
Now we apply m times the formula

a,b
a+1,b b a+1,b+1
F
,z = F
, z zF
,z ,
c
c
c
c+1
(44)
see Eq. 9.137,12 of [16], and obtain a sum of 2m series of hypergeometric functions, of
which each series can be summed with 9.182,8 of [16], to give directly (36).
4. Results for the propagators

4.1. The scalar case
Now that we have all these formulae at hand, the calculation of the scalar bulk-to-bulk
propagator is ultra-simple. We convolute a bulk-to-boundary propagator of dimension
with a bulk-to-boundary propagator of dimension along the boundary, i.e., integrate
over the d-dimensional boundary variable and get with (38) for s = 0
A, (x1 , x3 ) = 0,0 (x1 , x3 ) + { }
0( )
0,0 ( ) + { }.
=
0()0()
(45)
This is up to normalization just the AdS two-point function of a field with parameter plus
the two-point function of a field with parameter . Thus we see that in the computation of
the AdS two-point function by intertwining the legs of a conformal two-point function we
automatically obtain as an artefact of the technique the two-point function of the field with
the shadow parameter.
424
4.2. The vector case

Now we consider the convolution of two vector bulk-to-boundary propagators, i.e., we
consider the integral

(1)
(1)
b ; x12)K (1) (b,
c; x23)
A, (x1 , x3 ) = d d x2 K (a,

=

b)
2a, x12(
x12 ,
b)
(
a
,
2 )
2
(x12
x12

1
x30
2(b, x23)x23 , c

(b, c) .
2
2
(x23 )
x23
d d x2
1
x10
(46)
The idea of calculating this integral is to write each bulk-to-boundary propagator as a

differential operator with respect to the exterior variables acting on a linear combination of
scalar bulk-to-boundary like terms. We then get

x
(1)
d
b , 1 )x 2 + 1 x 1 (
b )x 2
A, (x1 , x3 ) = d x2 a0 10 (
a,
12
12
10

1
x10
b , 1 )x 2(1)
+
(
a , 1 )(
12
2( 1)2

x
3 )x 2 + 1 x 1 (b,
c)x 2
c0 30 (b,
23
23
30

1
x30
2(1)
3 )(
+
(b,
c, 3 )x12
.
(47)
2( 1)2
Since all derivatives act on variables which are not integrated, we can take them in front
of the integrals and perform these in terms of the functions r,s , where we restrict on
the direct term. The r,s are functions of x13 , thus we can use := 1 = 3 . The b are
contracted and we are left with

(1)
(1)
(1)
(1)
(1)
A,
(48)
= A, + A, + A, + A, ,
direct
a0
c0
a0 c0
where we expanded in powers of a0 and c0 . The first term is given by

(1 )(1 ) 1

(1)
A, (x1 , x3 ) = (
a , c)
0,0 (x1 , x3 )
1
x10x30

)

(,
c, )

1,1 (x1 , x3 )
+ (
a , )(
2
2 ( 1)2 ( 1)2

1
1
1
1
+
1,0 (x1 , x3 ) +
0,1 (x1 , x3 ) .
x30 2( 1)2
x10 2( 1)2
(49)
Now we observe that the first line can be presented directly as a Legendre function by (36).
c, )
we use (31) on the first summand in [ ], and find
For the term proportional to (
a , )(
425
that we can apply (36) on the result

(1 )(1 ) 1

(1)
a , c)
0,0 (x1 , x3 )
A, (x1 , x3 ) = (
1
x10x30

1
c, )

1,1 (x1 , x3 )
+ (
a , )(
( 1)2

1
1
1
1
1,0 (x1 , x3 ) +
0,1 (x1 , x3 )
+
x30 2( 1)2
x10 2( 1)2

1
1
0( )

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

(1 )(1 )
0,0 ( ) .
+ (
a , c)
(50)
x10 x30
The second term in (48) is given by

1
(1)
)
0,1 1 1 0,0 .

(,
c, )
A, = a0 (
a0
2( 1)2
x30
(51)
We observe that both terms can be summed up by (37), to result in

(x
,
x
)
A(1)

10
30
,
=
a0
0( )

a0 (
c, )
0( + 1)0( + 1)

1
1
2
2
2,1 ( ) +
0,0 ( ) +
1,1 ( ) .
x30
x30 1
x10
In a similar manner we obtain for the third term in (48)

(x
,
x
)
A(1)

10
30
,
c0

1
1 1

(, )1,0
0,0
= c0 (
a , )
2( 1)2
x10
0( )

=
a , )
c0 (
0( + 1)0( + 1)

1
2
1
2
0,0 ( ) +
1,1 ( ) .
2,1 ( ) +
x10
x10 1
x30
(52)
(53)
We check that this one is the same as (52) after interchanging a c and x1 x3 , up to
the sign. This must be the case, because the derivative = 1 acting on x13 equals 3 ,
therefore, the sum of (52) and (53) is symmetric under the above permutation, which is
evident from the defining integral.3 Finally, the fourth term in (48) is given by the Laplacian
3 Strictly speaking, it may happen that a possible asymmetric part is cancelled with the asymmetric part in
the shadow term. However, from the comparison with the vector propagator in the literature we know that this
assumption is indeed justified.
426
acting on a scalar bulk-to-bulk propagator

1
0( )

(1)
)
0,0 ( ). (54)
=
A, (x10, x30 )
(, )0,0 =
a0 c0 (,
a0 c0
0( + 1)0( + 1)
Let us now write the vector two-point function in terms of the fundamental bitensors
(13). The two fundamental bitensors for tensor rank 1 are a, c, and a, c, ,
where the unprimed variables correspond to x1 and the primed ones to x3 . We perform the
differentiations of the Legendre functions s,t in Eqs. (50), (52), (53), (54) with (33), e.g.,
let us look at (50)

A(1)
, (x1 , x3 )
1
0( )
=
0( + 1)0( + 1)

2
)(
3,2 ( ) (
a ,
c, )
2 ( 1)1,1 ( ) +
1

2
(
a , c)
+ ( 1)0,0 ( ) +
2,1 ( )
,
1
x10 x30
(55)
add the various contributions up, solve (15) for the d-dimensional quantities, and insert
them into the sum. We then obtain for the complete vectorial two-point function

0( )
(1)
=
A,
direct
0( + 1)0( + 1)

2
3,2 ( ) I2
2 ( 1)1,1 ( ) +
1

2
2,1 ( ) I1 .
( 1)0,0 ( ) +
(56)
1
All other terms vanish, as can be checked with the identities (35). Comparing (55) with (56)
one notes, that the result is completely determined by the principal contribution, i.e.,
terms of maximal degree in a and c. This is clear, because when we substitute the ddimensional expressions composed of a and c by the (d + 1)-dimensional invariants, we
add components with positive degree in a0 or c0 , which must cancel the contributions from
(52)(54), due to symmetry. Thus we only have to determine the principal contributions.
4.3. The symmetric traceless tensor of rank 2
In this case the complete calculations are unappealingly lengthy, but with our experience
gained from the vectorial case we know how to shorten the route. In contrast to the case of
the vector we now have to take the subtraction of traces into account. First, we write down
the bulk-to-boundary propagator
x12)
K (a, b;
2 2
2
2

x10
2a, x12(
x12 , b)
x12
b 2a, x12

= 2
(
a , b)
a
2
2
d
(x12)
x12
x12
(2)
427

x12 )2
x12 )
a , x12 )(b,
(
a , x12 )2 (b,
(
a , b)(
2
4
= 2
4
+ (
a , b)
4
2
(x12)
x12
x12

a , x12 )2
b2 2
2 (
+ terms of positive order in a0 , c0 ,

a 4x10
4
d
x12
2
x10
(57)
which is traceless with respect to the bulk and the boundary variables. As before we write
down the integral for the bulk-to-bulk propagator, restrict to the direct terms and expand
into powers of a0 and c0

(2)
2A, (x1 , x3 )

=
direct

b ; x12)K (2)(b,
c; x23)
d d x2 K(2)(a,

(2)
= A, (x1 , x3 )
p.c.
direct
+ terms of positive order in a0 , c0 ,
(58)
where p.c. denotes the principal contribution, which is the only one we have to take into
b with b.
account. The factor 2 on the left-hand side comes from the two contractions of
(2)
The calculation of A, |p.c. is similar to the case of the vector propagator, i.e., express
it in terms of derivatives of r,s and manipulate the resulting expressions with the formula
(31) until we can write the sum of derivatives of the r,s functions in terms of the Legendre
(2)
functions s,t with (36) and (37). We decompose A, |p.c. into summands of different
numbers of derivatives:

(2)
A,
p.c.

(2)
= A,
2 (

(
a,)
c,)

(2)
+ A,

(2)
+
A
,
2
2 c 2
(
a ,)
c,)(
a,
(
a ,)(
c)

(2)
+ A,

a 2 (
c,)

(2)
+ A,

(2)
+
A

,
2
a 2 c 2
(
a ,
c )2
(59)
Before turning to the calculation, note that the invariant bitensors we have to expect for
the symmetric tensor of rank 2 are I12 , I1 I2 , I22 , and I3 , I4 for the removal of the traces of
the former invariants. The terms in the second line of (59) contain a 2 and c2 (after adding
appropriate terms with a0 and c0 ), so they remove the traces of the terms in the first line
and can consequently be computed from them. Therefore it is sufficient to consider only
the terms in the first line.
Let us start with the calculation of the first term in (59):

A(2)
, (0, 0; x1, x3 )
2 (
2
(
a ,)
c,)
2 (
2
= (
a , )
c, )
428
)
2
)

2(,
(6 4)(,
1
(x
,
x
)
+
2,1 (x1 , x3 )
2,2
1
3
24 ( 2)4 ( 2)4
x30 23 ( 2)4 ( 1)3
)

1
1
( 1)2
(6 4)(,
+ 2
2,0 (x1 , x3 ) +
1,2 (x1 , x3 )
3
x10 2 ( 1)3 ( 2)4
x30 2( 2)4 ()2
+
( 2)( 2) 1 +
1,1 (x1 , x3 )
(x10 x30) ( 1)3 ( 1)3
)

(,
( 1)2
0,2 (x1 , x3 ) +
2,0 (x1 , x3 )
2()2 ( 2)4
2d( 2)4 ()2

1 d 2( 1)
1,0 (x1 , x3 )
+
x10 d( 1)3 ()2
0( )
2 (
2
(
a , )
=2
c, )
0( + 2)0( + 2)

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

( 1)
1
1
+
1
0,1 ( ) + 2 ( 1)2 2,2 ( ) .
1
d
2
+
2
x10
(60)
The second term in (59) is given by

(2)
A, (x1 , x3 )

(
a ,)(
c,)(
a,
c)
c, )(
a , c)
= (
a , )(

1 2( 1)( 1)2

1,0 (x1 , x3 )
2
( 1)3 ()2
x10x30
1
2 x
x10
30
2( 1)2 ( 1)
0,1 (x1 , x3 )
()2 ( 1)3

( 1)( 1)
1
(, )1,1 (x1 , x3 )
x10 x30 ( 1)3 ( 1)3
0( )
c, )(
a , c)
(
a , )(
=2
0( + 2)0( + 2)

1
1
1,0 ( ) .
( 1)2 1,1 ( ) + 2
x10 x30
1
+
(61)
Finally, the third summand in (59) is

A(2)
(x
,
x
)

1
3
,
2
(
a ,
c)
= (
a , c)2
=2
( 1)2 ( 1)2
0,0 (x1 , x3 )
()2 ()2
2 x2
x10
30
0( )
0( + 2)0( + 2)
(
a , c)2
( 1)2 ( 1)2
0,0 ( ).
2 x2
x10
30
(62)
429
Now it is straightforward to write down the full propagator for a symmetric traceless tensor
field of rank 2. We perform the differentiations in (60) and (61), add them up together
with (62) and sort them with respect to the invariants I12 , I22 , and I1 I2 . After some not
very labourious algebra we then find the propagator of a symmetric traceless tensor field
of rank 2

(2)
A, (x1 , x3 )
direct
0( )
=
0( + 2)0( + 2)

d 1
d 1
4,4 ( ) + 23
4,3 ( )
24
d
1

+ 22 ( 1)2 2,2 ( ) I22

d 1
+ d 1
3,3 ( ) 23
3,2 ( )
+ 25
d
1

22 ( + 1)( 1)2 1,1 ( ) I1 I2

d 1
2,2 ( ) + 22
2,1 ( )
+ 23
d
1

+ ()2 ( 1)2 0,0 ( ) I12

traces .
(63)
Acknowledgements
This work is supported in part by the German Volkswagenstiftung. The work of R.M.
was supported by DFG (Deutsche Forschungsgemeinschaft). W.R. thanks Professor JianZu Zhang from the East-China University of Science and Technology for the hospitality at
his institute.
Appendix A. The trace terms

Consistency of the group approach to traceless symmetric tensor fields on AdS space
requires that the bitensor propagators can be made traceless using only the basis of
geometric bitensors I1 , I2 , I3 and I4 from Eqs. (14) and (18). We apply the Laplacian a
to Eq. (19) and express the result as a linear combination of the independent monomials
m
m
n
n
I1 1 I2 2 I3 1 I4 1 R1,2
(A.1)
430
with
R1 :=
1 2
I2c ,
z02
R2 :=
1
I4 .
a2
(A.2)
In fact, the basic formula is

1
a I1m1 I2m2 I3n1 I4n2
2

m1
(R1 + R2 )I1m1 2 I2m2 I3n1 I4n2
=
2

m2
+ m1 m2 R1 I1m1 1 I2m2 1 I3n1 I4n2 +
( 2 1)R1 I1m1 I2m2 2 I3n1 I4n2
2

n1
R1 + ( 2 1)R2 + n1 (d + 1)R1 + ( 2 1)R2
+ 4
2

+ 2m1 n1 R1 + 2m2 n1 R1 + ( 2 1)R2 I1m1 I2m2 I3n1 1 I4n2

n2
m1 m2 n1 n2 1
R
+ 4
2 + n2 (d + 1)R2 + 2(m1 + m2 + 2n1 )n2 R2 I1 I2 I3 I4
2

n1
n1
m1 m2 +2 n1 2 n2
+4
I3
I4 4
R2 I1 I2
( 2 1)R1 I1m1 I2m2 I3n1 2 I4n2 +1
2
2
m1 1 m2 +1 n1 1 n2
I2
I3
I4 .
+ 2m1 n1 R2 I1
(A.3)
In the case l = 2, we have (l1 , l2 ) {(2, 0), (1, 1), (0, 2)} and we calculate the matrix V
representing 12 a in the bases {I1l1 I2l2 } and {R1 , R2 }, respectively. We obtain
R1
R2
(2, 0)
1
1
(1, 1)
(0, 2)
2 1
,
0
(A.4)
where we indicated the respective basis elements by the labels above each column to the
left of the rows. For the second term in (19) we need the matrix M representing 12 a
on span(I3 , I4 ), and the image is again span(R1 , R2 ). This must be the case, because
otherwise there were no solution to (19). We get from (A.3)

d +1
0
M=
(A.5)
2 1 d + 1
and collect the unknown coefficients in a matrix
(2,0)
(1,1)
(0,2)
A0010
A0010
A0010
.
A=
(2,0)
(1,1)
(0,2)
A0001 A0001 A0001
(A.6)
Then (19) takes the form of a matrix equation

V = MA,
(A.7)
which is easy to solve thanks to the triangular shape of M:

1
2 1
1
1
A=M V =
.
2 1
2 1
2 1)2
d + 1 1 d+1
d+1
(d+1
431
(A.8)
In the case l = 3 we also get a triangular 4 4 matrix M acting on the trace terms and
the matrix A of unknowns comes out as
(30)
(12)
(03)
A1010 A(21)
1010 A1010 A1010
(30)
(21)
(12)
(03)
A1001 A1001 A1001 A1001
(30)
(12)
(03)
A0110 A(21)
A
A
0110
0110
0110
(21)
(12)
(03)
A(30)
0101 A0101 A0101 A0101
3
2
2 1
0
2 1
2 1)2

2 1
2 d+3
(d+3
0
1
.
3 1 d+3
=
(A.9)
2
0
1
2
3( 1)
d +3
2 3
2 1
2 1)2
6
1 7d+3
8 d+3
9 (d+3
d+3
In the case l = 4 a dramatic change occurs since the matrix M representing 12 a on
the rank 4 trace terms becomes a 9 10 rectangular matrix, i.e., there are 9 trace terms
to cancel the traces of the 10 invariants of rank 4, thereby leading to 10 equations for the
unknown entries of A. In general, for l = 2p even, the number of trace terms and the
number of invariants differ at p 2:

1
1
3
Number of trace terms = p p2 + p +
,
3
2
6

1
2
.
Number of equations = p(p + 1) p +
(A.10)
3
3
Thus the set of equations for A contains

1
1 2 1
p p p+
3
2
6
(A.11)
linearly dependent equations. We compute the linear system for l = 4 and show thereby
that there is 1 linearly dependent equation. The matrix V representing 12 a on

span I14 , I13 I2 , I12 I22 , I1 I23 , I24
is calculated by (A.3) to result in a 10 5 matrix, which decomposes into

V0
,
V=
0
(A.12)
(A.13)
432
where
6
6
0
V0 =
0
0
0
3
0
3
3
0
0
2 1
0
4
0
1
1
0
0
3( 2 1)
0
3
0
0
0
0
.
0
2 1
0
(A.14)
The basis in the preimage of 12 a is ordered as in (A.12), and the 10 basis elements
{Ii Ij Rk } in the image are ordered lexicographically, i.e., I12 R1 , I12 R2 , . . . , I4 R2 . By the
simple transformation
4
(l l )
A 12
d + 5 0020
we can bring the system of equations into the block form

0
A0
M0
V0
,
=
0
A1
B M1
(l l )
(l l )
1 2
1 2
A0201
+
A0201
(A.15)
(A.16)
where the 4-tupels (m1 , m2 , n1 , n2 ) are ordered as

(2010), (2001), (1110), (1101), (0210), (0201), (0020), (0011), (0002).
(A.17)
Then M0 has triangular shape
d +5
0
0
0
2 1 d + 5
0
0
0
0
d
+
5
0
M0 =
2 1) d + 5
4
0
3(
0
0
0
0
0
0
2
0
thus M0 can be inverted easily to give
6
3

2 1
6 1 2 1
3 d+5
d+5
0
3
1
A0 =

2 3
24
d +5
3 1 7d+5
d+5
0
0
6
0
d+5
0
0
0
0
d +5
5( 2 1)
0
2
4
16
( 2 1)2
d+5
1
1
(A.18)
2 1
1)
(d+5
0
0
0
,
0
0
d +5
13 2 5
d+5
0
.
2
6( 1)
( 2 1)2
30 d+5
0
3( 2 1)
1)
9 (d+5
2
3
1
21 d+5
2
(A.19)
Finally, it remains to solve
BA0 = M1 A1 .
(A.20)
433
1)
We manipulate both sides the following way: multiplying the first row with (d+5
, the
4
and adding both to the third row results in a zero row, which we
second with 1 + d+5
skip on both sides of (A.20). We thus get the transformed equations
2
1 ,
M1 M
with
0
BA0 BA
2(d + 3)
0
1 = 6( 2 1)
d +5
M
0
( 2 1)
and
0 =
BA
6
6
d +5

6 1
2 1
d+5
(A.21)
0
2(d + 3)
(A.22)
2(3 2 1) 6( 2 1)
2 1
0
6
3
1
3 d+5
2
1)
(d+5
2
6( 2 1)2
0
.
0
(A.23)
We evaluate then
0
1 BA
A1 = M
1
(A.24)
and get
A1 =
d + 5
3
d+3

3( 2 1)
6
d+5 1 d+3

2( 2 1)
3( 2 1)2
3
d+3 1 d+5 + (d+3)(d+5)
3 2 1
d+3
2 1
6(3 2 1)
d+5 1
d+3
( 2 1)2
3(3 2 1)
(d+3)(d+5) 1 +
d+3
3
d+3
3
6( 2 1)
d+5 1 d+3
3( 2 1)
3( 2 1)
(d+3)(d+5) 1 + d+3
3( 2 1)
d+3
18( 2 1)2
(d+3)(d+5)
9( 2 1)3
(d+3)2 (d+5)
3( 2 1)2
d+3
18( 2 1)3
(d+3)(d+5)
9( 2 1)4
(d+3)2 (d+5)
(A.25)
At the end, we have to invert the transformation (A.15).
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] C. Fronsdal, Massless fields with integer spin, Phys. Rev. D 18 (1978) 3624.
[3] M.A. Vasiliev, Nonlinear equations for symmetric massless higher spin fields in (A)dS(d), hep-th/0304049.
[4] I.R. Klebanov, A.M. Polyakov, AdS dual of the critical O(N ) vector model, Phys. Lett. B 550 (2002) 213,
hep-th/0210114.
[5] K. Lang, W. Rhl, The critical O(N ) sigma model at dimensions 2 < d < 4: Fusion coefficients and
anomalous dimensions, Nucl. Phys. B 400 (1993) 597.
434
[6] K. Lang, W. Rhl, The critical O(N ) sigma model at dimensions 2 < d < 4: A list of quasiprimary fields,
Nucl. Phys. B 402 (1993) 573.
[7] T. Leonhardt, A. Meziane, W. Rhl, On the proposed AdS dual of the critical O(N ) sigma model for any
dimension 2 < d < 4, Phys. Lett. B 555 (2003) 271, hep-th/0211092.
[8] C.P. Burgess, C.A. Lutken, Propagators and effective potentials in anti-de Sitter space, Phys. Lett. B 153
(1985) 137.
[9] B. Allen, T. Jacobson, Vector two-point functions in maximally symmetric spaces, Commun. Math.
Phys. 103 (1986) 669.
[10] E. DHoker, D.Z. Freedman, S.D. Mathur, A. Matusis, L. Rastelli, Graviton and gauge boson propagators in
AdS(d + 1), Nucl. Phys. B 562 (1999) 330, hep-th/9902042.
[11] B. Allen, M. Turyn, An evaluation of the graviton propagator in de Sitter space, Nucl. Phys. B 292 (1987)
813.
[12] M. Turyn, The graviton propagator in maximally symmetric spaces, J. Math. Phys. 31 (1990) 669.
[13] V.K. Dobrev, Intertwining operator realization of the AdS/CFT correspondence, Nucl. Phys. B 553 (1999)
559, hep-th/9812194.
[14] E. Witten, Anti-de Sitter space and holography, Adv. Theor. Math. Phys. 2 (1998) 253, hep-th/9802150.
[15] D.Z. Freedman, S.D. Mathur, A. Matusis, L. Rastelli, Correlation functions in the CFT(d)/AdS(d + 1)
correspondence, Nucl. Phys. B 546 (1999) 96, hep-th/9804058.
[16] I.S. Gradshteyn, I.M. Ryzhik, Table of Integrals, Series, and Products, Academic Press, San Diego, CA,
1965.

From dynamical to numerical R-matrices:

a case study for the Calogero models
Michael Forger, Axel Winterhalder
Departamento de Matemtica Aplicada, Instituto de Matemtica e Estatstica, Universidade de So Paulo,
Caixa Postal 66281, BR-05311-970 So Paulo, SP, Brazil
Received 23 January 2003; accepted 17 June 2003
Abstract
Within the class of integrable Calogero models associated with (semi-)simple Lie algebras and
with symmetric pairs of Lie algebras identified in a previous paper, we analyze whether and to
what extent it is possible to find a gauge transformation that takes the traditional Lax pair with
its dynamical R-matrix to a new Lax pair with a numerical R-matrix.
PACS: 02.30.Ik
1. Introduction
In a recent paper [1], we have performed a systematic analysis of the CalogeroMoser
Sutherland models, or Calogero models, for short, which constitute an important class of
completely integrable Hamiltonian systems. Our work follows the traditional Lie algebraic
approach outlined long ago by Olshanetsky and Perelomov [24] which is based on the
use of (semi-)simple Lie algebras and, more generally, of symmetric pairs, extending it
so as to encompass the construction not only of a Lax representation for the equations of
motion but also that of a dynamical R-matrix. The existence of these structures was found
to depend on the possibility of solving a simple set of algebraic constraints for a certain
function F or K that assigns to each root a generator F or K in the pertinent Cartan
Work supported by CNPq (Conselho Nacional de Desenvolvimento Cientfico e Tecnolgico), Brazil, and
by FAPESP (Fundao de Amparo Pesquisa do Estado de So Paulo), Brazil.
E-mail addresses: forger@ime.usp.br (M. Forger), winter@ime.usp.br (A. Winterhalder).
doi:10.1016/S0550-3213(03)00536-4
436
M. Forger, A. Winterhalder / Nuclear Physics B 667 [PM] (2003) 435483
subalgebra h:1 these state that for any two roots and
g (F ) g (F ) = , ,
(1)
in the case of (semi-)simple Lie algebras g and
,
g (K ) g (K ) = ,
(2)
are defined in
in the case of symmetric pairs (g, ), where the coefficients , and ,
terms of the structure constants N, of g and the coupling constants g of the model by
, = g+ N,
(3)
and by
1
,
= (g+ N, + g+ N, + g+ N, + g+ N, ),
4
(4)
respectively.2 In the first case, it was found that a solution of these constraints exists only
for the Lie algebras sl(n, C) of the A-series, whereas in the second case, explicit solutions
were found for the complex Grassmannians SU(p + q)/S(U (p) U (q)) of the A III-series
when |p q| 1.
Following a somewhat different direction, several authors [5,6] have recently shown
by explicit matrix computations that the Calogero models based on sl(n, C), degenerate
as well as elliptic, admit a gauge transformation taking the dynamical R-matrix into a
numerical one: this is achieved by explicitly constructing a group-valued function on
the configuration space which is used to conjugate the standard Lax pair and dynamical
R-matrix of the model into a new Lax pair and a numerical R-matrix.
In the present paper, we systematize the method of Fehr and Pusztai [6], adapting
it to the formalism developed in our previous work [1]: this allows us to extend it from
the degenerate to the elliptic models as well as from the case of Lie algebras to that of
symmetric pairs. In all cases, we find that the existence of a gauge transformation with the
desired property can be reduced to a set of purely algebraic constraints which are similar
to but not identical with the integrability constraints (1) and (2) found in Ref. [1]. In the
case of Lie algebras, it turns out that these various constraints all have one and the same
solution, thus confirming the previous results of other authors [5,6] that the well-known
dynamical R-matrices of the Calogero models based on sl(n, C) can be gauge transformed
to numerical R-matrices. In the case of symmetric pairs, however, we find extra constraints
on the root system, over and above those that guarantee integrability. In particular, for the
A III-series of complex Grassmannians SU(p + q)/S(U (p) U (q)) where integrability
has in Ref. [1] been shown to hold when |p q| 1, these constraints exclude the case
1 In this paper, we adopt a slightly modified notation: for reasons to become clear towards the end of the paper,
we shall in the case of symmetric pairs denote the generators F of Ref. [1] by K .
2 In the case of symmetric pairs, it is also assumed that the root generators E in g can be and have been
chosen so that E = E for all , implying that N , = N, for all , . As explained in the
erratum to Ref. [1], this is not always possible but is a necessary condition for our proof of integrability, and it
can always be arranged to hold for all roots if it can be made to hold for all real roots , that is, all roots in
satisfying = .
437
|p q| = 1 but allow for a solution in the case p = q. This means that the dynamical Rmatrices for the Calogero models associated with the classical root systems can be gauge
transformed to numerical R-matrices for the Cn and Dn systems but not for the Bn and
BCn systems; remarkably, the latter are just the ones containing an explicit dependence on
the coupling constants.
The paper is organized as follows. In Section 2, we give a brief summary of the method
of gauge transforming Lax pairs and dynamical R-matrices in integrable systems such as
the Calogero models; moreover, we collect a number of identities to be used repeatedly
later on. In Section 3, we present our calculations for the case of Lie algebras and in
Section 4 those for symmetric pairs. Finally, in Section 5 we draw our conclusions and
comment on perspectives for further work.
2. Gauge transformations
Consider an integrable model with a finite-dimensional phase space which we assume
to be the cotangent bundle T Q of a configuration space Q. Integrability is encoded into
the existence of a Lax representation for the equations of motion,
L = [L, M],
(5)
together with that of an R-matrix whose role is to control the Poisson brackets between the
components of the Lax matrix L, according to the formula [7]
{L1 , L2 } = [R12 , L1 ] [R21 , L2 ].
(6)
T Q
into a given Lie algebra g whereas R will in general

Here, L and M are maps from
be a map from T Q into the second tensor power U (g) U (g) of the universal enveloping
algebra U (g) of g; as usual, L1 = L 1, L2 = 1 L etc. The choice of g is far from
obvious; it reflects the hidden symmetries that are present in the model. Moreover, even if
one fixes g and a connected Lie group G that has g as its Lie algebra, L, M and R are not
uniquely determined. In particular, we are free to perform a gauge transformation by an
arbitrary function g on T Q with values in G, as follows:
L = gLg 1 ,

1 1
1 1 1
1

R12 = g1 g2 R12 + g1 {g1 , L2 } + g1 g2 {g1 , g2 }, L2 g1 g2 .
2
(7)
(8)
The second transformation law is dictated by the requirement that the fundamental Poisson
bracket relation (6) should be preserved under this transformation, which is easy to check.
Note that in general, L, M and g may depend on a spectral parameter u, in which case R
will depend on two spectral parameters u and v.
In the case of the Calogero models of interest here, g is a simple complex Lie algebra,
with Cartan subalgebra h and corresponding root system fixed once and for all, Q is an
open subset in a real subspace of h in which we fix a basis {H1 , . . . , Hr }, L is of the form
L(q, p; u) =
r

j =1
pj Hj +
L (q, u)E ,
(9)
438
with functions L whose explicit form will be needed only later, and R is independent
of the momentum variables. (For details, see Ref. [1].) Our aim in what follows will be to
determine g in such a way that R becomes constant (as a function on phase-space). To this
end, we shall assume that g is also independent of the momentum variables and introduce
the gauge potentials
Aj (u) = g 1 (u)j g(u).
(10)
Reverting to ordinary tensor notation, we get

r
r

1
(u) 1 g(u)
L(v) =
g (u) pj , g(u) Hj =
Aj (u) Hj ,
,
j =1
j =1
so Eq. (8) simplifies to

r

Aj (u) Hj g 1 (u) g 1 (v) .
R (u, v) = g(u) g(v) R(u, v)

j =1
This implies

1
g (u) g 1 (v) k R (u, v) g(u) g(v)

r

= k R(u, v)
Aj (u) Hj
j =1

r

1

1
+ g (u) g (v) k g(u) g(v) , R(u, v)
Aj (u) Hj ,
j =1
so the condition that the partial derivatives k

requiring

r

k R(u, v)
Aj (u) Hj
R (u, v)
of
R (u, v)
j =1
R(u, v)
r

all vanish amounts to

Aj (u) Hj , Ak (u) 1 + 1 Ak (v) = 0.
j =1
Using the integrability condition

k Al (u) l Ak (u) + Ak (u), Al (u) = 0
(11)
that follows from Eq. (10), this can be rewritten in the form
k R(u, v)
r

j Ak (u) Hj R(u, v), Ak (u) 1 + 1 Ak (v)
j =1
r

j =1

Aj (u) Hj , Ak (v) = 0.
(12)
439
In order to compute the content of Eqs. (11) and (12), we shall in what follows expand the
gauge potential according to

h
Aj (u)E ,
Aj (u) = Aj (u) +
(13)
which allows us, in particular, to decompose Eq. (11) into its Cartan part

h
h
Ak (u)A
k Al (u) l Ak (u) +
l (u)H = 0
(14)
and its root part

h
h
k Al (u) l Ak (u) + Ak (u) Al (u) Al (u) Ak (u)

+
N, Ak (u)Al (u) = 0.
(15)
,
+ =
In what follows, we shall analyze under what conditions this system of equations admits
solutions when we insert the explicit expressions for R given in Ref. [1] and evaluate the
commutators in Eq. (12) using the usual abbreviation j = (Hj ) and the relations
[Hj Hj , E 1] = j E Hj ,
[Hj Hj , 1 E ] = j Hj E
(16)
(no summation over j ),

[F E , Hj 1] = 0,
[F E , 1 Hj ] = j F E ,
[F E , E 1] = (F )E E ,
[F E , 1 E ] = 0,
[F E , 1 E ] = F H ,
[F E , 1 E ] = N , F E +
if = 0
(17)
(valid for any set of generators F belonging to the Cartan subalgebra h) and
[E E , Hj 1] = j E E ,
[E E , 1 Hj ] = j E E ,
(18)
[E E , E 1] = 0,
[E E , E 1] = H E ,
[E E , E 1] = N , E + E
if = 0,
(19)
[E E , 1 E ] = E H ,
[E E , 1 E ] = 0,
[E E , 1 E ] = N , E E +
if = 0.
(20)
440
The second step would be to determine g itself and, from there, find L and R : this
question will be addressed elsewhere in order not to overload our presentation here.
Concluding this section, let us for later use collect the functional identities satisfied
by the coefficient functions L that appear in Eq. (9) above. For the degenerate models,
L (q, u) = ig w((q)) where w is an odd function of its argument,
w(t) = w(t),
(21)
that satisfies the differential equation

w
= w2 ,
w
(22)
as well as the functional equation

w (s) w (t)
+
w(s + t) + w(s)w(t) = 0,
w(s)
w(t)
(23)
already employed in Ref. [1]. For the elliptic models, L (q, u) = ig ((q), u) where
and the closely related Weierstrass zeta function satisfy the symmetry properties
(z1 , z2 ) = (z1 , z2 ),
(z) = (z)
(24)
and the functional equations

(s, u)(s, u) = (s) (u),

(25)

(s, u) (s, u) (s, u)(s, u) = (s),

(s, u) (t, u) (s, u)(t, u) = (s) (t) (s + t, u),
(26)
(s, v u)(s + t, v) + (t, u v)(s + t, u) = (s, u)(t, v),

(s, u v)(s, u) + (v u) + (u) (s, v) = (s, v),
(28)
already employed in Ref. [1], as well as the additional functional equations

(s, u) = (s + u) (s) (s, u),

(s, u)(t, u) = (s) + (t) + (u) (s + t + u) (s + t, u),
(27)
(29)
(30)
(31)
denotes the derivative of with respect to the first argument; all of these can be
where
derived from the representation of and in terms of the Weierstrass function:
(z)
(z1 + z2 )
,
(z) =
.
(z1 ) (z2 )
(z)
Note that in the degenerate case, the spectral parameter drops out. In fact, all of the
calculations to be presented in what follows can be carried out for the degenerate case
in exactly the same manner as for the elliptic case, provided one performs the following
substitutions:
(z1 , z2 ) =
(s, u), (s, v)
w(s),
w (s)
,
w(s)
(u), (v), (u v), (v u), (s + u), (s + t + u)
(s, u v), (s, v u), (s)
0.
(32)
441
Therefore, we shall suppress the calculations for the degenerate models, except at the few
points where substantial differences arise.
3. Calogero models for semi-simple Lie algebras

According to Ref. [1], the standard Lax matrix L and the dynamical R-matrix for the
Calogero models associated with the root system of a simple complex Lie algebra g read
L=
r

ig w (q) E ,
(33)

w ((q))

E E ,
w (q) F E +
w((q))
(34)
j =1
R=
pj Hj +
for the degenerate model and

L(u) =
r

pj Hj +
j =1
R(u, v) =

ig (q), u E ,
(35)
r

j =1

(u v) + (v) Hj Hj

(q), v F E

(q), u v E E ,
(36)
for the elliptic model. As has been shown in Ref. [1], integrability requires the generators
F hR appearing in Eqs. (34) and (36) to satisfy the constraints (1). Moreover, writing
1
F = (F F ),
2
(37)
we also impose the condition

F+ = 0,
(38)
which follows from Eq. (1) by setting = when g = 0 but turns out to be true in
general, independent of this hypothesis.
In order to compute the content of Eqs. (11) and (12), we further expand the Cartan part
of the gauge potential according to
h
Aj (u) =
r

k=1
Akj (u)Hk .
(39)
442
Then inserting Eqs. (36), (13) and (39) into Eq. (12), we obtain

k (q), v F E
k (q), u v E E
0=
+
+
r

j Ak (q, u) Hj
j =1
r

r

j Ak (q, u)E Hj
j =1
j =1
r

j =1
r

j =1
r

(u v) + (v) Ak (q, u)[Hj Hj , E 1]

(u v) + (v) Ak (q, v)[Hj Hj , 1 E ]

j
(q), v Ak (q, u)[F E , Hj 1]

j
(q), v Ak (q, v)[F E , 1 Hj ]
j =1
+
+

(q), v Ak (q, u)[F E , E 1]

(q), v Ak (q, v)[F E , 1 E ]
,
r

j =1
r

j
(q), u v Ak (q, u)[E E , Hj 1]

j
(q), u v Ak (q, v)[E E , 1 Hj ]
j =1
+
+
+

(q), u v Ak (q, u)[E E , E 1]

(q), u v Ak (q, v)[E E , 1 E ]
,
r

j Ak (q, v)Aj (q, u) E
j =1
r

j Aj (q, u)Ak (q, v)E E .
j =1 ,
Using Eqs. (16)(20) to carry out the commutators, together with the relation
r

j =1
j Hj = H ,
(40)
443
we can collect the terms to identify the components of Eq. (12) along the various subspaces
of g g: those along h Hj (1 j r),

h
j Ak (q, u) +
(41)
j (q), v A
k (q, v)F = 0,
those along h g ( ),

k (q), v F + (u v) + (v) Ak (q, v)H

h

+ (q), v Ak (q, v) F
N , (q), v Ak (q, v)F
,
+=
r

h
(q), u v Ak (q, u)H +
j Ak (q, v)Aj (q, u) = 0,
(42)
j =1
those along g Hj ( , 1 j r),

j Ak (q, u) + j (q), u v Ak (q, v)

j (u v) + (v) Ak (q, u) = 0,
(43)
those along g g ( ),
r

j Aj (q, u)Ak (q, v) = 0,
(F ) (q), v Ak (q, u)
(44)
j =1
those along g g ( ),

k (q), u v + (F ) (q), v Ak (q, u)

h

h

+ (q), u v Ak (q, u) Ak (q, v)
+
r

j Aj (q, u)A
k (q, v) = 0,
(45)
j =1
and finally those along g g with , , = 0,

(F ) (q), v Ak (q, u)

+

+

N, (q), u v Ak (q, v) (q), u v Ak (q, u)
r

j Aj (q, u)Ak (q, v) = 0.
(46)
j =1
This is a complicated set of equations which we shall solve in a series of steps.

We begin by considering the differential equation (43) for the root part of the gauge
potential, which by using the functional equation (29) (with u and v interchanged) is seen
to have the simple solution

Ak (q, u) = (q), u ak ,
(47)
444
where the coefficients ak are constants that must be determined from the remaining
equations. For what follows, we shall find it convenient to assemble these constants into a
vector in hR by writing, for any ,
a =
r

aj Hj ,
(48)
j =1
so that of course
ak = (Hk , a ).
(49)
In analogy with Eq. (37), we also introduce the abbreviation

1
a = (a a ).
2
Now we are ready to state the first main result of this section.
(50)
Proposition 1. The integrable Calogero model associated with the root system of a simple
complex Lie algebra g admits a gauge transformation g from the standard Lax pair of
Olshanetsky and Perelomov and the dynamical R-matrix of Ref. [1], as given by Eqs. (33)
(36), to a new Lax pair with a numerical R-matrix if and only if the set of generators
F hR appearing in Eqs. (34) and (36) satisfies the algebraic constraints

F+ = 0,
(51)
1
F =
(52)
H ,
2 ||
(F )F (F )F = N, F+
for , such that = ,
as well as the additional algebraic constraints

H F F = 0,
(53)
(54)
N, F F = H F F H ,
(55)
,
+ =
to be imposed in the case of the elliptic model, where F = 1/2(F F ) as above, with
1 = 1. In this case, the root part and the Cartan part of the gauge potential Ak = g 1 k g
associated with this gauge transformation g are given by

Ak (q) = w (q) (Hk , F ),
(56)
and
h
Ak (q) =
w ((q))
(Hk , F )F ,
w((q))
(57)
for the degenerate model and by

Ak (q, u) = (q), u (Hk , F ),
and
h
Ak (q, u) =

(q) (Hk , F )F (u)Hk ,
445
(58)
(59)
for the elliptic model.

Note. As we shall show after completing the proof of Proposition 1, Eq. (53)
forces all
roots in to have the same length (which by convention we fix to be 2 ) and also
allows for a choice of basis in which the signs 1 are independent of so that Eq. (52) can
be simplified as follows:
1
F = 1H .
2
(60)
Proof. With the vector notation introduced above, we can first of all reduce Eq. (44) to a
single algebraic constraint:
(a ) = (F ).
Note that replacing by and adding/subtracting the two equations, we get

F+ = a+ ,

F = a .
Using Eq. (38), the first of these can be sharpened as follows:

F+ = 0 = a+ .
(61)
(62)
(63)
(64)
Next, inserting Eq. (47) together with the functional equation (25) into the differential
equation (41) for the Cartan part of the gauge potential, we see that this equation can be
solved by setting

h
h
(q) ak F ak (u),
Ak (q, u) =
(65)
h
ak (u)
where the
are constants that must be determined from the remaining equations,
provided we assume the coefficients ak to satisfy the relation

(66)
j ak F = 0 for 1 j, k r.
Converted into a tensor equation, it reads

H F a = 0,
(67)
which leads back to Eq. (66) by taking the scalar product with Hj in the first and with
Hk in the third tensor factor. Note that in the degenerate case, the same argument works,
446
but Eq. (66)/(67) is not needed. Similarly, inserting Eq. (47) together with the functional
equation (29) into Eq. (45), we obtain

k (q), u (q), v + (u) (v) (q), u v

h
h
(F )ak (q), u (q), v (q), u v ak (u) ak (v)

(a )ak (q), u (q), v = 0.
Obviously, the terms proportional to ((q), u v) cancel provided we set
h
ak (u) = (u)Hk ,
(68)
and the remaining terms cancel if we impose the relation

k + (F )ak + (a )ak = 0.
(69)
Converted to a vector equation in hR , it reads

H + (F )a + (a )a = 0,
(70)
which leads back to Eq. (69) by taking the scalar product with Hk . Even simpler to handle
is Eq. (46), which by insertion of the functional equation (28) reduces to the relation
+
(F )ak N, ak
(a )ak = 0
for , such that = .
(71)
Converted to a vector equation in hR , it reads

(F )a N, a+ (a )a = 0
for , such that = ,
(72)
which leads back to Eq. (71) by taking the scalar product with Hk .
Before proceeding to the solution of the remaining equations, let us pause to draw a few
consequences of the algebraic constraints (61)(64) and (69)/(70) derived so far; this will
help us considerably to simplify our further work. First of all, Eqs. (61)(64) state that
(F ) = (F ) = (a ) = (a ),
implying that Eq. (70) can be reduced to
(a )(a a ) = H .
Applying to this relation and using the previous equation again, we conclude that
||
(a ) = 1
2
1
and a =
H ,
2 ||
(73)
where 1 = 1 is a sign factor (1). Next, we simplify all these equations by showing
that Eqs. (71)/(72) and (73) in fact force the vectors a and F to be equal. To prove this,
we begin by symmetrizing Eq. (72) with respect to the exchange of and , obtaining
(F a )a + (F a )a = 0.
447
Symmetrizing with respect to the exchange of and and inserting Eq. (73) gives

1
F+ a+ a +
(74)
(F a )H = 0.
2 ||
Symmetrizing again with respect to the exchange of and and inserting Eq. (73) then
leads to

1
1
F+ a+ H +
F+ a+ H = 0.
2 ||
2 ||
But H and H are linearly independent since, as stated in Eq. (72), the roots and are
supposed to be non-proportional, so the coefficients must vanish separately, that is, for any
two roots , , we have

F+ a+ = 0
whenever is not proportional to and, according to Eq. (64), also when is proportional
to . Since generates hR , this simply means that a+ = F+ . Inserting this conclusion
back into Eq. (74) and applying once more the same argument, we arrive at the result
that a = F . With this result, Eqs. (61)(63) and (70) reduce to trivial identities whereas
Eqs. (64), (73), (72), (67), (47) with (49) and (65) with (68) assume the form given in
Eqs. (51), (52), (53), (54), (58) and (59), respectively.
Let us summarize the results obtained so far. With the exception of Eq. (42), the system
of Eqs. (41)(46) has been completely solved in terms of the explicit formulae (56)(59)
for the gauge potential with the explicit formula (52) for the odd part F of the coefficient
vectors F and the algebraic constraints (51), (53) and (54). Thus we are left with the task
of verifying the implications of Eqs. (14), (15) and (42).
Beginning with Eq. (14), we use the functional equation (25) to compute

h
h
Ak (q, u)A
k Al (q, u) l Ak (q, u) +
l (q, u)H
=

(q) l (Hk , F )F k (Hl , F )F

(q), u (q), u (Hk , F )(Hl , F )H

1
=
(q) +l (Hk , F )F l (Hk , F )F
2
(u)
=
k (Hl , F )F + k (Hl , F )F

+ (Hk , F )(Hl , F )H (Hk , F )(Hl , F )H
(Hk , F )(Hl , F )H

(q) l Hk , F F+ + l Hk , F+ F

+ k Hl , F F+ k Hl , F+ F

+ Hk , F Hl , F+ H Hk , F+ Hl , F H
(u)
(Hk , F )(Hl , F )H
448
and can use Eqs. (52) and (54) to verify that the terms under the first sum cancel mutually in
pairs whereas the second sum vanishes. Note that in the degenerate case, the same argument
works, but Eq. (54) is not needed.
For the proof of Eq. (15), the trick is to split the sum over roots coming from the
third and fourth term into contributions with = , which cancel mutually, contributions
with = , which combine with the contributions coming from the first and second
term (transformed using the functional equation (30)), and the remaining contributions
with = : these can be complemented by terms that also cancel mutually (marked by
underlining) and then be combined with the contributions from the fifth term (transformed
using the functional equation (31)):
h

h

k Al (q, u) l Ak (q, u) + Ak (q, u) Al (q, u) Al (q, u) Ak (q, u)

+
N, Ak (q, u)Al (q, u)
,
+ =

= (q), u k (Hl , F ) (q), u l (Hk , F )

+ (q), u
(q) (F )(Hk , F )(Hl , F ) (u)k (Hl , F )

(q), u
(q) (F )(Hl , F )(Hk , F ) (u)l (Hk , F )

N , (q), u (q), u (Hk , F )(Hl , F )
,
+=

= (q), u (q), u (u) (Hk , H )(Hl , F ) (Hl , H )(Hk , F )

+ (q), u (q)

+(F )(Hk , F )(Hl , F ) (F )(Hk , F )(Hl , F )

(F )(Hl , F )(Hk , F ) + (F )(Hl , F )(Hk , F )

+ (q), u
(q) (Hk , F ) (F )(Hl , F ) + (F )(Hl , F )
=

(q), u
(q) (Hl , F ) (F )(Hk , F ) + (F )(Hk , F )
=

(q) + (q) N , (Hk , F )(Hl , F )
+ (q), u
,
+=

(q), u ((q) + u) (u)
N , (Hk , F )(Hl , F )
,
+=

= (q), u (q) + u (q) (u)
449

(Hk , H ) Hl , F+ Hk , F+ (Hl , H )

2 F Hk , F Hl , F+ + 2 F Hk , F+ Hl , F

+ (q), u (q) + u (u)

2 F Hk , F Hl , F+ 2 F Hk , F+ Hl , F

N, (Hk , F )(Hl , F )
,
+ =

+ (q), u
(q) (Hk , F )
=

Hl , (F )F + (F )F + N, F

(q), u
(q) (Hl , F )
=

Hk , (F )F + (F )F N, F .
The last two terms vanish due to Eq. (53), while the first term vanishes due to Eq. (52).
The same reasoning shows that the second term will vanish provided we impose the
condition (55). Note that in the degenerate case, the same argument works, but Eq. (55)
is not needed.
The proof of Eq. (42) proceeds along similar lines, using the functional equations (29)
(31):

k (q), v F + (u v) + (v) Ak (q, v)H

h

+ (q), v Ak (q, v) F
N , (q), v Ak (q, v)F
,
+=
r

h
(q), u v Ak (q, u)H +
j Ak (q, v)Aj (q, u)
j =1

= (q), v k F + (u v) + (v) (q), v (Hk , F )H

+ (q), v
(q) (F )(Hk , F )F (q), v (v)k F

N , (q), v (q), v (Hk , F )F
,
+=

(q), u v (q), u (Hk , F )H
r

(q), v
(q) j (Hj , F )(Hk , F )F
j =1
450

r

(q), v (u)
j (Hk , F )Hj

j =1

= (q), v (q), v (v) (Hk , H )F

+ (u v) + (v) (u) (q), v

(q), u v (q), u (Hk , F )H

+ (q), v (q) +(F )(Hk , F )F (F )(Hk , F )F

(F )(Hk , F )F + (F )(Hk , F )F

+ (q), v
(q) (Hk , F ) (F )F + (F )F
=
((q), v)

(q) Hk , (F )F + (F )F F
=

(q), v
(q) + (q) N , (Hk , F )F
,
+=

+ (q), v (q) + v (v)
N , (Hk , F )F
,
+=

= (q), v (q) + v (q) (v)

(Hk , H )F (Hk , F )H

+ (q), v (q) 2 F Hk , F F+ Hk , F+ F

(q) (Hk , F )
+ (q), v
=

(F )F + (F )F N, F

(q), v
(q)
=

Hk , (F )F + (F )F + N, F F

+ (q), v (q) + v (v)
N , (Hk , F )F
,
+=

= (q), v (q) + v (v)

(Hk , H )F (Hk , F )H
N, (Hk , F )F ,
,
+ =
451
where in the last step, we have used Eqs. (53) and (52). Again, this expression will vanish
provided we impose the condition (55). Note that in the degenerate case, the same argument
works, but Eq. (55) is not needed.
Having concluded the proof of Proposition 1, we pass to analyzing the implications
of the algebraic constraints that we have derived. As it turns out, the conditions stated in
Proposition 1 are sufficiently strong to allow for a complete classification of all possible
solutions. As a by-product, we shall be able to reduce Eq. (52) to the form given in Eq. (60).
A first step in this direction is taken by observing that the signs 1 that appear in Eq. (52)
may without loss of generality be assumed to be independent of :
1 = 1
for all .
(75)
This freedom of choice follows from the possibility of performing a transformation that
changes the signs of the root generators without modifying any of the relations between
generators and structure constants used in the preceding calculations: it is given by
E
E = 11 E ,
H = H ,
= 11 1 N
N,
, ,
1+
N,
F
F = 11 F
and, omitting the primes, brings Eq. (52) into the form
1
H .
F =
2 ||
(76)
Next, let us write down the system obtained from Eq. (53) upon replacing by and
by :
(F )F (F )F = N, F+ ,
(F )F + (F )F = N, F ,
(F )F (F )F = N, F+ ,
(F )F + (F )F = N, F .
(77)
Adding these four equations gives

1
1
F+ F F+ F = N, F+
+ N, F
.
2
2
Inserting Eq. (52) and separating the coefficients of H and H , we conclude that

1
1
1
2 F+ =
N,
N, ,
||
| |
| + |
(78)
plus the same equation with and interchanged. It is to be noted that this derivation
is only valid when = , as stated in Eq. (53): this supplementary condition is also
needed in order to guarantee that H and H are linearly independent but can in fact
452
be eliminated from Eq. (78) since this formula is automatically satisfied when = .
(Indeed, for = the r.h.s. is understood to vanish since 2 and 0 do not belong to the
root system , whereas the l.h.s. vanishes as a consequence of Eq. (51).)
The algebraic equation (78) is identical with a special case of Eq. (42) of Ref. [1],
obtained by replacing the coupling constants g by 1/| |. As has been shown in Section 2.2
of Ref. [1], there is only one type of simple complex Lie algebra g for which there exists a
solution, namely those of the A-series. In particular, g is simplylaced, and all its roots have
the same length, which according to our convention equals 2, and Eq. (76) simplifies
to the form given in Eq. (60). Explicitly, if g = sl(n, C) with hR consisting of the real
diagonal (n n)-matrices, we have = {ab | 1 a = b n} with ab (H ) = Haa Hbb
for H hR and take Eab = Eab where Eab is the matrix whose entry in the ath row and
bth column is 1 while all other entries are 0; then the structure constants Nab,cd = Nab ,cd
are given by
Nab,cd = bc ad ,
and writing
Fab
= Fab ,
(79)
we have
1
1
+
Fab
= (Eaa + Ebb ) + 1n ,
2
n
(80)
and
1
Fab
= (Eaa Ebb ),
2
implying that for 1 = +1,
1
Fab = Ebb + 1n ,
n
while for 1 = 1,
(81)
(82)
1
Fab = Eaa + 1n .
(83)
n
It is then easy to check that F , as defined by Eqs. (80)(83), satisfies all the conditions
stated in Proposition 1. To see this, assume for simplicity that 1 = +1 (noting that the case
1 = 1 can be obtained from this one by replacing F by F , which does not affect the
validity of any of Eqs. (53)(55)). Then assuming, for example, that = ab and = cd ,
the l.h.s. ab (Fcd )Fab cd (Fab )Fcd and r.h.s. Nab,cd Fab +cd of Eq. (53) are both equal
to

1
1
ad Ebb 1n bc Edd 1n ,
n
n
except when a = d and b = c ( = ), where the r.h.s. is understood to vanish while the
l.h.s. does not. Similarly, the formula

n

1
Eaa 1n = 0,
n
a=1
453
allows us to verify Eqs. (54) and (55): the l.h.s. of Eq. (54) becomes

1
1
(Eaa Ebb ) Ebb 1n Eaa 1n
n
n
1a=bn

1
1
=
Eaa Eaa 1n Eaa 1n
n
n
a=1

n

1
1
Ebb Ebb 1n Ebb 1n ,
+
n
n
n

b=1
which vanishes as required, while that of Eq. (55), for = ab , becomes

(Fac Fcb Fcb Fac )
1cn
c=a, c=b

1
1
1
1
Ecc 1n Ebb 1n Ebb 1n Ecc 1n
n
n
n
n
1cn
c=a, c=b

2
1
= Eaa + Ebb 1n Ebb 1n
n
n

1
2
+ Ebb 1n Eaa + Ebb 1n
n
n

1
1
= (Eaa Ebb ) Ebb 1n + Ebb 1n (Eaa Ebb )
n
n
= Hab Fab Fab Hab ,
as required. In this way, we have rederived the main result of Refs. [5,6], which states
that the dynamical R-matrix of the integrable Calogero model associated with the root
system of the simple Lie algebra g = sl(n, C) of the A-series can be gauge transformed to
a numerical R-matrix.
4. Calogero models for symmetric pairs

According to Ref. [1], the standard Lax matrix L and the dynamical R-matrix for the
Calogero models associated with the root system of a symmetric pair (g, ) read
L=
r

pj Hj +
j =1

ig w (q) E ,

1 w ((q))
(E E + E E ),
w (q) K E +
R=
2
w((q))
(84)
(85)
454
for the degenerate model and

L(u) =
r

pj Hj +
j =1

ig (q), u E ,
(86)
r

1
(u v) + (u + v) Hj Hj
R(u, v) =
2
j =1

1
(u v) (u + v) + 2(v) Cz
2

+
(q), v K E

1
(q), u v E E + (q), u v E E ,
2
(87)
for the elliptic model, where = 0 with
0 = { | = },
= { | = }.
As has been shown in Ref. [1], integrability requires the generators K ib0 appearing in
Eqs. (85) and (87) to satisfy the constraints (2). Moreover, writing
1
K = (K K ),
2
we also impose the condition

/ ,
K+ = 0 for such that
(88)
(89)
which follows from Eq. (2) by setting = when g = 0 but turns out to be true in
general, independent of this hypothesis.
Before proceeding with the calculations, we pause to note that the R-matrices given by
Eqs. (85) and (87) have certain symmetry properties with respect to the automorphism
that we want to be preserved under the gauge transformation to R (u, v): in the degenerate
case, R takes values in k m whereas in the elliptic case, R(u, v) is even under the action
of 1 and odd under the action of 1 when these are combined with a change of sign
in the corresponding spectral parameter. This can be achieved by imposing a restriction on
the action of on g or, equivalently, on the gauge potentials Aj : in the degenerate case, g
should take values in K and the Aj should take values in k, whereas in the elliptic case, we
require that

g(u) = g(u),
Aj (u) = Aj (u),
(90)
or in terms of the components of the gauge potentials in the expansion (13),
h

h
Aj (u) = Aj (u),
A
j (u) = Aj (u).
(91)
455
In order to compute the content of Eqs. (11) and (12), we further expand the Cartan part
of the gauge potential according to
h
Aj (u) =
r+s

Akj (u)Hk .
(92)
k=1
Then inserting Eqs. (87), (13) and (92) into Eq. (12), we obtain
0=

k (q), v K E

1
k (q), u v E E
2

1
k (q), u v E E
2
r

j Ak (q, u) Hj
j =1
r

j Ak (q, u)E Hj
j =1

1
(u v) + (u + v) Ak (q, u)[Hj Hj , E 1]
2
r
j =1

1
(u v) + (u + v) Ak (q, v)[Hj Hj , 1 E ]
2
r
j =1
j
1
(u v) (u + v) + 2(v) Ak (q, u)[Cz , Hj 1]
2
r+s
j =1
j
1
(u v) (u + v) + 2(v) Ak (q, v)[Cz , 1 Hj ]
2
r+s
j =1

1
(u v) (u + v) + 2(v) Ak (q, u)[Cz , E 1]
2

1
+
(u v) (u + v) + 2(v) Ak (q, v)[Cz , 1 E ]
2
r+s

j
(q), v Ak (q, u)[K E , Hj 1]
j =1
r+s

j =1

j
(q), v Ak (q, v)[K E , 1 Hj ]
456

(q), v Ak (q, u)[K E , E 1]

(q), v Ak (q, v)[K E , 1 E ]
j
1
(q), u v Ak (q, u)[E E , Hj 1]
2
r+s
j =1
j
1
(q), u v Ak (q, v)[E E , 1 Hj ]
2
r+s
j =1
j
1
(q), u v Ak (q, u)[E E , Hj 1]
2
r+s
j =1
j
1
(q), u v Ak (q, v)[E E , 1 Hj ]
2
r+s
j =1

1
(q), u v Ak (q, u)[E E , E 1]
2

1
(q), u v Ak (q, v)[E E , 1 E ]
2

1
(q), u v Ak (q, u)[E E , E 1]
2

1
(q), u v Ak (q, v)[E E , 1 E ]
2

r

j =1
r

j =1 ,
Using the definition of Cz ,

Cz =
r+s

j =r+1
Hj Hj +

0
E E ,
(93)
457
and Eqs. (16)(20) to carry out the commutators, together with the relation
r+s

r

j Hj = H ,
j =1
j Hj = (H )a ,
j =1
r+s

j Hj = (H )b ,
j =r+1
we arrive at

0=
k (q), v K E

1
k (q), u v E E
2

1
k (q), u v E E
2
r

j Ak (q, u) Hj
j =1
j Ak (q, u)E Hj
j =1
r
1
r

j (u v) + (u + v) Ak (q, u)E Hj
j =1

1
j (u v) + (u + v) Ak (q, v)Hj E
2
r
j =1
r+s
j
1
j (u v) (u + v) + 2(v) Ak (q, u)E E
2
j =1 0
r+s
j
1
j (u v) (u + v) + 2(v) Ak (q, v)E E
2
j =1 0
r+s

1
j (u v) (u + v) + 2(v) Ak (q, u)E Hj
2
j =r+1
r+s

1
j (u v) (u + v) + 2(v) Ak (q, v)Hj E
2
j =r+1

1
(u v) (u + v) + 2(v) Ak (q, u)H E
2
0

1
+
(u v) (u + v) + 2(v) Ak (q, v)E H
2
0
1
+
2
, 0
+

N , (u v) (u + v) + 2(v) Ak (q, u)E + E
(94)
458
1
2

N , (u v) (u + v) + 2(v) Ak (q, v)E E +
, 0

r+s

j
j (q), v Ak (q, v)K E
j =1

(K ) (q), v Ak (q, u)E E

(q), v Ak (q, v)K H

N , (q), v Ak (q, v)K E +

,
+

j
1
j (q), u v Ak (q, u)E E
2
r+s
j =1

j
1
+
j (q), u v Ak (q, v)E E
2
r+s
j =1

j
1
( )j (q), u v Ak (q, u) E E
2
r+s
j =1

j
1
j (q), u v Ak (q, v) E E
2
r+s
j =1

1
(q), u v Ak (q, u)H E
2

1
2

N , (q), u v Ak (q, u)E + E

,
+

1
(q), u v Ak (q, v)E H
2

1
2

N , (q), u v Ak (q, v)E E +

1
(q), u v Ak (q, u) H E
2

1
2
459

N , (q), u v Ak (q, u)E + E

,
+

1
2

1
+
2

N , (q), u v Ak (q, v)E E +
r

j =1
r

j =1 ,
Relabelling summation indices and using cyclicity and antisymmetry of the structure
constants, we can bring this expression into the following form:

0=
k (q), v K E

1
k (q), u v E E
2

1
k (q), u v E E
2
r

j Ak (q, u) Hj
j =1
+
+
r
1
j Ak (q, u)E Hj
j =1

j (u v) + (u + v) Ak (q, u)E Hj
j =1
r
1
r

j (u v) + (u + v) Ak (q, v)Hj E
j =1
r+s
j
1
j (u v) (u + v) + 2(v) Ak (q, u)E E
2
j =1 0
r+s
j
1
+
j (u v) (u + v) + 2(v) Ak (q, v)E E
2
j =1 0
r+s

1
j (u v) (u + v) + 2(v) Ak (q, u)E Hj
2
j =r+1
460
r+s

1
j (u v) (u + v) + 2(v) Ak (q, v)Hj E
2
j =r+1

1
(u v) (u + v) + 2(v) Ak (q, u)H E
2
0

1
(u v) (u + v) + 2(v) Ak (q, v)E H
2
0
2
+
1
2

+
N, (u v) (u + v) + 2(v) Ak (q, u)E E
, 0
+

+
N, (u v) (u + v) + 2(v) Ak (q, v)E E
0 ,
+
r+s

j
j (q), v Ak (q, v)K E
j =1

(K ) (q), v Ak (q, u)E E

(q), v A
k (q, v)K H

N , (q), v Ak (q, v)K E +

,
+

j
1
j (q), u v Ak (q, u)E E
2
r+s
j =1

j
1
j (q), u v Ak (q, v)E E
2
r+s
j =1

j
1
( )j (q), u v Ak (q, u) E E
2
r+s
j =1

j
1
j (q), u v Ak (q, v) E E
2
r+s
j =1

1
(q), u v Ak (q, u)H E
2
1
2
461

+
N, (q), u v Ak (q, u)E E
,
+

1
2
1
2

+
N, (q), u v Ak (q, v)E E

,
+

1
(q), u v Ak (q, u) H E
2
1
2

+
N, (q), u v Ak (q, u) E E
,
+

1
2
1
2

+
N, (q), u v Ak (q, v) E E

,
+
r

j =1
r

j =1 ,
Noting that for 0 , we have j = 0 for 1 j r and (H )a = 0, we can now identify

the components of Eq. (12) along the various subspaces of g g:
The components along h Hj lead to the following system of equations.
For 1 j r:
h
j Ak (q, u) +

j (q), v A
k (q, v)K = 0.
(95)
For r + 1 j r + s:

j (q), v A
k (q, v)K = 0.
The components along h g lead to the following system of equations.
(96)
462
For :

1
k (q), v K + (u v) + (u + v) Ak (q, v)(H )a
2

1
+ (u v) (u + v) + 2(v) Ak (q, v)(H )b
2

h

+ (q), v Ak (q, v) K
N , (q), v Ak (q, v)K

,
+=

1
1
+ (q), v u Ak (q, u)H + (q), v + u Ak (q, u) H
2
2
r

h
+
j Ak (q, v)Aj (q, u) = 0.
(97)
j =1
For 0 :

1
(u v) (u + v) + 2(v) Ak (q, v) Ak (q, u) H
2

N , (q), v Ak (q, v)K = 0.
(98)
,
+=
The components along g Hj lead to the following system of equations.

1 j r:
For ,

1
j Ak (q, u) j (u v) + (u + v) Ak (q, u)
2

1
+ j (q), u v Ak (q, v) + (q), u + v Ak (q, v) = 0.
2
For 0 , 1 j r:
j Ak (q, u) = 0.
r + 1 j r + s:
For ,

(u v) (u + v) + 2(v) Ak (q, u)

(q), u v Ak (q, v) (q), u + v Ak (q, v) = 0.
(99)
(100)
(101)
For 0 , r + 1 j r + s:
Ak (q, u) Ak (q, v) = 0.
The components along g g lead to the following system of equations.
For 0 and = :

h
h
Ak (q, u) Ak (q, v) = 0.
For , 0 with + = 0, no new condition arises, due to Eq. (102).
(102)
(103)
463
For 0 , :

+
1
N, (u v) (u + v) + 2(v) Ak (q, v)
2

(K ) (q), v Ak (q, u)

+

+

1
+ N, (q), v u Ak (q, u) + (q), v + u Ak (q, u)
2
r

j Aj (q, u)Ak (q, v) = 0.

+
(104)
j =1
0 :
For ,

+
N, (u v) (u + v) + 2(v) Ak (q, u)

+
N, (q), u v Ak (q, v)

+

(q), u + v Ak (q, v) = 0.
(105)
if = = :
For , :

(K ) (q), v Ak (q, u)

1
N, (q), u + v A+
(q, u) A+
(q, v)
k
k
2
r

j Aj (q, u)Ak (q, v) = 0,
(106)
j =1
if = (independently of whether = or = ):

1
k (q), u + v (K ) (q), v Ak (q, u)
2
h

h

1
+ (q), u + v Ak (q, u) ( ) Ak (q, v)
2

1
N, (q), u v A
(q, u) (q), u v A
(q, v)
k
k
2
r

j Aj (q, u)A
(q, v) = 0,
+
(107)
k
j =1
if = = :

(K ) (q), v Ak (q, u)

1
+ N, (q), u v A+
(q, u) A+
(q, v)
k
k
2
r

+
j Aj (q, u)A
k (q, v) = 0,
j =1
(108)
464
if = (independently of whether = or = ):

1
k (q), u v + (K ) (q), v Ak (q, u)
2
h

h

1
+ (q), u v Ak (q, u) Ak (q, v)
2

1
(q, u) (q), u + v A
(q, v)
N, (q), u + v A
k
k
2
r

+
(109)
j Aj (q, u)A
k (q, v) = 0,
j =1
if = and = :

(K ) (q), v Ak (q, u)

+

+

1
N, (q), v u Ak (q, u) + (q), u v Ak (q, v)
2

+

+

1
(q, u) (q), u + v Ak
(q, v)
N, (q), u + v Ak
2
r

(110)
j Aj (q, u)Ak (q, v) = 0.
j =1
This is a complicated set of equations which we shall solve in a series of steps.

We begin by considering the algebro-differential equations (99)(102) for the root part
of the gauge potential, which we claim to have the simple solution

((q), u)Mk for
Ak (q, u) =
(111)
,
for 0
Mk
where the coefficients Mk are constants that must be determined from the remaining
equations, subject to the constraint
Mk = Mk
for ,
(112)
imposed in order to guarantee the validity of Eq. (91). (The corresponding constraint for
0 is empty.) Indeed, the statement of Eq. (111) for 0 follows directly from
Eqs. (100) and (102). Similarly, the statement of Eq. (111) for is an immediate
consequence of Eq. (99) in the degenerate case but is somewhat harder to prove in the
elliptic case. To this end, we recast the functional equation (29) into the form

(q), u v (q), v

= (u v) (v) (q), u (q), u
and use the ansatz

Ak (q, u) = (q), u Mk (q, u) for
465
to rewrite the differential equation (99) in the form

(q), u j Mk (q, u) + j (q), u Mk (q, u)

1
j (u v) + (u + v) (q), u Mk (q, u)
2

1
+ j (u v) + (v) (q), u (q), u Mk (q, v)
2

1
+ j (u + v) (v) (q), u (q), u Mk (q, v) = 0,
2
while the algebraic equation (101) takes the form

(u v) (u + v) + 2 (v) (q), u Mk (q, u)

(u v) + (v) (q), u (q), u Mk (q, v)

+ (u + v) (v) (q), u (q), u Mk (q, v) = 0.
Both of these equations can be simplified by using the functional equation (30) and
subtracting (1/2)j times the second from the first, with the result that

j Mk (q, u) = j (u v) + (v) + (q) (q) + u

Mk (q, u) Mk (q, v) ,
(113)
while

(u v) (u + v) + 2 (v) Mk (q, u)

(u v) + (v) (q) + u + (q) Mk (q, v)

+ (u + v) (v) (q) + u + (q) Mk (q, v) = 0.
(114)
Antisymmetrizing the last equation with respect to the exchange of u and u gives

(u v) (u + v) + 2 (v) Mk (q, u) Mk (q, u)

= (u v) + (u + v) + (q) u (q) + u

Mk (q, v) Mk (q, v) .
(115)
In order to analyze the consequences of this relation, we shall use the identity
(x + y) (x) (y) =
1 (x) (y)
,
2 (x) (y)
which can also be written as

(u v) (u + v) + 2 (v) =
(v)
,
(u) (v)
implying that
(u v) + (u + v) + (s u) (s + u)
(u)
(u)
(u)( (s) (v))
=
+
=
.
(u) (v) (s) (u) ( (s) (u))( (u) (v))
466
After rearranging coefficients, we conclude that Eq. (115) can be reformulated as stating
that the function

((q)) (u)
Mk (q, u) Mk (q, u)

(u)
is independent of u. Putting u = (q), we see that it must actually vanish identically, which
is only possible if
Mk (q, u) Mk (q, u) = 0.
Inserting this result back into Eq. (114), we get
Mk (q, u) Mk (q, v) = 0.
Now Eq. (113) implies that Mk is in fact a constant.
For what follows, we shall find it convenient to assemble the constants introduced in
Eq. (111) above into a vector in a0 by writing, for any ,
M =
r

Mj Hj ,
(116)
j =1
so that of course
Mk = (Hk , M ).
(117)
In analogy with Eq. (88), we also introduce the abbreviation

1
M = (M M ).
(118)
2
the generators M a0 are in a sense complementary to the
Note that, for ,
generators K ib0 . (This observation will come to play an important role later on.) In
particular, Eq. (112) amounts to the condition
M = M
for
(119)
which is analogous to the condition

K = K
for
(120)
of Ref. [1].
Now we are ready to state the first main result of this section. Our terminology will
follow that of Ref. [8], according to which roots are called imaginary if = ,
real if = and complex if and are linearly independent, whereas two roots
and are called strongly orthogonal if both + and are not roots (as is well
known, this implies that and are orthogonal in the usual sense).
Proposition 2. The integrable Calogero model associated with the root system of a
symmetric pair (g, ) admits a gauge transformation g from the standard Lax pair of
Olshanetsky and Perelomov and the dynamical R-matrix of Ref. [1] to a new Lax pair
with a numerical R-matrix if and only if (a) the automorphism acts on the root system
in such a way that
467
contains no imaginary roots, i.e., 0 = and = ,

for any complex root in , and are strongly orthogonal, i.e.,
/ ,
and (b) the set of generators K ib0 appearing in Eqs. (85) and (87) can be
complemented by a set of generators M a0 which, taken together, satisfy the algebraic
constraints

M+ = 0,
K+ = 0,
(121)
1
1
(H )b ,
(H )a ,
M =
K =
(122)
2 ||
2 ||
1
(K )K (K )K = (N, K+ + N, K+ ),
2
1
(K )M (M )M = (N, M+ N, M+ )
2
for , such that = , = ,
(123)
as well as the additional algebraic constraints

(H )a K M = 0,
(H )b M M = 0,
(124)
N, M K = (H )a K M (H )b ,
,
+ =
N, M M = (H )a M M (H )a ,
(125)
,
+ =
to be imposed in the case of the elliptic model, where K = (1/2)(K K ) and

M = (1/2)(M M ) as above, with 1 = 1. In this case, the root part and the
Cartan part of the potential Ak = g 1 k g associated with this gauge transformation g are
given by

Ak (q) = w (q) (Hk , M )
(126)
and
h
Ak (q) =
w ((q))
(Hk , M )K
w((q))
(127)
for the degenerate model and by

Ak (q, u) = (q), u (Hk , M )
and
h
Ak (q, u) =

(q) (Hk , M )K (u)Hk
for the elliptic model.
(128)
(129)
468
Note. As we shall show after completing the proof of Proposition 2, Eq. (123)
forces all
roots in to have the same length (which by convention we fix to be 2 ) and also
allows for a choice of basis in which the signs 1 are independent of , so that Eq. (122)
can be simplified as follows:
1
K = 1(H )b ,
2
1
M = 1(H )a .
2
(130)
Proof. With the vector notation introduced above, we can first of all use Eq. (102), with
replaced by + , to conclude that the middle terms in Eqs. (106) and (108) vanish
identically and thus reduce both of them to a single algebraic constraint:
(K ) = (M )
for such that = .
Note that replacing by and adding/subtracting the two equations, we get

K+ = M+ for such that = ,

K = M for such that = .
Using Eq. (89), the first of these can be sharpened as follows:

K+ = 0 = M+ for such that
/ .
(131)
(132)
(133)
(134)
(Indeed, if = so that Eq. (131) no longer applies, Eq. (134) remains correct because
in this case M+ = 0, according to Eqs. (118) and (119).) Next, inserting Eq. (111) together
with the functional equation (25) into the differential equation (95) for the Cartan part of
the gauge potential, we see that this equation can be solved by setting

h
h
(q) Mk K Mk (u),
Ak (q, u) =
(135)
where the Mk (u) are constants that must be determined from the remaining equations,
provided we assume the coefficients Mk to satisfy the relation

j Mk K = 0 for 1 j, k r.
(136)
Converted into a tensor equation, it reads

(H )a K M = 0,
(137)
which leads back to Eq. (136) by taking the scalar product with Hj in the first and with
Hk in the third tensor factor. Note that in the degenerate case, the same argument works,
but Eq. (136)/(137) is not needed. Moreover, Eq. (96) is satisfied as a consequence of the
identity

j f (q) Mk K = 0 for r + 1 j r + s,
469
which is valid for any even function f (such as ): this is easily shown by replacing
by in the sum and noting that ( )j = j for r + 1 j r + s, ( )(q) = (q),
Mk = Mk and K = K , so that

j f (q) Mk K =
j f (q) Mk K for r + 1 j r + s.
Similarly, inserting Eq. (111) together with the functional equation (29) into Eqs. (107)
and (109), we obtain

1
k (q), u (q), v + (u) + (v) (q), u + v
2

(K )Mk (q), u (q), v
h
h
1
(q), u + v Mk (u) ( ) Mk (v)
2

1
+ N, Mk (q), u v 2(q), u
2

(q), v u 2(q), v

(M )Mk (q), u (q), v = 0
and

1
k (q), u (q), v + (u) (v) (q), u v
2

(K )Mk (q), u (q), v
h
h
1
(q), u v Mk (u) Mk (v)
2

1
N, Mk (q), u + v 2(q), u
2

(q), u v 2(q), v

(M )Mk (q), u (q), v = 0
respectively. In both cases, the terms proportional to ((q), u v) cancel provided we
set
h
Mk (u) = (u)Hk ,
(138)
which also guarantees that Eq. (103) is valid, and due to the functional equation (28), the
remaining terms then cancel if we impose the relation
1
1
k + (K )Mk + (M )Mk + N, Mk = 0 for .

2
2
Converted to a vector equation in a0 , it reads
(139)
1
1
(H )a + (K )M + (M )M + N, M = 0 for ,
(140)
2
2
which leads back to Eq. (139) by taking the scalar product with Hk . Even simpler to handle
is Eq. (110), which by insertion of Eq. (111) together with the functional equation (28)
470
reduces to the relation

1
+
+
(M )Mk = 0
N, Mk
(K )Mk N, Mk
2
for , such that = , = .
(141)
Converted to a vector equation in a0 , it reads

1
(K )M (N, M+ N, M+ ) (M )M = 0
2
(142)
which leads back to Eq. (141) by taking the scalar product with Hk . Finally, inserting
Eq. (111) together with the functional equation (29) (more precisely, the difference of two
copies of Eq. (29): one with v u, u v and one with v u, u v) into Eq. (105)
and dividing by ((q), u), we obtain
+
N, Mk
0 ,
= 0 for ,
(143)
which in turn reduces Eq. (104) to
(K )Mk = (M )Mk
for 0 , .
(144)
In the degenerate case, the same conclusion is reached along a slightly different path, since
in this case Eq. (105) is void while Eq. (104) takes the form

+
(K )w (q) Mk + N, Mk w (q) (M )w (q) = 0,
which again leads to Eqs. (143) and (144) since w and w are functionally independent.
Before proceeding to the solution of the remaining equations, let us pause to draw a few
consequences of the algebraic constraints (131)(134), (139)/(140) and (143) derived so
far; this will help us considerably to simplify our further work. For this purpose, we must
distinguish between real and complex roots in :

For real roots ( = ), Eqs. (119) and (120), together with the fact that
K ib0 and hence K = K , imply that M = M , K = K and (K ) =
( K ) = (K ) = (K ), so (K ) = 0. Thus in this case, the last term and
the second term in Eq. (140) drop out, so we get
1
(M )M = (H )a .
2
Applying to this relation and using that in this case, (H )a = H , we conclude that
2(M )2 = (H ) = (, ),
i.e.,
1
(M ) = ||
2
1
and M =
(H )a ,
2 ||
where 1 = 1 = 1 = 1 is a sign factor (1).
(145)
471
For complex roots ( = ), Eq. (131) implies that (K ) = (M ), so

we may rewrite Eq. (140) in the form

1
1
(H )a + 2 M+ M+ 2 M M + N, M = 0.
2
2
But under the substitution , the first three terms in this equation are odd while
the last is even, which forces them to vanish separately. Now we must distinguish two
cases:
if is not a root, the last term drops out and, according to Eq. (134), so does
the second. Thus we get

1
M M = (H )a .
4
Applying to this relation, we conclude that

2
1

4 M = (H )a = (, ),
2
which can be further simplified because + is never a root [9, Ex. F.2, p. 530]
so that in this case, and are (strongly) orthogonal, leading to

1
1
(H )a ,
M = || and M =
(146)
2 2
2 ||
where 1 = 1 = 1 = 1 is a sign factor (1);
if is a root, we arrive at a contradiction, since in this case is a real
root, so that according to the previous item, M cannot vanish. Therefore, this
possibility must be excluded.
Moreover, we see that M can never vanish, so the only way to guarantee the validity of
Eq. (143) is to assume that the sum of a root in and a root in 0 is never a root. Since
we may freely change the sign of , this forces all roots in to be (strongly) orthogonal to
all roots in 0 , which is only possible if one of these two sets is empty, since g is supposed
to be simple and hence must be irreducible. This proves the two restrictions on the action
of on stated in the proposition, namely
0 = ,
= ,
(147)
and, for any root in ,

/
and either = or .
Moreover, Eqs. (145) and (146) can be unified into a single formula
1
(H )a ,
M =
2 ||
(148)
(149)
where 1 = 1 = 1 = 1 is a sign factor (1).

Let us summarize the results obtained so far. With the exception of Eq. (97),
the system of Eqs. (95)(110) has been completely solved in terms of the algebraic
conditions (147), (148), the explicit formulae (126)(129) for the gauge potential with the
472
explicit formula (149) for the odd part M of the coefficient vectors M and the algebraic
constraints (121), (133), (136)/(137) and (141)/(142). Thus we are left with the task of
verifying the implications of Eqs. (14), (15) and (97).
Beginning with Eq. (14), we use the functional equation (25) and Eqs. (147)(149) to
compute

h
h
Ak (q, u)A
k Al (q, u) l Ak (q, u) +
l (q, u)H
=

(q) l (Hk , M )K k (Hl , M )K

(q), u (q), u (Hk , M )(Hl , M )H

1
=
(q) +l (Hk , M )K l (Hk , M )K
2
k (Hl , M )K + k (Hl , M )K
(u)
=
+ (Hk , M )(Hl , M )H (Hk , M )(Hl , M )H
(Hk , M )(Hl , M )H

(q) l Hk , M K+ + l Hk , M+ K
(u)

+ k Hl , M K+ k Hl , M+ K

+ Hk , M Hl , M+ H Hk , M+ Hl , M H
(Hk , M )(Hl , M )(H )b

1
1
+
+
l k K +
k l K
=
(q)
2 ||
2 ||

+
(q) +l Hk , M+ K k Hl , M+ K

(u)
+
2 ||
k Hl , M+
(H )b
2 ||

l Hk , M+ (H )b
(Hk , M )(Hl , M )(H )b .
Obviously, the whole expression will vanish provided we assume that

1
K =
(H )b ,
2 ||
which is complementary to the condition (149) derived previously and that

(H )b M M = 0,
(150)
(151)
473
which is complementary to the condition (137) derived previously. Note that in the
degenerate case, the same argument works, but Eq. (151) is not needed. Note also that
Eq. (133) can now be eliminated because it follows from Eqs. (148), (149) and (150).
For the proof of Eq. (15), the trick is to split the sum over roots coming from the third
and fourth term into various pieces: the contribution with = and, if = , also
the contribution with = , which cancel mutually, the contribution with = and,
if = , also the contribution with = , which combine with the contributions
coming from the first and second term (transformed using the functional equation (30)), and
finally the remaining contributions with = and = : these can be complemented
by terms that also cancel mutually (marked by underlining) and then be combined with the
contributions from the fifth term (transformed using the functional equation (31)):
h

h

k Al (q, u) l Ak (q, u) + Ak (q, u) Al (q, u) Al (q, u) Ak (q, u)

+
N, Ak (q, u)Al (q, u)
,
+ =

= (q), u k (Hl , M ) (q), u l (Hk , M )

+ (q), u
(q) (K )(Hk , M )(Hl , M ) (u)k (Hl , M )

(q), u
(q) (K )(Hl , M )(Hk , M ) (u)l (Hk , M )

N , (q), u (q), u (Hk , M )(Hl , M )
,
+=

= (q), u (q), u (u) (Hk , H )(Hl , M ) (Hl , H )(Hk , M )

+ (q), u (q)

+(K )(Hk , M )(Hl , M )
(K )(Hk , M )(Hl , M )
(1 , )(K )(Hk , M )(Hl , M )
+ (1 , )(K )(Hk , M )(Hl , M )
(K )(Hl , M )(Hk , M )
+ (K )(Hl , M )(Hk , M )
+ (1 , )(K )(Hl , M )(Hk , M )
(1 , )(K )(Hl , M )(Hk , M )

+ (q), u
(q) (Hk , M )
=, =

(K )(Hl , M ) + (M )(Hl , M )
474

(q), u

(q) (Hl , M )
=, =

(K )(Hk , M ) + (M )(Hk , M )

(q) + (q) N , (Hk , M )(Hl , M )
+ (q), u
,
+=

N , (Hk , M )(Hl , M )
(q), u (q) + u (u)
,
+=

= (q), u (q) + u (q) (u)

(Hk , H ) Hl , M+ Hk , M+ (Hl , H )

2(2 , )(K ) Hk , M Hl , M+

+ 2(2 , )(K ) Hk , M+ Hl , M

+ (q), u (q) + u (u)

+2(2 , )(K ) Hk , M Hl , M+

2(2 , )(K ) Hk , M+ Hl , M

,
+ =

+ (q), u

N, (Hk , M )(Hl , M )

=, =

(q), u

(q) (Hk , M )

Hl , (K )M + (M )M

+ N, M

(q) (Hl , M )
=, =

Hk , (K )M + (M )M

N, M

(1 , ) (q), u (q)

+N, (Hk , M )(Hl , M )
N,+ (Hk , M )(Hl , M+ )
+ N, (Hl , M )(Hk , M )

N+, (Hl , M )(Hk , M+ ) .
475
The last term vanishes due to the condition (148), whereas the previous two terms vanish
due to Eq. (142). The first term also vanishes because, as already observed before, M+ = 0
for real roots while, according to Eqs. (133), (134) and (146),

4(K )M = 4 K M = 4 M M = (H )a
for complex roots . Finally, the same reasoning shows that the second term will vanish
provided we assume that

1
N, M M = ||(M M M M ),
(152)
2
,
+ =
which is easily reduced to the second equation in Eq. (125) by noting that
1
(M M M M )
2
= M M+ M+ M = M M M M
and using Eq. (122). Note that in the degenerate case, the same argument works, but
Eq. (152) is not needed.
The proof of Eq. (97) proceeds along similar lines, using the functional equations (29)
(31):

1
k (q), v K + (u v) + (u + v) Ak (q, v)(H )a
2

1
+ (u v) (u + v) + 2(v) Ak (q, v)(H )b
2

h

+ (q), v Ak (q, v) K
N , (q), v Ak (q, v)K
,
+=

1
1
+ (q), v u Ak (q, u)H + (q), v + u Ak (q, u) H
2
2
r

h
+
j Ak (q, v)Aj (q, u)
j =1

= (q), v k K

1
+ (q), v (u v) + (u + v) (Hk , M )(H )a
2

1
+ (q), v (u v) (u + v) + 2(v) (Hk , M )(H )b
2

+ (q), v
(q) (K ) Hk , M K (v)k K

,
+=

N , (q), v (q), v (Hk , M )K
476

1
+ (q), v u (q), u (Hk , M ) (H )a + (H )b
2

1
(q), v + u (q), u (Hk , M ) (H )a (H )b
2
r

(q), v
(q) j (Hj , M )(Hk , M )K
j =1
r

(q), v (u)
j (Hk , M )Hj

j =1

= (q), v (q), v (v) (Hk , H )K

1
(u v) + (u + v) 2(u) (q), v
+
2

+ (q), v u (q), u

(q), v + u (q), u (Hk , M )(H )a

1
(u v) (u + v) + 2(v) (q), v
+
2

+ (q), v u (q), u

+ (q), v + u (q), u (Hk , M )(H )b

+ (q), v (q)

+(K )(Hk , M )K (K )(Hk , M )K

(1 , )(K )(Hk , M )K
+ (1 , )(K )(Hk , M )K
(M )(Hk , M )K + (M )(Hk , M )K
+ (1 , )(M )(Hk , M )K

(1 , )(M )(Hk , M )K

(q) (Hk , M ) (K )K + (K )K
+ (q), v

(q), v
=, =

(q) Hk , (M )M + (K )M K
=, =

(q) + (q) N , (Hk , M )K
(q), v
,
+=

+ (q), v (q) + v (v)
N , (Hk , M )K
,
+=
477

= (q), v (q) + v (q) (v)

(Hk , H )K (Hk , M )(H )b

(2 , ) (K ) (M ) (Hk , M )K
+ (2 , )(K )(Hk , M )K

(2 , )(M )(Hk , M )K

+ (q), v (q) + v (v)

+(2 , ) (K ) (M ) (Hk , M )K
(2 , )(K )(Hk , M )K
+ (2 , )(M )(Hk , M )K

N, (Hk , M )K
,
+ =

+ (q), v
=, =

(q), v

(q) (Hk , M )

(K )K + (K )K N, K

(q)
=, =

Hk , (M )M + (K )M

+ N, M K

+ (1 , ) (q), v (q)

+N, (Hk , M )K N+, (Hk , M )K+

+ N, (Hk , M )K N,+ (Hk , M+ )K .
The last term vanishes due to the condition (148), whereas the previous two terms vanish
due to Eq. (142) and provided we impose the relation
1
(K )K (N, K+ + N, K+ ) (K )K = 0
2
(153)
which is complementary to it. The first term also vanishes because, according to Eqs. (149)
and (150),
(Hk , H )K (Hk , M )(H )b

= (Hk , H )K+ Hk , M+ (H )b

= 1 2 || Hk , M K+ Hk , M+ K

1
= || (Hk , M )K (Hk , M )K ,
2
478
whereas for real roots ( = ), we have K = K , M = M , (K ) = 0

and hence by Eq. (145)

(2 , ) (K ) (M ) (Hk , M )K
+ (2 , )(K )(Hk , M )K
(2 , )(M )(Hk , M )K
= (M )(Hk , M )K (M )(Hk , M )K

1
= || (Hk , M )K (Hk , M )K ,
2
while for complex roots ( = ), we have (K ) = (M ) and hence by
Eq. (146)

(2 , ) (K ) (M ) (Hk , M )K
+ (2 , )(K )(Hk , M )K
(2 , )(M )(Hk , M )K
= 2(M )(Hk , M )K 2(M )(Hk , M )K

1
= || (Hk , M )K (Hk , M )K .
2
Finally, the same reasoning shows that the second term will vanish provided we assume
that

1
(154)
N, M K = ||(M K M K )
2
,
+ =
which is complementary to the condition (152) derived previously and is easily reduced to
the first equation in Eq. (125) by noting that
1
(M K M K )
2
= M K+ M+ K = M K M K
and using Eq. (122). Note that in the degenerate case, the same argument works, but
Eq. (154) is not needed.
Having concluded the proof of Proposition 2, we pass to analyzing the implications
of the algebraic constraints that we have derived. The first thing that suggests itself is to
combine the generators K and M into generators
F = K + M
(155)
which, according to Eqs. (119) and (120), define a -covariant map from to hR :
F = F .
(156)
Then Eq. (121) becomes equivalent to Eq. (51) and Eq. (122) becomes equivalent to
Eq. (52). The relation between Eq. (123) and Eq. (53), however, is more intricate.
479
To approach this question, note that Eq. (53) decomposes naturally into a component
along ib0 ,
(F )K (F )K = N, K+
(157)
and a component along a0 ,

(F )M (F )M = N, M+
for , such that = , = .
(158)
Now observe that Eq. (119) can be used to show that the second equation in Eq. (123) is
equivalent to Eq. (158). Indeed, antisymmetrizing Eq. (123) with respect to the exchange
of and eliminates one of the two terms containing structure constants and leads to
Eq. (158), and conversely, substituting by in Eq. (158) and subtracting the result, we
are led back to the second equation in Eq. (123). On the other hand, using Eq. (120) and
applying the same argument, the first equation in Eq. (123) turns out to be a consequence
of Eq. (157) but is apparently weaker. Similarly, Eq. (124) is part of Eq. (54), from which
it can be obtained by projecting from hR onto a0 in the third tensor factor, that is, by
applying the operator 1 1 (1 )/2, and Eq. (125) is part of Eq. (55), from which it
can be obtained by projecting from hR onto a0 in the first tensor factor, that is, by applying
the operator (1 )/2 1.
Although the conditions stated in Proposition 2 thus seem to be weaker than those stated
in Proposition 1, it turns out that they are still sufficiently strong to allow for a complete
classification of all possible solutions. As a by-product, we shall be able to reduce Eq. (122)
to the form given in Eq. (130). The arguments employed to achieve this are essentially
the same as the ones in the previous section. First, we argue that, as before, the signs 1
that appear in Eq. (122) may without loss of generality be assumed to be independent
of . Next, writing down the system obtained from Eq. (158) upon replacing by
and by , adding the resulting four equations, inserting Eq. (122) and separating
the coefficients of (H )a and (H )a , we arrive at the same formula as in the previous
section, Eq. (78). Once again, it is to be noted that this derivation is only valid when
= , = , as stated in Eq. (158): this supplementary condition is also needed to
guarantee that (H )a and (H )a are linearly independent but can in fact be eliminated from
Eq. (78) since this formula is automatically satisfied when = or = . (Indeed,
for = or = the r.h.s. is understood to vanish since 2, 0 and do not
belong to the root system , whereas the l.h.s. vanishes as a consequence of Eq. (121).)
The statement that for = and = , the generators (H )a and (H )a are
linearly independent, used in the derivation of Eq. (78) given here, can be proved indirectly,
as follows. Suppose that for some pair of roots , satisfying = and = ,
these generators were linearly dependent. Since 0 is empty so that (H )a and (H )a are
both non-zero, this amounts to assuming that there exists a non-zero real number such
that (H )a = (H )a , or equivalently,
= ( ).
(159)
480
Obviously, if both roots are real, Eq. (159) reduces to = , with = 1, a

contradiction. Similarly, if one of the two roots is complex while the other is real, we also
get a contradiction since if, for example, is complex and is real, Eq. (159) becomes
= (1/2)( ) which is excluded since and being strongly orthogonal implies
that the only linear combinations of and which are roots are and . To handle
the case where both roots are complex and hence and as well as and are
strongly orthogonal, we begin by noting that cannot be orthogonal to both and
since otherwise, would be so as well and hence would be orthogonal to ,
which contradicts Eq. (159). Exchanging with and with if necessary, we may
assume without loss of generality that is not orthogonal to and that the factor in
Eq. (159) is positive. With these conventions, taking the scalar product of Eq. (159) with
and with gives
(, ) (, ) = (, ),
(, ) = (, ) (, ),
(160)
implying
(, ) = 2 (, ).
In the root system of an arbitrary simple complex Lie algebra, this forces 2 to be 1, 2, 3,
1/2 or 1/3. But Eq. (159) excludes the possibility of 2 being different from 1 since the
root system of any simple complex Lie algebra is contained in an appropriate lattice formed
by the integer linear combinations of vectors (1/2)ei where the ei are an orthonormal basis
of Rn , so an equation of the form (159) with an irrational value of can only hold if both
sides vanish, which is impossible since 0 is empty. Thus we conclude that = 1, so ,
, and all have the same length and Eq. (159) becomes
= .
(161)
This allows us to determine the -string through . First, cannot be a root since
if it were, it would belong to 0 which is empty. Second, + must therefore be
a root, since and are not orthogonal. Third, + 2 cannot be a root since if it
were, we would have 2(, )/(, ) 2, implying | + |2 0, which is absurd.
Hence and generate a root system of type A2 for which they act as simple roots;
in particular, 2(, )/(, ) = 1. Inserting this conclusion back into Eq. (160), we see
that 2(, )/(, ) = 3, which is only possible if the -string through consists of
four roots, namely , + , + 2 and + 3 (recall that any root string has length
at most 4). But this requires the angle between and to be 150 and forces and
to have different length, contrary to a conclusion reached before.
In this way, we arrive once again at the conclusion that the simple complex Lie algebra
g must belong to the A-series. Moreover, the automorphism that defines the symmetric
pair (g, ) is further restricted by various additional constraints. The first such condition
is that the root generators E in g can be chosen so that E = E for all , which
according to the erratum of Ref. [1] is not only sufficient but also necessary to guarantee
that the proof of integrability given in Ref. [1] really works: this excludes the symmetric
pairs of the A I-series SL(n, R)/SO(n) for which all roots are real and E = E for
481
all . The second condition is that there should be no imaginary roots: this excludes
the symmetric pairs of the A II-series SL(n, H)/Sp(n) as well as the symmetric pairs of the
A III-series of complex Grassmannians SU(p, q)/S(U (p) U (q)) with |p q| 1. The
third and final condition is that for all complex roots , should be strongly orthogonal
to : this excludes the symmetric pairs of the A III-series of complex Grassmannians
SU(p, q)/S(U (p) U (q)) with p = q. On the other hand, it is clear that the symmetric
pairs associated with the Grassmannians SU(n, n)/S(U (n) U (n)) do provide a nontrivial solution: explicitly, we have in the notation employed at the end of the previous
section (with n replaced by 2n) and in Section 3.2 of Ref. [1]
1
1
+
= (Eaa + Ebb + E(a)(a) + E(b)(b)) + 12n ,
Kab
4
2n
1
+
Mab
= (Eaa + Ebb E(a)(a) E(b)(b)),
4
(162)
and
1
= (Eaa Ebb + E(a),(a) E(b)(b)),

Kab
4
1
Mab = (Eaa Ebb E(a)(a) + E(b)(b)),

4
implying that
(163)
1
1
Kab = (Ebb + E(b)(b)) + 12n ,
2
2n
1
Mab = (Ebb E(b)(b)),
2
when 1 = +1, while
(164)
1
1
Kab = (Eaa + E(a)(a)) + 12n ,
2
2n
1
Mab = (Eaa E(a)(a)),
2
when 1 = 1.
(165)
5. Conclusions and outlook

Our analysis of the question whether the known dynamical R-matrices for integrable
Calogero models can be gauge transformed to numerical R-matrices has revealed that this
is possible in some cases but not in alla conclusion that could definitely not be reached
by looking at the standard model associated with the root system of the A-series alone.
In fact, it had been known from previous work that (a) the Calogero models associated
with the root systems of simple complex Lie algebras g are integrable, in the sense of
admitting a Lax representation with a dynamical R-matrix, if and only if g = sl(n, C)
[1] and (b) that this dynamical R-matrix can be gauge transformed to a numerical one
[5,6]. The results reported in this paper show that for the Calogero models associated
482
with the root systems of symmetric pairs (g, ), the situation is more intricate. First of
all, there is still no complete answer to the question which of these models are integrable,
in the sense of admitting a Lax representation with a dynamical R-matrix: the only case
that has been analyzed completely is that of the A III-series of complex Grassmannians
SU(p, q)/S(U (p) U (q)), where integrability has been shown to occur if and only if
|p q| is either 0 or 1. We strongly suspect that this is in fact the only class of symmetric
spaces where integrability prevails, but a rigorous proof of this conjecture is still missing.
What is shown in this paper is that dynamical R-matrices of integrable Calogero models
associated with non-Grassmannian symmetric pairsshould they existcannot be gauge
transformed to numerical R-matrices and, more importantly, that the dynamical R-matrices
of the Grassmannian Calogero models can be gauge transformed to numerical R-matrices
if p = q but not if |p q| = 1. The first case includes the Cn and Dn models, whereas the
second case includes the Bn and BCn models.
In summary, our results show that the question which originally motivated our work
on integrability of the Calogero models, namely the search for an understanding of the
mathematical nature and role of dynamical R-matrices, is still far from a definite answer,
since the attempt to reduce them to numerical R-matrices via gauge transformations is only
partially successful.
Accepting the fact that the role of dynamical R-matrices for our understanding of
integrable systems can apparently not be reduced to that of numerical R-matrices in
disguise, there are many questions that gain new impetus. Continuing to use the Calogero
models as a guideline, we believe that there are several directions in which future work
will be capable of providing new insights into the problem. One of them is the question
of what should be the algebro-differential constraints to be satisfied by a truly dynamical
R-matrix, or in other words, what is the real mathematical status and interpretation of the
dynamical YangBaxter equation. A remarkable fact is that, as will be shown in a separate
publication [10], there is a natural candidate which is gauge invariant. Another promising
direction for research is a further clarification of the relation between Calogero models
and the geodesic flow on symmetric spaces subjected to MarsdenWeinstein phase-space
reduction: this relation should also shed new light on the role of the recently introduced
spin Calogero models [11].
References
[1] M. Forger, A. Winterhalder, Dynamical R-matrices for Calogero models, Nucl. Phys. B 621 (2002) 523
570.
[2] M.A. Olshanetsky, A.M. Perelomov, Completely integrable Hamiltonian systems connected with semisimple Lie algebras, Invent. Math. 37 (1976) 93108.
[3] M.A. Olshanetsky, A.M. Perelomov, Classical integrable finite-dimensional systems related to Lie algebras,
Phys. Rep. 71 (1981) 313400.
[4] A.M. Perelomov, Integrable Systems of Classical Mechanics and Lie Algebras, Vol. 1, Birkhuser, Basel,
1990.
[5] B.Y. Hou, W.L. Yang, The non-dynamical R-matrix structure of the elliptic CalogeroMoser model, Lett.
Math. Phys. 44 (1998) 3541, solv-int/9711008;
B.Y. Hou, W.L. Yang, The non-dynamical R-matrix structure of the elliptic An1 CalogeroMoser model,
J. Phys. A: Math. Gen. 32 (1999) 14751486, q-alg/9711010.
483
[6] L. Fehr, B.G. Pusztai, On the classical R-matrix of the degenerate CalogeroMoser models, Czech. J.
Phys. 50 (2000) 5964, math-ph/9912021v2;
L. Fehr, B.G. Pusztai, The non-dynamical R-matrices of the degenerate CalogeroMoser models, J. Phys.
A: Math. Gen. 33 (2000) 77397759, math-ph/0005021v2.
[7] O. Babelon, C.-M. Viallet, Hamiltonian structures and Lax equations, Phys. Lett. B 237 (1990) 411416.
[8] A.W. Knapp, Lie Groups Beyond an Introduction, Birkhuser, Basel, 1996.
[9] S. Helgason, Differential Geometry, Lie Groups and Symmetric Spaces, Academic Press, New York, 1978.
[10] M. Forger, A. Winterhalder, Gauge transformations of the dynamical YangBaxter equation, in preparation.
[11] L.-C. Li, P. Xu, Spin CalogeroMoser systems associated with simple Lie algebras, C. R. Acad. Sci. Paris
Sr. I 331 (2000) 5560, math.SG/0009180;
L.-C. Li, P. Xu, Integrable spin CalogeroMoser systems, math.QA/0105162.

Complex multiplication symmetry

of black hole attractors
Monika Lynker a , Vipul Periwal b , Rolf Schimmrigk c
a Indiana University South Bend, South Bend, IN 46634, USA
b Gene Network Sciences, Ithaca, NY 14850, USA
c Kennesaw State University, Kennesaw, GA 30144, USA
Received 2 April 2003; accepted 19 May 2003
Abstract
We show how Moores observation, in the context of toroidal compactifications in type IIB string
theory, concerning the complex multiplication structure of black hole attractor varieties, can be
generalized to CalabiYau compactifications with finite fundamental groups. This generalization
leads to an alternative general framework in terms of motives associated to a CalabiYau variety
in which it is possible to address the arithmetic nature of the attractor varieties in a universal way via
Delignes period conjecture.
PACS: 11.25.-w; 11.25.Hf; 11.25.Mj
1. Introduction
During the past few years number theoretic considerations have become useful in string
theory in addressing a variety of problems in string theory, such as the understanding of the
underlying conformal field theory of CalabiYau manifolds [14], the nature of black hole
attractor varieties [5,6], and the behavior of periods under reduction to finite fields [7]. Our
aim in the present paper is to further develop and generalize some of the observations made
by Moore in his analysis of the arithmetic nature of so-called black hole attractor varieties.
The attractor mechanism [813] describes the radial evolution of vector multiplet scalars
of spherical dyonic black hole solutions in N = 2 supergravity coupled to Abelian vector
E-mail addresses: mlynker@iusb.edu (M. Lynker), vipul@gnsbiotech.com (V. Periwal),
netahu@yahoo.com, rschimmr@kennesaw.edu (R. Schimmrigk).
doi:10.1016/S0550-3213(03)00454-1
M. Lynker et al. / Nuclear Physics B 667 [PM] (2003) 484504
485
multiplets. Under particular regularity conditions the vector scalars flow to a fixed point in
their target space. This fixed point is determined by the charge of a black hole, described
by a vector in the lattice of electric and magnetic charges of the N = 2 Abelian
gauge theory. If the N = 2 supergravity theory is derived from a type IIB string theory
compactified on a CalabiYau space, the vector multiplet moduli space is described by the
moduli space M of complex structures of X, and the dyonic charge vector takes values in
the lattice = H3 (X, Z).
Moore observed in the context of simple toroidal product varieties, such as the triple
product of elliptic curves E3 , or the product K3E of a K3 surface and an elliptic curve,
that the attractor condition determines the complex moduli
of the tori to be given by
D),
obtained
by adjoining to the
algebraic numbers in a quadratic imaginary
field
Q(
rational numbers Q an imaginary number D, where D < 0. This is of interest because

for particular points in the moduli space elliptic curves exhibit additional symmetries, they
admit so-called complex multiplication (CM). For compactifications with toroidal factors
Moores analysis then appears to indicate an interesting link between the attractiveness
of varieties in string theory and their complex multiplication properties.
CalabiYau varieties with elliptic factors are very special because they have infinite
fundamental group, a property not shared by CalabiYau manifolds in general. Other
special features of elliptic curves are not present in general either. In particular Calabi
Yau spaces are not Abelian varieties, and they do not, in any obvious fashion, admit
complex multiplication symmetries. Hence, it is not clear how Moores observations can be
generalized. It is this problem which we wish to address in the present paper. In order to do
so we adopt a cohomological approach and view the modular parameter of the elliptic curve
as part of the primitive cohomology. In the case of elliptic curves E this is simply a choice
of view because there exists an isomorphism between the curve itself and its Jacobian,
defined by J (E) = H1 (E, C)/H1 (E, Z) described by the AbelJacobi map j : E J (E).
These varieties are Abelian.
The Jacobian variety of an elliptic (or more general) curve has a natural generalization
to higher-dimensional varieties, defined by the intermediate Jacobian of Griffiths. It would
be natural to use Griffiths construction in an attempt to generalize the elliptic results
described above. In general, however, the intermediate Jacobian is not an Abelian variety
and does not admit complex multiplication. For this and other reasons we will proceed
differently by constructing a decomposition of the intermediate cohomology of the Calabi
Yau, and using this decomposition to formulate a generalization of the concept of complex
multiplication of black hole attractor varieties. To achieve this we formulate complex
multiplication in this more general context by analyzing in some detail the cohomology
group H3 (X) of weighted Fermat hypersurfaces.
The paper is organized as follows. In order to make the presentation more self-contained
we briefly review in Section 2 the physical setting of black hole attractors in type IIB
theories, as well as Moores solution of the K3 E solution of the attractor equations. In
Section 3 we describe the necessary background of Abelian varieties, and in Section 4 we
show how Abelian varieties can be derived from CalabiYau hypersurfaces by showing
that the cohomology of such varieties can be constructed from the cohomology of curves
embedded in these higher-dimensional varieties. This leads us to Abelian varieties defined
by the Jacobians of curves. Such Abelian varieties do not, in general, admit complex
486
multiplication. What can be shown, however, is that Jacobians of projective Fermat curves
split into Abelian factors which do admit complex multiplication. We briefly describe this
construction and generalize the discussion to curves of BrieskornPham type. Combining
these results shows that we can define the complex multiplication type of CalabiYau
varieties with finite fundamental groups via the CM types of their underlying Jacobians. In
Section 5 we indicate some of the arithmetic consequences for CalabiYau varieties that
derive from the emergence of Abelian varieties with complex multiplication in the context
of black hole attractors.
In the process of our analysis we will recover the same fields which Moore uncovered by
considering the fields generated by periods of higher-dimensional varieties. Even though
our approach is very different from Moores, it is not completely unexpected that we should
be able to recover the field of periods by considering the complex multiplication type. The
reason for this is a conjecture of Deligne [14] which states that the field determined by the
periods of a critical motive is determined by its L-function. Because Delignes conjecture
is important for our general view of the issue at hand, we briefly describe this conjecture
in Section 6 in order to provide the appropriate perspective. Delignes conjecture is in fact
a theorem in the context of projective Fermat hypersurfaces [16], but has not been proven
in the context of weighted hypersurfaces. Mathematically, our results in essence can be
viewed as support of this conjecture even in this more general context. In Section 7 we
summarize our results and indicate possible generalizations.
2. Arithmetic of elliptic attractor varieties

2.1. Attractor varieties
In this paper we consider type IIB string theory compactified on CalabiYau threefold
varieties. The field content of the string theory in 10D space X10 splits into two sectors
according to the boundary conditions on the world sheet. The NeveuSchwarz fields
are given by the metric g (X10 , T X10 T X10 ), an antisymmetric tensor field
B (X10 , 2 ) and the dilaton scalar C (X10 , R). The Ramond sector is spanned by
even antisymmetric forms Ap (X10 , p ) of rank p zero, two, and four. Here p X
denotes the bundle of p-forms over the variety X.
In the context of the black hole solutions considered in [8], the pertinent sectors
are given by the metric and the five-form field strength F of the RamondRamond 4form A4 . The metric is assumed to be a static spherically symmetric spacetime which
is asymptotically Minkowskian and describes an extremally charged black hole, leading to
the ansatz

ds 2 = e2U (r) dt dt + e2U (r) dr dr + r 2 2 ,
(1)
where r is the spatial three-dimensional radius, 2 is the 2D angular element, and the
asymptotic behavior is encoded via eU (r) for r . The expansion of the fiveform F leads to a number of different 4D fields, the most important in the present context
being the field strengths F L of four-dimensional Abelian fields, the number of which
depends on the dimension of the cohomology group H3 (X) via

L
A4mnp (x, y) =
A4L
(x)mnp (y),
487
(2)
where {L }L=1,...,b3 is a basis of H3(X). This is usually written in terms of a symplectic

basis {a , a }a=0,...,h2,1 , for which X a b = ab , leading to an expansion of the field
strength of the form
F(x, y) = Fa (x) a Ga (x) a .
(3)
Being a five-form in ten dimensions the field strength F admits (anti)self-duality constraints
with respect to Hodge duality, F = 10 F. The ten-dimensional Hodge operator 10
factorizes into a 4D and a 6D part 10 = 4 6 . A solution to the anti-selfduality constraint
in 10D as well as the Bianchi identity dF = 0 can be obtained by considering [5]

F = Re E 2,1 + 0,3 ,
(4)
where [17]
e2U (r)
(5)
dt dr
r2
is a 2-form for which the four-dimensional Hodge duality operator leads to 4 E = iE.
Two standard maneuvers to derive the dynamics of a string background configuration
are provided by the reduced IIB effective action with a sort of small superspace ansatz [11],
and the supersymmetry variation constraints of the fermions in non-trivial backgrounds, in
particular the gravitino and gaugino variations. We adopt the notation of [18]. Defining an
inner product , on H3 (X) via

, = ,
(6)
E q sin d d iq
the gravitino equation involves the integrated version of the 5-form field strength defined
as [19,20]

T = eK/2 , F = eK/2 Ga Fa (x) za G
(7)
a (x) ,
with Khler potential

a z a Ga .
= i za G
eK = i,
(8)
Here the second equation is written in terms of the periods

za = , a = ,
Aa
Ga = , a =
(9)
Ba
with respect to a symplectic homological basis {Aa , Ba }a=0,...,h2,1 H3 (X) which is dual
to the cohomological basis {a , a }a=0,...,h2,1 H3 (X). The holomorphic three-form thus
488
can be expanded as
= za a Ga a .
(10)
The supersymmetry transformation of the gravitino A = A dx can then be written

in terms of the components of T as
A
A = DA + dx T
(5) ,
where denotes the covariant Dirac matrices and

1 ab
D = dx D = dx ab + iQ
4
(11)
(12)
is a derivative covariant with respect to both the Lorentz and the Khler transformations
in terms of the spin connection ab and the Khler connection Q . The variation of the
gaugino of the Abelian multiplets takes the form
i
(5)A .
iA = i zi 5 A + Gi
(13)
2
Plugging these ingredients into the supersymmetry transformation behavior of the gravitino
and the gaugino fields, and demanding that the vacuum remains fermion free, leads to the
following equations for the moduli zi and the spacetime function U (r)
dU
= eU |Z|,
d
dzi
= 2eU g i j j |Z|,
d
(14)
where = 1/r, gi j = i j K is the metric derived from the Khler potential K, and

Z( ) = eK/2 = eK/2
(15)
is the central charge of the cycle H3 (X) with Poincar dual H3 (X). To make
the moduli and charge dependence of the central charge explicit one can alternatively view
Z( ) as the integral of the graviphoton form

Z za , pa , qa = T = eK/2 Ga pa za qa
(16)
S2
in terms of the charges

a
a
p = F ,
qa = G
a.
S2
(17)
S2
The fixed point condition of the attractor equations can be written in a geometrical way
as the Hodge condition
H3 (X, Z) = 3,0 + 0,3 .
(18)
489
this can be formulated as

Writing 3,0 = i C
a C z a ,
ipa = Cz
a ,
a CG
iqa = CG
(19)
where C = eK/2 Z.
The system (19) describes a set of b3 (X) charges (pa , qa ) determined by the physical 4dimensional input, which in turn determines the complex periods of the CalabiYau variety.
Hence, the system should be solvable. The interesting structure of the fixed point which
emerges is that the central charges are determined completely in terms of the charges of the
four-dimensional theory. As a consequence the 4D geometry is such that the horizon is a
moduli independent quantity. This is precisely as expected because the black hole entropy
should not depend on adiabatic changes of the environment [21].
2.2. Arithmetic of attractor elliptic curves
In Ref. [5] Moore noted that two types of solutions of the attractor equations have
particularly interesting properties. The first of these is provided by the triple product
of a torus, while the second is a product of a K3 surface and a torus. Both solutions
are special in the sense that they involve elliptic curves. In the case of the product
threefold X = K3 E the simplifying feature is that via Knneths theorem one finds
H3 (K3 E)
= H2 (K3) H1 (E), and therefore the cohomology group of the threefold
in the middle dimension is isomorphic to two copies of the cohomology group H2 (K3)
because H1 (E) is two-dimensional. The attractor equations for such threefolds have been
considered in [22]. The resulting constraints determine the holomorphic form of both
factors in terms of the charges (p, q) of the fields. The complex structure of the elliptic
curve E = C/(Z + Z) is solved as

p q + Dp,q
p,q =
(20)
,
p2
where Dp,q = (p q)2 p2 q 2 is the discriminant of a BPS state labelled by
= (p, q) H3 (K3 E, Z).
(21)
The holomorphic two form on K3 is determined as 2,0 = C(q p), where C is a

constant.
Moore makes the interesting observation that this result is known to imply that
the elliptic curve determined by the attractor equation is distinguished by exhibiting a
particularly symmetric structure. Elliptic curves are groups and therefore one can consider
the endomorphism algebra End(E). This algebra can take one of three forms. Generally,
End(E) is just the ring Z of rational integers. For special curves however there are two
other possibilities for which End(E) is either an order of a quadratic imaginary field F ,
i.e., it is a subring OF which generates F as a Q-vector space and is finitely generated as a
Z-module, or it is a maximal order in a quaternion algebra. The latter possibility can occur
only when the field K over which E is defined has positive characteristics. Elliptic curves
490
are said to admit complex multiplication if the endomorphism algebra is strictly larger than
the ring of rational integers.1
The point here is that the property of complex multiplication appears if and only if the
j -invariant j ( ) of an elliptic curve E = C/(Z + Z) is an algebraic integer, i.e., it solves
a polynomial equation with rational coefficients such that the coefficient of the leading
term is unity. This happens if and only if the modulus is an imaginary quadratic number,
i.e., it solves an equation A 2 + B + C = 0 for A, B, C Z. The j -invariant of the elliptic
curve E can be defined in terms of the Eisenstein series

1
1
Ek ( ) =
(22)
2
(m + n)k
m,nZ
m,n coprime
as
j ( ) =
E4 ( )3
,
( )
(23)
where 1728( ) = E4 ( )3 E6 ( )2 . In general j ( ) does not take algebraic values, not

to mention values in an imaginary quadratic field. Even for algebraic is j ( ) in general a
transcendental number unless is imaginary quadratic. Thus we see that in the framework
of toroidal compactification the solutions of the attractor equations can be characterized
as varieties which are unusually symmetric
and which admit complex multiplication by a
quadratic imaginary field F = Q(i |D|).

Once this is recognized several classical results about elliptic curves with complex
multiplication are available to illuminate the nature of the attractor variety. Exploring these
consequences is of interest because they provide tools that allow a characterization of
attractor varieties that involve elliptic factors. The nature of attractor varieties without
elliptic factors is at present not understood. Generalizations of the arithmetic results
obtained in the elliptic context provide a framework in which CalabiYau varieties with
finite fundamental groups (which may be trivial) can be explored.
One of the important number theoretic results associated to elliptic curves with complex
multiplication is that the extension F (j ( )) obtained by adjoining the j -value to F is the
maximal unramified extension of F with an Abelian Galois group, i.e., the Hilbert class
field. Geometrically there is a Weierstrass model, i.e., a projective embedding of the elliptic
curve of the form
y 2 + a1 xy + a3 y = x 3 + a2 x 2 + a4 x + a6
(24)
that is defined over this extension F (j ( )).

Even more interesting is that it is possible to construct from the geometry of the elliptic
curve the maximal Abelian extension Fab of F by considering the torsion points Etor on the
curve E, i.e., points of finite order with respect to the group law. The field F (j ( ), Etor ),
defined by the j -function and the torsion points, is in general not an Abelian extension of
F , but contains the maximal Abelian extension Fab . It is possible to isolate Fab by mapping
1 A brief review of the arithmetic theory of elliptic curves can be found in [23]. A more extended source [24].
491
the torsion points via the Weber function

E : E P1 ,
(25)
of the curve E, whose definition depends on the automorphism group of the curve.
Assuming that the characteristic of the field F is different from 2 or 3, the elliptic curve
can be embedded via the simplified Weierstrass form
y 2 = x 3 + Ax + B.
If the discriminant

= 16 4A3 + 27B 2
(26)
(27)
does not vanish the elliptic curve is smooth, and the automorphism group Aut(E) can be
shown to take one of the following forms, depending on the value of the j -invariant
if j (E) = 0 or 1728, i.e., AB = 0

{1}
if j (E) = 1728, i.e., B = 0
,
Aut(E) = {1, i}
(28)
{1, 3 , 32 } if j (E) = 0, i.e., A = 0

where 3 = e2i/3 is a primitive third root of unity.
The Weber function can then be defined as
AB
x(p) if j (E) = 0 or 1728
2
A
2
E (p) = x (p) if j (E) = 1728
B 3
x
(p)
if
j
(E)
=
0
(29)
and the Hilbert class field F (j ( )) can be extended to the maximal Abelian extension Fab
of F by adjoining the Weber values of the torsion points

Fab = F j ( ), E (t) t Etor .
(30)
We see from these results that the attractor equations pick out special elliptic curves with
an enhanced symmetry group. This is an infinite discrete group which in turn leads to a rich
arithmetic structure. It is this set of tools which we wish to generalize to the framework of
CalabiYau varieties proper, i.e., those with finite fundamental group.
3. Abelian varieties with complex multiplication

We first review some pertinent definitions of Abelian varieties. An Abelian variety over
some number field K is a smooth, geometrically connected, projective variety which is
also an algebraic group with a group law A A A defined over K. A concrete way
to construct Abelian varieties is via complex tori Cn / with respect to some lattice ,
that is not necessarily integral, and admits a Riemann form. The latter is defined as an
R-bilinear form , on Cn such that x, y takes integral values for all x, y , and
satisfies the relations x, y = y, x. Furthermore, x, iy is a positive symmetric form,
not necessarily non-degenerate. The result then is that a complex torus Cn / has the
492
structure of an Abelian variety if and only if there exists a non-degenerate Riemann form
on Cn /.
A special class of Abelian varieties are those of CM type, so-called complex
multiplication type. The reason these varieties are special is because, as in the lowerdimensional case of elliptic curves, certain number theoretic question can be addressed
in a systematic fashion for this class. Consider a number field F over the rational numbers
Q and denote by [F : Q] the degree of the field F over Q, i.e., the dimension of F over
the subfield Q. An Abelian variety A of dimension n is called a CM-variety if there exists
an algebraic number field F of degree [F : Q] = 2n over the rationals Q which can be
embedded into the endomorphism algebra End(A) Q of the variety. More precisely, a
CM variety is a pair (A, ) with : F End(A) Q an embedding of F . It follows
from this that the field F necessarily is a CM field, i.e., a totally imaginary quadratic
extension of a totally real field. The important ingredient here is that restriction to (F )
End(A) Q is equivalent to the direct sum of n isomorphisms 1 , . . . , n Iso(F, C)
such that Iso(F, C) = {1 , . . . , n , 1 , . . . , n }, where denotes complex conjugation.
These considerations lead to the definition of calling the pair (F, {i }) a CM type, in the
present context, the CM type of a CM variety (A, ).
The context in which these concepts will appear in this paper is provided by varieties
which have complex multiplication by a cyclotomic field F = Q(n ), where n denotes
the cyclic group generated by a primitive nth root of unity n . The field Q(n ) is the
imaginary quadratic extension of the totally real field Q(n + n ) = Q(cos(2/n)) and
therefore is a CM field. The degree of Q(n ) is given by [Q(n ) : Q] = (n), where
(n) = #{m N | m < n, gcd(m, n) = 1} is the Euler function. Hence, the Abelian
varieties we will encounter will have complex dimension (n)/2. Standard references for
Abelian varieties with complex multiplication have been provided by Shimura [2527].
In the following sections we first reduce the cohomology of the BrieskornPham
varieties to that generated by curves and then analyze the structure of the resulting weighted
curve Jacobians.
4. Abelian varieties from BrieskornPham type hypersurfaces

4.1. Curves and the cohomology of threefolds
The difficulty of higher-dimensional varieties is that there is no immediate way to
recover Abelian varieties, thus making it non-obvious how to generalize the concept of
complex multiplication from one-dimensional CalabiYau varieties, which are Abelian
varieties, to K3 surfaces and higher-dimensional spaces. As a first step we need to
disentangle the Jacobian of the elliptic curve from the curve itself. This would lead us
to the concept of the middle-dimensional cohomology, more precisely the intermediate
(Griffiths) Jacobian which is the appropriate generalization of the Jacobian of complex
curves. The problem with this intermediate Jacobian is that it is not, in general, an Abelian
variety.
We will show now that it is possible nevertheless to recover Abelian varieties as the
basic building blocks of the intermediate cohomology in the case of weighted projective
493
hypersurfaces. The basic reason for this is that the cohomology H3 (X) for these varieties
decomposes into the monomial part and the part coming from the resolution. The monomial
part of the intermediate cohomology can be obtained from the cohomology of a projective
hypersurface of the same degree by realizing the weighted projective space as a quotient
variety with respect to a product of discrete groups determined by the weights of the
coordinates. For projective varieties

d
Xdn = (z0 , . . . , zn+1 ) Pn+1 z0d + + zn+1
= 0 Pn+1
(31)
it was shown in [29] that the intermediate cohomology can be determined by lowerdimensional varieties in combination with Tate twists by reconstructing the higherdimensional variety Xdn of degree d and dimension n in terms of lower-dimensional
varieties Xdr and Xds of the same degree with n = r + s. Briefly, this works as follows.
The decomposition of Xdn is given as

Xdr+s
= BZ1 ,Z2 Y1 Xdr Xds /d ,
(32)
which involves the following ingredients.

(1) Y1 (Xdr Xds ) denotes the blow-up of Xdr Xds along the subvariety
Y = Xdr1 Xds1 Xdr Xds .
(33)
The variety Y is determined by the fact that the initial map which establishes the
relation between the three varieties Xdr+s , Xdr , Xds is defined on the ambient spaces
as
(x0 , . . . , xr+1 ), (y0 , . . . , ys+1 ) (x0 ys+1, . . . , xr ys+1, xr+1 y0 , . . . , xr+1 ys ).
(34)
This map is not defined on the subvariety Y ;
(2) Y1 (Xdr Xds )/d denotes the quotient of the blow-up Y1 (Xdr Xds ) with respect
to the action of

d : (x0 , . . . , xr , xr+1 ), (y0 , . . . , ys , ys+1 )

(x0 , . . . , xr , xr+1 ), (y0 , . . . , ys , ys+1 ) ;
(3) BZ1 ,Z2 ((Y1 (Xdr Xds ))/d ) denotes the blow-down in Y1 (Xdr Xds )/d of the two
subvarieties
Z1 = Pr Xds1 ,
Z2 = Xdr1 Ps .
This construction leads to an iterative decomposition of the cohomology which takes

the following form. Denote the Tate twist by
Hi (X)(j ) := Hi (X) W j
(35)
494
with W = H2 (P1 ) and let Xdr+s be a Fermat variety of degree d and dimension r + s.
Then
r
s

Hr+s2j Xdr1 (j )
Hr+s2k Xds1 (k)
Hr+s Xdr+s
j =1
k=1

= Hr+s Xdr Xds d Hr+s2 Xdr1 Xds1 (1).
(36)
This allows us to trace the cohomology of higher-dimensional varieties to that of

curves.
Weighted projective hypersurfaces can be viewed as resolved quotients of hypersurfaces
embedded in ordinary projective space. The resulting cohomology has two components,
the invariant part coming from the projection of the quotient, and the resolution part.
As described in [30], the only singular sets on arbitrary weighted hypersurface Calabi
Yau threefolds are either points or curves. The resolution of singular points contributes to
the even cohomology group H2 (X) of the variety, but does not contribute to the middledimensional cohomology group H3 (X). Hence, we need to be concerned only with the
resolution of curves (see, e.g., [31]). This can be described for general CY hypersurface
threefolds as follows. If a discrete symmetry group Z/nZ of order n acting on the threefold
leaves invariant a curve then the normal bundle has fibres C2 and the discrete group induces
an action on these fibres which can be described by a matrix

mq
0
,
(37)
0
m
where is an nth root of unity and (q, n) have no common divisor. The quotient
C2 /(Z/nZ) by this action has an isolated singularity which can be described as the singular
set of the surface in C3 given by the equation

nq
.
S = (z1 , z2 , z3 ) C3 z3n = z1 z2
(38)
The resolution of such a singularity is completely determined by the type (n, q) of the
action by computing the continued fraction of qn
n
1
= b1
[b1, . . . , bs ].
q
b2 1
.. 1
.
(39)
bs
The numbers bi specify completely the plumbing process that replaces the singularity
and in particular determine the additional generator to the cohomology H (X) because
the number of P1 s introduced in this process is precisely the number of steps needed in
the evaluation of qn = [b1, . . . , bs ]. This can be traced to the fact that the singularity is
resolved by a bundle which is constructed out of s + 1 patches with s transition functions
that are specified by the numbers bi . Each of these glueing steps introduces a sphere,
which in turn supports a (1, 1)-form. The intersection properties of these 2-spheres are
described by HirzebruchJung trees, which for a Z/nZ action is just an SU(n + 1) Dynkin
diagram, while the numbers bi describe the intersection numbers. We see from this that
495
the resolution of a curve of genus g thus introduces s additional generators to the second
cohomology group H2 (X), and g s generators to the intermediate cohomology H3 (X).
Hence, we have shown that the cohomology of weighted hypersurfaces is determined
completely by the cohomology of curves. Because the Jacobian, which we will describe in
the next subsection, is the only motivic invariant of a smooth projective curve this says that
for weighted hypersurfaces the main motivic structure is carried by their embedded curves.
We will come back to the motivic structure of CalabiYau varieties in Section 6.
4.2. Cohomology of weighted curves
For smooth algebraic curves C of genus g the de Rham cohomology group H1dR (C)
decomposes (over the complex number field C) as

H1dR (C)
(40)
= H0 C, 1 H1 (C, O).
The Jacobian J (C) of a curve C of genus g can be identified with
J (C) = Cg /,
(41)
where is the period lattice

:= . . . , i , . . .
i=1,...,g

a H1 (C, Z), i H0 C, 1 ,

(42)
where the i form a basis. Given a fixed point p0 C on the curve there is a canonical
map from the curve to the Jacobian, called the AbelJacobi map
: C J (C),
(43)
defined as

p
p . . . ,

i , . . . mod .
(44)
p0
We are interested in curves of BrieskornPham type, i.e., curves of the form

Cd = x d + y a + zb = 0 P(1,k,G)[d],
(45)
such that a = d/k and b = d/G are positive rational integers. Without loss of generality we
can assume that (k, G) = 1. The genus of these curves is given by
(d k)(d G) + (kG d)
1
g(Cd ) = (2 ) =
(46)
.
2
2kG
For non-degenerate curves in the configurations P(1,k,G)[d] the set of forms

H1dR P(1,k,G)[d]

1 r d 1,

= r,s,t = y s1 zt d/Gdy r + ks + Gt = 0 mod d, 1 s dk 1,
(47)
d
1t G 1
496
defines a basis for the de Rham cohomology group H1dR (Cd ) whose Hodge split is given
by

H0 Cd , C1 = {r,s,t | r + ks + Gt = d},
H1 (Cd , OC ) = {r,s,t | r + ks + Gt = 2d}.
(48)
In order to show this we view the weighted projective space as the quotient of projective
space with respect to the actions Zk : [0 1 0] and ZG : [0 0 1], where we use the abbreviation
Zk = Z/kZ and for any group Zr the notation [a, b, c] indicates the action

[a, b, c] : (x, y, z) a x, b y, c z ,
(49)
where is a generator of the group. This allows us to view the weighted curve as the
quotient of a projective Fermat type curve

0 1 0
P(1,k,G)[d] = P2 [d]/Zk ZG :
(50)
.
0 0 1
These weighted curves are smooth and hence their cohomology is determined by
considering those forms on the projective curve P2 [d] which are invariant with respect
to the group actions. A basis for H1dR (P2 [d]) is given by the set of forms

H1 P2 [d] = r,s,t = y s1 zt d dy 0 < r, s, t < d,

r + s + t = 0 (mod d), r, s, t N .
(51)
Denote the generator of the Zk action by and consider the induced action on r,s,t
Zk : r,s,t s r,s,t .
(52)
It follows that the only forms that descend to the quotient with respect to Zk are those for
which s = 0( mod k). Similarly we denote by the generator of the action ZG and consider
the induced action on the forms r,s,t
ZG : r,s,t t d r,s,t .
(53)
Again we see that the only forms that descend to the quotient are those for which
t = 0 (mod G).
4.3. Abelian varieties from weighted Jacobians
Jacobian varieties in general are not Abelian varieties with complex multiplication.
The question we can ask, however, is whether the Jacobians of the curves that determine
the cohomology of the CalabiYau varieties can be decomposed such that the individual
factors admit complex multiplication by an order of a number field. In this section we show
that this is indeed the case and therefore we can define the complex multiplication type of
a CalabiYau variety in terms of the CM types induced by the Jacobians of its curves.
497
It was shown by Faddeev [32]2 that the Jacobian variety J (Cd ) of Fermat curves
Cd P2 splits into a product of Abelian factors AOi

J (Cd )
A Oi ,
(54)
=
Oi I /(Z/dZ)
where the set I provides a parametrization of the cohomology of Cd , and the sets Oi are
orbits in I of the multiplicative subgroup (Z/dZ) of the group Z/dZ. More precisely it
was shown that there is an isogeny

i : J (Cd )
(55)
A Oi ,
Oi I /(Z/dZ)
where an isogeny i : A B between Abelian varieties is defined to be a surjective

homomorphism with finite kernel. In the parametrization used in the previous subsection I
is the set of triplets (r, s, t) in (51) and the periods of the Fermat curve have been computed
by Rohrlich [35] to be

s t
1
(56)
,
1 s 1 t j s+kt ,
r,s,t = B
d
d d
Aj B k
where is a primitive dth root of unity, and

1
B(u, v) =
t u1 (1 v)v1 dt
(57)
is the classical beta function. A, B are the two automorphism generators

A(1, y, z) = (1, y, z),
B(1, y, z) = (1, y, z)
(58)
and is the generator of H1 (Cd ) as a cyclic module over Z[A, B]. The period lattice of
the Fermat curve therefore is the span of

1
r s
,
, 0 j, k d 1.
. . . , j r+ks 1 r 1 s B
,...
1r,s,t d1
d
d d
r+s+t =d
(59)
The Abelian factor A[(r,s,t )] associated to the orbit Or,s,t = [(r, s, t)] can be obtained as the
quotient
A[(r,s,t )] = C(d0 )/2 /r,s,t ,
(60)
where d0 = d/ gcd(r, s, t) and the lattice r,s,t is generated by elements of the form

as at
as
at 1
B
,
1
a (z) 1
(61)
,
d
d
d
2 More accessible are the references [3335] on the subject.
498
where z Z[d0 ], a Gal(Q(d0 /Q)) runs through subgroups of the Galois group of the
cyclotomic field Q(d0 ) and x is the smallest integer 0 x < 1 congruent to x mod d.
We adapt this discussion to the weighted case. Denote the index set of triples (r, s, t)
parametrizing the one-forms of the weighted curves Cd P(1,k,G)[d] again by I. The cyclic
group (Z/dZ) again acts on I and produces a set of orbits

Or,s,t = (r, s, t) I/(Z/dZ) .
(62)
Each of these orbits leads to an Abelian variety A[(r,s,t )] of dimension
1
dim A[(r,s,t )] = (d0 ),
(63)
2
and complex multiplication with respect to the field F[(r,s,t )] = Q(d0 ), where d0 =
d/ gcd(r, ks, Gt). This leads to an isogeny

A[(r,s,t )].
i : J P(1,k,G)[d]
(64)
[(r,s,t )]I /(Z/dZ)
The complex multiplication type of the Abelian factors Ar,s,t of the Jacobian J (C) can be
identified with the set

Hr,s,t := a (Z/dZ) ar + aks + aGt = d
(65)
via a homomorphism from Hr,s,t to the Galois group. More precisely, the CM type is
determined by the subgroup Gr,s,t of the Galois group of the cyclotomic field that is
parametrized by Hr,s,t

Gr,s,t = a Gal Q(d0 )/Q a Hr,s,t
(66)
by considering

F, {a } = Q(d0 ), a a Gr,s,t .
(67)
5. Arithmetic of Abelian varieties

Abelian varieties with complex multiplication have special properties because of their
particular symmetries. It turns out that even though the theory of CM fields associated to
higher-dimensional varieties is not as complete as the theory associated to elliptic curves
with complex multiplication, a number of key results of the elliptic theory have been
generalized to Abelian varieties, mostly by Shimura.
Suppose that the Abelian variety A of dimension n has complex multiplication by the
ring of integers OF of some CM field F (or by an order in F ), and that the CM type
of the variety is given by (F, {i }i=1,...,n ). The arithmetic structure induced by higherdimensional varieties is concerned not with F itself but the so-called reflex field F , which
depends
! not only on the field F , but also on the CM type of F . F is defined as the extension
Q( ni=1 a i ) of the field of rational numbers by adjoining the traces of elements a F .
The higher-dimensional analog of the elliptic field of moduli then gives an unramified
Abelian extension of the field F [26]. Even though ramified class fields over F can be
499
obtained as well, the theory leads to less complete results because it does not give all
Abelian extensions of F . For this reason Hilberts twelfth problem is still not solved for
CM fields.
The basic question is whether there is a simple way to characterize the kind of subfield
of the maximal Abelian extension Fab of a CM field F that can be obtained by adjoining
to the reflex field the moduli fields of Abelian varieties as well as the points of finite order.
A nice result in this direction has been obtained by Wei [36]. In brief, her theorem states
that given a CM field F with totally real subfield FR , the field Fmod generated by the
moduli and torsion points of all polarized Abelian varieties of CM type whose reflex field
is contained in F , is the subfield of Fab that is fixed under the subgroup H of the Galois
group Gal(Fab /F ) generated by the verlagerungs map

R Gal Q/F

Ver : Gal Q/F
(68)
.
ab
More concisely,
Fmod = (Fab )H .
(69)
The verlagerungs map Ver involved here is a general construction which assigns to a
subgroup H of a group G a homomorphism between the Abelianizations Gab = G/(G, G)
and Hab = H /(H, H ) of the pair of groups
Ver : Gab Hab .
(70)
Consider a system of C of representatives for the left cosets of H in G. For each g G

decompose the translate ga for any a C as ga = a ! ga , with ga H and a ! C. The
verlagerungs map is then defined as

ga mod (H, H ).
Ver g mod (G, G) =
(71)
aC
More details can be found in [37,38].

We see from this that, even though the results are weaker, the generalization from the
imaginary quadratic fields of one-dimensional Abelian varieties to the CM fields of higher
dimensions allows for a fairly nice characterization.
6. Delignes period conjecture

Our focus in this paper is on the fields of complex multiplication that we derive from
the Abelian varieties which we construct from CalabiYau varieties. In Moores analysis of
higher-dimensional manifolds the focus is on fields derived from the periods of the variety.
In this section we describe how the period approach can be recovered from our higherdimensional complex multiplication point of view via Delignes conjecture formulated
in [14]. Precursors to Delignes formulation can be found in Shimuras work [3941].
Delignes conjecture in its motivic formulation is also useful in the present context
because it allows us to provide a general perspective for our results which will furnish
what we expect to be a useful general framework in which to explore further the arithmetic
500
nature of attractor varieties. Motives are somewhat complicated objects whose status is
reminiscent of string theory: different realizations are used to probe what is believed to
be some yet unknown unifying universal cohomology theory of varieties which satisfies
a number of expected functorial properties. More precisely, motives are characterized
by a triplet of different cohomology theories together with a pair of compatibility
homomorphisms. In terms of these ingredients a motive then can be described by the
quintuplet of objects
(MB , MdR , MG , IB, , IG, ),
(72)
where the three first entries are cohomological objects constructed via Tate twists from
the Betti, de Rham, and tale cohomology, respectively. Furthermore IB, describes a map
between the Betti and de Rham cohomology, while IG, is a map between Betti and tale
cohomology.3 The focus in the present paper is mostly on motives derived from the first
(co)homology groups H1 (A) and H1 (A) of Abelian varieties A, as well as the primitive
cohomology of Fermat hypersurfaces.
The second ingredient in Delignes conjecture is the concept of an L-function. This can
be described in a number of equivalent ways. Conceptually, the perhaps simplest approach
results when it can be derived via Artins zeta function as the HasseWeil L-function
induced by the underlying variety, i.e., by counting solutions of the variety over finite
fields.4 The complete L-function receives contributions from two fundamentally different
factors, (M, s) = L (M, s)L(M, s). The infinity term L (M, s) originates from those
fields over which the underlying variety has bad reduction, i.e., it is singular, while the
second term L(M, s) collects all the information obtained from the finite fields over which
the variety is smooth. The complete L-function is in general expected to satisfy a functional
equation, relating its values at s and 1 s. A motive is called critical if neither of the infinity
factors in the functional equation has a pole at s = 0.
The final ingredient is the concept of the period of a motive, a generalization of ordinary
periods of varieties. Viewing the motive M as a generalized cohomology theory, Deligne
formulates the notion of a period c+ (M) C /Q by taking the determinant of the
compatibility homomorphism
IB, : MB MdR
(73)
between the Betti and the deRham realizations of the motive M. Delignes basic conjecture
then relates the period and the L-function via L(M, 0)/c+ (M) Q. Contact with the
HasseWeil L-function is made by noting that for motives of the type M = H(X)(m) with
Tate twists one has L(M, 0) = L(X, m). When applied to motives of Abelian varieties
c+ (M) leads to the determinant of the standard period matrix [15].
Important for us is a generalization of this conjecture which involves motives with
coefficients. Such motives can best be described via algebraic Hecke characters, which
are of particular interest for us because they come up in the L-function of projective
Fermat varieties. Algebraic Hecke characters were first introduced by Weil and called
3 Detailed reviews of motives can be found in [28].
4 Ref. [1] contains a brief description of this construction.
501
Hecke characters of type A0 , which is what they are called in the older literature. The
beauty of algebraic Hecke characters is that one is immediately led to a clear distinction
between the defining field K and the field in which a character lives, i.e., the field F of
values. In the context of motives constructed from these characters the field F becomes the
field of complex multiplication. Delignes conjecture emerges in the following way.
Deligne period conjecture:
L(M, 0)
F.
c+ (M)
(74)
This shows why the Deligne conjecture is of interest to us. The period and the L-function
determine the same field, which is the CM field of the motive. Delignes conjecture has
been proven by Blasius for Fermat hypersurfaces [16].
7. Summary and generalizations

We have shown that the concepts used to describe attractor varieties in the context
of elliptic compactifications can be generalized to CalabiYau varieties with finite
fundamental groups. We have mentioned above that the Abelian property is neither carried
by the variety itself nor the generalized intermediate Jacobian

J n (X) = H2n1 (Xan , C)/H2n1 Xan , Z(n) + Fn H2n1 (Xan , C),
(75)
but by the Jacobians of the curves that are the building blocks of the middle-dimensional
cohomology HdimC X (X). These Jacobians themselves do not admit complex multiplication, unlike the situation in the elliptic case, but instead split into different factors which
admit different types of complex multiplication, in general. Furthermore the ring class field
can be generalized to be the field of moduli, and we can consider also points on the Abelian
variety that are of finite order, i.e., torsion points, and the field extensions they generate.
This allows us to answer a question posed in [5] which asked whether the absolute

Galois group Gal(K/K)
could play a role in the context of N = 2 compactifications of
type IIB strings. This is indeed the case. Suppose we have given an Abelian variety A
defined over a field K with complex multiplication by a field F . Then there is an action

of K on the torsion points of A.
of the absolute Galois group Gal(K/K)
of the closure K
This action is described by a Hecke character which is associated to the fields (K, F ) [26].
We have mentioned already that in general the (Griffiths) intermediate Jacobian is only
a torus, not an Abelian variety.
Even in those cases it is however possible to envision the existence of motives
via Abelian varieties associated to a variety X. Consider the Chow groups CHp (X)
of codimension p cycles modulo rational equivalence and denote by CHp (X)hom the
subgroup of cycles homologically equivalent to zero. Then there is a homomorphism, the
AbelJacobi homomorphism, which embeds CHp (X)hom into the intermediate Jacobian
: CHp (X)hom J p (X).
(76)
502
The image of on the subgroup Ap (X) defined by cycles algebraically equivalent to zero
does in fact define an Abelian variety, even if J p (X) is not an Abelian variety but only a
torus [42]. Hence we can ask whether attractor varieties are distinguished by AbelJacobi
images which admit complex multiplication.
Even more general, we can formulate this question in the framework of motives because
of Delignes conjecture. Thinking of motives as universal cohomology theories, it is
conceivable that attractor varieties lead to motives in the Abelian category with (potential)
complex multiplication. The standard cycle class map construction of CHp (X)hom is
replaced by the first term of a (conjectured) filtration in the resulting K-theory.
Combining the two threads of our analysis illustrates that the two separate discussions
in [5] characterizing toroidal attractor varieties via complex multiplication on the one
hand, and CalabiYau hypersurfaces via periods on the other, are two aspects of our
way of looking at this problem. This is the case precisely because of Delignes period
conjecture which relates the field of the periods to the field of complex multiplication via
the L-function of the variety (or motive). Thus a very pretty unified picture emerges.
Acknowledgements
We are grateful to G. Moore and P. Deligne for discussions. M.L. and R.S. thank the
Kavli Institute for Theoretical Physics, Santa Barbara, for hospitality and support through
KITP Scholarships during the course of part of this work. R.S. would also like to thank
the organizers of the Duality Workshop at the ITP in 2001. This research was supported
in part by the National Science Foundation under grant No. PHY99-07949. The work of
M.L. and R.S. has been supported in part by NATO under grant CRG 9710045. The work
of V.P. was supported by NSF under grant No. PHY98-02484.
References
[1] R. Schimmrigk, On the BeilinsonBloch conjecture and Rational Conformal Field Theory, Lecture at Bonn
University, 1995.
[2] R. Schimmrigk, Arithmetic geometry of CalabiYau manifolds and rational conformal field theory, J. Geom.
Phys. 44 (2003) 555, hep-th/0111226;
R. Schimmrigk, Aspects of conformal field theory from CalabiYau arithmetic, in: J. Lewis, N. Yui (Eds.),
The Proceedings of the Workshop on Arithmetic, Geometry, and Physics around CalabiYau Varieties and
Mirror Symmetry, 2003, in press, math.AG/0209168.
[3] R. Schimmrigk, S. Underwood, The ShimuraTaniyama conjecture and conformal field theory, J. Geom.
Phys., in press, hep-th/0211284.
[4] S. Gukov, C. Vafa, Rational conformal field theory and complex multiplication, hep-th/0203213.
[5] G. Moore, Attractors and arithmetic, hep-th/9807056;
G. Moore, Arithmetic and attractors, hep-th/9807087.
[6] S.D. Miller, G. Moore, LandauSiegel zeroes and black hole entropy, Asian J. Math. 4 (2000) 183212,
hep-th/9903267.
[7] P. Candelas, X. de la Ossa, F. Rodriguez-Villegas, CalabiYau manifolds over finite fields, hep-th/0012233.
[8] S. Ferrara, R. Kallosh, A. Strominger, N = 2 extremal black holes, Phys. Rev. D 52 (1995) 54125416,
hep-th/9508072.
503
[9] A. Strominger, Macroscopic entropy of N = 2 extremal black holes, Phys. Lett. B 383 (1996) 3943, hepth/9602111.
[10] S. Ferrara, R. Kallosh, Supersymmetry and attractors, Phys. Rev. D 54 (1996) 1514, hep-th/9602136;
S. Ferrara, R. Kallosh, Universality of supersymmetric attractors, Phys. Rev. D 54 (1996) 1525, hepth/9603090.
[11] S. Ferrara, Gibbons, R. Kallosh, Black holes and critical points in moduli space, Nucl. Phys. B 500 (1997)
75, hep-th/9702103.
[12] W.A. Sabra, General static N = 2 black holes, Mod. Phys. Lett. A 12 (1997) 2585, hep-th/9703101;
W.A. Sabra, Black holes in N = 2 supergravity theories and harmonic functions, Nucl. Phys. B 510 (1998)
247, hep-th/9704147.
[13] K. Behrndt, D. Lst, W.A. Sabra, Stationary solutions of N = 2 supergravity, Nucl. Phys. B 510 (1998) 264,
hep-th/9705169.
[14] P. Deligne, Valeurs de Fonctions L et Periodes dIntegrales, in: Proc. Sympos. Pure Math., Vol. 33, Amer.
Math. Soc., Providence, 1979, pp. 313346.
[15] D. Gross, L-function at the central critical point, in: U. Jannsen, S. Kleiman, J.-P. Serre (Eds.), Motives,
Amer. Math. Soc., Providence, 1994.
[16] D. Blasius, On the critical values of Hecke L-series, Ann. Math. 124 (1986) 2363.
[17] P. Fre, Supersymmetry and first order equations for extremal states: monopoles, hyperinstantons, black holes
and p-branes, Nucl. Phys. B (Proc. Suppl.) 57 (1997) 52, hep-th/9701054.
[18] P. Candelas, X. de la Ossa, Moduli space of CalabiYau manifolds, Nucl. Phys. B 355 (1991) 455.
[19] M. Billo, A. Ceresole, R. DAuria, S. Ferrara, P. Fre, T. Regge, P. Soriani, A. van Proeyen, A search for
non-perturbative dualities of local N = 2 YangMills theories from CalabiYau threefolds, Class. Quantum
Grav. 13 (1996) 831, hep-th/9506075.
[20] A. Ceresole, R. DAuria, S. Ferrara, The symplectic structure of N = 2 supergravity and its central
extensions, Nucl. Phys. B (Proc. Suppl.) 46 (1996) 67, hep-th/9509160.
[21] F. Larsen, F. Wilczek, Internal structure of black holes, Phys. Lett. B 375 (1996) 37, hep-th/9511064.
[22] L. Andrianopoli, R. DAuria, S. Ferrara, U-duality and central charges in various dimensions revisited, Int.
J. Mod. Phys. A 13 (1998) 431490, hep-th/9612105;
L. Andrianopoli, R. DAuria, S. Ferrara, Flat symplectic bundles of N-extended supergravities, central
charges and black hole entropy, in: Proceedings of the 5th Winter School on Mathematical Physics held
at the Asia Pacific Center for Theoretical Physics, February 1997, hep-th/9707203.
[23] J.H. Silverman, A survey of the arithmetic theory of elliptic curves, in: Modular Forms and Fermats Last
Theorem, Springer-Verlag, Berlin, 1997.
[24] J.H. Silverman, Advanced Topics in the Arithmetic of Elliptic Curves, Springer-Verlag, Berlin, 1994.
[25] G. Shimura, Y. Taniyama, Complex Multiplication of Abelian Varieties and its Application to Number
Theory, 1961.
[26] G. Shimura, Introduction to Arithmetic Theory of Automorphic Functions, Princeton Univ. Press, Princeton,
NJ, 1971.
[27] G. Shimura, Abelian Varieties with Complex Multiplication and Modular Functions, Princeton Univ. Press,
Princeton, NJ, 1998.
[28] U. Jannsen, S. Kleiman, J.P. Serre, Motives, Amer. Math. Soc., Providence, 1994.
[29] T. Shioda, T. Katsura, On Fermat varieties, Tohoku Math. J. 31 (1979) 97115.
[30] P. Candelas, M. Lynker, R. Schimmrigk, CalabiYau manifolds in weighted P4 , Nucl. Phys. B 341 (1990)
383, KEK archive 199001117.
[31] R. Schimmrigk, A new construction of a three-generation CalabiYau manifold, Phys. Lett. B 193 (1987)
175, KEK archive 198706386;
R. Schimmrigk, Heterotic RG flow fixed points with non-diagonal affine invariants, Phys. Lett. B 229 (1989)
227, KEK archive 198905565.
[32] D.K. Faddeev, On the divisor groups of some algebraic curves, Dokl. Tom 136 (1961) 296298, Sov. Math. 2
(1961) 6769 (in English);
D.K. Faddeev, Invariants of divisor classes for the curves y G = x k (1 x) in an G-adic cyclotomic field,
Trudy Mat. Inst. Steklov. 64 (1961) 284293 (in Russian).
[33] A. Weil, Sur les priodes des intgrales Abeliennes, Comm. Pure Appl. Math. 29 (1976) 813.
504
[34] B.H. Gross, On the periods of Abelian integrals and a formula of Chowla and Selberg, Invent. Math. 45
(1978) 193.
[35] D.E. Rohrlich, The periods of the Fermat curve, Invent. Math. 45 (1978) 208, Appendix to Gross 1978.
[36] W. Wei, Moduli fields of CM motives applied to Hilberts 12TH Problem, preprint.
[37] J. Neukirch, Algebraic Number Theory, Springer-Verlag, Berlin, 1999.
[38] D. Ramakrishnan, R. Valenza, Fourier Analysis on Number Fields, Springer-Verlag, Berlin, 1998.
[39] G. Shimura, On some arithmetic properties of modular forms of one and several variables, Ann. Math. 102
(1975) 491515.
[40] G. Shimura, The special values of the zeta functions associated with cusp forms, Commun. Pure Appl.
Math. 29 (1976) 783804.
[41] G. Shimura, On the periods of modular forms, Math. Ann. 229 (1977) 211221.
[42] H. Saito, Abelian varieties attached to cycles of intermediate Jacobians, Nagoya Math. J. 75 (1979) 95119.

CUMULATIVE AUTHOR INDEX B661B667
Abel, S.A.
Acquaviva, V.
Adams, D.H.
Akbar, M.M.
Akemann, G.
Alexandrov, S.Yu.
Alishahiha, M.
ALPHA Collaboration
lvarez, E.
Antoniadis, I.
Armoni, A.
Armoni, A.
Arutyunov, G.
Arutyunov, G.
Astefanesei, D.
Axenides, M.
B663 (2003) 197

B667 (2003) 119
B662 (2003) 220
B663 (2003) 215
B664 (2003) 457
B667 (2003) 90
B661 (2003) 174
B663 (2003) 3
B663 (2003) 365
B662 (2003) 40
B664 (2003) 233
B667 (2003) 170
B663 (2003) 163
B665 (2003) 273
B665 (2003) 594
B662 (2003) 170
Bais, F.A.
Barbieri, R.
Bardakci, K.
Bartolo, N.
Beccaria, M.
Becchi, C.
Becker, K.
Becker, M.
Beenakker, W.
Beisert, N.
Belitsky, A.V.
Bellucci, S.
Bellucci, S.
Bena, I.
Benakli, K.
Benson, D.
Berezin, V.
Bigi, I.I.
Bilal, A.
Blaek, T.
Blumenhagen, R.
Boer, D.
B666 (2003) 243

B663 (2003) 141
B661 (2003) 235
B667 (2003) 119
B663 (2003) 394
B664 (2003) 371
B666 (2003) 144
B666 (2003) 144
B667 (2003) 359
B664 (2003) 131
B667 (2003) 3
B663 (2003) 605
B665 (2003) 402
B664 (2003) 45
B662 (2003) 40
B665 (2003) 367
B661 (2003) 409
B665 (2003) 367
B663 (2003) 343
B662 (2003) 359
B663 (2003) 319
B667 (2003) 201
0550-3213/2003 Published by Elsevier B.V.

doi:10.1016/S0550-3213(03)00665-5
Bonciani, R.
Bourbonnais, C.
Bouttier, J.
Brandenburg, A.
Brax, P.
Brignole, A.
Brower, R.C.
Brower, R.C.
Bruckmann, F.
Buchbinder, I.L.
Buchmller, W.
B661 (2003) 289

B663 (2003) 568
B663 (2003) 535
B667 (2003) 394
B667 (2003) 149
B666 (2003) 105
B661 (2003) 344
B662 (2003) 393
B666 (2003) 197
B665 (2003) 402
B665 (2003) 445
Cacciari, M.
Campanario, F.
Cao, J.
Casas, J.A.
Chandrasekharan, S.
Chapovsky, A.P.
Chetyrkin, K.G.
Chitov, G.Y.
Choi, K.-S.
Conde, J.
Cvetic, M.
B664 (2003) 299

B663 (2003) 280
B663 (2003) 487
B666 (2003) 105
B662 (2003) 220
B667 (2003) 359
B666 (2003) 289
B663 (2003) 568
B662 (2003) 476
B663 (2003) 365
B662 (2003) 89
Daleo, A.
Darriulat, P.
Dasgupta, K.
de Azcrraga, J.A.
de Boer, J.
de Haro, S.
DElia, M.
Demasure, Y.
Denner, A.
Derkachov, S..
Deser, S.
Di Bari, P.
Di Francesco, P.
Dinh, P.N.
Dobashi, S.
Dolan, F.A.
B662 (2003) 334

B661 (2003) 3
B666 (2003) 144
B662 (2003) 185
B665 (2003) 545
B664 (2003) 45
B661 (2003) 139
B661 (2003) 153
B662 (2003) 299
B661 (2003) 533
B662 (2003) 379
B665 (2003) 445
B663 (2003) 535
B661 (2003) 3
B665 (2003) 94
B665 (2003) 273
506
Dorey, P.
Dorey, P.
Dorogovtsev, S.N.
Dotsenko, V.S.
Dubovsky, S.L.
Dung, N.T.
B661 (2003) 425

B661 (2003) 464
B666 (2003) 396
B664 (2003) 477
B664 (2003) 407
B661 (2003) 3
Emmanuel-Costa, D.
Engquist, J.
Espinosa, J.R.
Etesi, G.
Evlampiev, K.
B661 (2003) 62
B664 (2003) 439
B666 (2003) 105
B662 (2003) 511
B662 (2003) 120
Faisst, M.
Falkowski, A.
Farhi, E.
Feng, B.
Feverati, G.
Floratos, E.
Forger, M.
Freidel, L.
Freitas, A.
Fursaev, D.V.
Fyodorov, Y.V.
B665 (2003) 649

B667 (2003) 149
B665 (2003) 623
B661 (2003) 113
B663 (2003) 409
B662 (2003) 170
B667 (2003) 435
B662 (2003) 279
B666 (2003) 305
B664 (2003) 403
B664 (2003) 457
Ganjali, M.A.
Garca Canal, C.A.
Gardi, E.
Geyer, B.
Ghodsi, A.
Giudice, G.F.
Giusto, S.
Gomis, J.
Gonzlez, J.
Gorsky, A.S.
Gracey, J.A.
Gracey, J.A.
Graham, N.
Groot Nibbelink, S.
Groot Nibbelink, S.
Grozin, A.G.
Grozin, A.G.
Guitter, E.
Gustavsson, A.
B661 (2003) 174

B662 (2003) 334
B664 (2003) 299
B662 (2003) 531
B661 (2003) 174
B663 (2003) 377
B664 (2003) 371
B665 (2003) 49
B663 (2003) 605
B667 (2003) 3
B662 (2003) 247
B667 (2003) 242
B665 (2003) 623
B663 (2003) 60
B665 (2003) 236
B663 (2003) 280
B666 (2003) 289
B663 (2003) 535
B667 (2003) 111
Hall, L.J.
Harmark, T.
Herdeiro, C.A.R.
Hernndez, L.
Hieu, B.D.
Hiller, J.R.
Hollik, W.
Honecker, G.
Hou, B.Y.
B663 (2003) 141

B662 (2003) 3
B665 (2003) 189
B663 (2003) 365
B661 (2003) 3
B661 (2003) 99
B666 (2003) 305
B666 (2003) 175
B663 (2003) 467
Huber, P.
Huber, S.J.
Hwang, K.
B665 (2003) 487

B666 (2003) 269
B662 (2003) 476
Ichinose, I.
Imai, T.
Imbimbo, C.
Intriligator, K.
Isaev, A.P.
Ivanov, N.Ya.
Izquierdo, J.M.
B663 (2003) 520

B665 (2003) 520
B664 (2003) 371
B667 (2003) 183
B662 (2003) 461
B666 (2003) 88
B662 (2003) 185
Jack, I.
Jacobsen, J.L.
Jaffe, R.L.
Janik, R.A.
Jones, D.R.T.
B662 (2003) 63
B664 (2003) 477
B665 (2003) 623
B661 (2003) 153
B662 (2003) 63
Kamimura, K.
Kaminsky, K.
Kanaki, A.
Kawai, H.
Kazakov, V.A.
Kehagias, A.
Khater, W.
Khemani, V.
Kim, J.E.
Kimura, Y.
King, S.F.
Kitazawa, Y.
Klebanov, I.R.
Kleinert, H.
Knechtli, F.
Korchemsky, G.P.
Korchemsky, G.P.
Kostov, I.K.
Kotikov, A.V.
Kraus, E.
Kristjansen, C.
Krykhtin, V.A.
Khn, J.H.
Kuroki, T.
Kurylov, A.
Kutasov, D.
B662 (2003) 491

B663 (2003) 33
B667 (2003) 359
B664 (2003) 185
B667 (2003) 90
B662 (2003) 170
B661 (2003) 209
B665 (2003) 623
B662 (2003) 476
B664 (2003) 512
B662 (2003) 359
B665 (2003) 520
B664 (2003) 3
B666 (2003) 361
B663 (2003) 3
B661 (2003) 533
B667 (2003) 3
B667 (2003) 90
B661 (2003) 19
B661 (2003) 83
B664 (2003) 131
B665 (2003) 402
B665 (2003) 649
B664 (2003) 185
B667 (2003) 321
B666 (2003) 56
Lalak, Z.
Larosa, M.
Laugier, A.
Lee, H.K.
Lee, K.
Lehners, J.-L.
Leonhardt, T.
Li, Y.Q.
Lin, H.-Q.
Lin, H.Q.
B667 (2003) 149

B667 (2003) 261
B662 (2003) 40
B665 (2003) 153
B665 (2003) 179
B661 (2003) 273
B667 (2003) 413
B666 (2003) 337
B663 (2003) 487
B666 (2003) 337
Lindner, M.
Lindstrm, U.
Lipatov, L.N.
Livine, E.R.
Louapre, D.
Lowe, D.A.
L, H.
Ludwig, A.W.W.
Lst, D.
Lynker, M.
B665 (2003) 487

B662 (2003) 147
B661 (2003) 19
B663 (2003) 231
B662 (2003) 279
B667 (2003) 55
B662 (2003) 89
B661 (2003) 577
B663 (2003) 319
B667 (2003) 484
Maillard, T.
Majumdar, P.
Manashov, A.N.
Mannel, T.
Mannel, Th.
Manvelyan, R.
Marandella, G.
Marchi, M.
Martucci, L.
Masina, I.
Mastrolia, P.
Mastrolia, P.
Matarrese, S.
Mathur, S.D.
Meggiolaro, E.
Melles, M.
Melnikov, K.
Mendes, J.F.F.
Metzger, S.
Moore, J.E.
Morita, T.
Moriyama, S.
Morozov, A.
Mulders, P.J.
Mlsch, D.
B662 (2003) 40
B664 (2003) 213
B661 (2003) 533
B663 (2003) 280
B665 (2003) 367
B667 (2003) 413
B663 (2003) 141
B665 (2003) 425
B666 (2003) 230
B661 (2003) 365
B661 (2003) 289
B664 (2003) 341
B667 (2003) 119
B661 (2003) 344
B665 (2003) 425
B662 (2003) 299
B662 (2003) 409
B666 (2003) 396
B663 (2003) 343
B661 (2003) 514
B664 (2003) 185
B665 (2003) 49
B666 (2003) 311
B667 (2003) 201
B662 (2003) 531
Nadolsky, P.M.
Nadolsky, P.M.
Nan, C.M.
Nason, P.
Nastase, H.
Navarro, I.
Niarchos, V.
Niemi, A.J.
Nilles, H.P.
Ngrdi, D.
Nogueira, F.S.
Nomura, Y.
Nyawelo, T.S.
B666 (2003) 3
B666 (2003) 31
B663 (2003) 591
B667 (2003) 394
B667 (2003) 55
B666 (2003) 105
B666 (2003) 56
B666 (2003) 311
B665 (2003) 236
B666 (2003) 197
B666 (2003) 361
B663 (2003) 141
B663 (2003) 60
Okawa, Y.
Okui, T.
Oleari, C.
B663 (2003) 33
B663 (2003) 141
B667 (2003) 394
507
Olechowski, M.
Oliver, S.J.
Onorato, P.
Ooguri, H.
Oriti, D.
Osborn, H.
Osland, P.
Owen, A.W.
B665 (2003) 236

B663 (2003) 141
B663 (2003) 605
B663 (2003) 33
B663 (2003) 231
B665 (2003) 273
B661 (2003) 209
B663 (2003) 197
Papadopoulos, C.G.
Papucci, M.
Park, J.
Parvizi, S.
Pearce, P.A.
Periwal, V.
Phuong, P.T.
Picn, M.
Pijlman, F.
Pinsky, S.S.
Pittau, R.
Plmacher, M.
Pocklington, A.
Pocklington, A.
Pons, J.M.
Pope, C.N.
Pozzorini, S.
Pradisi, G.
Prokushkin, S.
B667 (2003) 359

B663 (2003) 141
B665 (2003) 49
B661 (2003) 174
B663 (2003) 409
B667 (2003) 484
B661 (2003) 3
B662 (2003) 185
B667 (2003) 201
B661 (2003) 99
B667 (2003) 359
B665 (2003) 445
B661 (2003) 425
B661 (2003) 464
B665 (2003) 129
B662 (2003) 89
B662 (2003) 299
B667 (2003) 261
B666 (2003) 144
Radu, E.
Ramgoolam, S.
Ramsey-Musolf, M.J.
Ravindran, V.
Remiddi, E.
Remiddi, E.
Renard, F.M.
Riccioni, F.
Riotto, A.
Rocek, M.
Roiban, R.
Roiban, R.
Rhl, W.
Rupp, C.
B665 (2003) 594

B667 (2003) 55
B667 (2003) 321
B665 (2003) 325
B661 (2003) 289
B664 (2003) 341
B663 (2003) 394
B663 (2003) 60
B667 (2003) 119
B662 (2003) 147
B664 (2003) 45
B665 (2003) 211
B667 (2003) 413
B661 (2003) 83
Sakaguchi, M.
Saleur, H.
Salvay, M.J.
Samukhin, A.N.
Santachiara, R.
Sasaki, R.
Sassot, R.
Savoy, C.A.
Schimmrigk, R.
Schwetz, T.
B662 (2003) 491

B663 (2003) 443
B663 (2003) 591
B666 (2003) 396
B664 (2003) 477
B663 (2003) 467
B662 (2003) 334
B661 (2003) 365
B667 (2003) 484
B665 (2003) 487
508
Segal, A.Y.
Seidensticker, T.
Seki, S.
Serbo, V.G.
Sezgin, E.
Shafi, Q.
Sharpe, E.
Shi, K.-J.
Shifman, M.
Shifman, M.
Shik, H.Y.
Shimada, H.
Sibiryakov, S.M.
Sibold, K.
Siegel, W.
Silva, P.J.
Sin, S.-J.
Singh, H.
Skenderis, K.
Smith, J.
Sokatchev, E.
Sokatchev, E.
Solodukhin, S.N.
Staudacher, M.
Stefanski Jr., B.
Stelle, K.S.
Stelle, K.S.
Striet, J.
Strumia, A.
Su, S.
Sudb, A.
Sugawara, Y.
Sundell, P.
B664 (2003) 59
B665 (2003) 649
B661 (2003) 257
B662 (2003) 409
B664 (2003) 439
B665 (2003) 469
B664 (2003) 21
B663 (2003) 487
B664 (2003) 233
B667 (2003) 170
B666 (2003) 337
B665 (2003) 94
B664 (2003) 407
B661 (2003) 83
B665 (2003) 179
B666 (2003) 230
B667 (2003) 310
B661 (2003) 394
B665 (2003) 3
B665 (2003) 325
B663 (2003) 163
B665 (2003) 273
B665 (2003) 545
B664 (2003) 131
B666 (2003) 71
B661 (2003) 273
B662 (2003) 89
B666 (2003) 243
B663 (2003) 377
B667 (2003) 321
B666 (2003) 361
B661 (2003) 191
B664 (2003) 439
Takayama, Y.
Takayanagi, T.
Takeda, K.
Talavera, P.
Tan, C.-I.
Tan, C.-I.
Tanaka, T.
Tatar, R.
Tateo, R.
Tateo, R.
Tavartkiladze, Z.
Taylor, M.
Taylor, T.R.
Thao, N.T.
Thieu, D.Q.
Thorn, C.B.
Thuan, V.V.
Tobe, K.
Tomino, D.
B665 (2003) 520

B662 (2003) 3
B663 (2003) 520
B665 (2003) 129
B661 (2003) 344
B662 (2003) 393
B662 (2003) 413
B665 (2003) 211
B661 (2003) 425
B661 (2003) 464
B665 (2003) 469
B665 (2003) 3
B663 (2003) 319
B661 (2003) 3
B661 (2003) 3
B661 (2003) 235
B661 (2003) 3
B663 (2003) 123
B665 (2003) 520
Trittmann, U.
Tseytlin, A.A.
Tsutsui, I.
B661 (2003) 99
B664 (2003) 247
B662 (2003) 447
Uchino, T.
Uraltsev, N.
B662 (2003) 447

B665 (2003) 367
van Baal, P.
van Holten, J.W.
van Neerven, W.L.
van Nieuwenhuizen, P.
Varela, O.
Veneziano, G.
Veretin, O.
Verzegnassi, C.
B666 (2003) 197

B663 (2003) 60
B665 (2003) 325
B662 (2003) 147
B662 (2003) 185
B667 (2003) 170
B665 (2003) 649
B663 (2003) 394
Walcher, J.
Waldron, A.
Walter, M.G.A.
Walter, W.
Wang, W.
Wang, Y.
Wecht, B.
Wehefritz-Kaufmann, B.
Weigel, H.
Weiglein, G.
Wells, J.D.
Wiese, K.J.
Wiesenfeldt, S.
Winter, W.
Winterhalder, A.
Witten, E.
Wolff, U.
Wu, J.-B.
Wu, J.-B.
Wu, X.
B665 (2003) 211

B662 (2003) 379
B665 (2003) 236
B666 (2003) 305
B667 (2003) 349
B663 (2003) 487
B667 (2003) 183
B663 (2003) 443
B665 (2003) 623
B666 (2003) 305
B663 (2003) 123
B661 (2003) 577
B661 (2003) 62
B665 (2003) 487
B667 (2003) 435
B664 (2003) 3
B663 (2003) 3
B663 (2003) 79
B663 (2003) 95
B665 (2003) 153
Yang, W.-L.
Ylmaz, N.T.
Yoneya, T.
Yuan, C.-P.
Yuan, C.-P.
Yue, C.
Yung, A.
B663 (2003) 467

B664 (2003) 357
B665 (2003) 94
B666 (2003) 3
B666 (2003) 31
B667 (2003) 349
B662 (2003) 120
ZeuthenRome (ZeRo) Collaboration

Zheng, Z.-J.
Zheng, Z.-J.
Zhu, C.-J.
Zhu, C.-J.
Znojil, M.
Zong, H.
B664 (2003) 276

B663 (2003) 79
B663 (2003) 95
B663 (2003) 79
B663 (2003) 95
B662 (2003) 554
B667 (2003) 349

Nucl - Phys.B v.667

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.667

Uploaded by

Copyright:

Available Formats

Nuclear Physics B 667 (2003) 354

Gauge/string duality for QCD conformal operators

Received 28 April 2003; accepted 18 June 2003

A.V. Belitsky et al. / Nuclear Physics B 667 (2003) 354

[C]0| tr P exp i dx A (x) |0,

A.V. Belitsky et al. / Nuclear Physics B 667 (2003) 354

A.V. Belitsky et al. / Nuclear Physics B 667 (2003) 354

of freedom at strong couplingthe string junctionsand elucidate the relation between

M(v  ) (0) (0) M(v) = ( )(v + v  ) .

A.V. Belitsky et al. / Nuclear Physics B 667 (2003) 354

In the heavy-quark limit, m , it depends only on the product of velocities v  v =

 ) v  [; 0]v [0; ] M(v)

A.V. Belitsky et al. / Nuclear Physics B 667 (2003) 354

from excited states |MX (v). The Wilson

A.V. Belitsky et al. / Nuclear Physics B 667 (2003) 354

where D = iA is a covariant derivative. Their dependence on the ultraviolet cut-off

A.V. Belitsky et al. / Nuclear Physics B 667 (2003) 354

OJ = H (p) W (iJ ) H (p) P exp i dx A (x) .

We would like to stress that Eq. (2.15) holds only for J  1.

A.V. Belitsky et al. / Nuclear Physics B 667 (2003) 354

A.V. Belitsky et al. / Nuclear Physics B 667 (2003) 354

Feynman diagram is straightforward and details can be found in Appendix A. It leads to

A.V. Belitsky et al. / Nuclear Physics B 667 (2003) 354

3. Conformal operators and integrable spin chains

A.V. Belitsky et al. / Nuclear Physics B 667 (2003) 354

A.V. Belitsky et al. / Nuclear Physics B 667 (2003) 354

The Lorentz operators (3.2) can be re-expressed in terms of conformal operators O J as

A.V. Belitsky et al. / Nuclear Physics B 667 (2003) 354

A.V. Belitsky et al. / Nuclear Physics B 667 (2003) 354

A.V. Belitsky et al. / Nuclear Physics B 667 (2003) 354

A.V. Belitsky et al. / Nuclear Physics B 667 (2003) 354

where {j } (j12 , . . . , j12N1 ) and PJ (x1 , . . . , xN ) is a homogeneous polynomial of

(L1 + + Lk ) PJ = J12k (J12k 1)PJ

with L2 = (L+ L + L L+ )/2 + L20 . At N = 2 the solution to (3.20) is given by Jacobi

A.V. Belitsky et al. / Nuclear Physics B 667 (2003) 354

where [d N x] = dx1 dxN (1 k xk ) and integration goes over 0 xk 1. Eq. (3.23)

under the additional (highest weight) condition

A.V. Belitsky et al. / Nuclear Physics B 667 (2003) 354

where k = /xk . The general expression for Hk,k+1 looks like

and, as a consequence, the Schrdinger equation for HN turns out to be completely

A.V. Belitsky et al. / Nuclear Physics B 667 (2003) 354

A.V. Belitsky et al. / Nuclear Physics B 667 (2003) 354

Remarkably enough, this Hamiltonian inherits integrability properties of the quantum

A.V. Belitsky et al. / Nuclear Physics B 667 (2003) 354

with n = 2, . . . , N . One observes that (qNcl ) =

and, therefore, the classical

cl defines a classical limit of the Heisenberg SL(2, R) spin magnet

A.V. Belitsky et al. / Nuclear Physics B 667 (2003) 354

4. Cusp anomaly at weak coupling

where 0|T Aa (x)Ab (y)|0 = g 2 ab D (x y) is a gluon propagator and t a t a = Nc is

A.V. Belitsky et al. / Nuclear Physics B 667 (2003) 354

A.V. Belitsky et al. / Nuclear Physics B 667 (2003) 354

A.V. Belitsky et al. / Nuclear Physics B 667 (2003) 354

one obtains the metric on this manifold as

Here the Hamiltonian H = L2 is the LaplaceBeltrami operator on the S 3 . It defines

HYm1 m2 (g) = j (j + 1)Ym1 m2 (g),

Ym1 m2 (g) = (2j + 1)1/2 dm1 m2 (g),

Here, the SU(2) character is defined as [62]

A.V. Belitsky et al. / Nuclear Physics B 667 (2003) 354

(4.11), one calculates the matrix element entering (4.15) as

(2j + 1)eij (j +1)

where q = ei and = . One can rewrite the same expression as

and the classical action (4.14) is given in the gauge e( ) = const by

M(v ) (0) (0) M(v) = ( )(v + v ) .

In the heavy-quark limit, m , it depends only on the product of velocities v v =

) v [; 0]v [0; ] M(v)

We would like to stress that Eq. (2.15) holds only for J 1.

for 1 k 1. As before, at k = 2 and k = N the coefficient in front of