Some Implications of Perturbative Approach To Ads/Cft Correspondence

Nuclear Physics B 794 (2008) 112
www.elsevier.com/locate/nuclphysb
Some implications of perturbative approach

to AdS/CFT correspondence
Hikaru Kawai a,b , Takao Suyama a,
a Department of Physics, Kyoto University, Kitashirakawa, Kyoto 606-8502, Japan
b Theoretical Physics Laboratory, The Institute of Physics and Chemical Research (RIKEN),
Wako, Saitama 351-0198, Japan

Received 21 September 2007; accepted 23 October 2007
Available online 28 October 2007
Abstract
We show some implications of the approach to AdS/CFT correspondence based on type IIB string in
the flat spacetime with D3-branes proposed in our previous paper. We discuss a correspondence for high
energy scattering amplitudes of N = 4 super-YangMills proposed recently. We also discuss AdS/CFT
correspondence at finite temperature. Our approach provides clear understanding of these issues.
2007 Elsevier B.V. All rights reserved.
1. Introduction
In our previous paper [1], we proposed how to understand the reason why AdS/CFT correspondence holds. Our discussion there was based on the perturbative type IIB string in the flat
spacetime with D3-branes introduced as boundaries of the worldsheets. We mainly discussed
the relation [2,3] between Wilson loops in N = 4 super-YangMills (SYM) in four dimensions
and minimal surfaces in the AdS5 S 5 spacetime, and showed that the correspondence is a
consequence of an approximate symmetry which exists in the worldsheet theory, if the large
N limit is taken with the t Hooft coupling kept finite but large, and if the worldsheets relevant for evaluating the Wilson loops and the minimal surfaces are restricted within a region near
D3-branes.
* Corresponding author.
E-mail addresses: hkawai@gauge.scphys.kyoto-u.ac.jp (H. Kawai), suyama@gauge.scphys.kyoto-u.ac.jp

(T. Suyama).
0550-3213/$ see front matter 2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.nuclphysb.2007.10.016
H. Kawai, T. Suyama / Nuclear Physics B 794 (2008) 112
In this paper, we show that our approach is useful for understanding the other issues discussed
in the literature. The first example, discussed in Section 2, is on the calculation of high energy
scattering amplitudes of the SYM, which was proposed in [4] and recently confirmed on the
strong coupling side at the level of four-gluon scattering in [5]. It will be shown that this proposal
is also justified by the scale invariance in a similar way to the case of the Wilson loop. The
second example is on AdS/CFT correspondence at finite temperature, discussed in Section 3.
The discussions on Wilson loops can be carried out also in this case. From the point of view of
the perturbative string with D3-branes, the issue on the entropy can be understood rather clearly.
2. High energy scattering amplitudes
Let us consider a scattering process of n massless open string states living on parallel N
D3-branes. At a low energy scale, the dynamics of open strings is governed by N = 4 SYM in
four dimensions. The scattering amplitude can be calculated from worldsheets with a number of
boundaries on which n vertex operators for the massless states are attached. We only consider
the case in which all the vertex operators are attached to a single boundary of the worldsheet.
Let us take the large N limit with the t Hooft coupling = gs N fixed. Due to taking this limit,
closed string handles of the worldsheets are suppressed, and therefore, the relevant topology
of the worldsheets is that of the disk with boundaries. These worldsheets correspond to planar
ribbon graphs of the SYM. Let (t ak )ij be the color-dependent factor of the kth vertex operator.
Then the amplitude considered here is proportional to Tr(t a1 t a2 t an ) which is the only factor
depending on the gauge group.
Suppose that one of the D3-branes is separated from the other N 1 D3-branes which are on
top of each other, and all of the external open string states are on the separated D3-brane. The
restriction of the external states only results in choosing a trivial prefactor, instead of a generic
Tr(t a1 t a2 t an ), and therefore, it is easy to deduce the generic amplitude from this special one.
We can also restrict the sum over worldsheets so that the boundaries, on which there is no vertex
operator, are on the coincident D3-branes. This restriction results in assigning N 1, not N , to
each color index loop, and therefore, this leads to a modification of the amplitude by an amount
of order N1 , which is negligible in the large N limit. Due to the separation of the D3-brane, the
strips of the string corresponding to the outermost loop are stretched between the D3-branes with
a finite width, and therefore, it is a massive state that is propagating along this loop. The situation
is depicted in Fig. 1. Fig. 2 shows the corresponding Feynman diagram.
The introduction of mass in this manner is expected to make the amplitude well defined in
the IR region. In terms of the N = 4 SYM, what we have done is the following. We first give a
non-zero background to one of the scalar fields as
1 ... n n + 1...N
..
n
,
n+1
..
..
.
.
N
0
1
..
.
(2.1)
where the situation depicted in Fig. 1 corresponds to n = 1, and one can consider a situation for
generic n. Here we assume n N , and consider only planar diagrams. We further assume that
Fig. 1. Worldsheet configuration for the regularized amplitude.
Fig. 2. Feynman diagram corresponding to the worldsheet in Fig. 1.
the colors of external particles are taken from the n n matrices. Therefore, in the double line
notation, the index of the outermost index loop runs from 1 to n. Furthermore we can constrain
the other index loops to run from n + 1 to N . This constraint does not affect the amplitude if
n N . In this way, we give non-zero mass to all the propagators that belong to the outermost
loop as in Fig. 2.
In order to illustrate the mechanism that mass of the outermost propagators regularizes the
IR divergence, we consider the following integral which corresponds to the Feynman diagram
depicted in Fig. 3,

1
1
1
1
1
1
I = d 4 k1 2
d 4 k2 2
. (2.2)
2
2
2
k1 (p1 k1 ) (p2 + k1 )
k2 (p1 k1 k2 ) (p2 + k1 + k2 )2
Here we assume k 2 = 0 for simplicity. If p12 = p22 = 0, the integral of k2 gives a term of order
log(k12 ), and the k1 integral is IR divergent. The regularization we are discussing is to give mass
to all propagators in the outermost loop. However, it is sufficient to consider only one propagator
in this simple case. Actually, if we give mass to the outer gluon, I is regularized to

1
1
1
Ireg = d 4 k1 2
k1 + 2 (p1 k1 )2 (p2 + k1 )2
d 4 k2
1
1
1
.
2
2
k2 (p1 k1 k2 ) (p2 + k1 + k2 )2
(2.3)
The k1 integral converges this time, although the k2 integral gives log(k12 ). The point is that the
mass of an outer propagator makes the whole diagram IR finite.1
Note that the mass introduced for the IR regularization must be much smaller than the string
scale, and therefore, the separation among D3-branes must be very small. Note also that the
momenta of the external states are also much smaller than the string scale so as not to produce
massive string states in the scattering process, although we call this process a high energy
scattering.
To relate the scattering amplitude to a classical worldsheet configuration, let us make the
following change of variables in the worldsheet theory,
X = XD ,
X
I
= XD
,
(2.4)
(2.5)
for = 1, . . . , 4 and I = 6, . . . , 10. This is nothing but the T-duality transformation of the worldsheet variables. The transformation of the fermionic variables is defined as
a
,
S a = SD
a
ab b
S =M S .
(2.6)
(2.7)
We have employed the GreenSchwarz formalism and taken the light-cone gauge. Since we consider worldsheets in the flat spacetime, this transformation obviously preserves the worldsheet
action. The boundary condition for X turns into the Dirichlet boundary condition since (2.4)
implies
X = XD ,
(2.8)
at the boundaries. As a result, the D3-branes in the original setup turn into D-instantons. It is
interesting to notice that the constant modes xD of XD are not fixed, since the boundary condition
is XD = 0. Therefore, xD should be integrated in the worldsheet path-integral which indicates

that the D-instantons are distributed uniformly along x -directions.
Fig. 3. A Feynman diagram which could have an IR divergence.

1 As far as we know, there is no proof for the justification of the regularization procedure described above. It is very
interesting to prove that our procedure works for generic amplitudes.
It is very interesting to compare this situation with the large N reduction of the SYM. In the reduced model, the momentum integrals for Feynman diagrams correspond to integrals of diagonal
elements of matrices [6]. If the reduced model is regarded as an effective theory of D-instantons,
then the diagonal elements of the matrices dictate the positions of the D-instantons. This chain of
correspondences also implies that the integration of the positions of the D-instantons is necessary
in the calculation of the high energy scattering amplitude.
In [5], classical solutions for the worldsheet are discussed in AdS5 S 5 background, and
then the T-dual transformation is performed to obtain explicit expressions of the solutions. This
corresponds, in our perturbative approach, to performing an anisotropic scale transformation of
[1] to go to the gravity region, and then perform the above transformation. In this paper, we
proceed through a different way; we perform the above T-dual transformation first, and then
perform a scale transformation defined below.
It should be noted that the transformation we would like to perform is not exactly the T-duality
transformation, since the x -directions are non-compact. The analysis below is thus valid only
when we restrict ourselves with the worldsheets which are topologically a disk with boundaries.
Inclusion of handles, corresponding to considering a finite N case, would require a more complicated analysis. In the following, however, we call this the T-duality transformation, which would
probably make no confusion.
From now on, we consider disk worldsheets in the presence of N D-instantons. As mentioned
above, this system is equivalent to the original system including D3-branes only in the large N
limit.
We consider the scale transformation discussed in [1] in this D-instanton system. The transformation properties of the T-dual variables can be easily derived from the original one. As will
be shown below, it is an isotropic scale transformation in this case. However, this transformation will turn out to be an approximate symmetry in a similar sense to [1]. As a result, we will
verify that the high energy scattering amplitude of N = 4 SYM can be calculated by a specific
configuration of the worldsheet in AdS5 , as is claimed in [5].
Recall that the scale transformation of the coordinate fields in the D3-brane setup is
X i ( ) = M ij X j ( ),
(2.9)
P ( ) = M P ( ),
S a ( ) = iM ab S b ( ),
i
ij
(2.10)
(2.11)
S a ( ) = iM ab S b ( ),
where i, j, a, b run from 1 to 8, and the 8 8 matrix

I44
0
M ij =
,
0
I44
ab

M ab = 1 2 3 4 ,
(2.12)
M ij
and
M ab
are defined as
(2.13)
(2.14)
where i are the SO(8) gamma matrices.

The T-duality transformation (2.4)(2.7) provides the scale transformation of the T-dual variables as
i
i
( ) = XD
( ),
XD
(2.15)
PDi ( ) = PDi ( ),
a
a
SD
( ) = i SD
( ),
(2.16)
(2.17)
Fig. 4. LSZ-like reduction. The cross represents the insertion of S or |B0 .

a
a
SD
( ) = iSD
( ).
(2.18)
Note that the transformation (2.4) exchanges the coordinates and the momenta which is clear
in (2.8). In this T-dual situation, the scale transformation is indeed the ordinary isotropic scale
transformation. One can easily obtain the scale transformation of the oscillators as
i
ni = n
,
ni
Sna
Sna
(2.19)
i
= n
,
a
= i Sn
,
a
= iSn .
(2.20)
(2.21)
(2.22)
Using these transformation rules, it is easy to show that the boundary state of a D-instanton

1
a a
|B = exp
(2.23)
i i iSn
Sn |B0 ,
n n n
n=1

I

i|a|
x = 0 p = 0
|B0 = |i|i
(2.24)
a
|0n ,
n=1
is invariant under the scale transformation. Note that

p = 0 = d 4 q x = q ,
(2.25)
describes the uniform distribution of the D-instanton.

To show the existence of the scale invariance in the D-instanton case, let us consider the free
energy in the D-instanton background defined as
F () =

Fn
n=0
n!
n ,
(2.26)
where Fn contains the contributions from worldsheets with n boundaries.

It is easy to calculate the variation S of the worldsheet action under the scale transformation,
and it can be shown that the variation is a sum of vertex operators corresponding to the state |B0 .
Since the boundary state is invariant, the variation of Fn under the scale transformation is obtained by inserting the state |B0 on the worldsheet, that is,

Fn = Fn |B0 .
(2.27)
To evaluate Fn (|B0 ), consider a worldsheet path-integral Fn+1 (z) with n + 1 boundaries, one
of whose boundary is placed at x I = zI with I = 5, . . . , 10. Note that the N 1 D-instantons we
have discussed so far are placed at x I = 0. In other words, we place another set of D-instantons
which are distributed parallel to the original D-instantons but their positions in x I -directions are
different. It is possible to obtain Fn (|B0 ) from Fn+1 (z) through an LSZ-like procedure:

Fn |B0 = d 6 z z Fn+1 (z),
(2.28)
where z is the Laplacian on R6 . See Fig. 4 for an image of this procedure. The variation of the
total free energy is therefore

F () = d 6 z z F (, z),
(2.29)
where
F (, z) =
n

n=0
n!
Fn+1 (z).
(2.30)
Since a single D-instanton is separated from the hypersurface along which

the other DI
instantons are distributed, the worldsheet is stretched in the region 0 |x | = x I x I r where
r = max{ls , h}
(2.31)
and h is the distance of the separated D-instanton from the hypersurface of the D-instantons. If
we take |zI | to be larger than r, then the corresponding worldsheet has a thin tube connecting
the boundary at x I = zI and the body of the worldsheet. The tube represents the propagation
of closed string states, and the dominant contribution comes from the massless propagation. As
a result, F (, z) behaves as |zI |4 for large |zI |. For the range 0 |zI | r, we assume that
F (, z) varies slowly with |zI |.
Let us calculate the integral

I = d 6 z z f (z),
(2.32)
where
f (z) =
f (0)
(0 |zI | r),
r4
f (0) (|zI | > r)
|zI |4
(2.33)
is a typical example of the function having the property assumed for F (, z). It is easy to check
that I = 4vol(S 5 )r 4 f (0). Similarly, the RHS of (2.29) would be estimated as

d 6 z z F (, z) r 4 C(, r)F (, z = 0),
(2.34)
where C(, r) is assumed to be of order one. Noticing that Fn+1 (z = 0) = Fn+1 , we obtain
F () r 4 C(, r) F (),
(2.35)
where
n

n=0
n!
Fn+1 = F ()
(2.36)
is used. In other words, the free energy transforms as

F () F + r 4 C(, r) ,
(2.37)
1
which indicates the existence of the scale invariance if r 4 . Note that the sum of the infinite
number of worldsheets with boundaries is crucial for the existence of this scale invariance.
It should be pointed out that our calculation has been done with an appropriate analytic continuation of . At first, the free energy is defined as a power series of as (2.26). This definition
is valid when the effects from the D-instantons are small. However, to estimate the variation of
the free energy, we assume the behavior of the sum F (, z), not each Fn (z), and therefore, this
estimate should be valid even beyond the convergence radius of the perturbative series (2.26).
1
The result that there exists a scale invariance if r 4 agrees with the existence of an isometry
of the corresponding metric when is large. To see the importance of the analytic continuation,
let us consider the metric of D-instantons distributed along a four-dimensional hypersurface [7]

1
ds 2 = H () 2 dx dx + d 2 + 2 d52 ,
(2.38)
where
H () = 1 +
C
.
4
(2.39)
In the region 4 , the metric (2.38) behaves as

dx dx + d 2
2
2
+ d5 ,
ds C
2
(2.40)
which is a metric on AdS5 S 5 . In this metric, the scale invariance is realized as the isometry
x = x ,
= .
(2.41)
From the perturbative string point of view, the metric (2.38) is given as a power series of .
1
By summing them up, we obtain the non-trivial function H () 2 . The possibility to have such
a closed expression enables us to go beyond the convergence radius of the perturbative series
where the isometry exists.
We consider a worldsheet which is relevant to a high energy scattering of open string states
on the D3-branes. As is explained above, one of the D3-brane is separated from the other N 1
D3-branes to implement the IR regularization, and all the external open string states are on the
separated D3-brane.
We make a scale transformation in the T-dual picture which is isotropic and brings the worldsheet to a region far away from the D-instantons. The situation is depicted in Fig. 5. If we take
to be large, then the worldsheet can be brought to a place far enough from the hypersurface of the
D-instantons so that the presence of the D-instantons is represented by the curved background
1
(2.38) while the scale invariance is still valid. In the region 4 , the background (2.38) is
approximated by AdS5 S 5 . Then, the classical solution of the string is the minimal surface
obtained in [5]. The size of the minimal surface is determined by the momenta of the external
open string states, which become larger and larger by the scale transformation. It should be noted
that, before the scale transformation, the size of the worldsheet is smaller than the string scale, as
mentioned at the beginning of this section. After the scale transformation, this classical solution
will dominate the summation over worldsheets. Note that, since the scale transformation for the
Fig. 5. Scale transformation in the T-dual picture.
D-instanton system is isotropic, the ratio of the size of the minimal surface to the distance from
the hypersurface does not change. This implies that the worldsheet always exists near the boundary ( = 0) of AdS5 . Notice that in the D-instanton case, the near horizon region corresponds
to the region of the boundary of AdS5 , as can be seen from the metric (2.38). This is in contrast
with the situation in [1] where D3-branes are placed at the center of the AdS5 spacetime from
the gravity point of view.
Since the scale transformation is a symmetry as long as the worldsheet is restricted in the
1
region |x I | 4 , the classical action for the minimal surface provides the high energy scattering
amplitude with which we started. In this way, the scale invariance found in [1] provides the
argument which verifies the relation claimed in [5].
One may think that the approximate symmetry is broken by the presence of the non-trivial
dilaton background
e = H ().
(2.42)
Since
Recall that the dilaton coupling in the worldsheet action is a sub-leading order term of
the actual expansion parameter is R 2 where R is a typical length scale of the target spacetime,
1
which is proportional to 4 in this case, the contribution from the dilaton background would be
negligible if we take a large .
3. Finite temperature
Next, we consider D3-branes at finite temperature. This is realized by considering a Euclideanized D3-brane which wraps on a circle with the circumference . AdS/CFT correspondence
has also been considered in this case [8]. The argument on the scale transformation can be also
carried out in the case of the finite temperature.
One of the crucial points of our perturbative approach is the transformation property of the
boundary state of the wrapped D3-brane. Since we impose the anti-periodic boundary condition for spacetime fermions in the Euclidean time direction on which the D3-brane wraps, the
fermionic worldsheet variables flip their signs as the string winds around the direction. The scale
transformation for these winding sectors is the same as that for the zero-winding sector, except
for the fact that the modings of S a and S a are half-integral if the winding number is an odd
10
integer. Due to the presence of these winding sectors, the boundary state has the form

|B; w,
|B =
(3.1)
wZ
where w is the winding number, and

1
j
ij i
ab a b
M n n iM Sn Sn |B0 ; 2k,
|B; 2k = exp
n
n=1

1
j
ij i
ab a
b
|B; 2k + 1 = exp
M n n iM Sn+ 1 Sn+ 1 |B0 ; 2k + 1.
n
2
2
(3.2)
(3.3)
n=1
The term |B; 2k has almost the same form with the supersymmetric boundary state [9], and the
transformation property is the same except for the obvious scaling of the circumference . It
might look non-trivial to analyze the transformation property of the term |B; 2k + 1, due to the
half-integral moding of the fermionic oscillators. However, one can check that the calculations
can be done similarly with the case of |B; 2k, and obtain
|B; 2k + 1 |B; 2k + 1.
(3.4)
Therefore, we can define the scale transformation so that the boundary state |B is scale invariant up to the scaling of the circumference . Then, AdS/CFT correspondence also follows at
finite temperature case, as in [1]. Note that the length scale of the background, if exists, is also
transformed by the scale transformation. This fact implies that, if one considers a Wilson loop
realized as in [1], then it is related to a minimal surface placed at the outside of the event horizon
of a black hole background, since the boundary of the worldsheet is always at the outside of the
event horizon.
There are other researches in the case at finite temperature which discuss the entropy of N = 4
SYM and that of the AdSSchwarzschild black hole [8,10]. Due to the limitation of the explicit
calculations, the entropy is calculated in the classical gravity only when is large, and in the
SYM only when is small. The corrections to these results have also been calculated in [1114]
which suggest that the entropy in the gravity region is smoothly interpolated to the entropy in
the SYM region by varying . These are the arguments supporting that the SYM entropy and the
black hole entropy are the same for any . In the following, we will show the coincidence of the
entropy at large .
Our point of view on AdS/CFT correspondence is based on D3-branes in the flat spacetime.
The temperature is encoded in the radius of the Euclideanized time direction. The temperaturedependence thus comes from strings which wind around the time direction. Let us consider the
free energy of strings in this case. In the large N limit with kept fixed, the winding closed
strings with the genus h 1 provide contributions of order gs2h2 2h2 N 22h which is at
most of order N 0 . On the other hand, the winding open strings may provide contributions of
order N 2 . Therefore, in the large N limit, the temperature-dependent part of the free energy is
dominated by open strings. When the temperature is small, then only massless open string states
contribute. As a result, the thermodynamical quantities which are obtained from the temperaturedependence of the free energy, for example the energy and the entropy, are equal to those of the
SYM in the limiting case mentioned above.
From an observer at the asymptotically flat region, the thermal D3-branes are regarded as a
non-extremal black hole

1
1
ds 2 = H (r) 2 f (r) dt 2 + d x 2 + H (r) 2 f (r)1 dr 2 + r 2 d52 ,
(3.5)
11
where
r04
4
,
f
(r)
=
1
.
(3.6)
r4
r4
The constant r0 is determined so that the Euclideanized version of (3.5) does not have a conical
singularity. If is taken to be large, then r0 can be written as
H (r) = 1 +
2
.
r0 =
(3.7)
Note that here is the circumference of the Euclidean time circle at the asymptotically flat
region. Therefore, this coincides with the one appeared in the D-brane setup. In this way, we
can equate the temperature of the gravity side with that of the gauge theory side.
The gravity description is valid when r0 is large. If we take to be large, we can also take
to be large while keeping r0 still large. This means that there exists a parameter region in which
both the gravity description and the SYM description are valid. (The description in terms of free
SYM cannot be valid in this region, of course.) The ADM mass of the black hole must be the
same with the total energy of the SYM plus the contribution from the tension of the D3-branes.
Since the entropy is obtained from the thermodynamic relation
S
1
(3.8)
= ,
E T
we can conclude that the entropies of the SYM and the black hole must be the same, due to
the coincidence of the energy and the temperature, although it is difficult to perform an explicit
evaluation of the entropy of the SYM with large .
The crucial points of our argument are that the temperature-dependence of the free energy is
dominated by open strings in the large N limit, and that there is a region of parameters where
both the gravity description and the SYM description are valid. It is also important that some
of physical quantities, the energy and the temperature for example, in two descriptions can be
compared directly with each other at the asymptotically flat region which is absent after taking
the near-horizon limit.
4. Discussion
We have shown that our point of view of AdS/CFT correspondence, based on the perturbative
type IIB string theory with D3-branes, can be applied to some situations discussed in the literature
of AdS/CFT correspondence. The amplitude of high energy scattering processes of the SYM can
be related to classical worldsheet configurations in the AdS5 space obtained in [5] by the scale
transformation, which clarifies the reason why such a correspondence holds. Thermal properties
of the SYM are also discussed in our point of view. It can be shown that the reduction of open
string system on D3-branes to the SYM occurs even in the case of large , and some thermal
properties of the SYM can be related to some gravitational quantities, the connection being made
at the asymptotically flat region far from the D3-branes.
The successful applications of our point of view indicate that it would enable us to make
further predictions for the correspondence which are not known so far. As long as the correspondence is based on our scale invariance, the coincidence of some quantities will have a firm
footing. We hope to provide some new and non-trivial correspondences between gauge theory
and gravity, possibly less supersymmetric, which could be checked by explicit calculations.
12
Acknowledgements
We would like to thank M. Fukuma, T. Matsuo, T. Takayanagi, T. Uematsu for valuable discussions. We would also like to thank L. Dixon for valuable comments on the IR regularization of the
SYM amplitudes. This work is supported by the Grant-in-Aid for the 21st Century COE Center for Diversity and Universality in Physics from the Ministry of Education, Culture, Sports,
Science and Technology (MEXT) of Japan. The research of T.S. is supported in part by JSPS
Research Fellowships for Young Scientists.
References
[1] H. Kawai, T. Suyama, AdS/CFT correspondence as a consequence of scale invariance, arXiv: 0706.1163.
[2] S. Rey, J. Yee, Macroscopic strings as heavy quarks: Large-N gauge theory and anti-de Sitter supergravity, Eur.
Phys. J. C 22 (2001) 379, hep-th/9803001.
[3] J.M. Maldacena, Wilson loops in large N field theories, Phys. Rev. Lett. 80 (1998) 4859, hep-th/9803002.
[4] Z. Bern, L.J. Dixon, V.A. Smirnov, Iteration of planar amplitudes in maximally supersymmetric YangMills theory
at three loops and beyond, Phys. Rev. D 72 (2005) 085001, hep-th/0505205.
[5] L.F. Alday, J. Maldacena, Gluon scattering amplitudes at strong coupling, arXiv: 0705.0303.
[6] T. Eguchi, H. Kawai, Reduction of dynamical degrees of freedom in the large N gauge theory, Phys. Rev. Lett. 48
(1982) 1063;
G. Parisi, A simple expression for planar field theories, Phys. Lett. B 112 (1982) 463;
G. Bhanot, U.M. Heller, H. Neuberger, The quenched EguchiKawai model, Phys. Lett. B 113 (1982) 47;
D.J. Gross, Y. Kitazawa, A quenched momentum prescription for large N theories, Nucl. Phys. B 206 (1982) 440.
[7] G.W. Gibbons, M.B. Green, M.J. Perry, Instantons and seven-branes in type IIB superstring theory, Phys. Lett.
B 370 (1996) 37, hep-th/9511080.
[8] E. Witten, Anti-de Sitter space, thermal phase transition, and confinement in gauge theories, Adv. Theor. Math.
Phys. 2 (1998) 505, hep-th/9803131.
[9] M.B. Green, M. Gutperle, Light-cone supersymmetry and D-branes, Nucl. Phys. B 476 (1996) 484, hep-th/9604091.
[10] S.S. Gubser, I.R. Klebanov, A.W. Peet, Entropy and temperature of black 3-branes, Phys. Rev. D 54 (1996) 3915,
hep-th/9602135.
[11] A. Fotopoulos, T.R. Taylor, Remarks on two-loop free energy in N = 4 supersymmetric YangMills theory at finite
temperature, Phys. Rev. D 59 (1999) 061701, hep-th/9811224.
[12] M.A. Vazquez-Mozo, A note on supersymmetric YangMills thermodynamics, Phys. Rev. D 60 (1999) 106010,
hep-th/9905030.
[13] C. Kim, S. Rey, Thermodynamics of large-N super YangMills theory and AdS/CFT correspondence, Nucl. Phys.
B 564 (2000) 430, hep-th/9905205.
[14] S.S. Gubser, I.R. Klebanov, A.A. Tseytlin, Coupling constant dependence in the thermodynamics of N = 4 supersymmetric YangMills theory, Nucl. Phys. B 534 (1998) 202, hep-th/9805156.

Stationary black holes and attractor mechanism

Dumitru Astefanesei a , Hossein Yavartanoo b,
a Global Edge Institute, Tokyo Institute of Technology, Tokyo 152-8550, Japan
b Center for Theoretical Physics and BK-21 Frontier Physics Division Seoul National University,
Seoul 151-747, South Korea

Received 4 July 2007; accepted 22 October 2007
Abstract
We investigate the symmetries of the near horizon geometry of extremal stationary black hole in fourdimensional Einstein gravity coupled to Abelian gauge fields and neutral scalars. Careful consideration of
the equations of motion and the boundary conditions at the horizon imply that the near horizon geometry
has SO(2, 1) U (1) isometry. This compliments the rotating attractors proposal of hep-th/0606244 that
had assumed the presence of this isometry. The extremal solutions are classified into two families differentiated by the presence or absence of an ergo-region. We also comment on the attractor mechanism of both
branches.
1. Introduction
The attractor mechanism plays a key role in understanding the entropy of non-supersymmetric
extremal black holes in string theory [1,2]. In certain cases, the macroscopic entropy of extremal non-supersymmetric attractor horizons can be matched to the weak coupling statistical
entropy despite the fact that these quantities do no seem to be protected by supersymmetry
[37].
It was originally noticed in [8] that the extremal four-dimensional Kerr and KerrNewman
black holes have an SO(2, 1) U (1) isometry. The last year, the authors of [9], found even
more four-dimensional extremal black holes had this isometry. Emboldened by this observation,
E-mail addresses: dumitru@th.phys.titech.ac.jp (D. Astefanesei), yavar@phya.snu.ac.kr (H. Yavartanoo).

doi:10.1016/j.nuclphysb.2007.10.015
14
D. Astefanesei, H. Yavartanoo / Nuclear Physics B 794 (2008) 1327
they found that, for four-dimensional stationary extremal black holes, in a theory of gravity
with neutral scalar fields non-minimally coupled to Abelian gauge fields, one can generalise
the entropy function formalism of [10] simply by assuming an SO(2, 1) U (1) near horizon
geometry.
The generalised entropy function is constructed, on an SO(2, 1) U (1) symmetric background, by taking the Legendre transform (with respect to the electric charges and angular
momentum) of the reduced Lagrangian evaluated at the horizon. Extremising the entropy function is equivalent to the equations of motion and its extremal value corresponds to the entropy.
Since the entropy function depends only on the near horizon geometry, its extremum and hence
the entropy is independent of the asymptotic data. This is precisely the attractor behaviour.
However, if the entropy function has flat directions something interesting happens: while the
extremum remains fixed, flat directions will not be fixed by near horizon data and can depend on
the asymptotic moduli.
There exist two distinct branches of stationary extremal black hole solutions which, in [9],
are dubbed ergo- and ergo-free branches according to their properties.1 The first branch, also
known as the fast branch, can exist for angular momentum of magnitude larger than a certain
lower bound and does have an ergo-region. On the other hand, the ergo-free branch can exist
only for angular momentum of magnitude less than a certain upper bound. The ergo-free branch
can also be smoothly connected to a static extremal black hole.
The entropy function has no flat directions for the ergo-free branch: the scalar and all other
background fields at the horizon are independent of the asymptotic data. However, there is a
drastic change for the ergo-branchthe entropy function has flat directions: despite the entropy
being independent of the moduli, the near horizon fields are dependent on the asymptotic data.
We find it significant that, the existence of an ergo-region allows energy to be extracted classically either by the Penrose process for point particles or by superradiant scattering for fields.
It is tempting to believe that the presence of the ergo-sphere is intimately related to the appearance of flat directions. One might say that the ergo-branch, not completely isolated from its
environment due to these processes, retains some dependence on the asymptotic moduli. From
this perspective, it is amazing that the black hole is isolated enough for the entropy to remain
independent.2
A consistent microscopic picture for KaluzaKlein (KK) black hole in agreement with the
macroscopic analysis of rotating attractors [9] was provided in [3,4]. That is, the D-brane model
reproduces the entropy of KK black hole, while the mass gets renormalized from weak to strong
coupling just for the ergo-branch black hole solutions in agreement with the existence of the
flat directions in the entropy function for this branch. Emparan and Maccarrone, [3], have also
provided a microscopic interpretation for the superradiant ergosphereeven if the temperature is
vanishing, the extremal black holes with ergosphere correspond to states with both left- and rightmoving excitations such that the open strings can combine and the emission of closed strings is
possible. The extraction of energy should reduce the angular momentum in such a way that the
event horizon area is increasing (it cannot decrease in the classical processes). Indeed, since
the left-moving excitations have spin, the emitted closed string will necessarily carry angular
momentum away from the black hole.
1 The existence of two branches in the moduli space of extremal rotating black holes was discussed for the first time
in [11].
2 It is possible that the addition of higher derivative terms might lift these flat directions. It would be interesting to see
whether this would erase the ergo-sphere.
15
In this note we fill up a gap in the proposal of [9] by proving that the near horizon geometry
of extremal rotating black holes in EinsteinHillbert gravity coupled to Abelian gauge fields and
neutral scalar fields has an enhanced SO(2, 1) U (1) symmetry. Unlike the static case where
the near horizon geometry is AdS2 S 2 , the AdS2 part does not decouple from the angular
part and the values of the moduli at the horizon have an angular dependence. Also, by adding
angular momentum to static black holes, the SO(3) symmetry of the sphere is broken to U (1).
However, the near horizon geometry is still universal in the sense that is still independent of the
coupling constants and is determined just by charges and angular momentum parameter. The
attractor mechanism is related to the extremality rather than to the supersymmetry property of
the theory/solution. Indeed, the enhanced symmetry of the near horizon geometry and the long
throat of AdS2 is at the basis of the attractor mechanism for stationary black holes [9,10].
2. Generalities
We consider a theory of gravity coupled to a set of massless scalars and vector fields, whose
general bosonic action has the form

1
A B
F
I G , i , AI = 2 d 4 x G R 2gij () i j fAB ()F
k
M

1
A B
fAB ()F F

(1)
,
2 G
A with A = (0, . . . , N ) are the gauge fields, i with (i = 1, . . . , n) are the scalar fields,
where F
and k 2 = 16G4 . The moduli determine the gauge coupling constants and gij () is the metric
in the moduli space. We use Gaussian units to avoid extraneous factors of 4 in the gauge fields,
and the Newtonss constant is set to G4 = 1.
Varying the action we obtain the following equations of motion for the metric, moduli, and
the gauge fields:

1
i
j
A
B
A
B
R 2gij = fAB 2F F G F F
(2)
2

1 fAB A
1 fAB A
1
F F B +
F F B , (3)
Ggij j =
i
4
G
8 G i

1
fAB F B
= 0.
G fAB F B +
(4)
2 G
To get the equations of motion, we have varied the moduli and the gauge fields independently.
The Bianchi identities for the gauge fields are F A [;] = 0.
We are interested in stationary black hole solutions to the equations of motion. In general
relativity the boundary conditions are fixed. However, in string theory one can obtain interesting
situations by varying the asymptotic values of the moduli and so, in general, the asymptotic
moduli data should play an important role in characterizing these solutions. Indeed, the nonextremal black hole solutions are characterized by the usual conserved charges and also by the
scalar chargesthe scalar charge is defined as the monopole in the multipoles expansion of
the scalar field at the boundary. Thus all its properties are moduli dependent, e.g., the entropy
depends by the asymptotic values of the moduli. However, the entropy of extremal solutions
obtained by taking the smooth limit when the temperature is vanishing is independent of the
16
asymptotic moduli data. We will see in the next section that the enhanced symmetry of their near
horizon geometry make them special in this regard.
Now let us write down the most general stationary black hole solution by using just its symmetries.3 An asymptotically flat spacetime is stationary if and only if there exists a Killing
vector field, , that is time-like at spatial infinityit can be normalized such that 2 = 1. It
was also been shown that stationarity implies axisymmetry [12] and so the event horizon is a
Killing horizon. Using the time-independence and axisymmetry we can write the most general
stationary metric with an axial Killing vector, , as
ds 2 = gtt dt 2 + 2gt dt d + g d 2 + grr dr 2 + g d 2 .
(5)
The event horizon of a stationary black hole is a Killing horizon of t + , where the
constant coefficient is the angular velocity of the horizon. It is convenient to rewrite the metric
(5) in the ADM form

ds 2 = N 2 dt 2 + ij dx i + N i dt dx j + N j dt

2
= N 2 dt 2 + g d + N dt + grr dr 2 + g d 2 ,
(6)
and so we obtain:
N2 =
(gt )2
gtt ,
g
N =
gt
,
g
ij = gij .
In this form, the event horizon corresponds to the surface N 2 = 0. The shift vector, N , evaluated
at the horizon reproduces the angular velocity of the horizon:

gt
.
= N H =
g H
By eliminating the conical singularity in the Euclidean ( = it, r) sector, we obtain the temperature

(N 2 )
1

=
T=
(7)
.

4 N 2 grr H
3. Near horizon geometry of extremal black holes
We consider a generic covariant two derivative gravity Lagrangian that has three basic components: metric, scalars, and gauge fields. We show that, given a few simple assumptions, the
near horizon geometry of a stationary, extremal spinning black hole solutions of this Lagrangian
necessarily has the near horizon symmetry SO(2, 1) U (1). To prove the previous statement,
we make use of the following ingredients:
Symmetries: we assume time independence and axisymmetry;

The black hole is extremalin other words the surface gravity (temperature) is zero;
We expand the fields near the horizon and take a scaling limit;
Gauge choices;
3 The thermodynamics of the non-extremal black hole solutions using the method developed in [1315] will be presented in [16].
17
Finiteness of certain physical quantities;

Equations of motion;
Spherical topology of the horizon.
3.1. Constraining the metric
As a warm-up exercise, we begin by examining 4-dimensional spherically symmetric black
holes by using the following ansatz:
ds 2 = a(r)2 dt 2 + a(r)2 dr 2 + b(r)2 d 2 .
(8)
The near horizon geometry of the extremal black holes can be obtained in two steps: first, take
the extremal limit when the temperature is vanishing (this is a smooth limit on the Lorentzian
section) and then obtain the near horizon geometry. We expand the metric components near the
outer horizon and for the non-extremal solution (r+ = r ) we obtain:

a 2 = f (r) = f0 + f1 + f2 2 + ,
( + )
+
b2 =
(9)
=
2
a
f 0 + f 1 + f2 2 +
and so the temperature is f0 = 4T . Here we used a coordinates system such that the horizon is
at = r r+ = 0 and defined the non-extremality parameter = r+ r .
The extremal limit is obtained for 4T = f0 0 and to obtain the near horizon geometry
we also take = 0. By changing the coordinate = t/f1 one can easily obtain the AdS2 S 2
explicitly

1
1
1
2
2
2
2
d + 2 d + d 2 .
ds =
(10)
f1
f1
We use a similar method to obtain the near horizon geometry of stationary extremal black
holes. However, the extremal limit in this case is more subtle since we should also consider the
non-diagonal component ( d dt) of the metric. Let us first rewrite the metric components in a
more useful form:
N = + (r r+ )(r, ),
N 2 = (r r+ )(r r )(r, ),
1
,
grr =
(11)
(r r+ )(r r )(r, )
where (r, ), (r, ), and (r, ) are regular functions. The temperature can be read off as
before and using Eq. (7) we obtain
r+ r
( )( ),
T=
(12)
4
where () and ( ) are the values of (r, ) and (r, ) at the outer horizon. For a nonextremal blackhole the temperature (surface gravity) is finite and constant on the horizon and
so we obtain ( )( ) = C, where C is a constant that depends on the charges P , Q and
the (mass and angular) parameters m, a. Expanding the ADM form of the metric near the outer
horizon we obtain the following metric:

2
(r r+ )(r r )( ) dt 2 + g d + (r r+ )( ) dt
1
(13)
+
dr 2 + g d 2 .
(r r+ )(r r )( )
18
To obtain the near horizon geometry, we first construct the following family of metrics
t
(14)
t ,
where is an arbitrary parameter. There is a smooth limit 0 for which the near horizon
geometry is obtained. Obviously, this is important for stationary field configurations where there
exist also terms of the form dr dt. This limit is especially useful when we consider the near
horizon expansion of the gauge fields.
Taking the extremal limit (r+ r ) and choosing a particular gauge we obtain

2 ( )
sin2
dr 2
d + r( ) dt + 2 d 2 ,
ds 2 = ( ) r 2 dt 2 + 2 2 +
(15)
( )
C r
C
r r+ + r,
where = t
. To obtain the above expression we use the appropriate coordinates system in
which g = r 2 grr and the gauge freedom to write ( )g ( ) = sin2 . Here, we have considered the metric in a rotating frame with respect to a distant observer with the angular velocity
equal to that of the black hole. For a horizon with spherical topology, we require

sin2
0,
2,
(16)
( )2 , ,
C 2 2 ( )
such that the deformed horizon, labelled by the coordinates (, ), is a smooth deformation of
the sphere. Unlike the static case, the fields at the horizon have an angular dependence and so
solving the attractor equations requires boundary conditions, i.e., the values of the fields at the
poles of the horizon.
Let us end up this subsection with an important comment about the extremal limit. For a
stationary black hole there are three intensive parameters associated to the horizon: the angular
velocity, the temperature, and the electric (magnetic) potential. Thus, there are two interesting
extremal limits T = 0 when the angular velocity is or is not vanishing. In the discussion section
we comment further on the physics of the extremal black holes.
3.2. Constraining the scalars and gauge fields
Let us start by investigating the scalar and the gauge fields configuration in the near horizon
limit. For simplicity, we do not carry on the moduli and gauge fields indices in this subsection
we specialize to one scalar and one gauge field configuration, but the generalization to a configuration with more than one scalar and one gauge field is straightforward. Expanding the scalars
at the horizon, r = 0, we obtain

(r, t) = r ( ) + r1 ( ) + O r 2 + .
(17)
Requiring that the scalars are finite at the horizon, implies 0 and then by taking the scaling
limit, r r, 0, we find

0 ( ), = 0,
=
(18)
0,
> 0,
in the near horizon region. We assume that the near horizon effective gauge coupling f (( )) is
well behaved and can be Taylor expanded around the poles, i.e.,

f ( ) = f0 + f1 + O 2 .
(19)
19
Let us turn to the gauge fields and perform a similar analysis. We impose that the gauge fields
are time-independent and start with the following ansatz
A = At (r, ) dt + Ar (r, ) dr + A (r, ) d + A (r, ) d,
(20)
that can be further simplified by choosing an appropriate gauge choice to fix A = 0 (or Ar = 0).
We can expand the gauge fields about the horizon as

dr

+ r a ( ) + O(r) d.
A = r at ( ) + O(r) r dt + r ar ( ) + O(r)
r
(21)
Requiring F 2 remains finite at the horizon implies , , 0. We take , , = 0 so that

after taking the scaling limit r r, t t/, 0 we obtain a non-zero result. With this
assumption the scaling limit gives
dr
+ a ( ) d + O().
r
The Einstein equations can be written as

1
R 2 = f 2F F g F F
.
2
A = at ( )r dt + ar ( )
(22)
(23)
The (r ) equation plays an important role in what follows: using the results from Appendix A
and the fact that = 0, we get
sin2 ( )
() ( ) = 0,
2 ( )
(24)
which implies () is, in fact, a constant. This was the last step in our proofit is straightforward
to check that the metric (15) with () a constant function has the SO(2, 1) U (1) isometry.
4. Attractor mechanism
In this section, we consider the attractor mechanism for static and stationary black holes. For
the static black hole solutions, we show the equivalence of the entropy function formalism and
the effective potential method. Entropy function formalism was generalized to stationary black
holes in [9]. We comment on the role played by the enhanced symmetry of the near horizon
geometry in decoupling the moduli equations of motion at the horizon from the bulk.
4.1. Static black holes
The Bianchi identity and equation of motion for the gauge fields can be solved by a field
strength of the form

1
F A = f AB QB fBC P C 2 dt dr + P A sin d d,
b
(25)
where P A , QA are constants that determine the magnetic and electric charges carried by the
gauge field F A , and f AB is the inverse of fAB .
As discussed in [17], the equations of motion for the moduli are

1 Veff
,
r a 2 b2 gij r j = 2
2b i
(26)
20
where Veff ( i ) is a function of scalars fields i given by

Veff (i ) = f AB QA fAC P C QB fBD P D + fAB P A P B .
(27)
It is clear from Eq. (26) that Veff ( i ) is an effective potential for the scalar fieldsit plays an
important role in describing the attractor mechanism [17,18].
For the attractor mechanism it is sufficient for two conditions to be met. First, for fixed
charges, as a function of the moduli, Veff must have a critical point. Denoting the critical values for the scalars as i = 0i we have,
i Veff (i0 ) = 0.
(28)
Second, the matrix of second derivatives of the potential at the critical point,
1
Mij = i j Veff (i0 ),
(29)
2
should have positive eigenvalues.
Once the two conditions mentioned above are met it was argued in [17] that the attractor
mechanism works and the entropy is given by the effective potential at the horizon.
The near horizon geometry is AdS2 S 2 and so we can apply Sens entropy function [10]
to investigate the attractor behaviour of static extremal solutions. All other background fields
respect the SO(2, 1) SO(3) symmetry of AdS2 S 2 . We keep the analysis general in order to
understand the role of Veff .
In [10], Sen found that the entropy of a spherically symmetric extremal black hole is the
Legendre transform of the Lagrangian densitythe only requirements are gauge and general
coordinate invariance of the action. In fact, this is similar with a generalization of the Walds
formalism for extremal black holes and it is based on the observation that there is a smooth
extremal limit on the Lorentzian section of a charged black hole.
The entropy function is defined as

F (
u, v , e , p)
= 2 ei qi f (
(30)
u, v , e , p)
= 2 ei qi d d GL ,
where d GL is the Lagrangian density, qi = f/ei are the electric charges, us are the
moduli values at the horizon, pi and ei are the near horizon radial magnetic and electric fields,
and v1 , v2 are the sizes of AdS2 and S 2 , respectively. Thus, F /2 is the Legendre transform of
and
f with respect to the variables ei . Then, for an extremal black hole of electric charge Q
magnetic charge P , Sen have shown that the equations determining u , v and e are given by:
F
= 0,
us
F
= 0,
vi
F
= 0.
ei
(31)
Thus, the black hole entropy is given by S = F (

u, v , e , p)
at the extremum (31). We observe that
the entropy function, F (
u, v , e , p),
determines the sizes v1 , v2 of AdS2 and S2 , and also the near
horizon values of moduli us and gauge field strengths ei .
Now, we are ready to apply this method to our action (1). The general metric of AdS2 S 2
can be written as

1
ds 2 = v1 2 d 2 + 2 d 2 + v2 d 2 + sin2 d 2 .
(32)
21
The field strength ansatz is

A
F A = Fr
dr d + P A sin d d = eA dr d + P A sin d d
(33)
and so

F (v1 , v2 , e, q, p) = 2 qA eA f (v1 , v2 , e, p) ,

8
v2 A B v1 A B
e e + p p
2fAB eA eB .
f (v1 , v2 , e, p) = 2 v2 + v1 fAB
v1
v2
k
(34)
The attractor equations are:

F
=0
v1
F
=0
v2
F
=0
i
F
=0
eA
v2
1
fAB eA eB fAB p A p B = 0,
2
v2
v1
1
v1
1 + fAB eA eB 2 fAB p A p B = 0,
v1
v2
fAB
fAB A B
p p eA eB = 2 i eA p B ,
i

16 v2
qA = 2
fAB eB fAB p B .
v1
k
(35)
(36)
(37)
(38)
By combining the first two equations we obtain v = v1 = v2 = fAB (eA eB + p A p B ) that is expecting also from our near horizon geometry analysis above. Its also easy to check that the
entropy is given by F at the attractor critical point:

16 2
(39)
fAB eA eB + p A p B = v.
k2
Using the electromagnetic field ansatz (25), it can be easily shown that S = Veff , qA = QA
(the sign appears because of our convention for FtrA ), and (38) matches the critical point
condition of Veff .
S=F =
4.2. Stationary black holes

We have shown in the previous section that the near horizon geometry of extremal spinning
black holes has the symmetries of AdS2 S 1 and can be written as

dr 2
ds 2 g dx dx = v1 ( ) r 2 dt 2 + 2 + 2 d 2 + 2 v2 ( )(d r dt)2
(40)
r
and the most general field configuration consistent with the SO(2, 1) U (1) symmetry of AdS2
S 1 is of the form:
i = ui ( ),

1 A
F dx dx = eA bA ( ) dr dt + bA ( ) d (d r dt),
2
(41)
where , and ei are constants, and v1 , v2 , ui , and bA are functions of . Here is a periodic
coordinate with period 2 and takes value in the range 0 .
22
Based on this observation, a generalized entropy function was proposed in [9]

F d d 2 J + QA eA d det GL ,
(42)
and so there is one more attractor equation associated to the angular momentum J . Thus, the
entropy and the near horizon background of a spinning extremal black hole are obtained by
extremaizing this entropy function that depends only on the parameters labelling the near horizon
background and the electric and magnetic charges and the angular momentum carried by the
black hole.
Interestingly, in all known cases, the appearance of flat directions in the entropy is associated
with the presence of an ergo-sphere. Since not all moduli are fixed at the horizon the mass is
not guaranteed to be fixed. The microscopic analysis of [3] confirms that in fact the mass is not
fixed and so there is a nice microscopic interpretation for the ergo-branch. On the other hand,
the slowly spinning extremal black holes in the ergo-free branch lack the rotational superradiance but can produce superradiant amplification of KK electric charged waves. However, this
phenomenon cannot be easily seen in the CFT since it is related to a modification of the central
charge.
One important question is if there is a similar effective potential for stationary black holes and
if one can use a similar analysis as in the static case to study the attractor mechanism. Unfortunately, at this point, we have just shown that the equations of motion at the horizon decouple from
the bulkwe hope to report a detailed analysis elsewhere. Here, let us just indicate the main step
in this analysis. We start by trying to solve the equations of motion and Bianchi identities for
the gauge fields. However, unlike the static case, we cannot obtain a general expression for the
gauge fields, but rather their expressions in terms of two unknown functions (A and u):
FrtM = r AM
FtM = AM
t (r, ),
t (r, ),

r uN (r, )
uN (r, )
F M r = f MN i
,
F M = f MN i
.
g
g
(43)
One can write down the expressions of the gauge fields in some concrete examples. However,
an interesting exercise is to work with this general form of the gauge fields and try to extract
as much information as possible from the equations of motion. For the moduli the equations of
motion become

1 fMN
1
r AM r AN + r 2 AM AN
G i =
i
2
G
1 f MN
+
(44)
r uM r uN + r 2 uM uN .
2 i
However, in this case, the moduli have also an angular dependence and the equations do not
decouple. In principle one should be able to read off the effective potential from the right-hand
side of this equation, but that is not straightforward in this casean effective potential for constant scalar fields was proposed in [9]. The best thing we can do is to check what is happening
in the near horizon limit. After some tedious manipulations we found that in the near horizon
limit the moduli equations are decoupled from the bulk. The scalar fields at the horizon have
also an angular dependence and we obtain a system of distributions rather than functions. Thus,
the boundary conditions, i.e., the values of the fields at the poles of the horizon, are important
and the equations are difficult to be solved in a general caseconcrete examples are presented
in [9].
23
5. Discussion
Recently, after the proposal of Sen [10], there was a lot of work on attractor mechanism
and entropy function (see, e.g., [19]). Motivated by the generalization of the attractor mechanism to non-supersymmetric extremal stationary black holes, we investigated the near horizon
geometry of spinning extremal black holes in a theory of gravity with uncharged scalar fields
non-minimally coupled to Abelian gauge fields. We found that the near horizon geometry of
these black holes has the symmetry of AdS2 S 1 the AdS2 part does not decouple from the
angular part. Consequently, the horizons are attractors for the moduli and their geometry is independent of the boundary moduli data. One subtlety is that the extremal spinning black holes
are further divided in two branches: ergo- and ergo-free branch, respectively. In both cases the
SO(2, 1) isometry of AdS2 is generated by the Killing vectors:
L1 = t ,
L0 = tt rr ,
2
L1 = (1/2) 1/r + t 2 t (tr)r + (/r) ,
(45)
but they have distinct properties. The former is characterized by an entropy function with flat
directions and for the latter there is no flat directions of the entropy function. If there are no flat
directions then, clearly, the entropy is independent of the moduli. On the other hand, if there are
flat directions, then the extremization of the entropy function does not determine all the moduli
values at the horizon. Location of these parameters along the flat directions may depend on the
asymptotic values of the moduli. But since the entropy function does not depend on the flat
directions, the entropy is still independent of the asymptotic values of the moduli, and so has an
attractor behaviour.
Let us comment now on the physics of the two branches. In general, in the supergravity
approximation,
the entropy is a function of the duality invariant combinations D(QA , P B ),

S = (|D| J 2 ) and the mass saturates an extremality bound that is independent of the angular momentum parameter, J 2 the plus sign corresponds to the ergo-free branch and the minus
sign to the ergo-branch. When D = J 2 the extremal horizon disappears and becomes a naked
singularitythis situation resembles the static case with one charge. Except this situation, the
extremal limit has finite area and zero surface gravity. The fastly spinning extremal black holes
have a non-zero horizon angular velocity and so their causal structure is similar with Kerr solution. Let us start with a non-extremal black hole that Hawking radiates. Clearly, Hawking
radiation carries away the angular momentum and so the black hole is slowing down. If the
black hole is radiating away all the angular momentum before reaching the extremal limit, then
the corresponding solution will be in the ergo-free branch. On the other hand, if the black hole
reaches the extremal limit and the angular velociy is non-zero, then there is radiation due to the
ergo-region. If the evaporating process is fine tuned such that the extremal limit is reached when
J = |D|, then the black hole behaves more as an elementary particle [20]there are potential
barriers outside the horizon which increase without bound.
In this paper we also tried to extend the analysis of the effective potential to extremal spinning black holes. We have not been able to conclusively construct an explicit effective potential,
mainly because of technical obstacles. In the static case one can explicitly check that the moduli are fixed at the attractor horizon that is a critical point for the effective potential. A similar
analysis is difficult for the stationary case. However, by studying the equations of motion for the
24
moduli, we concluded that they decouple from the bulk at the horizon.4 A complete determination of the scalar fields at the horizon needs also imposing the boundary conditions which are the
values of the fields at the poles of the horizon.
The near horizon geometry of a stationary extremal black hole is universal and so the entropy
does not depend of couplings. The extremality condition is very powerfull to force an attractor
behaviour of the horizonit is independent of the supersymmetry of the theory/solution. This
does not come as a surprise, though, since the near horizon geometry has an enhanced symmetry
and the long throat of AdS2 is the main ingredient for the existence of the attractor mechanism.
Acknowledgements
We thank Kevin Goldstein for collaboration in the initial stages of this work and for further
discussions. It is also a pleasure to thank Soo-Jong Rey, Ashoke Sen, and Sandip Trivedi for
useful conversations. D.A. would like to thank KIAS, Seoul for hospitality during part of this
work. D.A. has presented this work at ISM06 Puri (December 2006), KIAS, Seoul (February
2007), YITP, Kyoto (February, 2007), TITECH, Tokyo (May 2007) and he likes to thank the
audience at all these places for their positive feedback. The work of D.A. has been done with
support from MEXTs program Promotion of Environmental Improvement for Independence of
Young Researchers under the Special Coordination Funds for Promoting Science and Technology, Japan. D.A. also acknowledges support from NSERC of Canada. H.Y. would like to thank
the Korea Research Foundation Leading Scientist Grant (R02-2004-000-10150-0) and Star Faculty Grant (KRF-2005-084-C00003). While this paper was being completed, Ref. [21] appeared
which overlaps with the material presented in Section 3.
Appendix A. The Maxwell equations in the near horizon limit
In this appendix we explicitly obtain the equations of motion for the gauge fields in the near
horizon limit. These expressions are useful in Section 3.2for simplicity, we specialize again to
a configuration with one scalar and one gauge field, but the generalization is straightforward.
The non-zero components of the Maxwell tensor are given by

Fti = (at ( ), rat ( ), 0)
(A.1)
where i {r, , }.
Fi = (ar ( )/r, 0, a ( ))
Raising the indices we obtain

C2
( )
,0 ,
at ( ),
F ti = 2
r
( )

2
C
2 ( )
i
2
F = 2
a ( ) + ( ) ( ) ,
C rar ( ), 0,
( )
sin2
C2
F r = 2 ()rat ( ),
( )
(A.2)
(A.3)
(A.4)
where ( ) = at ( ) ()a ( ).

4 This is not a sufficient condition for the attractor mechanism to exist. However, a rigurous proof was given in [9] by
using the entropy function formalism.
25
Maxwells equations are

Gf F = 0.
(A.5)
From the r-component of Maxwells equation we obtain

1 ( ) sin( )f ( )ar ( ) = 0,
(A.6)
which can be integrated to give

ar ( ) = 1
1
( )
,
f sin
(A.7)
where we have assumed that the effective gauge coupling at the north pole, f ( = 0), is well
behaved. Now, for F 2 to be finite at = 0, we require 1 = 0, i.e., ar = 0. This in turn means
that ar ( ) does not contribute to the Maxwell tensor and can be gauged away.
Similarly from the t-component of Maxwells equation we obtain

1 sin( )f ( ) ( ) = 0,
(A.8)
which, by an argument similar to the one for ar above, implies is zero. Some important relations
used in this derivation are
G = C 2 ( ) sin
(A.9)
and
(s )2 = g
2
( )
1 1
r t + C 2 r 2 r2 + C 2 2 + 2 2 ,
=
( )
sin
2
2
2C
2C
F 2 = 2 at2 2 + C 2 ar2 + 2 (a )2 .
sin
(A.10)
(A.11)
References
[1] A. Dabholkar, A. Sen, S. Trivedi, Black hole microstates and attractor without supersymmetry, hep-th/0611143.
[2] D. Astefanesei, K. Goldstein, S. Mahapatra, Moduli and (un)attractor black hole thermodynamics, hep-th/0611140.
[3] R. Emparan, A. Maccarone, Statistical description of rotating KaluzaKlein black hole, Phys. Rev. D 75 (2007)
084006, hep-th/0701150.
[4] R. Emparan, G.T. Horowitz, Microstates of a neutral black hole in M theory, Phys. Rev. Lett. 97 (2006) 141601,
hep-th/0607023.
[5] G.T. Horowitz, D.A. Lowe, J.M. Maldacena, Statistical entropy of nonextremal four-dimensional black holes and
U-duality, Phys. Rev. Lett. 77 (1996) 430, hep-th/9603195.
[6] J.M. Maldacena, A. Strominger, Statistical entropy of four-dimensional extremal black holes, Phys. Rev. Lett. 77
(1996) 428, hep-th/9603060.
[7] A. Dabholkar, Microstates of non-supersymmetric black holes, Phys. Lett. B 402 (1997) 53, hep-th/9702050.
[8] J.M. Bardeen, G.T. Horowitz, Phys. Rev. D 60 (1999) 104030, hep-th/9905099.
[9] D. Astefanesei, K. Goldstein, R.P. Jena, A. Sen, S.P. Trivedi, Rotating attractors, JHEP 0610 (2006) 058, hepth/0606244.
[10] A. Sen, Black hole entropy function and the attractor mechanism in higher derivative gravity, JHEP 0509 (2005)
038, hep-th/0506177.
[11] D. Rasheed, The rotating dyonic black holes of KaluzaKlein theory, Nucl. Phys. B 454 (1995) 379, hep-th/
9505038.
[12] S. Hollands, A. Ishibashi, R.M. Wald, A higher dimensional stationary rotating black hole must be axisymmetric,
Commun. Math. Phys. 271 (2007) 699, gr-qc/0605106.
26
[13] D. Astefanesei, E. Radu, Quasilocal formalism and black ring thermodynamics, Phys. Rev. D 73 (2006) 044014,
hep-th/0509144.
[14] R.B. Mann, D. Marolf, Holographic renormalization of asymptotically flat spacetimes, Class. Quantum Grav. 23
(2006) 2927, hep-th/0511096.
[15] D. Astefanesei, R.B. Mann, C. Stelea, Note on counterterms in asymptotically flat spacetimes, hep-th/0608037.
[16] D. Astefanesei, R.B. Mann, C. Stelea, in preparation.
[17] K. Goldstein, N. Iizuka, R.P. Jena, S.P. Trivedi, Non-supersymmetric attractors, Phys. Rev. D 72 (2005) 124021,
hep-th/0507096.
[18] S. Ferrara, G.W. Gibbons, R. Kallosh, Black holes and critical points in moduli space, Nucl. Phys. B 500 (1997) 75,
hep-th/9702103.
[19] P.K. Tripathy, S.P. Trivedi, Non-supersymmetric attractors in string theory, JHEP 0603 (2006) 022, hep-th/0511117;
M. Alishahiha, H. Ebrahim, Non-supersymmetric attractors and entropy function, JHEP 0603 (2006) 003, hepth/0601016;
R. Kallosh, N. Sivanandam, M. Soroush, The non-BPS black hole attractor equation, JHEP 0603 (2006) 060, hepth/0602005;
B. Chandrasekhar, S. Parvizi, A. Tavanfar, H. Yavartanoo, Non-supersymmetric attractors in R2 gravities,
JHEP 0608 (2006) 004, hep-th/0602022;
S. Bellucci, S. Ferrara, A. Marrani, On some properties of the attractor equations, Phys. Lett. B 635 (2006) 172,
hep-th/0602161;
S. Ferrara, R. Kallosh, On N = 8 attractors, Phys. Rev. D 73 (2006) 125005, hep-th/0603247;
M. Alishahiha, H. Ebrahim, New attractor, entropy function and black hole partition function, JHEP 0611 (2006)
017, hep-th/0605279;
S. Kar, S. Majumdar, Noncommutative D(3)-brane, black holes and attractor mechanism, Phys. Rev. D 74 (2006)
066003, hep-th/0606026;
S. Ferrara, M. Gunaydin, Orbits and attractors for N = 2 MaxwellEinstein supergravity theories in five dimensions,
Nucl. Phys. B 759 (2006) 1, hep-th/0606108;
P. Kaura, A. Misra, On the existence of non-supersymmetric black hole attractors for two-parameter CalabiYaus
and attractor equations, Fortschr. Phys. 54 (2006) 1109, hep-th/0607132;
G.L. Cardoso, V. Grass, D. Lust, J. Perz, Extremal non-BPS black holes and entropy extremization, JHEP 0609
(2006) 078, hep-th/0607202;
H. Arfaei, R. Fareghbal, Double-horizon limit and decoupling of the dynamics at the horizon, JHEP 0701 (2007)
060, hep-th/0608222;
B. Chandrasekhar, H. Yavartanoo, S. Yun, Non-supersymmetric attractors in BI black holes, hep-th/0611240;
G.L. Cardoso, B. de Wit, S. Mahapatra, Black hole entropy functions and attractor equations, hep-th/0612225;
R. DAuria, S. Ferrara, M. Trigiante, Critical points of the black-hole potential for homogeneous special geometries,
hep-th/0701090;
S. Bellucci, S. Ferrara, A. Marrani, Attractor horizon geometries of extremal black holes, hep-th/0702019;
A. Sen, Entropy function for heterotic black holes, JHEP 0603 (2006) 008, hep-th/0508042;
A. Ghodsi, R4 corrections to D1D5p black hole entropy from entropy function formalism, Phys. Rev. D 74 (2006)
124026, hep-th/0604106;
P. Prester, Lovelock type gravity and small black holes in heterotic string theory, JHEP 0602 (2006) 039, hepth/0511306;
A. Sinha, N.V. Suryanarayana, Extremal single-charge small black holes: Entropy function analysis, Class. Quantum
Grav. 23 (2006) 3305, hep-th/0601183;
B. Sahoo, A. Sen, Higher derivative corrections to non-supersymmetric extremal black holes in, JHEP 0609 (2006)
029, hep-th/0603149;
B. Chandrasekhar, BornInfeld corrections to the entropy function of heterotic black holes, hep-th/0604028;
R.G. Cai, D.W. Pang, Entropy function for 4-charge extremal black holes in type IIA superstring theory, Phys. Rev.
D 74 (2006) 064031, hep-th/0606098;
G.L. Cardoso, J.M. Oberreuter, J. Perz, Entropy function for rotating extremal black holes in very special geometry,
hep-th/0701176;
S. Nampuri, P.K. Tripathy, S. Trivedi, On the stability of non-supersymmetric attractors in string theory, arXiv:
0705.4554 [hep-th];
A. Ceresole, G. DallAgata, Flow equations for non-BPS extremal black holes, JHEP 0703 (2007) 110, hepth/0702088;
L. Andrianopoli, R. DAuria, S. Ferrara, M. Trigiante, Black-hole attractors in N = 1 supergravity, hep-th/0703178;
27
K. Saraikin, C. Vafa, Non-supersymmetric black holes and topological strings, hep-th/0703214;

M.R. Garousi, A. Ghodsi, On attractor mechanism and entropy function for non-extremal black holes/branes, hepth/0703260;
J.H. Cho, S. Nam, Non-supersymmetric attractor with the cosmological constant, arXiv: 0705.2892 [hep-th];
M. Cvitan, P.D. Prester, S. Pallua, I. Smolic, Extremal black holes in D = 5: SUSY vs GaussBonet corrections,
arXiv: 0706.1167 [hep-th];
R.G. Cai, L.M. Cao, On the entropy function and the attractor mechanism for spherically symmetric extremal black
holes, arXiv: 0704.1239 [hep-th];
O. Dias, P.J. Silva, Attractors and the quantum statistical relation for extreme (BPS or not) black holes, arXiv:
0704.1405 [hep-th];
N.V. Suryanarayana, M. Wapler, Charges from attractors, arXiv: 0704.0955 [hep-th].
[20] C. Holzhey, F. Wilczek, Black holes as elementary particles, Nucl. Phys. B 380 (1992) 447, hep-th/9202014.
[21] H.K. Kunduri, J. Lucietti, H. Reall, Near-horizon symmetries of extremal black holes, arXiv: 0705.4214.

Two-dimensional black holes in a higher derivative

gravity and matrix model
Kwangho Hur, Seungjoon Hyun , Hongbin Kim, Sang-Heon Yi
Department of Physics, College of Science, Yonsei University, Seoul 120-749, Republic of Korea
Received 24 August 2007; accepted 23 October 2007
Abstract
We construct perturbatively a class of charged black hole solutions in type 0A string theory with higher
derivative terms. They have extremal limit, where the solution interpolates smoothly between near horizon
AdS2 geometry and the asymptotic linear dilaton geometry. We compute the free energy and the entropy of
those solutions using various methods. In particular, we show that there is no correction in the leading term
of the free energy in the large charge limit. This supports the duality of the type 0A strings on the extremal
black hole and the 0A matrix model in which the tree level free energy is exact without any corrections
in the leading order.
PACS: 04.70.Bw; 04.70.-s; 04.70.Dy; 11.25.Tq; 11.25.Pm
Keywords: Black holes; Noncritical 0A string theory; 0A matrix model
1. Introduction
Black holes are interesting objects in gravity and string theory, which may be awaiting the
complete understanding on the quantum nature of gravity. Since black holes have a large curvature value region near the singularity, the non-perturbative formulation of gravity theory or its
non-perturbative stringy generalization may be needed to understand the full quantum nature
of black holes. Though full non-perturbative descriptions of M/string theories are not avail* Corresponding author.
E-mail addresses: khur@physics.umn.edu (K. Hur), hyun@phya.yonsei.ac.kr (S. Hyun), hongbin@yonsei.ac.kr

(H. Kim), shyi@phya.yonsei.ac.kr (S.-H. Yi).
doi:10.1016/j.nuclphysb.2007.10.018
K. Hur et al. / Nuclear Physics B 794 (2008) 2845
29
able yet, there are several interesting toy string models whose non-perturbative descriptions
are known. One such class of toy models is the one of matrix models which correspond to the
non-perturbative formulation of noncritical string theories. Specifically, various two-dimensional
string theories on the flat background can be reformulated in terms of matrix models. In some
cases, the dual matrix models are believed to be a complete non-perturbative formulation of the
corresponding noncritical string theories. Then, one may naturally ask whether there are black
hole solutions in a noncritical string theory, and if there are, how they can be incorporated and
understood in the context of dual matrix model.
It was proposed in [1] that the type 0A matrix model with = 0 and non-zero RR fluxes
q+ = q = q is dual to 0A string theory on the extremal black hole [2]. Later on, the type 0A
matrix model was generalized [3] to incorporate two kinds of RR fluxes and found that it depends
only on the combined flux1 Q = q+ + q . Since the generalized matrix model is the same as the
matrix model with just one kind of RR flux and there is no evidence for the existence of black
holes in the type 0A matrix model side, it was argued that the matrix model is not related to the
0A strings on black holes [3]. Furthermore, the curvature radius of the black hole solution is the
order of string length scale, independent of charge, and therefore the low energy gravity cannot
be trusted and it is not clear whether the black hole exists at all. Nevertheless, there are some
pursuits of matching between the matrix model and type 0A strings on black holes or AdS space
[46] with partial success.
In this paper we will try to extend these efforts by including lowest order corrections in the
low energy effective gravity for type 0A string theory. As we mentioned, the curvature radius of
the black hole is the order of string length scale, and therefore higher derivative terms should be
taken into account. One of the motivation of this work is to determine how the behavior of the
black hole geometry is modified under the higher derivative correction.
We find the perturbative evidence of the existence of charged black holes even with the higher
derivative terms in the type 0A string theory, which interpolate smoothly between the near horizon geometry and the asymptotic geometry. In the case of the extremal black hole, the near
horizon geometry is AdS2 , as usual. This is in contrast to the four-dimensional cases where it is
not easy to find the interpolating solutions. We compute the exact entropy of this extremal black
hole using Sens formalism and find the condition on the coefficient a of the higher derivative
term in order to have the extremal black hole solution. We also compute the free energy of the
non-extremal black holes using Euclidean action approach and Walds Noether charge method.
In the extremal limit, the leading term, in the large charge limit, of the free energy turns out to
be unaffected under the -correction. This agrees with the result from the matrix model, which
supports the duality of those two models.
The organization of the paper is as follows. In Section 2, we briefly review some relevant
aspects on the type 0A string theory and its dual 0A matrix model. We also review the black
hole solutions in type 0A string theory. In Section 3, we construct charged black hole solutions
in the presence of higher derivative terms in the metric. We construct solutions, perturbatively
in the coefficient a, and find the black hole geometries which interpolate near horizon region
and asymptotic region, which is linear dilaton geometry. In Section 4, we compute the free energy and the entropy of the charged black hole solutions, for both extremal and non-extremal
cases using various approaches. The computation supports the duality between the 0A matrix
1 There is an additional term which depends on the difference between RR fluxes, but it was irrelevant to arguments
for black holes [3].
30
model and the 0A strings on the extremal black hole geometry. In Section 5, we draw some
conclusions. In Appendix A, we review some relevant results in the Noether charge method and
in Appendix B, we derive a generic relation between the temperature dependence of the radius
of the time-like Killing horizon for a non-extremal black hole and the curvature radius of the
near-horizon geometry of the corresponding extremal black hole.
2. Black holes in the 0A string theory and the 0A matrix model
The nonchiral projection of two-dimensional fermionic string theories gives rise to two type 0
string theories, so-called type 0A and type 0B. Only NSNS and RR sectors survive under the
projection, and thus the type 0 theories contain bosonic fields only. In type 0A string theory, the
NSNS sector contains a graviton, a dilaton and a tachyon, while the RR sector includes two
one-form gauge fields [7,8]. It was known that the low energy effective theory of the 0A strings
admits charged black hole solutions [2]. Based on the computation of the free energy, it was
argued that the 0A string theory on those black holes is dual to the 0A matrix model [1,9,10]. In
Section 2.1, we review the low energy effective theory of 0A strings and its black hole solutions.
In Section 2.2, we give some relevant features in the 0A matrix model for the comparison with
black hole side results.
2.1. Black holes at the lowest order in : Review
The low energy effective action of type 0A string theory at the lowest order in is given
by [7]

1 2
8
2
2
I0 = d x g
e
R + 4 + f1 (T )(T ) + f2 (T )
2 2

+ 2 2
2
2
+
(1)
q + F q F ,
f3 (T ) F
f3 (T ) F
4
4
where F denote field strengths of two RR gauge fields and q denote the corresponding charges,
respectively. The theory admits the following linear dilaton geometry as a vacuum solution:
= k,
ds 2 = dt 2 + d 2 ,
(2)
where all other fields vanish. It is also known that there exist charged black hole solutions in the
model [1,2]. When the tachyon field T is turned off, RR gauge fields can be easily solved as
q
(3)
,
T = 0,
2
which corresponds to the configuration of the background D0-branes with charges given by
q = q.
One may use this to integrate out RR gauge fields, and obtain the action of the form

1 2
2
I0 = d 2 x g
(4)
R
+
4
+
.
e
+
4k
2 2
+
F01
= F01
=
Here we denoted the original cosmological constant as k 2 = 2/ and new effective cosmological
constant, coming from the gauge field contributions, as = q 2 /(2 ). Therefore the low
31
energy effective theory of type 0A string theory reduces to the two-dimensional dilaton gravity
with two kinds of cosmological constants, one of which is related to the charges of RR gauge
fields. The theory admits the charged black hole solutions in which the dilaton is taken to be
proportional to the spatial coordinate ,
0 = k
(5)
while the metric is of the form

ds 2 = l0 () dt 2 +
d 2
,
l0 ()
(6)
with the factor l0 () given by2

1
(7)
(M0 ).
2k
As will be clear, M0 may be regarded as a mass of the black hole. One can also obtain the
extremal black hole solution where the position of the horizon and the mass are given in terms
of the charge as

1 ln 2 .
e2kex = 2 , Mex = ex + 2ke2kex =
(8)
2k
4k
4k
l0 () = 1 e2k
The near horizon geometry of the extremal black hole becomes AdS2 .
2.2. The free energy in the 0A matrix model
In this section we give the expression of free energy in the 0A matrix model [3,7]. The free
energy, F ln Z, of the 0A matrix model with the Fermi energy level and RamondRamond
(RR) flux q compactified at the radius R (or at the temperature 1/(2T ) = /(2)) is given by

t/2
3 F0A
2 t/2R
i 2 t 12 qt
e
=
dt
2R Im
.
(9)
2
sinh(t/2) sinh( t/2R)
3
2
0
In this integral form non-perturbative effects for R and are included. Since it is enough for us
to consider the perturbative effects, we perform the series expansion on x/ sinh x and integrate
term by term. This gives us the perturbatively expanded form as

q 2m1
3 F0A
m+1
=
(1)
(2m)!
2R Re
i
2
2
2
3
m=0
2n
m

/2
|1 212(mn) ||1 212n ||B2(mn) ||B2n |
(10)
.
[2(m n)]!(2n)!
R
n=0

Note that index m corresponds to the genus expansion in the 0A string theory as ( 2 i q2 )1
in the matrix model plays the role of the string coupling gs in the type 0A theory. On the other
hand, the summation over n corresponds to the -expansion.
2 We set 2 2 = 1 from now on.
32
Since the thermodynamic free energy, F0A is defined by

F0A F0A ,
we obtain

2 1
1
q 2
q
F0A =
ln
Re
i

2
2
2
2
2

2

/2
1
q
+
1+
i
ln
24
R
2
2

q 22m
m2
+
(1)
(2m 3)!
Fm (R) ,
i
2
2
(11)
(12)
m=2
where
2n
m

/2
|1 212(mn) ||1 212n ||B2(mn) ||B2n |
.
Fm (R)
[2(m n)]!(2n)!
R
(13)
n=0
To match the matrix model results with those of the type 0A string theory on the twodimensional extremal black hole, we should take some appropriate limits in the above. First
of all, we should take the infinite radius limit (i.e., zero temperature limit) with = 0. Furthermore, we should double the RR flux q to get the effects from two kinds of RR flux q+ = q = q.
After all these limits are taken, we obtain the tree level part of the thermodynamic free energy as

2

2 1
2 q2
1
tree
F0A =
=
Re
iq ln
iq
ln q 2 . (14)

4

2
2
2
=0
Note that we have suppressed the ambiguous quadratic terms on q, which may have corrections. They are related to the divergent parts which exist in the free energy expressions and may
be regularized by the explicit cutoff. One may also note that there is no further correction in
the tree level free energy of the type 0A matrix model, which is not the case for higher genus
ones. In the next section, we will consider higher derivative corrections to the low energy effective action of type 0A string theory and check that it gives the same free energy with the type 0A
matrix model.
3. Black holes in the higher derivative type 0A gravity
Now we would like to include the higher derivative correction, i.e., higher order correction
to the action given in the previous section. We restrict ourselves to the higher derivative terms in
the NSNS sector fields only.3 The unique higher derivative term for the metric in two dimensions, which appear in the next order in correction, is an R 2 term. Furthermore, the results
from function computation [11] tell us that we do not need to consider higher derivative terms
in . Henceforth, it is enough to add the following correction term to the action (4):

1
I1 = 2 d 2 x ge2 a R 2 ,
(15)
2
3 One may include the generic higher derivative terms in the RR sector as well.
33
where a is a certain dimensionless number which may be fixed by the computation of corrections in string theory.
3.1. Black hole solutions
The equations of motion of the metric and the dilaton are found to be

1
4a
0 = R + 2 + g 2 2 4()2 + 4k 2 + e2 2 e2 e2 R ,
2
k
2a
0 = R + 4 2 4()2 + 4k 2 + 2 R 2 .
(16)
k
The metric for the black hole solutions of the above equations of motion is chosen to be the
Schwarzschild type as
ds 2 = l() dt 2 +
d 2
.
l()
(17)
In general, we have three equations of motion in which only two of them are independent. After
simple manipulation, the equations of motion for the metric component gtt and dilaton in the
above Schwarzschild type metric are given by
0 = l 6l 4l + 8l( )2 8k 2 e2 +
0 = l 4l 4l + 4l( )2 4k 2
4a
(2l l l l ),
k2
2a 2
(l ) ,
k2
dl

where = d
d and l = d . It is not easy to obtain a black hole solution analytically. Instead, we
use perturbative approach to find out charged black hole solutions. We require that the solutions
reduce to the well-known black hole solutions as a 0. We also require that the geometry approaches, asymptotically, to the linear dilaton geometry (2). The word perturbative here means
the perturbative expansion in terms of variable a in Eq. (15). Let us recall a is just a number
fixed by correction and not a parameter we may vary arbitrarily. Furthermore there is no guarantee that it is small. Nevertheless we may still regard it as an adjustable variable and try to do
a perturbative analysis. The partial justification of this approach is given by the comparison of
results from this approach with those from the exact one in the near horizon limit of extremal
black hole.
First of all, the analysis of the asymptotic behavior of the dilaton and the metric leads to

() = k + O e4k ,

1
l() = 1 e2k (M ) + O e4k ,
(18)
2k
where we denote M as a mass of the black hole. As will be shown in later section, there is an
ambiguity in defining the ADM mass in two dimensions, due to the divergencies. However the
difference M Mex , where Mex is the corresponding quantity in the extremal black hole with
the same charge, is well defined as the energy above the extremal black hole, and thus it maybe
justified to call M as the mass of the black hole. The mass of the black hole depends on a and
may be written generically in the form
M = M0 + aM1 + a 2 M2 + .
34
We introduce, for clarity, a dimensionless mass m and a dimensionless cosmological constant

as
M
,
= 2.
k
k
Through the perturbative expansion in a, we obtain the solutions of the equations of motion,
up to second order in a, as

= k 16a 2 e4k (m0 k + )2 + O a 3 ,
(19)
3
2
l() = l0 () + al1 () + a l2 () + O a ,
(20)
m=
where

1
(m0 k) ,
l0 () = 1 + e
2

m1
+ e4k 2 + 2(m0 k)2 ,
l1 () = e2k
2

m2
2k
+ e4k 962 + 4(m0 k)(m1 48) 128(m0 k)2

l2 () = e
2

+ e6k 83 + 82 (m0 k) + 48(m0 k)2 + 16(m0 k)3 .
(21)
2k
At first glance, one may think that mi s are independent parameters describing the above
perturbative solutions. However, this is not the case and the mass of a black hole is expressed
in terms of just a single parameter m as can be seen from the fact that all the expressions of the
given solution can be rewritten as the function of a single variable m up to the given order. In
terms of mass parameter m, the solution can be rewritten as

= k 16a 2 e4k (m k + )2 + O a 3 ,
(22)

e2k
(m k) + ae4k 2 + 2(m k)2
2

32a 2 e4k 32 + 6(m k) + 4(m k)2

+ 8a 2 e6k 3 + 2 (m k) + 6(m k)2 + 2(m k)3

+ O a3 .
l() = 1
(23)
This solution has several distinct features. First of all, the only combination which appear in
the metric and the dilaton is of the form
p (m k)q e2(p+q)k .
an
(24)
in the dilaton include, generically, terms proportional to

Secondly, the coefficients of
e2lk , with l = 2, 3, . . . , n, while the coefficients of a n in the metric function l() contain terms
proportional to e2lk , with l = 2, 3, . . . , n + 1. These generic features can be shown to be true
in all orders of a from the recursive structure in the equations of motion. From these properties,
we find an additional characteristic feature of the solution: the solution depends only on one
combination m
of two parameters m and where

m 1
m
= ln .
2
4
35
In order to see this, we introduce the shifted spatial coordinate :

1
=
ln .
2k
4
Then the generic form displayed in (24) becomes (m
k )
q e2(p+q)k , which depends only on
the parameter m.
This property will be useful later on.
3.2. Horizon and temperature
The horizon is determined in terms of given variables m and by the condition
l(H ) = 0.
(25)
Using the perturbative solution of the function l() given in Eq. (23), up to a 2 order, one can
reorganize the above relation between the mass parameter m and the horizon H , which is more
suitable to the perturbative determination of H , as

m kH = 2e2kH 1 + a 8 + 2 e4kH

8a 2 32 + 24e2kH + 82 e4kH + 3 e6kH + O a 3 .
The perturbative expression for the position of the horizon, H , can be obtained by expanding it
as a series in a as

0
1
2
+ aH
+ a 2 H
+ O a3 ,
H = H
(26)
i as functions of m (i = 0, 1, 2 . . .), and, in principle, as a function of m.
which gives us H
i
H
Temperature of these non-extremal black holes can be obtained from the condition of the
absence of the conical singularity in the Euclideanized black hole geometry. In the case at hand,
the temperature is given by the relation
T=
1
l (H ; m, ).
4
1
Note that in terms of the shifted coordinate, it becomes T = 4
l(H ; m),
which tells us that
there is no -dependence in the relation between the Hawking temperature T and the position
of the horizon in the shifted coordinate, H . As we have seen, H is determined by the mass
or vice versa (m
= m(
H )) from Eq. (25). After expressing the
parameter m
(H = H (m))
mass parameter m
in the metric function l()
in terms of the position of the horizon, H , the
above relation reduces to the one between the position of the horizon in the shifted coordinate
and the Hawking temperature. Then, the formula can be inverted and H can be written in terms
of T only. Up to a 2 order the perturbative relation is given by

1
k H (T ) = kH (T ) ln
2
4

2

1
T
T
T
= ln 1 2
(27)
8a 1 4
+ 8 2
+ O a3 .
2
k
k
k
One can take the extremal limit of these charged black hole solutions, which is characterized
by the conditions
l(ex ) = 0,
l (ex ) = 0,
36
where the second condition tells that the Hawking temperature of the extremal black hole is zero.
Up to a 2 order, the perturbative solution of the horizon position is given by

1
+ 4a + 16a 2 + O a 3 ,
kex = ln
(28)
2
4
while the parameter mex is given by

12a + 16a 2 + O a 3 .
mex = 1 ln
2
4
Note that the following combination of two parameters depends only on a:

1
1
(mex kex ) = 1 + 32a + O a 3 .
2
This fact can be confirmed to hold for all orders in a by using the equations of motion.
(29)
(30)
4. Free energy and entropy of black holes

In this section we find the free energy and the entropy of our black hole solutions using various
approaches. First of all we use Sens entropy function formalism to obtain the full expression of
entropy of the extremal black hole. Then we use Euclidean action formalism and Walds Noether
charge method to obtain the free energy and the entropy of non-extremal black holes.
4.1. Entropy of extremal black holes
For the extremal case, the exact expression for the entropy can be obtained for given Lagrangian by Sens entropy function formalism [1215]. In this section, we use Sens formalism
to obtain the full expression of the entropy of the extremal black hole. In the next section, we
will obtain the perturbative entropy of non-extremal black holes and compare it with the results
in this section.
One can show that the following field configurations, which have the SO(2, 1) symmetry,

dx 2
ds 2 = v x 2 dt 2 + 2 ,
(31)
x
= u,
(32)
are the solutions of the equations of motion (16). These configurations may appear as the nearhorizon limit of our extremal black holes. We found that it is the case, up to a 2 order, and we
believe it holds, generically, all orders in a. This is supported by the fact that our perturbative
solutions interpolate smoothly between the near horizon region, which is AdS2 geometry, and the
asymptotic region, which is linear dilaton geometry. It is in contrast to the higher-dimensional
case [16], with higher derivative terms, where it is not easy to find the solution which interpolates
smoothly between the near horizon and the asymptotic regions.
The entropy function is given by the value of the total Lagrangian at the horizon

2 8a 1
F (v, u) = 2v e2u + 2 2 + 4k 2 + k 2 .
(33)
v
k v
The value of u and v are fixed by the extremum conditions:
F
= 0,
v
F
= 0,
u
which lead to
1
v = 2 (1 + 1 32a ),
4k
2u

2 1
8a
= 2
1 2
.
k v
k v
37
(34)
The entropy of the extremal black hole is the extremum value of the entropy function and is given
by
S = F (v , u ) =
(1 + 1 32a ).
2
(35)
1
for the entropy to be real, which sets the condition for the
Note that a should satisfy a 32
model to have an extremal black hole solution. In order to compare with the perturbative results
in the next section, we recover the original physical parameter q 2 = 4 and expand in a and
obtain

q2
(36)
1 8a 64a 2 + O a 3 .
4
As will be shown, it is consistent with the results for the non-extremal black holes in the extremal
limit using the Euclidean action method and the Noether charge one.
S=
4.2. Non-extremal black holes: Euclidean action approach

There are several ways to get the entropy and finite temperature free energy for non-extremal
black holes. We employ two well-known methods to calculate the finite temperature free energy
for charged black holes, namely Euclidean action approach [17] and Walds Noether charge
method [18,19]. In this section, we get the free energy expression for non-extremal black holes
by using the Euclideanized action. Before proceeding, let us recall that the naive ADM mass is
divergent for a two-dimensional black hole which asymptotes to the linear dilaton background.
To regularize this divergence, one should introduce a cut-off. One way to remove the cut-off
dependent terms is to subtract the value of the reference spacetime. One natural choice for the
reference spacetime is the extremal black hole which has the same charge in the background as
non-extremal ones [20] (see also [21]).
In the Euclidean action approach, we need the GibbonsHawkingYork [17,22] boundary
terms to get the well-defined variational problem and the correct free energy. Since we have a
higher derivative correction term in the bulk Lagrangian, we should consider the corresponding
correction in the boundary terms, as well. It turns out that this correction in the boundary terms
has no effect to the free energy expression in the case at hand. The boundary terms, when the
bulk Lagrangian is written in terms of an arbitrary function, f (R), of the Ricci scalar R, can be
easily found and the total action becomes

d
IE =
(37)
gf (R) 2
hf (R)K, f (R)
f (R),
dR
M
where K is the trace of the second fundamental form of the boundary M. The form of the
boundary terms is valid even when f (R) contain other fields, for instance, the dilaton field. In
our case, f (R) is given by

2a
f (R) = e2 R + 2 R 2 + 4 + 4k 2 + .
k
38
After plugging in our perturbative solution and taking the boundary at = ,

the action of the
bulk part contributes to

d

gf (R) = 4ke2k + 3k 2 + aI1 () + a 2 I2 () ,
H
(38)
where

I1 () = e2k 4k 2 + 2(m k)2 ,

I2 () = e2k 128k 52 9(m k) 5(m k)2

+ e4k 32k 3 + 52 (m k) + 12(m k)2 + 4(m k)3 .
The boundary term in the Euclidean action is given by

1
4
hf (R)K = = l ()e2 1 + a 2 R
.
2
k
=
After plugging in our perturbative solutions, the boundary part of the action becomes

Ibd = k m k + + O e2k
.
2
(39)
(40)
(41)
(42)
The value of the total Euclidean action diverges as the cut-off goes to infinity. We remove
these cut-off dependent terms by subtracting the action value of the reference extremal black hole
with the same charges. The periodicity ex in the Euclidean time direction for the extremal black
hole is chosen from the relation [9,23,24]

= g (),
ex gex ()
(43)
so that the proper length of the reference geometry in the Euclidean time direction at the boundary
coincides with the one of the non-extremal black hole geometry. Its asymptotic form becomes
1

ex M Mex 2k
+ O e4k .
=
e
4k
The regularized free energy is defined by

FBH () =
IE () IEex (ex )
,
ex
which is determined to be, up to second order,

FBH (T ) = H (T ) ex ,
(44)
where H (T ) and ex are given in (27) and (28). Note that the temperature dependence of the
free energy is entirely incorporated in the horizon position H up to the second order in a. We
conjecture the relation (44) is an exact one, holding for all orders in a. One of the reason that the
relation (44) might be an exact one comes from the relation, derived in Appendix B, between the
temperature dependence of the horizon radius, or the position of the horizon in two dimensions,
of the non-extremal black hole and the curvature radius of the AdS2 geometry, which is the
near-horizon geometry of the corresponding extremal black hole. Indeed, from the relation (44),
39
conjectured to hold all orders in a, we obtain the entropy of the extremal black hole as

F
H (T )
S=
=
= 2v = 2 (1 + 1 32a )

T T =0
T
2k
T =0
which agrees with the exact expression (35) of the entropy of the extremal black hole.
Via a standard thermodynamic relation, we obtain the entropy of the non-extremal charged
black holes as
F
S=
T

1

1 8( Tk ) + 8 2 ( Tk )2
T
= 2 1 2
1 8a
k
k
1 2( Tk )

T
2 T 2
3 T 3
4 T 4

2 1 12( k ) + 48 ( k ) 96 ( k ) + 64 ( k )
64a
(45)
+ O a3 ,
T 2
(1 2( k ))
and indeed, in the extremal limit, the entropy becomes

q2
1 8a 64a 2 + O a 3 ,
4
which agrees with the result (36) from the Sens formalism in the previous section.
S=
(46)
4.3. Non-extremal black holes: Noether charge method

Now, we turn to the Noether charge method pioneered by Wald [18,19,26] and obtain the same
results as Euclideanized action method if we take the same prescriptions for removing the cut-off
dependent terms. By using this method we may obtain more insights into the structure of free
energy. As was done in Euclidean action approach, we may regularize the divergent expression
by subtracting the reference spacetime which is given by the extremal black hole. For the basic
materials on the Noether charge method relevant for our discussions, see Appendix A.
The Noether charge for the action variation under the coordinate transformation by vector
field is

4a
4aR
.
Q = 2 e2 1 + 2 R [ ] + 2 [ ] e2 1 + 2
(47)
k
k
The relevant Killing vector field for the ADM-like mass and the free energy is the timelike one t .
Then, as shown in Appendix A, the energy is given by

E = Qt + B ,
(48)
where
t

B
=e

4a
4a
1 + 2 R l 2l 1 + 2 R ,
k
k
= e2k [2kl l 8k] .
After plugging in the solution, the energy of the charged black hole is found to be

16a
2k
2 2k
E = M 2ke + 8k e +
.
k
(49)
(50)
(51)
40
One may note that, as explained in Appendix A, there remain ambiguities in the choice of surface
term and, accordingly, potential for currents Q in the Noether charge method. In the
higher-dimensional case, where ADM mass is well defined and finite, those ambiguities do not
affect the ADM mass and therefore do not represent any ambiguity. In two-dimensional case, the
ADM-like mass is apparently divergent and, thus, should be taken with care. One may wonder
whether those apparent ambiguities really lead to any real ambiguity in the ADM-like mass. It
turns out that they affect only on the cut-off dependent part, which will be removed eventually,
leaving the essential part,
M +
16a
invariant.
In the Noether charge method, the entropy is given by the value of Noether charge at the
horizon and thus it becomes

4a
1 t
2
1 + 2 R .
S = Q = 4e
(52)
T
k
H
The entropy itself is finite and well defined. It is totally free from the ambiguities in defining
Noether charge mentioned above. Through the thermodynamic relation among the free energy,
the energy and the entropy as
FBH = E T S,
(53)
we obtain the free energy FBH with respect to the reference geometry:
ex
FBH = FBH FBH

4a
= (M Mex ) T 4e2 1 + 2 R
.
k
H
(54)
When our black hole solutions are inserted in the above equation, the same results as Euclidean
action approach are obtained, up to the second order in a, as

FBH (T ) = H (T ) ex .
(55)
So far we have used the extremal black holes as our reference geometry to remove the cutoff dependent terms in the mass of non-extremal black holes. This is the same prescription we
used in the Euclidean action formalism where it seems to be the only natural choice to deal with
cut-off dependent terms. As a result, the free energy for the extremal black hole is assigned to be
zero. But in the Walds Noether charge method, there seems to be more natural prescription in
dealing with the cut-off dependent terms. We may just remove all the cut-off dependent terms in
the mass formula for the non-extremal black hole.4 If we follow this prescription, the free energy
becomes

16a
4a
2
.
T 4e
1+ 2R
FBH (T ) = M +
(56)
k
k
H
In the extremal limit, the free energy is given by

16a
1 q2
ex
FBH = Mex +
=
ln q 2 + O q 2 .

k
4
2
4 For another consistent prescription extracting the finite part of the ADM-like energy, see [25].
(57)
41
In the leading term, i.e., q 2 ln q 2 , there is no -correction, which agrees with the result from the
matrix model. There is inherently an overall factor 2 difference between the free energy of the
extremal black hole and the matrix model. This may be resolved by rescaling the time coordinate
of black hole.
5. Discussion
In this paper we constructed charged black hole solutions in the low energy effective theory
of type 0A string theory with the correction term. Though they are not exact forms, but are
given only perturbatively in a, they interpolate smoothly between the near horizon geometry and
the asymptotic dilaton geometry, which strongly indicates the existence of the solutions. They
also have the smooth extremal limit whose near horizon geometry is AdS2 , which can be easily
confirmed to be the exact solution of the equations of motion with the higher derivative term if
a 1/32.
We computed the free energy and the entropy of the non-extremal black holes using Euclidean
action and Walds Noether charge approach. There are divergent or cut-off dependent terms in
the computation of ADM-like mass in two dimensions due to the linear dilaton and fluxes. We
discussed the subtleties in removing those cut-off dependent terms. The leading term of free
energy in the large charge limit is independent of the method of removing cut-off dependent
terms and thus shown to be well defined. It was shown to be the same as the one from matrix
model. This supports the duality between the 0A string theory and the 0A matrix model. Though
the geometry itself is given only perturbatively, we found the exact expression of the entropy
of the extremal black hole and also found an exact relation between the free energy of the nonextremal black hole and the position of Killing horizon.
It is believed that the matrix model does not admit the black hole solution. On the other hand,
there are exact black hole solutions in the low energy effective theory of the dual type 0A string
theory. One argument against the existence of black hole solutions is that the curvature radius is
the order of string length scale, independent of the charge, and thus low energy effective gravity
theory cannot be trusted. We considered a possible higher derivative term in type 0A gravity and
found the condition of the existence of the black hole solutions. By direct computations in string
theory, one may get the value of a to confirm the (non-)existence of the 0A black holes.
It would be very interesting to include the higher derivative terms in one-form gauge fields and
the combination of them with NSNS fields. Since the 0A string theory is not supersymmetric,
it is not clear how to incorporate those terms. Another interesting problem is whether the duality
holds when we include the quantum correction in the 0A string theory due to the tachyon field.
Works in this direction is under progress.
Acknowledgements
K. Hur was supported by the Basic Research Program of the Korea Science and Engineering
Foundation under grant number R01-2004-000-10651-0. The work of S. Hyun was supported by
the Science Research Center Program of the Korea Science and Engineering Foundation through
the Center for Quantum Spacetime (CQUeST) of Sogang University with grant number R112005-021. S. Hyun, H. Kim and S.-H. Yi were supported by the Korea Research Foundation
Grant funded by Korea Government (MOEHRD, Basic Research Promotion Fund) (KRF-2005070-C00030).
42
Appendix A
In this appendix, we summarize some relevant results in Noether charge method for the entropy and the free energy by Wald [18,19,2628].
Under a general field variation a , the variation of the Lagrangian is given by

( gL) = gEa a + g ,
where Ea denotes the equations of motion for various fields, a , and corresponds to the surface
term. The Noether current of a vector field for diffeomorphism invariance can be obtained as
J = L .
(A.1)
The potential Q for the current is defined by

Q = J .
(A.2)
Though there are ambiguities in the definition of the potential Q, Wald showed that the black
hole entropy and the free energy can be written unambiguously in terms of Q as

1
S=
(A.3)
d Q ,
F = E T S,
T
H
where H denotes bifurcate Killing horizon for Killing vector and E is the ADM-like energy
which can also be obtained by Noether charge method. More explicitly, it was shown that, whenever one can write

d B B = d ,
E can be defined by

E = d Q B B .
(A.4)
Since the ambiguities in Q may play some roles in our case, we explain them in some detail.
One of them comes from the total derivative terms added in the Lagrangian, which do not change
the equations of motion

gL gL + g .
This additional term modifies as
1
+ + g g .
2
Note that the transformation of the added term in under the coordinate transformation (i.e., its
Lie derivative) is given by
1
+ g g = + .
2
There is another ambiguity in from its definition. Namely, one can modify as
+ Y ,
43
where Y is an antisymmetric tensor. This leads to the modified form of the Noether current

J J + + Y .
The Noether charge itself has ambiguity in its definition, J = Q , up to total derivative
term
Q Q + Z ,
where Z is a totally antisymmetric tensor. Therefore, these ambiguities can modify the
Noether charge, Q as
Q Q + + Y + Z .
(A.5)
For completeness, we give an example how the above various tensors are related to Walds differential forms [18,19]. The tensor is related to the component of the differential form 1 D1
in D-dimensional spacetime case as
=
1
1 D1 1 D1 ,
(D 1)!
g 01D1 = 1.
Appendix B
In this appendix, we derive a somewhat generic relation between the temperature dependence
of the radius of the time-like Killing horizon for a non-extremal black hole and the curvature
radius of the near horizon geometry in the extremal limit.
For concreteness, let us take the metric as a Schwarzschild type as
ds 2 = l(r) dt 2 +
dr 2
2
.
+ r 2 dD2
l(r)
Now, let us recall relations leading the horizon and temperature

1
d
l (rH ), l l(r).
4
dr
If the radius of the horizon and the mass of non-extremal black holes are assumed to be written as
an analytic function of temperature, which is not a valid assumption in the case of Schwarzschild
black holes, the generic form of the expansions are

rH T , Qa = r0 Qa + f Qa T + O T 2 ,

m T , Qa = m0 Qa + c Qa T + O T 2 ,
l(rH ) = 0,
T=
where Qa denote various charges in the given setup. Note that r0 and m0 are the radius of the
horizon and the mass of the corresponding extremal black hole, respectively, and thus satisfy
l(r0 ; m0 ) = l (r0 ; m0 ) = 0.
Then, the temperature relation, when l (rH ) is expanded near r0 , leads to

l r0 ; m(T ), Q
T +O T2 ,
4T = f (Q)l (r0 ; m0 , Q) + c(Q)
m
T =0
44
where there is no temperature independent term on the right-hand side due to the extremal condition, l (r0 ; m0 , Q) = 0. On the other hand, the horizon condition becomes

l(rH ; m, Q)
T +O T2 ,
0 = l(rH ; m, Q) = c(Q)
m
T =0
which gives the condition
l = 0.
m
As a result, we get

rH (T )
4
,
= f (Q) =

T
l (r0 ; m0 , Q)
T =0
c=0
as
(B.1)
which shows the relation between the temperature dependence of the horizons radius of a nonextremal black hole and the curvature radius of the near horizon geometry in the corresponding
extremal black hole, since the near horizon geometry in the extremal limit under the assumption
l (r0 ; m0 , Q) = 0 is given by

2
dx 2
1
2
2 2
2
x dt + 2 + r02 dD1
dsnear =
, x l (r0 ; m0 , Q)r.
l (r0 ; m0 , Q)
2
x
References
[1] S. Gukov, T. Takayanagi, N. Toumbas, Flux backgrounds in 2D string theory, JHEP 0403 (2004) 017, hep-th/
0312208.
[2] N. Berkovits, S. Gukov, B.C. Vallilo, Superstrings in 2D backgrounds with RR flux and new extremal black holes,
Nucl. Phys. B 614 (2001) 195, hep-th/0107140.
[3] J.M. Maldacena, N. Seiberg, Flux-vacua in two-dimensional string theory, JHEP 0509 (2005) 077, hep-th/0506141.
[4] A. Strominger, A matrix model for AdS2 , JHEP 0403 (2004) 066, hep-th/0312194.
[5] O. Aharony, A. Patir, The conformal limit of the 0A matrix model and string theory on AdS2 , JHEP 0511 (2005)
052, hep-th/0509221.
[6] P. Horava, C.A. Keeler, Strings on AdS2 and the high-energy limit of noncritical M-theory, arXiv: 0704.2230 [hepth].
[7] M.R. Douglas, I.R. Klebanov, D. Kutasov, J.M. Maldacena, E. Martinec, N. Seiberg, A new hat for the c = 1 matrix
model, hep-th/0307195.
[8] T. Takayanagi, N. Toumbas, A matrix model dual of type 0B string theory in two dimensions, JHEP 0307 (2003)
064, hep-th/0307083.
[9] U.H. Danielsson, J.P. Gregory, M.E. Olsson, P. Rajan, M. Vonk, Type 0A 2D black hole thermodynamics and the
deformed matrix model, JHEP 0404 (2004) 065, hep-th/0402192.
[10] M.E. Olsson, The stringy nature of the 2d type-0A black hole, JHEP 0605 (2006) 032, hep-th/0511106.
[11] R.R. Metsaev, A.A. Tseytlin, Order (two loop) equivalence of the string equations of motion and the sigma model
Weyl invariance conditions: Dependence on the dilaton and the antisymmetric tensor, Nucl. Phys. B 293 (1987) 385.
[12] A. Sen, Black hole entropy function and the attractor mechanism in higher derivative gravity, JHEP 0509 (2005)
038, hep-th/0506177.
[13] A. Sen, Entropy function for heterotic black holes, JHEP 0603 (2006) 008, hep-th/0508042.
[14] A. Sen, Black hole entropy function, attractors and precision counting of microstates, arXiv: 0708.1270 [hep-th].
[15] S. Hyun, W. Kim, J.J. Oh, E.J. Son, Entropy function and universal entropy of two-dimensional extremal black
holes, JHEP 0704 (2007) 057, hep-th/0702170.
[16] A. Sen, How does a fundamental string stretch its horizon? JHEP 0505 (2005) 059, hep-th/0411255.
[17] G.W. Gibbons, S.W. Hawking, Action integrals and partition functions in quantum gravity, Phys. Rev. D 15 (1977)
2752.
[18] R.M. Wald, Black hole entropy in the Noether charge, Phys. Rev. D 48 (1993) 3427, gr-qc/9307038.
45
[19] V. Iyer, R.M. Wald, Some properties of Noether charge and a proposal for dynamical black hole entropy, Phys. Rev.
D 50 (1994) 846, gr-qc/9403028.
[20] H. Liebl, D.V. Vassilevich, S. Alexandrov, Hawking radiation and masses in generalized dilaton theories, Class.
Quantum Grav. 14 (1997) 889, gr-qc/9605044.
[21] J.L. Davis, L.A. Pando Zayas, D. Vaman, On black hole thermodynamics of 2-D type 0A, JHEP 0403 (2004) 007,
hep-th/0402152.
[22] J.W. York, Role of conformal three geometry in the dynamics of gravitation, Phys. Rev. Lett. 28 (1972) 1082.
[23] G.W. Gibbons, R.E. Kallosh, Topology, entropy and Witten index of dilaton black holes, Phys. Rev. D 51 (1995)
2839, hep-th/9407118.
[24] S.W. Hawking, G.T. Horowitz, S.F. Ross, Entropy, area, and black hole pairs, Phys. Rev. D 51 (1995) 4302, grqc/9409013.
[25] J.L. Davis, R. McNees, Boundary counterterms and the thermodynamics of 2-D black holes, JHEP 0509 (2005)
072, hep-th/0411121.
[26] V. Iyer, R.M. Wald, A comparison of Noether charge and Euclidean methods for computing the entropy of stationary
black holes, Phys. Rev. D 52 (1995) 4430, gr-qc/9503052.
[27] T. Jacobson, G. Kang, R.C. Myers, On black hole entropy, Phys. Rev. D 49 (1994) 6587, gr-qc/9312023.
[28] T. Jacobson, G. Kang, R.C. Myers, Black hole entropy in higher curvature gravity, gr-qc/9502009.

Joint resummation for slepton pair production

at hadron colliders
Giuseppe Bozzi a , Benjamin Fuks b , Michael Klasen b,
a Institut fr Theoretische Physik, Universitt Karlsruhe, Postfach 6980, D-76128 Karlsruhe, Germany
b Laboratoire de Physique Subatomique et de Cosmologie, Universit Joseph Fourier/CNRS-IN2P3,
53 Avenue des Martyrs, F-38026 Grenoble, France

Available online 1 November 2007
Abstract
We present a precision calculation of the transverse-momentum and invariant-mass distributions for supersymmetric particle pair production at hadron colliders, focusing on DrellYan like slepton pair and
sleptonsneutrino associated production at the CERN Large Hadron Collider. We implement the joint resummation formalism at the next-to-leading logarithmic accuracy with a process-independent Sudakov
form factor, thus ensuring a universal description of soft-gluon emission, and consistently match the obtained result with the pure perturbative result at the first order in the strong coupling constant, i.e. at O(s ).
We also implement three different recent parameterizations of non-perturbative effects. Numerically, we
production and compare the resummed cross section with the perturbative result.
give predictions for eR eR
The dependence on unphysical scales is found to be reduced, and non-perturbative contributions remain
small.
PACS: 12.38.Cy; 12.60.Jv; 13.85.Qk; 14.80.Ly
Keywords: Resummation; Supersymmetry; Hadron colliders; Sleptons
1. Introduction
One of the main tasks in the experimental programme of the CERN Large Hadron Collider
(LHC) is to perform an extensive and conclusive search for the supersymmetric (SUSY) partners
E-mail address: klasen@lpsc.in2p3.fr (M. Klasen).

doi:10.1016/j.nuclphysb.2007.10.021
G. Bozzi et al. / Nuclear Physics B 794 (2008) 4660
47
of the Standard Model (SM) particles predicted by the Minimal Supersymmetric Standard Model
[1,2]. Scalar leptons are among the lightest supersymmetric particles in many SUSY-breaking
scenarios [3,4]. Presently, the experimental (lower) limits on electron, muon, and tau slepton
masses are 73 GeV, 94 GeV, and 81.9 GeV, respectively [5]. Since sleptons often decay into
the corresponding SM partner and the lightest stable SUSY particle, the distinctive signature at
hadron colliders will consist in a highly energetic lepton pair and associated missing energy.
The leading order (LO) cross section for the production of non-mixing slepton pairs has been
calculated in [69], while the mixing between the interaction eigenstates was included in [10].
The next-to-leading order (NLO) QCD corrections have been calculated in [11], and the full
SUSY-QCD corrections with non-mixing squarks in the loops have been added in [12]. Recently, an accurate calculation of the transverse-momentum (qT ) spectrum including soft-gluon
resummation at the next-to-leading logarithmic (NLL) accuracy has been performed [13], allowing for the reconstruction of the mass and the determination of the spin of the produced particles
by means of the Cambridge (s)transverse mass variable [14,15] and for distinguishing thus the
SUSY signal from the SM background, mainly due to W W and t t production [16,17]. Very recently, the mixing effects relevant for the squarks appearing in the loops have been investigated
at NLO, and the threshold-enhanced contributions have been computed at NLL [18]. The numerical results show a stabilization of the perturbative results through a considerable reduction of
the scale dependence and a modest increase with respect to the NLO cross section.
Since the dynamical origin of the enhanced contributions is the same both in transversemomentum and threshold resummations, i.e. the soft-gluon emission by the initial state, it would
be desirable to have a formalism capable to handle at the same time the soft-gluon contributions in both the delicate kinematical regions, qT M and M 2 s, M being the slepton pair
invariant-mass and s the partonic centre-of-mass energy. This joint resummation formalism has
been developed in the last eight years [19,20]. The exponentiation of the singular terms in the
Mellin (N ) and impact-parameter (b) space has been proven, and a consistent method to perform
the inverse transforms, avoiding the Landau pole and the singularities of the parton distribution
functions, has been introduced. Applications to prompt-photon [21], electroweak boson [22],
Higgs boson [23], and heavy-quark pair [24] production at hadron colliders have exhibited substantial effects of joint resummation on the differential cross sections.
In this paper we apply the joint resummation formalism at the NLL level to the hadroproduction of slepton pairs at the LHC, thus completing our programme (started in Ref. [13] and
continued in Ref. [18]) of providing the first precision calculations including soft-gluon resummation for slepton pair production at hadron colliders. In Section 2, we briefly review the
theoretical formalism of joint resummation following Refs. [20,22]. We reorganize the terms of
the resummed formula in a similar way as it was done for transverse-momentum resummation in
[25]. The inverse transforms from the Mellin and impact-parameter spaces and the matching of
the resummed result with the fixed-order perturbative results are discussed in Section 3. Section 4
is devoted to phenomenological predictions for the LHC, together with a comparison of the three
types of resummation (transverse-momentum, threshold, and joint), showing their impact on the
qT -spectrum and on the invariant-mass distribution. Our results are summarized in Section 5.
2. Joint resummation at the next-to-leading logarithmic order
We consider the hard scattering process
ha (pa )hb (pb ) F (M, qT ) + X,
(1)
48
where F is a generic system of colourless particles, such as a Higgs boson or a DrellYan

(s)lepton pair, M is the invariant mass of the final state F , and qT is its transverse momentum.
Thanks to the QCD factorization theorem, the unpolarized hadronic cross section

d2
=
2
2
dM dqT
a,b

dxa
dxb fa/ ha (xa ; F )fb/ hb (xb ; F )
d2 ab
(z; s , R , F )
dM 2 dqT2
(2)
can be written as the convolution of the relevant partonic cross section ab with the universal
distribution functions fa,b/ ha,b of partons a, b inside the hadrons ha,b , which depend on the
longitudinal momentum fractions of the two partons xa,b and on the unphysical factorization
scale F . The partonic scattering cross section depends on the strong coupling constant s ,
the unphysical renormalization and factorization scales R and F , and on the scaling variable
z = M 2 /s, where s = xa xb S and S = (pa + pb )2 are the partonic and hadronic centre-of-mass
energies, respectively. The lower limits for the integration over the longitudinal momentum fractions contain the quantity = M 2 /S, which approaches the value = 1 when the process is
close to the hadronic threshold M 2 S. In Mellin N -space, the hadronic cross section naturally
factorizes
dN
d2
=
N fa/ ha (N + 1; F )fb/ hb (N + 1; F )
2i
dM 2 dqT2
a,b
C
d2 ab
(N; s , R , F ),
dM 2 dqT2
(3)
where the contour C in the complex N -space will be specified in Section 3 and the N -moments
of the various quantities are defined according to the Mellin transform
1
F (N) =
dx x N1 F (x)
(4)
for x = xa,b , z, and F = fa/ ha ,b/ hb , , , respectively. The jointly resummed hadronic cross
section in N -space can be written at NLL accuracy as [20,22,23]
d2 (res)
(N; s , R , F )
dM 2 dqT2
2
(0)
d b ibqT
cc Hcc (s , R )
Cc/ ha (N, b; s , R , F )
e
=
4
c

exp Ec(PT) (N, b; s , R ) Cc/
hb (N, b; s , R , F ).
(5)
The indices c and c refer to the initial state of the lowest-order cross section c(0)
c and can then
only be q q or gg, since the final state F is assumed to be colourless.
For slepton pair and sleptonsneutrino associated production at hadron colliders,
()
ha (pa )hb (pb ) li (p1 )lj (p2 ) + X,
(6)
49
we have M 2 = (p1 + p2 )2 , qT2 = (p1T p2T )2 , and

eq el ij (LqqZ + RqqZ )Re(Lli lj Z + Rli lj Z )
2 3 2 2
eq el ij +
2
9M
4xW (1 xW )(1 m2Z /M 2 )
2 )|L
2
(L2qqZ + RqqZ
li lj Z + Rli lj Z |
,
+
2 (1 x )2 (1 m2 /M 2 )2
32xW
W
Z

|Lqq W Lli l W |2
2 3
=
q(0)
,
q
2 (1 x )2 (1 m2 /M 2 )2
9M 2 32xW
W
W
q(0)
q =
(7)
(8)
where i, j denote slepton/sneutrino mass eigenstates with masses mi,j , mZ and mW are the
masses of the electroweak gauge bosons, is the electromagnetic fine structure constant, xW =
sin2 W is the squared sine of the electroweak mixing angle, and the velocity is defined as

= 1 + m4i /M 4 + m4j /M 4 2 m2i /M 2 + m2j /M 2 + m2i m2j /M 4 .

(9)
The coupling strengths of the left- and right-handed (s)fermions to the electroweak vector bosons
are given by

{Lff Z , Rff Z } = 2Tf3 2ef xW ff ,

f f
f f
{Lfi f Z , Rfi f Z } = Lff Z Sj 1 Si1 , Rff Z Sj 2 Si2 ,
j
j
{Lqq W , Rqq W } = { 2cW Vqq , 0},

l
2cW Si1
,0 ,
{Lli l W , Rli l W } =

q
q
{Lqi qj W , Rqi ql W } = Lqq W Si1 Sj 1 , 0 ,
(10)
where the weak isospin quantum numbers are Tf3 = 1/2 for left-handed and Tf3 = 0 for
right-handed (s)fermions, cW is the cosine of the electroweak mixing angle, and Vff are the
CKM-matrix elements. The unitary matrices S f diagonalize the sfermion mass matrices, since
in general the sfermion interaction eigenstates are not identical to the sfermion mass eigenstates
(see Appendix A).
The function Hcc in Eq. (5) contains the hard virtual contributions and can be expanded perturbatively in powers of s ,
Hcc (s , R ) = 1 +

s (R ) n
n=1
(n)
Hcc (R ).
(11)
The coefficients
Cc/ ha (N, b; s , R , F ) =

a,b

Cc/b N; s (M/) Ub/a (N; M/, F )fa/ ha (N + 1; F )
(12)
and Cc/
hb , defined analogously, allow to evolve the parton distribution functions fa,b/ ha,b from
the unphysical factorization scale F to the physical scale M/ with the help of the QCD evo-
50
lution operator
2
Ub/a (N; , 0 ) = exp

dq 2
b/a N; s (q)
2
q
(13)
20
and to include, at this scale, the fixed-order contributions

s n (n)
Cc/b (N ),
Cc/b (N; s ) = cb +
(14)
n=1
that become singular when qT 0 (but not when z 1). The QCD evolution operator fulfills
the differential equation

dUb/a (N; , 0 )
=
Ub/c (N; , 0 )c/a N; s () ,
2
d ln
c
(15)
where the anomalous dimensions c/a (N; s ) are the N -moments of the AltarelliParisi splitting
functions. The function
N)
= b +
(b,
N
N
1 + b/
with b bMeE /2 and N N eE ,
(16)
organizes the logarithms of b and N in joint resummation. Its exact form is constrained by the
requirement that the leading and next-to-leading logarithms in b and N are correctly reproduced
in the limits b and N , respectively. The choice of Eq. (16) with = 1/4 avoids the
introduction of sizeable subleading terms into perturbative expansions of the resummed cross
section at a given order in s , which are not present in fixed-order calculations [22].
The perturbative (PT) eikonal exponent
M 2
Ec(PT) (N, b; s , R ) =

M2

d2
()
ln
+
B
()
A
c s
c s
2
2
(17)
M 2 / 2
allows to resum soft radiation in the A-term, while the B-term accounts for the difference between the eikonal approximation and the full partonic cross section in the threshold region, i.e.
the flavour-conserving collinear contributions. In the large-N limit, these coefficients are directly
connected to the leading terms in the one-loop diagonal anomalous dimension calculated in the
MS factorization scheme [26]
Bc (s )
c/c (N; s ) = Ac (s ) ln N
+ O(1/N).
2
They can thus also be expressed as perturbative series in s ,

s n (n)
s n (n)
Ac (s ) =
Ac
and Bc (s ) =
Bc .
n=1
(18)
(19)
n=1
Performing the integration in Eq. (17), we obtain the form factor up to NLL,
Ec(PT) (N, b; s , R ) = gc(1) () ln + gc(2) (; R )
(20)
51
with
(1)
Ac 2 + ln(1 2)
,
0

(1)
Ac 1 1 2
2 + ln(1 2)
gc(2) (; R ) =
(1
2)
+
ln
2
1 2
03
(1)

(2)
(1)
Ac
M 2 Ac
Bc
2
+
ln 2 2
ln(1 2)
+ ln(1 2) +
0
1 2
0
R
0
gc(1) () =
(21)
and = 0 /s (R ) ln . The first two coefficients of the QCD -function are

1
1
(22)
(11CA 4TR Nf ) and 1 =
17CA2 10TR CA Nf 6CF TR Nf ,
12
24
Nf being the number of effectively massless quark flavours and CF = 4/3, CA = 3, and TR =
1/2 the usual QCD colour factors.
In order to explicitly factorize the dependence on the parameter , it is possible to reorganize
the resummation of the logarithms in analogy to the case of transverse-momentum resummation
[25,27]. The hadronic resummed cross section can then be written as
0 =

dN
d2 (res)
b db
N
=
fa/ ha (N + 1; F )fb/ hb (N + 1; F )
J0 (bqT )
2
2
2i
2
dM dqT
a,b C
0

Habcc (N; s , R , F ) exp Gc (ln ; s , R ) .
(23)
The function Habcc does not depend on the parameter and contains all the terms that are
constant in the limits b or N ,

s (R ) n (n)
(0)
Habcc (N; R , F ) .
Habcc (N; s , R , F ) = cc ca cb
(24)
+
n=1
(1)
At O(s ), the coefficient Habcc is given by

(1)
(1)
(1)
(1)
Habcc (N; R , F ) = ca cb
Hcc (R ) + ca Cc/b
Cc/a (N )
(N ) + cb
M2
(1)
(1)
+ ca c/b
c/a (N ) ln 2 .
(N ) + cb
F
(25)
The -dependence appearing in the C-coefficient and in the evolution operator U of Eq. (12) is
(PT)
factorized into the exponent Gc , which has the same form as Ec defined in Eq. (17) except for
the substitution
Bc (s ) B c (N; s ) = Bc (s ) + 2(s )
d ln Cc/c (N; s )
+ 2c/c (N; s ).
d ln s
(26)
(2)
At NLL accuracy, Eq. (20) remains almost unchanged, since only the coefficient gc of Eq. (21)
has to be slightly modified by
(1)
Bc(1) B c(1) (N ) = Bc(1) + 2c/c (N ).
(27)
52

(1)
(1)
Although the first-order coefficients Ca/b (N ) and Hcc (R ) are in principle resummationscheme dependent [27], this dependence cancels in the perturbative expression of Habcc [25].
In the numerical code we developed for slepton pair production, we implement the DrellYan
resummation scheme and take Hq q (s , R ) 1. The C-coefficients are then given by
(1)
Cq/q
(N ) =
2
2 8
+
3N(N + 1)
3
(1)
(N ) =
and Cq/g
1
.
2(N + 1)(N + 2)
(28)
3. Inverse transform and matching with the perturbative result

Once resummation has been achieved in N - and b-space, inverse transforms have to be
performed in order to get back to the physical spaces. Special attention has to be paid to the singularities in the resummed exponent, related to the divergent behaviour near = exp[/(20 s )],
i.e. the Landau pole of the running strong coupling, and near b = 2N and b = 4N , where
= 0 and infinity, respectively. The integration contours of the inverse transforms in the Mellin
and impact parameter spaces must therefore avoid hitting any of these poles.
The b-integration is performed by deforming the integration contour with a diversion into the
complex b-space [21], defining two integration branches
b = (cos i sin )t
with 0 t ,
(29)
valid under the condition that the integrand decreases sufficiently rapidly for large values of |b|.
The Bessel function J0 is replaced by two auxiliary functions h1,2 (z, v) related to the Hankel
functions
+iv
1
h1 (z, v)
h2 (z, v)
d eiz sin ,
iv
iv
d eiz sin .
(30)
+iv
Their sum is always h1 (z, v) + h2 (z, v) = 2J0 (z), but they distinguish positive and negative
phases of the b-contour, being then associated with only one of the two branches defined in
Eq. (29).
The inverse Mellin transform is performed following a contour inspired by the Minimal Prescription [28] and the Principal Value Resummation [29], where one again defines two branches
N = C + zei
with 0 z , > >
.
2
(31)
The parameter C is chosen in such a way that all the singularities related to the N -moments
of the parton densities are to the left of the integration contour. It has to lie within the range
0 < C < exp[/(20 s ) E ] in order to obtain convergent inverse transform integrals for any
choice of and .
A matching procedure of the NLL resummed cross section to the NLO result has to be performed in order to keep the full information contained in the fixed-order calculation and to avoid
possible double-counting of the logarithmically enhanced contributions. A correct matching is
53
achieved through the formula

d2
d2 (F.O.)
=
(s ) +
2
2
dM dqT
dM 2 dqT2
dN N
2i
CN
2 (res)
d
b db
(N, b; s )
J0 (bqT )
2
dM 2 dqT2

(N, b; s ) ,
2
d2 (exp)
dM 2 dqT
(32)
where d2 (F.O.) is the fixed-order perturbative result, d2 (res) is the resummed cross section
discussed above, and d2 (exp) is the truncation of the resummed cross section to the same perturbative order as d2 (F.O.) . Here, we have removed the scale dependences for brevity.
At NLO, the double-differential partonic cross section
(F.O.)
d ab
dM 2 dqT2

s (R ) (1)
(0)
(z; s , R ) = qT2 (1 z) ab +
ab (z) + O s2
(33)
receives contributions from the emission of an extra gluon jet and from processes with an initial
gluon splitting into a q q pair,
TR
(0)
Aqg (s, t, u)q q () ,
2s
TR
(1)
(0)
g q (z) =
Aqg (s, u, t)q q () ,
2s
CF
q(1)
(z) =
(M)
Aqq (s, t, u)q(0)
q ()
q ()
2s
with [30]

s
t
2uM 2
Aqg (s, t, u) =
+ +
,
t
s
st
Aqq (s, t, u) = Aqg (u, t, s).
(1)
qg
(z) =
(34)
(35)
(36)
(37)
(38)
The Mandelstam variables s, t, and u refer to the 2 2 scattering process ab , Z 0 , W + X

and are related to the invariant mass M (or scaled squared invariant mass z = M 2 /s), transverse
momentum qT , and rapidity y of the slepton pair by the well-known relations
s = xa xb S = M 2 /z,

t = M 2 S M 2 + qT2 xb ey ,

u = M 2 S M 2 + qT2 xa ey .
(39)
(40)
(41)
Integration over qT requires the cancellation of soft and collinear singularities with virtual contributions in order to arrive at the finite single-differential partonic cross section
(F.O.)

d ab
s (R ) (1)
(0)
(z; s , R , F ) = ab (1 z) +
ab (z; R , F ) + O s2 ,
2
dM
(0)
(42)
where the first term ab is defined in Eqs. (7) and (8) and the second term including the full
NLO SUSY-QCD corrections can be found in Ref. [18].
54
The expansion of the resummed result reads

d2 (exp)
(N, b; s , R , F )
dM 2 dqT2

(exp)
fa/ ha (N + 1; F )fb/ hb (N + 1; F ) ab (N, b; s , R , F ),
=
(43)
a,b
(exp)
where ab
is obtained by perturbatively expanding the resummed component

(0)
s (R ) n (n)
(exp)
ab (N, b; s , R , F ) =
cc ca cb
abcc (N, ln ; R , F )
+
c
n=1

(n)
(44)
+ Habcc (N; R , F ) .
The perturbative coefficients (n) are polynomials of degree 2n in ln , and H(n) embodies the
constant part of the resummed cross section in the limits b and N . In particular, the
first-order coefficient (1) is given by
(1)
(1;2)
(1;1)
abcc (N, ln ) = abcc ln2 + abcc (N ) ln ,
(45)
(1;2)
abcc = 2A(1)
and
c ca cb
(1)

(1;1)
(1)
(1)
abcc (N ) = 2 Bc ca cb
+ ca c/b
c/a (N ) .
(N ) + cb
(46)
with
4. Numerical results
We now present numerical results for
the production of a right-handed selectron pair at the
LHC for a centre-of-mass energy of S = 14 TeV. For the masses and widths of the electroweak gauge bosons and the mass of the top quark, we use the values mZ = 91.1876 GeV,
mW = 80.403 GeV, Z = 2.4952 GeV,W = 2.141 GeV, and mt = 174.2 GeV [5]. The electromagnetic fine structure constant = 2GF m2W sin2 W / is calculated in the improved Born
approximation using the world average value GF = 1.16637 105 GeV2 for Fermis coupling
constant, and sin2 W = 1 m2W /m2Z .
We choose the mSUGRA benchmark point BFHK B [31], which gives after the renormalization group evolution of the SUSY-breaking parameters [32] a light eR of mass meR = 186 GeV
and rather heavy squarks with masses around 800850 GeV. The top-squark mass eigenstate t1 is
slightly lighter, but does not contribute to the virtual squark loops due to the negligible top-quark
density in the proton. For the LO (NLO and NLL) predictions, we use the LO 2001 [33] (NLO
2004 [34]) MRST sets of parton distribution functions. For the NLO and NLL predictions, s is
n =5
evaluated with the corresponding value of f = 255 MeV at two-loop accuracy. We allow
MS
the unphysical scales F and R to vary between M/2 and 2M to estimate the perturbative
uncertainty.
In Fig. 1, we present the transverse-momentum spectrum of the selectron pair, obtained after
production threshold up
integrating the equations of Sections 2 and 3 over M 2 , from the eR eR
to the hadronic centre-of-mass energy. We plot the fixed order result at order s (dashed line),
the expansion of the resummed formula at the same perturbative order (dotted line), the total
55
at the LHC. NLL + LO matched (full), fixed

Fig. 1. Transverse-momentum distribution for the process pp eR eR
order (dashed) and asymptotically expanded results (dotted) are shown, together with three different parameterizations
of non-perturbative effects (insert).
NLL + LO matched result (solid line), and the uncertainty bands from the scale variation. The
fixed order result diverges as expected as qT tends to zero. The asymptotic expansion of the
resummation formula is in good agreement with the O(s ) result in this kinematical region, since
the cross section is dominated by the large logarithms that we are resumming. For intermediate
values of qT , we can see that the agreement between the expansion and the perturbative result
is slightly worse. This effect was not present in qT -resummation for such qT -values [13] and is
thus related to the threshold-enhanced contributions important in the large-M region. This can
also be seen in Fig. 2, where we directly compare the jointly- and qT -matched results, the latter
having been obtained with the qT -resummation formalism of Ref. [25]. The two approaches
lead to a similar behaviour in the small-qT region, but the jointly-resummed cross section is
about 510% lower than the qT -resummed cross section for transverse momenta in the range
50 GeV < qT < 100 GeV. However, the effect of the resummation is clearly visible in both
cases, the resummation-improved result being even 40% higher than the fixed-order result at
qT = 80 GeV. In Fig. 1, we also estimate the theoretical uncertainties through an independent
variation of the factorization and renormalization scales between M/2 and 2M and show that
the use of resummation leads to a clear improvement with respect to the fixed-order calculation.
In the small- and intermediate-qT regions the scale variation amounts to 10% for the fixed-order
result, while it is always less than 5% for the matched result.
The qT -distribution is affected by non-perturbative effects in the small-qT region coming,
for instance, from partons with a non-zero intrinsic transverse-momentum inside the hadron and
from unresolved gluons with very small transverse momentum. Global fits of experimental Drell
Yan data allow for different parameterizations of these effects, which can be consistently included
56
Fig. 2. Transverse-momentum distribution of selectron pairs at the LHC in the framework of joint (full) and qT (dashed)
resummation.
NP . We include
in the resummation formula of Eq. (23) through a non-perturbative form factor Fab
in our analysis three different parameterizations of this factor [3537],

bmax M
NP(LY-G)
2
Fab
(47)
(b, M, x1 , x2 ) = exp b g 1 + g 2 ln
bg 1 g 3 ln(100x1 x2 ) ,
2

bmax M
NP(BLNY)
(b, M, x1 , x2 ) = exp b2 g 1 + g 2 ln
+ g 1 g 3 ln(100x1 x2 ) ,
Fab
(48)
2

M
NP(KN)
(b, M, x1 , x2 ) = exp b2 a1 + a2 ln
+ a3 ln(100x1 x2 ) .
Fab
(49)
3.2 GeV
The most recent values for the free parameters in these functions can be found in Refs. [36,37].
We show in the upper-right part of Fig. 1 the quantity
=
d (res.+NP) (R = F = M) d (res.) (R = F = M)
,
d (res.) (R = F = M)
(50)
which gives thus an estimate of the contributions from the different NP parameterizations (LY-G,
BLNY and KN). They are under good control, since they are always less than 5% for qT > 5 GeV
and thus considerably smaller than the resummation effects.
The invariant-mass distribution M 3 d/dM for eR -pair production at the LHC is obtained after integrating the equations of Sections 2 and 3 over qT2 and is shown in Fig. 3. The differential
cross section d/dM has been multiplied by a factor M 3 in order to remove the leading mass
dependence of propagator and phase space factors. We can see the P -wave behaviour relative
to the pair production of scalar particles, since the invariant-mass distribution rises above the
57
Fig. 3. Invariant-mass distribution M 3 d/dM of eR pairs at the LHC. We show the total NLL + NLO jointly matched
(full), as well as the fixed-order NLO SUSY-QCD (dashed) and LO (dotted) results, with the corresponding scale uncertainties (vertically hashed, horizontally hashed, and shaded bands).
threshold at s = 2meR with the third power of the slepton velocity and peaks at about 200 GeV
above threshold (both for M 3 d/dM and the not shown d/dM differential distribution), before falling off steeply due to the s-channel propagator and the decreasing parton luminosity.
In the large-M region, the resummed cross section is 30% higher than the leading order cross
section, but this represents only a 3% increase with respect to the NLO SUSY-QCD result. In the
small-M region, much further then from the hadronic threshold, resummation effects are rather
limited, inducing a modification of the NLO results smaller than 1%. The shaded, horizontally,
and vertically hashed bands in Fig. 3 represent the theoretical uncertainties for the LO, NLO
SUSY-QCD, and the jointly-matched predictions. At LO the dependence comes only from the
factorization scale and increases with the momentum-fraction x of the partons in the proton (i.e.
with M), being thus larger in the right part of the figure. This dependence is largely reduced at
NLO due to the factorization of initial-state singularities in the PDFs. Including the dependence
due to the renormalization scale in the coupling s (R ), the total variation is about 711%. After
resummation, the total scale uncertainty is finally reduced to only 78% for the matched result,
the reduction being of course more important in the large-M region, where the resummation
effects are more important.
In Fig. 4, we show the cross section correction factors
d i /dM
(51)
d LO /dM
as a function of the invariant-mass M. i labels the corrections induced by NLO QCD, NLO
SUSY-QCD, joint- and threshold-resummation (as obtained in [18]), these two last calculations
Ki =
58
Fig. 4. K-factors as defined in Eq. (51) for eR pair production at the LHC. We show the total NLL + NLO jointly (full),
and threshold (dashed) matched results, as well as the fixed-order NLO SUSY-QCD (dotted) and QCD (dash-dotted)
results.
being matched with the NLO SUSY-QCD result. At small invariant mass M, the resummation
is less important, since we are quite far from the hadronic threshold, as shown in the left part
of the plot. At larger M, the logarithms become important and lead to a larger increase of the
resummed K-factors over the fixed-order one. We also show the difference between threshold
and joint resummations, which is only about one or two percents. This small difference is due
to the choice of the Sudakov form factor G and of the H-function, which correctly reproduce
transverse-momentum resummation in the limit of b , N being fixed, but which present
some differences in the pure threshold limit b 0 and N , as it was the case for joint
resummation for Higgs and electroweak boson production [22,23]. However, this effect is under
good control, since it is much smaller than the theoretical scale uncertainty of about 7%.
5. Conclusions
With this work we complete our programme of performing precision calculations for slepton
pair production at hadron colliders. Together with the previous papers on transverse-momentum
[13] and threshold [18] resummation, soft-gluon resummation effects are now consistently included in predictions for various distributions exploiting the qT , threshold, and joint resummation
formalisms. We found that the effects obtained from resumming the enhanced soft contributions
are important at hadron colliders, even far from the critical kinematical regions where the resummation procedure is fully justified. The numerical results show a considerable reduction of
the scale uncertainty with respect to fixed order results and also a negligible dependence on
non-perturbative effects, introduced through different Gaussian-like smearings of the Sudakov
59
form factors. These features lead to an increased stability of the perturbative results and thus to a
possible improvement of the slepton pair (sleptonsneutrino) search strategies at the LHC.
Acknowledgements
This work was supported by a PhD fellowship of the French Ministry for Education and
Research.
Appendix A. Sfermion mixing
The soft SUSY-breaking terms Af of the trilinear Higgssfermionsfermion interaction and
the off-diagonal Higgs mass parameter in the MSSM Lagrangian induce mixings of the leftand right-handed sfermion eigenstates fL,R of the electroweak interaction into mass eigenstates f1,2 . The sfermion mass matrix is given by [2]

M =
2
m2LL + m2f
mf mLR
mf mLR
m2RR + m2f
(A.1)
with

m2LL = m2F + Tf3 ef sin2 W m2Z cos 2,
(A.2)
m2RR = m2F + ef sin2 W m2Z cos 2,

cot for up-type sfermions,
mLR = Af
tan for down-type sfermions.
(A.3)
(A.4)
It is diagonalized by a unitary matrix S f , S f M2 S f = diag(m21 , m22 ), and has the squared mass
eigenvalues
m21,2 = m2f +
1
2
m + m2RR
2 LL
m2LL m2RR

+ 4m2f |mLR |2 .
(A.5)
For real values of mLR , the sfermion mixing angle f , 0 f /2, in

f
S =
cos f
sin f
sin f
cos f

f1
f cfL
with
=S
,
f2
fR
(A.6)
can be obtained from

tan 2f =
2mf mLR
m2LL m2RR
(A.7)
If mLR is complex, one may first choose a suitable phase rotation fR = ei fR to make the
mass matrix real and then diagonalize it for fL and fR . tan = vu /vd is the (real) ratio of the
vacuum expectation values of the two Higgs fields, which couple to the up-type and down-type
(s)fermions. The soft SUSY-breaking mass terms for left- and right-handed sfermions are mF
and mF , respectively.
60
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
H.P. Nilles, Phys. Rep. 110 (1984) 1.

H.E. Haber, G.L. Kane, Phys. Rep. 117 (1985) 75.
B.C. Allanach, et al., Eur. Phys. J. C 25 (2002) 113.
J.A. Aguilar-Saavedra, et al., Eur. Phys. J. C 46 (2006) 43.
W.M. Yao, et al., Particle Data Group, J. Phys. G 33 (2006) 1.
S. Dawson, E. Eichten, C. Quigg, Phys. Rev. D 31 (1985) 1581.
P. Chiappetta, J. Soffer, P. Taxil, Phys. Lett. B 162 (1985) 192.
F. del Aguila, L. Ametller, Phys. Lett. B 261 (1991) 326.
H. Baer, C.H. Chen, F. Paige, X. Tata, Phys. Rev. D 49 (1994) 3283.
G. Bozzi, B. Fuks, M. Klasen, Phys. Lett. B 609 (2005) 339.
H. Baer, B.W. Harris, M.H. Reno, Phys. Rev. D 57 (1998) 5871.
W. Beenakker, M. Klasen, M. Krmer, T. Plehn, M. Spira, P.M. Zerwas, Phys. Rev. Lett. 83 (1999) 3780.
G. Bozzi, B. Fuks, M. Klasen, Phys. Rev. D 74 (2006) 015001.
C.G. Lester, D.J. Summers, Phys. Lett. B 463 (1999) 99.
A.J. Barr, JHEP 0602 (2006) 042.
E. Lytken, Czech. J. Phys. 54 (2004) A169.
Yu.M. Andreev, S.I. Bityukov, N.V. Krasnikov, Phys. At. Nucl. 68 (2005) 340.
G. Bozzi, B. Fuks, M. Klasen, Nucl. Phys. B 777 (2007) 157.
H.N. Li, Phys. Lett. B 454 (1999) 328.
E. Laenen, G. Sterman, W. Vogelsang, Phys. Rev. D 63 (2001) 114018.
E. Laenen, G. Sterman, W. Vogelsang, Phys. Rev. Lett. 84 (2000) 4296.
A. Kulesza, G. Sterman, W. Vogelsang, Phys. Rev. D 66 (2002) 014011.
A. Kulesza, G. Sterman, W. Vogelsang, Phys. Rev. D 69 (2004) 014012.
A. Banfi, E. Laenen, Phys. Rev. D 71 (2005) 034003.
G. Bozzi, S. Catani, D. de Florian, M. Grazzini, Nucl. Phys. B 737 (2006) 73.
G.P. Korchemsky, Mod. Phys. Lett. A 4 (1989) 1257.
S. Catani, D. de Florian, M. Grazzini, Nucl. Phys. B 596 (2001) 299.
S. Catani, M.L. Mangano, P. Nason, L. Trentadue, Nucl. Phys. B 478 (1996) 273.
H. Contopanagos, G. Sterman, Nucl. Phys. B 419 (1994) 77.
R.J. Gonsalves, J. Pawlowski, C.F. Wai, Phys. Rev. D 40 (1989) 2245.
G. Bozzi, B. Fuks, B. Herrmann, M. Klasen, Nucl. Phys. B 787 (2007) 1.
W. Porod, Comput. Phys. Commun. 153 (2003) 275.
A.D. Martin, R.G. Roberts, W.J. Stirling, R.S. Thorne, Phys. Lett. B 531 (2002) 216.
A.D. Martin, R.G. Roberts, W.J. Stirling, R.S. Thorne, Phys. Lett. B 604 (2004) 61.
G.A. Ladinsky, C.P. Yuan, Phys. Rev. D 50 (1994) 4239.
F. Landry, R. Brock, P.M. Nadolsky, C.P. Yuan, Phys. Rev. D 67 (2003) 073016.
A.V. Konychev, P.M. Nadolsky, Phys. Lett. B 633 (2006) 710.

A dispersive approach to Sudakov resummation

Einan Gardi a,b,c, , Georges Grunberg d
a Cavendish Laboratory, University of Cambridge, J.J. Thomson Avenue, Cambridge, CB3 0HE, UK
b Department of Applied Mathematics & Theoretical Physics, Wilberforce road, Cambridge CB3 0WA, UK
c School of Physics, The University of Edinburgh, Edinburgh EH9 3JZ, Scotland, UK 1
d Centre de Physique Thorique, cole Polytechnique, CNRS, 91128 Palaiseau cedex, France

Abstract
We present a general all-order formulation of Sudakov resummation in QCD in terms of dispersion integrals. We show that the Sudakov exponent can be written as a dispersion integral over spectral density
functions, weighted by characteristic functions that encode information on power corrections. The characteristic functions are defined and computed analytically in the large-0 limit. The spectral density functions
encapsulate the non-Abelian nature of the interaction. They are defined by the time-like discontinuity of specific effective charges (couplings) that are directly related to the familiar Sudakov anomalous dimensions
and can be computed order-by-order in perturbation theory. The dispersive approach provides a realization
of dressed gluon exponentiation, where Sudakov resummation is enhanced by an internal resummation of
running-coupling corrections. We establish all-order relations between the scheme-invariant Borel formulation and the dispersive one, and address the difference in the treatment of power corrections. We find that
in the context of Sudakov resummation the infrared-finite-coupling hypothesis is of special interest because
the relevant coupling can be uniquely identified to any order, and may have an infrared fixed point already at
the perturbative level. We prove that this infrared limit is universal: it is determined by the cusp anomalous
dimension. To illustrate the formalism we discuss a few examples including B-meson decay spectra, deep
inelastic structure functions and DrellYan or Higgs production.
PACS: 12.38.Cy
Keywords: QCD; Resummation; Sudakov logarithms; Renormalons; Power corrections
* Corresponding author at: School of Physics, The University of Edinburgh, Edinburgh EH9 3JZ, Scotland, UK.
E-mail address: einan.gardi@cern.ch (E. Gardi).

1 Address after October 1st 2007.
doi:10.1016/j.nuclphysb.2007.10.022
62
E. Gardi, G. Grunberg / Nuclear Physics B 794 (2008) 61137
1. Introduction
An accurate theoretical description of inclusive differential cross sections and decay spectra
is essential for many aspects of collider and flavor physics; it is therefore one of the primary
goals of QCD perturbation theory. The main challenge in describing differential distributions is
associated with kinematic regions where there is a large hierarchy of scales, as occurs for example
in the production of a jet with a large energy and a small mass. Near the exclusive limit of phase
space, the so-called Sudakov region, only soft and collinear radiation is kinematically allowed,
and because of the dynamical enhancement of such radiation, multiple gluon emission becomes
important. In this way the hierarchy of scales translates into large logarithms that appear with
increasingly high powers in perturbation theory and spoil the convergence of the expansion. Even
a qualitative description of the distribution in this region require all-order resummation. Classical
examples where the Sudakov region has been thoroughly studied include deep inelastic structure
functions for x 1 [15], DrellYan and Higgs production near the partonic threshold [1,2,
613] or at small transverse momentum (see, e.g., [1418]), event-shape distributions [1929],
heavy-quark fragmentation [3034], and inclusive B decays into a light quarks [32,3544], e.g.,
B Xs or B Xu l .
From a theoretical perspective the Sudakov limit is very interesting. Despite the fact that one is
dealing with a complex multi-scale problem in a non-Abelian gauge theory, one can in fact control the dominant perturbative corrections to all orders and resum the series. The simplification of
the expansion in the Sudakov limit is a direct reflection of the factorization property of infrared
(soft and collinear) singularities. In infrared safe quantities the singularity associated with soft
and collinear radiation (integrated over phase space) cancel exactly, order-by-order in s , with
infrared singularities in virtual corrections. The structure of these singularities and their cancellation is encoded in the resummation formalism: the large logarithms are nothing but the finite
remainder in the sum of the real and virtual contributions, which are separately divergent.
Sudakov resummation is best formulated in moment (Mellin) space where the multi-gluon
phase space factorizes. In moment space the combined effect of soft and collinear radiation and
the corresponding virtual corrections appears an exponential Sudakov factor. The Sudakov factor
sums up to all orders the dominant radiative corrections that are enhanced by powers of ln N
(where N is the moment index), while neglecting corrections that are suppressed by powers of N
for N .
While Sudakov resummation is designed to deal with logarithmic singularities alone, infrared
sensitivity appears also through power-suppressed effects, O((/m)p ) [22,4553], where m is
the hard scale and where p is an integer. In the Sudakov region, power corrections are particularly
important: whereas perturbative corrections are enhanced at N by powers of the logarithm
ln N , power corrections are enhanced by powers of N , taking the form O((N /m)p ). This
parametric enhancement of non-perturbative corrections implies that for any fixed coupling (or
a fixed hard scale) the N limit itself is strictly beyond the reach of resummed perturbation
theory. Put differently, upon ignoring these power corrections one can only expect the large-N
resummation to be a valid approximation in a restricted range: N m/.
Quite remarkably, perturbation theory itself is sensitive [47] to the presence of these powersuppressed corrections [54]. This sensitivity appears through infrared renormalons [55,56], factorially increasing coefficients that emerge out of the integral over the running coupling. Despite
the fact that the Sudakov exponent can be uniquely computed to any logarithmic accuracy, the series as a whole diverges. The understanding that renormalons are an inherent part of the Sudakov
exponent [47,52,53] has prompted the development of a new resummation formalism, dressed
63
gluon exponentiation [22,23,57]; for a recent review, see [54]. DGE combines Sudakov resummation with an additional internal resummation of running-coupling corrections.2 The internal
resummation is based on an all-order calculation of the exponentiation kernel in the large-0
limit, which exposes the renormalons and thus opens the way for including non-perturbative
power corrections.
DGE has been applied and successfully compared with data in a variety of inclusive distributions: event-shape distributions [22,23], deep inelastic structure functions [5], heavy quark
fragmentation [30] and inclusive decay spectra [4044]. It has proven to provide a good description of these distributions over the entire Sudakov region, significantly extending the range of
applicability of resummed perturbation theory. In its first application, to event-shape distribution, DGE has opened the way for a quantitative description of the two-jet region, which had
been inaccessible to analytic methods before. In heavy-quark fragmentation and inclusive decay
spectra DGE provides a viable alternative to the conventional approach where the corresponding distributions (heavy-quark fragmentation function and shape function, respectively) have
been parametrized by some ad hoc functional forms. Further details can be found in recent review
talks [44,54] and in the original publications.
Technically these results have been obtained by trading [22] the integration over the running coupling in the Sudakov exponent for a scheme-invariant [64] Borel integration. The Borel
transform has been computed analytically in the large-0 limit, allowing to identify the renormalon singularities. The perturbative sum is then defined by the principal value prescription and
power corrections are parametrized based on the renormalon ambiguities. While the original calculations of event-shape distributions in Refs. [22,23] have been performed at next-to-leading
logarithmic (NLL) accuracy3 it was observed that the Borel formulation is in fact completely
general, and can accommodate subleading corrections with any logarithmic accuracy, provided
that the relevant Sudakov anomalous dimensions are known. This has been consequently used
in other applications [5,32,43,54] to perform calculations with formal NNLL accuracy while
preserving the pattern of renormalon singularities found in the large-0 limit.
In this paper we establish an alternative formulation of Sudakov resummation that captures
the same class of radiative corrections as well as the renormalon structure, using the dispersive
approach.4 Following [68,69] we show that the Sudakov exponent can be written, to any logarithmic accuracy, as a dispersive integral of the time-like discontinuity of specific non-Abelian
effective charges, integrated with respect to characteristic functions that are defined (and computed analytically) in the large-0 limit. This way the exponent in the non-Abelian theory is
written in full analogy with the way renormalon resummation has been formulated in Refs. [49,
61,62] at the level of a single dressed gluon (the large-0 limit) with one important difference:
the effective coupling is computed order-by-order in the non-Abelian theory.
2 Internal refers here to the fact that the argument of the relevant logarithms involves loop-momenta that are inte-
grated over, rather than external scales. Resummation of running-coupling corrections has been extensively studied in the
past, primarily in the context of single-scale quantities [45,49,55,56,5862]. See in particular the discussion in Section 7
of [58] and in [45]. For a related early discussion of Sudakov resummation for the electron form factor in QED see
also [63].
3 A proper definition of the QCD scale corresponding to the gluon bremsstrahlung coupling [12] (see Eq. (104)
in [22]) is sufficient to account for the entire non-Abelian correction, O(CA /0 ), at this order; is fixed by the NLO
correction to the cusp anomalous dimension [65,66].
4 A dispersive representation of the Sudakov exponent was first considered in the original DGE paper [22], see Eq. (43)
there. In that paper, however, the discussion was limited to NLL accuracy.
64
Refs. [68,69] emphasized the fact that Sudakov resummation leaves some freedom (see
also [70]) that does not allow to make conclusive statements about power corrections. Indeed, in
general there is a conflict between the heuristic argument for power corrections in expressions involving integration over the running coupling (e.g., [47]) and renormalon-based arguments [52].
It was observed that, at the perturbative level, the resummation of Sudakov logarithms can be
implemented in infinitely many ways, each of which consisting of an integral mapping of a perturbative object (the running Sudakov coupling) onto a fixed target: the Sudakov exponent.
To each mapping corresponds a different kernel. In Ref. [68] it was proposed to fix this ambiguity and select the unique resummation scheme by requiring that in the large-0 limit, the
running Sudakov coupling coincides with a dressed gluon propagator. The specified scheme
so determined is the dispersive approach, where the above conflict does not arise. In the present
paper we further develop this approach, establishing the connection with the physical Sudakov
anomalous dimensions.
We show that this dispersive formulation is in close correspondence with the scheme-invariant
Borel formulation and establish the relations between the two. We further show that the effective
charges appearing in the dispersive formulation are uniquely defined to any order in perturbation
theory and stand in one-to-one correspondence with the relevant Sudakov anomalous dimensions.
These effective charges provide a (process-dependent) generalization of the concept of the gluon
bremsstrahlung coupling, which was defined in [12] at the next-to-leading order. Moreover, we
find that the dispersive formulation is closely related to the joint resummation formalism of
Ref. [17], where the effective charges are given a direct diagrammatic interpretation in terms of
webs [71].
The dispersive approach proposed here is particularly suited to the implementation of the infrared finite coupling hypothesis of Ref. [49]. This possibility is of special interest in the context
of Sudakov resummation, since the relevant non-Abelian coupling can be identified to any order, and furthermore, may have an infrared fixed-point already at the perturbative level. In this
case the coupling may have a causal analyticity structure [7274], free of any Landau singularities.
We proceed as follows: in Section 2 we recall some general features of the Sudakov limit
and define the basic elements used in Sudakov resummation. Then, in Section 3 we analyze
the structure of the exponentiation kernel in momentum space. We first use general considerations of infrared factorization and renormalization-group invariance to identify the physical
Sudakov anomalous dimensions. We then employ all-order resummation in the large-0 limit arriving at a dispersive representation of the kernel. Next, in Section 4 we generalize the dispersive
representation to the full non-Abelian theory. We show that this leads to a unique (yet processdependent) generalization of the concept of a gluon bremsstrahlung coupling, the Sudakov
effective charges, which are computed order-by-order in perturbation theory. We then analyze
the evolution of these effective charges, finding that they may have a perturbative infrared fixed
point. We further prove that the corresponding fixed-point value is process-independent, and
compute its BanksZaks expansion. Then in Section 5 we use the tools of the previous sections
to derive a general dispersive representation of the Sudakov exponent. We also present explicit
results for the characteristic functions in several different processes and analyze their properties.
In Section 6 we explain the relation between the dispersive formulation and the scheme invariant
Borel formulation. Next, in Section 7 we address power corrections. Finally, in Section 8 we
summarize our conclusions.
65
2. General set-up: Sudakov resummation for inclusive distributions

Consider an inclusive infrared-safe distribution, d (m2 , r)/dr where the r 0 limit is characterized by Sudakov logarithms,

d (m2 , r)
= (r) 1 + O(s )
dr

s
ln(r) b1 d1
+ regular terms +
+
+ CR
r
r
+
(2.1)
whose maximal power grows with order as s n ln2n1 (r)/r due to multiple soft and collinear
radiation. The notation bi and di used for the coefficients is associated with their phase-space
origin; this will be elucidated in the next section. The integration prescription [ ]+ is defined by5
1
0

1

1
1
dr F (r) 1+u
= dr F (r) F (0) 1+u ,
r
r
+
(2.2)
where F (r) is a smooth test function. This prescription accounts for the divergent virtual corrections, which cancel against the singularity generated when integrating the real-emission contributions near r = 0.
In order to describe the distribution at small r we need to resum all the singular terms in (2.1)
to any order. To this end it is convenient to work in moment space, where the multi-gluon phase
space factorizes. The moments are given by:
N m
1
=
dr (1 r)N1
d (m2 , r)
,
dr
(2.3)
where the plus prescription in (2.1) guarantees that the moments are finite, cf. (2.7) below. Once
resummation has been performed analytically in moment space one recovers the distribution in
momentum space by an inverse Mellin transformation:

i

d (m2 , r)
dN
=
(1 r)N N m2 resummed ,

dr
2i
resummed
(2.4)
where the integration contour runs parallel to the imaginary axis in the complex N plane, to the
right of the singularities of the integrand.
We assume that the large-N limit singles out infrared singularities associated with two physical scales, which we call the jet scalemomenta of order O(m2 r)and the soft scalemomenta
of order O(m2 r 2 ). We therefore have a double hierarchy of scales:
m2 m2 r m2 r 2
m2 m2 /N m2 /N 2 .
(2.5)
While we are interested in large N , in order to apply perturbation theory we assume that all three
scales are within the perturbative regime, namely m2 /N 2 2 .
5 Differentiation with respect to u can be used to generate powers of ln(r). See Eq. (6.4) in [75] for a more general
definition.
66
This general scenario, with minor variations, is encountered in many different applications,
for example:
Inclusive B decays into light quarks [32,3544], e.g., B Xs or6 B Xu l . These decays are characterized by jet-like kinematics having a large hierarchy between the energy of
the hadronic system X and its mass, or equivalently, between the two lightcone components
of the jet, p E |p|,
namely p p + . Thus, in (2.5) the hard scale is the b-quark

mass or the large lightcone component of the jet (p ), while r is the ratio between the small
and large components r = p + /p . The intermediate scale is the jet mass p + p m2 r.
The soft scale m2 r 2 is associated with soft gluons that couple of the b quark, putting it
slightly off its mass shell prior to the decay. In perturbation theory the soft subprocess is
therefore associated with the momentum distribution of the b quark inside an initial on-shell
b quark, see [32] and references therein. Beyond perturbation theory this function is replaced
by the momentum distribution of the b quark in a meson, which differs from its perturbative
counterpart by power corrections.
Inclusive B production in e+ e annihilation, see, e.g., [3034]. Here r = 1 x where x =
2EB /Q, the energy fraction of the detected B meson and Q is the center-of-mass energy. At
a difference with (2.5) the three scales are Q2 Q2 (1 x) m2b (1 x)2 , where mb is the
b-quark mass. The soft scale m2b r 2 is associated with the heavy quark fragmentation process
(which, at the perturbative level, is similar to b decay [32]).
DrellYan or Higgs production in hadronic collisions in the DIS scheme [1,2,612], where
m = Q is the mass of the produced pair (or mH , in case of Higgs production), r = 1 Q2 /s
is the fraction of energy carried by soft gluon radiation into the final state, and s is the
partonic center-of-mass energy. Here the soft scale is genuinely associated with the Drell
Yan process while the jet mass scale enters via the quark distribution function in the DIS
factorization scheme (the x 1 limit of deep inelastic structure functions) and it can be
traded for 2F upon using the MS factorization scheme.
Event-shape distributions in e+ e annihilation, near the two-jet limit [1925,29]. Here r is
identified with a shape variable that vanishes in the two-jet limit, for example 1 T where
T is the thrust variable or C/6 where C is the C parameter; the hard scale m is the centerof-mass energy in the collision. Because jet observables are less inclusive [22,23,29,76], the
formalism we develop applies only to a limited logarithmic accuracy. Non-inclusive effects
can be accommodated in a class of event-shape variables (where Eqs. (2.7) and (2.8) below
are valid [19]) but this goes beyond the scope of the present work.
As we shall see the same resummation formalism applies in all these examples. Moreover, the
coefficients of the logarithms have a certain degree of universality: first, the leading logarithms
are always related to the cusp anomalous dimension [65,66], A(s ) in (2.8) below; second, the
physics on the jet-mass scale m2 r is the same in all these examples: it is associated with the production of a quark jet with a constrained invariant mass of O(m2 r). The corresponding Sudakov
logarithms are the same as in deep inelastic structure functions in the x 1 limit; they are generated, in all these examples, by the anomalous dimension function B(s ) in (2.8). In contrast,
6 Note that in case of the semileptonic decay b ul , resummation is applied to the triple differential width [42]. For
simplicity of the notation we do not write here explicitly derivatives with respect to other kinematic variables.
67
the physics on the soft scale m2 r 2 is rather different; this will be reflected in the resummation
formalism through a process-dependent Sudakov anomalous dimension D(s ).
Owing to the factorization property of soft and collinear radiation the moments (2.3) can be
written as follows:

N m2 = H m2 Sud m2 , N +
N m2 ,
(2.6)
where the Sudakov factor Sud(m2 , N) sums up to all orders in perturbation theory all the
terms that diverge for N , H (m2 ) is a hard coefficient function that includes constant
(N -independent) terms at each order in s ; such corrections arise from virtual diagrams, proportional to (r). Finally,
N (m2 ) = O(1/N) includes residual real-emission corrections that
fall at large N as 1/N (up to logarithms). In this paper we are interested in the Sudakov factor
Sud(m2 , N). Both H (m2 ) and
N (m2 ) can be computed order-by-order in s by matching
the resummation formula with the fixed-order expansion. We shall not discuss them further here.
The general formula [1,2] (see also [3,19,31,32,38,4143]) for the Sudakov factor is:
1

2
dr
(1 r)N1 1 R m2 , r ,
Sud m , N = exp CR
(2.7)
r
0
where the subtraction of 1 accounts for the virtual corrections encapsulated in the plus prescription in (2.1), and the momentum-space kernel, which is fully defined by the real-emission
contributions that are singular for r 0, takes the general form:
R(m2 , r) 1
=
CR
r
r
rm2

2
2 2
dk 2 2
,
A s k + B s rm D s r m
k2
(2.8)
r 2 m2
where A, B and D are Sudakov anomalous dimensions that can be computed order-by-order
in s . A(s ) is the universal cusp anomalous dimension [65,66], whose expansion, in the MS
scheme is known to three-loop order [77]:
A(s ) =

CR
a + a2 a 2 + a3 a 3 + ,
0
(2.9)
where a 0 sMS / . The other Sudakov anomalous dimensions appearing in Eq. (2.8) are
B(s ) =

CR
b1 a + b2 a 2 + b3 a 3 + ,
0
(2.10)
which is associated with the jet mass scale m2 r, and

D(s ) =

CR
d1 a + d2 a 2 + d3 a 3 + ,
0
(2.11)
which is associated with the soft scale m2 r 2 . The known expansion coefficients of these anomalous dimensions are summarized in Appendix A. As already mentioned, D is process dependent.
In Appendix A we quote the coefficients in two examples, one of DrellYan (or Higgs) production and the second of inclusive B decay (or heavy-quark fragmentationthe anomalous
dimension is the same [32]). These two examples will be used throughout this paper.
Next, note that CR , the overall coefficients of all the anomalous dimensions entering (2.8)
in a given process, depends on the color charge of the hard parton(s) that radiates. In inclusive
68
B decay or heavy-quark fragmentation CR = CF , while for DrellYan production CR = 2CF

and for Higgs production by gluongluon fusion CR = 2CA .
Finally, it is useful to introduce some alternative definitions for the Sudakov factor using the
Laplace weight instead of the Mellin one. The first observation is that at large N one can make
the following approximation of the real-emission weight factor in (2.7)
(1 r)N1 eN r ,
which only modifies the exponent by O(1/N) terms. Next one notes that as far as the realemission part is concerned the integral over r can be extended to r = , well beyond the physical
phase space r = 1: the resulting modification of the Sudakov exponent is exponentially small,
O(eN ). This, however, does not apply to the virtual terms, where any change of the upper limit
of integration is reflected in a modification of the constant O(N 0 ) term for N . Let us
therefore define:

2
2
N r
dr
m , N = exp CR
Sud
(2.12)
(r < 1) R m , r ,
e
r
0
which satisfies

m2 , N = ln Sud m2 , N + O(1/N).
ln Sud
(2.13)
The O(1/N) contributions by which Eq. (2.12) differs from Eq. (2.7), can be compensated in the
expression for the moments by a different remainder function. The physical moments N (m2 )
can be obtained by matching with the fixed-order result:

m2 , N +
N m2 ;
N m2 = H m2 Sud
(2.14)
N (m2 ) can be determined to fixed
where H (m2 ) is the same hard function as in (2.6), while
2
order from the perturbative expansion of N (m ).
It is useful to further define a Sudakov factor where the subtraction term too is integrated to
r = :

2

dr N r
(2.15)
Sud m , N = exp CR
er R m2 , r .
e
r
0
The suppression of the weight factor eN r er guarantees convergence for r . Again, the
manipulations used in obtaining (2.15) from (2.7) do not change the N divergent terms,
which all emerge from the r 0 limit. Here however, in contrast with (2.12), the constant terms
O(N 0 ) for N are modified.
Note that the moment-space exponent of Eq. (2.7) is composed of harmonic sums (powers
of the (N ) function, which differ from ln(N ) by constants and inverse powers of N ) whereas
the Laplace version, Eq. (2.15), involves strictly powers of ln(N ), with no additional constants
nor O(1/N) terms. Eq. (2.12) contains powers of ln(N ) and constant terms, with no additional
O(1/N) terms [68,78]. Both (2.15) and (2.7), but not (2.12), are defined such that for N = 1 the
exponent exactly vanishes so the Sudakov factor becomes 1.
Both the constant term, O(N 0 ), and the O(1/N) contributions by which Eq. (2.15) differs
from Eq. (2.7), can be compensated in the expression for the moments by a different multiplicative hard function and a different remainder function:

N m2 = H m2 Sud m2 , N +
N m2 ;
(2.16)
69
similarly to H (m2 ) and

N (m2 ) in (2.6), H (m2 ) and
N (m2 ) in (2.16) can be determined
to fixed order from the perturbative expansion of N (m2 ). Finally, Eq. (2.16), similarly to (2.14)
and (2.6), can be inserted into the inverse Mellin transformation (2.4) to generate the resummed
distribution in momentum space.
3. The exponentiation kernel
The key to constructing an effective resummation technique is the understanding of the allorder structure of the kernel. The purpose of this section is to analyze the momentum-space
kernel (2.8), first using the general considerations of infrared factorization and renormalizationgroup invariance, and then by considering the all-order calculation of the kernel in the large-0
limit.
3.1. Infrared factorization and renormalization-group invariance
Let us begin by noting that Sudakov logarithms in infrared and collinear safe distributions
arise from different regions of phase space, involving different physical scales. As follows
for (2.1), in the perturbative expansion of the kernel,

R(m2 , r)
ln(r) b1 d1 s (m)
= CR
+
+
CR
(3.1)
r
r
r
the two are mixed together. In contrast, upon taking one derivative (cf. Eq. (2.8))

dR(m2 , r)
= J rm2 S r 2 m2 ,
d ln m2
(3.2)
the dependence upon the jet (rm2 ) and soft (r 2 m2 ) scales is nicely disentangled into two
different functions corresponding to the two subprocess, J and S respectively. These are the
physical jet and soft Sudakov anomalous dimensions. They are related to the ones appearing
in (2.8) by the following relations:
dB(s (2 ))

,
CR J 2 = A s 2 +
d ln 2
dD(s (2 ))

.
CR S 2 = A s 2 +
d ln 2
(3.3)
We emphasize that J (2 ) and S(2 ) are renormalization-group invariant functions, while their
conventional decomposition into the cusp A, B and D (which are renormalization-schemedependent quantities) is specific to the MS scheme.
3.2. The physical interpretation of the Sudakov anomalous dimensions
The functions J (2 ) and S(2 ) have a clear physical meaning: they govern the scaledependence of certain physical quantities (the physical evolution kernels or physical anomalous dimensions) in the Sudakov limit. Let us illustrate this statement using a couple of
examples.
70
3.2.1. Deep inelastic structure functions

The scale-dependence of the (flavor non-singlet) deep inelastic structure function F2 can be
expressed in terms of F2 itself, yielding the following evolution equation (see, e.g., Refs. [63,79,
80]):
1
dF2 (x, Q2 )
=
d ln Q2
(x, Q2 )
KDIS(F2 )
moments by
F2 N, Q

dz
KDIS(F2 ) x/z, Q2 F2 z, Q2 .
z
(3.4)
is the physical evolution kernel; it is renormalization-group invariant. Defining
1
=

dx x N1 F2 x, Q2 /x,
(3.5)
Eq. (3.4) implies that the moment-space physical evolution kernel, the physical anomalous dimension, is:
K DIS(F2 ) N, Q
1

d ln F2 (N, Q2 )
dx x N1 KDIS(F2 ) x, Q2 /x =
.
d ln Q2
(3.6)
Let us now consider the x 1 limit. In this limit the evolution of the structure function takes
a simple from [4,5] (see Eq. (13) in [54]) and Ref. [70]:

d ln F2 (N, Q2 )
x N1 1
=
C
dx
J (1 x)Q2
F
2
1x
d ln Q
1
d ln H (Q; F )
+ O(1/N),
+
(3.7)
d ln Q2
where the first term, which includes the N divergent corrections to all orders, is controlled
by the Sudakov anomalous dimension J (2 ) of Eq. (3.3). The constant term can be written [70]
in terms of the quark electromagnetic form factor F (Q2 ) [8184]:
Q
d ln H (Q; F ) d ln(F (Q2 ))2
d2 2
=
+
C
J
F
d ln Q2
d ln Q2
2
0

= G 1, s Q2 , = 0 + B s Q2 ,
2
(3.8)
where each of the two terms in the first line is separately infrared divergent, but the divergence
cancels in the sum [70]; in the second line the result is expressed in terms of B of Eq. (2.10)
above and G(Q2 /2 , s (2 ), ), which is the finite part of d ln(F (Q2 ))2 /d ln Q2 as defined in
Ref. [84] using dimensional regularization.
Comparing (3.6) and (3.7) we therefore find the following relation in momentum space:

J ((1 x)Q2 ) d ln(F (Q2 ))2
KDIS(F2 ) x, Q2 = CF
(1 x) + O (1 x)0
+
1x
d ln Q2

J ((1 x)Q2 )
= CF
1x
+

+
d ln(F (Q2 ))2

+ CF
d ln Q2

Q

d2 2
J
(1 x)
2

infrared finite
+ O (1 x) ,
0
71
(3.9)
where the plus prescription is defined in (2.2) and the O((1 x)0 ) term neglected here is integrable for x 1 (although divergent: it contains powers of ln(1 x)). We thus find that
CF J ((1 x)Q2 )/(1 x) is the leading term in the expansion of the physical evolution kernel KDIS(F2 ) (x, Q2 ) in the x 1 limit with fixed jet mass (1 x)Q2 . The next term in this
expansion, proportional to (1 x), is comprised of purely virtual corrections associated with
the quark form factor. This term is infrared divergent, but as indicated in the second line in (3.9),
the singularity cancels exactly upon integrating over x with the divergence of the integral of
CF J ((1 x)Q2 )/(1 x) near x 1.
3.2.2. DrellYan cross section
The cross section of the DrellYan process, ha + hb e+ e + X, is:
2 dx dx

d
4em
j
i
=
fi/ ha (xi , F )fj/ hb (xj , F )gij , Q2 , 2F ,
2
2
xi xj
dQ
9Q s
1
(3.10)
i,j 0
where s is the hadronic center-of-mass energy, s = xi xj s is the partonic one, Q2 is the squared
mass of the lepton pair and = Q2 /s . The partonic threshold 1 is characterized by Sudakov
logarithms.
Let us define the Mellin transform of the quarkantiquark partonic cross section in (3.10),
e.g., in electromagnetic annihilation gq q (, Q2 , 2F ) = eq2 [(1 ) + O(s )], by
Gq q N, Q
, 2F
1

d N1 gq q , Q2 , 2F .
(3.11)
Considering the Sudakov limit one has

Gq q N, Q2 , 2F HDY Q2 , 2F SudDY N, Q2 , 2F + O(1/N)
(3.12)
with
1
SudDY = exp
0
N1 1
d
1
2 2
(1
) Q

dk 2
A s k 2
+ DDY s (1 )2 Q2
2
k
2F

.
(3.13)
Upon taking the logarithmic derivative we identify [68] the physical anomalous dimension SDY
of Eq. (3.3):

d ln Gq q (N, Q2 , 2F )
K DY N, Q2
2
d ln Q
1
0

d N1 KDY , Q2
72
1
2CF
d
0

N1 1
SDY Q2 (1 )2
1
d ln HDY (Q2 , 2F )
+
+ O(1/N),
d ln Q2
(3.14)
where KDY (, Q2 ) is the physical DrellYan evolution kernel defined for arbitrary ; in the
second line we considered the large-N limit where Eqs. (3.12) and (3.13) apply and then used
Eq. (3.3).
The constant term can be expressed in terms of the analytically-continued electromagnetic
quark form factor [70] (see also [6,8,9]):
d ln HDY (Q2 , 2F ) d ln |(F(Q2 ))|2
=
+ CF
d ln Q2
d ln Q2
Q

d2
SDY 2 ,
2
(3.15)
where the infrared singularities cancel in the sum, as in (3.8). Thus, in momentum space we have:

SDY (Q2 (1 )2 ) d ln |(F (Q2 ))|2
(1 ) + O (1 )0
+
KDY , Q2 = 2CF
2
1
d ln Q

SDY (Q2 (1 )2 )
= 2CF
1
+

+
d ln |(F (Q2 ))|2

+ CF
d ln Q2

Q
0

2

d2
SDY
(1 ) + O (1 )0 .
2
infrared finite
(3.16)
We see that the Sudakov anomalous dimension SDY controls the leading term in the expansion of
the physical DrellYan kernel (3.14) near threshold. The 1 limit is taken such that Q(1 ),
corresponding to the total energy carried by soft gluons to the final state, is kept fixed. The
next term in this expansion, proportional to (1 ), is determined by the quark form factor,
analytically-continued to the time-like axis. This purely virtual term is infrared singular, but
upon performing an integral over this singularity cancels with the one generated by integrating
the real-emission term 2CF SDY (Q2 (1 )2 )/(1 ) near 1.
3.3. The large-0 limit: A dispersive representation of the kernel
Upon disentangling the hard, jet and soft scales in (3.2) we have significantly reduced
the complexity of the problem: we are now dealing with two independent single-scale quantities,
J (rm2 ) and S(r 2 m2 ). At this point we need additional tools to examine the all-order structure of
these two anomalous dimensions. As these objects define the exponentiation kernel, an approximation based on a single dressed gluon appears most natural [22]; multiple emission is taken into
account by exponentiation. In this section we shall therefore consider the single-dressed-gluon
approximation, the large-0 limit, postponing the discussion of multiple emission to Section 5
where we shall perform exponentiation in moment space.
73
A convenient way to compute the Sudakov anomalous dimensions in the single-dressed-gluon

approximation is the dispersive technique.7 A brief description of the method is given in Appendix B. In order to compute the real-emission contribution to some physical quantity to all
orders within this approximation one first evaluates the single gluon emission diagrams with
an off-shell gluon, k 2 = 2 = 0. The result of this O(s ) calculation defines the characteristic
function F(2 /m2 , r). This function is then integrated with the discontinuity of the coupling,
generating the all-order sum. Here we are interested specifically in the singular contributions for
r 0. In taking this limit we expect to identify the dependence on the two scale rm2 and r 2 m2
corresponding to the anomalous dimensions J and S, respectively.
3.3.1. Dispersive representation of the kernel
The result for the characteristic function at small r takes the form:

s
s 1
CR F(, r) = CR
FJ (/r) FS /r 2 + regular terms,
r

(3.17)
Fsing. (,r)
where 2 /m2 , the ratio between the gluon virtuality 2 and the hard scale in the process,
m2 . The singular terms are distinguished by considering the limits limr0 (rF(, r)) with either
fixed /r or fixed /r 2 . The terms that are regular in these two limits are irrelevant for Sudakov
resummation. A detailed example of how Fsing. (, r) is constructed by considering these two
limits can be found in Section 3 of Ref. [24], where the kernel of the C parameter was computed
(note the difference in complexity(!) between the full F (, r) and Fsing. (, r)). Below we shall
summarize the results for Fsing. (, r) in some examples.
The central point is that Eq. (3.17) exhibits the general property of infrared factorization
summarized by Eq. (3.2): F(, r), which is a function of two variables can be decomposed in
this limit into a sum of two functions of a single variable: one depends on the jet mass scale and
the other on the soft scale. Such a separation of scales cannot be done in an ordinary fixed-order
calculation: it requires keeping an infrared regulator8 in place.
With a single gluon emission the contribution to the physical quantity itself and to the exponentiation kernel are the same (see the arguments leading to Eq. (3.38) below). We can therefore
use (3.17) to construct the resummed kernel in the large-0 limit as a dispersive integral according to Eq. (B.10):

2

R(m2 , r)
d
CR
CR
=
V 2 Fsing. 2 /m2 , r Fsing. (0, r) ,

2
r
0
large 0
(3.18)
where we neglected the power corrections associated with the analytization of the Landau
singularity; we shall revisit this issue in Section 7 and Appendix D. Here the spectral density
7 The dispersive technique is a general method to perform resummation of running-coupling corrections. This technique was developed in Refs. [49,61,62], with no specific consideration of the Sudakov limit; it was used to study power
corrections in a variety of applications. A well-known example is the average thrust [45,48,49]. In Ref. [22] the dispersive technique was used to compute the Sudakov exponentiation kernel for the thrust distribution based on the results
of Ref. [46] for the characteristic function. Similar calculations using the dispersive technique were later done for other
event-shape variables including the heavy-jet mass [23], the C parameter [24] and angularities [25], as well as for heavy
quark fragmentation [30].
8 As alternatives to the gluon mass regulator one can work in D-dimensions or use the Borel technique (see
Appendix B).
74
function V (2 ), as defined by Eq. (B.6), stands for the timelike discontinuity of the coupling in
the V-scheme, where
2V = 2 e5/3 ,
(3.19)
where 2
is defined in the MS scheme. As we shall see below (Eq. (4.20)) this is also the large-0
limit of the gluon bremsstrahlung coupling.
3.3.2. Dispersive representations of the Sudakov anomalous dimensions
Eq. (3.18) with Fsing. (2 /m2 , r) of (3.17) constitute the dispersive representation of the Sudakov evolution kernel in the large-0 limit; it involves two physical scales. We now wish to
construct similar representations for the jet and soft Sudakov anomalous dimensions (3.3)
each involving just one scale.
According to the leading-order result, Eq. (3.1), Fsing. (2 /m2 , r) must reach a finite limit for
2
0, namely:
ln(r) b1 d1
(3.20)
+
.
r
r
This, together with the functional form of Fsing. (2 /m2 , r) summarized by (3.17), implies that
the two functions FJ and FS are separately logarithmically divergent for 2 0,
Fsing. (0, r) =
FJ () = ln() + b1 + O();
FS () = ln() + d1 + O().
(3.21)
Therefore, Eq. (3.18), as written, cannot be split into the jet and soft contributions.
An alternative representation of the kernel can be obtained using integration-by-parts; following [49] we define the timelike coupling aVMink (2 ) through
da Mink (2 )
= V
;
d ln 2
aVMink
dm2 2
V m ,
m2
(3.22)
obtaining

2

R(m2 , r)
d Mink 2
CR
Fsing. 2 /m2 , r
=
aV
CR

2
r
0
large 0
0
CR 1
=
0 r
2
2
d2 Mink 2
FJ
a
FS 2 2 ,
2 V
rm2
r m
(3.23)
where we use the usual convention,

dF (, r)
F (, r)
d ln 1/
and similarly
d
F J () FJ ();
d
(3.24)
d
F S () FS (),
d
(3.25)
so

FJ () =

dy
FJ (y);
y

FS () =

dz
FS (z).
z
(3.26)
75
In the second line of Eq. (3.23) we inserted F sing. (2 /m2 , r) according to (3.17).
Infrared safety of (3.23) implies that F sing. (2 /m2 , r) should vanish for 2 0. Obviously
this is not true for the separate terms F J and F S in the square brackets, which do not vanish for
2 0 but rather tend to a constant; according to (3.21) they obey
F J (0) = F S (0) = 1.
(3.27)
Thus, the integral (3.23) cannot be simply split into two finite integrals one corresponding to the
jet contribution and one to the soft. According to (3.2), however, such a split must be possible
upon taking one logarithmic derivative. Indeed using (3.23) one gets

2
2

d
1
1 dR(m2 , r)
sing. 2 /m2 , r ,
F
=
(3.28)
V

r d ln m2 large 0 0
2
0
and therefore

2
2
2
2
d
1
dR(m, r)
F
=
F
,
V
J
S
d ln m2 large 0 0
2
rm2
r 2 m2
(3.29)
which, in virtue of (3.27), can be split into two finite integrals by subtracting F J (0) = 1 and
adding F S (0) = 1 inside the square brackets. By comparing with (3.2), Eq. (3.29) implies that
the jet and soft anomalous dimensions have the following dispersive representation in the
large-0 limit:

1
J k 2 large =
0
0

d2 2 2 2
V FJ /k F J (0) ,
2
(3.30)

d2 2 2 2
V FS /k F S (0) ,
2
(3.31)
and

1
S k 2 large =
0
0

0
where we used (3.27). Thus F J and F S can be identified as the characteristic functions associated to the jet and soft Sudakov anomalous dimensions J and S, respectively. The
corresponding representations in term of aVMink (2 ) are:

1
J k 2 large =
0
0
d2 Mink 2 2 2
FJ /k ,
a
2 V
(3.32)
d2 Mink 2 2 2
a
FS /k ,
2 V
(3.33)
and

1
S k 2 large =
0
0

0
where
d
d
FJ () = F J ();
(3.34)
FS () = F S ().
d
d
These representations of the Sudakov anomalous dimensions will be generalized in Section 4 to
full QCD.
76
3.3.3. Dispersive representation of Sudakov anomalous dimensions as limits of the full physical
evolution kernels
Eqs. (3.30) and (3.31) can be obtained as r 0 limits of the corresponding finite-r dispersive
representations of the physical evolution kernels in momentum space. We illustrate this statement
(and identify the precise limit to be taken) with the examples of deep inelastic structure functions
and the DrellYan cross section. This will also provide an illustration of the statement preceding
Eq. (3.18) above.
3.3.3.1. Deep inelastic structure functions Let us consider the large-0 limit but finite x dispersive representation [49] of the physical evolution kernel. We note that at the level of the partonic
calculation, the convolution on the r.h.s. of (3.4) becomes trivial in the large-0 limit since the
O(s ) corrections to F2 (z, Q2 ) generate terms that are subleading by powers of 1/0 . Therefore,
in this limit KDIS(F2 ) (x, Q2 ) is directly proportional to the partonic dF2 (x, Q2 )/d ln Q2 , so

(1/x 1)KDIS(F2 ) x, Q2 large
CF
0

d2 2
x) ,
V (1 x) F 2 /Q2 , x F(0,
2
(3.35)
where F(2 /Q2 , x) is the standard notation for the characteristic function corresponding to
F2 (x, Q2 ) (see Eq. (4.27) in [49]). In Eq. (3.35) we multiplied both sides by a factor of r = 1 x
in order to get a finite limit for the integrand for x 1 (see Eq. (3.45) below), in agreement with
the general result Eq. (3.17). Upon taking this limit under the integral with a fixed invariant mass
W 2 = Q2 (1 x)/x one obtains
lim
x1
fixed W 2

(1/x 1)KDIS(F2 ) x, Q2 large
CF
=
0

d2 2 2
V FDIS /W 2 F DIS (0) ,
2
(3.36)
where we substituted

F DIS 2 /W 2 =
lim
x1
fixed W 2

(1/x 1)F 2 /Q2 , x .
(3.37)
Eq. (3.36) shows in particular that the specified limit of (1/x 1)KDIS(F2 ) (x, Q2 )|large 0 does
exist. This (large-0 ) result is consistent with the more general result Eq. (3.9) (valid also at finite 0 ). Moreover, Eq. (3.9) allows to connect directly the momentum-space physical quantity
KDIS(F2 ) (x, Q2 ) in the Sudakov limit with the physical Sudakov anomalous dimension, i.e., the
exponentiation kernel. Comparing with Eq. (3.9), we thus deduce Eq. (3.30), with the identification

F J 2 /k 2 = F DIS 2 /k 2 .
The explicit result [57,70] will be given in Eq. (3.45) below.
(3.38)
77
3.3.3.2. DrellYan cross section In quite a similar way, we start from the large-0 limit but
finite dispersive representation of the quarkantiquark partonic cross section gq q (, Q2 , 2F )
in (3.10), which was calculated in Ref. [49]. By taking a logarithmic derivative with respect to Q2
we obtain the large-0 result for the physical DrellYan evolution kernel (3.14),

(1 )KDY , Q2 large
0
CF
0

d2 2
V (1 ) F DY 2 /Q2 , F DY (0, ) ,
2
(3.39)
where FDY (2 /Q2 , ) is the characteristic function corresponding to gq q (, Q2 , 2F ) and where

we multiplied by a factor of r = 1 . Next, by taking the 1 limit under the integral while
fixing the total energy radiated into the final state, EDY = Q(1 ), we obtain:

(1 )KDY , Q2 large
lim
1
fixed EDY
CF
=
0

d2 2 2 2
V 2FDY /EDY 2F DY (0) ,
2
(3.40)
where we substituted

2
=
2F DY 2 /EDY
lim
1
fixed EDY

(1 )F DY 2 /Q2 , ,
(3.41)
and the limit can be shown [70] to exist, see Eq. (3.46) below. Comparing with Eq. (3.16), we
deduce Eq. (3.31) for S = SDY , with the identification:

F S 2 /k 2 = F DY 2 /k 2 .
(3.42)
3.4. Characteristic functions of Sudakov anomalous dimensions: results
So far we have analyzed the properties of the physical Sudakov anomalous dimensions and
their dispersive formulation in the large-0 limit based on general considerations. Let us now
summarize the results for the characteristic functions based on explicit calculations. These calculations have been usually done considering a specific distribution at finite r, computing the
full characteristic function F (, r) by evaluating the squared matrix element with an off-shell
gluon, and only at the end identifying the terms that are singular at r 0 as demonstrated in
Eqs. (3.37) and (3.41) above. There is however an alternative: it was shown in [57] that there
exists a systematic approximationwhere one lightcone component of the gluon is taken small
together with its transverse momentum squared and its virtualitythat allows a direct calculation of the r 0 singular terms, namely FJ ,S . The calculation of the soft characteristic function
FS can be simplified further, taking all the gluon momentum components to be small. This is
the standard soft approximation albeit with an off-shell gluon. Here the hard partons that emit
the radiation can be replaced by Wilson lines. The soft function can then be obtained through
the renormalization of a corresponding Wilson-loop operator defined by the trajectories of the
colored hard partons [32,54,85,86].
A compilation of the expressions for the characteristic functions in a few examples is given in
Table 1. The results for the derivative F J ,S , defined by Eq. (3.25) or (3.26), and the corresponding Borel function, defined by Eq. (6.7) below, are summarized in Table 2.
78
Table 1
Results for the momentum-space characteristic functions of some inclusive distributions
FJ (y = /r)
Process
(y < 1)[ ln(y) 34 + 12 y + 14 y 2 ]
Jet function (e.g., DIS)
FS (z = /r 2 )
Process
B decay; HQ fragmentation
SQD
1
ln(1 + 1/z) + 1+z
DrellYan; gg Higgs
SDY
(z < 1/4)[2 ln 1+ 214z ln(z)]
e+ e jets C parameter (r = c = C/6)
Sc
(z < 4) ln(z) + (z > 4)[2tanh1 ( 1 4/z) ln(z)]
e+ e jets thrust (r = t = 1 T )
St
(z < 1) ln(z)
Table 2
Results for the logarithmic derivatives of momentum-space characteristic functions of some inclusive distributions and
the corresponding Borel functions (6.7)
F J (y = /r)
Process
J

Process
BJ (u)
y2
(y < 1)[1 y2 2 ]
1( 1 +
1
2 1u
1u/2 )
F S (z = /r 2 )
BS (u)
(1 u) sin u u
SQD
DrellYan; gg Higgs
SDY
e+ e jets C parameter (r = c = C/6)
Sc
e+ e jets thrust (r = t = 1 T )
1
(1+1/z)2
(4z<1)
14z
1 (z>4)
14/z
St
1 (z > 1)
2 (1u)
(12u)
2 (1+u)
(1+2u)
3.4.1. Inclusive B decay and heavy-quark fragmentation

Let us consider the characteristic function corresponding to the exponentiation kernel in inclusive B decays. The simplest calculation is of the photon-energy spectrum d (m2b , x)/dx
in radiative B decays B Xs , where x = 2E /mb . This spectrum was computed in the
large-0 limit in [40]. The original calculation was done using the Borel representation of the
dressed gluon propagator. In the dispersive formulation we obtain the following singular terms
for r = 1 x 0:

1
3 1 1 2
Fsing. (, r) =
( < r) ln(/r) +
+
r
4 2r
4 r2

FJ (/r)

1
ln 1 + r 2 / +
,
1 + /r 2

(3.43)
FSQD (/r 2 )
with = 2 /m2b , where we have already split the result according to (3.17) and identified the
two functions FJ (/r) and FSQD (/r 2 ). Equivalently, for the derivative we have:
F sing. (, r)
dFsing. (, r)
d ln 1/

1
1
/r 2
1 1 2
+
( < r) 1
.
r
2r
2 r2
1 + /r 2 (1 + /r 2 )2

F J (/r)
79
(3.44)
F SQD (/r 2 )
Here the superscript QD stands for Quark Distribution, indicating that this particular soft function is related to the longitudinal momentum distribution of an off-shell b quark inside an on-shell
b-quark state. The same result for the singular terms (3.43) was obtained for the triple differential
rate in b Xu l ; here r is defined as the ratio between the two lightcone components p + /p
of the Xu jet, p = E |p|,
and the hard scale is p .

Next we comment that both FJ (/r) and FS (/r 2 ) can be extracted from the calculation
of heavy-quark fragmentation in e+ e annihilation, Eq. (54) in [30]. The fact that FS (/r 2 ) is
the same for heavy quark decay and fragmentation follows from the general equality of the two
Sudakov anomalous dimensions, proven in [32].
Finally, the quark distribution function at finite mass and arbitrary momentum fraction has
been computed in the large-0 limit in Ref. [32] using the Borel regularization. Repeating the
calculation with a massive gluon and identifying the terms that are singular for r 0 we recover
FSQD (/r 2 ) of (3.43).
3.4.2. The jet function
As already mentioned in the introduction, the jet function appears in many different inclusive
distributions where Sudakov logs emerge from a constraint on the invariant mass of a jet in the
final state [57]. The simplest example is deep-inelastic structure functions, where J is the only
source of Sudakov logarithms [5]. According to Eq. (3.37), the jet characteristic function is
obtained upon considering the x 1 limit of the F2 (x, Q2 ) characteristic function (Eq. (4.27)
in [49]) with a fixed invariant mass Q2 (1 x)/x. The result is [57,70]:

lim
(1 x)F(, x)
FJ /(1 x) =
x1
fixed /(1x)

3 1
1 2

+
+
,
= ( < 1 x) ln
1x
4 2 1 x 4 (1 x)2
(3.45)
where = 2 /Q2 . As anticipated, this result is identical to the jet-function contribution to various infrared safe observables including inclusive B decay (see [40]), Eq. (3.43) above with
r = 1 x, and a range of distributions in e+ e annihilation into hadrons [57]: event-shape
variablese.g., Eq. (16) in [23] and Section 3 in [24], single particle inclusive cross section
Eq. (28) in [57], and heavy-quark production cross section.
3.4.3. The DrellYan soft function
The DrellYan (or Higgs) production case offers the basic example of a soft function. Using in Eq. (3.41) above the explicit expression from Ref. [49] for the DrellYan characteristic
function FDY (, ) with = 2 /Q2 , one obtains [70]:

1
FSDY /(1 )2 =
2
lim
1
/(1 )2 fixed
1
= 2 tanh

(1 )FDY (, )

4
1
4/(1 )2 < 1 .
2
(1 )
(3.46)
(3.47)
80
Upon defining z /(1 )2 ,
FSDY (z) = 2 tanh1 ( 1 4z ) (z < 1/4)

1 + 1 4z
= 2 ln
ln(z) (z < 1/4)
2
(3.48)
and therefore

(4/(1 )2 < 1)
F SDY /(1 )2 =
.
1 4/(1 )2
(3.49)
The same result has been obtained in Ref. [52] (see Eq. (4.6) there) and Ref. [57] (see Eq. (69)
and the discussion following it) by considering directly the renormalon calculation in the soft
approximation. Finally, note that in the DrellYan case the second derivative FSDY is singular
and therefore one must use (3.31) rather than (3.33).
3.4.4. Event-shape distributions
Beyond the universal jet function discussed above, event-shape distributions in e+ e annihilation are sensitive to large-angle soft gluon emission from the back-to-back recoiling quarks.
The soft function strongly depends on the way the shape variable weighs the parton momenta.
This is reflected in shape-variable dependent subleading Sudakov logarithms as well as power
corrections.
Before considering specific examples, a general comment is due concerning the notcompletely-inclusive nature of event-shape variables: in contrast with the inclusive cross sections
and decay spectra discussed above, event-shape variables do distinguish at some level between
an off-shell gluon and the final-state particles to which it decays. Differences arise already in the
large-0 limit. The characteristic functions for event-shape variables are defined in the inclusive
approximation, neglecting any such differences.
Let us first recall the result in the case of the thrust (or the heavy jet mass) which are particularly simple. Starting from the exponentiation kernel in the large-0 limit (e.g., Eq. (20) in [23])
we have, in full analogy with Eq. (3.23),

R(m2 , t)
CR

t
large 0
CR
=
0

0

d Mink 2 1
1 1 2
Q
aV
1
2 ( < t) > t 2 ,

t
2t
2t

(3.50)
F sing. (,t)
where = 2 /Q2 , Q2 is the center-of-mass energy and t = 1 T , where T is the thrust variable. For thrust CR = 2CF . In order to extract the separate jet and soft contributions to the
characteristic function we use the identity:

( < t) > t 2 = ( < t) 1 > t 2
and note that the terms proportional to /t or 2 /t 2 in (3.50) contribute only at the = t
phase-space limit (they are power-suppressed at the = t 2 limit). This leads to the following
81
decomposition of F sing. (, t) in (3.50)

1
1 1 2
2
Fsing. (, t) =
( < t) 1
2 1 >t
,
t
2t
2t

2
F J (/t)
(3.51)
FSt (/t )
where we neglected regular contributions that are irrelevant for Sudakov resummation. Here we
recognize the characteristic function of the jet from (3.44) and (3.45) while for the soft contribution we have:

F St /t 2 = 1 /t 2 > 1 = /t 2 < 1 .
(3.52)
In a similar way we can write down the characteristic function of the C parameter using the
results of Ref. [24]. Upon neglecting regular contributions, we obtain (see Eq. (3.5) in [24]):

4c2 /
1 1 2
2 1

+
1
Fsing. (, c) = ( < c) > 4c

c
2 c 2 c2
1 4c2 /(1 + 1 4c2 / )

1
( > 4c2 )
1 1 2

(3.53)
( < c) 1
c
2 c 2 c2
1 4c2 /

F J (/c)
F Sc (/c2 )
with c = C/6, where C is the conventional normalization of the C parameter (see [24]). Again
we recognize the familiar jet function and find to be:

(/c2 > 4)
.
F Sc /c2 = 1
1 4c2 /
(3.54)
Note the similarity [54] with the case of the DrellYan soft function, Eq. (3.49). Also here the
second derivative FSc is singular, so one can only use (3.31).
4. From the large-0 limit to the non-Abelian theory
The dispersive technique is a convenient way to derive the all-order result in the large-0 limit.
But, in fact, it is much more than that: a dispersive representation, when it exists, summarizes
the analytic structure of the observable in the complex momentum plane as well as its infrared
sensitivity. Moreover, it is particularly well-suited to parametrizing power corrections by means
of infrared-finite coupling.
The purpose of the present section is to show that the dispersive representation of the Sudakov
anomalous dimensions, Eqs. (3.30) and (3.31), which emerges out of the calculation in the large0 limit, can be readily generalized to the full theory. In this generalization the characteristic
functions, which have been computed analytically in the large-0 limit, are kept fixed, while
the V-scheme coupling aVMink , which was so far identified only to NLO in the large-0 limit, is
replaced by specific effective charges that are defined order-by-order in QCD and capture the
non-Abelian nature of the interaction. After presenting the general formalism we will examine a
few examples where corrections are known to the NNLO and beyond. The investigation of the
evolution of these effective charges leads to interesting observations.
82
4.1. A general dispersive representation of Sudakov anomalous dimensions

The generalization of Eqs. (3.30) and (3.31) (as well as (3.32) and (3.33)) beyond the large-0
limit is straightforward:

1
J k2 =
0
=
0

1
0

d2 2 2 2
J FJ /k F J (0)
2
d2 Mink 2 2 2
FJ /k ,
a
2 J
(4.1)
and

1
S k2 =
0
=
1
0

0

d2 2 2 2
S (0)
/k
S
S
2
d2 Mink 2 2 2
FS /k ,
a
2 S
(4.2)
where, in full analogy with (3.22),

J ,S

Mink 2
daJ
,S
d ln 2
Mink
aJ
,S

dm2
J ,S m2 ,
2
m
(4.3)
and, similarly to (B.6), J ,S (2 ) correspond to the timelike discontinuities of some Euclidean

effective charges, originally defined for spacelike momenta:
Eucl
aJ
,S
2
k =

0
d2 Mink 2
2 /k 2
a
=
2 J ,S
(1 + 2 /k 2 )2

0

d2
J ,S 2 ,
2
2
+k
(4.4)
Eucl (k 2 ) that would violate the physical

where we ignored potential Landau singularities in aJ
,S
analytic properties of these effective charges (cf. Eq. (B.5)). Of course, such singularities may
appear at any given order in perturbation theory. We shall return to this issue in Sections 4.3
and 7.
Let us note in passing that while the Minkowskian representation of the Sudakov anomalous
dimensions, Eqs. (4.1) and (4.2), can always be written, in general there are no equivalent EuclidEucl (k 2 ). This fact reflects
ean representations consisting of integrals over the effective charges aJ
,S
the timelike nature of these quantities.
Obviously, we require that in the large-0 limit S (2 ) and J (2 ) would both coincide with
the discontinuity of the V-scheme coupling, V (2 ) of Eq. (B.6). In Section 4.2 we will show,
order-by-order in perturbation theory, that a generalization according to (4.1) and (4.2) is indeed
possible, and then further characterize these effective charges. We stress that the order-by-order
analysis is entirely independent of the assumed analytic structure, as Landau singularities can
only modify the dispersion relations (4.4) by power-suppressed terms.
83
4.2. From anomalous dimensions to effective charges: Order-by-order relations

We wish to show that the perturbative expansion of the anomalous dimensions, S and J , and
consequently their conventional decomposition in MS into the cusp A, B and D, Eq. (3.3), are effectively translated by (4.1) and (4.2), order-by-order in perturbation theory, into an expansion of
Mink (2 ). Thus, by virtue of the dispersion relations (4.4),
the corresponding effective charges, aJ
,S
Eucl (2 ).
they uniquely translate into the expansion of the Euclidean effective charges, aJ
,S
Eucl (2 ) in the MS scheme
Defining the expansion coefficients of aJ
,S

Eucl 2
aJ
= a + j2E a 2 + j3E a 3 + j4E a 4 + ,

aSEucl 2 = a + s2E a 2 + s3E a 3 + s4E a 4 + ,
where a 0 sMS (2 )/ , we have

2 3
5
Mink 2
= a + j2E a 2 + j3E
a + j4E 2 j2E + a 4 + ,
aJ
3
6

2

5
Mink 2
E 2
E
3
E
2
E
aS
= a + s2 a + s3
a + s4 s2 + a 4 + ,
3
6
(4.5)
(4.6)
where we used (4.4) and the perturbative expansion of the running coupling in the MS scheme,
where:

a z2 = a 2 ln(z)a 2 2 + ln2 (z) ln(z) a 3 2

5
+ ln3 (z) + ln2 (z) 2 ln(z) a 4 2 + ,
(4.7)
2
where the function is given by

d a
= a 2 1 + a + 2 a 2 + ,
2
d ln
(4.8)
with = 1 /0 2 and n = n /0 n+1 .

Next we note that Eqs. (4.1) and (4.2) can be written as
1
=
0
dy Mink 2
y FJ (y)
a
y J
(4.9)
dz Mink 2
z FS (z).
a
z S
(4.10)
and
1
=
0

0
Substituting (4.7) into (4.6) we get:

2

Mink
y = a 2 + j2E ln(y) a 2 2
aJ

2
E
2
E
+ ln (y) 2 ln(y)j2 ln(y) a 3 2
+ j3
3
84

5
E
+ ln (y) + 3j2 + ln2 (y) + 2j2E 3j3E + 2 2 ln(y)
2

5
+ j4E 2 j2E + a 4 2 + ,
(4.11)
6
3
and a similar expression for aSMink (z2 ) where jnE are replaced by snE . We observe that upon
inserting these expansions into (4.9) and (4.10) we would need to evaluate the log-moments of
FJ (y) and FS (z), respectively,

Jk
dy k
ln (y)FJ (y),
y

Sk
dz k
ln (z)FS (z).
z
(4.12)
The first few log-moments in some examples are given in Table 3 (see discussion below).
Using Eq. (4.11), Eq. (4.9) yields the following expansions:
CR J

2
E
CR
2
E
E
3
=
+ J2 2J1 j2 J1 a +
a + j2 J1 a + j3
0
3

CR
=
a + a2 a 2 + a3 a 3 + a 2 + a 3 + b1 + 2b2 a +
0
d a(
2 ) dB(s (2 ))
,
= A s 2 +
d a
d ln 2
(4.13)
where we have identified the expansions of the anomalous dimensions A(s (2 )) and B(s (2 ))
of Eqs. (2.9) and (2.10), respectively. At order O(a 2 ) we find b1 = J1 . Eq. (4.13) yields the
Eucl in (4.5):
following expressions for the coefficients for the Euclidean effective charge aJ
j2E = a2 ,
1
j3E = a3 J2 + 2b1 a2 + 2 2b2 ,
3
j4E
= a4 + 6b12 a2

5
+ (2a2 3J2 6b2 + 3a3 )b1 + 3a2 J2 + J3
2
5
+ 2 3b3 + 2 a2 2b2 ,
6
(4.14)
where the log-moments Jk are defined in (4.12) and an and bn are the MS coefficients of (2.9)
and (2.10), respectively.
Similarly (4.10) yields:

CR

2
+ S2 2S1 s2E S1 a 3 +
CR S 2 =
a + s2E S1 a 2 + s3E
0
3
85
Table 3
Summary of results for the first few log-moments of the characteristic functions, Jk and Sk with k = 0 to 5, of some
inclusive distributions (cf. Table 2). The log-moments are defined in Eq. (4.12). They can be most easily extracted
using (6.7) by expanding the Borel functions in Table 2
Process
J0
J1 = b1
J2
J3
J4
J5
3/4
5/4
27/8
51/4
495/8
S0
S1 = d1
S2
S3
S4
S5
SQD
2 /3
7 4 /15
7 4 /3
DrellYan; gg Higgs
SDY
2 /3
123
3 4 /5
40 2 3 + 7205
123
3 4 /5
40 2 3 7205
Process
e+ e jets C parameter
Sc
2 /3
e+ e jets thrust
St

CR
a + a2 a 2 + a3 a 3 + a 2 + a 3 + (d1 + 2d2 a + )
0
d a(
2 ) dD(s (2 ))
= A s 2 +
,
d a
d ln 2
(4.15)
where the expansions of the anomalous dimensions A(s (2 )) and D(s (2 )) of Eqs. (2.9)
and (2.11), respectively, have been identified. At order O(a 2 ) we find d1 = S1 . Eq. (4.15) yields
the following coefficients for aSEucl in (4.5)
s2E = a2 ,
1
s3E = a3 S2 + 2d1 a2 + 2 2d2 ,
3
s4E
= a4 + 6d12 a2

5
+ (2a2 3S2 6d2 + 3a3 )d1 + 3a2 S2 + S3
2
5
+ 2 3d3 + 2 a2 2d2 ,
(4.16)
6
where the process-dependent log-moments Sk are defined in (4.12) and an and the processdependent dn are the MS coefficients of (2.9) and (2.11), respectively.
Obviously, such an identification can be done to arbitrarily high order: according to (4.1)
and (4.2), the information contained in J and S at a given order in the perturbative expansion is
Eucl and a Eucl , as well as a Mink and a Mink , to that order.
sufficient to determine aJ
S
J
S
Finally, note that following the identification we made above we can write the small- expansion of the characteristic functions (3.21) as

FJ () = ln() + J1 + O() = ln() +
dy
ln(y)FJ (y) + O(),
y
(4.17)
dz
ln(z)FS (z) + O(),
z
(4.18)

FS () = ln() + S1 + O() = ln() +
0
where the constant terms b1 = J1 and d1 = S1 were fixed by the first log-moment of the characteristic function itself.
86
4.2.1. Cusps and NLO universality of the gluon bremsstrahlung coupling

Eucl (2 ) are closely related to the cusp anomalous
We observe that the effective charges aJ
,S
dimension [65,66]:
A

CR Eucl CR Eucl CR
a + a2 a 2 + ;
aS
aJ
0
0
0
(4.19)
their expansions start deviating [68,69] only at the NNLO. We emphasize that the NLO universality of the effective charges holds despite the fact that the source of the soft radiation can be quite
different: it applies to the entire class of inclusive distributions discussed in this paper, including
B decay spectra, event-shape distributions, deep inelastic structure functions, single-particle inclusive cross sections, DrellYan production and heavy quark fragmentation. To NLL accuracy,
the only way by which the non-Abelian nature of the interaction affects the exponent is through
the cusp anomalous dimension (see, e.g., [32,85,86]), and therefore this effect is independent of
the details of the configuration of color sources emitting the radiation.
This NLO universality explains and justifies a posteriori some results obtained in the past
in several occasions [12,22,49]. In particular, in [12] it was proposed to absorb the NLO term
into the definition of the coupling, to define the gluon bremsstrahlung coupling. This can be
formulated as a rescaling of the dimensional transmutation scale as follows (see also Eq. (60)
in [12]):

MS
5 CA 1 2
2
2
2
+
,
GB = exp a2 = exp
(4.20)
3
0 3 12
where the subscript GB stands for gluon bremsstrahlung and where 2 is defined in MS.
Thanks to this NLO universality, in the original DGE paper [22] it was possible to generalize the exponentiation of a single dressed gluon to full next-to-leading logarithmic accuracy,
without having to introduce two separate effective charges. At this level, differences between
observables, which affects subleading logarithms as well as power corrections, are encoded only
in the different functional form of the characteristic function. Beyond the NLO the soft and the
jet effective charges start differing from the cusp anomalous dimension, which also becomes
scheme dependent.
Note that differences with respect to the cusp anomalous dimension appear already in the
large-0 limit. Nevertheless, we expect that similarly to the cusp anomalous dimension, the
Eucl (2 ) and a Eucl (2 ) would be renormalon free.9 Thus, the expansions
effective charges aJ
S
Eucl (2 ) and
(4.5) should not develop large subleading corrections. Absence of renormalons in aJ
Eucl
2
2
2
aS ( ) implies that also J ( ) and S( ) are renormalon-free, since the small 2 /k 2 expansion of the characteristic functions in (4.1) and (4.2) does not give rise to any non-analytic terms.
These relations will be further discussed in Section 6.1 using the Borel formulation.
4.2.2. Examples
Let us now use the explicit expressions for the characteristic functions in Section 3.4 together
with the known perturbative expansions of the Sudakov anomalous dimensions, Eqs. (2.9), (2.10)
and (2.11) with the coefficients in Appendix A, to derive the corresponding expansions of the
Eucl and a Eucl .
S
9 It should be emphasized that absence of renormalons in a Eucl (2 ) and a Eucl (2 ) is obvious in the large- limit,
0
J
S
where these functions coincide with the one-loop coupling (4.39), but it is far from obvious beyond this limit.
4.2.2.1. The jet function Starting with the jet function case, from (3.44) we get:

y y2
1
F J (y) = (y < 1) 1
FJ (y) = (y < 1)y
y ,
2
2
2
87
(4.21)
so using (4.9) we obtain

1
J 2 =
0

0
dy Mink 2
1
y FJ (y) =
aJ
y
0
1

2
1
Mink
y
y+
dy aJ
.
2
(4.22)
The corresponding log-moments defined in (4.12) are:

Jk
0
dy k
ln (y)FJ (y) =
y
1

1
dy lnk (y) y +
= (1)k k! 1 + 2k /2.
2
(4.23)
Note that these coefficients increase factorially. The first few numerical values are summarized
in Table 3.
Using now (4.14) with the MS bn in Appendix A we get:
Eucl
= a + j2E a 2 + j3E a 3 + j4E a 4 +
aJ

101 3
2
= a + a2 a + a3 +
a2
18
2

1
3
2
73
+
53 CA +
+ 33 CF a 3
0
72
16
4

9
3
901
+ a4 a 3 + + 2 a 2 +
+ 23
4
8
216

5 2
1
15CA
9CF
a2
3
0
8
8

1979 37 2 50
13 4
993 45 2 23
+
3 +
CA +
3 CF
72
54
3
120
32
32
4

21 2 33
1
C A + C F CA a2
+ 2
32
32
0

385 905
37 2 41 4 11 2
87
+
+ 3 + 5 CA 2
216
48
432
960
24
8

19 723 3 2 245
17 4 1 2
45
+
+
3 +
+ 3 + 5 CF CA
768
4
24
240
8
8

2
4
105 3
3
45
87
1 2
2
+
+ 3 +
3 5 CF
128
64
16
40
4
4

1795 17
7 2
35
511
1
+ 3 CA 3 +
+ 3 +
CF CA 2
+ 3
1152 16
2304
8
64
0

33
11 2
33
+
3 +
CF 2 CA a 4 + ,
256 16
64
(4.24)
88
where the coefficients an , which are known in full to three-loop order (see Appendix A) have
not been substituted. As with the anomalous dimension J itself, the only missing ingredient in
Eucl (2 ) at four-loop order (N3 LO) is the four-loop coefficient of the cusp anomalous dimenaJ
Eucl / A is known to this order.
sion, a4 : the combination CR aJ
0
The fact that the four-loop coefficient is not known in full is rather unimportant in practice:
the series for the cusp anomalous dimension A in MS converges very well, and a4 can be reliably
estimated using Pad approximants [87] as discussed in Section 4 in [7]. For example, for Nc = 3
and Nf = 4 we get:

CR
1.341 [1, 2] Pad
2
3
4
A=
(4.25)
+ ,
a + a 0.962 + a 1.159 + a
1.397 [2, 1] Pad
0
where the difference between the two Pad approximant predictions can be used as a rough
measure of uncertainty.
Using these coefficients of A in (4.24) we get10 :
Eucl
= a + a 2 0.962 a 3 1.019
aJ

3.221 using [1, 2] Pad for a4 in (4.25)
4
a
+ .
3.165 using [2, 1] Pad for a4 in (4.25)
(4.26)
Note that this expansion converges very well, better than the corresponding Minkowskian coupling:
Mink
= a + a 2 0.962 a 3 4.309
aJ

18.798 using [1, 2] Pad for a4 in (4.25)
a 4
+ ,
18.742 using [2, 1] Pad for a4 in (4.25)
(4.27)
and the physical anomalous dimension J itself:

0 J = a + a 2 1.712 a 3 1.061

17.606 using [1, 2] Pad for a4 in (4.25)
4
a
+ ,
17.549 using [2, 1] Pad for a4 in (4.25)
(4.28)
which both contain large 2 terms (see (4.6)) owing to the analytic continuation to the timelike region. It should be emphasized that (at least in the large-0 limit) none of the three has
renormalons, despite the fact that the log-moments Jk increase factorially. The most convenient
way to see this is by considering the relations between these quantities in terms of their Borel
representations, see Section 6.1 below.
4.2.2.2. The quark distribution function (B decay) Consider next the soft function associated
with the quark distribution in an on-shell heavy quark. From (3.44) we get:
F SQD (z) = 1
1
(1 + 1/z)2
FSQD (z) =
2z2
.
(1 + z)3
(4.29)
10 Note in contrast with (4.25), a direct Pad approximant prediction of the N3 LO coefficient in a Eucl is unreliable: the
J
[1, 2] and [2, 1] approximants differ by much. A direct prediction for (4.27) or (4.28) below is even worse owing to the
2
large terms related to the analytic continuation.
89
Using (4.10) we get

SQD
1
=
0

0
dz Mink 2
1
z FSQD (z) =
a
z SQD
0
2
z
dz aSMink
QD
2z
.
(1 + z)3
(4.30)
The corresponding log-moments Sk are given by

SkQD
dz k
ln (z)FSQD (z) =
z

dz lnk (z)
0
2z
.
(1 + z)3
Their first few values are summarized in Table 3.

Using (4.14) we get the following results for coefficients in Eq. (4.5):
2
aSEucl
= a + s2E a 2 + s3E a 3 +
QD

10
22 CA 5 2 7 9
2
= a + a2 a + a3 + a2
+
+ 3 a 3 + .
3
9
0 18
9 2
(4.31)
(4.32)
Numerically, we find again very good convergence in the first few orders; for Nc = 3 and Nf = 4
we get:
= a + a 2 0.962 a 3 0.799 + .
aSEucl
QD
(4.33)
4.2.2.3. The DrellYan soft function Consider now the soft function associated with DrellYan
production or Higgs production through gluongluon fusion. The characteristic function is given
by (3.49). Here the second derivative is singular, but using (3.31) one can verify that the relations
in Eq. (4.16) hold. Then, using (6.7) to extract the log-moments Sk (see Table 3) one readily
obtains:
2
aSEucl
= a + s2E a 2 + s3E a 3 + s4E a 4 +
DY

28 CA 8 7
+
3 a 3
= a + a2 a 2 + a3 +
9
0 9 2

10
116
+ a4 + + 2a2 2 +
+ 23
3
27

1
2905
247 2 23 4 245
+
+
3 +
CA
0
216
120
12
144

645 19
4
+
3
CF
32
2
20

1
269 581
155 11
11 4
2
+ 2
+
+ 3 + 95 +
3 CA 2
240
864 24
864
48
0

6839 73
11 4
+
+ 3 +
CF C A
384
12
240

1
11 77
49
7
+ 3 + 3 CA 3 + + 3 CF CA 2 a 4 + , (4.34)
18 32
18 32
0
90
where we used the known coefficients (A.3). Using (A.1), Eq. (4.34) yields the following expansion for Nc = 3 and Nf = 4:
aSEucl
= a + a 2 0.962 a 3 0.508
DY

9.07 using [1, 2] Pad for a4 in (4.25)
4
+ .
a
9.01 using [2, 1] Pad for a4 in (4.25)
(4.35)
4.3. Renormalization-group evolution of the Sudakov effective charges

In order to evaluate the anomalous dimensions entering the Sudakov exponent one needs to
perform integrals over the discontinuity of the running coupling in (4.1) and (4.2). The scale
dependence of the relevant Euclidean couplings can be computed directly [63] by solving the
appropriate renormalization-group equations. Upon taking the derivative of the effective charges
Eucl (2 ) and a Eucl (2 ) with respect to ln 2 and expressing the result in terms of the effective
aJ
S
charge itself one arrives at the following renormalization-group equations:
Eucl (2 )
daJ

Eucl 2 2 J
Eucl 2 2
Eucl 2
1 + aJ
+ aJ
= aJ
2
Eucl 2 3 J

+ aJ
3 + ,
2
2J = 2MS + j3E j2E j2E ,
3
2
3J = 3MS + 2j4E + 4 j2E + j2E 6j2E j3E 2j2E 2MS ,
(4.36)
2

2

daSEucl (2 )
1 + aSEucl 2 + aSEucl 2 2S
= aSEucl 2
2
d ln
3

+ aSEucl 2 3S + ,
2
2S = 2MS + s3E s2E s2E ,
3
2
3S = 3MS + 2s4E + 4 s2E + s2E 6s2E s3E 2s2E 2MS .
(4.37)
d ln 2
These equations can be truncated at any given order, and integrated. The first approximation
is obtained setting = n = 0 which is also the result of the large-0 limit,

Eucl 2
large = aSEucl 2 large =
aJ
0
a(
2)
1 + 53 a(
2)
(4.38)
Upon using the universal NLO correction of (4.19) to fix the initial condition one obtains:

Eucl 2
one-loop =
aSEucl 2 one-loop = aJ
1
ln 2 /2GB
(4.39)
where GB is defined in Eq. (4.20). At the next truncation order ( = 0 but 2,3,... = 0) the
Eucl (2 ) and a Eucl (2 ) still coincide. The analytic solution can be written in
S
terms of the Lambert W function [72,73], facilitating exact analytic continuation to the time-like
axis.
Eucl (2 ) and a Eucl (2 ) start differing from each other. It is only at this
Beyond this order aJ
S
level that the process-dependent nature of soft gluons radiation reveals itself. Indeed in the large0 limit n = 0. The evolution equations (4.36) and (4.37) directly reflect the non-Abelian nature
91
of the interaction. It therefore becomes particularly important to study the higher-order corrections to these equations.
Here comes a very interesting observation: in contrast with the MS coupling and with other
J ,S
are not all positive.
physical Euclidean effective charges [72], the higher-order coefficients 2,3,...
Most importantly 2J ,S is negative in all the known examples (see below). This opens up a
possibility that an infrared fixed-point would be realized already at the perturbative level. In this
case Landau singularities may not appear in the entire first sheet of the complex momentum
plane, leading to a causal analyticity structure [72,74], where the entire complex momentum
plane is mapped into a compact region in the complex coupling plane. A Landau singularity does
appear in the one-loop coupling (4.39), in the large-0 limit, as well as in the two-loop coupling
with any realistic number of light flavors. Its absence would therefore be a direct consequence of
the non-Abelian nature of soft-gluon interaction, which may be different in different processes.
If this scenario is realized the perturbative solution for the effective charge can genuinely be
a good approximation to the full, non-perturbative effective charge down to the infrared limit. In
this case the Sudakov factor computed as a dispersive integral will be in very good control, far
beyond what can be achieved in a fixed-logarithmic-accuracy approach.
Of course, in reality we know, at best, just 2J ,S and 3J ,S and therefore it is hard to make any
firm conclusions about the size and perturbative stability of the fixed point. Obviously, a fixed
point can always be washed out by sufficiently large subleading corrections. Nevertheless, some
of the examples considered below (in particular the DrellYan case) are quite suggestive of this
scenario.
It is important to stress that the infrared finite coupling scenario discussed here has nothing
to do with Analytic Perturbation Theory (APT) coupling [88], which becomes finite only owing
to the analytization of the Landau singularities by imposing the dispersion relation [33,38,88].
This model implicitly assumes the existence of power corrections that are not of infrared origin.
See further discussion of this issue in Section 7 and Appendix D.
To examine the effective-charge beta function in Eqs. (4.36) and (4.37) it is useful to introduce
a decomposition of the coefficients into powers of 0 (eliminating the Nf dependence in favor
1
of 0 = 11
12 CA 6 Nf ). To this end we define first the decomposition of the coefficients of the
anomalous dimensions ai , bi and di in the MS scheme, Eqs. (2.9), (2.10) and (2.11), respectively:
ai =
1
0 i1
i1

ai,k 0 k ;
bi =
k=0
i1

1
0 i1
k=0
bi,k 0 k ;
di =
i1

0 i1
k=0
di,k 0 k
(4.40)
and the MS beta function

iMS =
i

MS j
i,j
0
(4.41)
j =0
with iMS = iMS /0 i+1 . In what follows we omit the superscript MS, and simply use i,j when
referring to this scheme. Using similar notation we also define the decomposition of the coefficients of the effective-charge beta function in Eqs. (4.36) and (4.37):
iJ =
i
iJ
1 J j
=
i,j 0 ;
0 i+1 0 i+1
j =0
iS =
i
iS
1 S j
=
i,j 0 ,
0 i+1 0 i+1
j =0
(4.42)
92
where the fact that the sum over j goes only up to j = i, as in the MS scheme (4.41), rather
than j = i + 1 (as occurs in most physical effective charges) is a reflection of the fact that in the
large-0 limit the Sudakov effective charges are simply related to the MS coupling, see (4.38)
above.
Considering the first scheme-dependent coefficient,
2J =
J
J
J
2,2
2,1
2,0
2J
J
=
+
+
+
;
2,3
0
0 3
0 2
0 3
2S =
2S
0 3
S
S
S
2,2
2,1
2,0
S
= 2,3
+
+ 2+ 3
0
0
0
(4.43)
J ,S
J
S and J = S are uniwe observe, as anticipated, that 2,3
= 0. We also find that 2,0
= 2,0
2,1
2,1
versal, i.e., they depend only on the coefficients of the cusp anomalous dimension and the MS
J ,S
beta function (they do not depend on the details of the soft function considered) while 2,2
are process dependent, depending on b1 = J1 and b2 or on d1 = S1 and d2 , respectively. The
expressions are:
S
2,0
= a2,0 1,0 + 2,0 ,
S
2
2,1
= a2,0
+ 2,1 a2,1 1,0 a2,0 1,1 + a3,0 ,
S
= 2a2,0 a2,1 + 2S1 a2,0 + a3,1 a2,1 1,1 + 2,2 2d2,0 ,
2,2
S
2,3
= S2 + 2S1 a2,1 +
2
2
+ a3,2 = 0,
2d2,1 a2,1
3
(4.44)
J
and similarly for 2,j
, with the obvious replacement of the coefficients dk by the corresponding bk
J ,S
and Sk by Jk . Note that the vanishing of 2,3
can be verified by taking the large-0 limit of 2J ,S
in Eqs. (4.36) and (4.37) and using j3E and s3E of Eqs. (4.14) and (4.16), respectively.
At the next order,
J
J
J
J
3,3
3,2
3,1
3,0
J
3J = 3,4
+
+ 2 + 3 + 4;
0
0
0
0
0
S
S
S
S
3,3
3,2
3,1
3,0
S
3S = 3,4
+
+ 2+ 3+ 4
0
0
0
0
(4.45)
J ,S
J
S is universal, while the
so we find, as expected, that 3,4
= 0. We also observe that 3,0
= 3,0
J ,S
J ,S
J ,S
other coefficients (3,1
and 3,2
and 3,3
) are process dependent, depending on d1 = S1 , d2
and d3 as well as on S2 . The general expressions are:
S
2
= 22,0 a2,0 + 1,0 a2,0
+ 3,0 ,
3,0
S
2
3
= 2a4,0 + 1,1 a2,0
+ 4a2,0
+ 3,1 22,0 a2,1
3,1
+ 21,0 a2,0 a2,1 + 4a2,0 1,0 S1 41,0 d2,0 22,1 a2,0 6a3,0 a2,0 ,
93

S
2
3,2
= 6a3,0 12a2,0
+ 4a2,0 1,1 + 4a2,1 1,0 S1 + 2a4,1
2
+ 3,2 6a3,0 a2,1 + 12a2,0
a2,1 +
5 2
1,0 6d3,0
3
2
22,1 a2,1 + 1,0 a2,1
22,2 a2,0
6a2,0 a3,1 41,0 d2,1 41,1 d2,0 + 21,1 a2,0 a2,1 + 12a2,0 d2,0 51,0 S2 ,
S
3,3
= 12S12 a2,0 + (24a2,0 a2,1 + 6a3,1 12d2,0 + 4a2,1 1,1 )S1

2
2
+ 3,3 + 12a2,0 a2,1
22,2 a2,1 6d3,1 + 1,1 a2,1
51,1 S2 41,1 d2,1
5 2
+ 2a4,2 + 12a2,0 d2,1 6a2,0 a3,2 +
1,1 + 12a2,1 d2,0 6a2,1 a3,1 ,
3

S
2
3,4
= 12S12 a2,1 + 12a2,1
+ 6a3,2 6S2 12d2,1 S1 + 2S3
3
6a2,1 a3,2 6d3,2 + 2a4,3 + 4a2,1
+ 12a2,1 d2,1 = 0
(4.46)
J ,S
J
with the obvious replacements. Again, the vanishing of 3,4
can be verified
and similarly for 3,j
by taking the large-0 limit of 3J ,S in Eqs. (4.36) and (4.37) and using j4E and s4E in Eqs. (4.14)
and (4.16), respectively.
We comment that the above universality of 2,0 , 2,1 and 3,0 is directly related to the universality of the BanksZaks fixed point [68,89], which will be proven (to all orders) in Section 4.4
below. This relation follows immediately from Eqs. (3.3) and (3.6) of [67], taking into account
the universality of the BanksZaks critical exponent.
4.3.1. Three-loop examples

Let us now examine the coefficients of the effective-charge functions for the examples considered in Section 4.2. Using Eqs. (4.44) and (4.44) with the explicit coefficients in Appendix A
and Table 3 (or, alternatively, using in (4.36) and (4.37) the explicit coefficients (4.24), (4.32)
and (4.34)) together with the MS -function coefficients, we obtain the following results for the
three-loop coefficients for the effective-charge functions: for the jet function (4.24) we find
2J
with
J
2,2
J
2,1
0 2
J
2,0
0 3

53 3
1
49 1 2
3 + 2 CA +
CF ,
32 2
8
16 4

1 4
5
73 11
J
=
+ 2 +
3 CA 2
2,1
120
48
96
4

235 11
1
3
+
+ 3 + 2 CF CA CF 2 ,
96
4
16
32

301
11
11
7 2
11 2
J
=
CA 3 +
CF CA 2 +
CF 2 CA .
2,0
512 192
64 192
128
J
=
2,2
(4.47)
(4.48)
Another constraint on the jet Sudakov anomalous dimension is the color structure. In particular,
for Nf = 0 we have:

121 121 2 121
J
+
3 CA 2 CF
2 N =0 =
f
768 576
48
94

1729 121
21 2
11 4
+
3 +
+
CA 3 ,
1152
32
128
1440
(4.49)
where the CF 3 and the CA CF 2 components vanishes. In (4.47) the latter implies the following
relation between the coefficients of (4.48):
11 J (CF 2 )
J (C 2 C )
+ 2,0 F A = 0.
2,1
(4.50)
12
This relation involves only the universal components of 2J that coincide with the corresponding
ones in 2S , so it holds for any Sudakov anomalous dimension.
Let us turn now to the soft anomalous dimensions. For the quark-distribution function (4.32)
we find
S
2 QD
with
2,2QD
0

2,2QD =
2,1QD
0 2
2,0QD
0 3
(4.51)

23
97
33 CF +
3 CA
8
32
(4.52)
J
J
and 2,1QD = 2,1
and 2,0QD = 2,0
, given in Eq. (4.48) above. For the DrellYan soft function
(4.34) we find
2SDY
with
SDY
=
2,2
SDY
2,2
0

SDY
2,1
0 2
SDY
2,0
0 3
(4.53)

23
65
33 CF + CA
8
32
(4.54)
SDY
SDY
J
J
and 2,1
= 2,1
and 2,0
= 2,0
, given in Eq. (4.48) above.
The constraint on the color structure of the soft anomalous dimensions is more stringent than
for the jet: they are maximally non-Abelian. In particular, for Nf = 0 we have:

523
11 4
17 2 121
SDY
2 N =0 =
+

3 +
CA 3 ,
f
288 288
48
1440

85
11 4
17 2 121
S
+

3 +
CA 3 ,
2 QD N =0 =
(4.55)
f
32 288
36
1440
where the CF 3 , the CF 2 CA and the CF CA 2 components all vanish. Thus, beyond (4.50), which is
S is universal:
obviously realized, there is an additional relation, namely the CF component of 2,2
144 J ,S (CF CA 2 ) 12 J ,S (CA CF ) 23
(4.56)
2,1
=
33 .
121 2,0
11
8
This general property is of course realized in the examples above, Eqs. (4.52) and (4.54). Thus,
S distinguishes between different soft functions.
at this order, only the CA component of 2,2
S (CF )
2,2
Incidentally, the numerical values of 2J 11 and 2S are also not far; for example for Nc = 3 and
S
Nf = 4 (where = 0.7392) one gets: 2J 1.954, 2 QD 1.734, and 2SDY 1.443.

11 The corresponding value quoted in [68] is incorrect.
95
As anticipated, the numerical values of 2J and 2S are all negative, allowing for a perturbative infrared fixed point. It thus becomes interesting to examine higher-order corrections to the
effective-charge functions.
4.3.2. Four-loop examples
In case of the jet function and the soft function associated with DrellYan production we do
have sufficient information to determine the corresponding four-loop coefficients 3 of Eq. (4.46)
using the Pad approximant predictions for a4 in (4.25). For Nc = 3 and Nf = 4 we get the
following effective-charge functions:

Eucl (2 )
daJ

Eucl 2 2
Eucl 2 2
Eucl 2
1.954 aJ
1 + 0.739 aJ
J
2
d ln

4.003 [1, 2] Pad for a4 Eucl 2 3
+
+
(4.57)
aJ
4.116 [2, 1] Pad for a4
and
(2 )
daSEucl
DY
d ln 2

2 2
2

2 2
1.443 aSEucl
= aSEucl
1 + 0.739 aSEucl
DY
DY
DY

10.532 [1, 2] Pad for a4 Eucl 2 3
+ .
aSDY
10.645 [2, 1] Pad for a4
(4.58)
In case of the jet function effective charge (4.57) the positive 3 coefficient does not support
the fixed-point scenario; a more definite conclusion requires higher orders. However, in the case
of the soft DrellYan effective charge 3 is negative and sufficiently large to bring the fixedpoint value to the perturbative regime: at this truncation order we get a zero for the function
at aSDY 0.46 corresponding to sSDY 0.7. Of course, since there is no perturbative stability
at this order, higher-order corrections may well change this infrared fixed-point value. Further
insight on this issue will be gained in the next section.
4.4. The universal infrared limit of Sudakov effective charges
If we assume that the Sudakov effective charges do indeed admit a finite infrared limit, it
becomes natural to ask what this limit is, and further, to what extent it is universal. The most
natural tool to address these questions is the BanksZaks expansion [7274,9093].
4.4.1. BanksZaks fixed point
Let us briefly recall some terminology. A BanksZaks fixed point [90] is a non-trivial infrared
fixed point, ((2 0)) = 0, occurring in asymptotically free gauge theories with a sufficient amount of matter. In QCD, such a conformal infrared limit characterizes the theory with
Nf /Nc 33/2 where the beta function
ds (2 )/
2 =
d ln 2
2
3
4
= 0 s 2 / 1 s 2 / 2 s 2 / +
(4.59)
has 0 > 0 and 1 < 0, so it is negative for a vanishingly small s (2 ) admitting asymptotic
freedom, but then changes sign again at s (2 ) 0 /1 , implying that theory is conformal in
96
the infrared limit, where the coupling saturates at s (2 0) 0 /1 . In the 0 0 limit the
infrared physics is fully under control of perturbation theory. In real-world QCD both 0 and 1
are positive, and for most physical effective charges so are the higher-order coefficients 2 , 3 ,
etc.; then there is no infrared fixed point.
The BanksZaks expansion [7274,9093] amounts to expressing the infrared fixed-point
value of the coupling as a systematic expansion in (positive) powers of 0 . To construct the
1
expansion one eliminates the Nf dependence in favor of 0 = 11
12 CA 6 Nf , and then solves the
)
= 0 (the tilde denotes a particular renormalization scheme) order-by-order in 0 ,
equation (
using the fact that in the formal 0 0 limit the infrared coupling itself is of the order of 0 .
Using the notation of the previous section for the decomposition of the effective-charge beta
function coefficients, the resulting expansion takes the form (see, e.g., Section 3.1 in [72]):

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

2
3,0 + 31,1 2,0 22,0 3
2
+ 2,1 + 1,1
+ 2
bBZ
1,0
1,0

3
+ 2,2 1,0 + 1,1
+ 32,1 1,1 + 3,1
2
41,1 3,0 + 61,1
2,0 + 42,1 2,0 + 4,0
1,0
3
2 + 5
101,1 2,0
52,0
2,0 3,0
4
+
+ ,
bBZ
3
2
1,0
1,0
(4.60)
where bBZ 0 /1,0 . This expansion converges well for an appropriate number of light quark
flavors (Nf /Nc 33/2) where a fixed point exists (and is realized perturbatively) while for a
realistic number of light flavors, its convergence strongly depends on the observable considered [72]; this is consistent with the expectation that real-world QCD does not have a conformal
infrared limit. Physical quantities may have a finite infrared limit also in the confining phase,
which is not driven by a conformal fixed point and is usually inaccessible to perturbation theory. Yet, it is possible that certain physical effective charges can be described by perturbation
theory down to the infrared limit and then the BanksZaks expansion provides a natural tool to
compute it.
4.4.2. The universal infrared limit of Sudakov effective charges
In general the infrared limit of the coupling is scheme dependent. This is true also for physical effective charges and it is reflected in observable-dependent coefficients of the BanksZaks
expansion starting at NLO [72,91]. Quite remarkably, and in sharp contrast with the general situation, Sudakov effective charges have a common infrared limit that is uniquely determined by
the cusp anomalous dimension; as we shall see, their BanksZaks expansion is universal to all
orders in perturbation theory.
This type of universality has been first observed in Ref. [89] where it was noted that the
BanksZaks expansion of the Sudakov effective charges relevant to deep inelastic structure functions and DrellYan production coincide up to the N3 LO. The issue was further developed in
97
Ref. [68] where a connection was made with the infrared limit of the cusp anomalous dimension.
Here we give a general proof12 of the above statements.
To prove the universality of the infrared limit note first that the relations between the physical
Sudakov anomalous dimensions and the MS ones (3.3) involve a momentum derivative, thus
the function itself. Upon assuming that all effective charges involved admit an infrared fixed
point, Eq. (3.3) immediately implies that in this limit the term containing the function vanishes,
so [68]:

C R J 2 0 = C R S 2 0 = A 2 0 .
(4.61)
The next observation is that the dispersion relations (4.1) and (4.2) necessarily relate the infrared
limit of the Sudakov anomalous dimension with that of the corresponding Sudakov effective
charge:

1 Mink 2
1 Eucl 2
J 2 0 = aJ
0 = aJ
0 ;
0
0

2
1 Mink 2
1
0 = aSEucl 2 0 ,
S 0 = aS
0
0
(4.62)
where the equality of the Minkowskian and Euclidean effective charges [72] follows directly
form (4.4). It is worthwhile emphasizing that this equality holds despite the fact that the function of the Minkowskian effective charge differs from the corresponding Euclidean one by large
2 terms at any order (at three-loops and beyond). The BanksZaks expansion of a Minkowskian
effective charge is identical to the corresponding Euclidean one (Section 3.3 in [72]).
Using in (4.60) the results for the effective-charge beta function of the Sudakov effective
charges, Eqs. (4.44) and (4.46), we find, as implied by (4.61) and (4.62), that all the processdependent coefficients drop out: at each order in the BanksZaks expansion only the cusp
anomalous dimension and the QCD beta function coefficients appear. The result for the first
few orders is:

1 Eucl 2
aJ ,S 0
0

2,0 2
22,0
= bBZ + a2,0 + 1,1
b + 21,1
a2,0 a2,1 1,0
1,0 BZ
1,0

3,0 31,1 2,0 22,0 2 3
+ a3,0 + 2,1 + 1,1 2 +
+
b
1,0
1,0 2 BZ

81,1 2,0 23,0 52,0 2
2
+ 22,1 + 31,1 +
+
a2,0
1,0
1,0 2

32,0
+ (22,0 21,0 1,1 )a2,1 +
+ 31,1 a3,0 1,0 a3,1 2,2 1,0
1,0
+ a4,0 + 1,1 3 + 3,1 + 32,1 1,1
42,1 2,0 + 61,1 2 2,0 + 41,1 3,0 + 4,0

1,0
12 A different argument, which also applies to the more general situation where the infrared fixed points are of a nonperturbative origin and not necessarily related to the cusp anomalous dimension (which may have no fixed point at the
non-perturbative level) is given in [68]: see Eq. (27) there, where instead a relation to the quark form factor is established.
98

101,1 2,0 2 + 52,0 3,0 52,0 3 4
b + ,
1,0 2
1,0 3 BZ
(4.63)
where the coefficients of the cusp anomalous dimension ai,j and the QCD beta function i,j are
defined in (4.40) and (4.41), respectively. Both these quantities are defined in the MS scheme,
but this scheme dependence cancels at each order in the expansion, as both the l.h.s. and the
expansion parameter are physical quantities.
Finally, for Nc = 3 QCD we find the following numerical values:

1 Eucl 2
aJ ,S 0
0

21 947 1 2 2
bBZ
= bBZ +
10 272 4

98 425 175 21 947 2 275
11
3
+

3 + 4 bBZ
17 585 664 20 544
214
80

16
25 685 4
4 955 688 493 5335
+ a4,0 +
4,0 +
+
3 2
107
54 784
422 055 936
856

4 750 031 643 031 75 857 683
4
+
+
3 bBZ
1 083 839 643 648
3 297 312
2
3
4
bBZ 0.3308bBZ
+ 6.902bBZ
+ [a4,0 + 0.14954,0 167.43]bBZ
+ .
(4.64)
For Nf = 4 we obtain the following numerical values at the first three truncation orders:
Eucl (2 0)/ = 0.312, 0.279 and 0.488, corresponding to a Eucl (2 0) = 0.649, 0.582
aJ
0
,S
J ,S
and 1.017, respectively. The higher-order corrections are certainly significant, in particular the
NNLO ones. The results suggest a fixed point value that is marginally perturbative. This is consistent with the conclusion we reached in Section 4.3 by considering directly the fixed-point
solution of the effective-charge beta function.
5. A dispersive representation of the moment-space Sudakov exponent
In Section 4.1 we have written a dispersive representation for the physical Sudakov anomalous
dimensions. Its ingredients have been studied in detail, the characteristic functions in Section 3
and the corresponding effective charges in Section 4. We are now ready to use these tools to
construct a dispersive representation of the Sudakov exponent based on the definitions in Section 2. We proceed as follows: in Section 5.1 we present a short derivation of the main result,
a general dispersive representation of the Sudakov exponent. Then, in Section 5.2 we present
explicit results for the moment-space characteristic functions entering the Sudakov exponent in
various examples and analyze their limits. Finally, in Section 5.3 we return to the general discussion by considering in detail the role of virtual corrections in the large-N limit. This discussion
elucidates the interpretation of the dispersive formula.
5.1. The Sudakov exponent: A general dispersive representation
The crucial step here is going from momentum space, where the real-emission contribution
is finite and can be considered separately, to moment space, where an infrared singularity is
generated, which requires cancellation with virtual corrections. This cancellation is incorporated
in the Sudakov exponent of Eq. (2.7), or equivalently in (2.12) or (2.15).
99
It is convenient to consider first the Sudakov evolution equation obtained through the logarithmic derivative of (2.15):
d ln Sud(m2 , N)
= CR
d ln m2
dR(m2 , r)
dr N r
er
e
r
d ln m2

= CR

dr N r
er J rm2 S r 2 m2
e
r

= CR

dr N r
er J rm2 CR
e
r

0

dr N r
er S r 2 m2 ,
e
r
(5.1)
where in the second line we have inserted the derivative of the momentum-space kernel (3.2) that
admits infrared factorization. Owing to the cancellation between the real and virtual terms for
r 0, the jet and soft contributions to (5.1) are separately infrared finite. This explains the
third line. The purely real Laplace integral,

dr N r 2
J rm ,
e
r
however, is infrared divergent. Here, the dispersive integral becomes handy. Substituting into
(5.1) the dispersive representation of the Sudakov anomalous dimensions according to (4.1)
and (4.2), respectively, we obtain:

d 2

d ln Sud(m2 , N) CR
dr N r
r
=
e
e
J m F J (/r) F J (0)
2
0
r

d ln m
0

dr N r
er
e
r

d 2 2
S m FS /r F S (0) ,

(5.2)
where, as usual, = 2 /m2 . Next, changing the order of integration and defining the momentspace characteristic functions by the Laplace transform of the momentum-space ones:
GJ (N ) =
dr N r
FJ (/r),
e
r

GS N 2 =
dr N r 2
FS /r ,
e
r
we obtain an elegant dispersive representation of the Sudakov evolution equation:

d ln Sud(m2 , N) CR
d 2
=
J m GJ (N ) GJ () + ln N
2
0

d ln m
0
(5.3)
100

d 2 2
S m GS N GS () + ln N ,

(5.4)
where the ln N terms originate in the terms proportional to F J (0) and F S (0) in (5.2), where we
have evaluated the trivial integral

dr N r
er = ln N
e
r
and used Eq. (3.27) setting F J (0) = F S (0) = 1. We have explicitly written the jet and soft
contributions to the r.h.s. of (5.4) as two separate integrals, in order to emphasize the fact that
they are separately finite: the combinations

GJ (N ) GJ () + ln N and GS N 2 GS () + ln N
(5.5)
are each falling as a power of ; indeed the expansion of the characteristic functions at small
arguments starts as
GJ (N ) = ln(N ) + J1 E + O(N ),

1
1/2

1
GS N 2 = ln N 2 + S1 E + O N 2
(5.6)
,
2
2
where S1 and J1 are the log-moments defined in (4.12). These expansions are most easily derived
using Eq. (6.19) below. Explicit examples will be given in Section 5.2.
Because jet and soft contributions to the r.h.s. of (5.4) are separately finite, evolution equations similar to (5.4) can be written separately for the jet and soft Sudakov factors (see, e.g.,
Eqs. (8) and (22) in Ref. [54]) whose product is Sud(m2 , N). The definition of such factors, however, requires to introduce a factorization scale. This will not be necessary in the derivation we
present below, where we shall combine the two dispersive integrals in (5.4) into one.
Note that integrating Eq. (5.4) by parts we obtain an alternative expression in terms of the
Sudakov effective charges:

d Mink 2
d ln Sud(m2 , N) CR
m GJ (N ) GJ ()
=
aJ
2
0

d ln m
0

d Mink 2 2
m GS N GS () .
a
S
(5.7)
The final step in deriving a dispersive representation of the exponent is to integrate the evolution equation (5.4). Quite conveniently, the entire m2 dependence of the r.h.s. in (5.4) is through
J (m2 ) and S (m2 ). Using (4.3) we can therefore readily integrate (5.4) getting:

2

d Mink 2
CR
m GJ (N ) GJ () + ln N
Sud m , N = exp
aJ
0

0

,
a Mink m2 GS N 2 GS () + ln N
(5.8)
S
101
Table 4
Summary of results for the moment-space characteristic functions GJ ( = N ) and GS ( 2 = N 2 ) of some inclusive
distributions
Process
Moment-space characteristic function, GJ ( )
2
(1 + 2 4 ) Ei(1, ) + ( 4 34 )e
Process
Moment-space characteristic function, GS ( 2 )
1 + [ cos() sin()](Si() ) [ sin() + cos()] Ci()

2
2
2
2
DrellYan; gg Higgs
K0 (2)
e+ e jets C parameter (c = C/6)
ln() 2 F
E
4
8 3 2
e+ e jets thrust (t = 1 T )
Ei(1, )
1, 1
2
3 , 3 , 2 16
2 2
1, 3
3 F
2
+ 192
3 2
2, 2, 52
2

16
which is the main result of this section. Explicit results for the moment-space characteristic
functions GJ ,S in a few examples will be computed and analyzed in the next section; the final
expressions are compiled in Table 4.
The exponent (5.8) is finite owing to the exact cancellation of logarithmic singularities to
any order in perturbation theory; there are two levels of cancellation: first between real and
virtual terms within the separate jet and soft parts, and second between these terms. The latter
cancellation mixes the infrared and ultraviolet in an interesting way. To see this note that for
0 the convergence of the separate jet and soft parts in (5.8) follows from the convergence
of (5.4) (the power-like falloff of the combinations in (5.5) above). On the other hand, in contrast
with (5.4), for the integrals corresponding to the jet and the soft parts in (5.8) are
not separately finite. For the GJ ,S terms convergence for is guaranteed by the Laplace
weight in (5.3). The ultraviolet divergence is therefore entirely due to the ln N term, identifying
its origin as the cusp singularity. In conclusion, the ln N terms are required for the separate
jet and soft integrals to converge for 0, but render each of them divergent for .
Mink (m2 )
Nevertheless, the exponent as a whole is finite owing to the fact that aSMink (m2 ) and aJ
are equal at leading and at next-to-leading order (see Section 4.1) and therefore their difference
behaves as s3 (2 ) at large 2 .
Mink (m2 ) =
Note that a somewhat different picture emerges in the approximation where aJ
Mink (m2 ) = a Mink (m2 ), which is valid to NLL accuracy (see (4.39)) or in the large- limit.
aJ
0
GB
In this case, the ln N terms cancel, and we obtain

Sud m2 , N NLL

2

d Mink 2
CR
m
GJ (N ) GJ () GS N GS () .
= exp
(5.9)
a
0
GB
0
Here each of the terms is separately finite for , but the 0 singularities cancel in two
levels: first between real and virtual, and second between the jet and soft terms.
5.2. Moment-space characteristic functions and their expansions
Let us examine now some explicit examples. Having determined the functional form of the
momentum-space characteristic functions (Table 2) we can readily compute the Laplace integrals (5.3) to determine the moment-space ones. The final results are collected in Table 4.
102
In what follows we give further details on the calculation of the characteristic functions and
address the convergence of the dispersive integral in the exponent (5.8) considering the asymptotic behavior of these functions at small and large . In the DrellYan case we extend the
discussion and compare the dispersive formalism developed here with the joint resummation
formalism or Ref. [17]. For DrellYan we also derive an alternative representation of the exponent in terms of a Euclidean integral [68]. Such a representation does not exist in the other
examples.
5.2.1. The jet function
Let us begin by constructing an explicit expression for the moment-space characteristic function of the jet, GJ (N 2 /m2 ), based on the momentum-space one given in (3.44). According
to (5.3) we have:
GJ ( ) =
2
dr N r
FJ
e
r
m2 r

=
0

1 2 2
dr N r
1 2
2
2
r > /m 1
e
r
2 m2 r 2 m2 r

dw w
1 1
11
e
1
w
2 w 2 w2

= 1+

2
3
Ei(1, ) +
e ,
2
4
4 4
(5.10)
where N2 /m2 and where the exponential-integral function is defined as usual by

Ei(p, )
d
e
.
p
(5.11)
Let us now examine the convergence of the integral in (5.8) for the jet part. First, for
there is convergence for each of the separate GJ ( ) terms owing to the behavior of this function
for large :

3 1
1
GJ ( )
(5.12)
+ O 3 e .
2
2
In the 0 limit there is convergence in (5.8) only owing to cancellations between the
separate terms. We obtain

3
GJ (N ) = E ln(N ) ln()
4

1
+ 2 E + ln(N ) + ln() N + O 2 ln ,
(5.13)
2
making the combination GJ (N ) GJ () + ln N power-suppressed as required for the integral
to converge.
103
5.2.2. The quark distribution function (B decay)

Let us consider now the soft function associated with the quark distribution in an on-shell
heavy quark. Using in (5.3) the explicit expression for the momentum-space characteristic function in (3.44) we obtain:

GSQD 2 =
2
dr N r
FSQD
e
r
m2 r 2

=
0
2

d
2
+
e
1 + 2 (1 + 2 )2
i i

1 1 i
+ e Ei(1, i) + ei Ei(1, i) +
e Ei(1, i) ei Ei(1, i)
2 2

4

1
= +
cos() sin() Si()
sin() + cos() Ci(), (5.14)

2
2
2
2
where we have defined = N /m = N and changed variables to 2 = r 2 /. The

exponential-integral function is defined in (5.11) above, and
=

Si()
dt
sin(t),
t

Ci() E + ln() +

dt
cos(t) 1 .
t
(5.15)
Let us examine the convergence of the integral in (5.8) for the soft part. For we find

2
1
GSQD () 2 + O 3 ,
(5.16)
which guarantees convergence separately for each of the GSQD terms. As expected, in the 0
limit there is convergence in (5.8) only owing to cancellations between the terms. We obtain

1 1
GSQD N 2 = E ln(N ) + ln() + N 1/2

2 2
4

1
N 2 + N 3 3/2 + O 2
(5.17)
4
24
making the combination GS (N 2 ) GS () + ln N power-suppressed.
QD
QD
5.2.3. The DrellYan soft function

Next, let us consider the soft function associated with DrellYan production. Using in (5.3)
the expression for the momentum-space characteristic function (3.49), we obtain:

GSDY 2 =

0
2
1
dr N r
d
1
2/
FSDY
= K0 (2),
e
=
e
r
2
m2 r 2
1
0
(5.18)
104
where we have defined = N /m = N and changed the integration variable from r to =

2
4/r . Here K0 is the modified Bessel function of the second kind.
To examine the convergence
of the dispersive integral in the exponent (5.8) for , let us
expand (5.18) at large = N . The result is:

1
1 1
1
1
+O 2
.
GSDY 2 = e2
(5.19)
2
16
Obviously, the integrals over the GSDY terms converge for large .
Next, consider the 0 limit. Here the expansion yields

2
GSDY N 2 = ln(N ) E + ln(N ) E + 1 N 2 + O N 2
(5.20)
making the combination GSDY (N 2 ) GSDY () + ln N power-suppressed at small , as required

for convergence. Note that only even powers of appear in this expansion.
Let us return now to the physical DrellYan kernel. The Laplace equivalent of (3.14), which
differs from it only by O(1/N) terms for N , is
DY (Q2 , N) d ln HDY (Q2 , 2 )
d ln Gq q (N, Q2 , 2F ) d ln Sud
F

+
+ O(1/N),
d ln Q2
d ln Q2
d ln Q2
(5.21)
with
DY (Q2 , N)
d ln Sud
d ln Q2

dr N r
= 2CF
(r < 1) SDY Q2 r 2
e
r
0
2CF
=
0
=
2CF
0
2CF
0
2CF
=
0

0

2
2
dp 2
p
p
F
(0)
F
S
SDY
SDY
p 2 DY
Q2 r 2

2 2
2
2
dp 2
N p
1
p
S p GSDY
FSDY
+ E + ln(N )
2
p 2 DY
Q2
Q2

2 2
2
dp 2 Mink 2 d
N p
1
p
S
p
a
F
G
SDY
DY
S
2
2
2
DY
2
p
d ln Q
Q
Q2

!
1

dp 2 Mink 2 d
p
a
K0 2 N 2 p 2 /Q2 p 2 /Q2 < 1/4
2
p 2 SDY
d ln Q2
2 ln
2CF

0

dr N r
(r < 1)
e
r
1+
2
Q
/4

1 4p 2 /Q2
ln p 2 /Q2
2
!

!

dp 2 Mink 2 d
2
2
2
p
K0 2 N p /Q ln Q2 /p 2 ,
a
p 2 SDY
d ln Q2
(5.22)
105
where in the second line we used the dispersive representation of SDY from (4.2), in the third (see
derivation in Section 5.3) we performed the integration over r and substituted the Laplace transform of F DY according to (5.3) (and used F DY (0) = 1), in the fourth we performed integrationby-parts and in the fifth we inserted the explicit expressions from (5.18) and (3.48). Finally, in
the last line we have changed the upper limit of integration over the K0 term, which modifies the
result by an exponentially
small correction at large N , and neglected the first term in the square

brackets, 2 ln((1 + 1 4p 2 /Q2 )/2), which generates O(2 /Q2 ) power-suppressed corrections.
Note that the logarithmic derivative with respect to Q2 cannot be pulled out of the p 2 integral in (5.22), which would then diverge. This divergence is the usual collinear divergence
associated with the evolution of the quark distributionin the Sudakov exponent of infrared
and collinear safe distributions, Eq. (5.37), it cancels with the jet subprocess. In order to obtain
DY (Q2 , N, 2 ) one would obviously need to introduce a factorization scale.
Sud
F
It is interesting to compare our final result for the large-N DrellYan kernel (5.22) with the
equivalent expression derived in the joint resummation formalism [17]. To this end consider the
result for the Eikonal cross section, Eq. (48) in [17], where dimensional regularization (D =
4 2) is used to regularize the collinear divergence in the exponent:
q q (N, bQ, )
Q2

d 22 k
2
2
q q k 2 , k 2 + k
, 2 , k 2 + k
p2
12
0

!
!

ibk
K 2 N 2 p 2 /Q2 ln
Q2 /p 2 ,
e
0
= exp 2
dp
dk
(5.23)
where b is the Fourier conjugate to the transverse momentum of the system, and q q is the
web [71] corresponding to radiation from the annihilating lightlike quark and antiquark, defined at fixed transverse momentum. One immediately recognizes that upon considering the
small b limit, corresponding to the total cross section (or to a case where the resummation of
transverse momentum logarithms is unimportant compared to threshold logarithms, bm N ),
Eq. (5.23) reproduces Eq. (5.22) up to13 terms that are finite at N , provided one makes the
following identification14 :
2
2 2
p = lim aSMink
p , ,
aSMink
(5.24)
DY
DY
0
where

Mink 2
2
CF aSDY p , ,
0
p2
22

d
k
2
2
dk 2
q q k 2 , k 2 + k
, 2 , k 2 + k
p2 .
12
(5.25)
13 Finite terms are generated by a different upper limit on the momentum integration, Q2 in (5.23) versus Q2 /4 in
(5.22), which amounts to a different split of the virtual corrections between the Sudakov factor and the hard function that
multiplies it.
14 In the 0 limit, the result is independent on the dimensional regularization scale . For example, in case of a
single gluon emission (Eq. (45) in [17]), the r.h.s. in (5.25) is lim0 CF (s /)(4 eE 2 /p2 ) /p2 .
106
This gives a direct diagrammatic interpretation to the Minkowskian effective charge in terms of
webs.
Eq. (5.22) is the standard dispersive representation of the kernel that involves the Minkowskian
effective charge. In the DrellYan case, the exists an alternative Euclidean representation [68].
To derive it one defines a Euclidean characteristic function by the following dispersion relation:
GSDY () =

0
dy
G Eucl (y).
( + y)2 SDY
(5.26)
Substituting this expression for GSDY in the soft part of (5.4) (after integration by parts) one gets:
d ln SudDY (Q2 , N)
d ln Q2
2
2

Eucl 2

d Mink 2 dy
2CF
yQ2
G
yN GSEucl
=
a
(y)
DY
0
y (1 + 2 )2 SDY
2 SDY
yQ2
0
2CF
0
dy Eucl 2 Eucl 2 Eucl

yQ GSDY yN GSDY (y) ,
a
y SDY
(5.27)
where in the second line we used the relation between the Minkowskian and the Euclidean effective charges, Eq. (4.4).
According to (5.18), the explicit result for GSEucl
(y), defined in (5.26), is [68,94]
DY
1
(y) = J0 (2 y ),
GSEucl
(5.28)
DY
2
where J0 is the Bessel function of the first kind.
Finally, note that when using the Euclidean representation (5.27), power-like infrared sensitivity (i.e., infrared renormalons) is directly reflected in the expansion of the Euclidean characteristic function near zero. This should be contrasted with the Minkowskian representation where only
non-analytic terms in the expansion are associated with renormalons (see Section 7 below). In the
case of DrellYan production, only even powers of N /m appear [52], as can be seen [68] from
the fact that J0 is an even function of its argument. Equivalently in the Minkowskian representation the non-analytic terms in the expansion of K0 appear with even powers (see Eq. (5.20))this
has been pointed out in [17,95].
5.2.4. The soft functions in event-shape distributions
Let us compute the moment-space characteristic functions of the thrust and the C parameter.
The case of the thrust is simple: F St (z) = (z < 1). Using (5.3) we obtain
1
GSt 2 =
2

0
dz /z
1
FSt (z) =
e
z
2
1
dz /z
= Ei(1, ).
e
z
(5.29)
The case of the C parameter is slightly more complicated. Let us write the momentum-space
characteristic function as

1
FSc (z) = (z < 4) + (z > 4) 1

1 4/z
107
and split the integral accordingly:

1
GSc 2 =
2
dz /z
FSc (z)
e
z
1
=
2
4
dz /z
+
e
z

4

dz /z
1
e
1
z
1 4/z

dx x/2
1
e
1
x
1 x2
0

3
2
3
1, 1 2
1, 2
F
F
E ln()
+
=
3 2 3 3
3 2
16
,
,
2
2,
2, 52
4
8
192
2 2
1
= Ei(1, /2) +
2

16 .
(5.30)
Let us examine the convergence of the integral in (5.8). For we find exponential suppression for the thrust and power suppression for the C parameter:

1
1
+O 2 ;
GSt 2 e

2
1
GSc 2 2 + O 4 .
(5.31)
In either case each of the GS terms in (5.8) converge.

For 0 we find

1
1
GSt 2 ln() E + 2 + 3 + O 4 ;
4
18
2

3
1
2
GSc ln() E + +
(5.32)
+ O 4 ,
4
8
192
making the combinations GS (N 2 ) GS () + ln N power-suppressed at small , as required for
convergence.
5.3. The large-N limit, strict factorization and finite terms
As discussed in Section 2, the Sudakov exponent in a given process is unique as far as the
terms that diverge at large-N are concerned, but it is subject to convention in what concerns finite
terms for N as well as any O(1/N) corrections. This has been illustrated by introducing
three definitions for the Sudakov factor in Eqs. (2.7), (2.12) and (2.15). Our purpose here is to
identify, in the framework of the dispersive approach, the unique large-N limit of the Sudakov
exponent, which in contrast with Eq. (5.9) above, involves two scales only: m2 /N and m2 /N 2
(such a strict scale separation is a must in an effective field theory framework). We will also show
that a convention-dependent part that violates this strict scaling is in fact necessary to render the
Sudakov exponent finite.
5.3.1. The large-N limit: Disentangling N -dependent and constant terms
Finite terms for N are generated by purely virtual corrections, which are proportional
to (r) in momentum space. Purely virtual corrections are partially accounted for within the
108
exponent and partially in the hard function H (m2 ) in (2.6) or (2.14). This split depends on the
convention15 adopted, except for the requirement that all infinities are assigned to the exponent
where they cancel against the integral over the real-emission part. Let us make this dependence
parametric by generalizing Eq. (2.12) though a modification of the cutoff value:
m2 , N, r0 = exp CR
Sud

2
dr N r
(r < r0 ) R m , r ,
e
r
(5.33)
with

m2 , N, r0 +
N m2 .
N m2 = H m2 , r0 Sud
(5.34)
For r0 = 1 the constant term O(N 0 ) in the exponent matches the original Mellin-transform definition (2.7).
It is straightforward to repeat the derivation of the dispersive representation with this definition. Proceeding as in (5.1) and (5.2) we obtain:
2 , N, r0 )
d ln Sud(m
d ln m2

dR(m2 , r)
dr N r
= CR
(r < r0 )
e
r
d ln m2
0
CR
=
0

0
d

dr N r
(r < r0 )
e
r

J m2 F J (/r) F J (0) S m2 F S /r 2 F S (0)

CR
d 2
=
J m HJ (N, , r0 ) S m2 HS (N, , r0 ) ,
0

(5.35)
with

HJ (N, , r0 )

dr N r
(r < r0 ) F J (/r) F J (0)
e
r
= GJ (N ) FJ (/r0 ) + ln(r0 ) + E + ln(N ),

dr N r
HS (N, , r0 )
(r < r0 ) F S /r 2 F S (0)
e
r
0

1
= GS N 2 FS /r02 + ln(r0 ) + E + ln(N ),
2
(5.36)
15 However, convention independence, as well as a neat separation between purely real (assigned to the Sudakov exponent) and purely virtual contributions can be achieved [6870] by taking two derivatives of the exponent.
109
where in the last step in (5.35) we performed the integration over r substituting GJ (N ) and
GJ (N 2 ) according to (5.3) and used (3.26) to write
r0
dr
FJ (/r) = FJ (/r0 );
r
r0
dr 2 1 2
FS /r = FS /r0 .
r
2
Note that HJ ,S in Eq. (5.36) are infrared finite owing to the subtraction of (r < r0 ) in the
square brackets; they are ultraviolet finite because of the differences F J (/r) F J (0) and
F S (/r 2 ) F S (0), respectively. Finally, performing the ln m2 integration in (5.35) we obtain
the result for the Sudakov factor:

m2 , N, r0
Sud

2

d Mink 2
CR
Mink
m HJ (N, , r0 ) aS
m HS (N, , r0 ) .
a
= exp
(5.37)
0
J
0
Explicit results for HJ ,S (N, , r0 ) can be readily obtained by substituting GJ ,S and FJ ,S of

Tables 4 and 1, respectively into Eq. (5.36).
Let us now check the convergence of the integral over in Eq. (5.37). Consider first the 0
limit. Using the small- expansion of GJ ,S in (5.6) and that of FJ ,S in (4.17) we observe that the
combinations in Eq. (5.36) are power-suppressed at 0. Next consider the limit. GJ ,S
and FJ ,S are (at least) power-suppressed at large . The remaining terms ln(r0 ) + E + ln(N ) are
common to the jet and soft part, and therefore the integrand is proportional to the difference
Mink (m2 ) a Mink (m2 ), which behaves as 3 (2 ) at large 2 .
of the two effective charges, aJ
s
S
Thus, the integral (5.37) is well defined.
Let us now isolate the N -dependent terms in the exponent. This can be elegantly done by
considering the limit r0 . Expanding HJ ,S (N, , r0 ) at large r0 we obtain:
HJ (N, , r0 1) =
GJ
(r+v)
(N ) + O(/r0 ),

(r+v)
N 2 + O /r02 ,
HS (N, , r0 1) =
GS
(5.38)
where we used the expansion of FJ ,S at small arguments, Eq. (4.17), and defined
(r+v)
GJ

(N )

dr N r
1 F J (/r) F J (0)
e
r
= GJ (N ) + ln(N ) J1 + E ,

2

dr N r
(r+v)
N
1 F S /r 2 F S (0)
e
GS
r
0

1
1
= GS N 2 + ln N 2 S1 + E ,
(5.39)
2
2
where the superscript (r + v) indicates that in contrast with GJ ,S of (5.3), which consists solely
of real emission contributions, through the subtraction of 1 in the square brackets (5.39) includes
(r+v)
also virtual corrections. This guarantees convergence for r 0. Note that
GJ (N ) and
(r+v)
(N 2 ) converge for r owing to the subtraction of F J (0) and F S (0), respectively.
G
S
110
Thus, similarly to HJ ,S (N, , r0 ) these objects are well defined. Based on these definitions
and (5.3) it is straightforward to recover the expressions in terms of GJ (N ) and GS (N 2 ),
respectively.
(r+v)
(N ) and
GS(r+v) (N 2 ) depend on a
In contrast to HJ ,S (N, , r0 ) the functions
GJ
single argument. Using (5.38) we therefore find that the N -dependence of the jet function in
(5.37) involves only the jet mass scale m2 /N while that of the soft function only the soft scale
m2 /N 2 . This is the infrared factorization property we have already encountered in Section 3
when working in momentum space. We also observe, however, that despite the fact that the
integral in (5.37) is well defined for any finite r0 , the r0 limit does not exist: by taking
r0 inside the integral a divergence is generated for . Thus, the O(/r0 ) and
O(/r02 ) contributions that distinguish HJ ,S (N, , r0 ) from their simple r0 limits (thus
violating strict factorization) are in fact necessary for convergence.
Having separated the N -dependent terms from the cutoff-dependent constant terms, Eq. (5.37)
can be elegantly written as follows:

m2 , N, r0
Sud

CR
d Mink 2 (r+v)
= exp
m
GJ (N )
FJ (/r0 )
aJ
0

0

1
(r+v)
m2
GS
N 2
FS /r02
,
2

Mink
aS
(5.40)
where for the r0 dependent terms we used the following definitions

/r0

FJ (/r0 )

dy
FJ (y) F J (0) = FJ (/r0 ) + ln(/r0 ) J1 ,
y

2
/r02
FS /r0

dy
FS (y) F S (0) = FS /r02 + ln /r02 S1 .
y
(5.41)
0
(r+v)
(r+v)
Note that
GJ (N ) and
GS
(N 2 ) in (5.39) are essentially GJ (N ) and GS (N 2 ), respectively, minus the leading terms in their expansions, Eq. (5.6). Similarly
FJ (/r0 ) and
FS (/r02 ) in (5.41) are FJ (/r0 ) and FS (/r02 ) minus the leading terms in their expansions,
Eq. (4.17). Consequently, the convergence of (5.40) for 0 is guaranteed for each of the separate terms, which all fall as 1 in this limit. In contrast, the convergence for involves
cancellation between all the terms.
5.3.2. Re-derivation of the exponent using strictly factorized components

We saw that upon removing the regulator r0 one recovers strictly factorized functions that
fully capture the N -dependent terms of the Sudakov exponent at large N . We can now repeat
the derivation of the dispersive representation of the exponent with no constant terms, Eq. (5.8),
using strictly factorized components. To this end let us define the formal jet and soft Sudakov
111
exponents:
EJ m /N = CR
2

dr N r
1 J rm2
e
r
(5.42)
and
ES m /N

= CR

dr N r
1 S r 2 m2 ,
e
r
(5.43)
which include both real and virtual corrections. Owing to the virtual term (1) these expressions
are infrared finite, but since the integration extends to infinity they are ultraviolet divergent. In
Appendix C we show that these formal definitions correspond to taking the large N -limit of the
finite-N physical evolution kernel, where the limit is taken with fixed N 2 /m2 for the jet and
fixed N 2 2 /m2 for the soft function. The characteristic functions of Eq. (5.39) are then identified
with the corresponding large-N limits of the full finite-N characteristic functions computed in
Ref. [49].
The Sudakov evolution equation, Eq. (5.1), can be expressed in terms of finite combinations
of these quantities, where the ultraviolet divergences cancel out. We have

d ln Sud(m2 , N) 2
(5.44)
= EJ m /N EJ m2 ES m2 /N 2 ES m2 .
d ln m2
Substituting into (5.42) and (5.43) the dispersive representation of the Sudakov anomalous dimensions according to (4.1) and (4.2), respectively, we obtain the moment-space representations:
CR
EJ m /N =
0
ES m /N
CR
=
0
d 2 (r+v)
J m
GJ (N ),

d 2 (r+v) 2
N ,
S m
GS

(5.45)
(r+v)
(r+v)
(N 2 ) have been defined in (5.39).
where, as usual, = 2 /m2 , and
GJ (N ) and
GS
Note that here the integrands are well defined and integrable near 0 but, as expected based
on the definitions (5.42), the integrals diverge for .
Integrating (5.45) by parts we obtain an alternative form in terms of the Sudakov effective
charges:
CR
EJ m /N =
0
ES m /N
CR
=
0
d Mink 2 (r+v)
m
GJ (N ),
a
J
d Mink 2 (r+v) 2
m
GS
N .
a
S
(5.46)
Note that Eqs. (5.42) and (5.43) on the one hand, and (5.46) on the other hand, give an example
Mink ,
of two different integral mappings of different running Sudakov couplings (J , S and aJ
,S
112
respectively) onto the same Sudakov exponents targets EJ ,S mentioned in the introduction section.
Next, substituting in Eq. (5.44) the dispersive representations (5.45) of the jet and soft Sudakov exponents, we get the dispersive representation of the Sudakov evolution equation:

d 2 (r+v)
d ln Sud(m2 , N) CR
(r+v)
=
J m
GJ (N )
GJ ()
2
0

d ln m
0

d 2 (r+v) 2
(r+v)
N
GS
() .
S m
GS

(5.47)
0
(r+v)
Using the expressions for
GJ ,S in Eq. (5.39) this yields Eq. (5.4). Finally, integration over
ln m2 gives:

2

d Mink 2 (r+v)
CR
(r+v)
m
GJ (N )
GJ ()
Sud m , N = exp
aJ
0

0

2 (r+v) 2

(r+v)
Mink
a
(5.48)
m
G
N
G
() ,
which yields Eq. (5.8). Thus, we have recovered the final result of Section 5.1 in terms of the
(r+v)
characteristic functions
GJ
,S that include virtual corrections. This derivation elucidates the
origin of the ln(N ) terms in Eq. (5.8). Eq. (5.48) also exhibits explicitly the fact that the integral
converges for 0: here each term is separately power-suppressed in this limit. Convergence
in the limit relies on cancellation between the terms.
6. Relation with the Borel representation of the Sudakov exponent
The Borel formulation of Sudakov resummation has been the subject of much theoretical work
over the past few years [5,22,23,32,43,54]. It led to a successful phenomenology in a range of
applications. The purpose of the present section is to study in depth the relation between this
formulation and the dispersive one, which was presented in the previous sections. We will show
that the two provide different ways of summing up the same set of radiative corrections. In the
next section we shall address potential differences at the power level.
6.1. Borel representation of the anomalous dimensions
Let us begin by introducing the scheme-invariant Borel representations of the Euclidean effective charges (4.5):
Eucl
aS

=
2
du T (u)
2
u

B aSEucl (u),

Eucl 2
=
aJ

du T (u)
0
2
2
u
Eucl
(u),
B aJ
(6.1)
113
where 2 is defined in MS. We emphasize that the Borel integrals do not necessarily converge
for arbitrary 2 : all that is required is that they exist for sufficiently large 2 . As usual, while the
integrand in (6.1) is scheme invariant [64], the separation between T (u) and the Borel function
Eucl ](u) itself is scheme dependent. In practice (see, e.g., Eq. (2.8) in [41]) it is convenient
B[aJ
,S
to use the t Hooft scheme where T (u) absorbs [64] the dependence on the two-loop coupling
= 1 /0 2 ,
T (u) =
|u|u eu
.
(1 + u)
(6.2)
Note that in the large-0 limit T (u)|large 0 = 1 and

Eucl

5
B aJ
(u) large = B aSEucl (u) large = e 3 u .
0
0
(6.3)
According to (4.4) the Borel representations of the timelike couplings corresponding to (6.1)
are obtained by analytic continuation:
Mink
aS

=
2 u
sin u Eucl
du T (u)
B aS (u),
2
u
0
Mink
aJ

=
2 u
sin u Eucl
du T (u)
B aJ (u).
u
2
(6.4)
Substituting these expressions into (4.1) and (4.2), respectively, we obtain the Borel representation of the physical Sudakov anomalous dimensions of (3.3) [5,54]:

1
J 2 =
0
1
=
0

0

2 u
du T (u)
BJ (u),
2

2
du T (u)
2
u
(6.5)
BS (u),
where
sin u Eucl
B aJ (u) BJ (u),
u
sin u Eucl
BS (u) =
B aS (u) BS (u),
u
BJ (u) =
(6.6)
with

BJ (u) = u
0

BS (u) = u
0
dy u
y FJ (y) =
y
dz u
z FS (z) =
z
0

(1)k uk
dy u
y FJ (y) =
Jk ,
y
k!
k=0
(1)k uk
dz u
z FS (z) =
Sk ,
z
k!
k=0
(6.7)
114
where F J ,S (z) are the characteristic functions defined in the large-0 limit (3.23) and Jk and Sk
are the corresponding log-moments defined in (4.12). Thus, the expansion of BJ ,S (u) also provides a convenient way to extract the log-moments. Obviously, the same information about higher
orders and infrared sensitivity (power corrections) contained in the characteristic functions F J ,S
is encapsulated by the corresponding Borel functions BJ ,S (u). Table 2 summarizes the explicit
results for BJ ,S (u) in various inclusive distributions.
Note that

sin u 5 u
e 3 BJ ,S (u).
BJ ,S (u) large =
(6.8)
0
u
Although BJ ,S (u) have singularities at certain integer values of the Borel variable, these do not
generate renormalons at large 0 in the anomalous dimensions J (k 2 ) and S(k 2 ) because of the
sin(u) factor associated with the timelike nature of these quantities.
Eucl ](u) and thus, in virtue of (6.6), also B
We further expect that B[aJ
J ,S (u) are renor,S
malon free at finite 0 . Absence of renormalons in anomalous dimensions is expected on general
grounds,16 but a formal proof in the non-Abelian case is still absent and would be very important.
When using the Borel formulation, the deviation of BJ ,S (u) from their large-0 limits is
computed order-by-order in perturbation theory. It amounts to a multiplicative modification of
the large-0 result [5] by a factor VJ ,S (u), see, e.g., Eqs. (19) and (26) in [54], which we can
now readily identify as
VJ ,S (u) =
Eucl ](u)
B[aJ
,S
Eucl ](u)|
B[aJ
large 0
,S
Eucl
53 u
= B aJ
.
,S (u)e
(6.9)
This gives a new perspective on the physical meaning of this function.

6.2. The Sudakov exponent
To demonstrate the correspondence between the dispersive representation of the Sudakov exponent and the Borel one, let us now re-derive the Borel formula for the exponent (cf. Eq. (27)
in [54]) following the steps of Section 5 above. We will then present a second derivation starting
from the dispersive formula (5.8).
6.2.1. Direct derivation
Inserting the Borel representation of the Sudakov anomalous dimensions in (6.5) into (5.1)
we obtain:
d ln Sud(m2 , N)
d ln m2

dr N r
dr N r
= CR
er J rm2 CR
er S r 2 m2
e
e
r
r
0
CR
0

0

dr N r
er
e
r
2 u
du T (u)
BJ (u)
rm2
16 Note however we deal here with physical anomalous dimensions, for which this expectation might not be realized.

dr N r
er
e
r
2
du T (u) 2 2
r m
115
u
BS (u) ,
(6.10)
where the two terms corresponding to the jet and soft contributions to the r.h.s. are separately
finite. Here one can change the order of integration using u as an infrared regulator. Using

dr N r u
r = N u (u),
e
r
we obtain an elegant Borel representation of the evolution equation:

2 u

d ln Sud(m2 , N) CR
=
du
T
(u)
BJ (u) (u) N u 1
2
2
0
d ln m
m
0

2 u
2u
du T (u)
BS (u) (2u) N 1 ,
m2
(6.11)
where, as before, the two Borel integrals are separately finite, for u 0. We note that the terms
proportional to N u and N 2u in Eq. (6.11), respectively, are the Borel representations of the formal
jet and soft Sudakov exponents (5.42) and (5.43), where u serves as a regulator.
It is straightforward to see the relation with Eq. (5.4)17 :

d 2
J m GJ (N ) GJ () + ln N

=

d Mink 2
m GJ (N ) GJ ()
aJ

=
2 u

du T (u)
BJ (u) (u) N u 1
2
m
(6.12)
and

d 2 2
S m GS N GS () + ln N

=

d Mink 2 2
m GS N GS ()
aS

=
2 u

du T (u)
BS (u) (2u) N 2u 1 .
m2
17 Note that upon using (6.4), Eqs. (6.12) and (6.13) can be derived from (6.16) below.
(6.13)
116
Note that in (6.11), (6.12) and (6.13) one must assume some prescription for the Borel singularities at positive integer and half integer values of u.
The final step in the derivation of the Sudakov factor is to integrate (6.11) over ln m2 . As
in (5.8) this requires to combine the two integrals over u in (6.11). The result (cf. Eq. (27)
in [54]) is:
2 u

2
u

du
CR
B
Sud m , N = exp
(u)
(u)
N
1
T (u)
J
0
u
m2
0

2u

BS (u) (2u) N 1 .
(6.14)
This is the Borel equivalent of (5.8).
6.2.2. Derivation based on the dispersive formula
An alternative route is to insert the Borel representation of the Minkowskian couplings (6.4)
directly into Eq. (5.8). This yields:

Sud m2 , N
2 u

d
sin u Eucl
CR
du T (u)
B aJ (u) GJ (N ) GJ () + ln N
= exp
2
0

u
m
0

2 u

sin u Eucl 2
du T (u)
.
B aS (u) GS N GS () + ln N
u
m2
(6.15)
Upon combining the two Borel integrals and changing the order of integration the ln N terms
cancel and the integrals over the GJ ,S terms take the form:

d u
1
GJ (N ) GJ () = N u 1 (u)BJ (u),

u

d u 2
1
GS N GS () = N 2u 1 (2u)BS (u),

u
(6.16)
where we have computed the integrals using the relations (5.3) and (6.7), which together yield a
direct relation between the moment-space characteristic functions GJ ,S () and the Borel functions BJ ,S (u) (all functions are defined in the large-0 limit):

"

/y F (y)
GJ () = 0 dy
d u
1
J
y e
GJ () = (u)BJ (u) (6.17)

" dy u

u
BJ (u) = u 0 y y FJ (y)
0
and

"
1 dz / z
FS (z)
2 0 z e
"
u
BS (u) = u 0 dz
z z FS (z)
GS () =

0
d u
1
GS () = (2u)BS (u).

u
(6.18)
117
It is straightforward to check this correspondence in the explicit examples considered above.

Using (6.16) in (6.15) we immediately recover the known form of the exponent in (6.14).
Finally, it is useful to write the inverse relations to (6.17) and (6.18):
1
GJ () =
2i
1
GS () =
2i
i
i
i
i
du u
(u)BJ (u),
u
du u
(2u)BS (u).
u
(6.19)
Using these relations with the u 0 expansion of BJ ,S (u) in (6.7) we can easily derive the
O( 0 ) term in the expansion of GJ ,S () in (5.6) by applying the Cauchy theorem.
7. Power corrections in the Sudakov exponent
As discussed in the introduction infrared renormalons are present in the Sudakov exponent.
This is evident in Eq. (6.14). As always, infrared renormalons reflect sensitivity to large-distance
dynamics and indicate the presence of non-perturbative power corrections. From (6.14) one can
immediately deduce that these power corrections (1) exponentiate together with the logarithms;
and (2) are enhanced at large N : the jet and soft contributions scale as integer powers of
N2 /m2 and N/m, respectively. Such parametrically-enhanced power corrections appear exclusively through the Sudakov factor and have a significant impact on the distribution in the
threshold region.
The renormalon technique offers here a unique window into the non-perturbative side of the
problem. Renormalon analysis allows us to find the specific pattern of power corrections [54]:
in a given process only certain power corrections appear while others are absent. This pattern of
power corrections is linked with the symmetry properties of the source of soft gluon radiation,
and it is very much process dependent. The absence of specific power terms can be seen as an
extension of the notion of infrared finiteness beyond the logarithmic level, i.e., infrared safety
at the power level.
In the Borel formulation the presence of infrared renormalon ambiguities is transparent: the
Borel integral in (6.14) is obstructed by simple poles at integer and half-integer values of u,
except where the Borel transforms of the Sudakov anomalous dimensions BS (u) and BJ (u)
of (6.6) vanish. It is natural then [22,54] to take the principal value of the integral in (6.14) as
a definition of the perturbative sum, and use the renormalon ambiguities as a basis for parametrization of power corrections.
In the dispersive approach, power-like sensitivity is somewhat less obvious. Consider for example the soft (S) Sudakov factor as can be obtained from (5.4) through integration-by-parts (an
example is provided by Eq. (5.22) for the DrellYan case):

d ln Sud(m2 , N)
d Mink 2 2
CR
m GS N GS ()
=
aS

2
0

d ln m
S
0

EMink m2 , N .
(7.1)
Only non-analytic terms in the small- behavior of the Minkowskian characteristic function
GS () are related with renormalons [45,49,58,61,62], indicating corresponding power correc-
118
tions. In a Euclidean formulation,

CR
d ln Sud(m2 , N)
ym GS N y GS (y) ,
=
a

2
y S
d ln m
0
S
(7.2)
such as the one of Eq. (5.27) in the DrellYan case, the relation is simpler: all the terms in the
small-y expansion of aSEucl are related with renormalons. Although a Euclidean formulation (7.2)
does not exist in generalamongst the examples considered in this paper it exists only in the
DrellYan caseit is instructive to consider it first since, in contrast with the Minkowskian one,
it facilitates separating between the infrared and ultraviolet contributions in a straightforward
manner by splitting the momentum integral:

2

2

d ln Sud(m2 , N)
IR
UV
m , N ; 2I + EEucl
m , N ; 2I EEucl m2 , N
= EEucl
(7.3)

2
d ln m
S
where
IR
EEucl
, N ; 2I
CR
yI

ym GS N y GS (y) ,
a
y S
(7.4)
where I Q/N is a momentum cutoff and yI = 2I /m2 . Using (7.2) and (7.3) the remainder is given by
UV
EEucl
, N ; 2I
CR
=
0

ym GS N y GS (y) .
a
y S
(7.5)
yI
One can now expand the Euclidean characteristic function at small y under the integral (7.4), getting the power corrections. According to Eq. (5.26) the Euclidean characteristic function GSEucl (y)
is the integral over the discontinuity of the Minkowskian one:

d GSEucl (y)
1
= Im GS (y i0) ,
d ln 1/y

d 1
Im GS ( i0) .
GSEucl (y) =

(7.6)
This is why only non-analytic terms in the expansion of the Minkowskian characteristic function
GS () are relevant.
The extension to the general case, where (7.2) does not exists, is based on (7.3) where
IR (m2 , N ; 2 ) is still defined by (7.4) but E UV (m2 , N ; 2 ) does not have a representation of
EEucl
I
I
Eucl
the form (7.5); it must instead be defined by first reverting to a Borel representation [45,58]. Since
in this section we are focussing on power corrections, we do not write down the explicit form
of the latter, but instead refer the reader to [45,58]. There is one point that needs to be stressed,
however: Eq. (7.3) will not coincide with the Minkowskian dispersive integral (7.1) unless the
EuclideanSudakov coupling is causal.
A causal coupling admits the dispersion relation (4.4): it is analytic except for the cut on
the timelike axis. In this case the entire first sheet of the complex momentum plane is mapped
onto a compact domain in the complex coupling plane whose boundary is the image of the
119
time-like axis (the simplest example where this is realized is a two-loop coupling with negative
1 /0 , see Fig. 2 in Ref. [74]). The three- and four-loop coefficients computed in Section 4.3 in
a few examples, have been found to have opposite signs to the one- and two-loop ones, opening
the possibility that a non-trivial infrared fixed-point appears in these effective charges within
perturbation theory. While the stability of this fixed point and the presence or absence of Landau
singularities obviously depends on higher orders, a causal structure becomes possible. In this
scenario the dispersion representation (4.4) holds for the perturbative coupling itself, which does
not need to be modified by any power terms, and then

EMink m2 , N = EEucl m2 , N .
EMink (m2 , N) still differs from the principal value of the Borel sum. But this difference is then
entirely due to power corrections that are related to infrared renormalons [58,9799], and genuinely reflect sensitivity to long-distance dynamics. The physical distribution is expected to differ
from this perturbative result by power corrections whose pattern follows the renormalons.
In contrast if aSEucl (k 2 ) has Landau singularities in the first sheet of the complex k 2 plane, as
indeed occurs in the large-0 limit (see Appendix B), then

EMink m2 , N = EEucl m2 , N ,
and, although EMink (m2 , N) in (7.1) is still finite and yields a unique result, the dispersion integral (7.1) differs from the corresponding Borel sum by power corrections that are not related
with infrared renormalons [45,49,58,61,62].
In the presence of Landau singularities in the Euclidean couplings, the Minkowskian couMink (k 2 ) are still finite in the infrared limit. One may therefore be tempted to consider
plings aJ
,S
the result of (5.8) or even the approximation of (5.9) as a non-perturbative model.18 One should
be aware, however, that this result involves additional power corrections that are not associated
with long-distance dynamics. As we recall in Appendix B, using the dispersive integral amounts
to analytization of the coupling that removes the Landau singularity. This is why the dispersive integral differs from the Borel sum by power terms that are not all related to renormalons.
In the large-0 limit this difference is given by the second term in the curly brackets in (B.10).
An explicit example is given in Appendix D.
In the following we derive the small- expansion of the Minkowskian characteristic function,
and then deduce from it the corresponding expansion of its discontinuity, namely of the Euclidean
characteristic function.
The relations (6.19) can readily be used to see that the singularities of (u)BJ (u) and
(2u)BS (u) at integer or half integer values of u, u = k/2 (k is a positive integer), are associated with terms of order O( k/2 ) in the expansion of the characteristic functions GJ ,S ()
at small . Non-analytic terms scaling as k/2 in this expansion are related to infrared renormalons at u = k/2 and therefore indicate O((/m)k ) power corrections in the exponent. Owing
to the N -dependence of the exponent (5.8) through GJ (N ) and GS (N 2 ), power corrections
associated with the jet and soft subprocesses will scale as (N )k/2 and (N 2 )k/2 , respectively.
To see the precise relation between Borel singularities in (6.14) and non-analytic terms
in (5.8), let us first recall the Borel singularity structure and then translate it into the terminology
of the dispersive approach. The universal (u) and (2u) factors give rise to poles in (6.14).
18 A related model has been proposed in [39], where the fixed-logarithmic-accuracy formula for the exponent has been
rewritten in terms of the analytic coupling (B.8).
120
The functions BJ ,S (u) multiplying them may nevertheless vanish, cancelling the potential pole.
According to (6.6) this depends on the properties of BJ ,S (u) as well as the Borel transform of
Eucl ](u). We assume, as in previous work [54],
the corresponding Sudakov effective charge B[aJ
,S
that the latter is innocuous: in the large-0 limit it is given by (6.3), so it has no poles nor zeros
at integer or half-integer locations, and we assume that singularities or zeros will not develop
there beyond this limit. In particular, absence of singularities is equivalent to the (already stated)
Eucl (2 ) are free of renormalons. In contrast
assumption that the Sudakov effective charges aJ
,S
BJ ,S (u) may have poles at integer values of u, as well as zeros at integer and half integer values
of u. Note that the assumption above implies that the zeros of BJ ,S (u) present at large-0 , which
cancel some potential renormalon singularities in the exponent, are still present in the full theory.
Thus, according to (6.19) non-analytic terms in the characteristic function emerge in the following cases:
In GJ (), when BJ (u) has a pole, i.e., when (u)BJ (u) has a double pole. This occurs
at u = 1 and u = 2 (see Table 2). Indeed, upon expanding the explicit result for GJ () we
get:

3
1
1
GJ () E ln()
+ 2 E ln()
4
2
2

3 1
1
5
+ + E + ln() 2 3 + O 4 .
(7.7)
8 4
4
72
The non-analytic terms ln() and 2 ln() in this expansion indicates, respectively,
N2 /m2 and N 2 4 /m4 power corrections in the exponent (5.8); see, e.g., [5,57,68,96].
In GS (), at all half integer values of u, u = k/2 where k is an odd number,
except
when BS (u) vanishes. In this case GS () develops a square-root singularity ( )k . Thus,
GS (N 2 ) in (5.8) gives rise to O((N /m)k ) corrections.
In GS (), at integer values of u (u = k/2 with even k) where BS (u) has a pole (so
(2u)BS (u) has a double pole). In this case GS () develops a logarithmic singularity:
k/2 ln(), leading to O((N /m)k ) corrections in (5.8).
The three cases above are precisely the ones where BJ ,S (u)|large 0 in Eq. (6.8) does not vanish,
i.e., where the singularities in (u) and (2u) are not cancelled.
Altogether the soft characteristic function has the following generic structure at small :

1
1
GS () = ln() + S1 E +
Ck ( )k
2
2
k odd

k/2
+
Ck ln() + analytic terms
(7.8)
k even
or
k
1 k
Ck ( )k +
Ck k/2 ln() + analytic terms.
GS () = +
2
2
2
k odd
k even
Thus, upon extracting the discontinuity according to (7.6) one obtains:

d GSEucl (y)
1
= Im GS (y)
d ln 1/y
(7.9)
121
k sin(k/2)
k
Ck
( y )k
Ck y k/2
2
(7.10)
k even
k odd
or
GSEucl (y) =
Ck

(1)(k1)/2 k
Ck y k/2
ck y k/2 .
( y) +
k even
k odd
(7.11)
k=1
The coefficients Ck of the first few non-analytic terms in specific examples are summarized
in Table 5. The table clearly demonstrates that the renormalon singularity structure is process
dependent.
The small-momentum expansion of the Euclidean characteristic function (7.11) can be readily
used inside the integral in Eq. (7.4) to obtain:
IR
EEucl
, N ; 2I

dy k/2 Eucl 2
CR

ck N k 1
y aS ym .
0
y
k=1
0

yI
(7.12)
Mk
Obviously, in the infrared region the coupling may differ from its perturbative part. This is encapsulated in the moments. In this way the moment Mk = O((I /m)k ) can be used to parametrize
non-perturbative power corrections of order O((N /m)k ) with any k in the Sudakov exponent.
Recall that the k = 1 moment (with a choice of I = 1 or 2 GeV) has been used in [49] and
following work (see, e.g., [2629]) to parametrize power corrections to event-shape distributions. The framework presented here allows one to identify this coupling to higher orders in
perturbation theory as well as to parametrize in a similar manner the corrections that scale as
O((N /m)k ) using the higher moments.
Eucl have a causal analyticity structure one can directly paraAssuming that the couplings aJ
,S
metrize these functions in the infrared consistently with the dispersion relation (4.4) as well as
their known ultraviolet evolution of Eqs. (4.36) and (4.37); a simple one-loop model of this kind
Mink one
has been considered in Ref. [100]. Upon taking the time-like discontinuity to obtain aJ
,S
can readily evaluate the Minkowskian integral in the exponent, Eq. (5.8), getting the all-order
sum already including power corrections of the renormalon type. An even simpler possibility
Mink in the infrared, consistently
would be to parametrize directly the Minkowskian couplings aJ
,S
Table 5
Summary of results for the first few coefficients of non-analytic terms in the small- expansion of the moment-space
characteristic functions GJ ( = N ) and GS ( 2 = N 2 ) of some inclusive distributions (cf. Tables 4 and 2)
ln( )
2 ln( )
3 ln( )
4 ln( )
5 ln( )
6 ln( )
1/2
1/4
2 ln()
4 ln()
6 ln()
SQD
/4
/24
1/24
/160
1/360
1/36
Process
Process
DrellYan; gg Higgs
SDY
1/4
e+ e jets C parameter
Sc
/4
/192
/20480
e+ e jets thrust
St
1/18
1/600
122
with their ultraviolet evolution.19 In this way one can set up a simple yet fully consistent powercorrection phenomenology without ever dealing with individual powers and without introducing
any momentum cutoff.
8. Conclusions
We have presented a general formalism for Sudakov resummation based on dispersion integrals. The expression for the Sudakov factor in infrared and collinear safe distributions involving
a jet function (J ) and a soft function (S) is summarized by Eqs. (5.48) and (5.8) or (5.40). This
formulation consists of two ingredients [68,69]: the first, GJ ,S , are characteristic functions that
are computed analytically in the large-0 limit and encode information on power corrections; the
second, aJ ,S , are Sudakov effective charges that are defined order-by-order in perturbation theory and encapsulate the non-Abelian nature of the interaction. In what concerns the soft function
(S) both these ingredients are process-dependent owing to the fact that large-angle soft gluon
radiation depends on the spacetime geometry of the hard partons that radiate. In contrast, there
is a unique jet function (J ) characterizing soft and collinear radiation from a jet with a fixed invariant mass. Explicit results for the characteristic functions in a variety of processes have been
compiled in Table 4, and the corresponding Sudakov effective charges have been computed and
analyzed in Section 4.
The Sudakov effective charges are directly related to the physical anomalous dimensions.
They can be computed to any order based on the conventional anomalous dimensions defined by
dimensional regularization. Furthermore, comparison with the joint resummation formalism [17]
leads to a direct diagrammatic interpretation of the Sudakov effective charges in terms of webs,
see (5.25) above.
The dispersive approach presented here provides a realization of DGE [22,23,54,57]: it goes
beyond the resummation of Sudakov logarithms per se by incorporating an internal all-orders
resummation of running-coupling corrections. This resummation guarantees renormalizationgroup invariance. Owing to the enhancement of subleading logarithms that are associated with
the running of the coupling [22,57], this additional resummation leads to a significant improvement over the conventional approach to Sudakov resummation where a renormalization-schemedependent truncation is performed, guided solely by the logarithmic accuracy criterion. Beyond
the perturbative level, the present approach facilitates a systematic analysis of power corrections based on renormalon ambiguities which reveal themselves through the discontinuities of
the characteristic function GJ ,S .
We have shown that there is a direct correspondence between the scheme-invariant Borel
formulation and the dispersive one, and derived all-order relations between their ingredients.
The two formulations capture the same set of radiative corrections. Yet, at the power level they
provide different regularizations of the sum. As far as renormalon singularities are concerned,
this difference does not pose any difficulty: the two regularizations are in principal equivalent, as
they differ by power corrections of the same parametric form that need be introduced to account
for genuine non-perturbative effects. In contrast, if the Sudakov effective charges have Landau
singularities, as occurs for example in the large-0 limit, the dispersion relation is violated. Then,
the dispersion integral differs from the Borel sum by additional power corrections that are not
19 In this case one should make sure that the resulting Euclidean coupling is causal at the perturbative level, and does
not involve extra unwelcome power corrections, as in the APT example!
123
related with renormalons. This complication does not arise if the Sudakov effective charges have
a causal analyticity structure, making them consistent with the dispersion relation.
The dispersive approach offers a convenient way to parametrize the non-perturbative power
corrections exposed by the renormalons by means of integrals over the coupling in the infrared
region [49]. We find that in the context of Sudakov resummation, the infrared-finite-coupling
approach by Dokshitzer, Marchesini and Webber is of special interest: in contrast with the general
situation here the couplingthe Sudakov effective chargecan be systematically identified to
any order. Moreover, as discussed in Section 4.3, there are indications that these effective charges
may reach a finite limit in the infrared already within perturbation theory. Remarkably, despite
the fact that the evolution of the Sudakov effective charges becomes process-dependent at three
loops, the infrared limit itself turns out to be universal: it depends only on the cusp anomalous
dimension.
A particularly attractive example from the point of view of the dispersive approach is that
of DrellYan or Higgs production. This has two aspects: first, there exists a Euclidean representation (5.27) where the identification of large-distance effects is transparent, and second the
relevant Sudakov effective charge admits an infrared fixed point already at the perturbative level
as both the three-loop and four-loop coefficients of the effective-charge beta function are negative, see Eq. (4.58).
A significant effort has been put in the past decade in examining the hypothesis of a universal
[2628,49] infrared-finite coupling in the context of event-shape distributions. Experimental data
were primarily used to test the assumed universality of the first power moment of this effective
coupling. Indeed, overall this assumption is supported by data [29]. Despite the different resummation formalism utilized in these studies compared to the one presented here, the notion of the
infrared-finite coupling, and the way it relates to power corrections, are the same. So far this effective coupling has only been identified to NLO, and its universality has not been established
theoretically.20 This has now changed: the present formalism uniquely identifies the effective
charges relevant for Sudakov resummation to all orders in perturbation theory, and shows that
their universality does not extend beyond the NLO. It nevertheless shows that their infrared limit
is universal. Approximate universality of power corrections, determined by the first few power
moments of the coupling, is expected as a by-product.
DGE, in its Borel formulation, has been successfully applied to phenomenology in a range of
processes [54]. Most importantly, it extends the range of applicability of resummed perturbation
theory closer to threshold, far beyond what can be achieved by a conventional, fixed-logarithmicaccuracy approach. This should be attributed to two main factors: first the additional resummation
performed, and second the possibility to identify correctly the pattern of power corrections,
especially the first one or two renormalon singularities. The main challenge has been to find
an effective parametrization of power corrections that captures the dependence on N /m for
N m/, where the power expansion completely breaks down. Even if one assumes that nonperturbative corrections are directly proportional to the corresponding renormalon residues, it
remains difficult to control them in practice since the residues vary going beyond the large-0
limit. Indeed, little is known about the Borel transform B[aSEucl ](u) in (6.14) beyond the large-0
limit and away from the vicinity of the origin. When using the dispersive formulation instead, the
problem translates into the parametrization of the Sudakov effective charge aSEucl (2 ) itself in the
20 See however Refs. [101,102] (as well as earlier work, especially [2022]) where universality of the leading power
correction (the shift) in a class of event-shape distribution is established independently of any infrared-finite-coupling
hypothesis.
124
infrared region. This should be an easier function to constrain, especially if aSEucl (2 ) tends to a
finite limit in the infrared already within perturbation theory, making it a slowly varying function.
Constraining the coupling over the infrared region may be further helped by the universality of
its infrared limit. It should be nevertheless stressed that the dispersive approach is advantageous
only upon assuming that the relevant couplings are causal, free of Landau singularities. In the
opposite case using the Borel formulation [54] is more straightforward.
Finally, parametrization of the Sudakov effective charges over the infrared region provides
an alternative to the shape-function approach, one that matches smoothly onto the perturbative
description, does not require the introduction of an explicit infrared cutoff, and most importantly,
is better constrained.
Acknowledgements
E.G. wishes to thank Bryan Webber for useful discussions. E.G. is grateful to the cole Polytechnique for hospitality when this work was initiated, the University of Edinburgh particle
physics group for hospitality in its final stages, and finally the HEP group in Cavendish Laboratory for a very enjoyable time over the past four years. G.G. benefited from early discussions
with Yuri Dokshitzer, Pino Marchesini and George Sterman.
Appendix A. Coefficients of the Sudakov anomalous dimensions in MS
The coefficients of the cusp anomalous dimension, corresponding to (2.9) are known to threeloop order [77]:
a1 = 1,

5 CA 1 2
,
a2 = +
3
0 3 12

1
253 5 2 7
1
55
a3 = +
33 CF +
+ 3 CA
3 0
16
72
18
2

7
11 4
2 11
1
605 11
2
+ 3 CA CF +
3 +
CA .
+ 2
192
4
18 18
4
720
0
(A.1)
The coefficients of the jet-function anomalous dimension B (associated with an unresolved

jet with a constrained mass) are known to three-loop order [7,70]:
3
b1 = ,
4

2 247 CA 73
5
3
3
CF 2
3 +
3 ,
b2 =
6
72
0 144 2
0 8
32 2
4357 29 2 2
+ 3
b3 =
648
36
3

1
5501 25 2 1
+
+ 3 CF
0
576
32
3

1807 283 2 335
13 4
+
+
+
3
CA
216
648
36
360

1
4
15
3 2 17
1 2
29
2
3
3
5 CF
128 64
16
40 12
4
0 2

55 543
245
17 4
1
15
1
+
+ 2
3
2 3 5 CA CF
6912
32
72
720
24
8

4891
41 4 11 2
29
115 2 605
+
+
3 +
3 5 CA 2 .
10 368 2592
144
2880
72
8
125
(A.2)
The coefficients of the DrellYan anomalous dimension DDY (associated with large-angle soft
radiation from two lightlike partons that annihilate to produce a non-colored heavy object) are
known to three-loop order [8,9,70]:
d1DY = 0,

14 1 2 1 CA
8 7
= + +
+ 3 ,
9
3
2 0
9 2

116
10
1
2345
23 4 517 2 245
DY
2
+ 109 + 3 +
+
+
3
CA
d3 =
81
3
20
180
324
18
216

1711 19
4 2
+
+ 3 +
+
CF
144
3
30
2

1
11 4
485
11 2
371
223 2
2
+
+
5
3
3 CA
1296
360
1296
36
72
20 2

11
18 821 209
11 4
+ 2 +
(A.3)
3
CF CA .
24
1728
36
360
d2DY
The coefficients of the anomalous dimension DQD corresponding to the heavy-quark distribution function [86] as well as the heavy-quark fragmentation function are known to two-loop
order [32]:
d1QD = 1,
d2QD =

1 CA 2 11 9
+
3 .
9
0 12 18 4
(A.4)
Appendix B. Renormalon sum: Borel and dispersive representations

There exist two convenient representations of a dressed gluon which result in two different
formulations of resummation formulae:
The Borel method: The running coupling, which include the effect of dressing, can be written as
sV (k 2 ) s (2 )
1
1
=
=
1 + (k 2 ) 0

0
2 u
V
du T (u)
,
k 2
(B.1)
where k is the gluon momentum and (k 2 ) is the vacuum polarization function, renormalized
at 2 . The coupling is defined in the spacelike region k 2 > 0 and then analytically continued to
the complex momentum plane. The superscript V stands for the V scheme (defined by the poten-
126

5
tial between two heavy quarks) where 2V = 2 e 3 (cf. (4.20)). For one-loop running coupling,
T (u) = 1. In this case one recovers

sV (k 2 )
1
1
(B.2)
=
.

2 /2 )
ln(k
0
one-loop
V
An all-order resummation of running-coupling corrections in a given quantity R(Q2 ) with a
single dressed gluon can thus be achieved by performing the momentum integration with the
modified propagator
1
1
(B.3)
.
k 2 i0
(k 2 i0)1+u
This procedure directly yields the Borel representation of the perturbative sum in the large-0
limit in the form:

u

2
1

du T (u) Q2 /2
B(u).
R Q large =
(B.4)
0
0
0
The dispersive method: The dispersive representation of the dressed gluon [45,49,61,62]
takes the form

1
V (m2 ) dm2
1
1
sV (k 2 ) s (2 )
=
,
(B.5)
1 + (k 2 ) 0
m2 k 2
1 + k 2 /2V
0
where V (V < 0) is the spectral density function, defined by the discontinuity of the coupling
on the timelike axis,
0 V
1 0 s (2 ) Im{(m2 )}
V m2
.
Im s m2 i0 / =
|1 + (m2 )|2
Taking the time-like discontinuity of the one-loop running coupling (B.2) one obtains:

1
V m2 one-loop = 2
.
ln (m2 /2 ) + 2
(B.6)
(B.7)
The dispersive integral (the first term in the curly brackets in (B.5)) then yields21
sAPT (k 2 )
1
=
dm2
1
2
2
2
m k ln (m2 /2V ) + 2
0

1
1
1
=
+
0 ln(k 2 /2V ) 1 + k 2 /2V
(B.8)
that differs from the original one-loop coupling (B.2) by pure power terms that eliminate the
Landau singularity. The second term in the curly brackets in (B.5) cancels this additional term,
and thus restores the Landau pole, making (B.5) consistent with (B.2). Note that we shall be
using (B.5) rather than (B.8): we will not assume anything about the way analytic properties of
physical quantities are eventually restored in (non-perturbative) QCD.
21 The superscript APT on (B.8) stands for analytic perturbation theory [88]; it distinguishes this object from the
ordinary one loop coupling.
127
All-order resummation can be achieved using (B.5) by performing the momentum integration
with a massive gluon propagator:
1
1
2
.
(B.9)
i0
m k 2 i0
This results in the dispersive representation of the perturbative sum in the large-0 limit:

R Q2
k 2
1
=
0
1
=
0
large 0

dm2 2 2 2
V m F m /Q F(0) + F 2V /Q2 F(0)
m2

dm2 Mink 2 2 2
2
2
m F m /Q + F V /Q F(0) ,
a
m2 V
(B.10)
where in the second line we applied integration-by-parts using (3.22) and

d
F m2 /Q2 .
F m2 /Q2 m2
2
dm
Appendix C. Taking the large-N limits of finite-N characteristic functions
In Section 3 we identified the Sudakov limit of the characteristic functions in momentum
space. A similar identification can be done in moment space. Let us now show that the Sudakov
characteristic function in Eq. (5.39) can be recovered as the appropriate N limit of the
moment-space characteristic function of the corresponding physical evolution kernel. We will
also show that the formal jet and soft exponents (5.42) and (5.43), respectively, naturally
emerge as limits of these kernels. Again we illustrate these statements in the examples of deep
inelastic structure functions and the DrellYan cross section.
C.1. Deep inelastic structure functions
The finite-N moment-space characteristic function22
1
GDIS(F2 ) (, N ) =
dx x N1 FDIS(F2 ) (, x)
(C.1)
appears in the dispersive representation (valid in the large-0 limit) of the moment-space evolution kernel K DIS(F2 ) (N, Q2 ) (Eq. (3.6)):

K DIS(F ) N, Q2
large 0
CF
0

d2 2
V GDIS(F2 ) 2 /Q2 , N GDIS(F2 ) (0, N) .
2
(C.2)
(v)
(v)
22 In Ref. [49] G
DIS(F2 ) (, N ) is denoted by FN (), whereas GDIS(F2 ) () and GDY () are denoted by Vs () and Vt (),
respectively. Note also that our normalization of the characteristic functions is half the one in this reference.
128
This is the moment-space version of Eq. (3.35). Let us now split GDIS(F2 ) (, N ) into its real and
virtual components:
(r)
(v)
GDIS(F2 ) (, N ) = GDIS(F2 ) (, N ) + GDIS(F2 ) ().
(C.3)
Considering first the real contribution, one finds [70], using the explicit expression in [49] for the
characteristic function:

dr N r
(r)
(, N ) =
FJ (/r) GJ (N ),
e
lim G
N DIS(F2 )
r
N fixed
0
(C.4)
which is the analogue of the momentum space relation, Eq. (3.45). Taking one derivative, we
deduce:
(r)
lim GDIS(F2 ) (, N ) =
N
N fixed
dr N r
FJ (/r) = GJ (N )
e
r

N (N )2
N 3 N
= 1+
.
Ei(1, N) +
e
2
4
4
4
(C.5)
Let us turn now to the virtual corrections. Considering the same limit the virtual pieces G (v) () as
well as its derivative, G (v) () diverge. Indeed for 0 one gets, using the explicit expressions
in [49]:
1 2
3
2 7
(v)
GDIS(F
()

()
ln
ln()
2)
2
2
3
4
and thus
3
(v)
lim GDIS(F2 ) () ln(N ) + ln(N ) + + O(1/N).
2
N
N fixed
(C.6)
(C.7)
Next, consider the combination GDIS(F2 ) (, N ) GDIS(F2 ) (0, N) occurring in the dispersive representation Eq. (C.2) and decompose it as
(v)

(r)
GDIS(F2 ) (, N ) GDIS(F2 ) (0, N) = GDIS(F
(, N ) + GDIS(F2 ) () GDIS(F2 ) (0, N) .
2)
(C.8)
Using now Eq. (4.43) in [49], one gets
3
lim GDIS(F2 ) (0, N) ln N + E ,
4
(C.9)
and thus, using Eq. (C.7):

lim
N
N fixed

3
(v)
GDIS(F2 ) () GDIS(F2 ) (0, N) = ln(N ) + + E .
4
Using Eq. (C.5) we deduce:

lim GDIS(F2 ) (, N ) GDIS(F2 ) (0, N)
N
N fixed
(C.10)
3
= GJ (N ) + ln(N ) + + E
4

N (N )2
3
N 3 N
= 1+
+ ln(N ) + + E ,
Ei(1, N) +
e
2
4
4
4
4
and, comparing with Eq. (5.39), we find:

(r+v)
lim GDIS(F2 ) (, N ) GDIS(F2 ) (0, N) =
GJ (N ).
129
(C.11)
(C.12)
N
N fixed
Thus, as announced, the jet Sudakov characteristic function,

GJ (N ), is identified as the
large-N limit with fixed N of the full finite-N characteristic function of F2 . Note also that,
taking the derivative of Eq. (C.11), one gets [68]
(r+v)
lim GDIS(F2 ) (, N ) = GJ (N ) 1,
(C.13)
N
N fixed
where 1 is the virtual contribution. Finally, taking the limit N inside the integrand of
Eq. (C.2) (which yields an ultraviolet divergent integral), and comparing with Eq. (5.45), one
obtains [70]:

lim K DIS(F2 ) N, Q2 large = EJ Q2 /N large .
(C.14)
0
N
Q2 /N fixed
C.2. DrellYan
In a similar way the finite-N characteristic function in the DrellYan case
1
GDY (, N ) =
dx x N1 FDY (, )
(C.15)
appears in the moment-space equivalent of (3.39):

CF
K DY , Q2 large =
0
0

d2 2 2 2
GDY /Q , N GDY (0, N) .
V
2
(C.16)
Splitting the characteristic function into the real and virtual contributions we have:
(r)
(v)
GDY (, N ) = GDY (, N ) + GDY ().
(C.17)
For the real emission part, the large-N limit with fixed N 2 exists, and yields [70]:

dr N r
(r)
lim GDY (, N ) = 2
FSDY /r 2 2GSDY N 2 ,
e
N
r
N 2 fixed
0
(C.18)
which is the analogue of the momentum space relation, Eq. (3.46). Taking a derivative we get:
lim
N
N 2 fixed
(r)
GDY (, N ) = 2

0

dr N r 2
FSDY /r = 2GSDY N 2 = 2K0 (2N ).
e
r
(C.19)
130
(v)
The virtual contribution to the characteristic function GDY () as well as its derivative, G (v) ()
diverge in this limit. The small- expansion is [49]:
1
3
2 7
(v)
GDY () ln2 () ln() +
,
2
2
6
4
(C.20)
so

3
(v)
GDY () 2 ln(N ) + ln N 2 + + O(1/N).
2
lim
N
N 2 fixed
Decomposing the difference entering (C.16) according to

(v)

(r)
GDY (, N ) GDY (0, N) = GDY (, N ) + GDY () GDY (0, N) ,
(C.21)
(C.22)
and using the relation

GDY (0, N) = 2GDIS(F2 ) (0, N),
(C.23)
which follows from the results in Section 4.6 of [49] (and reflects the fact that the DrellYan
cross section in the DIS scheme is an infrared and collinear safe quantity) one gets:
3
lim GDY (0, N) 2 ln N + 2E + O(1/N).
N
2
Using Eq. (C.21), one thus finds
(v)

lim GDY () GDY (0, N) = ln N 2 + 2E ,
N
N fixed
which implies
lim
N
N 2 fixed

1

GDY (, N ) GDY (0, N) = 2 GSDY N 2 + ln N 2 + E
2

= 2 K0 (2N ) + ln(N ) + E ,
and, comparing with Eq. (5.39), we find

(r+v)
lim GDY (, N ) GDY (0, N) = 2
GSDY N 2 .
N
N 2 fixed
(C.24)
(C.25)
(C.26)
(C.27)
Thus, as announced the large-N limit of the full finite-N characteristic function with fixed N 2
(r+v)
reproduces the Sudakov characteristic function
GSDY (N 2 ). Note that in accordance with the
discussion following Eq. (5.22) above, the explicit result in Eq. (C.26) is identical to the function
appearing in Eq. (54) in [17] at b = 0. Note also that upon taking one derivative23 of Eq. (C.26)
one obtains [68]

lim GDY (, N ) = 2GSDY N 2 1,
(C.28)
N
N 2 fixed
23 Note that G
2
SDY (N ) does not satisfy the analogue of the Laplace representation Eq. (5.3), owing to the singular
behavior of the momentum space characteristic function in the DrellYan case.
131
where 1 is the virtual contribution. Finally, taking the large-N limit inside the integral in
Eq. (C.16) (which is yields an ultraviolet divergent integral) and comparing with Eq. (5.45) above
one finds [70]:

lim
(C.29)
K DY N, Q2 large = ESDY Q2 /N 2 large .
0
N
Q2 /N 2 fixed
Appendix D. Renormalons and Landau singularities: Example

Let us consider for example the power terms distinguishing between (5.9) and (6.14) in the
case of inclusive B decays under the assumption that the effective charges are given by (4.39),
namely they do have Landau singularities. We obtain:

ln Sud m2 , N
ln Sud m2 , N
Dispersive
2 /m2
CF
=
0
Borel

d
GJ (N ) GJ () GS N 2 GS () 2 =2
GB

2
2

GB
GB
CF 1 2
=
5
N 1 N ln(N ) (N 1) ln
+
E
2
20 2
m
m2

soft

+O
iCF
+
20

+O
jet

2GB 2
m2
2

2 1/2

2GB 3/2
GB
GB
1 3
(N 1)
N 1
(N 1)
18
m2
m2
m2

soft

2GB 2
,
m2
jet
soft
(D.1)
where the first few terms in the expansion have been computed explicitly using (5.13) and (5.17).
The first line summarizes the leading real power term that contributes to the (unambiguous)
difference between (5.9) and the principal value of the Borel sum (6.14); these terms are not
related with renormalons. The second line summarizes the imaginary power terms which represent the ambiguity of the Borel sum owing to infrared renormalons. They provide an indication
of the potential size of genuine non-perturbative effects. Note that these terms originate in the
non-analytic terms in the small 2 expansion of the characteristic function. Obviously, there is
one-to-one correspondence between these non-analyticities and the Borel singularities in (6.14).
The latter have been discussed in [4043,54].
It is straightforward to identify the origin of the different terms in (D.1), as indicated below
each term: terms that scale at large N as N /m are associated uniquely with the soft (quark
distribution) function, while those scaling as N 2 /m2 with the jet function. It is evident that
both these classes of corrections contribute to both the real and the imaginary parts of (D.1). In
practice, power corrections on the soft scale are important while those on the jet-mass scale can
usually be neglected. Let us therefore consider the former in some detail:
132
The leading power term, O(N/m), represents the u = 1/2 renormalon ambiguity. This
ambiguity has been shown [40] to cancel in the physical spectra with the ambiguity in defining the b quark pole mass, or equivalently = MB mb , where MB is the meson mass and
mb is the b quark pole mass.
In Refs. [4043,54] the principal value Borel sum was used to define both the Sudakov factor
and the pole mass, eliminating the ambiguity. In Appendix E we provide a definition of the
pole mass that suites the dispersive approach.
As a consequence of the Landau singularity in aSEucl (k 2 ) assumed above,24 a term of
O((N/m)2 ) appears as a real contribution to the difference between (5.9) and the principal value of the Borel sum (6.14). This is a parametrically large correction that directly
influences the width of the computed spectra. It should be explicitly subtracted out when using the dispersive technique: it cannot be absorbed into the definition of any non-perturbative
parameter, as there is no corresponding renormalon ambiguity.
Finally, power terms of O((N/m)p ) where p is an integer p 3, appear as renormalon
ambiguities. Corresponding non-perturbative corrections are expected to appear as a consequence of the interaction between the b quark and the light degrees of freedom in the meson
[40], distinguishing the quark distribution is the meson from that in an on-shell b quark.
These power corrections have been parametrized in [41]. As far as these corrections are concerned the dispersive representation of the exponent (or its quark distribution function part)
provides an alternative definition of the perturbative component, which is a priori as good as
the definition based on the principal value Borel sum.
To conclude, we have seen that
Genuine non-perturbative effects are related to renormalon singularities. As usual, the same
conclusions with regards to such corrections can be reached using the dispersive and the
Borel formulations. As far as the regularization of the renormalons is concerned the two definitions are equivalent in principal: they differ by power corrections of the same parametric
form one needs to introduce in order to parametrize genuine non-perturbative effects. Obviously, the non-perturbative parameters need to be defined according to the regularization
used.
Quite independently of this, when the effective charges have Landau singularities the dispersive integral differs from the Borel sum by an additional set of computable power terms
that are not related with renormalons; these include parametrically important contributions,
O(2 N 2 /m2 ).
Appendix E. Definition of the pole mass in the dispersive approach
In Refs. [4043,54] the principal value Borel sum was used to define both the Sudakov factor
and the pole mass, eliminating the ambiguity (see, e.g., Eq. (4.4) in [43]). The use of the dispersion integral to define the Sudakov exponent requires a corresponding regularization of the
Importantly, the relevant mass, which we shall refer to as the dispersive pole
pole mass (or ).
disp.
mass, mb , is uniquely fixed by the formalism, and similarly to the principal value pole mass,
24 This correction does not appear if a Eucl (k 2 ) has a causal analyticity structure, see [58] and the discussion in Section 7
S
above.
133
it can be accurately determined based on some short distance mass (e.g., the MS mass). The relation between the dispersive pole mass and the principal value pole mass becomes transparent
upon recalling the cancellation mechanism of the u = 1/2 renormalon. For concreteness let us
refer to the B Xs example.
By combining the Laplace weight of the inverse-Mellin transform in (2.4) with the Sudakov
factor (5.9):

2E N1
Sud m2 , N
mb
u=1/2

2E
MB
N1
2

I 2
N disp. CF
d Mink 2 N 2 2
exp
a
,
mb
4
2 SQD
m2b
(E.1)
where have used the relevant (square-root) term in the expansion of the characteristic function
GSQD (N 2 ) in (5.17) and omitted all other terms, which are irrelevant for the u = 1/2 renormalon. We consider the integration up to an arbitrary scale I that should be above ; the
specific scale choice will not affect the final result.
The equivalent integral in the scheme-invariant Borel formulation can be obtained by inserting
(6.4) into (E.1), which yields:

2E N1
Sud m2 , N
mb
u=1/2

N1
2E
N PV I N CF

exp
MB
mb
mb 2

2 u

du
sin u Eucl
PV
(E.2)
T (u)
B aSQD (u) ,
1 2u 2I
u
0
where we have chosen the principal value prescription for both and the u = 1/2 ambiguity in
the quark distribution function.
The key point is that the object considered in (E.1) and (E.2) is unambiguous, so the relation
between the two definitions of the pole mass can simply be read off comparing the two equations:
disp.
disp. PV = mPV
b mb
2
CF
= mb
4
I
0
+ I

d2 Mink 2 2
a
2 SQD
m2b
CF
PV
2

0
2 u
du
sin u Eucl
T
(u)
B aSQD (u).
1 2u 2I
u
(E.3)
Moreover, the r.h.s. of Eq. (E.3) does not depend on the cutoff I (see Eq. (B.19) in [58] and
Section 3.2 in [45]). We can thus set I = to get:
disp.
disp. PV = mPV
b mb
134
CF
=
4
2

d2 Mink 2 2
a
2 SQD
2
CF
+
PV
2

0

du
sin u Eucl
T (u)
B aSQD (u) ,
1 2u
u
(E.4)
exhibiting the fact that this difference is a number of order which is independent of any hard
scale such as I or mb . Note that these equations are valid even if the Euclidean coupling has
Landau singularities (such as the one loop coupling). The Minkowskian coupling is still infrared
finite in this case, and the convergence of the Borel integral is ensured by the oscillations of the
factor sin(u)/u.
It is also possible to give a Euclidean version of the previous definition, iff one assumes the
EuclideanSudakov effective charge is causal and infrared finite. It is straightforward to show
that the analogues of Eqs. (E.1), (E.3) and (E.4) are:

2E
mb

N1

Sud m2 , N
u=1/2,Eucl
2E
MB
N1
2

I 2
N disp. CF
dk Eucl 2 N 2 k 2
k
exp
a
,
mb
4
k 2 SQD
m2b
(E.5)
disp.
disp. PV = mPV
b mb
2
CF
= mb
4
I
0

dk 2 Eucl 2 k 2
k
a
k 2 SQD
m2b

CF
+ I
PV
2
2 u

du
(u),
T (u)B aSEucl
2
QD
1 2u I
(E.6)
and, setting I = , since the r.h.s. of Eq. (E.6) does not depend on the cutoff I in the case of
a causal coupling (see Eq. (3.19) in [58]):
disp.
disp. PV = mPV
b mb
CF
=
4
2

dk 2 Eucl 2
k2
a
(k
)
k 2 SQD
2
CF
+
PV
2

0

Eucl
du
T (u)B aSQD (u) .
1 2u
(E.7)
](u)
Note also the convergence of the Borel integral has to be insured by oscillations in B[aSEucl
QD
if aSEucl
(k 2 ) has an infrared fixed point (similarly to the Minkowskian coupling above). Finally,
QD
135
we note that, although Eqs. (E.1) and (E.5) define different u = 1/2 pieces of the Sudakov expodisp.
nent, the corresponding mb masses are the same: this result follows immediately comparing
Eqs. (3.19) and (B.19) in Ref. [58].
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
G. Sterman, Nucl. Phys. B 281 (1987) 310.

S. Catani, L. Trentadue, Nucl. Phys. B 327 (1989) 323.
A. Vogt, Phys. Lett. B 497 (2001) 228, hep-ph/0010146.
S. Forte, G. Ridolfi, Nucl. Phys. B 650 (2003) 229, hep-ph/0209154.
E. Gardi, R.G. Roberts, Nucl. Phys. B 653 (2003) 227, hep-ph/0210429.
T.O. Eynck, E. Laenen, L. Magnea, JHEP 0306 (2003) 057, hep-ph/0305179.
S. Moch, J.A.M. Vermaseren, A. Vogt, Nucl. Phys. B 726 (2005) 317, hep-ph/0506288.
S. Moch, A. Vogt, Phys. Lett. B 631 (2005) 48, hep-ph/0508265.
E. Laenen, L. Magnea, Phys. Lett. B 632 (2006) 270, hep-ph/0508284.
S. Catani, D. de Florian, M. Grazzini, P. Nason, JHEP 0307 (2003) 028, hep-ph/0306211.
G. Sterman, W. Vogelsang, Phys. Rev. D 74 (2006) 114002, hep-ph/0606211.
S. Catani, B.R. Webber, G. Marchesini, Nucl. Phys. B 349 (1991) 635.
V. Ravindran, Nucl. Phys. B 746 (2006) 58, hep-ph/0512249.
V. Ravindran, J. Smith, W.L. van Neerven, QCD threshold corrections to di-lepton and Higgs rapidity distributions
beyond N2 LO, hep-ph/0608308.
P. Bolzoni, Phys. Lett. B 643 (2006) 325, hep-ph/0609073.
J.C. Collins, D.E. Soper, G. Sterman, Nucl. Phys. B 250 (1985) 199.
E. Laenen, G. Sterman, W. Vogelsang, Phys. Rev. D 63 (2001) 114018, hep-ph/0010080.
G. Bozzi, S. Catani, D. de Florian, M. Grazzini, Higgs boson production at the LHC: Transverse-momentum
resummation and rapidity dependence, arXiv: 0705.3887 [hep-ph].
S. Catani, L. Trentadue, G. Turnock, B.R. Webber, Nucl. Phys. B 407 (1993) 3.
G.P. Korchemsky, G. Sterman, Nucl. Phys. B 555 (1999) 335, hep-ph/9902341.
G.P. Korchemsky, S. Tafat, JHEP 0010 (2000) 010, hep-ph/0007005.
E. Gardi, J. Rathsman, Nucl. Phys. B 609 (2001) 123, hep-ph/0103217.
E. Gardi, J. Rathsman, Nucl. Phys. B 638 (2002) 243, hep-ph/0201019.
E. Gardi, L. Magnea, JHEP 0308 (2003) 030, hep-ph/0306094.
C.F. Berger, L. Magnea, Phys. Rev. D 70 (2004) 094010, hep-ph/0407024.
Y.L. Dokshitzer, A. Lucenti, G. Marchesini, G.P. Salam, Nucl. Phys. B 511 (1998) 396, hep-ph/9707532;
Y.L. Dokshitzer, A. Lucenti, G. Marchesini, G.P. Salam, Nucl. Phys. B 593 (2001) 729, Erratum.
Y.L. Dokshitzer, A. Lucenti, G. Marchesini, G.P. Salam, JHEP 9801 (1998) 011, hep-ph/9801324.
Y.L. Dokshitzer, A. Lucenti, G. Marchesini, G.P. Salam, JHEP 9805 (1998) 003, hep-ph/9802381.
M. Dasgupta, G.P. Salam, J. Phys. G 30 (2004) R143, hep-ph/0312283.
M. Cacciari, E. Gardi, Nucl. Phys. B 664 (2003) 299, hep-ph/0301047.
M. Cacciari, S. Catani, Nucl. Phys. B 617 (2001) 253, hep-ph/0107138.
E. Gardi, JHEP 0502 (2005) 053, hep-ph/0501257.
U. Aglietti, G. Corcella, G. Ferrera, Nucl. Phys. B 775 (2007) 162, hep-ph/0610035.
M. Neubert, Factorization analysis for the fragmentation functions of hadrons containing a heavy quark, arXiv:
0706.2136 [hep-ph].
G.P. Korchemsky, G. Sterman, Phys. Lett. B 340 (1994) 96, hep-ph/9407344.
C.W. Bauer, S. Fleming, D. Pirjol, I.W. Stewart, Phys. Rev. D 63 (2001) 114020, hep-ph/0011336.
S.W. Bosch, B.O. Lange, M. Neubert, G. Paz, Nucl. Phys. B 699 (2004) 335, hep-ph/0402094.
U. Aglietti, G. Ricciardi, G. Ferrera, Phys. Rev. D 74 (2006) 034004, hep-ph/0507285;
U. Aglietti, G. Ricciardi, G. Ferrera, Phys. Rev. D 74 (2006) 034005, hep-ph/0509095;
U. Aglietti, G. Ricciardi, G. Ferrera, Phys. Rev. D 74 (2006) 034006, hep-ph/0509271.
U. Aglietti, G. Ferrera, G. Ricciardi, Nucl. Phys. B 768 (2007) 85, hep-ph/0608047.
J.R. Andersen, E. Gardi, JHEP 0701 (2007) 029, hep-ph/0609250.
136
[44]
[45]
[46]
[47]
[48]
[49]
[50]
[51]
[52]
[53]
[54]
[55]
[56]
[57]
[58]
[59]
[60]
[61]
[62]
[63]
[64]
[65]
[66]
[67]
[68]
[69]
[70]
[71]
[72]
[73]
[74]
[75]
[76]
[77]
[78]
[79]
[80]
[81]
[82]
[83]
[84]
[85]
[86]
[87]
[88]
[89]
[90]
[91]
[92]
[93]
[94]
E. Gardi, Inclusive B decays from resummed perturbation theory, hep-ph/0703036.

E. Gardi, G. Grunberg, JHEP 9911 (1999) 016, hep-ph/9908458.
G.P. Korchemsky, G. Sterman, Nucl. Phys. B 437 (1995) 415, hep-ph/9411211.
B.R. Webber, Phys. Lett. B 339 (1994) 148, hep-ph/9408222.
Y.L. Dokshitzer, G. Marchesini, B.R. Webber, Nucl. Phys. B 469 (1996) 93, hep-ph/9512336.
R. Akhoury, V.I. Zakharov, Phys. Lett. B 357 (1995) 646, hep-ph/9504248.
A.V. Manohar, M.B. Wise, Phys. Lett. B 344 (1995) 407, hep-ph/9406392.
M. Beneke, V.M. Braun, Nucl. Phys. B 454 (1995) 253, hep-ph/9506452.
Y.L. Dokshitzer, B.R. Webber, Phys. Lett. B 404 (1997) 321, hep-ph/9704298.
E. Gardi, Inclusive distributions near kinematic thresholds, in: the Proceedings of FRIF Workshop on First Principles Non-perturbative QCD of Hadron Jets, LPTHE, Paris, France, 1214 January 2006, p. E003, hep-ph/0606080.
M. Beneke, Phys. Rep. 317 (1999) 1, hep-ph/9807443.
M. Beneke, V.M. Braun, hep-ph/0010208.
E. Gardi, Nucl. Phys. B 622 (2002) 365, hep-ph/0108222.
G. Grunberg, JHEP 9811 (1998) 006, hep-ph/9807494.
S.J. Brodsky, J.R. Ellis, E. Gardi, M. Karliner, M.A. Samuel, Phys. Rev. D 56 (1997) 6980, hep-ph/9706467.
M. Neubert, Phys. Rev. D 51 (1995) 5924, hep-ph/9412265.
M. Beneke, V.M. Braun, Phys. Lett. B 348 (1995) 513, hep-ph/9411229.
P. Ball, M. Beneke, V.M. Braun, Nucl. Phys. B 452 (1995) 563, hep-ph/9502300.
G. Grunberg, Phys. Rev. D 29 (1984) 2315.
G. Grunberg, Phys. Lett. B 304 (1993) 183.
G.P. Korchemsky, A.V. Radyushkin, Nucl. Phys. B 283 (1987) 342.
G.P. Korchemsky, Mod. Phys. Lett. A 4 (1989) 1257.
G. Grunberg, JHEP 0107 (2001) 033, hep-ph/0106070.
G. Grunberg, Phys. Rev. D 74 (2006) 111901, hep-ph/0609309.
G. Grunberg, AIP Conf. Proc. 892 (2007) 268, hep-ph/0610310.
S. Friot, G. Grunberg, Constant terms in threshold resummation and the quark form factor, JHEP 0709 (2007) 002,
arXiv: 0706.1206 [hep-ph].
J.G.M. Gatheral, Phys. Lett. B 133 (1983) 90.
E. Gardi, M. Karliner, Nucl. Phys. B 529 (1998) 383, hep-ph/9802218.
E. Gardi, G. Grunberg, M. Karliner, JHEP 9807 (1998) 007, hep-ph/9806462.
E. Gardi, G. Grunberg, JHEP 9903 (1999) 024, hep-th/9810192.
P. Gambino, E. Gardi, G. Ridolfi, JHEP 0612 (2006) 036, hep-ph/0610140.
P. Nason, M.H. Seymour, Nucl. Phys. B 454 (1995) 291, hep-ph/9506317.
G. Grunberg, Infrared finite coupling in Sudakov resummation, hep-ph/0601140.
S. Catani, Z. Phys. C 75 (1997) 665, hep-ph/9609263.
W.L. van Neerven, A. Vogt, Nucl. Phys. B 568 (2000) 263, hep-ph/9907472.
A. Sen, Phys. Rev. D 24 (1981) 3281.
A.H. Mueller, Phys. Rev. D 20 (1979) 2037.
J.C. Collins, Phys. Rev. D 22 (1980) 1478.
L. Magnea, G. Sterman, Phys. Rev. D 42 (1990) 4222.
I.A. Korchemskaya, G.P. Korchemsky, Phys. Lett. B 287 (1992) 169.
G.P. Korchemsky, G. Marchesini, Nucl. Phys. B 406 (1993) 225, hep-ph/9210281.
J.R. Ellis, E. Gardi, M. Karliner, M.A. Samuel, Phys. Lett. B 366 (1996) 268, hep-ph/9509312.
D.V. Shirkov, I.L. Solovtsov, Phys. Rev. Lett. 79 (1997) 1209, hep-ph/9704333.
G. Grunberg, Evidence for infrared finite coupling in Sudakov resummation: A revised view-point, in: the Proceedings of FRIF Workshop on First Principles Non-perturbative QCD of Hadron Jets, LPTHE, Paris, France,
1214 January 2006, p. E001, hep-ph/0606033.
T. Banks, A. Zaks, Nucl. Phys. B 196 (1982) 189.
G. Grunberg, Phys. Rev. D 46 (1992) 2228.
P.M. Stevenson, Phys. Lett. B 331 (1994) 187, hep-ph/9402276.
S.A. Caveny, P.M. Stevenson, hep-ph/9705319.
G. Grunberg, Phys. Rev. D 73 (2006) 091901, hep-ph/0603135.
137
[95] E. Laenen, G. Sterman, W. Vogelsang, Power corrections in eikonal cross sections, in: Contributed to 30th International Conference on High-Energy Physics (ICHEP 2000), Osaka, Japan, 27 July2 August 2000, in: Osaka 2000,
High Energy Physics, vol. 2, 2000, pp. 14111413, hep-ph/0010183.
[96] E. Gardi, G.P. Korchemsky, D.A. Ross, S. Tafat, Nucl. Phys. B 636 (2002) 385, hep-ph/0203161.
[97] G. Grunberg, Phys. Lett. B 372 (1996) 121, hep-ph/9512203.
[98] Y.L. Dokshitzer, N.G. Uraltsev, Phys. Lett. B 380 (1996) 141, hep-ph/9512407.
[99] G. Grunberg, Fixed points and power corrections, in: Presented at International Workshop on Deep Inelastic Scattering and Related Phenomena (DIS 96), Rome, Italy, 1519 April 1996, in: Rome 1996, Deep Inelastic Scattering
and Related Phenomena, pp. 642646, hep-ph/9608375.
[100] B.R. Webber, JHEP 9810 (1998) 012, hep-ph/9805484.
[101] C. Lee, G. Sterman, in: the Proceedings of FRIF Workshop on First Principles Non-perturbative QCD of Hadron
Jets, LPTHE, Paris, France, 1214 January 2006, p. A001, hep-ph/0603066.
[102] C. Lee, Mod. Phys. Lett. A 22 (2007) 835, hep-ph/0703030.

Bulk entropy in loop quantum gravity

Etera R. Livine a, , Daniel R. Terno b
a Laboratoire de Physique ENS Lyon, CNRS UMR 5672, 46 Alle dItalie, 69364 Lyon Cedex 07, France
b Centre for Quantum Computer Technology, Department of Physics, Macquarie University,
Sydney NSW 2109, Australia

Received 21 June 2007; received in revised form 8 October 2007; accepted 30 October 2007
Abstract
In the framework of loop quantum gravity (LQG), we generalize previous boundary state counting for
black hole entropy [E.R. Livine, D.R. Terno, Quantum black holes: Entropy and entanglement on the horizon, Nucl. Phys. B 741 (2006) 131, gr-qc/0508085] to a full bulk state counting. After suitable gauge fixing,
we show how to compute the bulk entropy of a bounded region of space (the black hole) with fixed boundary conditions. This allows to study in detail the relationship between the entropy and the boundary area
and to identify a holographic regime for LQG where the leading order of the entropy scales with the area. In
this regime we can fine tune the factor between entropy and area without changing the Immirzi parameter.
PACS: 04.60.Pp; 04.70.Dy; 04.60.Nc
Keywords: Loop quantum gravity; Entropy; Black hole; Holographic principle; Spin network coarse-graining
1. Introduction
To understand the deep structure of Loop Quantum Gravity (LQG) [1], in particular the
holographic principle and quantum black holes, it is necessary to analyze in detail the entropy
counting. Most of the work on black hole entropy in LQG focuses on boundary state counting,
especially in the isolated horizon framework [2]. In the present work we propose to extend these
considerations to the bulk entropy.
E-mail addresses: etera.livine@ens-lyon.fr (E.R. Livine), dterno@physics.mq.edu.au (D.R. Terno).

doi:10.1016/j.nuclphysb.2007.10.027
E.R. Livine, D.R. Terno / Nuclear Physics B 794 (2008) 138153
139
We use the framework outlined in [3]. We consider a given spin network state for the 3d
geometry of the (canonical) hypersurface and focus on an arbitrary bounded region. The previous work [3] analyzed the boundary entropy of such a region assuming a totally mixed state, i.e.
having no knowledge about its interior. Then the area-entropy law was naturally recovered. In
the present work, we extend this calculation to the bulk entropy. More precisely, we retain the information about the graph (underlying the quantum geometry state) inside the considered region
and, under given boundary conditions, count all the distinct spin network states on that graph.
Thus the entropydefined as the logarithm of the number of statesdepends on the topology of
the graph through its number of loops.
For the trivial tree graph topology this entropy coincides with one obtained by counting
boundary states only [3]. As soon as the topology becomes non-trivial the entropy diverges.
Nevertheless, we identify a symmetry responsible for this divergence and get a finite entropy
after suitable gauge fixing. The entropy increases with the number of loops. We find that the entropy can grow arbitrarily large compared to the boundary area for a complicated enough graph
topology. However, we also identify a regime for which the entropy still scales as the area. We
call it the holographic regime. In this regime it is possible arbitrarily adjust the proportionality
factor between the entropy and the area without changing the Immirzi parameter. If quantum
gravity is to be a holographic theory, then the LQG dynamics should ensure that projecting on
physical states selects this regime.
Finally, the gauge fixing that we use seems to be related to a gauge fixing of the Hamiltonian
constraint, but more investigations are required to understand that relationship.
2. Graph topology and state counting
Let us start with an arbitrary spin network state on the canonical hypersurface . Consider a
connected bounded region R of its graph1 that includes a finite set of vertices and the edges that
connect them. The boundary R is the set of edges which have only one end vertex laying in R.
We can picture R as a 3-ball and R as its boundary 2-sphere punctured by the boundary edges
(see Fig. 1).
Let us call the graph inside R and assume that the boundary R is made of n edges.
The spin network state carries spin labels (SU(2) representations) attached to each edge of the
Fig. 1. The internal edges are shown as regular lines, the boundary edges are dashed, and the selected exterior edges are
rendered in bold.
1 It is usually assumed that the graph is locally finite, i.e. that the valency of each vertex is finite.
140
graph. In particular, the boundary data consists of the spin labels j1 , . . . , jn of the n boundary
edges puncturing R. We assume these labels fixed once and for all as defining the boundary.
We further take them all equal to the lowest spin j1 = = jn = 1/2. As it was shown in [3],
this hypothesis simplifies the calculations and it is straightforward to generalize the calculations
to any generic configuration. For such a choice of spins the number of boundary edges n is
necessarily even, and we therefore assume in the following that the boundary is made of 2n
edges.
In [3] we motivated defining boundary states as intertwiners between the 2n boundary
representationsor equivalently as singlet states in the tensor product of the 2n spin j1 , . . . , j2n .
Counting the dimension N of the intertwiner space2 gives the (boundary) entropy S in term of
n :
the binomial coefficient C2n

1
3
2n
2n log 2 log n + .
S log N = log
(1)
n + 1 n n
2
Assuming that every j = 1/2 puncture contributes a microscopic area a 1 lP2 in Planck units, we
2
find that the entropy satisfies the usual proportionality law at the leading order, and has a 3/2
logarithmic correction:
S
A log 2 3
log A + ,
2
lP2 a 1
(2)
where the total boundary area is A = 2na 1 lP2 . In the standard LQG framework, the microscopic
2
area is given in terms of the Immirzi parameter a 1 = 3/2. Then the semi-classical relation
2
S A/4lP2 can be used to fix the value of the Immirzi ambiguity . The isolated horizon framework presents a different though similar calculation for the entropy. It leads to a different ratio
S/A and thus a different value of , but also to a different value of the logarithmic correction
usually 12 instead of the present 32 .
This boundary entropy calculation can be interpreted as computing the number of spin network states assuming that the graph inside the region R is reduced to a single vertex. Assuming
that the outside observer has no access to any information about the graph inside R, we have indeed coarse-grained this graph to a single point: this can be dubbed the black point model.
In the present work we investigate the effects of keeping a non-trivial graph on the entropy
calculation. Our bulk entropy is the maximal von Neumann entropy of the system
S log N,
(3)
where N is number of orthogonal states satisfying the imposed constraints. We do not impose
the (exterior) bulk to the surface matching, therefore we have no reason to restrict our counting
only to the surface states.
Consider with V vertices, E internal edges and E = 2n external legs. Then the number of
independent loops is L = E V + 1. When is a tree, L = 0, and the entropy is exactly the
2 In [3], we assumed the black hole state defined by the totally mixed state 1 on the intertwiner space. This is
naturally the state seen by the external observer who does not have any information on the internal black hole state. It is
also a static state, which does not evolve under unitary evolution, and thus corresponds to the physical set-up of a black
hole. Then the entropy of this state is of course the logarithm of the intertwiner space dimension.
141
same as for the trivial graph with a single vertex. If the graph possesses a non-trivial topology,3
L 1, we obtain an infinite number of states and thus of degrees of freedom. Indeed, loops of
a spin network can carry arbitrary spins j N/2 independently of each other. To overcome this
obstacle, we note that this divergence is associated to a further gauge invariance. This gauge
invariance is generated by the action of the holonomy operators on the loops of the spin network
state. To count physical degrees of freedom and to get a meaningful finite entropy, we gauge fix
this action [4]. We use the simplest gauge fixing and fix the spin associated to each loop of .
It actually requires fixing the spin of only one link of each loop. Having done this we indeed
obtain a finite dimensional Hilbert space (as we will see in the following section). The resulting
entropy does not depend on the choice of the spins used in the gauge fixing and only depends on
the size/area of the boundary E = 2n and on the complexity of the graph described by the
number of loops L.
This gauge fixing of the holonomy operators on the graph loops can be related to the gauge
fixing of the Hamiltonian constraint. In the context of the topological BF theory, this would
actually exactly coincide with the action of the Hamiltonian constraint.4 In the case of LQG, it is
more subtle and we should investigate further the validity and physical interpretation of our gauge
fixing. While we postpone this for future work, we insist that in our framework this gauge fixing is
natural. First, we gauge fix the holonomies on loops inside the considered region R. This a priori
does not affect the dynamics of the exterior of R. On the other hand, if we did not gauge fix them
we would be over-counting the number of states (seen by an external observer). What would be
non-trivial is to gauge fix the action of holonomies on loops crossing the boundary/horizon R.
Second, we do check that the number of degrees of freedom does not depend on the gauge fixing
and that the action of the holonomies creates isomorphic copies of the same physical Hilbert
space. In short, we are disregarding the internal excitations that do not couple to the exterior of
the region and that would produce an overcounting in the entropy ascribed to R by an external
observer.
In the following sections we study in detail the one-loop case and then show that the calculations can be straightforwardly generalized to an arbitrary graph topology. Finally we discuss the
different regimes of entropy corresponding to the different scaling of the number of loops L with
the size of the boundary n.
3. The one-loop case
Let us start by reviewing the black-point case, L = 0. The Hilbert space of spin networks of
a tree is isomorphic to the space of intertwiners between the boundary representations [3]. Its
dimension is given by:

1
2n
dg 1 (g)2n =
,
dim H0 =
(4)
2
n+1 n
SU(2)
3 Let us point out that the graph topology does not a priori have any relation to the topology of the spatial hypersurface.
There are two points of view about the manifold topology. It is either assumed right at the start and we consider embedded
graphs. Otherwise we consider abstract graphs, containing only combinatorial and algebraic data, and the hypersurface
topology is an emerging semi-classical notion. We favor the latter point of view.
4 The action of the holonomy operators in the BF theory generates the translational symmetry on the B field. It needs
to be dealt with in order to get the physical degrees of freedom of the theory.
142
Fig. 2. One-loop spin networks with the 2n boundary links: the internal loop carries the spin j .
where dg is the normalized Haar measure on the SU(2) Lie group and 1/2 (g) is the character in
the fundamental spin-1/2 representation. This formula can be interpreted in terms of a random
n give the number of returns to the
walk with a mirror in origin. The binomial coefficients C2n
origin for the usual random walk, while the mirror introduces the factor 1/(n + 1).
Consider now the one-loop diagram on the left of Fig. 2. It is a loop to which 2n legs are
attached, so all the vertices are 3-valent.5 Pick an arbitrary link e0 and fix the spin which it
carries, je0 = j . This obviously fixes the action of the holonomy operator on the loop: acting
with a holonomy carrying a spin J on this state would change the spin-j representation into the
tensor product representation j J .
Assume that j n/2. Moving from the initial link e0 to the next link e1 , we see that e1 can
carry either the spin j 1/2 or j + 1/2 since the external leg between e0 and e1 injects a
spin 1/2 into the diagram. Going on one finally arrives back to the initial link e0 and the initial
representation j . Therefore, the number of possible spin labeling j, j 12 , . . . , j 12 is exactly
the number of returns to the origin of a random walk after 2n iterations:

2n
dim H1 =
(5)
.
n
We get a finite result which is different from the tree case. Moreover, it does not depend on the
chosen spin j . This Hilbert space leads to the following asymptotic behavior of the entropy:
1
(6)
log n + .
2
It does not affect the leading order proportional to the area but only the logarithmic correction.
There is nevertheless a subtlety: the assumption that the spin j is large enough compared to
the boundary size n. Indeed if j is smaller than n/2, there exists the possibility that the random
walk along the loop will ascribe a spin 0 to a link. In such a case, we cannot move down to a
1/2 spin but can only go back up to +1/2. Then we get a smaller Hilbert space. We discard
this case because we do not consider it as a true one-loop graph anymore. Indeed, if a link is
labeled with a spin 0, it is just as if that link did not exist since the corresponding spin network
wave function would not depend on the holonomy on that link. Then if we remove that link from
the graph, we end up with a tree again. Therefore, we interpret this situation when j n/2 as
describing a superposition of a 0-loop and 1-loop graphs. However, we can still compute exactly
the dimension of the Hilbert space. For this purpose, it is more convenient to use the another
one-loop diagram that is shown on the right of Fig. 2.
We now look at the one-loop diagram with two vertices: all the boundary links merge into a
single link to which the loop is then attached. Assume that the loop still carries a fixed spin j and
S1 2n log 2
n
5 An arbitrary one-loop graph is equivalent to this form due to the SU(2) gauge invariance at every vertex (see e.g. [4]).
143
we call k the spin carried by the intermediate link. On one side, we have an intertwiner between
the 2n spin 1/2 and the spin k. As long as k n, the dimension of this intertwiner space is [3]:

2n
2n
2k + 1
2n
(n)
dk =
(7)
=
.
n+k
n+k+1
n+k+1 n+k
On the other side, we have a unique 3-valent intertwiner between two representations j and
the same representation k. The corresponding intertwiner space is of dimension one as long as
k 2j . The total number of spin network states amounts to summing over all possible ks from
0 to the maximal allowed spin kmax min(n, 2j ):
(j )
N1 =
k
max
(n)
dk =
k=0

2n
2n
.
n
n + kmax + 1
(8)
(j )
When j is larger than n/2, we recover the previous result with N1 = N1 =
2n
. At the other
2n
end of the spectrum, when j
= N0 = 2n
n n+1 .
Finally, we see that the values of the spin j carried by the loop between 0 and n/2 interpolate
between the tree case and the 1-loop case. This allows us to interpret this intermediate situation as a superposition of the 0-loop and 1-loop cases, and moreover to identify the true 1-loop
case as the j n/2 regime when the entropy does not depend anymore on the specific value
of j .
To conclude this section, we showed that considering a one-loop graph over a tree increases
the entropy. After carefully gauge fixing, we obtain a finite entropy which differs from the 0-loop
case only by the logarithm correction, 1/2 instead of 3/2.
In the following sections, we generalize this analysis to an arbitrary number of loops. We will
show that if the number of loops is fixed it will only affect the logarithm correction while if we
allow the number of loops to scale with n it may well also affect the leading order.
(j )
vanishes, we recover the 0-loop case with N1
4. The two-loop case

We now move to the entropy counting in the two-loop case. We work with the graph shown
on Fig. 3 and we fix the spins along the two loops to the same value j for the sake of simplicity.
We further assume as previously that j n/2 so that we consider a pure two-loop graph and not
a superposition with a 0-loop or 1-loop configuration. The dimension of the Hilbert space of spin
networks on is given by the following integral:

N2 = dg 1 (g)2n j (g)4 .
2
Fig. 3. Two-loop graphs with boundary links.
144
Fig. 4. Fully gauge-fixed two-loop spin network: we fix the spins carried by the three edges on the loops.
For computational purpose, it is convenient to express as a sum over the intermediate spin
label k. It is straightforward to obtain

n

3
(n)
N2 =
(9)
dk (2j + 1)(2k + 1) k(k + 1) .
2
k=0
The two terms can be computed exactly6 :

n

(n)
(2k + 1)dk = 22n ,
n

k=0
k=0
k(k
(n)
+ 1)dk

2n
=n
n22n .
n
(10)
As we see we have a residual j -dependent term, which actually diverges as j grows to infinity.
Neglecting7 the extra-term k(k + 1), the j -dependence can be factorized. Therefore, we interpret
this as a symmetry that we have not gauge fixed yet. Indeed we have gauge fixed the action of
the holonomy operator along the loops a and b by requiring ja = jb = j , but the action of the
holonomy along the loop a b is not yet fixed. This leaves a freedom in the intertwiner which
leads to that (2j + 1) factor, resulting in a over-counting in the Hilbert space dimension.
This can easily be seen considering the two-loop diagram as shown in Fig. 4. J can still vary
from 0 to (2j + 1). This is the freedom that requires gauge-fixing. It is possible to compute
the intertwiner dimension for different values of J . It is possible to compute the intertwiner
dimension for different values of J . For instance, we get:
(J =j )
N2
n

(n)
(2k + 1)dk ,
(J =2j )
N2
k=0
n

(n)
(k + 1)dk .
k=0
(n)
Both these examples, have the same behavior up a factor with a leading order given by k kdk .
They give the same entropy in the asymptotic limit n + (same leading order and logarithmic
6 A useful identity for the degeneracy coefficients is
n

(n)
dk sin(2k + 1) = 22n sin cos2n ,
k=0
which is derived by computing the trace of SU(2) group elements in the representation ( 12 ) 2n. Differentiating this
(n)
equation, we obtain the following formula for the polynomial averages over the dk distribution, for l N:
n

(n)
(2k + 1)2l+1 dk

= (1)l 22n sin cos2n =0 .
k=0
7 The second term in k(k + 1) is indeed be neglected if j goes to infinity faster than n. This condition is automatically
satisfied since we assume that j n/2.
145
correction):
S2 log 22n 2n log 2 +
(11)
with a vanishing logarithmic correction (they are of course sublog corrections). We see that
adding one loop from the one-loop case corresponds to implementing a + 12 factor in the logarithmic correction without affecting the leading order proportional to the area. We generalize this
statement to arbitrary number of loops in the following section.
5. Gauge fixing and generic bulk entropy
5.1. Informal arguments
For a generic graph with L loops, following the one-loop and two-loop cases, it is reasonable
to expect the gauge-fixed dimension at leading order to scale as
NL
n

(n)
k L1 dk .
(12)
k=0
Assuming that L is fixed, the asymptotics is in the large n limit only changes the logarithmic
correction of the entropy:

L
1 log n + .
SL log NL 2n log 2 +
(13)
2
Using the same techniques as in [3], it is straightforward to approximate the sum over k by an
integral and then compute the asymptotics of NL by a saddle point approximation. Then only the
(n)
logarithmic correction changes by a factor + 12 with eachloop because the distribution dk looks
like a Gaussian peaked at k = 0 with a width scaling as n.
(n)
More precisely, denoting x k/n [0, 1], we can approximate the dimensions dk for large n
using the Stirling formula:
x
2 22n
(n)
en(x) ,
dk
n (1 + x) 1 x 2
(14)
with the exponent (x) given by:

(x) = (1 + x) log(1 + x) + (1 x) log(1 x).
(15)
Therefore, the full Hilbert space dimension can be approximated by an integral:

1
2
NL 22n nL 2
1
dx
0
xL
en(x) .
(1 + x) 1 x 2
The exponent (x) is always positive for x [0, 1]. It has a unique fixed point at x = 0. For
n, we can use the Gaussian approximation, (x) = x 2 + O(x 3 ). Then we need to evaluate
large
L
2
(L+1)/2 . Finally, this leads to the asymptotics for the
0 dx x exp(nx ), which scales as n
entropy given above in (13).
So far we have kept the number of loops L fixed as the boundary area n was taken to infinity.
This number does not affect the leading order of the entropy, but only changes the pre-factor of
146
the logarithmic correction. However, we see that if we allow the graph complexity to grow with
the boundary size, we are able to change the leading order behavior of the entropy and change
the proportionality factor between the entropy S and the area n.
Below we make this argument more precise and compute the dimension and entropy exactly
for a specific choice of gauge-fixing. We find a missing factor 1/(L 1)! in the guessed dimension (12). This changes the asymptotical behavior when L is allowed to be arbitrarily large. Then
we keep the asymptotics described above for fixed L while allowing the loop number L to scale
as n changes the leading order factor8 between the SL and n.
5.2. The entropy formula
Let us now consider the L-loop graph with three vertices A, B, C as shown on Fig. 5. The 2n
boundary links combine into intermediate spin k at the vertex A. On the other side, the L loops
combine at the vertex B to that same intermediate link. Finally the intertwiner at the C node
closes the graph and describes how the L loops are coupled to each other.
Our gauge fixing, or more precisely, the choice of a sector of the spin network Hilbert space,
is achieved by fixing the spin labels on the L + 1 edges defining the L loops. We label them
sequentially by representations j, j, 2j, 4j, 8j, . . . , 2L1 j . As long as j is assumed larger than
n/2 as before, this leads to an entropy which does not depend on the gauge-fixing representation j
but only on the number of loops L and the number of boundary edges n.
We need to underline that we are not only fixing the action of the holonomy operator along
the L loops. Indeed, our specific choice of representation labels is not only convenient for computational purposes but also it fully fixes the unique intertwiner at the node C. This amounts
to fixing L 2 representation labels within the node C. This means that we are truly fixing
(L + 1) + (L 2) = 2L 1 quantum numbers, and not only L as we first expected. We are in
fact counting the number of possible intertwiners at the vertices A and B. At a nave level, the
Fig. 5. Gauge-fixed spin networks with L loops: the L + 1 edges of carry the labels j, j, 2j, . . . , 2L1 j .
8 Assuming that the number of loops goes as L = n with a fixed ratio > 0, it is straightforward to extract the
asymptotics of the entropy:

n

1
1
(n)
L1
S log
n log n + ,
k
dk
(L 1)!
2
k=0
where the precise value of > 0 depends on but is generically different from 2 log 2. The key to this calculation is that
= (x) log x. The fixed point x0 of

the weight k L1 now contributes to the exponent which get modified to (x)
is not x = 0 anymore. In particular (x
0 ) = 0 and this leads to a term proportional to n in the Gaussian approximation.
147
entropy that we compute counts only the different ways that the internal loops couple to the external links while we disregard the number of ways that the loops couple to each other by fixing the
intertwiner at the internal node C. At a mathematical level, if one fixes less representation labels
than we do (less than 2L 1), one gets an entropy which depends on j and diverges as j grows
large. This is a symptom of insufficient gauge-fixing. Nevertheless, there remains the open issue
of understanding the physical interpretation and relevance of such a gauge-fixing. In particular,
what is the precise symmetry that we are gauge-fixing? We leave this question for future investigation. However, even without such an interpretation in term of symmetry and gauge-fixing, we
still have identified sectors of the spin network Hilbert space with different asymptotic behaviors
of the entropy of the considered region which only depend on the number of loops of the graph
(supporting the spin network states) inside that region.
To compute the entropy, we need to find the number of intertwiners at the vertices A and B.
At the vertex A we have the same degeneracy dk(n) as previously. As for the vertex B, having
L loops leads to the following degeneracies for the spin-k subspaces as long as k 2L1 j for
L 1:

k+L1
j (L)
dk =
.
(16)
k
k+L1 K+L
This formula is straightforward to prove by induction using the identity K

= K .
k=0
k
As a result, the dimension of the spin network space inside R is
NL =
n

(n) j (L)
dk ,
(17)
dk
k=0
where we have assumed n 2j . For L 2 this can be slightly simplified9 :

n
n

2n
k+L2
(n) k + L 1
NL =
dk
=
.
k
n+k
k
k=0
(19)
k=0
For a small number of loops,10 we easily get the exact results for NL , using properties of the
(n)
binomial coefficients [5]. For L = 1, we recover the previous result N1 = k dk = 2n
n , with
1
an asymptotic expression for the entropy S1 2n log 2 2 log n. For L = 2 one obtains

n

1 2n
2n
2n1
N2 =
(20)
+
,
=2
2 n
n+k
k=0
9 This sum has an explicit closed form in terms of hypergeometric functions,
NL =
(2n)!
F1 (n, L 1, 1 + n; 1).
(n!)2 2
(18)
10 To reach higher values of L, the number of states N can be calculated by induction with the help of the following
L
recurrence relation:

n

2n
2n
km
k m2 n2 n2 k 2
=
n+k
n+k
k
k=0

2n
2n 2
2n(2n 1)
.
= n2
k m2
k m2
n+k
n1+k
k
148
with the entropy S2 having no logarithmic correction for large n (of course, it still contains sublogarithmic corrections). For L = 3, we can also compute:

n

2n
n + 1 2n
N3 =
(21)
+ 22n1 ,
(k + 1) =
2
n
n+k
k=0
with a logarithmic correction being + 12 .

5.3. The entropy asymptotics
In the analysis of the entropy asymptotics, we distinguish between three cases. First, if the
number of loops L is held fixed (or more generally L stays negligible compared to n), we recover
the results given above: the leading order 2n log 2 does not change, while the non-trivial topology
of the graph affects the logarithmic correction which becomes (L/2 1) log n.
In the second case L is scaling as n: the entropy depends on the limit of the ratio L/n
as n goes to infinity. In the third case L grows very large compared to n: if the ratio L/n is
unbounded, the leading order of the entropy drastically changes and will not scale proportionally
to n anymore.
The asymptotic expression
above cases can be derived from a saddle point approx 2nallthe
for
k+L2
imation of the sum NL = k n+k
as shown in details in appendix. It also allows to
k
obtain the sub-leading terms with an excellent precision.
When L n we get a compact expression,
SL = log NL ,
NL
22nL3 nL/21
,
(L/2)
(22)
where (z) = (z 1)!. This still gives a good estimate of the sub-leading terms: for example,
when L = 7 and n = 10 000, the difference between the exact and asymptotic values is S =
0.354794, while S 1.39 104 .
When L = n with a fixed ratio , we can still apply the same saddle point technique. Using
Stirling formula, we approximate the dimension NL by the integral expression,
3
22n nL 2
NL
(L 2)!eL2
1
0
dx
en(x)
,
3
2
x 1 x (x + ) 2
(23)
with the new -dependent exponent,

(x)
= (x) + x log x (x + ) log(x + ).
(24)
As shown in details in appendix, the Gaussian approximation controls the asymptotic behavior
of the entropy:
1
(25)
log n + .
2
The important new result is that the leading order factor S/n depends on and is not fixed
to the standard factor 2 log 2 anymore. We show on Fig. 6 a plot with Maple numerics for the
entropy SL (n) for L = 0 (as a reference) and = 1, 2. While the leading behavior is always
linear, the slop S/n clearly depends on the value of the ratio = L/n.
SL=n n
149
Fig. 6. Plot of the entropy SL (n) for L = 0, L = n and L = 2n for the number of punctures n running from 0 to 100. The
leading behavior is linear in n but the slope depends on the scaling L/n. The numerics give 1.386 for the boundary
entropy L = 0, while we get 2.074 for L = n and 2.400 for L = 2n. These numerical values fit the analytical
result (up to two decimals) derived in Appendix A.
The final case is when the complexity of the graph L grows much faster than the boundary
size n. In this case, the entropy grows faster than linearly compared to the boundary area:
SLn n(log L log n) + n
1
log n + .
2
(26)
6. The regimes of bulk entropy

After a suitable gauge fixing, we have managed to compute the finite bulk entropy counting
the number of spin network states supported by a fixed graph inside the considered bounded
region R with fixed boundary conditions. The entropy S only depends on the number of links n
puncturing the boundary R and the number of loops L of which quantifies the complexity of
the graph. Computing the asymptotics of SL , we distinguish three regimes.
The logarithmic regime: The number of loops L is held fixed while n is take to infinity.
The leading order of SL is linear in n and is exactly the same as for the boundary entropy [3].
The complexity of the graph only affects the factor in front of the logarithmic correction log n,
which increases with L. Therefore, if a (quantum) black hole is belongs to this regime, fixing the
smallest spin j = 12 fixes the Immirzi parameter.
The holographic regime: The number of loops L scales proportionally to the horizon area n.
The entropy still grows linearly in n in the leading order but the proportionality factor changes
and depends on the ratio = L/n. This is a sector of the spin network space where the entropy still scales with the boundary area and not faster. If the full LQG theory is to be (strongly)
holographic, then the dynamics should restrict the physical states to this regime: the graph complexity cannot grow faster than the boundary size. On the other hand, even if the LQG theory is
not restricted to this regime, a quantum black hole state should lay in this regime if we want to
respect the semi-classical area-entropy law. Then since the factor S/n depends on the ratio ,
we can adjust this new parameter so to get S A/4 without changing neither the Immirzi
150
parameter nor the area quantum a 1 . In this scenario, the black hole entropy calculation does not
2
fix the Immirzi parameter. From another perspective, we can say that the factor renormalizes
the Immirzi parameter . Finally we need the dynamics to select the physical sector of the spin
network space and provide us with the right value of . Of course, this scenario only holds if
we use the bulk entropy and do not restrict ourselves to work with the boundary entropy.
The non-linear regime: If the number of loops grows much faster than n, i.e. L/n ,
then the entropy is free to grow non-linearly with n. Since the leading order is n log L, we can,
e.g.,
get an entropy scaling with the volume n3/2 for an exponential growth of the type L
n
2 . Of course, one should keep in mind that the volume of R actually depends on the bulk
state and does not always scale as n3/2 . It could well be larger (or smaller) if the space is tightly
curved.
7. Conclusions
We explored the notion of bulk entropy in the framework of loop quantum gravity, in order
to generalize the calculations of boundary entropy of [3]. Our aim was to compute the number
of (spin network) states describing the quantum geometry of a bounded region of space, living
on a fixed graph and with fixed boundary conditions. The resulting entropy is an upper bound
on the quantum-mechanical entropy of the system, which may or may not be equal to the thermodynamic entropy. The relationship with latter can be settled only after a rigorous black hole
thermodynamics analysis, imposing all the constraints and deriving the classical limit. It is worth
noting that the von Neumann (quantum information-theoretical) entropy is well-defined and useful even when thermodynamic description does not exist [6].
Counting the spin network states living on a fixed open graph , the bulk entropy is obviously
related to the complexity of that graph. Since a spin network wave functions is labeled by one spin
representation per edge of (and one intertwiner per vertex), it might seem at first that degrees of
freedom are attached to these edges (and vertices). However, due to the gauge invariance, degrees
of freedom are truly carried by the loops of the graph. As a consequence, the bulk entropy was
found to depend only on the number of loops L of the underlying graph. A straightforward
calculation then gives an infinite entropy. Nevertheless, after fixing a certain number of labels of
the spin network states (corresponding to gauge fixing holonomy operators acting on the loops
of ), we were able to extract meaningful finite results. This lead to the identification of different
sectors of the spin network Hilbert space depending on L with different asymptotical behavior
of the entropy.
We can actually have an entropy growing as large as we want if we choose a large enough
number of loops L. Two sectors are particularly interesting. If the graph complexity is fixed while
the boundary area grows, then the bulk entropy S has the same leading order as the boundary
entropy but the logarithmic correction to the entropy changes and increases with L. On the other
hand, if L grows linearly with the boundary area A, then the bulk entropy also grows linearly
with the area but the ratio S/A depends explicitly on the ratio L/A. We call this sector the
holographic regime.
If we want black holes to satisfy the area-entropy law, the quantum black hole state should
necessarily be in the holographic regime. However, this regime is generic enough to allow for
fine-tuning of the ratio S/A to 1/4 without changing the Immirzi parameter (i.e. the value of
the minimal quantum of area). Indeed, we can always change the area-complexity ratio L/A to
adjust the value of S/A. Our point of view is that it is the LQG dynamic that is supposed to select
the physical regime and actual value of L/A.
151
There are two important issues to address in this scenario. First, should the considered entropy
for the space region be its boundary entropy or its bulk entropy? We see a priori no reason why
the degrees of freedom inside the region would not couple to the space outside. Then the bulk
degrees of freedom would affect the black hole evaporation. However, the case of a black hole
may (should?) be drastically different from a generic region of space. This is a question that needs
to be addressed by the LQG dynamics. The second issue is directly related to our approach. Our
entropy calculation relies on a partial fixing of the spin network labels. We discussed that it could
be interpreted as a gauge fixing of the Hamiltonian constraint, but this needs to be worked out in
details. Thus, the limits of validity of the present work depend on understanding the legitimacy of
this gauge fixing.Indeed, we have identified the different sectors of the spin network kinematical
space with different bulk entropy behavior, but we still need to provide these different sectors
with a proper physical interpretation.
Appendix A. Computing the asymptotics of the entropy
The asymptotic expressions for n and the different regimes of L = L(n) are based on
the use of Stirling formula, replacement of the sum by the (EulerMacLaurent) integral and the
saddle point (Gaussian) approximation. It is convenient for the analysis to introduce a shifted
loop number l = L 2 and consider the ratio l/n.
The first factor in the sum of Eq. (19) takes the form

2n
22n
f (k/n)en(k/n) ,
(A.1)
n+k
n
where
1
,
(x) = (1 + x) log(1 + x) + (1 x) log(1 x).
1 x2
The second factor becomes

(k + l) (k + l)l+k
nl
k+l
1
= l g(k/n, l/n)en(k/n,l/n) ,
l
k
e l!
k
k
e l!
k
f (x) =
where
(A.2)
(A.3)
(x + )
,
(x, ) = x log x (x + ) log(x + ).
x
The sum of Eq. (19) is replaced by the integral
g(x, ) =
22n nl+1
Nl
n el l!
1
dx f (x)g(x, )en(x,)
,
(A.4)
(A.5)
with = (x) + (x, ). The saddle point approximation requires the knowledge of the derivatives of the exponent:
+ 2x + (2 )x 2
.
(x + )(x x 3 )
The unique fixed point, (x
0 ) = 0, in the interval [0, 1] is

1 2
x0
+ 8 ,
4
x = log
x(1 + x)
,
(x + )(1 x)
x2 =
(A.6)
(A.7)
152
and it is easy to check that 2 is always positive on [0, 1].

In particular, for 0 < < the minimum of is inside the interval and we use the Gaussian
approximation
1
dx y(x)en(x)
y(x0 )
2
n 2 (x
0)
0)
en(x
,
that gives the following asymptotics for the entropy:

S(l, n) = log Nl ,
Nl
0 ,)
22n+1/2 nl
en(x

)g(x
,
)
f
(x
.
0
0
el l!
2 (x
0 , )
(A.8)
Hence the leading order term of the entropy

1
S(l, n) n f log n
2
(A.9)
is given in terms of :
2 log 2 (x
0 ) log

= 5 log 2 log 4 + log 3 + ( + 8)

log 8 4 + ( + 8) ,
(A.10)
while the subleading terms have a more cumbersome appearance.

As a matter of fact, this approximation is in an excellent agreement with the exact results:
e.g., S(l = 10, n = 10 000) 1.39 104 and the error is S 0.0077, increasing only to
S(l = 10, n = 40 000) 0.0080. At the other extreme, S(l = 1.5 105 , n = 1000) 6022.7
with S 8.3 105 .
When runs from 0 to , the fixed point x0 varies from 0 to 1. In the special case = 1
when l = n,
1
x0 = ,
2
(x
0 ) = log 2,
2 (x
0 ) = 4,
and we get simple asymptotics:

S(l = n) 3n log 2
1
1
log n log .
2
2
(A.11)
In this case, we can actually give a faster proof of the asymptotics:

n

(2n)! n
2n
n+k
(2n)! n
Nl=n =
2 .
=
=
n+k
k
(n!)2 k
(n!)2
k=0
In the limit regime l n, we have

1
,
(x
0 ) (log 2 1 log ),
x0
0
2
2
which improves the estimate of Eq. (22) to
0) 4
2 (x
5
2 +
,
2
l
l
l
1
1
log n log l + (1 log 2) log l log(4)
2
2
2 2
2

1
l
l
1
5
l2
9 l
1
+ l
+
+O 2 .
12l
n
2 n 4 2 n 4n 32 n
153
S(l n) = 2n log 2 +
(A.12)
Finally, in the case that l n, we have

2
5
,
(x
0 ) ( + 1) log + (2 log 2 1) ,
7
1
0) +
,
2 (x
2 2 2
and Eq. (A.8) yields a simple expression for the entropy,

2

n
1
1
log n + n log 2 + (6n 1)/12l + O 2 .
S(l n) n log l n +
2
2
l
x0 1
(A.13)
References
[1] C. Rovelli, Loop Quantum Gravity, Cambridge Univ. Press, Cambridge, 2004;
T. Thiemann, Lectures on loop quantum gravity, Lect. Notes Phys. 631 (2003) 41, gr-qc/0210094;
A. Ashtekar, J. Lewandowski, Background independent quantum gravity: A status report, Class. Quantum Grav. 21
(2004) R53, gr-qc/0404018.
[2] A. Ashtekar, J. Baez, K. Krasnov, Quantum geometry of isolated horizons and black hole entropy, Adv. Theor. Math.
Phys. 4 (2000) 1, gr-qc/0005126.
[3] E.R. Livine, D.R. Terno, Quantum black holes: Entropy and entanglement on the horizon, Nucl. Phys. B 741 (2006)
131, gr-qc/0508085.
[4] L. Freidel, E.R. Livine, Spin networks for non-compact groups, J. Math. Phys. 44 (2003) 1322, hep-th/0205268.
[5] R.L. Graham, D.E. Knuth, O. Patashnik, Concrete Mathematics, AddisonWesley, 1994.
[6] D.R. Terno, Phys. Rev. Lett. 93 (2004) 051303, hep-th/0403142.

Hard spectator interactions in B at order s2

Volker Pilipp
Arnold Sommerfeld Center, Department fr Physik, Ludwig-Maximilians-Universitt Mnchen, Theresienstrasse 37,
80333 Mnchen, Germany
Institute of Theoretical Physics, Universitt Bern, Sidlerstrasse 5, 3012 Bern, Switzerland
Received 4 October 2007; accepted 30 October 2007
Abstract
I compute the hard spectator interaction amplitude in B at NLO, i.e., at O(s2 ). This special
part of the amplitude, whose LO starts at O(s ), is defined in the framework of QCD factorization. QCD
factorization allows to separate the short- and the long-distance physics in leading power in an expansion
in QCD /mb , where the short-distance physics can be calculated in a perturbative expansion in s .
In this calculation it is necessary to obtain an expansion of Feynman integrals in powers of QCD /mb .
I will present a general method to obtain this expansion in a systematic way once the leading power is given
as an input. This method is based on differential equation techniques and easy to implement in a computer
algebra system.
The numerical impact on amplitudes and branching ratios is considered. The NLO contributions of the
hard spectator interactions are important but small enough for perturbation theory to be valid.
PACS: 12.15.Hh; 12.39.St; 13.20.He
Keywords: QCD factorization; Hard spectator interactions; Feynman integrals; Differential equations; Method of
regions
1. Introduction
In the last decades B physics has proven to be a promising field to determine parameters of
the flavour sector with high precision. On the theoretical side QCD factorization [1,2] has turned
out to be an appropriate tool to calculate B decay modes from first principles. Though the decay
* Correspondence address: Institute of Theoretical Physics, Universitt Bern, Sidlerstrasse 5, 3012 Bern, Switzerland.
E-mail address: volker.pilipp@itp.unibe.ch.

doi:10.1016/j.nuclphysb.2007.10.028
V. Pilipp / Nuclear Physics B 794 (2008) 154188
155
Fig. 1. Tree level, vertex correction and penguin contraction. These diagrams contribute to T I .
of the B-meson is caused by weak interactions, strong interactions play a dominant role. It is
however not possible to handle the QCD effects completely perturbatively. This is due to the
energy scales that are contained in the B-meson: Whereas s at the mass of the b-quark is a
small parameter, the bound state of the quarks leads to an energy scale of O(QCD ) which spoils
perturbation theory. The idea of QCD factorization is to separate these scales. At leading power
in QCD /mb we obtain the amplitude for B in the following form:
1
|H|B F
dx T I (x)f (x)
B
0
1
+
dx dy d T II (x, y, )fB B1 ( )f (x)f (y).
(1)
Two different types of quantities enter this formula. On the one hand the hadronic physics is
contained in the form factor F B and the wave functions B1 and , which will be defined
more precisely in the next section. These quantities contain the information about the bound
states of the mesons. They have to be determined by non-perturbative methods like QCD sum
rules or lattice calculations. Alternatively, because they are at least partly process independent,
they might be extracted in the future from experiment. On the other hand the hard scattering
I
II
kernels
T and T contain the physics of the hard scale O(mb ) and the hard collinear scale
O( mb QCD ) and can be calculated perturbatively.
The Feynman diagrams that contribute to B can be distributed into two different
classes. The class of diagrams where there is no gluon line connecting the spectator quark with
the rest of the diagram (Fig. 1) contributes to T I . We obtain T II by evaluating the hard spectator scattering diagrams, which are shown in LO in s in Fig. 2. The order s2 corrections of
the soft momentum l of the constituent quark of
T II are the topic of the present work. Through

the B-meson the hard collinear scale QCD mb comes into play. This leads to the fact that in
contrast to T I , which is completely governed by the scale mb , T II comes with formally large logarithms. These logarithms cannot be resummed in the present QCD calculation. It will be shown
by numerical analysis that the scale dependence of the hard spectator scattering amplitude and
the absolute size of its NLO corrections are small enough for perturbation theory to be valid.
My calculation of the hard spectator scattering amplitude is not the first one as it has been
calculated recently by [3,4]. It is however the first pure QCD calculation, whereas [3,4] used
the framework of soft-collinear effective theory (SCET) [57] an effective theory, where the
expansion in QCD /mb is performed at the level of the Lagrangian rather than of Feynman
integrals. It is the main result of this paper to confirm the results of [3,4] and to show by explicit
calculation that pure QCD and SCET lead to the same result in this case.
156
Fig. 2. Hard spectator interactions at O(s ). This is the LO of T II .
Hard spectator scattering corrections to the penguin diagram (third diagram of Fig. 1) are
beyond the scope of this publication. This class of diagrams does not influence the cancellation
of the scale dependence of the tree amplitude, i.e., the diagrams of Fig. 2 and higher order s
corrections. For phenomenological applications, however, the hard spectator penguin amplitudes,
which have been recently calculated in [8], should be taken into account. Also the order s2 of
T I is important for phenomenological applications. This calculation has been partly performed
by [9,10]. There the complete imaginary part and a preliminary result of the real part of the
amplitude is given.
From a technical point of view the calculation in this paper consists of the evaluation of
about 60 one-loop Feynman diagrams. The challenges of this task are due to the fact that these
diagrams come with up to five external legs and three independent ratios of scales. In order to
reduce the number of master integrals and to perform power expansions of the Feynman integrals,
integration by parts methods and differential equation techniques will prove appropriate tools.
They provide a general method to obtain higher powers of a Feynman integral once the leading
power is given.
The paper is organized as follows: I define my notations in Section 2. In Section 3 I show how
to get T II at LO. In Section 4 I present a method which uses differential equation techniques and
allows for the extraction of higher powers of Feynman integrals once the leading power is given.
Sections 5 is dedicated to the technical details of the calculation. After some remarks how to
evaluate the Feynman diagrams occuring at NLO I show how to deal with meson wave functions
at NLO. Especially the correct treatment of evanescent structures will be explained in detail. In
Section 6 the analytic results of the calculation will be given, whereas Section 7 will provide the
numerical analysis. I end up with the conclusions.
2. Notation and basic formulas
2.1. Kinematics
For the process B we will assign the momenta p and q to the pions (Fig. 2) which
fulfill the condition
p 2 , q 2 = 0.
(2)
157
This is the leading power approximation in QCD /mb . Let us define two Lorentz vectors n+ , n
by:
n+ (1, 0, 0, 1),
n (1, 0, 0, 1).
(3)
In the rest frame of the decaying meson p can be defined to be in the direction of n+ and q to be
in the direction of n . Light cone coordinates for the Lorentz vector z are defined by:
z0 + z3
z0 z3
z ,
,
2
2
So one can decompose z into:

zp zq
z =
q +
p + z ,
pq
pq
z+

z 0, z1 , z2 , 0 .
(4)
(5)
such that
z p = z q = 0.
(6)
2.2. Colour factors

In our calculations we will use the following three colour factors, which arise from the SU(3)
algebra:
1
CN = ,
2
CF =
Nc2 1
2Nc
and CG = Nc ,
(7)
where Nc = 3 is the number of colours.

2.3. Meson wave functions
The pion light cone distribution amplitude is defined by
if
(p
/ 5 )
(p)|q(z)
[. . .]q (0) |0z2 =0 =
4
1
dx eixpz (x).
(8)
The ellipsis [. . .] stands for the Wilson line

1
[z, 0] = P exp

dt igs z A(zt) ,
(9)
which makes (8) gauge invariant. For the definition of the B-meson wave function B1 we need
the special kinematics of the process. Following [2] let us define

d 4 l ilz
e
B (l, pB ).
B (z, pB ) = 0|q (z)[. . .]b (0)|B(pB ) =
(10)
(2)4
In the calculation of matrix elements we get terms like:

d 4l
d 4l
d 4 z eilz tr A(l)B (z) .
tr A(l)B (l) =
(2)4
(2)4
(11)
158
We have to consider only the case that the amplitude A depends on l only through l p:
A = A(l p).
(12)
In this case we can use the B-meson wave function on the light cone which is given by [2]:

0|q (z)[. . .]b (0)|B(pB )
z ,z =0
1
ifB
=
(p
/ B + mb )5
4
+
d eipB z B1 ( ) + n
/ + B2 ( ) ,
(13)
where
1
1
d B1 ( ) = 1 and
d B2 ( ) = 0.
(14)
It is now straightforward to write down the momentum projector of the B-meson:

d 4l
tr A(2l p) (l)
(2)4
1

ifB
=
/ + B2 ( ) A m2B
tr( p
/ B + mB )5 d B1 ( ) + n
4
(15)
At this point we give the following definitions

mB
1
d
B1 ( ),
(16)
B
n
mB
1
d n
ln B1 ( ).
(17)
3. Hard spectator interactions at LO

The effective weak Hamiltonian we deal with is given by [11]:
GF
Heff = Vud
Vub [C1 O1 + C2 O2 ] + h.c.,
2
where
V A (pb)
V A ,
O1 = (dp)
O2 = (di pj )V A (p j bi )V A .
(18)
(19)
Explicit expressions for the short-distance coefficients Ci can be obtained from [11]. The decay
amplitude of B is given by
A(B ) |Heff |B.
(20)
For later convenience we define

A(B ) A(B )I + A(B )II ,
(21)
159
where AI (AII ) belongs to the first (second) term of (1). Because AI and AII contain different
hadronic quantities, the renormalisation scale dependence of both of them has to vanish separately. So we can set their scales to different values I and II . As in AI there occurs
only the
mass scale mb we can set I = mb . In AII there occurs also the hard-collinear scale QCD mb .
As we will see this scale is an appropriate choice for II .
Because we only deal with the tree amplitude and do not consider penguin contractions only
the matrix elements of the operators O1 and O2 are taken into account. These operators close
under renormalisation such that the corresponding hard spectator amplitude is independent of the
renormalisation scale.
The decay amplitudes of B can be written in terms of ai as follows [12]:

p

p
A B 0 + =
u a1 +
p a4 + r a6 A ,

2A B 0 =
u (a1 + a2 )A ,

p
p
A B 0 0 0 =
u a2 +
p a4 + r a6 A ,
(22)
where

GF
A = i m2B m2 f+B f
2
and
r () =
2m2
.
m
b ()(m
u () + m
d ())
(23)
For the LO and NLO results of the ai I refer to [12]. We define analogously to (21)
ai = ai,I + ai,II
(24)
AI
AII .
where the labels I and II refer to the contribution to

and
The leading order of the hard spectator interactions which start at O(s ) is shown in Fig. 2.
The hard spectator scattering kernel T II , which does not depend on the wave functions, can be
obtained by calculating the transition matrix element between free external quarks, to which we
assign the momenta shown in Fig. 2. The variables x, x 1 x, y, y 1 y are the arguments
of T II , which arise from the projection on the pion wave function (8). In the sense of power
counting we count all components of l of O(QCD ), while the components of p and q are
O(mb ) or exactly zero. We define the following quantities
lp
,
pq
lq
.
pq
(25)
We will see that in the end the dependence on vanishes in leading power such that we can
use (15).
We consider the three cases B 0 + , B 0 0 0 and B 0 . In the case, that the
external quarks come with the flavour content of B 0 + , the LO hard spectator amplitude
for the effective operator O2 reads:

(1)
Aspect. B 0 +

O2 d(l)b(p
xp)u(xp)
d(
u(
yq)d(yq)
+ q l) spect.
160
d(l) d(xp)
u(xp)
(1 5 )b(p + q l)
x
m2B

2/
p g
p
/
(1 5 )u(yq),
d(yq)
(26)
y
y y
where the quark antiquark states in the input and output channels of the matrix element form
colour singlets. The subscript spect. means that only diagrams with a hard spectator interaction
are taken into account. The amplitude of O1 vanishes to this order in s . In the case of B 0
0 0 we get the tree amplitude from the matrix element of O1 . The case B 0 does not
need to be considered separately, because from isospin symmetry follows [12,13]:

2A B 0 = A B 0 + + A B 0 0 0 .
(27)
= 4s CF Nc
On the other hand the full amplitude is the convolution of T II with the wave functions, given
by (1). To extract T II from (26) we need the wave functions with the same external states we
have used in (26), i.e., we have to calculate the matrix elements (8) and (10), where the pion or
B-meson states are replaced by free external quark states. To the order O(s0 ) we get
(0)

y d(z q)eizqy u(
yq)d(yq)|
di (z)ui (0)|0z ,z =0
= 2Nc (y
y)d (yq)u (yq),
+ (x
) = 2Nc (x
x)u (xp)d (xp),
+
(0)
B (l
) dz+ eil z 0|d (z)i b (0)i |d(l)b(p
+ q l)z ,z =0
(0)
= 2Nc (l
l )d (l)b. (p + q l).
By using
A(1)
spect.

=
(28)
(0)
II(1)
dx dy dl (0)+
(x)(0)
(y)B
(x, y, l )

,

(l )T
(29)
we finally obtain:
CF
1

(1 5 )

3
2
2
(2) Nc xm
B

2/
p g
p
/
(1 5 )
y
y y
T II(1) (x, y, l )

= 4s
(30)
It should be noted that only the first summand of the above equation contributes after performing
the Dirac trace in four dimensions. The second summand is evanescent. This will be important,
when we will calculate the NLO corrections of the wave functions (see Section 5.2).
If we plug the hadronic wave functions defined by (8) and (15) into (29), i.e., we calculate
the matrix element (26) between meson states instead of free quark states, we get for the LO
amplitude1 :
(1)
Aspect.
if 2 fB CF
= 2 4s
4Nc
1
dx dy d B1 ( ) (x) (y)
0
1
.
x y
(31)
1 A
spect. is used for the matrix elements of the operators Oi between both free external quarks and hadronic meson
states. It should become clear from the context what is actually meant.
161
Following (1) and the conventions of [3] we write our amplitude in the form:
1
Aspect.i = im2B
dx dy d TiII (x, y, )fB B1 ( )f (x)f (y),
(32)
where in the case of B + we define

Aspect.1 = O2 spect. ,
and in the case B
0 0
Aspect.2 = O1 spect.
(33)
we define
Aspect.1 = O1 spect. ,
Aspect.2 = O2 spect. .
(34)
Because we use the NDR-scheme which preserves Fierz transformations for O1 and O2 , TiII has
the same form for both decay channels. From (31) and (32) we get:
II(1)
T1
= 4s
CF
1
,
2
4Nc x ym
2B
II(1)
T2
= 0.
(35)
4. Calculation of Feynman diagrams with differential equations

In this section I will discuss the extraction of subleading powers of Feynman integrals with
the method of differential equations [1416]. This method will prove to be easy to implement in
a computer algebra system. The idea to obtain the analytic expansion of Feynman integrals by
tracing them back to differential equations has first been proposed in [14]. This method, which
is demonstrated in [14] by the one-loop two-point integral and in [15] by the two-loop sunrise
diagram, uses differential equations with respect to the small or large parameter, in which the
integral has to be expanded.
In contrast to [14,15] I will discuss the case where setting the small parameter to zero gives
rise to new divergences. In this case the initial condition is not given by the differential equation
itself and also cannot be obtained by calculation of the simpler integral that is defined by setting
the expansion parameter to zero. It is not possible to give a general proof, but it seems to be a
rule, that one needs the leading power as a boundary condition. An efficient way to calculate
the leading power of Feynman integrals is provided by the method of regions [1720], whereas
the subleading powers can be obtained from a differential equation. In the present section I will
discuss which conditions the differential equation has to fulfill in order for this to work.
Although the examples I use below are taken from the present calculation, this method is
very general and can be used in any case in which the expansion of Feynman integrals in small
parameters is needed.
4.1. Description of the method
We start with a (scalar) integral of the form

1
dd k
I (p1 , . . . , pn , m1 , . . . , mn ) =
,
(2)d D1 . . . Dn
(36)
where the propagators are of the form Di = (k + pi )2 m2i . We assume that there is only one
mass hierarchy, i.e., there are two masses m M such that all of the momenta and masses pi
162
m
and mi are of O(m) or of O(M). We expand (36) in M
by replacing all small momenta and
masses by pi pi and expand in . After the expansion the bookkeeping parameter can be
set to 1.
We obtain a differential equation for I by differentiating the integrand in (36) with respect
to . This gives rise to new Feynman integrals with propagators of the form 12 and scalar prodDi
ucts k pi in the numerator. Those Feynman integrals, however, can be reduced to the original
integral and to simpler integrals (i.e., integrals that contain less propagators in the denominator)
by using integration by parts identities.
Finally we obtain for (36) a differential equation of the form
d
I () = h()I () + g(),
(37)
d
where h() contains only rational functions of and g() can be expressed by Feynman integrals
with a reduced number of propagators. It is easy to see that h and g are unique if and only if I and
the integrals contained in g are master integrals with respect to IBP-identities, i.e., they cannot
be reduced to simpler integrals by IBP-identities. If I () is divergent in = 4d
2 , I , h and g have
to be expanded in :

Ii i ,
h=
hi i ,
g=
gi i .
I=
(38)
i
Plugging (38) into (37) gives a system of differential equations for Ii , similar to (37). In the next
paragraph we will consider an example for this case.
First let us assume that h() and g() have the following asymptotic behaviour in :

g() =
j g (j ) (ln ),
h() = h(0) + h(1) + ,
(39)
j
i.e., h starts at 0 , and we allow that g starts at a negative power of . We count ln as O(0 ) so
the g (j ) may depend on ln . This dependence, however, has to be such that
lim g (j ) (ln ) = 0.
(40)
The condition (40) is fulfilled, if the g (j ) are of the form of a finite sum
m

an lnn .
(41)
n=n0
The limit m however can spoil the expansion (39). E.g., e ln = 1 so the condition (40) is
not fulfilled, which is due to the fact that we must not change the order of the limits 0 and
m .
Further we assume that also I () starts at 0
I () = I (0) (ln ) + I (1) (ln ) +
and plug this into (37) such that we obtain an equation which gives I (i) recursively:

i1

i1
h(j ) I (i1j ) ln
+ g (i1) ln
.
i I (i) = d

j =0
(42)
(43)
163
I want to stress that, because h starts at O(0 ), (43) is a recurrence relation, i.e., I (j ) does not
mix into I (i) if j i. As the integral is only well defined if i 1, we need the leading power I (0)
as boundary condition and (43) will give us all the higher powers in . It is easy to implement
(43) in a computer algebra system, because we just need the integration of polynomials and finite
powers of logarithms.
A modification is needed if h starts at 1 , i.e.,
n
h = h(0) + .
By replacing I n I we obtain the differential equation

d
n
I=
+ h I + n g,
d
(44)
(45)
which is similar to (37) and leads to

i+n (i)

i+n1
i+n1

(j ) (i1j )
ln + g
(i1)
ln
(46)
j =0
which is valid for i 1 n. So, if I starts at O(n ), the subleading powers result from the
leading power.
4.2. Examples
We start with a pedagogic example:
Example 4.1.

1
dd k
,
I=
d
2
2
(2) k (k )(k 2 1)
(47)
where 1. The exact expression for this integral is given by:

I=

i
ln
i
ln 1 + + 2 + .
=
2
2
(4) 1 (4)
(48)
We see that I diverges for 0. As described,

e.g., in [20] we can obtain the leading power by
expanding the integrand in the regions k and k 1. This leads in the first region to

i
1
1
dd k
=
(1 + )
+ 1 ln
(49)
(2)d k 2 (k 2 )

(4)2
and in the second region to

i
1
1
dd k
=
(1
+
)
+
1
,
(2)d k 4 (k 2 1) (4)2

(50)
such that we finally obtain

I (0) (ln ) =
i
ln .
(4)2
(51)
164
This is the result we obtain from the leading power of (48). We write the derivative of I with
respect to in the following form:

dd k
d
1
1
(52)
I=
I
.
d
1
(2)d k 2 (k 2 )2
d
We obtained the right-hand side of (52) by decomposing d
I into partial fractions. Of course
this decomposition is not unique which is due to the fact that I itself is not a master integral but
can be further simplified by partial fractioning. From (52) and (37) we get:
1
= 1 + + 2 + ,
1

1
i
i 1
g=
(53)
+ 1 + + ,
=
2
2
(4) (1 ) (4)
such that the coefficients in the expansion in according to (39) do not depend on the power
label (k):
h=
i
.
(4)2
We obtain for the recurrence relation (43):
k1

1
i
(k)

k1
(k1j )
I
(ln ) +
.
I = k d
(4)2
h(k) = 1
and g (k) =
(54)
(55)
j =0
Using the initial value (51) it is easy to prove by induction

i
ln , k 0.
(4)2
This result coincides with (48).
I (k) (ln ) =
(56)
The first nontrivial example, we want to consider, is the following three-point integral:
Example 4.2.

1
dd k
.
I=
d
2
(2) k (k + un + l)2 (k + n+ + n )2
(57)
Here n+ and n are collinear Lorentz vectors, which fulfill n2+ = n2 = 0 and n+ n = 12 ,
u is a real number between 0 and 1 and l is a Lorentz vector with l 2 = 0 and l 1. Furthermore
we define
= 2l n+
and = 2l n .
(58)
We expand I in l, so we make the replacement l l and differentiate I with respect to . The

integral is not divergent in such that we obtain a differential equation of the form (37) where
the Taylor series of h() starts at 0 as in (39). In g() only two-point integrals occur, which
are easy to calculate. I do not want to give the explicit expressions for h and g because they are
complicated, their exact form is not needed to understand this example and they can be handled
by a computer algebra system. Because the leading power of I is of O(0 ), (43) gives all of the
subleading powers.
165
We obtain the leading power as follows: First we have to identify the regions, which contribute
at leading power. If we decompose k into
k = 2k n+ n + 2k n n+ + k ,
(59)
we note that the only regions, which remain at leading power, are the hard region k 1 and the
hard-collinear region
k n ,
k .
k n+ 1,
(60)
The soft region k leads at leading power to a scaleless integral, which vanishes in dimensional regularisation. In the hard region we expand the integrand to
1
k 2 (k
+ un
)2 (k
+ n+ + n )2
(61)
By introducing a convenient Feynman parametrisation we obtain for the (4 2)-dimensional

integral over (61):

1 ln(1 u) 1 2
i
(1
+
)
exp(i)
(1
u)
.
ln
(62)
u

2
(4)2
In the hard-collinear region we expand the integrand to
k 2 (k
1
.
+ un + n+ )2 (2k n+ + 1)
(63)
The integral over (63) gives:

i
1 ln(1 u)
1
(1 + ) exp(i)
+ 2Li2 (u) + ln2 (1 u)
2
u

2
(4)

+ ln u ln(1 u) + ln(1 u) ln .
(64)
Adding (62) and (64) together we get the leading power of (57):
I (0) =

i 1
2 Li2 (u) + ln u ln(1 u) + ln(1 u) ln .
2
(4) u
By plugging (65) into (43) we obtain I at O():

i 1
ln(1 u) ln
ln(1 u) ln u
2 Li2 (u)
(1)
2 + ln u +
+
+ ln +
I =
u
u
u
(4)2 u

ln u
ln(1 u) ln(1 u) ln
ln(1 u) ln u
+2
+
+
1u
u
u
u

ln
2 Li2 (u)
+
+
.
1u
u
(65)
(66)
Now we want to consider the following four-point integral

Example 4.3.

1
dd k
I=
,
(2)d k 2 (k + n )2 (k + l n+ )2 (k + l un+ )2
(67)
166
where we used the same variables, which were introduced in (57). This example is very special,
because in this case our method will allow us to obtain not only the subleading but also the
leading power in l. I is divergent in such that we obtain after the expansion (38) a system of
differential equations of the following form:
d
d
I1 = h0 I1 + g1 ,
I0 = h0 I0 + h1 I1 + g0 .
(68)
d
d
It turns out that in our example h takes the simple form
2 + 2
,
such that analogously to (45) we can transform (68) into

d 2
d 2
I1 = 2 g1 ,
I0 = 2I1 + 2 g0 .
d
d
This system of differential equations can easily be integrated to:
h=
(i)
I1
i+2
(69)
(70)
i+1 (i1)
g1 ,
0
(i)
I0
i+2
i+1
(i)
(i1)
2I1 + g0
(71)
where the superscript (i) denotes the order in as in (39) and (42). Both I1 and I0 start at
O(1 ). Because (71) is valid for i 1, it gives us the leading power expression, which reads:

i
2 1
ln u
(1)
=
(1 + )
1
ln ,
I
(72)
u
1u
(4)2
where = 2l n+ as in the example above. The exact expression for (67) can be obtained
from [21]. Thereby (72) can be tested.
In the last paragraph I want to return to Example 4.2. I will show how we can use differential
equations to prove that the integral (57) depends in leading power only on the soft kinematical
variable = 2l n and not on = 2l n+ . We need derivatives of the integral with respect
to and , which we have to express through derivatives with respect to l . These derivatives
can be applied directly to the integrand, whose dependence on l is obvious. We start from the
following equations:

I + 2 I,
n+ I =
l

I + 2 I,
n I =
l
l I = I + I + 2l 2 2 I,
(73)
l
l
which lead to

1
I = n+ + n + l I,
2
l

1
I = n+ n + l I,
(74)
2
l
167
where we have set l 2 = 0 in (74). Using (74) we can show that in leading power (57) depends
only on and not on . So we can simplify the calculation of the leading power by making the
replacement l n+ . The proof goes as follows: From (42) we see that the statement I (0)
does not depend on is equivalent to
I ( , ) = O().
(75)
Using the first equation of (74) we get
I ( , ) = O()I ( , ) + O().
(76)
Because we know (e.g., from power counting) that I ( , ) starts at 0 , (75) is proven.
5. Technical details of the NLO calculation
5.1. Evaluation of the Feynman diagrams
The diagrams that contribute to T1II at NLO are listed in Fig. 35. The diagrams are evaluated
in leading power in QCD /mb . We do not use an effective theory like SCET but evaluate the
diagrams in full QCD using free external quark states, to which we assign the momenta given
in Fig. 2. The Feynman integrals have to be evaluated in leading power in QCD /mb . After
reducing the number of Feynman integrals by integration by parts (IBP) identities [22,23], we
get the leading power of the master integrals using the method of regions (see, e.g., [20]). In
many cases the method of regions allows for a further reduction of master integrals. E.g., the first
diagrams in the second line of Fig. 5 comes with the scalar integral

1
dd k
.
(77)
d
2
2
(2) k (k + xp
l) (k + xp
+ yq
l)2 ((k + p + q l)2 m2b )
This integral can be calculated in leading power by setting = 0, i.e., we make the replacement
l q where and are defined in (25). This can be seen as follows: Counting soft momenta
as O() and hard momenta as O(mb ) the regions of space where (77) gives a leading power
contribution are
k ,
k p ,
k ,
k q mb .
k mb ,
In these regions l occurs only in the combination l p. So we can make the replacement l
q . It can be easily seen that this replacement allows to reduce (77) to three-point functions by
decomposing the integrand into partial fractions. Alternatively one can use the exact expression
(A.6) for the four-point integral with one massive propagator line, which is given in Appendix A.
After taking the leading power it can easily be seen that we get the same result as by just making
the replacement l q .
Fig. 3. Gluon self energy.
168
Fig. 4. Abelian diagrams.
Fig. 5. Non-Abelian diagrams.
dd k
Tensor integrals of the form (2)
d (k , k k , . . .)/Denominator can be reduced to scalar integrals using the methods of [24]. This reduction may make it necessary to calculate subleading
powers of Feynman integrals to obtain some diagrams in leading power. Furthermore, subleading
powers of Feynman integrals are necessary in some diagrams, where the leading power vanishes
169
because of the equations of motion. The methods of the last section, however, allow for an extraction of the subleading powers once the leading powers have been calculated.
It is instructive and helps to avoid mistakes to obtain the Feynman integrals in two independent ways. Instead of using arguments depending on power counting we can obtain the power
expansion of the integrals by calculating the exact integrals and expanding the results. The exact expressions of massless four-point integrals are given in [21]. Quite general expressions for
four-point integrals with one massive propagator are given in Appendix A. All of the integrals
that were used for the present calculation have passed this independent test.
5.2. Wavefunction contributions
5.2.1. General remarks
It has already been demonstrated in Section 3 how in principle we can extract the scattering
kernel T II of (1) from the amplitude if we know the wave functions. T II does not depend on the
hadronic physics and on the form of the wave function and B in particular, so we can get T II
by calculating the matrix elements of the effective operators between free quark states carrying
the momenta shown in Fig. 2. Because we calculate T II in NLO we need unlike as in Section 3
the wave functions up to NLO. Let us write the second term of (1) in the following formal way:
Aspect. = B T II .
(78)
All of the objects arising in (78) have their perturbative series in s , so (78) becomes
(1)
(0)
(2)
(0)
Aspect. = (0) (0) B T II(1) ,

(0)
Aspect. = (1) (0) B T II(1) + (0) (1) B T II(1)

(1)
(0)
+ (0) (0) B T II(1) + (0) (0) B T II(2) ,
(79)
where the superscript (i) denotes the order2 in s . In order to get T II(2) we have to calculate
(2)
(1)
(1)
Aspect. , and B for our final states. Then T II(2) is given by
(0)
(0) (0) B T II(2)

(2)
(0)
(0)
= Aspect. (1) (0) B T II(1) (0) (1) B T II(1)

(1)
(0) (0) B T II(1) .
(80)
At this point a subtlety occurs. Let us have a closer look to the factorization formula (1). By
calculating the first order in s of the partonic form factor F B,(1) , which is defined by free
quark states instead of hadronic external states, we see that it can be written in the form
(0)
(1)
F B,(1) = (0) B Tformfact. .

(1)
(81)
But Tformfact. is not part of T II . So we have to modify (80) insofar as we have to subtract the
right-hand side of (81) from the right-hand side of (80):
2 Note that the hard spectator scattering kernel starts at O( ). So we call T II(1) the LO and T II(2) the NLO.
s
170
Fig. 6. Example for diagrams which obviously belong to the form factor.
Fig. 7. NLO contributions to the meson wave functions. The dashed line stands for the eikonal Wilson line which makes
the wave functions gauge invariant.
(0)
(0) (0) B T II(2)

(2)
(0)
(0)
= Aspect. (1) (0) B T II(1) (0) (1) B T II(1)

(1)
(0)
(1)
(0) (0) B T II(1) (0) (0) B Tformfact. T I(1) .
(82)
In (82) we did not include the term F B,(2) T I(0) , because it is obviously identical with the
diagrams where the gluons do not interact with the emitted pion (e.g., those of Fig. 6). Those
diagrams where not considered in the last section. So we do not have to consider them here. The
wave functions for free external quark states are given at LO by (28). At NLO there exist three
possible contractions: The two external quark states can be connected by a gluon propagator
or one of the external quarks can be connected to the eikonal Wilson line of the wave function
(Fig. 7). The diagrams of Fig. 7(a), (b) and (c) give the order s of the pion wave function
for free quarks, i.e., we have replaced the pion final state (p)| in (8) by the free quark state
(1)
q
(xp)q(xp)|.
The Fourier transformed wave function (x

) is defined analogously to (28).
171
For the diagrams in Fig. 7(a), (b) and (c) respectively we get:
1
k/ x/
p
d d k (x x p+ ) (x x)
q (xp)
(2)d
k2k+

,
+ q
(xp)
(a),(1)
(x
) = 8 2 is CF Nc
(b),(1)
(x
) = 8 2 is CF Nc

+
q(xp)
dd k
(x x

(2)d

1
q
(xp),
k/ x/
p
k+
) (x
p+
k2k+
x
)
d d k (x x + p+ )
(2)d
k2

1
1
q(xp)
.
q (xp)
x/
p k/ k/ + x/
p
(c),(1)
(x ) = 8 2 is CF Nc
(83)
5.2.2. Evanescent operators

At NLO the convolution of the wave functions with the tree level kernel T II,(1) gives rise
to new Dirac structures, which, however, can in four dimensions be reduced to the tree level
Dirac structures. So we obtain the tree level Dirac structures plus further evanescent structures,
which vanish for d = 4 but give finite contributions if they are multiplied by UV-poles. We
define our renormalisation scheme such that we subtract the UV-poles and these finite parts of
the evanescent structures.
The tree level kernel (30) contains two Dirac structures where the second one is evanescent
(after the projection on the wave functions). We write T II(1) in the following form:

2p
/ g p
/
1
(1 5 )
T II(1) (x, y, l )
(84)
(1 5 ),
xl
y
y y
stands for the wrong contraction of the Dirac indices, i.e., the Dirac
where the symbol
indices are given by
1
2 3

= 1
2
3
,

(85)

as in (30). The right contraction is defined by the symbol , i.e., writing the Dirac indices
explicitly
1
2 3

= 1
2
3
.
(86)
In d = 4 the wrong and the right contraction are related by Fierz transformations. It is convenient
and commonly used to define the renormalised wave functions in terms of the right contraction,
q
(0) instead of q(z)
i.e., to define ren. by renormalising the operator q(z)
q
(0) . This is
5
why we define our renormalisation scheme such that only the UV-finite part of the right contraction operators remains: Using the notation of (84)(86) we define the following operators:
O0 (x, y, l )
1
2p
/
(1 5 ) (1 5 )
(1 5 ),
2l x
y
(87)
172
2p
/
(1 5 ),
y
/
p
1
(1 5 )
O2 (x, y, l )
(1 5 ).
l x
y y
O1 (x, y, l )
l x
(1 5 )

The matrix elements of these operators are defined analogously to (29):

Oi dx
dy
dl

(x
)
(y
)B
(l
)Oi

(x
, y
, l
).
(88)
(89)
(90)
Note that O1 + O2 is just the convolution of the tree level kernel (84) with the wave functions.
Furthermore by using Fierz identities it is easy to prove that we have in four dimensions
O0 = O1 ,
(91)
O2 = 0.
(92)
So we define the following evanescent operators:

E1 O2 ,
(93)
E 2 O 1 O0 ,

1
(2 d)2

E3
(1
)
+
(1
(1

5
5
5
x yl

2
2/
p(1 5 ),
1
(1 5 ) p

/ (1 5 ),
E4
xy
yl

(94)
(95)
(96)
where we have defined E3 and E4 for later convenience. Using those operator definitions we
define our renormalisation scheme such that we subtract the UV-pole of O0 and the finite
1
E, where E is an arbitrary evanescent structure.
parts of Ei , i.e., terms of the form UV
It is important to note that we do not subtract IR-poles, because they depend not only on the
operator but also on the external states the operator is sandwiched in between. They have to
vanish in (82) such that the hard scattering kernel is finite. Finally we obtain the same result as if
we had regularised the IR-divergences by small quark and gluon masses because the evanescent
structures vanish in d = 4. The renormalisation scheme defined above is the same scheme that
was used in [3].
In the next step we will calculate the convolution integral of T II,(1) with the NLO wave functions given by (83), i.e., we have to calculate the renormalised matrix elements of O1 + O2 at
NLO.
5.2.3. Wave function of the emitted pion
First we consider the renormalisation of the emitted pion wave function: Because the contri(a)
(b)
bution of the wave functions and does not change the Dirac structure of the operators,
we do not need to consider evanescent operators when we calculate the diagrams of Fig. 7(a),
(b). So for the emitted pion wave function these diagrams give after renormalisation:

ren.
(1),(a),(b) 2s
UV ln y + 2y
1
O1 + O2ren. emitted =
+ 2 ln
CF
O1 (0)
4
IR
IR
y

1
(0)
+ 2 ln IR (2 + ln y + ln y)O

(97)
,
2
IR
173
where the LO matrix elements Oi (0) can be obtained from (26). Note that we kept the IR-pole
times the evanescent matrix element O2 (0) . This is needed for consistency because we also
kept similar terms in the QCD-calculation of Aspect. . Furthermore it allows us to show that all
IR-divergences vanish.
The diagram in Fig. 7(c) mixes different Dirac structures. So we have to include evanescent
operators in the renormalisation. In the case of the emitted pion wave function the operator O1
does not mix under renormalisation with the evanescent operator E1 (93). We obtain for the
renormalised matrix element:

ren. (1),(c)
y ln y
2s
UV
1
+ 2 ln
CF
O1 (0) .
O1 emitted =
(98)
4
y
IR
IR
The matrix element of E1 however has an overlap with O1 :

s
1
UV
1
(1),(c)
(0)
E1 emitted =
+ 2 ln
CF
(2y ln y 2y ln y)E
1
4
UV IR
IR

y ln y
(0)
4 ln y +
(99)
O1
.
y
The renormalisation prescription tells us to subtract the UV-pole and the UV-finite part of E1 .
So we obtain after renormalisation:

ren. (1),(c)
s
UV
1
(0)
+ 2 ln
CF
(2y ln y 2y ln y)E
E1 emitted =
1
4
IR
IR

y ln y
+ 4 ln y +
(100)
O1 (0) .
y
y
(0)
Note that the evanescent operator E1 leads to a finite term 4(ln y + y ln
y )O1 , which we would
have missed if we had just dropped the evanescent operators.
5.2.4. Wave function of the recoiled pion

In the next step we consider the NLO contribution of the recoiled pion wave function. As
in the case of the emitted pion the diagrams Fig. 7(a), (b) do not lead to a mixing between the
operators. Therefore we get:

ren.
(1),(a),(b) s
2 ln x + 4x
UV
1
O1 + O2ren. recoiled =
+ 2 ln
CF
O1 (0)
4
x
IR
IR

1
+ 2 ln IR O2 (0) .
(101)
IR
Other than in the case of the emitted pion the operators O1 and O2 mix the spinors of the
recoiled pion and the B-meson. Therefore we have to work in the operator basis of O0 and
the evanescent operators and define our renormalisation scheme such that the finite parts of the
matrix elements of the evanescent operators vanish. The diagram Fig. 7(c) contributes to the
matrix element of the renormalised operator O0ren. :

ren. (1),(c)
s
UV x ln x
1
2 ln
O0 recoiled = 2 CF
(102)
O0 (0) .
4
IR
IR
x
In the case of the evanescent operators we keep the IR-pole:

ren. (1),(c)
1 s
1
x ln x
E1 recoiled =
(103)
+ 2 ln IR
CF
E4 (0) ,
2 4
IR
x
174
(1),(c)
E2ren. recoiled

1 s
1
x ln x
=
+ 2 ln IR
CF
E3 (0) .
4 4
IR
x
At the end of the day we obtain a contribution from diagram Fig. 7(c):
(1),(c)
ren.
O0 + E1ren. + E2ren. recoiled

x ln x
1 s
UV
1
1

=
2 ln

CF
2 4
x
IR
IR
xl

2p
/ g p
/ (0)
+ 8O1 (0) .
y
y y
(104)
(105)
The very complicated but also very explicit form, in which the above equation was given, is
(2)
rather convenient, because the QCD amplitude Aspect. (see (78)(80)) comes with the same Dirac
structure and cancels the IR-pole of (105).
5.2.5. Wave function of the B-meson
The s corrections of the wave function of the B-meson are given by the second row of Fig. 7.
For the diagrams (d), (e) and (f) respectively they read:

d d k (l
l k ) (l
l )
(d),(1)
B (l
) = 8 2 is Nc CF
(2)d
k2k

1
q(l)
b (p + q l),
k/ + /l

d d k (l
l k ) (l
l )
(e),(1)
B (l
) = 8 2 is Nc CF
(2)d
k2k

1
q (l)
b(p + q l) ,
k/ p
/ q/ + /l + mb

d
d k (l l k )
(f),(1)
1
q(l)
B (l
) = 8 2 is Nc CF
(2)d
k/ + /l
k2

1
b(p + q l) .
(106)
p
/ + q/ /l k/ mb
In the case of the B-meson only the diagrams in Fig. 7(d), (e) give rise to UV-poles. Those
diagrams however do not lead to a mixing of O0 and the evanescent operators and we do not
have to deal with evanescent operators.
First let us have a look at the convolution integral which belongs to the diagram in Fig. 7(f):

dd k
1
2 2
2 1
O1 + O2 (1),(f)
=
(4)
i
N
C
s c F
B
x
(2)d 2(k + l) pk 2
k/ + /l
p
/ + q/ /l k/ + mb
bqe
qc
qr qr (1 5 ) 2
2
(k + l)
k 2k (p + q l)

2/
p
p
/
(107)
g
(1 5 )qe
,
y
y y
(
)
(
)
where qc , qr and qe are the spinors carrying the flavour quantum numbers of the light constituent quark of the B-meson, the recoiled pion and the emitted pion respectively. In leading
175
Fig. 8. Two diagrams which correspond in leading power to the contribution of the B-meson wavefunction (107).
Fig. 9. s contributions to the form factor.
power (107) is identical to the contribution of the two diagrams shown in Fig. 8, which is given
by:

dd k
1
2 2
2
(4) is Nc CF
(2)d k 2 (k + l xp)
2
k/ + /l
p
/ + q/ /l k/ + mb
b
qc
qr qr (1 5 ) 2
(k + l)2
k 2k (p + q l)

y q/ + x/
p k/ /l
y q/ + x/
p k/ /l
qe
(108)
(1 5 )qe .
(yq + xp
k l)2
(yq
+ xp
k l)2
In (108) the leading power comes from the region where k is soft. In this region of space the
integrand gets the form of the integrand in (107), so both contributions cancel. As we did not
include the diagrams of Fig. 8 in the last section we can skip the contribution of (107) here.
The remaining contributions are the diagrams in Fig. 7(d) and (e). Together they read:
(1),(d),(e)
O1 + O2 B

1
p
/
2p
/
qc qr
qr (1 5 )bqe
g
qe
x
y
y y

UV
IR
1
1
2 2
2
+ 2 ln
+ 2 ln
2 ln .
(4 + 2 ln ) 2
+4
UV
mb
IR
mb
3
(109)
= s2 Nc CF2
5.2.6. Form factor contribution

Finally we have to calculate the contribution of (81). It is given by
(0)
(1)
Aformfact. (0) (0) B Tformfact. T I(1)

CF s (1),
(1)
qe (yq) (1 5 )qe
(yq)T
(y).
f
4
(110)
176
The form factor f (1), is the s correction of the matrix element

qr (xp)q
r (xp) qr (1 5 )b b(p + q l)qc (l) ,
where T (1) (y) is given by [12]:

1
2
1 2y
(1)
T (y) = 6
+ ln 2 18 + 3
ln y i

y
mb

2 ln y
(3 + 2i) ln y (y y)
.
+ 2 Li2 (y) ln2 y +
y
(111)
(112)
We get f (1), by evaluating the diagrams in Fig. 9 and obtain finally:

/l
xp
/ + q/ + 1
2
2 1

b
Aformfact. = s Nc CF qc qr qr (1 5 ) (1 5 )
x
x
qe (1 5 )qe
T (1) (y).
(113)
6. NLO results
6.1. Analytical results for T1II and T2II
After the analysis of the last chapter we finally obtain the O(s2 ) results for the hard spectaII which are defined by (32). Those expressions appear in convolution
tor scattering kernels T1,2
integrals with wave functions, where x, y and are the integration variables as defined in (32).
The ultraviolet divergences are renormalised in the MS-scheme. The infrared divergences drop
out after subtracting the wave function contributions from the amplitude. The infrared finiteness
together with the finiteness of the convolution integrals ensures that the framework of QCDfactorization works at this order in s .
II read:
The explicit O(s2 ) contributions for T1,2
II(2)
Re T1

s2 CF
16 ln x 40 ln mb
80
16 ln
+
+
N
3x y
3x y
3x y
9x y
4Nc2 m2B

4 ln x 4 ln y
30
4 ln
+ CF
+
+
+
ln
x y
x y
x y
x y
mb

2
ln
2 ln x 2 ln x
5
+ ln 2
x y
x y
x y
x y

2x 2
4x
2
2(5x 2)
2x
2
+
Li2 x
(y x)
3 (y x)
2 y x (y x)x y x 2
y x 2

2x 2
4x
2
4
2x
4
2
+
+
+
+
Li2 y
(y x)
3 (y x)
2 y x (y x)x x y x 2 y x 2

2(x 2) 2
+
+
Li2 (xy)
x
x 2 y

2x 2
4x
2
xy
+
Li2
x
(y x)
3 (y x)
2 y x

y x
2
2
2
2
2x
+
Li2
+ + 2 + 2
Li2 (x y)
+
(y x)x x y
y
x x y x y

x y
2x
2
Li2
+
(y x)x x y
x

2
2x
4x
2
x y
+
+
+
Li
2
y
(y x)
3 (y x)
2 y x

2
2(3x 2) ln x ln x
2
2
2
ln x ln y
ln x ln y +
+
+
+
y x x
x y x 2
x 2 y

2x 2
2
2
4x
2
2x
+
ln y ln y
+
+
(y x)
3 (y x)
2 y x x y x 2 y x 2

2 ln x ln y ln2 x ln2 y ln2 x 2 ln2 y

+
+
x y
x y
x y
x y
x y

2(3x 2)
3
2x
3
4 3x 3
ln x +
+
ln y
+
+
+ 2
x
x y
x 2 y
(y x)
2 y x x

2x
1 3x
9x 1
3
+
ln x
+
+
x x y
(y x)
2 y x
x 2 y x

4
4
4
1
2
+ 2
ln y +
ln(1 xy)
+
x y x y
x x y x x y

3x 1 3x 2 2x 1 3
2
2(x 2 1)
+ 2
+
2 +
ln(1 x y)
x x y x(y x)
x y x
x 2 (y x)
2

2 2(2 2 x + 63x 63)
2 x 2
2 2 x
2
+
+
+
+
3x
3(y x)
3 3(y x)
2 3(y x)
3y x 2

80 ln mb
2 ln x
1
22
+ 2
ln
CG
2
3x y
3x y
x y

2x
2(5x 2)
2x
2
4
+
+
+
2 Li2 x
+
(y x)x (y x)
2 y x
x 2 y
y x

2x
4
2
2x
4
2(x 3)
Li2 y
+
+
+
+
+
x 2 y
(y x)
2 y x (y x)x x y x 2

xy
2
4
2x
2(x 2) 2
Li2 (xy) +
+
Li2
+
+
+
x
x
x 2 y
(y x)
2 y x x

y x
2
2
2
2
2x
+
Li2
+ + 2 + 2
Li2 (x y)
+
(y x)x x
y
x x y x y

x y
2
2x
Li2
+
(y x)x x
x

x y
2
2x
4
Li2
+
y
(y x)
2 y x x

2
2
2 ln x ln y 2(3x 2) ln x ln x
+
ln y ln x
+
x y
x y x
x 2 y

2
2
2x
2 ln x ln y
4
2
ln y ln y
+
+
+
+
+
x 2 y
(y x)
2 y x x y x 2 y x 2
177
178
2 ln x ln y ln2 x ln2 y ln2 x ln2 y

+
+
x
x y
x
x y
x

3 5x
2
(4 3x) ln x
ln y
+ 2
y x
x 2 y
x y

2
31
2
2
2
+
ln x
+
ln(1 xy) +
x x y x x 2 y
xy x 3x y y x

2 ln y
2(x + 1)
2
2
ln(1 x y)
x(y x)
xy x x 2 y
y x 2

2
2 x
2 2
2(3 2 x + 166x + 3 2 166)
+
, (114)
3x
9x 2 y
3(y x)
2 3(y x)

x ln x
1
x 2
2s2 CF
2x
1
II(2)
+
ln y
C
+
Im T1
=
+
+
F
x 2 y
(y x)
3 (y x)
2 y x y x
4Nc2 m2B

x 2
x
1
2x
1
+
ln x
(y x)
3 (y x)
2 y x (y x)x y x

1
x
2
x
3
+
ln y
+
+
(y x)x x y
y x
(y x)
2 2(y x)

x ln x
1
x
2
ln y
CG 2 +
2
x y
(y x)
2 y x

x
1
x
2
+
ln x
+
+
(y x)x (y x)
2 y x x y

3
1
x ln y
+
+
,
+
(115)
(y x)x 2x y y x
s2 CF CN
4Nc2 m2B

12 ln mb
2x 2
2x
2
4x
2
+
Li2 x
x y
(y x)
3 (y x)
2 y x (y x)x y x

xy
2x 2
4x
2
2x Li2 y
+
Li
+
2
(y x)x
x
(y x)
3 (y x)
2 y x

y x
2
2x
+
Li2
+
(y x)x x y
y

2
2x 2
4x
2
+
Li2 y
+
+
+
(y x)
3 (y x)
2 y x y x

x y
2
2x
Li2
+
(y x)x x y
x

x y
2x 2
4x
2
Li
+
+
+
2
y
(y x)
3 (y x)
2 y x
Re T2II =
2 ln x ln y 2 ln x ln y ln2 y ln2 y (2 3x) ln x

+
+
x y
x y
x y
x y
x 2 y
179

x 2
2x
3
x
+
+
+
+
ln y
x 2 y 2
(y x)
2 y x x 2 y

3
2
2x
2
+
+
ln x
ln(1 xy) +
+
x x y x x 2 y
(y x)
2 y x

1
2(x 2 1)
1
3x 2 2x 1
+
+
ln(1 x y)
+
x 2 (y x)
x 2 x 2 y x(y x)
2 x x 2 y 2

2
16
3
2
,
ln y +
+
x y x y
x y

x 2
2s2 CF CN
2x
1
1
+
+
+
ln y
Im T2II =
(y x)
3 (y x)
2 y x y x
4Nc2 m2B

2x
1
x
1
x 2
ln x
+
(y x)
3 (y x)
2 y x (y x)x y x

3
x
3
x ln y
+
.
+
(y x)x (y x)
2y x
2 2(y x)
(116)
(117)
The s2 corrections of the hard spectator interactions have already been calculated in [3,4].
However both of these calculations have been performed in the framework of SCET, while my
result is a pure QCD calculation. In order to compare (114)(117) to [3,4] we have to take into
account the definition of B . The SCET calculation naturally uses the B defined by the HQET
field for the b-meson, while I define B by QCD-fields. Those two definitons differ at O(s ),
which has been discussed in [25]. The difference in the logarithmic moments of the B-meson
wave function does not play a role, because these moments occur first at NLO. Using the results
of [25] it is easy to figure out with the help of a computer algebra system, that (114)(117)
reproduce the results of [3,4].
6.2. Convolution integrals and factorizability
By looking at the hard scattering kernels of (114)(117) it is not obvious that there remain
no singularities in the convolution integrals over wave functions (32). It is however possible to
perform the integration analytically, which proves the factorizabilty.
Regarding the B-meson wave function we will obtain the result in terms of the quantities B
and n , which are defined in (16) and (17). The -meson wave function is given in terms of
Gegenbauer polynomials:

(3/2)
an Cn (2x 1) .
(x) = 6x x 1 +
(118)
n=1
Due to the symmetry properties of the pion the first nonvanishing moment is a2 . We neglect an
for n > 2 and using (32) we get for the NLO of Aspect. :

2
mB
(2)
2 if fB
CF
481 + 152
Aspect.1 = s
CN 120 ln
B
mb
4Nc2

92 + 54 + 6 2 1
+ CF (162 + 361 ) ln
mb

1566 1008
18 3
2
+
(3) + 27 + i 9 +
5
5
5
180

1
2101
CG 240 ln
+ 102 + 6 2 1 +
2
mb
5

18
1008
(3) + 18 2 + i 9 + 3
5
5

961 + 404
+ a2 CN 240 ln
mb

741
14809
2
182 +
+ 42 1
+ CF (174 + 721 ) ln
mb
2
35

45072
1362
(3) + 204 2 + i 338 +

3
35
35

22299 43992
1
+ 504 + 42 2 1 +
(3) + 161 2
CG 480 ln
2
mb
35
35

1482 3
+ i 292 +
(119)
35
and
if2 fB
mB
CF CN
B
4Nc2

1467 252
12 3
2
108 ln
+
+
(3) 6 + i 54
mb
10
5
5

40281
108 3
29268
2
+
+
(3) 112 + i 118
.
+ a2 216 ln
mb
140
35
35
(120)
The finiteness of the above equations proves factorization of the hard spectator interactions at
NLO.
(2)
(2)
Including the contributions of Aspect.1 and Aspect.2 the quantities a1,II and a2,II defined in (22)
and (24) are
(2)
Aspect.2 = s2
a1,II =
a2,II =
i
f f+B m2B
i
f f+B m2B

(1)
(2)
(2)
C2 Aspect.1 + C2 Aspect.1 + C1 Aspect.2 ,

(1)
(2)
(2)
C1 Aspect.1 + C1 Aspect.1 + C2 Aspect.2 ,
(121)
where
A(1)
spect.1 =

iCF s fB f2 mB
2
9
1
+
a
.
2
B
Nc2
(122)
7. Numerical analysis
7.1. Input parameters
For my numerical analysis I use the parameters given in Table 1. The decay constant fB
fB
and the ratio B
have been obtained by QCD sum rules in [27] and [28], respectively. The
f+ B
logarithmic moments 1 and 2 where calculated in [3] using model light-cone wave functions
181
Table 1
Input parameters, which were used in the numerical analysis. All parameters given without explicit citation can be found
in [34]. Unless otherwise stated scale dependent quantities are given at = 1 GeV
CKM-parameters
Vud [26]
0.974
|Vub /Vcb | [26]

0.09 0.025
Vcb [26]
0.041
Vcd
0.23
(70 20) deg
Parameters of the B-meson

mB
fB [27]
5.28 GeV
(210 19) MeV
fB
B [28]
f+
B
1.56 0.17
1 [3]
2 [3]
B 0
3.2 1
11 4
1.67 ps
1.54 ps
Parameters of the -meson

f+B [2931]
0.28 0.05
f
131 MeV
a1
0
m
130 MeV
a2 [32,33]
0.3 0.15
Quark and W-boson masses

mb (mb )
4.2 GeV
mt (mt ) [12]
167 GeV
mc (mb )
(1.3 0.2) GeV
MW
80.4 GeV
Coupling constants
(5)
MS
225 MeV
GF
1.16639 105 GeV2
for the B-meson [3538]. For the form factor f+B I use the value from [31], which has been
obtained by QCD sum rules. This value is consistent with quenched and recent unquenched
lattice calculations [29,30]. The first Gegenbauer moment of the pion wave function is zero due
to G-parity while the second moment has been obtained by lattice simulations [32,33].
7.2. Amplitudes a1 and a2
The QCD amplitudes a1 and a2 are defined in [12]. Their hard scattering parts a1,II and a2,II ,
i.e., the parts of a1 and a2 , which contribute to AII (see (21)), are plotted in Fig. 10 as functions
of the renormalisation scale . The strong dependence on of the real part of LO is reduced
at NLO. Taking the twist-3 contributions into account does not increase the -dependence too
much. The imaginary part, which occurs first at NLO, is strongly dependent on the renormalisation scale. An appropriate choice for the scale of the hard scattering amplitude is the hard
collinear scale
hc = 1.5 GeV.
(123)
In the following numerical calculations we will evaluate a1,II and a2,II at hc . The vertex corrections AI will be evaluated at
b = 4.8 GeV.
Using the parameters of Table 1 we obtain
a1 = 1.015 + [0.039 + 0.018i]V + [0.012]tw3 + [0.029]LO
+ [0.010 0.031i]NLO ,
(124)
182
Fig. 10. Contribution of the hard spectator corrections to a1 and a2 as a function of the renormalisation scale . The
upper two figures show the real part, where the LO is given by the dashed line, while the sum of LO and NLO is shown
by the thick solid line. The twist-3 corrections are included in the graph given by the thin solid line. The third figure
shows the imaginary part, which occurs first at O(s2 ). So no distinction between LO and NLO is made.
183
a2 = 0.184 + [0.171 0.080i]V + [0.038]tw3 + [0.096]LO

+ [0.021 + 0.045i]NLO .
(125)
These equations are given in a form similar to (61) and (62) in [3]. The first number gives the
tree contribution, the vertex corrections are indicated by the label V , the twist-3 contributions
are labelled by tw3. The hard scattering part is separated into LO and NLO. The hadronic input
parameters I used are slightly different from [3] and in contrast to [3] I evaluated all quantities,
which belong to the hard scattering amplitude, at the hard collinear scale hc . This is why the
values I get for a1 and a2 are different from [3].
The hard scattering amplitudes a1,II and a2,II together with their numerical errors read:
a1,II = 0.051 0.011(param.)+0.026
0.005 (scale) 0.012(tw3)

+ 0.031 0.008(param.)+0.024
0.031 (scale) 0.012(tw3) i,
a2,II = 0.15 0.03(param.)+0.01
0.04 (scale) 0.04(tw3)

+ 0.045 0.012(param.)+0.040
0.033 (scale) 0.038(tw3) i.
(126)
The first error comes from the error of the input parameters in Table 1. The scale uncertainty
is obtained by varying hc between 1 and 6 GeV. The error labelled by tw3 gives the error of
the twist-3 contribution. Within the scale uncertainty (126) is compatible with [3]. The result I
obtained in QCD comes with formally large logarithms ln QCD /mb . Without resummation these
logarithms might spoil perturbation theory. However the error coming from the scale uncertainty
in (126) as well as the relative size of the NLO contributions are small enough for perturbation
theory to be valid.
7.3. Branching ratios
The dependence of the CP-averaged branching ratios on the hard collinear scale is shown in
Fig. 11. It is obvious that the NLO corrections reduce this dependence significantly.
From the parameter set Table 1 we obtain the following CP-averaged branching ratios

+2.90
+0.18
106 BR B + + 0 = 6.05+2.36
1.98 (had.)2.33 (CKM)0.31 (scale) 0.27(sublead.),

+4.00
+1.07
+1.13
106 BR B 0 + = 9.41+3.56
2.99 (had.)3.46 (CKM)3.93 (scale)0.70 (sublead.),

+0.20
+0.17
+0.20
106 BR B 0 0 0 = 0.39+0.14
(127)
0.12 (had.)0.17 (CKM)0.06 (scale)0.08 (sublead.).
The origin of the errors are the uncertainties of the hadronic parameters and the CKM parameters, the scale dependence and the subleading power contributions, i.e., twist-3 and annihilation
contributions. The error arising from the scale dependence was estimated by varying b between
2 and 8 GeV and hc between 1 and 6 GeV. If we compare (127) to the experimental values [39]:

106 BR B + + 0 = 5.5 0.6,

106 BR B 0 + = 5.0 0.4,

106 BR B 0 0 0 = 1.45 0.29,
(128)
we note that BR(B + + 0 ) is in good agreement with the data. For B + + 0 and B 0
+ QCD-factorization is expected to work well, because at tree level Wilson coefficients
occur in the so-called colour allowed combination C1 + C2 /Nc 1, while B 0 0 0 comes
184
Fig. 11. CP-averaged branching ratios as functions of the hard collinear scale hc in units of 106 . In the graph with
the dashed line only the leading order of the hard spectator scattering is contained, while in the solid line hard spectator
scattering is taken into account up to NLO.
185
at tree level with C2 + C1 /Nc 0.2 such that subleading power corrections are expected to be
more important. On the other hand there are big uncertainties in the parameters occurring in
fB
the combinations |Vub |f+B , B
and a2 . In [40] and [3] these parameters were fitted by the
f+ B
experimental values (128) of BR(B + + 0 ) and BR(B 0 + ). Setting

a2 (1 GeV) = 0.39,
(129)
leads to

|Vub |f+B 0.80 |Vub |f+B default ,

fB
fB
2.89
.
f+B B
f+B B default
This leads to the following branching ratios:

106 BR B + + 0 = 5.5 0.2(param.)+0.5
0.3 (scale) 0.6(sublead.),

0
+0.8
+0.9
6
+
10 BR B = 5.00.9 (param.)0.2 (scale)+0.9
0.6 (sublead.),

0
+0.2
6
0 0
10 BR B = 0.77 0.3(param.)0.3 (scale)+0.3
0.2 (sublead.).
(130)
(131)
The uncertainties of the quantities that occurred in (129) and (130) have not been considered
in the estimation of the errors in (131). The B 0 0 0 branching ratio obtained in (131) is
compatible with the value obtained in [3]. Though it is too low, due to the theoretical and experimental errors it is compatible with (128).
8. Conclusions
QCD factorization has turned out to be an appropriate tool to calculate B decay modes from
first principles, because it allows us to disentangle systematically the perturbative physics and
the non-perturbative physics. The present calculation showed that the hard spectator scattering
amplitude factorizes up to O(s2 ), i.e., all infrared divergences cancel and there are no remaining endpoint singularities. The former point is obvious after the explicit calculation of T II and
the latter point was shown by evaluating the convolution integral (1) analytically. The explicit
expressions for the hard spectator scattering kernel (114)(117) confirmed the result of [3,4]. So
they are also a confirmation that the leading power of the amplitudes can be obtained by performing the power expansion at the level of Feynman integrals rather than at the level of the QCD
Lagrangian using an effective theory like SCET, which was done in [3,4].
The main challenges in the evaluation of Feynman integrals were due to the fact that the
Feynman integrals came with up to five external legs and three independent rations of scales. The
calculation of the Feynman integrals was made possible with the help of tools like integration
by parts identities and differential equation techniques: In Section 4 it was shown how to get
the expansion of Feynman integrals in powers of QCD /mb by differential equations once the
leading power is given.

Because next to the mb -scale also the hard-collinear scale QCD mb enters the hard spectator scattering amplitude, large logarithms could spoil perturbation theory. However the numerical
analysis showed a strong reduction of the scale uncertainty of T II . It also confirmed the observation of [3], that the NLO of T II is numerically important but small enough for perturbation
theory to be valid.
186
Finally it is important to note that in the present calculation the contributions of penguin contractions and the effective penguin operators were not considered. Actually they play a dominant
role in the branching ratios of B K and CP asymmetries of B and should be taken
into account in phenomenological applications. They have recently been published in [8]. Also
the O(s2 ) corrections of T I were not part of the present work. These contributions have been
calculated in [9,10].
Acknowledgements
I would like to thank Gerhard Buchalla for proofreading the drafts and comments on the manuscript. I am grateful to Guido Bell, Matthus Bartsch and Sebastian Jger for many instructive
and helpful discussions.
Appendix A. Massive four-point integral
We consider the following massive four-point integral in d = 4 2 dimensions (Fig. 12):

dd k
1
I4 (p1 , p2 , p3 , p4 ) = 2
(A.1)
,
d
(2) D1 D2 D3 D4
where
D1 = k 2 + i,
D2 = (k + p1 )2 + i,
D4 = (k + p1 + p2 + p3 + p4 )2 m2 + i.
D3 = (k + p1 + p2 )2 + i,
(A.2)
Following [21] we introduce the external masses

pi2 = m2i
(i = 1, 2, 3, 4)
(A.3)
and the Mandelstam variables

s = (p1 + p2 )2 ,
t = (p2 + p3 )2 .
(A.4)
Furthermore we consider only the case, where

m22 = 0 and m24 = m2 .
(A.5)
Fig. 12. Basic one-loop four-point integral. The massive line, which carries the mass m, is indicated by the thick line.
187
The integral (A.1) can be evaluated using the method of [21]. This paper gives explicit expressions for massless one-loop box integrals. It is however possible to extend the single steps of
this paper to our case.
So finally we obtain:
I4 (p1 , p2 , p3 , p4 )

I4 s, t, m21 , m23 , m2
=
(1 + )(42 )
i
(4)2 m2 (s m21 ) st + m21 m23

1
ln(s i) + ln m2 t i ln m2 m23 i ln m21 i

+ ln2 m2 m23 i + ln2 m21 i ln2 (s i) ln2 m2 t i

+ ln m2 i ln(s i) + ln m2 t i

ln m2 m23 i ln m21 i

m2 m23 i
m2 t i
2 Li2 1
+ 2 Li2 1
s i
m21 i

2

m

2
+ 2 Li2 1 m3 m + i f + 2 Li2 1 m21 + i f m

m
2
m
2 Li2 1 t m + i f 2 Li2 1 (s + i)f
(A.6)
,
where f m =
s+tm21 m23
2
m (m21 s)+stm21 m23
The case m21 = 0 gives rise to further divergences and has to be considered separately:

I4 s, t, m21 = 0, m23 , m2
=
i (1 + )(42 )
(4)2
s(m2 t)

1

3
1
2+
2 ln m2 t i ln m2 i

2
2

2

2
+ ln(s i) ln m m3 i

2 2 1 2 2
+ ln m i ln2 m2 t i + ln2 m2 m23 i ln2 (s i)
3
4
2

+ ln m i ln(s i) ln m2 m23 i

m2 m23 i
+ 2 Li2 1 m23 m2 + i f m
2 Li2 1
s i

2 Li2 1 t m2 + i f m 2 Li2 1 (s + i)f m ,
(A.7)
+
where f m =
s+tm23
.
s(tm2 )
188
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
M. Beneke, G. Buchalla, M. Neubert, C.T. Sachrajda, Phys. Rev. Lett. 83 (1999) 1914.
M. Beneke, G. Buchalla, M. Neubert, C.T. Sachrajda, Nucl. Phys. B 591 (2000) 313.
M. Beneke, S. Jger, Nucl. Phys. B 751 (2006) 160.
N. Kivelm, hep-ph/0608291.
C.W. Bauer, S. Fleming, D. Pirjol, I.W. Stewart, Phys. Rev. D 63 (2001) 114020.
C.W. Bauer, D. Pirjol, I.W. Stewart, Phys. Rev. D 65 (2002) 054022.
M. Beneke, A.P. Chapovsky, M. Diehl, T. Feldmann, Nucl. Phys. B 643 (2002) 431.
M. Beneke, S. Jger, Nucl. Phys. B 768 (2007) 51.
G. Bell, arXiv: 0705.3133 [hep-ph].
G. Bell, arXiv: 0705.3127 [hep-ph].
G. Buchalla, A.J. Buras, M.E. Lautenbacher, Rev. Mod. Phys. 68 (1996) 1125.
M. Beneke, G. Buchalla, M. Neubert, C.T. Sachrajda, Nucl. Phys. B 606 (2001) 245.
M. Gronau, O.F. Hernandez, D. London, J.L. Rosner, Phys. Rev. D 50 (1994) 4529.
E. Remiddi, Nuovo Cimento A 110 (1997) 1435.
M. Caffo, H. Czyz, S. Laporta, E. Remiddi, Nuovo Cimento A 111 (1998) 365.
T. Gehrmann, E. Remiddi, Nucl. Phys. B 580 (2000) 485.
S.G. Gorishnii, Nucl. Phys. B 319 (1989) 633.
M. Beneke, V.A. Smirnov, Nucl. Phys. B 522 (1998) 321.
V.A. Smirnov, Commun. Math. Phys. 134 (1990) 109.
V.A. Smirnov, Springer Tracts Mod. Phys. 177 (2002) 1.
G. Duplancic, B. Niic, Eur. Phys. J. C 20 (2001) 357.
K.G. Chetyrkin, F.V. Tkachov, Nucl. Phys. B 192 (1981) 159.
F.V. Tkachov, Phys. Lett. B 100 (1981) 65.
G. Passarino, M.J.G. Veltman, Nucl. Phys. B 160 (1979) 151.
V. Pilipp, hep-ph/0703180.
CKMfitter Group, J. Charles, et al., Eur. Phys. J. C 41 (2005) 1, updates at http://ckmfitter.in2p3.fr.
M. Jamin, B.O. Lange, Phys. Rev. D 65 (2002) 056005.
A. Khodjamirian, T. Mannel, N. Offen, hep-ph/0611193.
A. Abada, et al., Nucl. Phys. B 619 (2001) 565.
E. Dalgic, et al., Phys. Rev. D 73 (2006) 074502.
A. Khodjamirian, R. Rckl, S. Weinzierl, C.W. Winhart, O.I. Yakovlev, Phys. Rev. D 62 (2000) 114002.
M. Gckeler, et al., Nucl. Phys. B (Proc. Suppl.) 161 (2006) 69.
V.M. Braun, et al., Phys. Rev. D 74 (2006) 074501.
Particle Data Group, W.M. Yao, et al., J. Phys. G 33 (2006) 1, updates at http://pdg.lbl.gov.
A. Khodjamirian, T. Mannel, N. Offen, Phys. Lett. B 620 (2005) 52.
V.M. Braun, D.Y. Ivanov, G.P. Korchemsky, Phys. Rev. D 69 (2004) 034014.
A.G. Grozin, M. Neubert, Phys. Rev. D 55 (1997) 272.
S.J. Lee, M. Neubert, Phys. Rev. D 72 (2005) 094028.
Heavy Flavor Averaging Group (HFAG), E. Barberio, et al., hep-ex/0603003, updates at http://www.slac.stanford.
edu/xorg/hfag.
[40] M. Beneke, M. Neubert, Nucl. Phys. B 675 (2003) 333.

Regge behavior of gluon scattering amplitudes

in N = 4 SYM theory
Stephen G. Naculich a,,1 , Howard J. Schnitzer b,2
a Department of Physics, Bowdoin College, Brunswick, ME 04011, USA
b Theoretical Physics Group, Martin Fisher School of Physics, Brandeis University, Waltham, MA 02454, USA
Received 6 September 2007; received in revised form 29 October 2007; accepted 30 October 2007
Abstract
It is shown that the four-gluon scattering amplitude for N = 4 supersymmetric YangMills theory in the
planar limit can be written, in both the weak- and strong-coupling limits, as a reggeized amplitude, with
a parent trajectory and an infinite number of daughter trajectories. This result is not evident a priori, and
relies crucially on the fact that the leading IR-divergence and the finite log2 (s/t)-dependent piece of the
amplitude are characterized by the same function for all values of the coupling, as conjectured by Bern,
Dixon, and Smirnov, and proved by Alday and Maldacena in the strong-coupling limit. We use the Alday
Maldacena result to determine the exact strong-coupling Regge trajectory.
PACS: 11.30.Pb; 12.40.Nn; 11.25.Tq
1. Introduction and conclusion

In this note, we analyze the Regge behavior of the four-gluon scattering amplitude for N = 4
supersymmetric SU(N ) YangMills theory in the planar (large-N ) limit, using the conjectured
ansatz of Bern, Dixon, and Smirnov [1] and the recent strong-coupling results of Alday and
Maldacena [2] obtained via the AdS/CFT correspondence. (Other recent applications of this work
include Refs. [3,4]. Reggeization of the gluon in nonsupersymmetric YangMills theories [5] as
E-mail addresses: naculich@bowdoin.edu (S.G. Naculich), schnitzr@brandeis.edu (H.J. Schnitzer).

1 Research supported in part by the NSF under grant PHY-0456944.
2 Research supported in part by the DOE under grant DE-FG02-92ER40706.
doi:10.1016/j.nuclphysb.2007.10.026
190
S.G. Naculich, H.J. Schnitzer / Nuclear Physics B 794 (2008) 189194
well as supersymmetric YangMills theories [6] has long been a subject of interest.) The Regge
limit corresponds to center-of-mass energy squared u with fixed spacelike momentum
transfer s < 0, where s = (k1 + k2 )2 , t = (k1 + k4 )2 , and u = (k1 + k3 )2 are Mandelstam variables
obeying s + t + u = 0. We show that in the Regge limit the color-ordered four-gluon amplitude
approaches
(s)

u
A4 (s)
(1.1)
+
u
s
where the leading Regge trajectory has the form

1
1
s
1
(s) = 1 + f (1) () f () log
+ g()
4
4
2
2
(1.2)
and
(s) = (const)A4div (s)eC()
(1.3)
with representing an infinite sum of subleading trajectories. The functions (s) and (s), like
the scattering amplitude itself, exhibit infrared divergences, which we regulate using dimensional
regularization in d = 4 2 dimensions. The four-dimensional t Hooft coupling = g 2 N is
dimensionless, and a scale is introduced to allow the coupling to be defined away from four
dimensions. The functions f () and g() characterize the IR divergence of the amplitude [1,2,7]:
f () is proportional to the cusp anomalous dimension [8], and g() is the function G0 defined in
Ref. [1]. The form of g() is dependent on the choice of scale [2]. Finally f (1) () is defined
via

d
(1.4)
f (1) () = f ()
d
and Adiv (s) and C()

are defined in Eqs. (2.3) and (2.1).
We emphasize that the Regge behavior of A4 that we have demonstrated is not a priori evident
from the results of Refs. [1,2], and in fact appears inconsistent with the fact that the exponent of
the amplitude (2.1) goes as log2 (t/s), whereas Regge behavior would seem to require log(t/s)
dependence. The Regge behavior of the amplitude (1.1) only occurs because the function f ()
that characterizes the leading IR divergence also multiplies the finite log2 (s/t)-dependent piece
of the amplitude, as conjectured in Ref. [1].
The Regge trajectory function (1.2) and residue (1.3) are exact (to all orders in the coupling)
in the planar limit, depending only on the forms of the functions f () and g(). Since these
functions are known in the weak-coupling [1] and strong-coupling [2] limits, we may determine
the exact trajectory function explicitly in both these limits. To lowest order in , we have

1
s
(s) = 1 +
(1.5)
log
+ O 2 .
8 2
2
This is equivalent to the result found in Ref. [9], where a different regularization scheme was
used (see also Refs. [5,6]). The Regge trajectory function in the strong-coupling limit is

1
1
s
(1 log 2)
(s)
(1.6)
log
+
2 4
4
2
where we have used the results of Alday and Maldacena [2].
191
A linear Regge trajectory (s) s would imply stringy behavior, with string tension
1/ . But Eq. (1.2) goes as log(s/2 ), rather than linearly in s, suggesting that we are in
the = 0 or infinite-tension limit of a string theory, with no Regge recurrences. This is not unexpected since N = 4 super-YangMills theory is a conformal theory, without massive states.3
After this paper was typed, we became aware that similar conclusions were reached using
different methods in Ref. [4].
2. The BDS ansatz
To derive the result (1.1), we begin with the conjecture of Bern, Dixon, and Smirnov (BDS) for
the exact form of the scattering amplitude in the planar (large-N ) limit of N = 4 supersymmetric
SU(N ) YangMills theory [1]. In the kinematic region where s and t are both spacelike (s, t < 0),
the color-ordered planar four-gluon amplitude (expressed using the notation of Ref. [2]) is given
by

f ()
2
2
2 s
log
+ C() .
A4 = Atree Adiv (s)Adiv (t) exp
(2.1)
8
t
The tree amplitude is
Atree =
4iK
st
(2.2)
where the definition of K may be found in Ref. [10]. The IR divergent contributions Adiv (s) and
Adiv (t) are rendered finite in d = 4 2 dimensions, and take the form

1
1 (1) 2
2
g
Adiv (s) = exp 2 f (2)

(2.3)
(s)
4
(s)
8
where the functions f (2) () and g (1) () are related to f () and g() via
d
d
2
f (2) () = f (),

d
g (1) () = g().
d
(2.4)
To lowest order in , the functions in Eq. (2.1) are given by [1]

f () =

+ O 2 ,
2
2

g() = O 2 ,
C()
=
2

4
+ O 2
2
3
16
so that the weak-coupling scattering amplitude reads

2
4 2
2 s
A4 = Atree A2div (s)A2div (t) exp
+
+
O
log
t
3
16 2
(2.5)
(2.6)
with

Adiv (s) = exp

2
2
.
+
O
16 2 2 (s)
3 We thank L. Dixon for a correspondence on this point.
(2.7)
192
3. The strong-coupling limit

Alday and Maldacena subsequently computed the planar four-point amplitude at strong coupling using the AdS/CFT correspondence, obtaining the result [2]

2
2
2 s
log
+ C()
A4 = Atree Adiv (s)Adiv (t) exp
(3.1)
8
t
with

2
1
(1 log 2) 2
Adiv (s) = exp
4
(s)
2 2 (s)
and (correcting a small error)

C()
=
1
2 log 2 + (log 2)2
4
3
which is fully consistent with the BDS conjecture (2.1), with
(1 log 2)
,
g()
.
f ()

(3.2)
(3.3)
(3.4)
4. Regge limit of the BDS ansatz

The purpose of this note is to examine the four-gluon scattering amplitude in the Regge limit
of large u, with fixed s < 0. Since s + t + u = 0, this corresponds to the limit t of the
expression (2.1). The large-t behavior of Atree is given by
t
Atree (const) .
(4.1)
t
s
To extract the behavior of Adiv (t) as t , we expand in
2
2
2

1 2
(2)
(2)
(1)
2
f
() + f
() log
+ f () log
+ O 3 ,
=f

(t)
t
2
t
2
2

+ O 2 ,
= g (1) () + g() log
g (1)
(4.2)
(t)
t
where f (1) () is defined in Eq. (1.4). Hence

1 (1)
t
1
t
2
2
() log
Adiv (t) = Adiv (s) exp
f
+ g() log
4
s
2
s

1
1
2 t
2 s
f () log
+ f () log
.
8
8
2
2
Next, we rewrite the last term of Eq. (2.1) as

s
s
f ()
t
f ()
f ()
2 t
log2
log
log
log
+
+
C()
.
exp
8
4
8
2
2
2
2
(4.3)
(4.4)
Because both the leading IR-divergence and the IR-finite log2 (s/t)-dependent term are controlled by the same function f (), we observe that the leading log2 (t/2 ) dependence cancels
193
out of the full scattering amplitude, so that the large-t behavior is determined by the log(t/2 )dependent terms. Collecting the leading large-t contributions to the amplitude, we obtain
(s)
t
A4 (s)
(4.5)
t
s
where

s
1 (1)
1
1
(s) = 1 + f
() f () log
+ g()
4
4
2
2
(4.6)
and
(s) = (const)A4div (s)eC() .
(4.7)
Writing the result in terms of the center-of-mass energy squared u, we obtain for the colorordered planar four-gluon amplitude
(s)

(s)

u
u
1
A4 (s)
(4.8)
= (s)
+ .
u
s
s
The full planar four-gluon amplitude is then obtained by summing over color-ordered amplitudes
multiplied by the associated trace over gauge group generators. The function (s) then describes
the leading Regge trajectory, in the adjoint channel. Any subleading terms that survive would
represent (an infinite sum of) daughter trajectories, again in the adjoint channel.
References
[2] L.F. Alday, J. Maldacena, Gluon scattering amplitudes at strong coupling, JHEP 0706 (2007) 064, arXiv: 0705.0303
[hep-th].
[3] Z. Bern, J.J.M. Carrasco, H. Johansson, D.A. Kosower, Maximally supersymmetric planar YangMills amplitudes
at five loops, arXiv: 0705.1864 [hep-th];
S. Abel, S. Forste, V.V. Khoze, Scattering amplitudes in strongly coupled N = 4 SYM from semiclassical strings in
AdS, arXiv: 0705.2113 [hep-th];
E.I. Buchbinder, Infrared limit of gluon amplitudes at strong coupling, arXiv: 0706.2015 [hep-th];
A. Brandhuber, P. Heslop, G. Travaglini, MHV amplitudes in N = 4 super-YangMills and Wilson loops, arXiv:
0707.1153 [hep-th];
F. Cachazo, M. Spradlin, A. Volovich, Four-loop collinear anomalous dimension in N = 4 YangMills theory,
arXiv: 0707.1903 [hep-th];
L. Cornalba, M.S. Costa, J. Penedones, Eikonal approximation in AdS/CFT: Resumming the gravitational loop
expansion, arXiv: 0707.0120 [hep-th];
R.C. Brower, M.J. Strassler, C.-I. Tan, On the eikonal approximation in AdS space, arXiv: 0707.2408 [hep-th];
S.J. Brodsky, G.F. de Teramond, Light-front dynamics and AdS/QCD: The pion form factor in the space- and
time-like regions, arXiv: 0707.3859 [hep-th];
M. Kruczenski, R. Roiban, A. Tirziu, A.A. Tseytlin, Strong-coupling expansion of cusp anomaly and gluon amplitudes from quantum open strings in AdS5 S 5 , arXiv: 0707.4254 [hep-th];
Z. Komargodski, S.S. Razamat, Planar quark scattering at strong coupling and universality, arXiv: 0707.4367 [hepth];
L.F. Alday, J. Maldacena, Comments on operators with large spin, arXiv: 0708.0672 [hep-th];
A. Jevicki, C. Kalousios, M. Spradlin, A. Volovich, Dressing the giant gluon, arXiv: 0708.0818 [hep-th];
A. Mironov, A. Morozov, T.N. Tomaras, On n-point amplitudes in N = 4 SYM, arXiv: 0708.1625 [hep-th].
[4] J.M. Drummond, G.P. Korchemsky, E. Sokatchev, Conformal properties of four-gluon planar amplitudes and Wilson
loops, arXiv: 0707.0243 [hep-th].
194
[5] M.T. Grisaru, H.J. Schnitzer, H.-S. Tsao, Reggeization of YangMills gauge mesons in theories with a spontaneously broken symmetry, Phys. Rev. Lett. 30 (1973) 811814;
M.T. Grisaru, H.J. Schnitzer, H.-S. Tsao, Reggeization of elementary particles in renormalizable gauge theories:
Vectors and spinors, Phys. Rev. D 8 (1973) 44984509;
V.S. Fadin, E.A. Kuraev, L.N. Lipatov, On the Pomeranchuk singularity in asymptotically free theories, Phys. Lett.
B 60 (1975) 5052;
V.S. Fadin, R. Fiore, M.I. Kotsky, Gluon Regge trajectory in the two-loop approximation, Phys. Lett. B 387 (1996)
593602, hep-ph/9605357;
I.A. Korchemskaya, G.P. Korchemsky, Evolution equation for gluon Regge trajectory, Phys. Lett. B 387 (1996)
346354, hep-ph/9607229.
[6] M.T. Grisaru, H.J. Schnitzer, Dynamical calculation of bound state supermultiplets in N = 8 supergravity, Phys.
Lett. B 107 (1981) 196;
M.T. Grisaru, H.J. Schnitzer, Bound states in N = 8 supergravity and N = 4 supersymmetric YangMills theories,
Nucl. Phys. B 204 (1982) 267;
H.J. Schnitzer, Reggeization of N = 8 supergravity and N = 4 YangMills theory, hep-th/0701217.
[7] J.C. Collins, Sudakov form factors, Adv. Ser. Direct. High Energy Phys. 5 (1989) 573614, hep-ph/0312336;
L. Magnea, G. Sterman, Analytic continuation of the Sudakov form-factor in QCD, Phys. Rev. D 42 (1990) 4222
4227;
S. Catani, The singular behaviour of QCD amplitudes at two-loop order, Phys. Lett. B 427 (1998) 161171, hepph/9802439;
G. Sterman, M.E. Tejeda-Yeomans, Multi-loop amplitudes and resummation, Phys. Lett. B 552 (2003) 4856, hepph/0210130.
[8] S.S. Gubser, I.R. Klebanov, A.M. Polyakov, A semi-classical limit of the gauge/string correspondence, Nucl. Phys.
B 636 (2002) 99114, hep-th/0204051;
M. Kruczenski, A note on twist two operators in N = 4 SYM and Wilson loops in Minkowski signature, JHEP 0212
(2002) 024, hep-th/0210115;
A.V. Kotikov, L.N. Lipatov, A.I. Onishchenko, V.N. Velizhanin, Three-loop universal anomalous dimension of the
Wilson operators in N = 4 SUSY YangMills model, Phys. Lett. B 595 (2004) 521529, hep-th/0404092;
Y. Makeenko, P. Olesen, G.W. Semenoff, Cusped SYM Wilson loop at two loops and beyond, Nucl. Phys. B 748
(2006) 170199, hep-th/0602100;
B. Eden, M. Staudacher, Integrability and transcendentality, J. Stat. Mech. 0611 (2006) P014, hep-th/0603157;
A.V. Belitsky, Long-range SL(2) Baxter equation in N = 4 super-YangMills theory, Phys. Lett. B 643 (2006)
354361, hep-th/0609068;
Z. Bern, M. Czakon, L.J. Dixon, D.A. Kosower, V.A. Smirnov, The four-loop planar amplitude and cusp anomalous
dimension in maximally supersymmetric YangMills theory, Phys. Rev. D 75 (2007) 085010, hep-th/0610248;
F. Cachazo, M. Spradlin, A. Volovich, Four-loop cusp anomalous dimension from obstructions, Phys. Rev. D 75
(2007) 105011, hep-th/0612309;
M. Kruczenski, R. Roiban, A. Tirziu, A.A. Tseytlin, Strong-coupling expansion of cusp anomaly and gluon amplitudes from quantum open strings in AdS5 S 5 , arXiv: 0707.4254 [hep-th].
[9] H.J. Schnitzer, Reggeization of N = 8 supergravity and N = 4 YangMills theory II, arXiv: 0706.0917 [hep-th].
[10] Z. Bern, L.J. Dixon, D.C. Dunbar, M. Perelstein, J.S. Rozowsky, On the relationship between YangMills theory
and gravity and its implication for ultraviolet divergences, Nucl. Phys. B 530 (1998) 401456, hep-th/9802162.

A dual algorithm for non-Abelian YangMills coupled

to dynamical fermions
J. Wade Cherrington
Department of Applied Mathematics, University of Western Ontario, London, Ontario, Canada
Received 7 October 2007; received in revised form 31 October 2007; accepted 1 November 2007
Abstract
We extend the dual algorithm recently described for pure, non-Abelian YangMills on the lattice to
the case of lattice fermions coupled to YangMills, by constructing an ergodic Metropolis algorithm for
dynamic fermions that is local, exact, and built from gauge-invariant bosonfermion coupled configurations.
For concreteness, we present in detail the case of three dimensions, for the group SU(2) and staggered
fermions, however the algorithm readily generalizes with regard to group and dimension. The treatment of
the fermion determinant makes use of a polymer expansion; as with previous proposals making use of the
polymer expansion in higher than two dimensions, the critical question for practical applications is whether
the presence of negative amplitudes can be managed in the continuum limit.
PACS: 11.15.Ha
Keywords: Latice gauge theory
1. Background
Despite continued progress in algorithms and hardware, the inclusion of dynamical fermions
in lattice gauge calculations continues to incur significant computational expense. To motivate
our proposal for a novel fermion algorithm, we briefly review how dynamical fermions are currently addressed. Recall that dynamic fermions coupled to a gauge field on a D-dimensional
hypercubic lattice for D 2 are governed by an action of the form
S[ge , v , v ] = SG [ge ] + SF [ge , v , v ],
E-mail address: jcherrin@uwo.ca.
doi:10.1016/j.nuclphysb.2007.11.006
(1)
196
J.W. Cherrington / Nuclear Physics B 794 (2008) 195215
where the ge are valued in the gauge group G at the edges of the lattice and v are the fermion
fields defined at the vertices of the lattice.
Unlike gauge group variables, it is not practical to directly simulate Grassmann variables on
the computer. A common approach to dynamical fermion simulation starts by integrating out the
fermion variables appearing in SF [ge , v , v ], to give a function of the gauge variables known
as the fermion determinant (its specific form is reviewed in Section 2). The fermion determinant
can be combined with the kinetic part of the gauge boson amplitude eSG [ge ] to give an effective
action for the gauge variables from which simulations on a computer can in principle proceed.
However, the fermion determinant renders this effective action non-localit couples together
gauge variables that are arbitrarily distant in the lattice. This poses a considerable problem for
the simulation, since computing the change in the effective action due to a small change in any
variable becomes very expensive, growing prohibitively with increasing lattice volume. A variety
of algorithms have been devised to work with the fermion determinant; a description of some of
the methods commonly employed can be found for example in [1].
After reviewing the description of single component, staggered free fermions in terms of selfavoiding polymers (as was done for example in [2]), we review what happens when a similar
procedure is applied to multi-component fermion fields minimally coupled to gauge fields. In
this case each polymer configuration corresponds to a Wilson loop functional; i.e., the trace of
a product of representation matrices around the polymer. Because there is more than one component of the fermion fields in the non-Abelian coupled case, the strict self-avoiding constraint
of the single component case is weakened; that is, for an n-component fermion, up to n directed
polymer lines can enter and leave a given vertex. The picture has long been knownit is essentially that of a hopping parameter expansion of the fermion determinant, described for example
in [3]. Unlike many past applications, in the present case no cut-off in the power of the hopping parameter or otherwise is applied. Because we seek an exactly dual model, all polymers are
included in the configurations considered.
For each polymer diagram that arises in the free case, upon applying the duality transformation for the group-valued field the result is a sum of configurations consisting of all closed,
branched, colored surfaces (spin foams) with open one-dimensional boundaries defined by the
polymer diagram. The totality of spin foams associated with all polymer diagrams (including the
trivial empty polymer) defines the joint configuration space. Crucially, local changes to the dual
configurations (either polymer or surface structure) lead to local changes in the dual amplitude.
The two theoretical inputs for this construction, a polymer decomposition of the fermion determinant (i.e., hopping parameter expansion as described in [3]) and a dual non-Abelian model
(e.g., [4] and references therein), have been present in the literature for some time, and as we
shall see the construction of the joint dual model at the formal level is a rather straightforward
synthesis of these constituent models. However, unlike (the simplest implementations of) conventional lattice gauge simulations, finding any practical algorithm for a dual model has proven
somewhat non-trivial in the non-Abelian case for dimensions greater than two. The algorithm
proposed here builds upon the dual non-Abelian algorithm of [5] that has recently been tested in
the pure YangMills sector. In addition to pure spin foam moves, we construct a set of moves that
act on polymer structure and specify the type of vertex amplitudes that arise due to the charges
carried by the polymer. Currently, an implementation of this algorithm is being tested and will
be reported on in a forthcoming work.
For context, it should be noted that a similar picture was present in the work of Aroca et
al. [6] and Fort [7], which dealt with the Abelian case of U (1) and proposed using a Hamiltonian
that leads to a different Lagrangian formulation, where the ensemble is built from a restricted
197
subset of the configurations that arise in the KogutSusskind case. In future work, we believe the
non-Abelian generalization of [6] may be a very interesting alternative to the KogutSusskind
formulation used here, particularly if the imbalance between negative and positive amplitudes
and the reduction of species doubling described in [6] can be carried over to the non-Abelian
case.
The outline of the paper is as follows. In Section 2, we review the origin of the fermion determinant and discuss its expansion in terms of polymers. In Section 3, we briefly review the dual
computational framework for non-Abelian, pure YangMills theory on the lattice. In Section 4,
we show a natural way to combine these frameworks and formulate ergodic moves for the coupled fermionboson system. In Section 5, we offer some conclusions and describe our program
for ongoing numerical work based on this algorithm and its extensions. Appendix A describes
the vertex amplitudes that arise in the joint case, while Appendix B expands on Section 2 to
describe the trace structure of polymers with multiply occupied vertices.
2. Polymer description of fermions on the lattice
In this section, we start from a conventional lattice discretization of free fermions, following [8] in essentials and notation. In the usual manner, the fermion determinant is arrived at by
exact integration of the Grassmann variables. Following [2], the fermion determinant is then expanded into states, each of which is represented by a family of disjoint, closed oriented loops,
including trivial and degenerate loops, the monomers and dimers, respectively. A typical polymer configuration (in the D = 2 massive case) is illustrated.1 We now review explicitly how the
fermion determinant and polymer picture come about. Note this section is purely for pedagogical
purposes and to fix notation to be used later; those familiar with the hopping parameter expansion
of the fermion determinant can safely skip it.
2.1. KogutSusskind staggered fermionsfree case
To illustrate the concept and introduce terminology, we treat a single species of fermions with
no additional indices. We start with the naieve lattice field action for free staggered fermions

S=
(2)
x ( + m) x ,
xV
where are the (Euclidean) Dirac matrices and V is the set of lattice vertices.
Using the central difference for the partial derivative, this becomes

D

1
D
S=
a m( x x )
(x x+ x+ x ) ,
2a
xV
(3)
=1
where D is the dimension2 of the hyper-cubic lattice, m is the mass, a is the lattice spacing, and
labels one of the D directions of the lattice; is the unit lattice vector associated to the th
direction. Following [8], we change to a basis which diagonalizes the gamma matrices, rewriting
1 Fig. 1 shows the fermion state on part of a larger lattice to which periodic boundary conditions are applied.
2 In three dimensions, the continuum limit of the staggered fermion action contains flavours corresponding to two
inequivalent representations of the Dirac algebra [9].
198
Fig. 1. Part of a typical configuration in the polymer expansion of the fermion determinant in two dimensions (massive
case).
the free action as

D

x ( x x+ x+ x ) ,
M( x x ) K
S=
xV
(4)
=1
with x (1)x1 ++x1 for {1, 2, . . . , D} where the xi are the components of the latM
tice site four-vector and 2K
= ma. The edge dependent sign factor x arises from the chosen
diagonalization.
To express the result of integration over the Grassmann variables it is convenient to introduce
the quark matrix Q, defined in terms of lattice regularized action as

SF [ge , v , v ] =
(5)
y Q[ge ]yx x .
x,yV
For later reference, we include dependence on the gauge degrees of freedom in our definition
of Q. We next apply the well-known result [8] that the Grassmann integral over the fermion
fields at every vertex evaluates to

SF [ge ,v , v ]
dv d v e
=
dv d v e x,yV y Q[ge ]yx x = det Q[ge ],
vV
vV
(6)
the determinant of the quark matrix.

Next we recall the continuum form of the action for massive fermions coupled to a gauge
field. In the massive case, the quark matrix can be written as
Qyx = Myx Kyx ,
(7)
199
where M is the fermion mass and Kxy is the hopping matrix that is non-zero for nearest neighbor
pairs (x, y). By inspection of (4), we write the quark matrix as
Qyx = Myx K
D
x (y,x y,x
).
(8)
=1
We now apply a well-known identity for the determinant of (any) matrix Q,

sgn()
Qx(x) ,
det Q =
(9)
where ranges over the set of all permutations of the indices of Q and sgn() is the sign of
the permutation. In the case of the quark matrix Q, the matrix indices
being permuted correspond

to vertices of the lattice. Thus, one is led to consider the product x Qx(x) for every permutation
of lattice vertices.
The next step is to recognize that every permutation
can be decomposed into a composition
of disjoint, non-trivial cyclic permutations c , = i ic . For a matrix with all entries nonzero, these permutations may involve sets of vertices that are arbitrarily separated on the lattice.
However, the quark matrices that arise in practice have a very specific structure (originating in
the lattice discretization from nearest neighbor approximations of the derivative operator); the
only non-zero matrix elements consist of nearest-neighbor pairs (and in the massive case, ondiagonal). Thus, the non-zero contributions can be analyzed as follows.
A given permutation affects any vertex trivially (the vertex is sent to itself) or as part of
a non-trivial cyclic permutation. In the massive case, vertices that are permuted trivially give
monomer factors equal to the mass M. In the massless case where diagonal entries vanish, any
trivial permutation will lead to a vanishing contribution; thus every vertex must participate in a
cyclic permutation in the massless case.
For the non-trivial permutations, it is useful to distinguish two cases that a vertex may participate in. Permutations that swap a pair of neighboring vertices are referred to as dimers; all other
non-trivial permutations consist of non-trivial loops of edges on the lattice; we shall refer to these
as cycles. Given these observations on the structure of Q, we can now write an expression for
det Q in more explicit detail as

sgn()M Nm
Kxy ,
det Qyx =
(10)
(xy)p
where Nm is the number of monomers. The product is over all directed edges (xy) that are part
of a dimer or cycle of the permutation.
2.2. Coupled case
To couple fermions to the gauge fields, the ordinary derivative is replaced by the covariant
one, thereby introducing the gauge variables ge which act on the fermions through the matrices
of the representation corresponding to the charge of the fermion. For specificity, we will consider
fermions charged in the representation of G labelled by c. One can show [8] that the lattice action
for staggered fermions coupled to the gauge field becomes

D

x x Ux x+ x+ Ux x .
M(x x ) + K
S=
(11)
x
200
In contrast to the (simplified one component) free case, there is in general a vector possessing
multiple Grassmann variable components at each vertex, and matrices U (ge ) that act non-trivially
on this vector. In terms of our permutation expansion, the quark matrix now takes on component
as well as vertex labels as follows:
j
ji
SF [ge , v , v ] =
(12)
y Q[ge ]yx xi .
x,yV
Comparing (12) to (11) we identify the multicomponent quark matrix as

ji
Qyx [ge ] = Myx K
D
ij

.
x Ux
y,x (Ux )ij y,x
(13)
=1
The ge dependence is through the representation matrices Ue , where e is labelled in one of the
two conventions introduced above. The determinant formula can again be applied; upon doing
so, the expansion into permutations takes on the form

i (x,i)
det Q[ge ] =
(14)
sgn()
QxIV (x,i) .
x,i
Observe that the permutations now act on both vertex and component indices of Q; the action
of on indices and vertices can be separated into the maps I and V , respectively. As in the
free case, the locality structure allows one to identify non-vanishing, polymer-like contributions.
For where vertex index mapping is trivial, the only non-vanishing entries are those for which
I (x, i) = i as the massive term is an inner product with no cross-terms. Thus, the monomers
contribute factors Mn, where n are the number of components of the fermion vectors.
As in the single-component case, for vertices that participate in non-trivial permutations, one
still has that only permutations which move x (x) where (x) is a nearest neighbor are
non-vanishing. However, due to the presence of multiple components to the fermion field, permutations can shift both components at a vertex simultaneously.
Configurations which involve non-trivial shifts of more than a single component (multiply
occupied vertex contributions) are discussed in Appendix B. In the remainder of this section,
we will restrict ourselves to the case where only a single component participates in the shift, to
illustrate in a simple setting how the trace of a product of Ue matrices comes about.
By inspection of (11), we see for a given nearest neighbor vertex shift, all possible permuij
tations of indices are allowed, since in the general case Ux has all non-vanishing entries. To
continue our analysis we factor the permutation into a part that acts on vertices V (these correspond to the dimers and cycles of the free case) and a part that acts on component indices I .
We now write the fermion determinant as

i (x,i)
det Q[ge ] =
sgn(V )
QxIV (x) + (multiply occupied vertex contributions).
V
(15)
Focusing our attention on the first term, we note that only one component per vertex is shifted
in each of the products of the sum (multiple component shifts at a vertex are precisely what is
included in the second term, discussed in Appendix B). For a given V , we can represent I
as an ordered sequence of arrows through the discrete n-point space living above every vertex
acted on by V ; see Fig. 2. Note that for a given dimer or cycle V , all possible index sequences
correspond to permutations of the same order (thus sgn(V ) can be factored out). The final step
201
Fig. 2. Graphical representation of one (solid) of all possible permutations (dashed) associated with a cycle of vertices of
length L.
Fig. 3. Visualization of 2-component fermion field on 163 lattice. Polymers that appear to have open ends close on the
opposite side of lattice due to periodic boundary conditions.
in the analysis is to recognize that the sum over all paths is simply the trace of the matrix product
of representation matrices around the cycle.
We define D1 as the restriction of det Q[ge ] to permutations involving singly occupied vertices; that is, the first term of (15). D1 can be constructed out of loops with a single associated
trace as follows:

i (i)
D1 =
sgn(V )
QxIV (x)
V
sgn(V )(nM)Nm K Ne

V

i i
i i
i1 i2
iL i1
(xy) U(x
U 2 3 U 3 4 U(x
L x1 )
1 x2 ) (x2 x3 ) (x3 x4 )
(xy)V
Nm
sgn(V )(nM)
Ne

(xy)V

(xy)

U(xy) ,
Tr
(16)
(xy)V
where Ne is the total number of edges where vertices are shifted in the permutation. In the second
line, the Einstein summation convention for repeated indices is used. Observe that (xy) denotes
an oriented edge. Depending on whether (xy) is along or opposing the canonical orientation, one
. The visualization of a typical polymer configurahas either a product of U(xy) or U(yx) = U(xy)
tion on a D = 3 lattice (including those with multiply occupied vertices) appears in Fig. 3.
So far our discussion has been generic with regard to group, dimension, and fermion charge;
the only major choice has been staggered fermions rather than an alternative lattice discretization.
202
Fig. 4. Visualization of pure YangMills vacuum on 163 lattice.
In the remainder of this work, we restrict our attention to D = 3, G = SU(2), and n-component
massive fermion fields charged with half-integer spin c and minimally coupled to SU(2).
3. Spin foam description of pure gauge theory on the lattice
Having described what we shall refer to as the free (no coupling to gauge fields) fermion
partition function in terms of closed lattice polymers, in this section we briefly review the
dual formulation of pure (no coupling to fermions) YangMills, which leads to closed, colored,
branched lattice surfaces. We shall see in the next section that these pictures naturally combine
to give the full interacting partition function in terms of a space of coupled configurations.
It can be shown (see for example [4,10], and [5] for detail on G = SU(2) in three dimensions)
that starting from the lattice discretized action for pure YangMills

ZB =
(17)
dge e pP S(ge ) ,
eE
one can transform to a spin foam formulation expressing the partition function in terms of dual
variables as follows:

2 jp (jp +1)
v
e
1
18j (iv , jv )
N (ie , je )
e
(2jp + 1) ,
ZB =
(18)
j
i vV
eE
pP
here V , E, and P denote the vertices, edges, and plaquettes of the lattice, respectively. The
summations over i and j range over all possible edge and plaquette labellings, respectively.
A plaquette labelling j assigns an irreducible representation of SU(2) to each element of P .
These representations are labelled by non-negative half-integers (we will denote this set by 12 N),
also referred to as spins; a labelling j is thus a map j : P 12 N. In the SU(2), D = 3 case,
edges are also labeled by half-integer representations (see Fig. 4). Thus an edge labelling is a
map i : E 12 N.
Following [5], we define a (vacuum) spin foam configuration as one summand in (18), i.e.,
a labelling of both plaquettes and edges by spins and intertwiners, respectively. The amplitude
assigned to a spin foam factors into a local product of amplitudes. The vertex amplitude (18j
symbol) depends on the 12 plaquettes and 6 edges incident to a vertex; the Ne factors depend on
203
the 4 plaquettes and the intertwiner labelling of an edge, and there is a product of local plaquette
factors.
The locality of the spin foam formulation was applied in [5] to perform computations that
were verified against conventional methods.
4. Dual fermionboson simulations
In this section we describe how the dual pictures of lattice fermions and gauge bosons presented in the previous two sections can be combined to form a joint dual partition function, built
up of gauge-invariant configurations with discrete occupancy and representation labels.
4.1. The joint partition function
Using the action S[ge , v , v ] for the full theory we write the partition function as follows:

ZJ =
dge
dv d v eSF [ge ,v ,v ] eSG [ge ]
eE

vV
dge det Q[ge ]eSG [ge ] ,
(19)
eE
where we have integrated out the fermionic variables to get the fermion determinant. We next
use the polymer expansion for the determinant in the case of gauge-coupled Q, as described in
Section 2.2:

ZJ =
dge det Q[ge ]eSG [ge ]
eE

dge
eE

Nm
sgn( )(nM)
NK

(xy)

(xy)

U(xy) eSG [ge ] .
Tr
(xy)
(20)
Here NK is the number of K factors (one per unit of edge occupancy) in the polymer configuration; the definition of polymer configuration for D = 3, G = SU(2) is given in Section 4.3.1.
In the second line, we have substituted the polymer expansion for fermion determinant. Next,
we recall the form of the character expansion (see [5] and references therein) for the amplitude
based on the heat kernel action at a plaquette p,

2
1
1
eSp (g) =
(21)
(2j + 1)e 2 j (j +1) j (g), j = 0, , 1, . . . .
2
2
K(I, 2 ) j
Substituting the character expansion into the previous equation, we have

ZJ =
dge
sgn( )(nM)Nm K NK
(xy) Tr
U(xy)
eE
(xy)

2
(2jp + 1)e 2 jp (jp +1) j (g),
(xy)
(22)
pP jp
2
where an overall constant factor of K(I, 2 ) per plaquette has been discarded. We show in
Appendix A that the group integrals over products of traces and characters in each term of
204
the character expansion can be evaluated exactly in terms of charged 18j symbols, provided
a sum over intertwiner labels is made at each edge. Using the vertex and edge amplitudes of
Appendix A, we can exhibit the joint dual partition function as

ZJ =
18j v (iv , jv , )
N e (ie , je , )1
s( )
P j
vV
2 jp (jp +1)

(2jp + 1) ,
eE
(23)
pP
where s( ) sgn( ) (xy) (xy) (nM)Nm K NK combines the two sign factors and a product
of M and K factors. As we shall see in the next section, joint configurations associated to a
polymer carry in general three rather than a single intertwiner label ie for each edge belonging
to the polymer; i here ranges over all the intertwiner labels.
Although the overall dual amplitude is still a product of local amplitudes as in the pure
YangMills case, the presence of the Wilson loop functionals associated to non-trivial polymers requires the vacuum vertex and edge amplitudes to be modified in a way that we define in
Appendix A; the result is a product of modified 18j v symbols and edge amplitudes N e that are
charged according to the polymer content of the configuration.
The joint ensemble that results here can be viewed as a generalization of the usual definition
of spin foams to include one-dimensional structure corresponding to the presence of fermionic
charge. For a given polymer, there is a sum over all spin foams satisfying admissibility, which is
modified at the polymer edges. From the worldsheet point of view [11], a polymer loop acts as
the source or sink of 2c fundamental sheets, where c is the half-integer charge of the fermion.
4.2. The joint fermionboson configurations
In this section we define explicitly the set of configurations that include all those that give
non-zero contributions3 to the joint dual partition function.
Specifically, for a given polymer , we introduce definitions that will allow us to characterize
the set of spin foam configurations (plaquette colorings) that are admissible in the presence of .
As in the pure YangMills case [5], we assume a splitting has been made for each edge with
j1 and j2 on one side and j3 and j4 on the other. Because of the presence of charges c1 and c2 ,
the intertwiner is generally 6-valent and three splittings have to be made. For discussing edge
admissibility, we assume the splitting is such that c1 and c2 are in the middle of the channel as
shown in Fig. 5. Less symmetric splittings are possible, but we restrict our attention to this case
in the following.
Let denote a polymer configuration of charge c. We define the set of -admissible plaquette
configurations as those configurations whose labellings satisfy c-edge admissibility at every edge
in the lattice, where c-admissibility is defined as follows:
Definition 4.1 (c-edge admissibility). The spins assigned to plaquettes incident to an edge are
said to be c-edge admissible if the parity and triangle inequality conditions are satisfied. The
3 Due to exceptional zeros there may be configurations that are admissible by the conditions defined in this section,
but are nonetheless zero. As in the pure YangMills case [5], we assume exceptional zeros are sufficiently isolated that
ergodicity on admissibles is equivalent to ergodicity on non-zero configurations.
205
Fig. 5. Symmetric splitting of a 6-valent SU(2) spin network vertex.
charge insertions c1 and c2 may be 0 or c, according to whether the edge is uncharged, singly
charged, or doubly charged. Writing j1 , j2 , j3 and j4 for the four spins incident to a given edge,
these conditions are
(1) Parity:
j1 + j2 + j3 + j4 + c1 + c2
is an integer.
(2) Triangle inequality: for each permutation x (x1 , x2 , x3 , x4 , x5 , x6 ) of the charge and spin
variables (c1 , c2 , j1 , j2 , j3 , j4 ) we have
x 1 + x2 + x 3 + x 4 + x5 x 6 .
These conditions are equivalent to the existence of a non-zero invariant vector in the SU(2)
representation c1 c2 j1 j2 j3 j4 .
The allowed range of intertwiner labels i1 , ic and i2 depend on the incident spin labels, on
each other and on c through the vertices where the charges c1 and c2 enter the diagram. We now
state the condition for and admissible spin foam (plaquette and edge intertwiner labelling) in the
presence of an arbitrary polymer :
Definition 4.2 ( -admissible spin foam). A spin foam is -admissible if and only if for every
edge e E:
(1) The plaquettes incident to e are c-edge admissible in the sense of Definition 4.1, with c1 and
c2 assigned depending on the occupancy of e by .
(2) Each vertex of Fig. 5 is admissible. Explicitly, the following conditions are simultaneously
satisfied:
i2 + j1 + j2 ,
i1 + ic + c1 ,
i2 + ic + c2
and i1 + j3 + j4
are integers
and

i1 |j1 j2 |, j1 + j2 |ic c1 |, |ic + c1 | ,

ic |i1 c1 |, i1 + c1 |i2 c2 |, |i2 + c2 | ,

i2 |j3 j4 |, j3 + j4 |ic c2 |, |ic + c2 | .
For a given polymer , we denote the set of -admissible spin foams by FA .
206
4.3. The joint moves and algorithm

In this section we define moves that transform from one joint configuration to another. Together, they are ergodic and obey detailed balance and can thus be used in a Metropolis or other
Markov chain Monte Carlo algorithms.
Definition 4.3 (Pure spin foam move). A pure spin foam move consists of a single application of
the cube, edge, or homology move. In terms of their effect on plaquette spins, these moves are as
defined in [5]. However, their effect on intertwiner labels needs to be generalized to account for
the extra intertwiner labellings introduced by polymers.
Because polymer moves require simultaneous changes in spin foam structure, we use the term
pure to distinguish spin moves that leave the polymer structure unchanged. We now describe
the generalization of each pure spin foam move to account for extra intertwiner labels.
Definition 4.4 (Generalized cube move). As described in Definition 2.4 and Appendix A.3 of [5].
With reference to compatible intertwiner moves of type A, no changes in intertwiner labels are
necessary. For type B edges, all three labels are increased or decreased by the same half-unit of
spin.
Definition 4.5 (Generalized edge move). As described in Definition 2.5 of [5]. Rather than
changing the single intertwiner label, the three intertwiner labels i1 (e), i2 (e) and ic (e) are each
randomly changed by 2, 0, or 2 (to preserve parity) units of spin.
Definition 4.6 (Generalized homology move). As described in Definition 2.6 of [5], but all three
intertwiner labels are increased or decreased by one half-unit of spin.
4.3.1. Polymer moves
In Section 2.2, we saw how sums over permutations in can be encoded into traces of matrix
products, ordered according to the orientation of the permutation. Our example was restricted to
singly occupied vertices, and is generalized in Appendix B. Combining these cases, we see that
the sum over all permutations in can be represented by traces of products of matrices, if we include diagrams corresponding to all possible routings at multiply occupied vertices. This new set
of objects, oriented diagrams with routings at multiply occupied vertices, we refer to as polymers,
and denote by P. It is important to distinguish the polymers from their finer-grained constituents,
the permutations , as polymers can be coupled naturally into the spin foam partition function,
whereas individual permutations cannot be using the methods here.
In this section we present a set of moves that are ergodic on the space of polymers P on a
3-dimensional hypercubic lattice. The polymer states at an edge in the 2-component case considered here are as follows. Assuming a global orientation has been selected for the edges, an edge
can be unoccupied, singly occupied, or doubly occupied with the occupied cases carrying both
positive and negative orientation. The occupancy data at each edge can thus be assigned from the
set {2, 1, 0, 1, 2}. We shall use the term line of flux interchangeably with directed polymer
line.
Definition 4.7 (Plaquette move). A plaquette and plaquette orientation (clockwise or counterclockwise) is randomly selected. To each edge, a delta occupancy of +1 or 1 (with signs given
207
Fig. 6. Stretching of a loop by a plaquette move.
according to plaquette orientation) is assigned, and added to the present occupancy. If the resulting occupancy on any edge has magnitude greater than 2, the move is immediately rejected.
At each multi-valent vertex, a choice of routing is made with equal probability. A proposed
move that removes occupancy of edges incident on multiply occupied vertices must make further random choice of routing that matches the routing present or be rejected. This is necessary
to preserve detailed balance. If the move is not rejected, the spin of the selected plaquette is randomly decreased or increased by c to satisfy parity. Because a change in the polymer occupancy
forces a change in both the plaquette spin and the charge structure at an edge, the affected intertwiner labels (at each edge of the affected plaquette) must change in a way that is compatible; if
not the result will be immediately c-edge inadmissible for the new charging.
This is the most fundamental polymer-changing move, and connects a very large region of
the space of polymers contributing to the fermion determinant. One can see trivially that the
plaquette move can create fundamental loops of either orientation when applied to empty space
(plaquettes of zero occupancy edges). Fig. 6 illustrates how the plaquette move can deform an
existing cycle. The same plaquette move of opposite orientation would lead to one of multiple
routings with a doubly occupied edge, as shown by Fig. 10 in Appendix B.
Because an equally weighted routing choice is made amongst several alternatives when the
plaquette move of (for example) Fig. 10 is made, the move that reverses a particular routing to
get the initial loop back should occur with proportional probability, in order to satisfy detailed
balance. This will unfortunately lead to a lowered acceptance rate, particularly in regimes with
high occupancy.
To identify intertwiner moves compatible with a plaquette move, one must give the choice
of splitting (grouping of j1 , . . . , j4 into pairs) around the edges of a plaquette, in the same way
that intertwiner moves compatible with a cube move depends on the splitting (see Appendix A.3
of [5]). For conciseness and generality, we have made the present definition of the algorithm splitting independent. Forthcoming work on numerical simulations with this algorithm will evaluate
alternative splittings and describe the appropriate compatible moves.
Definition 4.8 (Global circle move). For integer valued charge c, a global circle moves adds a
single line of charge and a minimal cylinder of 2c charged plaquette spins spanning the lattice.
The origin and orientation of the cylinder is randomly selected to lie on a plaquette of one of
three orthogonal lattice planes through the origin.
For half-integer c, two lines of charge spanning the lattice are added and a minimal surface
consisting of a line of c-charged plaquettes is added between the charge lines. The position and
208
orientation of the sheet is randomly selected to lie on an edge of one of the three orthogonal
lattice planes through the origin.
As in the pure gauge theory case, the non-trivial global topology of lattices with periodic
boundary conditions leads one to move that create and destroy structure on a global scale, in this
case lines of charge.
In the half-unit charge case, a second line needs to be added to absorb the flux introduced by
the first; the smallest possible global move places a second line of half-unit charge immediately
beside the first.
In the case of integer valued charge, the lowest energy (and hence smallest change in amplitude) structure satisfying admissibility is formed by wrapping the smallest possible cylinder
supported by the lattice; because there are 2n fundamental irreps, n can be incident on one side
normal to the line, wrap into a cylinder, and enter the other to satisfy c-edge admissibility.
Definition 4.9 (Dimer move). An edge is randomly selected and a single unit of occupancy is
added or removed. If the result is not consistent with the presence of dimers and cycle edges
(i.e., a cycle that ran through the edge is broken by the removal of occupancy), the move is
rejected.
Singly and doubly charged dimers cannot be constructed by plaquette moves. Note a singly
charged dimer contributes Tr(Ue Ue ) = n with weight K 2 while a doubly charged dimer
contributes Tr(Ue Ue ) Tr(Ue Ue ) Tr(Ue Ue Ue Ue ) = n2 n with weight K 4 (unlike general
polymers, we combine the two routings into one configuration). Both types of dimers evaluate to constants with respect to the gauge variables, and thus do not couple to the gauge
bosons.
Definition 4.10 (Junction move). A vertex is randomly selected, and if multiply occupied, the
routing of flux is changed.
A multiply occupied vertex in the case of a 2-component fermion field has two flux paths,
which can be routed in two different ways. When a multiply occupied vertex first appears as a
result of a polymer move, one routing is randomly selected (similarly in the inverse case, with
weightings to preserve detailed balance as discussed above). Thus, the junction moves are strictly
speaking unnecessary for ergodicity, but may be used to improve performance of Metropolis
algorithm.
4.3.2. The algorithm
Combining the polymer and spin foam moves, given, we give a statement for a Metropolis
algorithm ergodic on the joint ensemble.
Algorithm 1 (Joint fermionboson algorithm). An iteration of the joint algorithm consists of
choosing one of the seven previously defined moves, which can be organized as follow:
209
The algorithm can be tuned to improve acceptance rate by adjusting the relative frequency of
the attempted move types.
The algorithm also tracks the sign of the configuration changes with the creation and destruction of fermion loops, which can occur with any of the polymer moves. The product of monomer
factors M and hopping factors K are also updated for polymer moves. It is important to emphasize that changes in both the sign and other factors require only local consideration of the
polymer moves.
With regard to locality, one sees that the (pure spin foam) homology move and global circle
moves will lead to updates costing on the order of L2 and L respectively, where L is a characteristic side length of the lattice. In the pure YangMills case analyzed numerically in [5], the
homology moves have negligible influence beyond very small lattice sizes. It remains to be seen
how this is modified in the joint case, and how large an influence the global circle moves have.
Within the scope of the current work, the expectation values of observables depending on
dual degrees of freedom are computed in the usual manner, by averaging the observable over the
Markov chain generated by the Metropolis algorithm. For Wilson loop type observables commonly studied, the expectation value is actually a ratio of dual charged and dual vacuum partition
functions, with static charge corresponding to the Wilson loop observable present in the charged
partition function and a vacuum partition function given by ZJ of Eq. (23). The computation
of Wilson loop observables of pure YangMills and dynamical fermions will be reported on in
forthcoming work by the author.
5. Outlook and conclusions
We present here a local, exact algorithm for Metropolis simulation of the fermionboson vacuum. The details have been provided for the case of D = 3 staggered KogutSusskind fermions
coupled to a YangMills SU(2) field; however the algorithm has a straightforward generalization
to other dimensions and gauge groups.
A limitation of the algorithm as currently given is the species doubling inherent in the
(unrooted) KogutSusskind formulation (e.g., in four dimensions there will be four species).
210
An alternative to KogutSusskind fermions which addresses species doubling was developed by

Aroca et al. [6] and Fort [7]. We are currently investigating this modified fermion action in the
non-Abelian, higher-dimensional context. Another approach would be to go through a similar
procedure using Wilson fermions, in which the unwanted doublers become very heavy in the
continuum limit.
The crucial question for any new fermion algorithm is its performance relative to the highly
evolved dynamic fermion methods that exist within the conventional lattice gauge community
today. In three dimensions, slow-down at weaker coupling has been observed in recent work on
the pure YangMills case [5]. The situation in D = 4 is not well understood and is currently the
subject of numerical work by the author, as are improvements in the original D = 3 case.
With regard to the continuum limit, we expect the most critical question for the algorithm
proposed here is the seriousness of the sign problem. A hard sign problem has been discussed
as a general feature of the polymer expansion in the continuum (small mass) limit [12]. While
oscillating signs can be overcome for lattice fermions in certain two-dimensional theories [12,
13], the author is not aware of methods that have successfully addressed the sign problem for
D > 2. As both the fermion and dual YangMills (spin foam) amplitudes can carry negative
signs, an important question is how the signs interact; i.e., the problem of signs may be harder or
easier than for either the free fermion polymer expansion or dual YangMills alone, depending
on how the signs correlate.
Although numerical developments are required to begin evaluating this proposal, we believe
the approach may be of considerable interest. We find it remarkable that the fermion expansion
into polymers and the gauge field dualization into spin foams (both of which have been extensively explored on there own), combine together in a way that is very compelling geometrically,
and allows a local, exact Metropolis simulation using gauge-invariant configurations carrying
entirely discrete labels.
Acknowledgements
The author would like to thank Dan Christensen, Florian Conrady, and Igor Khavkine for
valuable discussions. The author was supported by NSERC.
Appendix A. Charged nJ symbols
A.1. The dual model with charges
In this appendix, we deal specifically with D = 3, G = SU(2). Following the discussion in
the appendix of [5], we recall that the dual partition function (in the absence of charge) has the
form

Z=
(A.1)
dge
cjp jp (gp ),
{jp }
eE
pP
where summation over jp is over unitary irreducible representations of SU(2). At this point it is
convenient to specialize to a D = 3 cubic lattice with periodic boundary conditions; orientation
choice is as given in the appendix of [5]. With this choice of orientation, the holonomy around a
plaquette p is gp = g1 g2 g31 g41 , where g1 , g2 , g3 and g4 are the group elements associated to
the edges of the plaquette p, starting with an appropriate edge and going cyclically. Recall that
211
the inverse gi1 is used if the orientation of edge i does not agree with that of p. Thus

d

a
jp (gp ) = Ujp (g1 )ba Ujp (g2 )cb Ujp g31 c Ujp g41 d ,
(A.2)
where Uj (g)ba denotes a matrix element with respect to a basis of the j representation. If we
insert (A.2) into (A.1) and collect together factors depending on the group element ge , we get a
product of independent integrals over the group, each of the form

b

b
dge Uj1 (ge )ba11 Uj2 (ge )ba22 Uj3 ge1 a3 Uj4 ge1 a4 =
3

dge
(A.3)
Here and below we use a graphical notation for tensor contractions, as in [5].
Eq. (A.3) defines a projection operator on the space of linear maps j4 j3 j1 j2 , so it
can be resolved into a sum over a basis of intertwiners Ii : j4 j3 j1 j2

=
dge
Ii I

i
=
Ii , Ii
i
(A.4)
where the intertwiners Ii : j1 j2 j4 j3 are chosen such that the trace Ii
, Ii of the composite Ii
Ii is zero whenever i
= i and non-zero if i
= i. The projection property is readily
verified.
We next define Z , the partition function charged according to the polymer , as follows
Z =

{jp }
eE

i+
dge Tr
Uce (ge )ie
cjp jp (gp ).
e
(A.5)
pP
Collecting matrix factors by dependence on edge variable ge , we find in addition to the matrices from the four incident plaquettes, a matrix from the edge e with charge c belonging to the
polymer 4 :

dge Uj1 (ge )ba11 Uj2 (ge )ba22 Uj3
b

b
i+
ge1 a3 Uj4 ge1 a4 Uce (ge )ie
3
4
e

=
dge
. (A.6)
4 In the case where an edge is doubly occupied, the integral involves a sixth matrix (and the resolving intertwiners
an additional input and output arrow). It is straightforward to generalize the present analysis to this case; the resulting
doubly charged 18j symbols are shown in Appendix B.
212
As in the pure case, the group integral can be resolved into invariant intertwiners

dge
Ii I

i
=
Ii , Ii
i
(A.7)
If, for each edge of the lattice, we fix a term i in the above summation, the intertwiners Ii and
Ii can be contracted with those coming from the other edges, leading to a sum over intertwiner
labellings at every edge. Observe that at edges occupied by polymers, there is more than a single
intertwiner spin label due to the additional splittings (see Fig. 5) introduced by the charge lines.
At each vertex of the lattice, there will be six intertwiners Ii (some carrying multiple labels), and
their contraction can be graphically represented as an octahedral network plus additional lines
depending on how the polymer passes through the vertex. As well, at each edge there will be a
normalization factor corresponding to the denominator of Eq. (A.7).
We consider first the vacuum case. In this case, each edge carries only a single intertwiner
label. The result is the 18j symbol central to pure YangMills spin foams,
(A.8)
The vertices are labelled by the directions of the associated lattice edges emanating from the
given lattice vertex, namely x, y, and z. The value of the 18j symbol depends on the
choice of basis elements Ii and Ii
in (A.7), the six summation indices i labelling the edges, and
the 12 incident plaquette labels j .
We now turn to the case where there is a (single) polymer along one or more of the edges
incident to a vertex. Each charged normalization factor N Ii , Ii in the denominator of (A.4)
depends on the charge c of the fermion at that edge, the intertwiner labels i on that edge, and
the labels of the four plaquettes incident on that edge. In the presence of external charges, the
vacuum 18j is modified depending on whether the line of charge proceeds directly through the
vertex or turns, leaving in a direction perpendicular to entry direction. We call these cases charged
18j symbols and denote them by an overline, 18j (jv , ie , ) where the additional dependence
on gamma reflects how the polymer charge is routed through the octahedral network. Typical
charged 18j symbols are shown in Fig. 7. Cases where the direction of the arrow on the charged
lines is flipped will also occur, depending on the polymer orientation. Additionally, as discussed
in the next section, more than a single line of flux can pass through a vertex, leading to charged
18j symbols of the form shown in Figs. 8 and 11.
213
Fig. 7. Charged 18j symbols with flux lines passing through at right angles (left) and straight through (right).
Fig. 8. Charged 18j symbols with two pairs of flux. Cases with flux not in the same plane are also possible.
In implementing numerical code for this algorithm, a choice of splitting (grouping of the
four plaquettes into (j1 , j2 ) and (j3 , j4 ) pairs on opposite sides of the splitting) is made and
each vertex resolved into a 3-valent sub-network with up to three non-trivial intertwiner labels,
as shown in Fig. 5. At this point, recoupling moves (see Appendix A.2 of [5], and references
therein) can be used to reduce the spin network to sums and products of know spin networks
such as the 6j and theta networks, for which efficient algorithms are available. It should be
noted, however, that different splittings lead to differing efficiency in implementation, so some
care and experimentation should be applied to finding an efficient splitting. Specific splitting
schemes and their performance evaluation will be reported on in forthcoming numerical work by
the author and collaborators.
Appendix B. Polymers with multiply occupied vertices
In order to couple to the spin foam representation of the gauge theory, we seek to collect
the permutation contributions to the fermion determinant into traces of products of Ue matrices
around closed, oriented loops of edges. The case of permutations where a single component is
shifted was discussed above in Section 2.2. For polymers where more than one component is
shifted at a vertex, recovering a trace formula is somewhat more subtle.
In Fig. 9, we illustrate a case where a vertex is multiply occupied; the orientation is such that
there are two possible routings that resolve the ambiguity at that vertex. Neither diagram by itself
corresponds to the desired sum of permutation contributions. In the matrix multiplications and
traces (viewed as a sum over all paths around a loop), there are terms in each corresponding to
214
Fig. 9. Two routings associated with a polymer that self intersects once at a point.
Fig. 10. A move introducing a doubly occupied edge, for which there are two distinct routings.
Fig. 11. A charged 18j symbol containing a 6-valent node incoming from a doubly charged edge.
paths that are not permutations. However, the same undesired terms occur with opposite sign in
the two diagrams (as one involves paths that form a single loop, the other paths that lie in two
disjoint loops) so the sum of both captures the sum of permutations associated with the polymer.
A similar cancellation occurs when two loops share a single edge, i.e., the edge is multiply
occupied. Because there are two multiply occupied vertices, there are 22 = 4 routings possible,
215
however only two are topologically distinct; two representatives appear in Fig. 10. The cancellation of unphysical paths between the traces over differently routed polymers is well known from
the hopping parameter expansion (HPE) of the fermion determinant as discussed for example
in [3]. As shown in Fig. 11, the presence of a doubly charged edge leads to a charged 18j spin
network with a 6-valent node. As well, the charged normalization factor N on a doubly charged
edge is as given in the denominator of Eq. (A.4), but with an additional c-charged line parallel to
the original c-charged line.
References
[1] T. DeGrand, C. DeTar, Lattice Methods for Quantum Chromodynamics, World Scientific, Singapore, 2006.
[2] M. Karowski, R. Schrader, H.J. Thun, Monte Carlo simulations for quantum field theories involving fermions,
Commun. Math. Phys. 97 (1) (1985) 529.
[3] H.J. Rothe, Lattice Gauge Theories: An Introduction, World Scientific, 1992.
[4] R. Oeckl, H. Pfeiffer, The dual of pure non-Abelian lattice gauge theory as a spin foam model, Nucl. Phys. B 598
(2001) 400426.
[5] J.W. Cherrington, J.D. Christensen, I. Khavkine, Dual computations of non-Abelian YangMills on the lattice,
arXiv: 0705.2629 [hep-lat], Phys. Rev. D (2007), in press.
[6] J.M. Aroca, H. Fort, R. Gambini, Worldsheet formulation for lattice staggered fermions, hep-lat/9607050.
[7] H. Fort, The worldsheet formulation as an alternative method for simulating dynamical fermions, Phys. Lett. B 444
(1997) 174178.
[8] I. Montvay, G. Munster, Quantum Fields on the Lattice, Cambridge Univ. Press, Cambridge, 1994.
[9] C. Burden, A.N. Burkitt, Lattice fermions in odd dimensions, Europhys. Lett. 3 (1987) 545552.
[10] F. Conrady, Geometric spin foams, YangMills theory, and background-independent models, gr-qc/0504059.
[11] F. Conrady, I. Khavkine, An exact string representation of 3d SU(2) lattice YangMills theory, arXiv: 0706.3423
[hep-th].
[12] I. Montvay, Simulation of staggered fermions by polymer algorithms, in: Probabilistic Methods in Quantum Field
Theory and Quantum Gravity, Plenum Press, New York, 1990.
[13] U. Wolff, Cluster simulation of relativistic fermions in two spacetime dimensions, arXiv: 0707.2872.

N = 2 quiver gauge model and partial supersymmetry

breaking
H. Itoyama a,b , K. Maruyoshi a, , M. Sakaguchi c
a Department of Mathematics and Physics, Graduate School of Science, Osaka City University, Japan
b Osaka City University Advanced Mathematical Institute (OCAMI), 3-3-138 Sugimoto, Sumiyoshi-ku,
Osaka 558-8585, Japan

c Okayama Institute for Quantum Physics, 1-9-1 Kyoyama, Okayama 700-0015, Japan
Received 27 September 2007; accepted 5 November 2007

Abstract
We construct an action of N = 2 affine An quiver gauge model having non-canonical kinetic terms and
equipped with electric and magnetic FI terms. N = 2 supersymmetry is shown to be broken to N = 1
spontaneously and N = 1 multiplets realized on the vacua are given. We also mention the models with
different gauge groups. It is argued that the affine A1 quiver gauge model provides a dynamical realization
to approach the KlebanovWitten N = 1 fixed point.
PACS: 11.30.Pb; 12.60.Jv
Keywords: Quiver gauge theory; Partial supersymmetry breaking; N = 2 supersymmetry
1. Introduction
Supersymmetry has become one of the most remarkable and attractive ideas in theoretical
physics. In particular, various investigations beginning with [1,2] have been made on N = 2
supersymmetric YangMills theory in four dimensions, taking advantage of its powerful properties. Furthermore, we can extract the important information of N = 1 super-YangMills theory,
E-mail addresses: itoyama@sci.osaka-cu.ac.jp (H. Itoyama), maruchan@sci.osaka-cu.ac.jp (K. Maruyoshi),

makoto_sakaguchi@pref.okayama.jp (M. Sakaguchi).
doi:10.1016/j.nuclphysb.2007.11.001
H. Itoyama et al. / Nuclear Physics B 794 (2008) 216230
217
such as the low energy effective superpotential, breaking N = 2 supersymmetry to N = 1 by

a superpotential [24].
On the other hand, in view of the fact that superstring theories produce, in some backgrounds,
extended supersymmetry in four dimensions and have no adjustable parameter, it is natural to
consider spontaneous breaking of the extended supersymmetry so as to obtain more realistic
N = 1 supersymmetric models. Although it had been argued that, based on the supercharge
algebra, rigid N = 2 supersymmetry is not spontaneously broken to N = 1, a loophole has been
first pointed out in [5] by the argument based instead on the supercurrent algebra which has
been modified by an additional spacetime independent term. In [6] and [79], N = 2, U (1) and
U (N ) gauge models with N = 2 vector multiplet only have been constructed, establishing this
modification of the algebra by introducing magnetic FayetIliopoulos (FI) term. It was shown
that the partial breaking of N = 2 supersymmetry indeed occurs in such models. (See also [10
12] for related discussions and [13] for supergravity.)
In the U (N ) gauge theory which contains only the N = 2 vector multiplet, the magnetic FI
term which causes the partial breaking can be easily introduced in the harmonic superspace formalism [14] (see [15] for a review) as a constant shift of the auxiliary field [9,16]. In addition, it
was shown that partial supersymmetry breaking can occur even in the presence of hypermultiplets
in the adjoint representation. However, the addition of hypermultiplets in fundamental representation makes it difficult, as pointed out in [17,18]. In this paper we overcome this difficulty by
considering a model with hypermultiplets in bi-fundamental representation, whose matter content is described by a quiver diagram. This model is, therefore, a quiver gauge model. We will
show that, in addition to electric FI term, it is possible to introduce a magnetic FI term for any
N = 2 quiver gauge theory. This statement leads to the conclusion that in generic N = 2 quiver
gauge theory with these terms, N = 2 supersymmetry can be broken to N = 1 spontaneously.
As an illustration, we will describe this explicitly in a specific model, affine A1 quiver gauge
model, focusing on the Coulomb branch.
This model may seem reminiscent of the one discussed in [19]: a flow, by a mass deformation,
from the N = 2 affine A1 theory on the world volume of the D3-branes at C2 /Z2 orbifold
singularity to N = 1 quiver gauge theory on that at conifold singularity. Indeed, we can show
that in special points of the Coulomb branch the mass spectrum is the same as that of the theory
at the conifold singularity. The remarkable point of our model is that the masses are produced
dynamically, and thus we can dynamically approach the theory on conifold, namely conifold
geometry.
The organization of this paper is as follows. In Section 2, we construct N = 2 affine An1
quiver gauge model equipped with electric and magnetic FI terms. The necessary condition to
introduce the magnetic FI term without breaking N = 2 supersymmetry in the action is examined. We also mention the cases of different types of quiver gauge theories. As an illustration,
we mainly consider one of the simplest models, affine A1 quiver gauge model, in the subsequent
sections. In Section 3, we derive the scalar potential which is needed in the analysis of the vacua.
We show, in Section 4, that N = 2 supersymmetry is broken to N = 1 spontaneously on the
Coulomb branch by observing the mass spectrum and the appearance of the NambuGoldstone
fermion. As an application of our model, in Section 5, we consider the dynamical realization of
N = 1 quiver gauge theory on the world volume of the D3-branes at the conifold singularity
which has been considered in [19]. The notations on the harmonic superspace used in this paper
are collected in Appendix A.
218
2. N = 2 quiver gauge model

When we place N D3-branes at An1 orbifold C2 /Zn , the gauge theory realized on the world
volume is N = 2 affine An1 quiver gauge theory [20], namely U (N )n gauge theory composed
of n vector multiples VI++ in adjoint representation and n hypermultiplets qI+ in bi-fundamental
representation, where I = 1, . . . , n. In this section, we will construct N = 2 affine An1 quiver
gauge model, which has the same matter content as above, but with non-canonical kinetic terms
and electric and magnetic FI terms. We examine a necessary condition to introduce the magnetic
FI term without breaking N = 2 supersymmetry in the action. This magnetic FI term causes the
partial spontaneous breaking of N = 2 supersymmetry as will be seen in the next section.
2
In the case with additional D5-branes
wrapping on non-trivial S s, as in [20,21], the gauge
group of the world volume theory is i U (Ni ) and the rank of gauge group of each node is
in general different. Although we will not consider the gauge model with this matter content
explicitly, we will mention the condition necessary to introduce the magnetic FI term at the end
of this section.
The action for the vector multiplet of the U (N )n gauge symmetry is given by

n

i
4
4 F I (W I )
SV =
(2.1)
(D)4 FI (WI ) (D)
d x
4
I =1
1
(D + )2 (D )2 . See Appendix A for the concrete
where WI is the curvature of VI++ and (D)4 = 16
form. FI is the prepotential for the I th U (N ) which is denoted by U (NI ).1 VI++ is composed
of a complex scalar I , SU(2) doublet Weyl fermions A
I (A = 1, 2 labels SU(2) automorphism
of N = 2 supersymmetry) and a real auxiliary field DI(AB) , which transform as adjoint representation of U (NI ). The U (NI ) gauge group is generated by taI (a = 0, 1, . . . , NI2 1), and
t0I represents the overall U (1) part of U (NI ). As was done in [9], (2.1) reduces in components to

n

1
4
=
d
x
m Dm bI A
SV
Im FI |ab Dm Ia Dm Ib F I |ab aA
I
2
I =1
+
+
1
1
i
m
aAB b
bB c
b
DI AB + FI |abc aA
FI |ab aA
I Dm I A + Im FI |ab DI
I I DI AB
2
4
4
i
1
1
a
b mn
bB c
Re FI |ab mnpq vIa mn vIb pq
FI |abc aA
I I DI AB Im FI |ab vI mn vI
4
4
8

i
mn b A
I F I |abc aA
mn bI A vIc mn
FI |abc aA
I
I
4

i
i
FI |abcd aI A bI B cI A dI B + F I |abcd aI A bI B cI A dI B
12
12
aA b c d A
c d
1
b
Im FI |ab I fcd i 2I I + aA
I fcd i 2I I A ,
2
1
a c d b e f
(2.2)
Im FI |ab fcd
I I fef I I ,
2
where the symbol FI |ab... denotes the derivative of FI with respect to Ia , Ib , . . . .

1 From now on we use this notation. But keep in mind that we are considering U (N )n gauge model with the same
ranks.
219
. . . , N I +1 ) transformLet us introduce the N = 2 hypermultiplets qI+ i j (i = 1, . . . , NI , i = 1,
ing as (NI , NI +1 ) under U (NI ) U (NI +1 ):

q1+
q2+
..
.
qn+
U (N1 )
N1
1
U (N2 )
2
N
1
N
N2
U (N3 ) U (Nn )
1
1
N3
1
Nn .
The matter part of the action is

n

+ i ++ + i
Sq = du d (4)
qI i D qI i
(2.3)
I =1
d (4)
where

= d 4y d 4 +
+ i
D ++ qI
and
i
I +1 j + i
= D ++ qI+ i i + iVI++a taI j qI+ j i iVI++a

i qI j .
+1 ta
(2.4)
As explained in Appendix A, after eliminating infinitely many auxiliary fields, a hypermultiplet

qI+ contains SU(2) doublet complex scalars QI A , and a pair of Weyl fermions, I and I . In
components, the matter action Sq reduces to

n

i i m
i
i
4
m
i
A
Dm I i i I i i m Dm I i i
Sq = d x
Q
I i D Dm QI A i I i
2
2
I =1
ABj
i
i
+ i Q I A i j DI i QI B i i i Q I A j i DIAB
+1 j QI B i

k
k A i
Q I A i j I j i Q I A j i I +1 i j I i k QA
I i I +1 i QI k

k
k
Ai
Q I A i j I j i Q I A j i I +1 i j I i k QA
I i I +1 i QI k
i
j
i
i
i
i
j
Ai
Aj
j A i
+ i I i j A
I i QI A i i QI j I A i I i i I i I +1 j QI A i + i QI i I +1A j I i
i
j A i
i
i
i
j
Ai j i
Aj
+ iI i j A
I i QI A i + i QI j I A i I i iI i I +1 j QI A i i QI i I +1A j I i
1
1
1
+ I i j I j i I i i + I i j I j i I i i I j i I +1 i j I i i
2
2
2
I j i I +1 i j I i i .
(2.5)
2
By construction the action SV + Sq is invariant under the N = 2 supersymmetry transformation law:
I = i 2AB A B
(2.6)
I ,
A

B
A
B
vI m = iAB m I + I m ,
(2.7)
1
m n A
vI mn + 2 m A Dm I iA [I , I ] + DI A B B ,
A
(2.8)
I=
2

A m B
A B
A B
DIAB = 2iA m Dm B
I + 2iDm I + 2 2 I , I + 2 2 I , I ,
(2.9)
i
Ai
A
A i
QI i = I i + I i ,
(2.10)
220
I i i = 2i m A Dm QI A i i 2 2A I i j QI A j i + 2 2A I +1 j i QI A i j ,
I i i = 2i i m Dm QI A i i + 2 2 A I i j QI A j i 2 2 A I +1 j i QI A i j .
(2.11)
(2.12)
2.1. Electric and magnetic FI terms

We introduce the electric FI term

n
n

AB 0

Se = du d (4)
TrU (NI ) I++ VI++ + h.c. = d 4 x
I DI AB + h.c.
I =1
I++
+
IAB u+
A uB
(2.13)
I =1
=
and
is the electric FI parameter of U (NI ) gauge group. In the
where
three vector notation, the electric FI parameter can be written as IAB = iI ( )A
B where ( =
1, 2, 3) are the Pauli matrices. Se causes a constant shift of the auxiliary fields in the dual vector
multiplets and thus the magnetic FI term Sm is introduced to shift the auxiliary field in the original
vector multiplets by a constant [6,9,17,18]. We shall shift the auxiliary field as
IAB
D aAB
= DIaAB + 4iDI AB 0a ,
I
aAB
D
= DIaAB 4i DI AB 0a ,
I
so that the supersymmetry transformation law (2.8) changes to

a A B
a A B
Aa
Aa
B + ,
B + .
I = DI
I = DI
(2.14)
(2.15)
It is easy to see that the action with the shift (2.14)

SV |DD =
i
4

d 4x
n

4 F I (W I )
(D)4 FI (W I ) (D)
(2.16)
I =1
where W W |DD is invariant under the N = 2 supersymmetry transformations with (2.15).

In addition obviously Se |DD is N = 2 superinvariant. As we will see in the next section, the
magnetic FI term introduced above causes partial spontaneous supersymmetry breaking.
Next we examine Sq in (2.5). It is known [17,18] that there is a difficulty in introducing the
magnetic FI term in the presence of hypermultiplets in fundamental representation. However we
find that the magnetic FI term can be introduced without breaking N = 2 supersymmetry in
the case with hypermultiplets in bi-fundamental representation. Let us show this explicitly. We
examine the following terms contained in Sq ,
ABj
i
i
i Q I A i j DI i QI B i i i Q I A j i DIAB
+1 j QI B i
i
i
j
i
i
i
j
A i
A j
j A i
+ i I i j A
I i QI A i i I i I +1 j QI A i i QI j I A i I i + i QI i I +1A j I i
j A i
i
j
i
i
i
A i j i
A j
+ iI i j A
I i QI A i iI i I +1 j QI A i + i QI j I A i I i i QI i I +1A j I i .
(2.17)
Under the shift (2.14), (2.17) acquires additional terms

I A i i DI DI AB t0I i j QI B j + 2Q
I A i i I +1 I +1 AB t I +1 j QI B i .
2Q
i
j
i
D
D
0
(2.18)
Now for the N = 2 invariance of the action with the replacement (2.14), the following terms
have to vanish

I A i i DI DI AB t0I i j QI B j 2Q
I A i i DI DI AB t0I i j QI B j
2 Q
i
i
I +1
AB I +1 j
I
+1
i
i
t0
+ 2 Q I A i D D
i QI B j
AB I +1 j

i
t0
+ 2Q I A i i DI +1 DI +1
i QI B j
i

j

4 I i i DI A B B t0I j QI A j i + 4 I i i DI +1A B B t0I +1 i QI A i j
I
I i
I +1
I +1 j
i
i
B
B
i
A
t0 j I j i 4Q
t0
+ 4Q A
I i D AB
I i D AB
i I j
i

j

4I i i DI A B B t0I j QI A j i + 4I i i DI +1A B B t0I +1 i QI A i j

I i

I +1 j i
i I
i I +1
B
B
t0 j I j i + 4Q A
t0
4Q A
I i D AB
I i D AB
i I j .
221
(2.19)
We find that this is achieved if we choose the magnetic FI parameters such that
DI = DI +1
(2.20)
where (t0I )ij = ji / 2N . A bi-fundamental hypermultiplet interacts with two different gauge
sectors, and thus we can introduce the magnetic FI terms such that the effect from the shift of the
auxiliary field of one gauge sector and that of the other sector cancel out with each other. We can
also see that the matter part does not contribute to the magnetic FI term as the additional terms
(2.18) cancels out for (2.20).
Summarizing the action of the N = 2 quiver gauge model is given by
S = SV + Sq + Se + Sm = [SV + Sq + Se ]|DD .
(2.21)
Each part is given in (2.1), (2.3) and (2.13), and D is subject to (2.20). We have seen that this is
invariant under the N = 2 supersymmetry transformation with the replacement (2.14).
We comment on the case with hypermultiplets in fundamental representation. As pointed out
in [17,18], it is hard to introduce the magnetic FI term which causes the partial spontaneous supersymmetry breaking.2 This can be seen as follows. In this case, terms with DI +1 = DI +1 = 0 in
(2.19) have to be deleted for the N = 2 superinvariance. This forces us to set the magnetic FI parameter DI AB to be imaginary. However, the real part of the magnetic FI parameter causes partial
spontaneous supersymmetry breaking [6], and thus for imaginary magnetic FI parameter N = 2
supersymmetry remains unbroken in the vacua. In other words, when we introduce the real part
of the magnetic FI parameter in the presence of hypermultiplets in fundamental representation,
the action is no longer invariant under the N = 2 supersymmetry transformation.
In this paper we mainly consider affine An1 quiver gauge models. However, it is now obvious that we can introduce the magnetic FI term in any N = 2 quiver gauge model with any
number of nodes (gauge sectors) and any number of arrows (bi-fundamental hypermultiplets).
Let us comment on two generalizations of our model among them. The first one is the case when
the rank of each gauge group is different. Such an N = 2 quiver gauge model is obtained by considering the additional D5-branes wrapping on non-trivial S 2 s (though we have non-canonical
kinetic terms) and is also interesting because, in contrast to the superconformal case above, we
have running gauge couplings. Even in this case, the argument on the magnetic FI term is similar
2 In [9], the magnetic FI term is introduced even in the presence of hypermultiplets in fundamental representation.
However, as explained below, the magnetic FI parameter is imaginary, and thus N = 2 supersymmetry remains unbroken
in the vacua.
222
to that given above. The only difference is that the condition of the magnetic FI parameter (2.20)
is changed as
I
I +1
.
D = D
NI +1
NI
(2.22)
This change comes from the normalization of the generator: (t0I )i j = i j / 2NI . So we have
no difficulty in adding the magnetic FI term. The second one is the case when the gauge group
is different from An1 . In fact, we can also construct N = 2, Dn , E6 , E7 and E8 quiver gauge
models. Even in these cases, all we have to do is to relate the FI parameters in accordance with
(2.20) or (2.22).
3. The minimal model
We will show that in our model N = 2 supersymmetry is partially broken to N = 1 spontaneously. As an illustration we will focus on the affine A1 quiver gauge model to which we refer
as the minimal model in the following sections.
The minimal model is composed of a pair of hypermultiplets qI+ and a pair of vector multiplets
2 ) and (N
1 , N2 ),
VI++ (I = 1, 2). q1+ and q2+ , respectively, transform as bi-fundamental, (N1 , N
++
and VI transform as adjoint under U (NI ). The action of this model is given by (2.21), with
summing only over I = 1, 2.
Let us write down the scalar potential in component. The scalar potential is
(1)
(2)
VI + VI ,
V=
(3.1)
I =1,2

1
1
VI(1) = gI ab PIa PIb + gI ab DIaAB DIb AB 2iDI AB DI AB FI |00 | + 2i DI AB DI AB F I |00 |
2
4
AB I
(3.2)
4iI D AB + 4i IAB DI AB ,
i A k

(2)
i
j
j
i
k
A
i
I A i I +1 I k QI I +1 QI
VI = Q I A j I i Q
j
i
i
k

i A k

i
j
j
i
k
i
I A i I +1 I k QI I +1 QA
+ Q I A j I i Q
(3.3)
j
i I k
i
where PIa = ifIabc Ib Ic , FI |ab... | represents FI |ab... evaluated at = = 0, and gI |ab =
Im FI |ab | is the Khler metric. DIaAB is obtained by solving the equation of motion as

AB
AB
AB
DIaAB = 2gIab (I + I )AB b0 + IAB
(3.4)
D FI |0b | + I D FI |0b | + Q1I |b + Q2I |b
where Q1I and Q2I are the contributions from the hypermultiplets q1+ and q2+ , respectively:

i A i 1 i B j
Q1 i tb j Q1 i + (A B) ,
2

i A j 1 i B i
AB
Q21|b = Q
2 i tb j Q2 i + (A B) ,
2

i A j 2 i B i
AB
Q12|b
= Q
i tb j Q1 i + (A B) ,
2 1

i i 2 i B j
AB
= Q A
Q22|b
2 i tb j Q2 i + (A B) .
2
AB
Q11|b
=
(3.5)

(1)
We can rewrite VI
(1)
VI
223
in (3.1) as
b

1
1
D
I AB 2i IAB IAB DI AB + DI AB
= gI ab PIa PIb + gI ab D aAB
I
2
4
(3.6)
where

= D aAB + 4i I AB a
D aAB
D
I
I
0

AB

AB
AB
F I |0b | + Q1I
= 2gIab (I + I )AB b0 + DI + DI
|b + Q2I |b .
(3.7)
4. Vacua of the minimal model

In this section, we will find the N = 1 supersymmetric vacua in the Coulomb branch QI = 0
by analyzing the condition stabilizing the scalar potential derived in the previous section.
The constraint, gI ab PIa PIb = 0, can be satisfied by vanishing non-diagonal components of
the vacuum expectation value of I , that is, Ir = 0 where trI represent non-Cartan generators of
the gauge group U (NI ). Then, we consider the condition to stabilize the scalar potential. While
the derivative of the scalar potential V with respect to the hypermultiplet scalar Q is trivially zero
in the Coulomb branch QI = 0, the non-trivial vacuum condition is derived from the derivative
with respect to I in the vector multiplet

i
c
0=
(4.1)
FI |abc |D b
V
=
I DI ,
a
I
4
a
A
where D aA
I B = iD I ( ) B . Note that the index I is not summed over here.
Let us examine the case with the single trace prepotential of degree nI + 2,
FI =
n
I +1
k=1
gk
Tr W k+1
(k + 1)!
(4.2)
I , i = 1, . . . , N , be the fundamental matrix of gauge group U (N ) which

for concreteness. Let Eij
I
I
has 1 at the (i, j ) component and 0 otherwise. Cartan generators can be written as tiI = EiiI . We
have V /Ir = 0 because FI |rii = D rI = 0. Noting that the points specified by FI |iii = 0
correspond to the unstable vacua, we derive the vacuum conditions as follows,
i i
D I D I = 0, for all i and I.
(4.3)
As in [9], we can choose the FI parameters by using SU(2) rotation as

I
(I + I ) = (0, eI , I ),
D + DI = (0, mI , 0).
(4.4)
Furthermore, we can set mI /I < 0, without loss of generality. In these choices of the FI parameters, we obtain the following vacuum condition,

eI
I
FI |ii = 2
(4.5)
+i
.
mI
mI
Note that the minus sign in front of iI /mI has been excluded by the positivity criterion of the
Khler metric: gI |ii = ImFI |ii > 0. In the original bases, this means

eI
I
+i
.
FI |00 =
(4.6)
mI
mI
224
The vacuum expectation values of the diagonal components of I are determined

from the above
nI
.
Thus,
the
gauge
symmetry
U
(N
)
is
broken
to
U
(N
equations
of
degree
n
I
I
I |i ) with NI =
i=1
nI
N
.
i=1 I |i
We can easily evaluate the vacuum energy

(I I ) DI + DI
4mI I 4i
V =
(4.7)
I =1,2
which comes from the last two terms in (3.6). As pointed out in [6], using the freedom to choose
the imaginary part of I , we can obtain the vanishing vacuum energy; if we set (I I )2 =
iI , then the vacuum energy is zero. The vanishing vacuum energy may indicate that N = 1
supersymmetry remains in the vacuum.
In the subsequent subsections, we will show that the mass spectrum on the vacuum can be
written in terms of N = 1 multiplets, and that a linear combination of fermions becomes the
NambuGoldstone fermion associated with the partial supersymmetry breaking.
4.1. Mass spectrum
The masses of the component fields contained in the vector multiplets are similar to those
in the pure U (N ) YangMills case [7,8]. The
I N = 2, U (NI ) vector multiplet decomposes into
U (NI |i ) massless vector multiplet, N = 1 masthree types of N = 1 multiplets: N = 1, ni=1
sive chiral multiplet in adjoint representation with mass MI = mI gI FI |0 where tI represent
unbroken generators, and N = 1 massive vector multiplets which correspond to the broken generators when the gauge symmetry is broken.
Let us turn to the matter part. The mass of the scalar components QI A i i is easily obtained by
(1)
evaluating the second derivative of V . First we examine VI in (3.6). We observe its vacuum
expectation value vanishes as follows. Since the first and the last terms in (3.6) do not contain
the scalar QI , only the second term can contribute to the mass. However, it vanishes as

aCD b

gI |ab D I
Q A i i QBj
D I CD
J
= 4
K j

CD

CD
Re D aICD Q A i i QBj Q1I
|a + Q2I |a
K j
1 i j

j
1
= 4iJ1 K
ta j i Re D a1AB ta2 i i j Re D a2AB
1 i j

j
2
ta j i Re D a1AB + ta2 i i j Re D a2AB
4iJ2 K
= 0.
(4.8)
In the last equality, we have used Re D aIAB = 2i(2 )AB mI 0a and the relation (2.20). Next,
we examine VI(2) in (3.3). It is easy to see that the scalar components Qi1Aj and Qi2Aj have the
same mass m2i j = 2|a1i a2j |2 where
I = diag(aI 1 , . . . , aI NI ).
(4.9)
The masses of the fermions I and I can be seen from the following terms of Sq (2.5)

1
1
j
j
1 i i 1 i j j i 2 i ji 1 j j + 2 i i 2 i j j i 1 i i j 2 j j + h.c.
2
2
(4.10)
225
Let us examine the first term from which masses of 1 and 1 are determined. As I is diagonal
(4.9), we can rewrite it as

i

1
0
a1i a2j
1 i
1 i i 1 i i
a1i a2j
0
1 i i
2 2

a1i a
1
0
+1 i i
2j
= +1 i i 1 i i
(4.11)
0
a1i a2j
1 i i
2 2
where we have defined ( )/ 2. By taking the normalization ofthe kinetic terms into
account, one sees that the masses of +1 i j and 1 i j can be evaluated as 2|a1i a2j |2 . In the
same way, by examining the second term in (4.10), the mass of +2 i j and 2 i j is found to be
the same as that of 1 .
2 ) representation is decomposed
Thus in the vacuum, N = 2 hypermultiplet q1+ in (N1 , N
into various
massive
multiplets
according
to
the
branching
rule
and the massive multiplets with

2
mass 2|a1i a2j | transform as (N1|i , N2|j ). The same is true for another hypermultiplet q2+
1 , N2 ).
in (N
4.2. NambuGoldstone fermion
In the case of the N = 2, U (N ) gauge model with/without hypermultiplets in adjoint representation [79], a linear combination of the overall U (1) fermions in the N = 2 vector multiplet
becomes the NambuGoldstone fermion. One might think that as there are two gauge sectors,
U (N )2 , two NambuGoldstone fermions would emerge. However, this is not correct. We show
that only one combination of these fermions becomes the NambuGoldstone fermion.
The vacuum expectation values of the supersymmetry transformations of component fields
vanish except for A
I . In the choice of the FI parameters (4.4), we obtain

aA
a 1 1
D I B = imI 0
(4.12)
1 1
with m1 = m2 which follows from the condition (2.20). Thus, letting a
I

a
a+
I = 0.
I = i 2mI 0a 1 2 ,
0+
Furthermore, combining 0+
1 and 2 , we find
0+

1 0+
= i 2(m1 m2 ) 1 2 = 0,
2

0+
= i 2(m1 + m2 ) 1 2 = 0
1 + 0+
2
1 (a1
2 I
a2
I ),
(4.13)
(4.14)
0+
where in the last equality we have used m1 = m2 . So, the fermion 0+
1 + 2 can be the Nambu
Goldstone fermion. In order to conclude that this is the NambuGoldstone fermion, we have to
show this fermion is exactly massless.
The mass term of 0 can be read off from (2.2) as
0 0
i
(4.15)
I A I B ,
FI 000 D 0AB
I
4
0
with the replacement (2.14). It is easy to see that this is proportional to imI 0
I I , and thus we
0+
0+
0+
0+
conclude that 0+
1 + 2 is massless as I are massless. As a result, we can identify 1 + 2
226
as the NambuGoldstone fermion associated with the partial spontaneous supersymmetry breaking.
Let us comment on the affine An1 quiver gauge model with n > 2. In this case 0+
I are
independent of I because of the condition (2.20). Therefore, only
the
vacuum
expectation
value

0+
of the supersymmetry transformation of the combination, = I 0+
,
is
not
zero.
As
I
I are
massless and so are, we may identify with the NambuGoldstone fermion associated with
partial supersymmetry breaking.
5. A dynamical realization of KlebanovWitten model
In [19], Klebanov and Witten considered the N = 2 affine A1 quiver gauge theory realized on
the world volume on D3-branes at orbifold singularity (dual to type IIB superstring in AdS5
S 5 /Z2 [22]), and discussed a flow to N = 1 fixed point by adding the mass operator [23] of
the chiral multiplet I in adjoint representation, which breaks N = 2 supersymmetry to N = 1.
It was shown that the superpotential of the effective theory at the fixed point can be regarded
as that of the world volume theory at the conifold singularity (dual to type IIB superstring in
AdS5 T 1,1 ).
As our minimal model is N = 2 affine A1 quiver gauge theory with non-canonical kinetic
terms and electric and magnetic FI terms, it is expected that our model might describe the N = 1
quiver gauge theory dual to type IIB superstring in AdS5 T 1,1 in some points of vacua. If so, the
minimal model may provide a dynamical realization of the statement of [19] because the N = 1
chiral multiplet I in adjoint representation becomes massive dynamically. We show that this is
the case.
First, we examine the matter content realized on the vacuum. Let us consider the following
point of vacua
1 = 2 = diag(a, . . . , a).
(5.1)
The
gauge symmetry is unbroken in this vacuum, while N = 1 chiral multiplet in the
adjoint representation becomes massive with mass MI = mI gIaa FI |0aa [8]. As was seen in Section 4.1, the N = 1 chiral multiplets (AI , B I ) in bi-fundamental representation acquire masses
m2i j = 2|a1i a2j |2 , but in the vacuum (5.1) these become massless. Thus, the massless mulU (N )2
tiplets in the vacuum are N = 1, U (N )2 vector multiplet, N = 1 chiral multiplets AI and BI

and (N,
N) representations, respectively. This is exactly the matter content
(I = 1, 2) in (N, N)
of the N = 1 quiver gauge theory dual to IIB superstring in AdS5 T 1,1 .
Next, we examine the superpotential realized in the vacuum. Expanding each field around its
vacuum expectation value, we obtain the fluctuation action in the N = 1 vacuum. The matter
part of the action can be written in terms of N = 1 multiplets. The superpotential term which
contains the chiral multiplets AI and BI in bi-fundamental representation is

Wmatter = Tr (A1 B1 A2 B2 )1 (B1 A1 B2 A2 )2
(5.2)
up to an overall numerical coefficient. Here we have not written the gauge index explicitly. Of
course, this is the ordinary superpotential of N = 2 supersymmetric gauge theory with hypermultiplets in N = 1 superspace formalism [19,23], because we are considering the Coulomb
branch. On the other hand, the mass term of the chiral multiplets I in the superpotential is
found to be
W = M1 Tr 12 + M2 Tr 22 .
(5.3)
227
The effective theory is described by the massless fields, and the massive chiral multiplets I
in adjoint representation should be integrated out. To achieve this, we note that the mass matrix MI is determined by the vacuum expectation value and the prepotential FI . Though
we are working in special points of vacua, MI still depends on the choice of the prepotential
function. Suppose that the prepotentials are set to satisfy M1 = M2 . By integrating out I using the equation of motion, we obtain the following superpotential up to an overall numerical
coefficient
Tr(A1 B1 A2 B2 B1 A1 B2 A2 )
(5.4)
which is the well-known superpotential of the world volume theory on the D3-branes at the
conifold singularity.
It is interesting to take the higher order terms in I into account, and examine the deformations of the superpotential (5.4).
Acknowledgements
We thank Yosuke Imamura, Kazutoshi Ohta and Yukinori Yasui for useful discussion. This
work is supported in part by the Grant-in-Aid for Scientific Research (18540285, 19540324,
19540304, 19540098) from the Ministry of Education, Science and Culture, Japan. Support from
the 21st century COE program Constitution of wide-angle mathematical basis focused on knots
is gratefully appreciated. We thank the Yukawa Institute for Theoretical Physics at Kyoto University, where the preliminary version of this work was presented, during the YITP-W-07-05 on
String Theory and Quantum Field Theory (6th10th August 2007). We wish to acknowledge
the participants for stimulating discussions.
Appendix A. Superfields in harmonic superspace
Harmonic superspace [14,15] is an extension of the N = 2 superspace [24] by including
harmonic variables u
A (A = 1, 2),

u+
A , uA SU(2),
u+A u
A = 1,
u+A = u
A
(A.1)
where the bar means the complex conjugation

QA = Q A ,
A,
B = AB BC Q C = Q
QA = AB QB = AB Q
AB = AB ,
AB = AC BD CD = AC BD CD = AC BD CE DF EF = AB .
(A.2)
We introduce an N = 2 vector multiplet V ++ = V ++a ta transforming as adjoint represena t . V ++ is the

tation under the gauge group: hermitian matrices (ta )i j generate [ta , tb ] = ifbc
a
+
++
+
++
analytic superfield satisfying D V

=D V
= 0. In the analytic basis

m m
y , , , u = x 2i A m B u+ u , A u , A u , u ,
(A.3)
A
D and D are given as
(A B)
228
D +
= ,
D+ =
+ 2i m m ,
+
y

m
D
,
= + 2i
y m
D =
(A.4)
and thus V ++ is a superfield in analytic subspace (y, + , + , u

i ). In the WessZumino gauge
V ++ is given as
2
2
2
V ++ = 2i + m + vm (y) i 2 + (y)
+ i 2 + (y) + 4 + + A (y)u
A
2
+ 2 + 2 AB

4 + + A (y)u
+
3
D
(y)u
u
,
(A.5)
A
A B
where vm , , A and D AB are vector, complex scalar, SU(2) doublet Weyl spinors and auxiliary
field, respectively. The curvature W of V ++ defined by

1 + 2
V ++ (v1 ) V ++ (vn )
dv1 dvn (i)n+1 + + + +

W = D
(A.6)
4
(u v1 )(v1 v2 ) (vn+ u+ )
n=1
is evaluated to give [9]

4
W = A mn A vmn i 2 + i 2( )4 mn Dm Dn + i A B Dm A m B 2 A A
3

2 A B
+ A B DAB
2 [, A B ] + i( )4 AB A , B + i 2( )4 , [, ]
3
+ ,
2i + [, ]
(A.7)
where vmn = m vn n vm + i[vm , vn ] and Dm = m + i[vm , ]. The ellipsis represents terms
which do not contribute to the action (2.1).
Next we introduce an N = 2 hypermultiplet q +i j transforming as bi-fundamental representation under the gauge group. The q + hypermultiplet is an analytic superfield satisfying
D + q + = D + q + = 0, and can be expanded as
2
2
q + = F + (y, u) + + (y, u) + + (y,
u) + + M (y, u) + + N (y, u)
+ 2 + (2)
2
+ i + m + A
(y, u) + + + (2) (y, u)

m (y, u) +
2 2
+ + + P (3) (y, u)
(A.8)
in the analytic basis. q + contains infinitely many auxiliary fields which are eliminated by solving
the equations of motion derived from (2.3): D ++ q + = 0 where D ++ is given in (2.4) and
D ++ = ++ 2i + m +
+ + + + ,
y m
++ = u+A
.
uA
(A.9)
We find that the auxiliary fields are eliminated by

i +
FI+ i i = QA
I (y) i uA ,
(A.10)
i = (y)
i = (y) i ,
i ,
i Aj

i
i
MI i = 2 I j QI i I +1 j i QA
I j uA ,

j
j
Ai
NI i i = 2 I i j QA
I i I +1 i QI j uA ,
i
Ai
i
A
I m i = 2Dm QI i uA ,
(A.11)
(A.12)
(A.13)
(A.14)

i Bj
A j B i
= 2i A
I j QI i I +1 i QI j uA uB ,
i Bj

(2)
A j
Bi
I i i = 2i A
I j QI i I +1 i QI j uA uB ,

(3)
j
AB j
Ci
PI i i = i DIAB i j QC
I i DI +1 i QI j uA uB uC .
(2) i
229
(A.15)
(A.16)
(A.17)
The physical fields are SU(2) doublet complex scalars QA and a pair of SU(2) isosinglet spinors,
and .
References
[1] N. Seiberg, Phys. Lett. B 206 (1988) 75.
[2] N. Seiberg, E. Witten, Nucl. Phys. B 426 (1994) 19, hep-th/9407087;
N. Seiberg, E. Witten, Nucl. Phys. B 430 (1994) 485, Erratum;
N. Seiberg, E. Witten, Nucl. Phys. B 431 (1994) 484, hep-th/9408099.
[3] J. de Boer, Y. Oz, Nucl. Phys. B 511 (1998) 155, hep-th/9708044;
F. Cachazo, K.A. Intriligator, C. Vafa, Nucl. Phys. B 603 (2001) 3, hep-th/0103067;
F. Cachazo, C. Vafa, hep-th/0206017.
[4] R. Dijkgraaf, C. Vafa, Nucl. Phys. B 644 (2002) 3, hep-th/0206255;
R. Dijkgraaf, C. Vafa, Nucl. Phys. B 644 (2002) 21, hep-th/0207106;
R. Dijkgraaf, C. Vafa, hep-th/0208048;
F. Cachazo, M.R. Douglas, N. Seiberg, E. Witten, JHEP 0212 (2002) 071, hep-th/0211170;
H. Itoyama, A. Morozov, Nucl. Phys. B 657 (2003) 53, hep-th/0211245;
F. Ferrari, JHEP 0606 (2006) 039, hep-th/0602249.
[5] J. Hughes, J. Polchinski, Nucl. Phys. B 278 (1986) 147;
J. Hughes, J. Liu, J. Polchinski, Phys. Lett. B 180 (1986) 370.
[6] I. Antoniadis, H. Partouche, T.R. Taylor, Phys. Lett. B 372 (1996) 83, hep-th/9512006;
I. Antoniadis, T.R. Taylor, Fortschr. Phys. 44 (1996) 487, hep-th/9604062.
[7] K. Fujiwara, H. Itoyama, M. Sakaguchi, Prog. Theor. Phys. 113 (2005) 429, hep-th/0409060;
K. Fujiwara, H. Itoyama, M. Sakaguchi, hep-th/0410132.
[8] K. Fujiwara, H. Itoyama, M. Sakaguchi, Nucl. Phys. B 723 (2005) 33, hep-th/0503113.
[9] K. Fujiwara, H. Itoyama, M. Sakaguchi, Nucl. Phys. B 740 (2006) 58, hep-th/0510255;
K. Fujiwara, H. Itoyama, M. Sakaguchi, Prog. Theor. Phys. Suppl. 164 (2007) 125, hep-th/0602267.
[10] P. Kaste, H. Partouche, JHEP 0411 (2004) 033, hep-th/0409303;
P. Merlatti, Nucl. Phys. B 744 (2006) 207, hep-th/0511280;
L. Girardello, A. Mariotti, G. Tartaglino-Mazzucchelli, JHEP 0603 (2006) 104, hep-th/0601078.
[11] K. Fujiwara, Nucl. Phys. B 770 (2007) 145, hep-th/0609039.
[12] H. Itoyama, K. Maruyoshi, Phys. Lett. B 650 (2007) 298, arXiv: 0704.1060 [hep-th].
[13] S. Ferrara, L. Girardello, M. Porrati, Phys. Lett. B 366 (1996) 155, hep-th/9510074;
P. Fre, L. Girardello, I. Pesando, M. Trigiante, Nucl. Phys. B 493 (1997) 231, hep-th/9607032;
M. Porrati, Nucl. Phys. B (Proc. Suppl.) 55 (1997) 240, hep-th/9609073;
J. Louis, hep-th/0203138;
H. Itoyama, K. Maruyoshi, Int. J. Mod. Phys. A 21 (2006) 6191, hep-th/0603180;
K. Maruyoshi, hep-th/0607047.
[14] A. Galperin, E. Ivanov, S. Kalitsyn, V. Ogievetsky, E. Sokatchev, Class. Quantum Grav. 1 (1984) 469;
A. Galperin, E.A. Ivanov, V. Ogievetsky, E. Sokatchev, Class. Quantum Grav. 2 (1985) 601;
A. Galperin, E. Ivanov, V. Ogievetsky, E. Sokatchev, Class. Quantum Grav. 2 (1985) 617.
[15] A.S. Galperin, E.A. Ivanov, V.I. Ogievetsky, E.S. Sokatchev, Harmonic Superspace, Cambridge Univ. Press, 2001.
[16] K. Fujiwara, H. Itoyama, M. Sakaguchi, hep-th/0611284.
[17] H. Partouche, B. Pioline, Nucl. Phys. B (Proc. Suppl.) 56 (1997) 322, hep-th/9702115.
[18] E.A. Ivanov, B.M. Zupnik, Phys. At. Nucl. 62 (1999) 1043, Yad. Fiz. 62 (1999) 1110, hep-th/9710236.
[19] I.R. Klebanov, E. Witten, Nucl. Phys. B 536 (1998) 199, hep-th/9807080.
[20] M.R. Douglas, G.W. Moore, hep-th/9603167;
M.R. Douglas, B.R. Greene, D.R. Morrison, Nucl. Phys. B 506 (1997) 84, hep-th/9704151.
[21] K. Ohta, T. Yokono, JHEP 0002 (2000) 023, hep-th/9912266;
230
I.R. Klebanov, M.J. Strassler, JHEP 0008 (2000) 052, hep-th/0007191;

F. Cachazo, S. Katz, C. Vafa, hep-th/0108120;
F. Cachazo, B. Fiol, K.A. Intriligator, S. Katz, C. Vafa, Nucl. Phys. B 628 (2002) 3, hep-th/0110028.
[22] S. Kachru, E. Silverstein, Phys. Rev. Lett. 80 (1998) 4855, hep-th/9802183.
[23] A. Hanany, A.M. Uranga, JHEP 9805 (1998) 013, hep-th/9805139;
S. Gukov, Phys. Lett. B 439 (1998) 23, hep-th/9806180.
[24] R. Grimm, M. Sohnius, J. Wess, Nucl. Phys. B 133 (1978) 275.

MHV amplitudes in N = 4 super-YangMills

and Wilson Loops
Andreas Brandhuber, Paul Heslop , Gabriele Travaglini
Centre for Research in String Theory, Department of Physics, Queen Mary, University of London, Mile End Road,
London E1 4NS, United Kingdom
Received 2 October 2007; accepted 5 November 2007
Abstract
It is a remarkable fact that MHV amplitudes in maximally supersymmetric YangMills theory at arbitrary
loop order can be written as the product of the tree amplitude with the same helicity configuration and a
universal, helicity-blind function of the kinematic invariants. In this note we show how for one-loop MHV
amplitudes with an arbitrary number of external legs this universal function can be derived using Wilson
loops. Our result is in precise agreement with the known expression for the infinite sequence of MHV
amplitudes in N = 4 super-YangMills. In the four-point case, we are able to reproduce the expression
of the amplitude to all orders in the dimensional regularisation parameter . This prescription disentangles
cleanly infrared divergences and finite terms, and leads to an intriguing one-to-one mapping between certain
Wilson loop diagrams and (finite) two-mass easy box functions.
1. Introduction
Even after many years, N = 4 supersymmetric YangMills (SYM) remains a fascinating
theory that constantly reveals new hidden structures and symmetries. In the last few years, substantial progress has been made in two seemingly unrelated areas of research of the theory.
On the one hand there is the conjecture of Bern, Dixon and Smirnov (BDS) of an exponential
formula for planar n-point MHV amplitudes in N = 4 super-YangMills (SYM) at large N [1].
According to this conjecture, higher-loop amplitudes are determined in terms of the one-loop am* Corresponding author.
E-mail addresses: a.brandhuber@qmul.ac.uk (A. Brandhuber), p.j.heslop@qmul.ac.uk (P. Heslop),

g.travaglini@qmul.ac.uk (G. Travaglini).
doi:10.1016/j.nuclphysb.2007.11.002
232
A. Brandhuber et al. / Nuclear Physics B 794 (2008) 231243
plitude together with four constants, which depend on the t Hooft coupling only. The finite parts
of these amplitudes obey a similar exponential formula: the all-loop finite parts are determined
purely by the one-loop finite part and two coupling-dependent constants. Furthermore, one of
these constants is a well-known physical quantity, known as the cusp anomalous dimension.
In parallel developments, integrability has been used, together with a number of further assumptions, to study the spectrum of gauge invariant operators in N = 4 SYM. Among the
impressive outcomes of this is the conjectured all-orders formula of [2] for the very same cusp
anomalous dimension appearing in the exponentiation formula, thus giving a tantalising potential
link between integrability and amplitudes. Calculations of four-point amplitudes using unitarity
methods at up to a highly impressive four-loop order have been performed [1,3], thus determining
the cusp anomalous dimension to this order (at least numerically) and leading to an expression for
the amplitude in terms of integral functions. These integral functions were calculated to higher
accuracy in [4], thus allowing for a more precise calculation of the cusp anomalous dimension
at four loops. Using the assumption of pseudo-conformality, originally observed at three [5] and
four loops [3], the expressions of the amplitude in terms of the integrals occurring at five loops
has been determined [6].
Very recently, Alday and Maldacena have been able to apply for the first time the AdS/CFT
correspondence to the calculation of amplitudes, and verified the form of the exponentiation of
the four-point amplitude at strong coupling [7]. Remarkably, it turns out that the computation
of amplitudes at strong coupling is dual to the computation of the area of a string ending on a
lightlike polygonal loop embedded in the boundary of AdS space. This, in turn, is equivalent to
the method for computing a lightlike polygonal Wilson loop at strong coupling using AdS/CFT.
The edges of the polygon are determined by the external momenta of the amplitude.
In [8] the same Wilson loop (with four lightlike segments) was considered in weakly-coupled
gauge theory and it was shown at one loop that it reproduces the known one-loop four-point
amplitude.1 The infrared divergent pieces come from propagators stretching between adjacent
edges, and the finite part of the amplitude comes from propagators stretching between opposite
edges.
In this paper we consider Wilson loops around arbitrary lightlike polygons with n sides at
one loop, and find precise agreement with one-loop n-point MHV amplitudes in N = 4 SYM
(divided by the tree amplitude). As for the four-point case, the infrared part is correctly reproduced by considering propagators between adjacent segments and the finite piece is obtained
from propagators between non-adjacent segments. This finite part is given by the sum of the
(finite) two mass easy box functions in four dimensions with coefficients unity. This ability to
separate the divergent and finite parts is a particularly nice feature, and enables us to calculate
finite parts purely in four dimensions without the need of a regulator. On the other hand, if we use
dimensional regularisation on the finite part as well, we reproduce the all orders in expression
for the two-mass easy box function, and hence the complete all-orders in expression for the
four-point MHV amplitude.
The paper is organised as follows. In Section 2 we review the form of n-point MHV amplitudes in terms of two-mass easy box functions, and discuss the BDS conjecture and the
strong-coupling calculation of amplitudes as lightlike Wilson loops. In Section 3 we present
the one-loop calculation of the infinite sequence of MHV amplitudes in N = 4 SYM using light1 More precisely, the calculations of [7] and [8], as well as ours, are insensitive to the polarisations of the particles
participating in the scattering. In particular, the tree-level ParkeTaylor amplitude, which appears as a common prefactor
in the N = 4 MHV amplitudes at any loop order, is not generated by the calculation.
233
Fig. 1. A two-mass easy box function. The momenta p and q are null, whereas, in general, P 2 = 0 and Q2 = 0. The
cases when either P 2 or Q2 , or both, are also null, correspond to the one-mass and zero-mass boxes, obtained as smooth
limits from the expression (2.3) of the two-mass box function.
like Wilson loops. We present our conclusions and comment on the result of our calculations in
Section 4.
2. One-loop MHV amplitudes in N = 4 SYM
In this section we briefly review the expression of the infinite sequence of MHV amplitudes
in N = 4 SYM amplitudes at one loop. These amplitudes were determined for the first time in
[9] using unitarity and collinear limits. Recently, their expression was re-derived in [10] using
one-loop MHV diagrams. In the following discussion we suppress constant factors connected
with dimensional regularisation when they are not essential.
The form of the n-point MHV amplitudes at L loops in N = 4 SYM is remarkably simple. It turns out that the amplitude is given by the tree-level amplitude, times a scalar function,
tree
(L)
A(L)
n = A n Mn .
(2.1)
(1)
At one-loop order, the function Mn is expressed as a sum of so-called two-mass easy box
functions F 2me [11], all with coefficient equal to one2 :

M(1)
(2.2)
F 2me (p, q, P , Q).
n =
p,q
The main characteristic of this function, depicted in Fig. 1, is that two opposite legs, p and q, are
massless, whereas the two remaining legs P and Q are massive. The summation in (2.2) is such
that each different two-mass easy function appears exactly once.3
A compact form of the two-mass easy box function containing only four dilogarithms was first
derived in [12]. This form was found independently in [10] in the context of MHV diagrams,
where an analytic proof of its equivalence with the conventional expression of, e.g., [11] was
given. Expressing the two-mass easy box as a function of the kinematic invariants s := (P + p)2 ,
2 In (2.2) we are suppressing a factor of c := (1 + ) 2 (1 )/[(4 )2 (1 2)].

n
2me
3 A more explicit way to write M(1) is M(1) = n [ 2 ]1 (1 (1/2)
n
n
n/21,r )Fn:r;i , where the relation to the
i=1 r=1
functions introduced in (2.2) and depicted in Fig. 1 is obtained by setting p = pi1 , q = pi+r , and P = pi + +
pi+r1 .
234
t := (P + q)2 and P 2 , Q2 , with p + q + P + Q = 0, it reads

F
2me
s, t, P , Q

t
P 2
Q2
+
2
2
2

+ Li2 1 aP 2 + Li2 1 aQ2
1
= 2

s
2

Li2 (1 as) Li2 (1 at),
(2.3)
where
a=
u
P 2 + Q2 s t
= 2 2
.
P 2 Q2 st
P Q st
(2.4)
For later use, we also quote the all-orders in expression of this function [13],
2me
s, t, P , Q

t
P 2
Q2
+
2
2
2

a2
1
F
,
,
1
+
,
1
2
1 aP 2
1 aP 2

a2
1
F
+
,
,
1
+
,
2 1
1 aQ2
1 aQ2

a2
1
2 F1 , , 1 + ,
1 as
1 as

a2
1
.
2 F1 , , 1 + ,
1 at
1 at
1
= 2

+

s
2

(2.5)
One important property of the MHV amplitudes is that they do not have multiparticle singularities. In particular, we note that, although each box function (2.3) contains poles in associated to
multiparticle invariants, after performing the sum (2.2) the infrared divergent terms only involve
two-particle invariants,
M(1)
n IR

n
1 sii+1
= 2
,

2
(2.6)
i=1
with sij := (pi + pi+1 )2 .

Recently, Bern, Dixon and Smirnov (BDS) proposed a remarkably simple conjecture for the
resummation at all loops of the planar MHV amplitudes in N = 4 SYM calculated at weak
coupling [1]. This conjecture, based on explicit calculations at two loops [14], was verified in
[1] up to three loops in the four-point case, and in [15] up to two loops at five points. Explicit
expressions of the four-point amplitudes at four and five loops were recently presented in [3] and
[6], respectively, and will allow for precise tests of the conjecture at four and five loops once the
relevant integral functions have been evaluated to the necessary degree of accuracy in .
235
The form of the BDS conjecture is inspired by the soft and collinear behaviour of amplitudes
in gauge theory [1623], and is expressed by
Mn := 1 +
a L M(L)
n ()
L=1

= exp
(L)
(L)
()M(1)
n (L) + C
+ En(L) ()

(2.7)
L=1
where a = [g 2 N/(8 2 )](4e ) . Here f (L) () = f0 + f1 + f2 2 is a set of functions,

one at each loop order, which make their appearance in the exponentiated all-loop expression for
the infrared divergences in generic amplitudes in dimensional regularisation [21]. In particular,
(L)
(L)
f0 = K /4, where K is the cusp anomalous dimension (related to the anomalous dimension
of twist-two operators of large spin). An important point of the conjecture is that the constants
C (L) do not depend on kinematics or on the number of particles n. The non-iterating contributions
(L)
En vanish as 0 and depend explicitly on n.
BDS also propose a very interesting form for the finite remainders of the MHV amplitude,
given by

1
Fn = exp K Fn(1) (0) + C ,
(2.8)
4
(L)
(1)
(L)
(L)
(1)
where Fn (0) is the finite remainder of Mn as defined in [1].

Motivated by the BDS conjecture, Alday and Maldacena have managed to reproduce the
exponential formula (2.7) at strong coupling for the four-point case using the AdS/CFT correspondence [7]. The AdS dual description of a planar colour-ordered amplitude in N = 4 SYM
is given by a classical open string worldsheet ending on a brane placed in the far infrared region
of AdS space. Specifically, using coordinates in which the metric of AdS space is
2
2
2
2 dx + dz
,
ds = R
(2.9)
z2
then the boundary of AdS space is at z = 0 and the infrared brane sits at z = .
It is convenient however to use a T-dual description of this string configuration. This T-duality
maps the AdS space into a new space with coordinates (y , r) which has the metric
2
2
2
2 dy + dr
,
d s = R
(2.10)
r2
where r = R 2 /z. We see that the new space is again AdS, but now the infrared region and
the boundary have been inverted. Therefore, the brane is located on the boundary in the new
coordinates. Furthermore, the momenta of the particles pi are expressed as differences of dual,
or region momenta yi [24], (2)pi = yi yi+1 . The calculation of the amplitude thus becomes
that of finding the classical action Scl of a string worldsheet whose boundary is a polygon with
vertices yi lying within the AdS boundary,
Mn eiScl .
(2.11)
For the four-point

amplitude, the corresponding string solution can be determined [7] giving
iScl = div +( /8) log2 (s/t) + C where div represents divergent terms. This agrees precisely
236
with the BDS conjecture (2.8) and predicts K = /4 at large = g 2 N , in agreement with
string calculations [25,26].
Now, the minimal area of a string ending on a path in the boundary of AdS space gives the
vacuum expectation value of the Wilson loop over the same path in the CFT at strong coupling
[27,28]. A subtlety arising in the case at hand is the presence of singular points or cusps in
the path, which lead to divergences [7,29]. Nevertheless the divergences can be regularised by
dimensional reduction even in the string calculation. Therefore, at least at strong coupling there
is evidence for a dual description of amplitudes as Wilson loops. In the next section we will
consider this possibility at small coupling. An important point to note here is that the string
calculation does not depend on the species or helicities of the particles in the amplitude. These
are subleading terms which would require corrections [7,30]. Our Wilson loop calculation is
also insensitive to the helicities of the scattered particles; thus, similarly to the AldayMaldacena
result, it does not generate the tree-level ParkeTaylor amplitude.
One mysterious and intriguing consequence of this dual description of amplitudes as Wilson
loops is the unexpected appearance of conformal symmetry. Wilson loops of smooth paths in
N = 4 SYM are conformally invariant objects (modulo an anomaly which does not depend
either on the shape or size of the loop [31,32]). However here the Wilson loop is divergent, since
the path is not smooth, and regularisation spoils the conformal symmetry. Nevertheless, a similar
pseudo-conformality seems to appear at weak coupling where all the integrals contributing to
four-point MHV diagrams can be determined by rewriting them using the region momenta and
appealing to off-shell conformality [3,5,8]. Furthermore at four points all conformal integrals of
a certain type and with certain singular properties appear with coefficients 1. The Wilson loop
picture would seem to suggest that this pseudo-conformal invariance should continue for n-point
functions. This point clearly deserves further investigation.
3. MHV amplitudes from a Wilson loop calculation
In this section we calculate one-loop corrections to the vacuum expectation value of a particular Wilson loop. In N = 4 SYM, the appropriate operator takes the form (suppressing fermions)
[3335]

i

W [C] := Tr P exp ig d A x( ) x ( ) + i x( ) y ( ) ,
(3.1)
C
where the i s are the six scalar fields of N = 4 SYM, and (x ( ), y i ( )) parametrise the loop C.
Importantly, when x 2 = y 2 , the Wilson loop (3.1) is locally supersymmetric. The specific form of
the contour C we choose is dictated by the gluon momenta p1 , . . . , pn . Specifically, the segment
associated to momentum pi will be delimited by ki and ki+1 ,
pi := ki ki+1 ,
(3.2)
and will be parametrised

as ki (i ) := ki + i (ki+1 ki ) = ki i pi , i [0, 1]. Momentum

conservation ni=1 pi = 0 implies that the contour is closed. In addition we set y i = 0 which
makes the Wilson loop locally supersymmetric, as the gluon momenta, and hence the segments
of the contour are null.4 However, we notice that each segment of the loop preserves a different
subset of supersymmetries, therefore supersymmetry is broken globally.
4 We thank George Georgiou for discussions on this point.
237
We notice that the coordinates ki can be interpreted as dual, or region momenta [24]. Indeed,
for any planar diagrams one can express the momentum carried by a line as the difference of
the momenta of the two regions of the plane separated by the segment. These coordinates have
been used in the context of amplitude calculations by Thorn and collaborators in [36,37] and,
more recently, in [38] in the context of MHV diagrams. They are also the T-dual coordinates
introduced in [7] which determine the classical string solutions, as discussed earlier, and appear
in the conformal integrals discussed in [5].
The four-particle case was recently addressed in [8], where it was found that the result of a
one-loop Wilson loop calculation reproduces the four-point MHV amplitude in N = 4 SYM.
Here we extend this result in two directions. First, we derive the four-point MHV amplitude to
all-orders in the dimensional regularisation parameter . Secondly, we show that this striking
agreement persists for an MHV amplitude with an arbitrary number of external particles.
Three different classes of diagrams give one-loop corrections to the Wilson loop.5 In the
first one, a gluon stretches between points belonging to the same segment. It is immediately
seen [8] that these diagrams give a vanishing contribution. In the second class of diagrams, a
gluon stretches between two adjacent segments meeting at a cusp. Such diagrams are ultraviolet
divergent and were calculated long ago [3946], specifically in [45,46] for the case of gluons
attached to lightlike segments.
In order to compute these diagrams, we will use the gluon propagator in the dual configuration
space, which in D = 4 2UV dimensions is

D
2 2
D
1
(z) :=
D
2
4 2
2
(z + i) 2 1
UV
=
(1 UV )
.
(3.3)
2
2
4
(z + i)1UV
A typical diagram in the second class is pictured in Fig. 2. There one has k1 (1 ) k2 (2 ) =
p1 (1 1 ) + p2 2 , where we used p1 = k1 k2 and p2 = k2 k3 . The cusp diagram then gives6
2 (1 UV )
ig UV
4 2UV
1
(p1 p2 )
[(p1 1 + p2 2 )2 ]1UV
0

1 (s)UV
UV 2 (1 UV )
= ig
.
2
2 UV
4 2UV
d1 d2
(3.4)
The UV divergence should be interpreted as a divergence at small differences of region momenta,

i.e., momenta, hence we interpret it as an infrared singularity in momentum space. Notice that
UV > 0, in order to regulate the divergence in (3.4). Furthermore the scale used in the Wilson
loop calculation is related to the scale used to regulate the amplitudes as = (c)1 (the precise coefficient c in front of can be reabsorbed into an appropriate redefinition of the coupling
constant).
The last class of diagrams consists of diagrams where the gluon connects non-adjacent segments, such as that pictured in Fig. 3. We denote by p and q the momenta carried by the
two segments, and calculate the one-loop contribution due to the gluon exchange. We also set
5 Notice that, for a Wilson loop bounded by gluons, we can only exchange gluons at one loop.
6 After changing variables 1 .
1
1
238
Fig. 2. A one-loop correction to the Wilson loop, where the gluon stretches between two lightlike momenta meeting at a
cusp. Diagrams in this class provide the infrared-divergent terms in the n-point scattering amplitudes, given in (2.6).
Fig. 3. Diagrams in this classwhere a gluon connects two non-adjacent segmentsare finite, and give a contribution
equal to the finite part of a two-mass easy box function F 2me (p, q, P , Q), second line of (2.3). p and q are the massless
legs of the two-mass easy box, and correspond to the segments which are connected by the gluon. The diagram depends
on the other gluon momenta only through the combinations P and Q.
p := kp kp+1 , q := kq kq+1 . The gluon propagator is a function of

q1
2

2
kp (p ) kq (q ) =
(ki ki+1 ) p p + q q

i=p
= p(1 p ) + qq +
q1
2
(ki ki+1 )
(3.5)
i=p+1
q1
We recognise that i=p+1 (ki ki+1 ) = P is the sum of the momenta between p and q, where,
in general, P 2 = 0. Hence
2

kp (p ) kq (q ) = P 2 + 2(pP )(1 p ) + 2(qP )q + 2(pq)p q

= P 2 + s P 2 (1 p ) + t P 2 q

+ s t + P 2 + Q2 p q ,
(3.6)
239
where we have re-expressed the result in terms of the invariants s := (P + p)2 , t := (P + q)2
defined earlier. We can also introduce u := s t + P 2 + Q2 .
The one-loop diagram in Fig. 3 is equal to

2 1 (1 UV )
ig UV
F (s, t, P , Q),
2 4 2UV
where F (s, t, P , Q) is the following two-dimensional integral,7
F (s, t, P , Q)
1
= dp dq
(3.7)
P 2 + Q2 s t
.
2
2
2
1+
+ (s
p + (t P )q + (s t + P + Q )p q )]
0
(3.8)
The integral is finite in four dimensions. We begin by calculating it in four dimensions setting
= 0 (and will come back later to the calculation for = 0). In this case, the result is
[(P 2
P 2 )
F=0 (s, t, P , Q)

= Li2 (as) + Li2 (at) Li2 aP 2 Li2 aQ2
(P 2 t)(Q2 t)
(P 2 s)(Q2 s)
+
log
t
log
P 2 Q2 st
P 2 Q2 st
2
2
(P s)(P t)
(Q2 s)(Q2 t)
2
log
Q
,
log
log P 2 log
P 2 Q2 st
P 2 Q2 st
where a is defined in (2.4). Using Eulers identity
+ log s log
Li2 (z) = Li2 (1 z) log z log(1 z) +
2
,
6
(3.9)
(3.10)
and noticing that [10] (1 as)(1 at)/[(1 aP 2 )(1 aQ2 )] = 1, we can rewrite

Li2 (as) + Li2 (at) Li2 aP 2 Li2 aQ2

= Li2 (1 as) Li2 (1 at) + Li2 1 aP 2 + Li2 1 aQ2 log s log(1 as)

log t log(1 at) + log P 2 log 1 aP 2 + log Q2 log 1 aQ2 .
(3.11)
Upon making use of the relations [10]
(t P 2 )(t Q2 )
(s P 2 )(s Q2 )
,
1
at
=
,
P 2 Q2 st
P 2 Q2 st
(s P 2 )(t P 2 )
(s Q2 )(t Q2 )
2
1 aP 2 =
,
1
aQ
,
=
P 2 Q2 st
P 2 Q2 st
we see that the terms in (3.9) containing logarithms cancel, and we are left with

F=0 = Li2 (1 as) Li2 (1 at) + Li2 1 aP 2 + Li2 1 aQ2 .
1 as =
(3.12)
(3.13)
We conclude that F=0 is precisely equal to the finite part of the two-mass easy box function
second line of (2.3). Finally, summing over all possible pairs of non-adjacent segments reproduces precisely the sum over box functions in (2.2).
7 In the following we set :=
UV < 0, where will correspond to the usual infrared regulator.
240
In the four-point case, a|P 2 =Q2 =0 . = 1/s + 1/t, and using

2
1
1
+ log2 (z) +
= 0,
Li2 (z) + Li2
z
2
6
(3.14)
one immediately finds [8]

1
2
2 s
+
,
F=0 P 2 =Q2 =0 = log
2
t
2
(3.15)
in complete agreement with the finite parts of the zero-mass box function (in the normalisations
of [9]).
Finally, we discuss the calculation at = 0. In this case one finds that

1
1
a
F = 2
2 F1 , , 1 + ,

1 aP 2
1 aP 2

a
1
1
a
+
F
F
,
,
1
+
,
,
,
1
+
,
1
1
2
2
1 as
1 as
1 aQ2
1 aQ2

a
1
(3.16)
.
2 F1 , , 1 + ,
1 at
1 at
This result is in precise agreement with the finite part of the all-orders in two-mass easy box
function, second and third line of (2.5). Notice that the expression in (3.16) is finite as 0.
For n > 4, simply replacing the = 0 expression of the box functions with their all-orders in
expression does not provide us with a complete, all-orders in expression for the amplitude,
as there are additional n-gon integrals which vanish as 0, which are not included. The fourpoint case is an exception, and in this case our calculation reproduces the expected all-orders in
result. Combining the infrared-divergent and finite terms, our result is

2
s
s
(1)
M4 () = 2
2 F1 1, , 1 , 1 +
t

2

t
t
+
(3.17)
,
2 F1 1, , 1 , 1 +
2
s
in agreement with [47].

Finally, we would like to mention that it is easy to see that splitting amplitudes and soft functions can also be derived at one loop using Wilson loops. Furthermore, they have an interesting
interpretation in terms of the geometry of the contour; for instance, splitting amplitudes arise
from two adjacent segments becoming nearly parallel to each other, therefore merging into a
single segment.
4. Conclusions
We found that a strikingly simple one-loop gluon exchange calculation of a Wilson loop,
whose (closed) boundary is defined by a set of lightlike segments p1 , . . . , pn , reproduces the npoint one-loop MHV amplitudes in N = 4 SYM (divided by the tree-level amplitude). We would
like to comment on this surprising result.
1. One of the important features of the calculation presented in this paper is that it neatly
separates the infrared-divergent terms from the finite parts. The latter can then be derived working
241
directly in four dimensions, which turns out to be a key calculational advantage. For this reason,
we are hopeful that this procedure could allow for a direct check of the exponentiation of the finite
remainders [1]. The perspective of deriving a field-theoretical proof of the all-loop expressions
of [1] from Wilson loopspossibly using the non-Abelian exponentiation theorem [48,49]is
an exciting one.
2. In [10], it was speculated that the two-mass easy box functions should emerge naturally
as Feynman diagrams of the perturbative description of a (possibly string) theory. This was motivated by the observation that the MHV amplitudes in N = 4 SYM at one loop are written as
sum of box functions, each appearing with coefficient one. In this paper, we have found that
the Wilson loop calculation gives a precise, one-to-one mapping of Wilson loop diagrams to the
finite part of two-mass easy box functions (or, in specific cases, the degenerate one-mass and
zero-mass functions). The massless legs of the box function, p and q, are simply those to which
the gluon is attached (see Fig. 3). The calculation is only sensitive to p, q, and the sum P of
the momenta between p and q. Therefore, the Wilson loop calculation seems to have provided
such a description where a two-mass easy box is a specific Wilson loop diagram (notice that for
the diagram with a gluon connecting segments p and q, the remaining part of the loop could
be deformed to the shape of a (generically) two-mass box function).8 It is tempting to speculate
that the observation of [3] that even at higher loops the MHV amplitudes are expressed as sums
of (conformal) integrals, each appearing with coefficients 1, could be explained in terms of
higher-loop Wilson loop diagrams.
3. We believe that the agreement found in this paper between our Wilson loop calculation and
the N = 4 MHV amplitude with an arbitrary number of points lends support to the conjecture
that the appropriate strongly-coupled string theory calculation at n points will confirm the BDS
conjecture for the full exponentiated expression of the n-point MHV amplitude.
Acknowledgements
It is a pleasure to thank James Drummond, George Georgiou, Valeria Gili, Gregory Korchemsky, Sanjaye Ramgoolam, Rodolfo Russo, Emery Sokatchev, Bill Spence and Costas Zoubos for
discussions. We would like to thank PPARC for support under a Rolling Grant PP/D507323/1
and the Special Programme Grant PP/C50426X/1. The work of P.H. is supported by an EPSRC
Standard Research Grant EP/C544250/1. G.T. is supported by an EPSRC Advanced Fellowship
EP/C544242/1 and by an EPSRC Standard Research Grant EP/C544250/1.
References
[2] N. Beisert, B. Eden, M. Staudacher, Transcendentality and crossing, J. Stat. Mech. 0701 (2007) P021, hep-th/
0610251.
[3] Z. Bern, M. Czakon, L.J. Dixon, D.A. Kosower, V.A. Smirnov, The four-loop planar amplitude and cusp anomalous
dimension in maximally supersymmetric YangMills theory, Phys. Rev. D 75 (2007) 085010, hep-th/0610248.
[4] F. Cachazo, M. Spradlin, A. Volovich, Four-loop cusp anomalous dimension from obstructions, Phys. Rev. D 75
(2007) 105011, hep-th/0612309.
8 We also note in passing that we have found a representation of the finite part of two-mass easy box functions in terms
of a very simple two-dimensional integral, see (3.8) and (3.13).
242
[5] J.M. Drummond, J. Henn, V.A. Smirnov, E. Sokatchev, Magic identities for conformal four-point integrals,
JHEP 0701 (2007) 064, hep-th/0607160.
[6] Z. Bern, J.J.M. Carrasco, H. Johansson, D.A. Kosower, Maximally supersymmetric planar YangMills amplitudes
at five loops, 0705.1864 [hep-th].
[7] L.F. Alday, J. Maldacena, Gluon scattering amplitudes at strong coupling, 0705.0303 [hep-th].
[8] J.M. Drummond, G.P. Korchemsky, E. Sokatchev, Conformal properties of four-gluon planar amplitudes and Wilson
loops, 0707.0243 [hep-th].
[9] Z. Bern, L.J. Dixon, D.C. Dunbar, D.A. Kosower, One-loop N -point gauge theory amplitudes, unitarity and collinear
limits, Nucl. Phys. B 425 (1994) 217, hep-ph/9403226.
[10] A. Brandhuber, B. Spence, G. Travaglini, One-loop gauge theory amplitudes in N = 4 super-YangMills from
MHV vertices, Nucl. Phys. B 706 (2005) 150, hep-th/0407214.
[11] Z. Bern, L.J. Dixon, D.A. Kosower, Dimensionally regulated pentagon integrals, Nucl. Phys. B 412 (1994) 751,
hep-ph/9306240.
[12] G. Duplancic, B. Nizic, Dimensionally regulated one-loop box scalar integrals with massless internal lines, Eur.
Phys. J. C 20 (2001) 357, hep-ph/0006249.
[13] A. Brandhuber, B. Spence, G. Travaglini, From trees to loops and back, JHEP 0601 (2006) 142, hep-th/0510253.
[14] C. Anastasiou, Z. Bern, L.J. Dixon, D.A. Kosower, Planar amplitudes in maximally supersymmetric YangMills
theory, Phys. Rev. Lett. 91 (2003) 251602, hep-th/0309040.
[15] Z. Bern, M. Czakon, D.A. Kosower, R. Roiban, V.A. Smirnov, Two-loop iteration of five-point N = 4 super-Yang
Mills amplitudes, Phys. Rev. Lett. 97 (2006) 181601, hep-th/0604074.
[16] A.H. Mueller, On the asymptotic behavior of the Sudakov form-factor, Phys. Rev. D 20 (1979) 2037.
[17] J.C. Collins, Algorithm to compute corrections to the Sudakov form-factor, Phys. Rev. D 22 (1980) 1478.
[18] A. Sen, Asymptotic behavior of the Sudakov form-factor in QCD, Phys. Rev. D 24 (1981) 3281.
[19] G.P. Korchemsky, Double logarithmic asymptotics in QCD, Phys. Lett. B 217 (1989) 330.
[20] S. Catani, L. Trentadue, Resummation of the QCD perturbative series for hard processes, Nucl. Phys. B 327 (1989)
323.
[21] L. Magnea, G. Sterman, Analytic continuation of the Sudakov form-factor in QCD, Phys. Rev. D 42 (1990) 4222.
[22] S. Catani, The singular behaviour of QCD amplitudes at two-loop order, Phys. Lett. B 427 (1998) 161, hep-ph/
9802439.
[23] G. Sterman, M.E. Tejeda-Yeomans, Multi-loop amplitudes and resummation, Phys. Lett. B 552 (2003) 48, hep-ph/
0210130.
[24] G. t Hooft, A planar theory for strong interactions, Nucl. Phys. B 72 (1974) 461.
[25] S.S. Gubser, I.R. Klebanov, A.M. Polyakov, A semi-classical limit of the gauge/string correspondence, Nucl. Phys.
B 636 (2002) 99, hep-th/0204051.
[26] S. Frolov, A.A. Tseytlin, Semiclassical quantization of rotating superstring in AdS5 S 5 , JHEP 0206 (2002) 007,
hep-th/0204226.
[27] S.J. Rey, J.T. Yee, Macroscopic strings as heavy quarks in large N gauge theory and anti-de Sitter supergravity, Eur.
Phys. J. C 22 (2001) 379, hep-th/9803001.
[29] E.I. Buchbinder, Infrared limit of gluon amplitudes at strong coupling, 0706.2015 [hep-th].
[30] S. Abel, S. Forste, V.V. Khoze, Scattering amplitudes in strongly coupled N = 4 SYM from semiclassical strings in
AdS, 0705.2113 [hep-th].
[31] J.K. Erickson, G.W. Semenoff, K. Zarembo, Wilson loops in N = 4 supersymmetric YangMills theory, Nucl. Phys.
B 582 (2000) 155, hep-th/0003055.
[32] N. Drukker, D.J. Gross, An exact prediction of N = 4 SUSYM theory for string theory, J. Math. Phys. 42 (2001)
2896, hep-th/0010274.
[34] S.J. Rey, J.T. Yee, Macroscopic strings as heavy quarks in large N gauge theory and anti-de Sitter supergravity, Eur.
Phys. J. C 22 (2001) 379, hep-th/9803001.
[35] N. Drukker, D.J. Gross, H. Ooguri, Wilson loops and minimal surfaces, Phys. Rev. D 60 (1999) 125006, hep-th/
9904191.
[36] D. Chakrabarti, J. Qiu, C.B. Thorn, Scattering of glue by glue on the light-cone worldsheet I: Helicity nonconserving amplitudes, Phys. Rev. D 72 (2005) 065022, hep-th/0507280.
[37] D. Chakrabarti, J. Qiu, C.B. Thorn, Scattering of glue by glue on the light-cone worldsheet II: Helicity conserving
amplitudes, Phys. Rev. D 74 (2006) 045018, hep-th/0602026.
[38] A. Brandhuber, B. Spence, G. Travaglini, K. Zoubos, One-loop MHV rules and pure YangMills, 0704.0245 [hepth].
243
[39] A.M. Polyakov, Gauge fields as rings of glue, Nucl. Phys. B 164 (1980) 171.
[40] I.Y. Arefeva, Quantum contour field equations, Phys. Lett. B 93 (1980) 347.
[41] S.V. Ivanov, G.P. Korchemsky, A.V. Radyushkin, Infrared asymptotics of perturbative QCD: Contour gauges, Yad.
Fiz. 44 (1986) 230;
S.V. Ivanov, G.P. Korchemsky, A.V. Radyushkin, Sov. J. Nucl. Phys. 44 (1986) 145.
[42] V.S. Dotsenko, S.N. Vergeles, Renormalizability of phase factors in the non-Abelian Gauge theory, Nucl. Phys.
B 169 (1980) 527.
[43] R.A. Brandt, F. Neri, M.A. Sato, Renormalization of loop functions for all loops, Phys. Rev. D 24 (1981) 879.
[44] G.P. Korchemsky, A.V. Radyushkin, Loop space formalism and renormalization group for the infrared asymptotics
of QCD, Phys. Lett. B 171 (1986) 459.
[45] I.A. Korchemskaya, G.P. Korchemsky, On lightlike Wilson loops, Phys. Lett. B 287 (1992) 169.
[46] A. Bassetto, I.A. Korchemskaya, G.P. Korchemsky, G. Nardelli, Gauge invariance and anomalous dimensions of a
light cone Wilson loop in lightlike axial gauge, Nucl. Phys. B 408 (1993) 62, hep-ph/9303314.
[47] M.B. Green, J.H. Schwarz, L. Brink, N = 4 YangMills and N = 8 supergravity as limits of string theories, Nucl.
Phys. B 198 (1982) 474.
[48] J.G.M. Gatheral, Exponentiation of eikonal cross-sections in non-Abelian gauge theories, Phys. Lett. B 133 (1983)
90.
[49] J. Frenkel, J.C. Taylor, Non-Abelian eikonal exponentiation, Nucl. Phys. B 246 (1984) 231.

Towards a fitting procedure for deeply virtual Compton

scattering at next-to-leading order and beyond
K. Kumericki a,b , D. Mller b,c,d , K. Passek-Kumericki b,e,
a Department of Physics, Faculty of Science, University of Zagreb, PO Box 331, HR-10002 Zagreb, Croatia
b Institut fr Theoretische Physik, Universitt Regensburg, D-93040 Regensburg, Germany
c Department of Physics and Astronomy, Arizona State University, Tempe, AZ 85287-1504, USA
d Institut fr Theoretische Physik II, Ruhr-Universitt Bochum, D-44780 Bochum, Germany
e Theoretical Physics Division, Rudjer Bokovic Institute, PO Box 180, HR-10002 Zagreb, Croatia
Received 11 October 2007; accepted 25 October 2007

Abstract
Combining dispersion and operator product expansion techniques, we derive the conformal partial wave
decomposition of the virtual Compton scattering amplitude in terms of complex conformal spin to twist-two
accuracy. The perturbation theory predictions for the deeply virtual Compton scattering (DVCS) amplitude
are presented in next-to-leading order for both conformal and modified minimal subtraction scheme. Within
a conformal subtraction scheme, where we exploit predictive power of conformal symmetry, the radiative
corrections are presented up to next-to-next-to-leading order accuracy. Here, because of the trace anomaly,
the mixing of conformal moments of generalized parton distributions (GPD) at the three-loop level remains
unknown. Within a new proposed parameterization for GPDs, we then study the convergence of perturbation
theory and demonstrate that our formalism is suitable for a fitting procedure of DVCS observables.
PACS: 11.25.Db; 12.38.Bx; 13.60.Fz
Keywords: Deeply virtual Compton scattering; Next-to-next-to-leading order corrections; Fitting procedure;
Generalized parton distributions; Conformal symmetry
* Corresponding author at: Theoretical Physics Division, Rudjer Bokovic Institute, PO Box 180, HR-10002 Zagreb,
Croatia.
E-mail address: passek@thphys.irb.hr (K. Passek-Kumericki).
doi:10.1016/j.nuclphysb.2007.10.029
K. Kumericki et al. / Nuclear Physics B 794 (2008) 244323
245
1. Introduction
The partonic content of the nucleon, other hadrons, and nuclei has been studied within hard
inclusive processes over almost four decades. Since partons are confined by the strong force,
their response to, e.g., an electromagnetic probe cannot be directly accessed in experiments;
the long-range interaction in their environment remains essential. Fortunately, based on factorization theorems [1], the short- and the long-distance interaction can often be separated. The
former depends on the specific process and can be systematically computed using perturbation
theory, while the latter is encoded in process-independent non-perturbative quantities, e.g., parton densities. These densities have a semiclassical interpretation within the parton picture, which
is, however, not independent of the conventions used in the evaluation of short-distance physics.
Together with an increasing amount of experimental data, the perturbative factorization approach
leads to a deeper and more precise insight into the hadronic world, which is essential even for
the search of new physics.
Most importantly, the factorization approach is a tool that relates observables measured in
various inclusive processes and thus has a predictive power. The improvement of quantitative
predictions requires, besides precise experimental measurements, also a refinement of the theoretical approach. This includes both the perturbative evaluation of radiative corrections to the
short-distance physics at higher orders and an understanding of the so-called power suppressed
corrections, which alter the factorization theorems. In particular, the effort in the perturbative
sector, which reached the three-loop level, led to a quantitative understanding of the inclusive
QCD physics at the level of a few percent. In practice, parton densities have for a long time
been extracted via a theoretically motivated functional ansatz, see, e.g., Ref. [2], depending on a
number of parameters, that is globally fitted to experimental data [310]. In such fits, based on
traditional 2 minimization, the error estimation using the hypothesis of linear error propagation
has become standard in the last few years [35,8]. To overcome drawbacks of this traditional
method, alternative frameworks of statistical inference [11] and neural network parameterization
[12,13] have been proposed.
In contrast to inclusive processes, the understanding and the theoretical description of exclusive ones remain poor. In general, there is a lack of experimental data and so the underlying
theoretical framework, e.g., the factorization of hadronic amplitudes in terms of distribution amplitudes [1418], cannot be quantitatively tested. Moreover, the theoretical framework at the
next-to-leading order (NLO) is developed for only a few processes and the applicability of the
factorization approach at accessible scales is controversially discussed in the literature [1921].
The old problem of whether all valence partons take part in the short-distance process or only the
active one (Feynman mechanism) remains open. A way out of this theoretical dilemma might be
provided by the use of light-cone sum rules [22,23].
Some years ago a new non-perturbative generalized distribution amplitude1 was proposed as
a means to access the partonic content of hadrons [24,25]. In particular, it was theoretically studied in connection with deeply virtual Compton scattering (DVCS), where the partonic content
of a hadron is probed by two photons. To leading order, the underlying picture portrays a parton
which is probed by a virtual photon, travels near the light cone, emits a real photon, and remains a constituent of the probed hadron. The partonic probability amplitude for such a process
1 This notion, proposed in Ref. [24], stood for expectation values of light-ray operators sandwiched between vacuum
and final hadronic or between initial and final hadronic states. Today, it denotes the distribution corresponding to the
former expectation value, while the distribution corresponding to the latter one is called generalized parton distribution.
246
is given by the generalized parton distribution (GPD) [2628]. It has soon been realized that
diffractive vector meson electroproduction [29], measured in the collider experiments H1 and
ZEUS at DESY (see, e.g., Refs. [3032]), can be described in terms of GPDs [33,34], too. The
usefulness of the GPDs has also been widely realized in connection with the spin problem, since
they encode the angular momentum carried by the individual parton species, as explicated by the
Jis sum rule [26]. During the last decade they have become an attractive object for theoretical
and experimental investigations. Today, they are considered as a new concept that provides a link
between different fields: exclusive and inclusive processes, perturbative and non-perturbative
physics (e.g., lattice simulations [3540] and model building [4149]). For comprehensive reviews, see Refs. [50,51].
In contrast to parton densities, which depend on the longitudinal momentum fraction of the
probed parton and on the resolution scale, GPDs encode also transversal degrees of freedom.
This allows a three-dimensional probabilistic interpretation of the parton distribution either in
the infinite momentum [5256] or in the rest frame [57]. Indeed, to some extent this information
can already be extracted from present experiments. The main theoretical complication arises from
the fact that GPDs depend on two longitudinal momentum fractions: x and skewness , which
are related to the s- and t -channel exchanges, respectively. The former one is either integrated out
during the convolution (in the real part of the scattering amplitude) or identified with the latter one
(in the imaginary part of the scattering amplitude). Even if we precisely knew the modulus and
the phase of hard exclusive leptoproduction scattering amplitudes,2 a complete reconstruction
of the GPD would not be uniquely possible, except with ideal data and employing evolution.
We note that it was pointed out that already the Fourier transform of the DVCS amplitude with
respect to the skewness parameter provides an image of the target with respect to the longitudinal
degrees of freedom [61,62]. The problem of deconvolution can be overcome to some extent
either by a realistic model, which we at present do not have, or by a hypothesis about the form
of skewness dependence. In the case of those GPDs that in the forward kinematics reduce to
parton densities, such a hypothesis can be tested by statistical analysis of inclusive and exclusive
data, e.g., by means of the 2 criteria. Certainly, if the skewness problem could be solved, the
extrapolation to the 0 case would be simple and would be a step towards the experimental
access of both Jis spin sum rule and a three-dimensional picture of the proton.
Among processes which enable us to access GPDs at present experiments, the DVCS is considered the theoretically cleanest one. Indeed, the first experimental DVCS data on the beam
spin asymmetry in fixed target experiments [63,64], or the cross section, measured by the H1
and ZEUS Collaborations [6567], could be successfully understood even in terms of oversimplified GPD anstze [68]. In contrast, the normalization of the cross sections of vector-meson
electroproduction, predicted to LO in the collinear factorization approach, in general overshoots
the H1 and ZEUS data. This process is widely believed to be affected by power-suppressed contributions. They are mainly related to the transversal size of the meson and appear so separately
as factorized contributions, which might be modelled by themselves [69,70]. On the other hand,
perturbative higher-order corrections for this process are large and reduce the size of the predicted cross sections [71]. However, to the best of our knowledge, a global NLO analysis of all
available experimental data has not been achieved so far. In our opinion, this is the only possible
2 We remark that the scan of the GPD shape in a certain region of the momentum fraction plane is possible only in the
so-called double DVCS process [5860] by variation of the incoming photon virtuality and the lepton pair mass.
247
way of confronting the collinear approach with experimental findings and before this is done, a
judgement about this approach can hardly be drawn.
With increasing amount and precision of experimental DVCS data [7275], there arises a
need for a better theoretical understanding of this process. Certainly, for the phenomenology of
GPDs it is essential to include and estimate perturbative and power-suppressed contributions as
well as to introduce a flexible parameterization of them. Besides a qualitative or a semi-analytic
understanding of observables in dependence on GPDs, see here, for instance, Refs. [68,76], also
fast and stable numerical routines are needed for a fitting procedure.
So far the perturbative contributions to the DVCS have been worked out to NLO accuracy
[7781], including the evolution [8284]. There exist numerical routines in the momentum fraction space [8587]. In this space the GPD ansatz is given in terms of a spectral function [24,88].
Here one has to model the functional dependencies on two momentum fraction variables and the
momentum transfer squared. Alternatively, one can work with the conformal moments of GPDs
[89], which diagonalize the LO evolution equation. This offers the possibility of a flexible GPD
parameterization, which covers the complete set of degrees of freedom and makes direct contact
with lattice results. The problem here is rather to find an appropriate and physically motivated
truncation in the parameter space. A partial wave expansion with respect to the angular momentum, the so-called dual GPD parameterization, has been proposed [90], where the partial wave
amplitudes are those of the mesonic exchanges in the t -channel with given angular momentum.
This also allows one to use the angular momentum as an expansion parameter [91] and Regge
phenomenology as a guideline towards realistic GPD anstze.
Unfortunately, if one works with discrete conformal moments, the GPDs are expanded in
terms of (mathematical) distributions, which live only in the so-called central region (
x ) of the whole support (1 x 1). This is quite analogous to the expansion of parton
densities with respect to the Dirac function and its derivatives, which live at the point x = 0.
A variety of approaches have been proposed to resum this formal series and thus to restore
the correct support of GPDs: smearing the expansion [89,90], an integral transformation [92,
93], Fourier transformation in the light-cone position space [9497], and a SommerfeldWatson
transformation [98]. Note that all these approaches should be mathematically equivalent and, in
particular, those of Refs. [96] and [98] lead to essentially the same representation. So far only the
smearing method has been extended to NLO accuracy in the MS scheme; however, speed and
stability remain restricted [99,100].
We believe that the GPD formalism that is based on the SommerfeldWatson transformation
and MellinBarnes integral representation [98] is suitable to satisfy the requirements which we
spelled out. Moreover, we can employ the power of conformal symmetry to investigate the convergence of the perturbative series up to NNLO [101,102]. Therefore, the fivefold goal of this
article is
to present a detailed derivation, based on analyticity and short-distance operator product
expansion, of the MellinBarnes integral representation for the twist-two Compton form
factors (CFFs),
to present the radiative corrections up to NNLO obtained using the predictive power of conformal symmetry, as well as to present the radiative corrections in the standard MS scheme
up to NLO accuracy, including the evolution,
to propose an intuitive ansatz for conformal GPD moments,
to investigate the convergency properties of the perturbative expansion,
248
to demonstrate the usefulness of the GPD fitting procedure based on this formalism, as well
as to discuss the requirements for a GPD ansatz.
The outline is as follows. In Section 2 we introduce the Compton scattering tensor and its decomposition in leading twist-two Compton form factors (CFFs). Employing their analytic properties,
we derive dispersion relations for CFFs. In Section 3 these dispersion relations together with
the short-distance operator product expansion are used to represent the CFFs as MellinBarnes
integrals. To diagonalize the evolution operator, we then introduce the conformal partial wave
expansion of CFFs. In Section 4 we employ conformal symmetry in the perturbative QCD sector
to find the Wilson coefficients for CFFs, expanded up to NNLO in the coupling. We also employ
the conformal partial wave expansion for the representation of the NLO corrections in the MS
scheme. An intuitive ansatz for the conformal GPD moments is introduced in Section 5. It is
based on the internal duality of GPDs and consists of an SO(3) partial wave expansion in the
t-channel. In Section 6 we discuss the convergency of the perturbative series. We then demonstrate in Section 7 on hand of the DVCS cross section, measured at high energies by the H1
and ZEUS Collaborations, that our formalism is suitable for a fitting procedure that is used to
extract GPD information from experimental data. Finally, we summarize and conclude. Three
appendices contain details on our conventions and the evaluation of conformal moments.
2. Compton scattering tensor and Compton form factors
2.1. General formalism
We are interested in the perturbative description of hard photon leptoproduction off a hadronic
target, e.g., a proton. Besides the BetheHeitler bremsstrahlungs process, parameterized by electromagnetic form factors, the DVCS process contributes [24,26,28]. The amplitude of the latter
is expressed by the Compton tensor. This tensor is defined in terms of the time-ordered product
of two electromagnetic currents, sandwiched between the initial and final hadronic state,

i
T (q, P , ) = 2 d 4 x eixq P2 , S2 |Tj (x/2)j (x/2)|P1 , S1 ,
(1)
e
where q = (q1 + q2 )/2 ( and q2 refer to the outgoing real photon), while P = P1 + P2 and
= P2 P1 . The incoming photon has a large virtuality q12 = Q2 and we require that, in the
limit q 2 = Q2 , the scaling variables
=
Q2
,
P q
q
,
P q
(2)
and the momentum transfer squared 2 are fixed. The dominant contributions arise then from
the light-cone singularities of the time-ordered product. This kinematics is a generalization of the
Bjorken limit, well-known from deep inelastic scattering (DIS). In particular, if the final photon
is on-shell, i.e., for the process we are interested in, the skewness parameter and the Bjorkenlike scaling parameter are equal to twist-two accuracy, i.e., = + O(1/Q2 ). Note that in the
following we assume P = P22 P12 = 0.
In the generalized Bjorken limit, we can employ the OPE to evaluate the Compton scattering tensor (1). Its dominant part is given by matrix elements of leading twist-two operators.
The use of the light-cone expansion, a resummed version of the short-distance operator product
expansion, allows a straightforward evaluation [24]. On the other hand, one might employ the
249
short-distance expansion, which yields a Taylor expansion of the Compton scattering tensor with
respect to the variable 1/ . Since this expansion converges only in the unphysical region, i.e.,
> 1, we need in addition a dispersion relation that connects the Mellin moments in the physical
region with the short-distance expansion. This technique well-known from deep inelastic scattering has been adopted for non-forward kinematics, e.g., in Ref. [103]. Below we use it within
a special short-distance OPE, namely, the conformally covariant one.
Let us introduce a parameterization of the Compton tensor in terms of the so-called Compton
form factors (CFFs), which has been employed for the evaluation of the differential cross section
[104,68]. To leading twist-two accuracy it reads
T
T (q, P , ) = g
q V
q A
+ ,
i qP
P q
(P q)2
(3)
T and

3
where g
qP q P are the transversal part of the metric and the Levi-Civita
tensor, respectively, which are contracted X P X P with projectors
q1 q2
(4)
q1 q 2
to ensure current conservation [104]. The ellipsis indicate terms that are finally power suppressed4 in the DVCS amplitude, or are determined by the gluon transversity GPD. The latter is
a twist-two contribution that enters at NLO, evaluated in [105,81,106], and is tied to a specific
azimuthal angular dependence in the cross section [107,68]. Hence, using an appropriate definition of observables, it can be separated from the other twist-two (and twist-three) contributions.
For the time being we consider its perturbative corrections beyond NLO more as an academic
issue and will not evaluate it here.
In the parity even sector the vector

i
U (P1 , S1 ) + ,
V = U (P2 , S2 ) H + E
(5)
2M
is decomposed into the target helicity conserving CFF H and the helicity flip one E. Analogously,
the axial-vector

5 + E 5 U (P1 , S1 ) + ,
A = U (P2 , S2 ) H
(6)
2M
where again higher-twist contributions are neglected. The
is parameterized in terms of H and E,
normalization of the spinors is U (P , S) U (P , S) = 2P .
It is convenient to introduce the nomenclature

E},
F = F V , F A , F V = {H, E}, F A = {H,

(7)
P = g
with the following choice of arguments: F(, , 2 , Q2 ). The variable

1 Q2
W2 u
P q
=
=
=
P2 P2
2 P2
with W 2 = (P1 + q1 )2 , u = (P1 q2 )2
(8)
3 The transversal part of the metric tensor reads g T = g

n n n n , where n = (1, 0, 0, 1)/ 2 and n =

(1, 0, 0, 1)/ 2 and at twist-2 we have n n = q P /(q P ).
4 Here we have neglected a further twist-two Compton form factor that can occur for virtual photons that are longi-
tudinal polarized. Obviously, it does not contribute to the DVCS amplitude, where the final photon state is of course
transversally polarized.
250
is in the Breit frame simply given by q0 , the conjugate variable of the time x0 . It is an appropriate
generalization of the photon energy in the rest frame for DIS. Instead of the skewness variable ,
depending on , we will employ the virtual photon mass asymmetry
=
q12 q22
q12
+ q22

+ O 2 /Q2 .
(9)
Obviously, = 0 in the forward case and = 1 for DVCS.

2.2. Analyticity and dispersion relations
Based on the analyticity of the Compton tensor (1), now we derive a dispersion relation for
the CFFs, introduced in Eqs. (5) and (6). From the definition of the Compton tensor (1), we read
off for complex valued q the following properties, see, for instance, Refs. [108,109],

T (q, P , ; S1 , S2 ) = T q , P , ; S2 , S1 ,
(10)
T (q, P , ; S1 , S2 ) = T (q, P , ; S1 , S2 ).
(11)
The former of these equations can be viewed as a generalization of the Schwarz reflection
principle, whereas the latter equation stems from crossing symmetry under the Bose exchange
(q1 , q2 , , ) (q2 , q1 , , ). Furthermore, in the spacelike region, i.e., 0 < Q2 , and for
0 2 2Max , where the upper limit is constrained by kinematics, the Compton tensor is
holomorphic in except for the branch cuts along the real axis. Its absorptive part

T (q, P , ; S1 , S2 ) T (q, P , ; S2 , S1 ) 4iW (q, P , ; S1 , S2 ),

(12)
is given by the commutator of the electromagnetic currents

1
d 4 x eixq P2 , S2 | j (x/2), j (x/2) |P1 , S1 .
W =
4e2
(13)
Finally, using the generalized Schwarz reflection principle (10), we have for the discontinuity,
now expressed in terms of the more appropriate variable ,
Disc T T ( + i, . . .) T ( i, . . .) = 4iW .
(14)
tell us
The spectral representation of the hadronic tensor W and baryon number conservation
that this discontinuity does not vanish for || cut with cut = (4Q2 + 2 )/4 P 2 , see Fig. 1.
Obviously, these analytic properties hold for the CFFs, too. The crossing relation (11) implies
the symmetry relations
F V (, . . .) = F V (, . . .),
F A (, . . .) = F A (, . . .).
(15)
To fix the form of the dispersion relation, we should estimate the high-energy behavior of the
CFFs from the common arguments of Regge phenomenology. The leading meson trajectories
2
give rise to a M ( ) behavior with M (2 ) M (0) 1/2 for 2 < 0. Thus, for the mesonexchange induced part it is appropriate to use Cauchys theorem with one subtraction, i.e., cf.
Fig. 1(a),

1
F , , 2 , Q2 =
2i

cut 0
cut +0

d
F , , 2 , Q2 + C , 2 , Q2 .

(16)
251
Fig. 1. The integration contour used in Eq. (16) and its deformation, employed to derive the dispersion relation (17), are
sketched in the left and right panel, respectively. Contribution of dashed segments vanishes.
In the axial-vector case, the subtraction constant C( ) F ( = 0, . . .) is zero for symmetry reasons, see Eq. (15). In the flavor singlet sector of parity even CFFs F V the pomeron
exchange appears. It induces a stronger power-like growth with an exponent P (2 ), where
P (2 ) P (0) 1. For this contribution a second subtraction should in principle be introduced in Eq. (16), giving rise to a C term. However, since a parity even CFFs is a even function
in , this constant C is zero. Hence, the analytic properties allow us to derive a singly subtracted
dispersion relation. Inflating the integration path to infinity, pictured in Fig. 1(b), we are left with
a path that encircles the cuts on the real axis. Picking up the discontinuity and employing now
the symmetry (15), we find for all CFFs and | | > 1 the dispersion relation5

F , , 2 , Q2
1
=
1
0

1
1
m F i0, , 2 , Q2 + C , 2 , Q2 .

+
(17)
Here, the upper and lower sign belong to the vector (V) and axial-vector (A) case, respectively,
where in the latter case the subtraction constant C vanishes. We have also used the fact that the
imaginary part m F ( , . . .) has the support | | 1/(1 + 2 /4Q2 ) 1.
The dispersion relation (17) will be used to relate the short-distance OPE, given as a Taylor
series with respect to the variable = 1/ < 1, to the imaginary part of the CFFs in the physical
region. If we choose to write the Taylor expansion of the CFFs with respect to as

1 (1)j j +1 Fj , 2 , Q2 + C , 2 , Q2 ,
F , , 2 , Q2 =
(18)
j =0
taking into account the symmetry relation (15), we arrive at

5 Although the function F is now expressed in terms of rather than , for simplicity we do not introduce a new
notation.
252
Fj ( )

1
d j +1

F(1/,
.
.
.)

j
+1
2(j + 1)! d
=0
1 (1)j
=
2
1
d j m F( i0, . . .).
(19)
This relation is valid for 0 j , where the subtraction term does not contribute. The latter enters
for j = 1, for which the first term in Eq. (17) drops out, and so we have the trivial identity

C , 2 , Q2 F V 1/, , 2 , Q2 =0 .
(20)
3. Operator product expansion to twist-two accuracy
Using the conformal OPE, presented in Section 3.1, and the dispersion relation, we derive in
Section 3.2 a MellinBarnes representation for CFFs. Then we give a detailed discussion of the
subtraction constant, which appears in a certain set of CFFs. On hand of a simple toy GPD model
we investigate some mathematical aspects of the MellinBarnes representation. In Section 3.3
we present the CFFs as conformal partial wave expansion in terms of conformal GPD moments,
depending on the complex conformal spin. We also provide the rotation of common Mellin GPD
moments to the conformal ones.
3.1. General form of the operator product expansion at twist-two level
The perturbative QCD predictions for the CFFs (7) might be derived by means of the OPE
for the time-ordered product of two electromagnetic currents, which is then plugged into the
Compton tensor (1). The predictive power of conformal symmetry allows for an economical
evaluation and is the key for the perturbative QCD predictions of the DVCS amplitude beyond
NLO, which will be presented in Section 4. Thus, we choose an operator basis in which this
symmetry can be manifestly implemented. The evaluation of the CFFs in this so-called conformal
OPE is straightforward and can be found in Refs. [110114,77,115]. In this section we mainly
present our conventions and give a short insight into several aspects of the conformal OPE.
The time-ordered product of two electromagnetic currents is expanded in the basis of so-called
conformal operators. The twist-two operators that are built of quark fields read at tree level
a V

(3/2)(1 + j )
+
O
3/2 D +
j a
(21)
=
)
(i
C
,
+
a OA
j
+ 5
2j (3/2 + j )
+
j
where, e.g., for three active quarks the non-singlet and singlet flavor matrices are

2 0
0
1 0 0
NS
= 0 1 0
and = 0 1 0 .
0 0 1
0 0 1
(22)
The gluon operators are expressed by the field strength tensor Ga = Ba Ba +gf abc Bb Bc
G
OV
G OA

j

(5/2)(j )
g
5/2 D +
j 1
(i+ ) G+
= j 2
Cj 1
G + .
i+
+
2 (3/2 + j )
(23)
The covariant derivatives either in the fundamental or adjoint representation contracted with the
light-like vector n are denoted as D + = D + D + and + = + + + for the total derivative.
253
Cj is the Gegenbauer polynomial of order j with index . The component is obtained by

contraction with the dual light-like vector n , i.e., n n = 1. The operators (21) and (23) are the
ground states of conformal multiplets (towers), which are labelled by the conformal spin j + 2.
Their Lorentz spin is J = j + 1 and higher states are obtained by multiplying the ground state
with total derivatives i+ . Hence, their spin J is increased by the number of total derivatives.
The advantage of such a basis is that in the hypothetical conformal limit conformal symmetry
guarantees that the renormalization procedure can be performed in such a way that only operators with the same conformal spin will mix. Beyond LO, this statement is not true in an arbitrary
scheme, and in particular not in the MS one. Suppose we rely on a scheme in which the conformal spin is a good quantum number. Then the only mixing problem occurs in the flavor singlet
sector, where quark and gluon operators with the same conformal spin will mix [114]. The 2 2
anomalous dimension matrix can be simply diagonalized by the transformation
+ I

OI
O
I
(24)
=
U
()
,
OI
G OI
j s
j
j
where the 2 2 matrices U Ij are expressed in terms of the anomalous dimensions. Here {+, }
labels the eigenfunctions of the renormalization group equation in this sector and I {V, A}.
Consequently, the renormalization group equation for all operators in question is then simply

a
d a I
(25)
O j () = a jI s () OI j ()
d
for a {NS, +, }. Moreover, if conformal symmetry is implemented in such a manifest way, it
can be employed to predict the Wilson coefficients of the OPE [110114,77,115].
With respect to the mixing properties of operators, it is convenient to perform also a decomposition in flavor non-singlet (NS) and singlet (S) CFFs and express the latter by the two
eigenmodes a {+, } of the evolution operator:
F = Q2NS NS F + Q2S S F,
F = F + G F = + F + F.
(26)
quarks6
The fractional charge factors are for three [four] active light

1 1
2 5
2
2
2
2
2
QNS =
,
QS = Q = QG = Q =
.
9 6
9 18
(27)
The general form of the short-distance OPE is governed by Lorentz invariance and is given as
sum over the irreducible representations of local operators. In our conformal operator basis this
expansion leads to the following leading twist approximation of the CFFs

Q2a a F,
F=
a
a=NS,

1 (1)j j 1 a CjI , Q2 /2 , s () a Fj , 2 , 2 .
(28)
j =0
2
Here, they are factorized in the perturbatively calculable Wilson coefficients a CjI (, Q
, s ())
2
and non-perturbative reduced matrix elements a Fj (, 2 , 2 ), defined below in Eqs. (31)
6 See Appendix A for more details. Basically, the squared fractional charges Q2 and Q2 are obtained from the
NS
decomposition Q2NS NS F + Q2 F = Q2u u F + Q2d d F + Q2s s F [+Q2c c F ].
254
and (32). These matrix elements carry conformal spin j + 2 and are nothing but the conformal moments of GPDs (see Appendix C). In Eq. (28) the sum runs over the conformal spin and
for I = V (A) only its odd (even) values contribute (in the following we will lighten our notation
by suppressing the superscript I {V, A}).
The = / dependence of the Wilson coefficients a Cj encodes information on how operators with the same spin, but a different number of total derivatives, or different conformal
spin projection, are arranged.7 This arrangement is indeed governed in the hypothetical conformal limit by conformal symmetry. For instance, to LO approximation this symmetry predicts
the form of the Wilson coefficients, that are expressed by hypergeometric functions, see e.g.,
Ref. [114], as

(1 + j )/2, (2 + j )/2 2
a tree
(29)
Cj () = a cjtree 2 F1
.
(5 + 2j )/2
The normalization a cjtree is not fixed. However, crucial for our following results is the fact that
it can be borrowed from deeply inelastic scattering. Namely, setting = 0, we realize that a cjtree
coincides with the forward Wilson coefficients
a tree
cj
a Cjtree ( = 0) = 1
for a = u, d, s, (c).
(30)
Such correspondence is also valid in all orders of perturbation theory, see next section and Appendix A.
The conformal moments Fj = {Hj , Ej , H j , E j } arise from the form factor decomposition of
the matrix elements of conformal operators (21) and (23). The reduced matrix elements
1
j
P+
P2 , S2 |a OjV |P1 , S1

a
2
2
a
2
2 i+
= U (P2 , S2 ) Hj , , + + Ej , ,
U (P1 , S1 ),
2M
1
j
P+
(31)
P2 , S2 |a OjA |P1 , S1

+ 5
= U (P2 , S2 ) a H j , 2 , 2 + 5 + a E j , 2 , 2
U (P1 , S1 ),
2M
(32)
of these operators are defined in such a way that in the forward limit we encounter the standard
Mellin moments of parton densities

a
(33)
qj 2 = lim a Hj , 2 , 2 and a qj 2 = lim a H j , 2 , 2 .
0
0
Note that the matrix elements of conformal operators are tensors of rank j + 1, contracted with
the light-like vector n. Hence the reduced ones (31) and (32) are polynomials in of order j + 1.
In particular, Hj and Ej are of order j + 1, whereas H j , E j and Hj + Ej are of order j , see, e.g.,
Ref. [116]. Moreover, time reversal invariance and hermiticity ensure that all these polynomials
are even under reflection . This then implies that for odd j conformal moments, such as
7 This can be easily seen from the Taylor expansion of the Wilson coefficients in powers of / . Taking the spin J
as summation label, the entire expansion is given in terms of inverse powers J . For each given value J there appear
J = {1, 2, 3, . . .} matrix elements n FJ n1 (, 2 , 2 ) with n = {0, 1, . . . , J 1}.
255
those appearing in the OPE of F V form factors, the combination Hj + Ej is actually of order
j 1.
3.2. Employing the dispersion relation
In this section, we combine the dispersion and the OPE techniques to restore both the imaginary and real part of the CFFs in the physical region. We also give an extended discussion of the
subtraction term.
3.2.1. MellinBarnes representation for CFFs
We express the Taylor expansion of the CFFs with respect to = 1/ in the convenient form
analogous to that introduced in Eq. (18)
a

1 (1)j j +1 a Fj , 2 , Q2 + a C , 2 , Q2 .
F , , 2 , Q2 =
(34)
j =0
Note that the CFFs and so the Taylor coefficients are physical quantities that are independent of
the renormalization/factorization scheme that is used for their evaluation.
First, we evaluate the Taylor coefficients a Fj , as defined in Eq. (19), from the conformal OPE
(COPE) result (28) taking / = = as an independent variable:
a
Fj

(n)
Cj +n , Q2 /2 , s () n a Fj +n 2 , 2 ,
n=0
even
(l)
Fj

1 dl
2
2
=
Fj , , ,
l! dl
=0
(35)
where j is (even) odd in the (axial-)vector case. Here we used the fact that all considered Fj are
even polynomials in of the order j or j + 1 and so Fj(l) = 0 for l > j or l > j + 1. Consequently,
the sum on the r.h.s. runs only over even n.
Next, we use the analytic properties of the CFFs which ensure that the Taylor coefficients (35)
are given by the Mellin moments of the imaginary part. Corresponding to Eq. (19), they then
read for general kinematics

d j m a F , , 2 , Q2 = a Fj , 2 , Q2 .
(36)
For technical reasons, we extend here and in the following the integration region to infinity using
m a F(, . . .) = 0 for > 1.
Now, we can combine Eqs. (35) and (36). For the DVCS kinematics ( = 1) we have with the
notation F (, 2 , Q2 ) F (, = 1, 2 , Q2 )

a (n)
2
2
a
d m F , , Q
Cj +n Q2 /2 , s () Fj +n 2 , 2 ,
j
(37)
n=0
even
where we denote the DVCS Wilson coefficients as Cj ( Q

, s ()) Cj ( = 1, Q
, s ()) and
2
2
2
make use of the photon virtuality Q2 2Q2 . It is worth mentioning that the Mellin moments
256
(a)
(b)
Fig. 2. (a) The integration contour in Eq. (38) parallel to the imaginary axis in the complex j -plane. Adding semicircles
results in an integration path which encloses the real axis. Contribution of dashed segments vanishes. (b) The integration
contour that will be used below in the numerical evaluation of CFFs.
(37) are directly measurable in single spin asymmetries. They contain the sum of all reduced
matrix elements built by operators with conformal spin equal or larger than j + 2 and so its spin
is larger than j .
In the next step we invert Eq. (37) following the standard mathematical procedure for inverse
Mellin transform. To do so we analytically continue the l.h.s. with respect to the conformal spin,
which is trivially done by considering j as complex valued. The Mellin moments are holomorphic functions in j , as long as the -integral containing the weight j (ln )N for N = 0, 1, 2, . . .
exists, and the condition e j c, with an appropriate constant c, is satisfied. The choice of c
should exclude the singularities in the complex j -plane arising from the endpoint at = 0. Notice also that in the limit e j the Mellin moments of a regular spectral function vanish,
except for a elastic pole contribution, i.e., a ( 1) term. The inverse Mellin transform reads

m F , , , Q
a
1
=
2i
c+i

dj j 1 a Fj , 2 , Q2 .
(38)
ci
Here all singularities of the integrand in the complex j -plane lie to the left of integration contour,
which is parallel to the imaginary axis, see Fig. 2(a). The rightmost lying singularity might be
related to the leading Regge trajectory, and so we have for c e j the condition8 (2 )1 < c.
There is one comment in order. The analytic continuation of the integrand in Eq. (38), known
for non-negative integer j , is not unique. According to Carlsons theorem [117,118], this ambiguity is resolved by specifying the behavior at j . Consequently, the analytic continuation
8 The leading Regge trajectory corresponds to (2 ) behavior of a F , i.e., j (2 ) behavior of the integrand in
(37), and the pole in j = (2 ) 1. Note that in perturbation theory the Wilson coefficients and anomalous dimensions
possess additional poles at negative integer j . Consequently, if a Fj is perturbatively expanded we have the condition
max((2 ) 1, 1) < c. Moreover, the flavor singlet anomalous dimension matrix in the parity even sector has poles
at j = 0. These poles could lie right to the pomeron pole j = P (2 ) 1, which appear in the non-perturbative ansatz
of GPD moments. Hence, we have in this sector the condition max(P (2 ) 1, 0) < c.
257
of the Mellin moments must be done in such a way that they vanish (or at least are bounded) in
the limit e j . Especially, one has to avoid an exponential growth, e.g., by a phase factor
eij . Obviously, then for > 1 the integrand in Eq. (38) does not contribute on the arc with
infinite radius that surrounds the first and fourth quadrants. Completing the integration path to a
semicircle, Cauchys theorem states that the imaginary part of the CFFs vanishes for > 1 as it
should be.
Finally, to restore the real part of the CFFs from the Mellin moments, we plug the imaginary part (38) into the dispersion relation (17), extending the integration interval to 0 ,
and perform the integration using the principal value prescription. The existence of the integral is ensured by appropriate bounds for e j . The lower one is max((2 ) 1, 1) < c or
max(P (2 ) 1, 0) < c, see footnote 8, and in addition for the upper one we have c < 1 (c < 0)
in the (axial-)vector case. The integration yields an additional weight factor in the MellinBarnes
integral (38) which is given by tan(j/2) and cot(j/2) for the vector and axial-vector case,
respectively. The whole amplitude for > 0 reads
a

F , , 2 , Q2
1
=
2i

j a
tan
dj j 1 i
Fj , 2 , Q2 + a C , 2 , Q2 ,
cot
2
c+i
(39)
ci
where the Mellin moments are defined by the series (35). The analogous expression for < 0
follows from the symmetry relations (15). We stress again that analyticity ties the real and imaginary part in a unique way. This is obvious in the axial-vector case, where the subtraction constant
is zero for symmetry reasons. The vector case will be discussed in the next section.
By comparing (39) with (34) one sees that with the help of dispersion relation technique the
analytic continuation of the CFFs in the unphysical region, given by the series (39), yields the
complex MellinBarnes integral (34). The corresponding Mellin moments a Fj are perturbatively
predicted by another series (35), which depends on the GPD moments a Fjn+n . This remaining
summation we postpone for later.
3.2.2. Discussing the subtraction constant
The evaluation of the subtraction constant C from the limit 0 deserves some additional
comments. The definition (20) of this constant suggests that it is entirely determined by shortdistance physics. The relevant local operator has dimension two (or spin zero) and simply does
not exist as a gauge invariant one.9 So on the first sight one would expect that, as in the forward
case, the subtraction constant is zero in the (generalized) Bjorken limit. Surprisingly, a more
careful look at the OPE result (28) shows that all (gauge invariant local) operators contribute,
because the highest possible order in = /, e.g., j + 1, of their conformal moments, cancels
the suppression factor j +1 in front of the Wilson coefficients. Hence, the subtraction constant
(j +1)
is in fact determined by light-cone physics and resuming all terms proportional to Fj
leads
to:
9 Such an operator might be interpreted as the expectation value of a gauge invariant non-local operator that carries
a (0)B a (0).
spin zero. In the light cone gauge such an operator is for instance given by the local two gluon operator g B
However, in a gauge invariant scheme its Wilson coefficient is zero.
258
C , , Q
2

(n)
Cn1 , Q2 /2 , s () n a Fn1 2 , 2 ,
n=2
even

a
C = 0, 2 , Q2 0.

(40)
(j +1)
Since Fj
appears only for j + 1 even, this constant vanishes for symmetry reasons in the
axial-vector caseas it mustsand is absent in the combination H + E. Moreover, it is predicted
to be zero if the virtualities of the photons are equal, i.e., for = 0. The expression (40) for the
subtraction constant has been also derived in the momentum fraction representation at tree level
[119], where it was expressed by a so-called D-term [120]. This term entirely lives in the central
GPD region and collects the highest possible order terms in . Hence, its conformal moments
(j +1)
provide us Fj
, e.g., for > 0,
3/2
dx j Cj

x
x
(j +1) j +1
.
D
= Fj
(41)
In the following we discuss the subtraction constant for the CFF E, rather than H. The reason
(j +1)
for doing so is that Ej
originally arises from a non-perturbatively induced helicity-flip, as we
will see in Section 5. We note that such non-perturbative effects are absent in linear combination
Hj + E j .
A subtraction constant is usually viewed as additional information that cannot be obtained
(j +1)
is the analytic continuation of Ej(n) , we should
from the imaginary part. However, if Ej
indeed be able to express the subtraction constant by the imaginary part of the CFF E.
Let us demonstrate such a possible cross talk between subtraction constant and imaginary
part in the most obvious manner. To do so we first make the mathematical assumption that
a E(, , . . .) vanishes for 0. In this case the subtraction in the dispersion relation (17) has
been overdone. Taking the 0 limit in the dispersion relation indeed leads to the desired
relation between subtraction constant and imaginary part:
a
2
C(, . . .) =
d 1 m a E(, , . . .).
(42)
Since the subtraction constant in the equal photon kinematics is vanishing, we derive the sum
rule
a
2
C( = 0, . . .) =
d 1 m a E(, = 0, . . .) = 0.
(43)
We remark, however, that this equality is not of physical interest, because Regge phenomenology
suggests that the small behavior of a E arises from the pomeron pole and so it will approximately grow with 1/ in the high-energy asymptotics. Hence, our mathematical assumption was
purely academic and the sum rule (43) does not apply.
Let us now consider a realistic behavior of the CFF E. Comparing the OPE prediction (40)
for the subtraction constant with the Mellin moments (35), one realizes that for j 1 the only
difference is the first term in the series (35). Thus, removing this term and taking the limit leads
259
to a formal definition of the subtraction constant

(0)
a
C , 2 , Q2 2 lim a Ej , 2 , Q2 a Cj , Q2 /2 , s () a Ej 2 , 2 . (44)
j 1
Here the separate terms on the r.h.s. can be singular and so the analytic continuation plays the
role of a regularization. We note that, corresponding to the high-energy ( 0) behavior, there
are also poles and perhaps branch cuts in the complex j -plane, which have to be surrounded
first, to arrive at e j = 1. This is certainly possible if the Mellin moments are meromorphic
functions. The situation is more tricky if a Ej contains singularities at j = 1 and the subtraction
procedure is maybe not uniquely defined. In particular, the appearance of a fixed singularity at
j = 1, e.g., a j,1 proportional contribution, might spoil the relation between the subtraction
constant and the imaginary part of the corresponding Compton form factor.
Let us explore the consequences in the case that both scheme dependent quantities a Cj =1
(0)
and a Ej =1 can be defined by analytic continuation. The GPD moment, appearing on the r.h.s.
of Eq. (44), might be expressed by the kinematical forward quantity
a
(0)
Ej

2 , 2
a E (
j
= 0, 2 , Q2 )
,
a C ( = 0, 2 , Q2 /2 , ())
j
s
see Eq. (35) with = 0. This provides an important relation, which formally reads

a
C , 2 , Q2

a C (, 2 , Q2 /2 , ())

j
s
a
2
2
.
=
0,

E
,
Q
= 2 lim a Ej , 2 , Q2 a
j
j 1
Cj ( = 0, 2 , Q2 /2 , s ())
(45)
Remarkably, only the ratio of Wilson coefficients appearing among physical quantities in Eq. (45)
is potentially plagued by ambiguities. Hence, we must conclude that in the j 1 limit this
ratio is scale and scheme independent and has the same value in any order of perturbation theory.
This value can thus be read off from the tree level coefficients (29):
2
2
2
j (, , Q / , s ())
j 1 a Cj ( = 0, 2 , Q2 /2 , s ())
lim
aC
= 1.
(46)
This formula has been checked to NLO accuracy. It turns out that radiative s corrections blow
up in the limit j 1; however, the leading singularities in j 1 are independent.
Relying on the validity of the prescription (45), we can calculate the subtraction constant from
the knowledge of the imaginary part. Taking into account Eq. (45), we formally write the desired
relation as

a
2
a
j
a
C(, . . .) =
d m E(, , . . .) E(, = 0, . . .)
,
lim
(47)
j 1
0
AC
where the integral is analytically regularized. That means that one first has to evaluate the integral for max((2 ) 1, 1) < e j and then perform the analytic continuation to j = 1.
Certainly, the prescription (47) relies on analyticity. If this property is absent, e.g., because of a
fixed pole or singularity at j = 1 with a -dependent residue, the subtraction constant a C is
only partially related to the imaginary part. We remark that in such a case the D-term for GPDs
260
[120], introduced as a separate contribution to cure the spectral representation [24,88], seems to
be justified on the first view. Otherwise the prescription of Ref. [121] for the restoration of the
highest possible order in is more appropriate.10
Let us remind that the existence of a fixed pole at angular momentum J = 0, i.e., j = 1
was already argued from Regge phenomenology inspired arguments at the end of the sixties
and it was proposed to access it by the uses of finite energy sum rules [123]. The analyses of
experimental measurements [124,125], although not fully conclusive, indicate a fixed pole at
J = 0. Interestingly, its residue might be related to the Thomson limit of the Compton amplitude
[126]. On the theoretical side studies in the partonic [127,128] or light-cone [129,130] approach
lead to a real term in the transversal part of the (forward) Compton amplitude that in the language
of Reggeization was interpreted as a J = 0 fixed pole. Nevertheless, via a subtracted sum rule
[131,132,127,128] it is related to the imaginary part of the Compton amplitude. This would imply
that there is no separate D-term contribution in the central GPD region.
3.3. Representation of CFFs in terms of conformal moments
In this section we derive a MellinBarnes representation alternative to Eq. (39), which expresses the CFFs in terms of conformal GPD moments rather than the usual Mellin moments.
The advantage of such a representation is that the CFFs are expanded in conformal partial waves,
which diagonalize the evolution operator to LO and facilitate direct application of conformal
symmetry beyond LO. In Section 3.3.1 we consider the well understood tree-level approximation to spell out the mathematical subtleties that appear in the mapping of conformal moments of
GPDs to the Mellin -moments (35) of CFFs. We propose then in Section 3.3.2 a Sommerfeld
Watson transformation which allows us to derive the desired MellinBarnes representation for
CFFs, which has been already derived previously by other methods [98,101]. Finally, we discuss
the MellinBarnes representation of conformal GPD moments themselves.
3.3.1. Lessons from a toy example
Let us now have a closer look at the series representation (37) of the Mellin moments a Fj ,
appearing in the MellinBarnes representation (39) for CFFs. To get some insight into its mathematical properties we consider the OPE at tree level. For simplicity, we will discard here the
2 -dependence, which appears only as a dummy variable, as well as the flavor index. The Wilson coefficients of the COPE are given in Eq. (29) and for DVCS kinematics, i.e., = 1, we
immediately have

Cjtree ( = 1) = 2 F1

2j +1 (5/2 + j )
(1 + j )/2, (2 + j )/2
.
1 =

(5 + 2j )/2
(3/2)(3 + j )
(48)
If we form partonic matrix elements of the conformal operators, e.g., for the CFF H, the con3/2
formal moments Hjtree are simply given in terms of Gegenbauer polynomials j Cj (1/).
10 The spectral representation is not uniquely defined [122]. Indeed, if one takes into account that both D-term and
spectral function are dependentthey arise from different projections of the same spectral function that enters in the
improved representationthe choice of a common spectral representation [24,88] plus D-term is equivalent to the
improved ones [121].

(n)tree
According to Eq. (35) the corresponding expression of Hj +n

Hjtree () =
261
reads

(3/2)(1 + j ) j 3/2 1
Cj
2j (3/2 + j )
(n)tree
Hj +n
(1)n/2 (3/2 + j + n/2)(1 + j + n)

.
2n (1 + j )(1 + n/2)(3/2 + j + n)
(49)
Consequently, the Mellin moments (35) are for our toy example given by the following series
Hjtree

n=0
even
(1)n/2 2j +1 (3 + 2j + 2n)(3/2 + j + n/2)

.
(1 + j + n)(2 + j + n)(1/2)(1 + j )(1 + n/2)
(50)
As one realizes from the asymptotics of the -functions, for large n with n j the summands
behave as nj 1/2 and for non-negative integer j this series must be resummed to arrive at a
meaningful result, which in our case is simply one. Hence, Hjtree = 1 and the inverse Mellin transform yields a -function. Note that the exponentially growing factor 2j in the Wilson coefficients
(48) is finally cancelled in the conformal Mellin moments (50) by an exponential suppression factor in the conformal GPD moments (49). This cancellation ensures that the imaginary part of the
CFF is vanishing for = > 1. We conclude that realistic conformal GPD moments for 1
are exponentially suppressed by a factor 2j .
The real part of the CFF can be obtained from the dispersion relation (17), while the subtraction constant follows from (45)

(0)tree
C tree = lim 2 Hjtree Cjtree (1)Hj

(51)
= 0.
j 1
This is in agreement with the fact that the highest order terms possible in are absent for our toy
(n)tree
conformal moments, i.e., Hn1 = 0 for n > 0, see Eq. (49). Consequently, the OPE prediction
(40) shows that the subtraction constant vanishes, too. Thus, we removed the ambiguity caused
by the subtracted dispersion relation, and arrive at the unique result:
1
d j m Htree ( ) = m Htree ( ) = (1 ),
0
e Htree ( ) =
2
.
1 2
(52)
Of course, this example is trivial and the findings coincide with the well-known answer,
1
H
tree
(, = 1) =
1

1
1
dx
H (x, ),
x i + x i
(53)
arising from the evaluation of the hand-bag diagram with a partonic GPD H (x, ) = (1 x).
If evolution is included, an explicit resummation of the series for the Mellin moments (37),
appearing in the MellinBarnes integral (39), cannot be achieved, since each term is labelled
by the conformal spin and will evolve differently. There are following possibilities to solve this
problem:
262
Numerical evaluation of the corresponding series.

Relying on an approximation of the series.
Arrangement of the integrand in the MellinBarnes integral (39) in such a way that the
integration, instead of Mellin moment, runs over the complex conformal spin
The third possibility will be worked out in the next section. In the following we shall study
whether an approximation of the series is appropriate. As we just saw, for complex j with e j
sufficiently small, the independent series (49) converges. If the first few terms are numerically
dominating, we might hope that these terms already induce a good approximation of the CFFs.
Certainly, if the dependence in the conformal moments is weak in the vicinity of = 0, i.e.,
2
2
if the Fj(n)
+n ( , ) are numerically small for larger n, the series might be approximated by a
finite sum. This should induce a good approximation for the CFFs for smaller values of .
Let us shortly demonstrate how this expansion works at tree level. We keep in the conformal
moments (49) only the leading term in 2 for 0, i.e., Hj(0)tree = 1 and all terms with n =
2, 4, . . . are neglected. Hence, the Mellin moments (50) read in this approximation

d j m Htree ( )
0
2j +1 (5/2 + j )
.
(3/2)(3 + j )
(54)
The inverse Mellin transform and the use of dispersion relation leads then for 0 to
3
m H
tree
2
( ) (2 )
,
2

e Htree ( ) 2 + O 3/2 .
(55)
The exponential factor 2j in the approximated moments (54) induces a wrong support. However,
we have still a useful approximation for small values of , where the deviation from the correct
result starts at order 3/2 . Naively, one might have expected that the accuracy is of order 2 . We
remark, however, that by taking into account the next order in the expansion (49), i.e., n = 2 as
well, the 3/2 proportional terms are annulled and the result is then valid up to order O( 5/2 ).
That the expansion of the conformal moments with respect to induces a systematic expansion
of the CFFs in powers of is perhaps not true in general.
3.3.2. Complex conformal partial wave expansion
We will now change the integration variable of the MellinBarnes integral (39) so that the
integration finally runs over the complex conformal spin and the non-perturbative input is given
in terms of conformal GPD moments. As mentioned above, in such a basis the evolution operator
is diagonal with respect to this quantum number. If this diagonality is not preserved in a given
scheme, we can always perform a scheme transformation so that the conformal GPD moments
evolve diagonally. Let us start with the SommerfeldWatson transformation of the series for the
Mellin moments Fj into a MellinBarnes integral. To do this we rewrite the series
a

Fj = 1, 2 , Q2 a Fj 2 , Q2

n=0
even

(n)
Cj +n Q2 /2 , s () a Fj +n 2 , 2
(56)
263
as an contour integral that includes the real positive axis and add two quarters of an infinite circle,
see Fig. 2, and arrive so at the MellinBarnes representation:
Fj , Q
2
d+i
i

4
dn

2 2

1
a
Cj +n Q2 /2 , s () a Fj(n)
+n , ,
sin(n/2)
(57)
di
(n)
(n)
where a Fj +n is the analytic continuation of (1)n/2 a Fj +n and the integration contour is chosen
(j +1)
so that all singularities lie to the left of it. Note that for j = 1 the expansion coefficient a Fj
a
is absent from this formula. Rather, it is contained in the subtraction constant C. Plugging this
representation into Eq. (39), we arrive after the shift of the integration variable j j n at a
MellinBarnes representation for the CFFs
a

F , 2 , Q2
1

2i
c+i
dj
j 1
tan
i+
cot

j
2

a

Cj Q2 /2 , s () a Fj , 2 , 2 .
(58)
ci
Here the conformal moments Fj are now written as a MellinBarnes integral
Fj = , ,
2
i
=
4
d+i
di
a F (n) (2 , 2 )
1 eij
j
n
dn
( i)
,
i(j
n)
sin(n/2)
1e
(59)
valid for 0 1. The (lower) upper sign corresponds to the (axial-)vector case. We included
in this definition a i prescription, which ensures that
1 eij
1 eij
n
(
i)
=
en(ln i)
1 ei(j n)
1 ei(j n)
(60)
vanishes also for n i in the complex n-plane rather than generate an exponential
j -dependent phase factor. We remark that the exponential suppression in the integral (59), arising
from 1/ sin(n/2) in the limit n i, is in general needed to cancel an exponential growing
(n)
of a Fj .
The MellinBarnes representation (58) for the CFFs, which we will from now on use in all our
numerical analysis, has been already derived in two alternative ways in Refs. [98,101]. As a gift
of the present approach we found for the conformal moments (31) and (32) the MellinBarnes
integral representation (59), which gives the analytic continuation of the conformal moments.
This issue has not been completely resolved before.
We will discuss briefly the representation (59) within our partonic toy GPD to shed light on
some tricky points. It can be shown that for the conformal moments of GPD H (x, ) = (1 x)
the integral (59) exists for 0 < < 1, where the integrand can be read off from the second
expression in Eq. (49). After completing the integration path, it reduces for complex valued j to
the function Hjtree in Eq. (49) and an addenda that arises from the additional phase factors in the
definition (59) of conformal moments. For 0 < 1 we might express our improved findings in
terms of hypergeometric functions
264
Hjtree () =

(3/2)(3 + j ) j
j, j + 3 1
2 F1
2
2
21+j (3/2 + j )

4j 1 3+2j
(1 + j )(3 + j )
3/2 + j/2, 2 + j/2 2
F
i0
.
2 1

5/2 + j
i + tan(j/2) (3/2 + j )(5/2 + j )
(61)
Hence for positive odd integer j this formula reduces to the Gegenbauer polynomials (times the
conventional normalization factor). The addenda is proportional to 1/(i + tan(j/2)), where this
factor cancels the corresponding one in the conformal partial wave expansion (58). We can again
close the integration path by an infinite arc so that the first and forth quadrant is encircled. It can
be shown that the infinite arc does not contribute. Since the integrand stemming from the addenda
is a holomorphic function in j , we conclude that the addenda finally vanishes for 0 < 1.
Consequently, we have found that the appropriate analytic continuation of the conformal GPD
moments is given by the first term on the r.h.s. of Eq. (61), i.e., the analytic continuation of
Gegenbauer polynomials is given by the replacement

(3/2)(1 + j ) j 3/2 1
(3/2)(3 + j ) j
j, j + 3 1
C
F
2 1
j
2
2
2j (3/2 + j )
21+j (3/2 + j )
for 0.
(62)
This result allows also for the extension of the region to 1. Of course, symmetry under
reflection allows for negative values of by replacement .
3.3.3. MellinBarnes representation of conformal GPD moments
First, notice that a Fj(n) from the MellinBarnes representation (59) is not an arbitrary function;
rather it must be guaranteed that a Fj
(i) is holomorphic in the first and fourth quadrant of the complex j -plane
(ii) reduces for non-negative integers j to a polynomial of order j or j + 1
(iii) possesses for j and 1 an exponential suppression factor 2j .
(n)
The requirement (i) is satisfied when it holds for a Fj . To show that also the second requirement
is fulfilled, let us suppose that the analytic continuation of a Fj(n) is defined in such a way that it
does not grow exponentially for n with arg(n) /2 and fixed j . Moreover, for simplicity
(n)
let us assume that a Fj is a meromorphic function with respect to n. For 0 < < 1 we can
then close the integration contour in Eq. (59) so that the first and forth quadrant are encircled.
Cauchys theorem states that a Fj is given as a sum of the residues in these two quadrants

2 2
n
1 eij
a
a (n)

.
Fj , 2 , 2 =
Res
,
(63)
F
2
1 ei(j n) sin(n/2) j
poles
This result is in general a series, defining for fixed a holomorphic function with respect to j .
For non-negative integer j , any contribution that does not arise from the poles at n = 0, 2, 4, . . .
will drop out, due to the factor 1 eij . Consequently, in this case only the poles on the real
(n)
n-axis contribute and with the requirement a Fj = 0 for n j + 2 we get the polynomial
a
j +p

n a (n) 2 2
Fj , 2 , 2 =
Fj ,
n=0
even
for j = {1, 3, 5, . . .} or j = {0, 2, 4, . . .},
(64)
265
where p = {0, 1}. Note that for p = 1 the highest possible term of order j +1 is now included.
Moreover, it is required that for = > 1 the following expression, which appears as integrand
in the MellinBarnes integral (58),

1 j 1
j a
tan
lim
(65)
Fj , 2 , 2 lnN j,
i+
cot
j j 2
2
vanishes in the limit j for arg(j ) /2. Here we took into
account that the Wilson coN
j/
j in the asymptotics we are
efficients in fixed order of perturbation
theory
behave
as
ln
considering. More precisely, 1/ j arises from the factor (j + 5/2)/ (3 + j ) in the normalization of the Wilson coefficients, see Eq. (48), and the logarithmical growing is induced by
radiative corrections. The condition (65) is necessary to ensure that the support of the imaginary
part of the CFFs is restricted to | | 1. Namely, for > 1 we can close the integration path to
form a contour that encircles the positive real axis. Obviously, the imaginary part drops out and
only the poles of the trigonometric functions contribute, yielding the series given by the COPE
(28). For = we precisely recover the subtraction constant a C, see Eq. (40), which
shows that it is already contained in the final representation (58) of the CFFs in the conformal
moments (59).
Moreover, we can easily generalize the formula (59) for the computation of the conformal moments for a given GPD, which has definite symmetry with respect to the momentum fraction x.
In this case the analytic continuation of the conformal moments are
a
Fj , ,
i(1 + j )
=
4(3/2 + j )
1
AC
d+i
dn
di
dx x
j n a
1 eij ( i)n 2n (3/2 + j n/2)

1 ei(j n) sin(n/2) (1 + j n)(1 + n/2)

F x, , ,
(66)
Here the Mellin moments in the second line appear in the form factor decomposition of the matrix
elements of local operators that are built exclusively out of covariant derivatives. Precisely these
matrix elements are measured on the lattice [3540]. It is understood that e(j n) is restricted
to such values that the integral exists. We note that the polynomiality of these moments is ensured
by the symmetry we assumed. So for instance for an (anti-)symmetric GPD we have for even
(odd) values of n
1
2
1
dx x
1
na
F x, , ,
1
=
n+p
a 2 2 i

dx x n a F x, , 2 , 2 =
Fni , ,
(67)
i=0
even
where p takes the values 0 and 1 for even and odd n, respectively. However, irrespective of
the symmetry p = 0 holds true for the combination H + E, H , and E.

4. DVCS amplitude up to next-to-next-to-leading order
In Section 4.1 we first discuss the predictive power of conformal symmetry in a conformally invariant theory for general kinematics. Here we rely on a special subtraction scheme,
the so-called conformal one, in which the renormalized operators are covariant under collinear
266
conformal transformations. This allows us to predict the functional form of the Wilson coefficients up to a normalization factor. This factor is simply the Wilson coefficient that appears in
DIS. To reduce the massless QCD to a conformally invariant theory, we intermediately assume
that the QCD has a non-trivial fixed-point so that the trace anomaly vanishes. Then we move to
the real world and discuss the inclusion of the running coupling. In Section 4.2 we present the
results for the DVCS kinematics. The recent progress in the evaluation of radiative corrections
in DIS allows us to present the CFFs to NNLO accuracy. However, the trace anomaly induces a
mixing of conformal GPD moments due to the evolution. Also note that the forward anomalous
dimensions of parity odd conformal GPD moments are mostly unknown beyond NLO. In Section 4.3 we present the NLO corrections to the twist-two CFFs in the MS scheme and discuss the
inclusion of mixing effects in the MellinBarnes representation.
4.1. Conformal OPE prediction and beyond
In the hypothetical conformal limit, i.e., if the function of QCD has a non-trivial fixed point
s = 0, the / dependence of the Wilson coefficients is predicted by conformal symmetry
[110113]. For all the cases we are considering, Wilson coefficients a C I , with a = {NS, +, }
and I = {V, A}, have the same general form. Thus, when there is no possibility of confusion, we
will simplify our notation by suppressing both of these superscripts. The COPE prediction for
general kinematics reads [114]

Cj /, Q2 /2 , s
2 j (s )/2

(2 + 2j + j (s ))/4, (4 + 2j + j (s ))/4 2
= c j s 2 F 1
,
2
(5 + 2j + j (s ))/2
Q2
(68)
where the normalization cj remains so far unknown. In the kinematical forward limit ( = 0) the
Wilson coefficients coincide with those known from DIS

2 j (s )/2
lim Cj /, Q2 /2 , s = cj s
(69)
,
0
Q2
see Appendix A. Neither Wilson coefficients (68) nor conformal GPD moments (31), (32) mix
under collinear conformal transformations. This implies that they evolve autonomously with respect to a scale change,

d
Fj . . . , 2 = j s Fj . . . , 2 ,
d

d
Cj . . . , Q2 /2 , s = Cj . . . , Q2 /2 , s j s .
d
(70)
(71)
Here the anomalous dimensions are the same as in DIS. We also realize that the conformal spin
j + 2 is a good quantum number.
Unfortunately, conformal symmetry is not manifest in a general factorization scheme, and
in particular not in the MS scheme. What is required is a special scheme, which ensures that
the renormalized conformal operators form an irreducible representation of the collinear conformal group SO(2, 1) [114]. The breaking of the covariant behavior with respect to dilatation,
e.g., shows up in the mixing of operators expectation values and of Wilson coefficients under
267
renormalization:

d
j k s () j k Fk . . . , 2 ,
F j . . . , 2 =
d
j
(72)
k=0

kj
d
2
2
2
2
Cj . . . , Q / , s () =
Ck . . . , Q / , s () kj s ()
,
d
(73)
k=j
respectively.11 This mixing arises in NLO and is induced by the breaking of special conformal
symmetry in LO. The mixing matrix, i.e., j k for j > k, contains besides a proportional term
also one that does not vanish in the hypothetical conformal limit. Knowing the form of the special
conformal symmetry breaking in the MS scheme to LO, it was shown that, after rotation, the
diagrammatic evaluation of NLO corrections to (flavor non-singlet) anomalous dimensions [133
135] and Wilson coefficients [7880] in the MS scheme coincide, for = 0, with the conformal
symmetry predictions [136,114,77]. A formal proof about the restoration of conformal symmetry,
valid to all orders of perturbation theory, has been given in Ref. [114]. Hence, we have no doubt
that conformal symmetry can be effectively used for the evaluation of higher order perturbative
corrections [137,101,102]. The interested reader can find a comprehensive review of the uses of
conformal symmetry in QCD in Ref. [115].
Let us suppose that we employ a scheme in which conformal symmetry is manifest in the
hypothetical conformal limit. Such a scheme we call a conformal subtraction (CS) scheme. We
discuss now the structure of the conformal OPE beyond this limit, i.e., the implementation of
the running coupling constant in the Wilson coefficients (68) and the evolution equation. In
the simplest case, i.e., forward kinematics, the conformal representation is trivial, as there are
only operators without total derivatives left, so we can restore the Wilson coefficients of the full
perturbative theory from the result (69) of the conformal limit. Namely, renormalization group invariance of the CFFs allows us to revive the -proportional term by expanding the exponential12

2 j (s )/2

d

c j s
cj s (Q) exp
j s ( )
(74)

Q2
Q
ln(Q2 /2 ).
in terms of
Of course, also the evolution of Mellin moments of parton densities is
governed by a diagonal evolution equation, i.e., by Eq. (72) with = 0. Finally, we are using the
normalization condition
cj (s ) = cjMS (s )
(75)
and thus we end up with the standard result for the Wilson coefficients of DIS, evaluated in the
MS scheme.
For general kinematics the proportional term is not automatically fixed, so different prescriptions might be used. For a discussion of this issue see Ref. [137]. We will point out here two
appealing possibilities.
11 Both these equations together ensure that the CFF j 1 C (. . . , Q2 /2 , ())F (. . . , ) is a renormalization
s
j
j
j
group invariant quantity.

12 We consider c and as independent quantities and so their perturbative expansion in the fixed-point and full
j
j
theory looks the same. Certainly, both of them contain to two-loop accuracy 0 proportional terms and the anomalous
dimensions to three-loop accuracy might be also rewritten in terms of the function expanded to two-loop accuracy.
268
Let us first assume that the conformal symmetry breaking can be entirely incorporated in the
running of the coupling so that the evolution equation for conformal GPD moments is diagonal:

d
Fj , 2 , 2 = j s () Fj , 2 , 2 .
d
(76)
Then also the Wilson coefficients in the full theory are simply obtained by replacing the expression for the resummed evolution logs. However, it is likely that the trace anomaly generates a
non-vanishing proportional addenda, which will appear at two-loop level. If it does, renormalization group invariance fixes the scale dependence and so the Wilson coefficients in such a
scheme read for the full theory

Cjfull /, Q2 /2 , s ()

= Cj /, 1, s (Q) + s (Q) Cj /, 1, s (Q)
g

d

exp
(77)
j s ( ) ,

Q
where Cj (/, 1, s ) is the COPE prediction (68). In the forward limit we again require that the
Wilson coefficients coincide with the DIS ones, evaluated in the MS. Thus, we have the condition
that the proportional addenda vanishes, i.e.,

lim Cj /, 1, s (Q) = 0.
(78)
The disadvantage of such a conformal scheme is that a consistent determination of the proportional addenda Cj requires a diagrammatical evaluation at NNLO, e.g., in the MS scheme.
Both the knowledge of the 0 -proportional terms in the two-loop corrections to the Wilson coefficients and the off-diagonal three-loop corrections to the anomalous dimensions are required.
While the former are known in the MS scheme [138,139], the calculation of the latter, although
reducible to a two-loop problem by means of conformal constraints, still requires a great effort.
Fortunately, to NNLO only the quark sector suffers from this uncertainty. The best what we can
do to NNLO accuracy is to minimize the effects of this lack of knowledge. To do so, we will
perform a transformation into a new scheme, called CS, where the normalization condition reads

Cj /, 1, s (Q) = 0,
(79)
while the mixing term is now shifted to the evolution equation. This mixing term will be suppressed at the input scale by an appropriate initial condition and we might expect that the
evolution effects are small, see discussion in Section 6.3. Hence the Wilson coefficients read
now

Cj /, Q2 /2 , s ()

kj

d

=
Ck /, 1, s (Q) P exp
,
kj s ( )
j s ( ) kj +

g
k=j
Q
(80)
269
where Ck (/, 1, s (Q)) is given in Eq. (68), kj is determined by the off-diagonal part of the
anomalous dimension matrix, and P denotes the path ordering operation.13 The renormalization
group equation for the conformal moments reads

d
F j . . . , 2
d
j 2

(s ()) j k

j k s () Fk . . . , 2 .
= j s () Fj . . . , 2
g()
(81)
k=0
Since this mixing term appears for the first time at NNLO accuracy, and is suppressed at the input
scale by the initial condition, we might expect that its effects will be small.
Formulae (80) and (81) are our main result that can be used for the numerical investigation of
radiative corrections to DVCS up to NNLO.
4.2. Perturbative expansion of Compton form factors
For DVCS kinematics the general expression for the Wilson coefficients (80) simplifies considerably. Since to twist-two accuracy = / = 1 is valid, the argument of hypergeometric
functions is unity and they simplify to products of functions. Having also in mind that in
DVCS the photon virtuality Q2 2Q2 is considered as the relevant scale, the COPE prediction
(68) together with Eq. (80) leads to the following expression for the DVCS Wilson coefficients

Cj Q2 /2 , s ()

Cj = 1, Q2 / 22 , s ()

2j +1+j (s (Q)) ( 52 + j + 12 j (s (Q)))

d

= cj s (Q)
(82)
j s ( )
exp

( 32 )(3 + j + 12 j (s (Q)))
Q
at the considered NNLO accuracy. Note that the off-diagonal part of the evolution operator does
not appear in this approximation, and that we have slightly changed the normalization condition
for = Q. In fact to arrive at the above result, Eq. (80) has to be multiplied by a factor
Q

j (s (Q))

d

2
exp
j s ( )

Q/ 2
=1+
s2 (Q)
(2)2

0 (0) 2
j ln (2) + O s3 ,
8
(83)
13 The path ordering operation defined by

P exp
Q

j

d
d
d
d

j k s ( ) = 1 +
j k s ( ) +
j m s ( ) mk s ( ) +

m=k
enables the compact representation of the solution of the non-diagonal renormalization group equations. Obviously, in
diagonal case, i.e. for j k = j k j , path ordering converts the evolution operator in a simple exponential function.
270
and such terms can be compensated by a change of the normalization condition (79). Here and in
the following the first expansion coefficient of (g)/g = (s /4)0 + O(s2 ) is 0 = (2/3)nf
11, where nf is the number of active quarks.
We now use perturbation theory to determine Cj (Q2 /2 , s ()) to NNLO accuracy. In contrast to Fj (. . . , 2 ), it is customary to express Cj (Q2 /2 , s ()) completely expanded in s (),
i.e., without resummation of leading logs from to Q. This actually corresponds to the result that
one would obtain from a diagrammatical calculation, without using the renormalization group
equation. Obviously, since the leading logarithms in Fj will be resumed and in Cj not, one will
end up with a residual dependence on the factorization scale . This scale might also be used
as the relevant scale in the running coupling constant. However, for generality, another scale r
is better suited for the expansion parameter s (r ). The truncation of perturbation theory then
implies that our results will depend also on this renormalization scale r .
The perturbative expansion of the DVCS Wilson coefficients (82) in terms of s (r ) is done
in a straightforward manner by means of
s () = s (r ) + 0
s2 (r ) 2
ln 2
4
r
(84)
and consequently we can expand the exponential factor

d
exp
j s ( )

Q
2
Q2

j (s (r ))/2
3
2 (r ) 0 (0) Q2 2 Q2
+
O
j ln 2 ln
1 s 2
s .
(2) 8
4r
Furthermore, we use the expansion

s (m) (Q2 /2 )
2j +1+j ( 52 + j + 12 j ) 2 j /2 2j +1 (5/2 + j )
j
=
[j ]m ,
(3/2)(3 + j )
2m m!
Q2
( 32 )(3 + j + 12 j )
(85)
(86)
m=0
where the so-called shift coefficients, which also include the factorization logs, are defined as

2

d m 2 4 (3 + j )(5/2 + j + )
(m) Q
sj
(87)
.
=
d m Q2
(5/2 + j )(3 + j + ) =0
2
The first two shift coefficients read in terms of harmonic sums as

Q2
sj(1) Q2 /2 = S1 (j + 3/2) S1 (j + 2) + 2 ln(2) ln 2 ,
(1) 2 2 2
(2) 2
2
S2 (j + 3/2) + S2 (j + 2),
sj Q / = sj Q /
(88a)
(88b)
where the analytical continuation of these sums are defined by

d
d2
2
(89)
ln (z + 1) + E and S2 (z) = 2 ln (z + 1) +
dz
6
dz
with E = 0.5772 . . . being the Euler constant. Taking into account the perturbative expansion of
anomalous dimensions and DIS Wilson coefficients, written as
S1 (z) =
j (s ) =

s (0)
s2
s3
(1)
(2)
+
j + O s4 ,
j +
j
2
3
2
(2)
(2)
271

s (1)
s2 (2)
cj + O s3 ,
cj +
2
2
(2)
the DVCS Wilson coefficients (82) take the form

2j +1 (j + 5/2) (0) s (r ) (1) 2 2
c +
Cj Q /
Cj =
(3/2)(j + 3) j
2

3
s2 (r ) (2) 2 2 2 2
C Q / , Q /r + O s ,
+
(2)2 j
where
cj (s ) = cj(0) +
(90)
(91a)
(1)
(1)
Cj
(1)
= cj
sj (Q2 /2 )
(0) (0)
cj j ,
2
(1)
(2)
sj (Q2 /2 ) (0) (0) 2
sj (Q2 /2 ) (0) (1)
(2)
(2)
(1) (0)
cj j + cj j +
cj j
Cj = cj +
2
8

Q2 1 (0) (0)
Q2
0
(1)
+
Cj Q2 /2 ln 2 + cj j ln2 2 .
2
4
r
(91b)
(91c)
Notice that the renormalization group logs ln(Q2 /2r ), caused by the running of s , and propor(2)
tional to 0 , appear for the first time in the NNLO correction Cj .
4.2.1. Flavor non-singlet Wilson coefficients
The flavor non-singlet Wilson coefficients have the form as indicated in Eqs. (91a)(91c). We
only have to decorate them and the anomalous dimensions with the corresponding superscripts.
It has been already taken into account that the DIS Wilson coefficients are normalized as
NS I(0)
cj
= 1.
(92)
The radiative corrections to NLO involve the Wilson coefficients of the DIS unpolarized structure
function F1 ,

3
9
5 2S1 (j )
NS V(1)
2
cj = CF S1 (1 + j ) + S1 (j + 2) +
S2 (j + 1) ,
(93)
2
2 2(j + 1)(j + 2)
for the vector case and those of the polarized structure function g1 ,

3
9
3 2S1 (j )
NS A(1)
2
cj = CF S1 (1 + j ) + S1 (j + 2) +
S2 (j + 1) ,
(94)
2
2 2(j + 1)(j + 2)
for the axial-vector one. The LO anomalous dimensions read in both cases

2
NS I(0)
(95)
j = CF 3 +
4S1 (j + 1) ,
(j + 1)(j + 2)
where CF = (Nc2 1)/(2Nc ) with Nc being the number of colors.
The expressions for two-loop quantities are lengthy. The non-singlet anomalous dimensions
(1)
j for the vector and axial-vector case are given, e.g., in Ref. [140], by analytic continuation of
(2)
j -odd and j -even expressions, respectively. The NNLO Wilson coefficients cj can be read off
for the vector and axial-vector cases from Refs. [141] and [142], respectively. Implementation of
evolution to NNLO accuracy requires also the anomalous dimensions to three-loop order [143].
Instead of the exact expressions one might alternatively use an analytic approximation, given for
the vector quantities in Refs. [144,143].
272
4.2.2. Flavor singlet Wilson coefficients

So far the {+, } basis, which resolves the mixing problem in the flavor singlet sector, has
been used for presenting the conformal predictions. However, the Wilson coefficients are in the
literature usually given in the {Q , G} basis. For convenience, we might express the flavor
singlet results in the basis of quark and gluon degrees of freedom. We write the Wilson coefficients as a two-dimensional row vector

C Ij Q2 /2 , s () = CjI , G CjI Q2 /2 , s () for I = {V, A}.
(96)
In the following we drop the superscript I. The rotation between the two bases is given by the
transformation matrix U j , defined in Eq. (24). Hence, for the Wilson coefficients we have

C j Q2 /2 , s () = + Cj , Cj Q2 /2 , s () U j (, Q),
(97)
where the {+, } entries are given in Eq. (82). Using the property (B.4) of the transformation
matrix within 0 = Q, derived in Appendix B, allows us to perform the transformation at the
normalization point = Q and afterwards restore the scale dependence by an backward evolution:

C j Q2 /2 , s () = + Cj , Cj Q2 /2 = 1, s (Q) U j (Q, Q)Ej1 (, Q).
(98)
Here Ej1 is the inverse of the evolution operator in the {, G} basis, defined below in Eq. (120).
Since U j (Q, Q) rotates the anomalous dimensions [depending on s (Q)], we can express for
= Q the flavor singlet contributions in the form

C j Q2 /2 = 1, s (Q)
(m)
2
2
2j +1 (5/2 + j ) sj (Q / = 1)
cj [ j ]m ,
=
(3/2)(3 + j )
2m m!
cj =

cj , G cj ,
(99)
m=0
where the anomalous dimension matrix is defined as

(0) G (0)
2
j
s j
j =
(0)
(0) + O s .
G
GG
2
j
j
(100)
This matrix valued expansion, although resulting from somewhat more involved derivation, is in
complete analogy to the flavor nonsinglet result (86).
Acting with the evolution operator Ej1 (, Q) on Eq. (99) removes the constraint Q2 /2 = 1
of the shift functions argument. As in the preceding section, we consequently expand with respect
to s (r ) and finally write the Wilson coefficients (96) in complete analogy to Eq. (91a) as

2j +1 (j + 5/2) (0) s (r ) (1) 2 2
c +
C j Q /
Cj =
(3/2)(j + 3) j
2

2 2 2 2
3
2 (r )
+ s 2 C (2)
(101a)
Q
+
O
/
,
Q
/
r
s ,
j
(2)
where

(1)
(1)
C j Q2 /2 = cj
(1)
sj (Q2 /2 )
2
(0) (0)
cj j ,
(101b)

(1)
273
(2)
sj (Q2 /2 ) (0) (0) 2

sj (Q2 /2 ) (0) (1)

(2)
(1) (0)
Q2 /2 = cj +
cj j + cj j +
cj j
2
8

0 (1) 2 2 Q2 1 (0) (0) 2 Q2
+
(101c)
C j Q / ln 2 + cj j ln 2 .
2
4
r
To LO the DIS Wilson coefficients are normalized as

(2)
Cj
V(0)
cj
A(0)
= cj
= (1, 0).
(102)
In NLO the quark Wilson coefficients correspond to the non-singlet ones given in Eqs. (93) and
(94),
V(1)
cj
A(1)
cj
V(1)
(103)
A(1)
(104)
= NS cj
= NS cj
while the gluon ones read

(4 + 3j + j 2 )S1 (j ) + 2 + 3j + j 2
(105)
,
(1 + j )(2 + j )(3 + j )

j
G A(1)
cj = nf
(106)
1 + S1 (j ) .
(1 + j )(2 + j )
The entries of the anomalous dimension matrices read to LO in the vector case:
4 + 3j + j 2
G V(0)
j
= 4nf TF
,
(107)
(j + 1)(j + 2)(j + 3)
4 + 3j + j 2
G V(0)
j
= 2CF
,
(108)
j (j + 1)(j + 2)

12
4
GG V(0)
j
= CA
+
4S1 (j + 1) + 0 ,
(109)
(j + 1)(j + 2) j (j + 3)
and in the axial-vector case:
j
G A(0)
j
= 4nf TF
,
(110)
(j + 1)(j + 2)
(j + 3)
G A(0)
(111)
j
= 2CF
,
(j + 1)(j + 2)

8
GG A(0)
j
= CA
4S1 (j + 1) + 0 ,
(112)
(j + 1)(j + 2)
where CA = Nc and TF = 1/2. At this order the anomalous dimensions in the quarkquark
channels are identical to the flavor non-singlet ones
G V(1)
cj
= nf
jV(0) = jA(0) = NS j(0) ,
(113)
given in Eq. (95). The NNLO DIS Wilson coefficients for the structure functions F1 and g1
are given in Refs. [141,142]. The NLO singlet anomalous dimensions for the vector and axialvector cases can be found in Refs. [140,145] and [146,147], respectively. To treat the evolution
to NNLO approximation we need also the NNLO flavor singlet anomalous dimension matrix.
It has been calculated for the vector case in Ref. [148]. We remark that all these quantities are
expressed in terms of rational functions and harmonic sums. Several numerical routines for them
are available, e.g., in Refs. [149152]. Here again one might rely on analytic approximations for
the quantities in question, see Refs. [153,148].
274
4.2.3. Expansion of the evolution operator

The evolution of the flavor non-singlet (integer) conformal moments in this CS scheme is
governed by

4
d
s () (0) s2 () (1) s3 () (2)
2
2
+
O
Fj , 2 , 2 =
j +
s Fj , ,
j
j
2
3
d
2
(2)
(2)
j 2

0 s3 () CS
j k + O(s ) Fk , 2 , 2 .
3
2 (2)
(114)
k=0
The solution of the renormalization group equation is given by the path-ordered exponential,
appearing in Eq. (80). Unfortunately, the mixing matrix CS
j k is not completely known, and so
we can here only deal with the solution for j k = 0:

Fj , ,
= Ej (, 0 )Fj ,
, 20

d

where Ej (, 0 ) = exp
j ( ) .

0
(115)
In the numerical analysis we will resum only the leading logarithms and expand the non-leading
ones. The result, see Ref. [137], reads

s () (1)
2 () (2)
Ej (, 0 ) = 1 +
Aj (, 0 ) + s 2 Aj (, 0 ) + O s3
2
(2)
(0)
s ()
s (0 )
j
,
(116)
where

(0)
(1)
j
s (0 )
1 j
(1)
Aj (, 0 ) = 1
,
s ()
20 0
0

(0)
(1)
(2)
j
2
1 (1)
s2 (0 ) 12 2 0 j
1 j
(2)
+
Aj (, 0 ) = Aj (, 0 ) 1 2
.
2
0
40 0
20
s ()
802
(117)
The expansion coefficients of the function are

2 ()
3 ()
s ()
=
0 + s 2 1 + s 3 2 + O s4 ,
g
4
(4)
(4)
2
38
325 2 5033
2857
0 = nf 11,
1 = nf 102,
2 =
nf +
nf
.
3
3
54
18
2
(118)
The flavor singlet case can be treated in the diagonal {+, } basis in the analogous way as the
non-singlet one. Here we present the solution in the {Q, G} basis:
Fj
GF
j

, 2 , 2 = Ej (, 0 )
Fj
GF
j

, 2 , 20 .
(119)
275
We will expand the evolution operator14

d

Ej (, 0 ) = P exp
s ( )
j
(120)
s2 ,
up to order
and while leading logs remain resummed the non-leading ones are expanded.
To have a condensed notation, we perturbatively expand the result in terms of the leading order
{+, } modes:

3
s () ab (1)
s2 () ab (2)
a
Ej (, 0 ) =
Aj (, 0 ) +
Aj (, 0 ) + O s
ab P j +
2
(2)2
a,b=
s ()
s (0 )
j
0
(121)
Here the projectors on the {+, } modes are
Pj =
(0)

1
j j 1 ,
j j
(122)
where the eigenvalues of the LO anomalous dimension matrix (i.e., j from Eq. (B.2)) are

G (0) G (0)

4

1
j
j
(0)
(0)
(0)
(0)
j =
j + GG j j GG j 1 +
, (123)
(0)
(0)
2
( j GG j )2
(0)
a straightforward calculation leads to the matrix valued coefficients

1 (0)
(1)
ab (1)
Aj = ab Rj (, 0 |1)a P j
j j bP j ,
(124)
20

c b

1
s (0 ) 0 + j j a
ab (2)
ab
ac
0
Aj =
Rj (, 0 |2) Rj (, 0 |1)
Pj
c b
()
0
j
j
s
c=

1 (0)
1 (0)
(1) c
(1) b
j
Pj
j
Pj
20 j
20 j
2

2 0 (0)
1 (1)
(2) b
ab Rj (, 0 |2)a P j 1
Pj,
(125)
j
j
20 j
402
where the dependence is accumulated in the functions
fined by
ab R
j (, 0 |n)
n0 + j

0
1
s (0 )
Rj k (, 0 |n) =
1
a
b
n0 + j k
s ()
a
ab
ab Rjj (, 0 |n) de-

.
(126)
The expansion of the evolution operator (116) or (121) will then be consistently combined with
the Wilson coefficients (91) or (101), respectively, see for instance Ref. [137].
14 Here the path ordering P is related to the non-diagonality of the anomalous matrix in {Q, G} basis.
276
4.3. MS results up to NLO order

The DVCS radiative corrections in the MS scheme are known only to NLO. The NLO corrections to the hard-scattering amplitude have been obtained by rotation from the conformal
prediction [114,77] and agree with the diagrammatical evaluation of Refs. [7880]. Employing
Eq. (C.6), the corresponding Wilson coefficients can be obtained by determining the conformal
moments of the hard-scattering amplitude, e.g., summarized in Ref. [154].
The integrals which are needed for the quark part are given in Appendix C of Ref. [137] for
integer conformal spin (see also the last paragraph in Appendix C.1). The analytical continuation
is straightforward and so in the MS scheme we have

V(1)
Q/2
Cj

9
5 4S1 (j + 1)
1
2
= CF 2S1 (1 + j ) +
+
2 2(j + 1)(j + 2) (j + 1)2 (j + 2)2
+
(0)
j
A(1)
ln

2
2
,
Q2
(127)
Q/

9
3 4S1 (j + 1)
1
+
= CF 2S12 (1 + j ) +
2 2(j + 1)(j + 2) (j + 1)2 (j + 2)2
Cj
(0)
j
ln
2
,
Q2
(128)
in vector and axial-vector case, respectively, while NS CjI(1) = CjI(1) .

The integral expressions needed for the determination of the singlet gluon contributions we
list in Appendix C.2. For the gluon conformal moments we obtain

G V(1)
Q/2
Cj
(4 + 3j + j 2 )[S1 (j ) + S1 (j + 2)] + 2 + 3j + j 2
+
(1 + j )(2 + j )(3 + j )

G A(1)
Q/2
Cj
= nf

j
= nf
1 + S1 (j ) + S1 (j + 2) +
(1 + j )(2 + j )
G A(0)
j
G V(0)
j
ln
2
.
Q2
ln
2
,
Q2
(129)
(130)
The complete anomalous dimension matrix in MS scheme is known to two-loop accuracy

[136,155,82]. We present here the more involved case of the evolution in singlet sector and will
just state the substitutions needed to obtain non-singlet case results; for the extensive account of
the singlet case evolution see also Ref. [100]. Note that our normalization, adopted from DIS,
differs from the original one. Anomalous dimensions used in this work are related to those in
Ref. [100], denoted as
j k = N j j k |[100] N 1
k .
The transformation matrix is

1 0
N j = N (j )
,
0 j6
(131)
(132)
with the normalization factor

(3/2)(j + 1)
N(j ) = j
,
2 (j + 3/2)
277
(133)
originating from the altered definitions (21) and (23) of conformal operators. Explicitly, we have
for non-diagonal entries

k QG
j k |[100]
2k (j + 1)(k + 3/2) QQ j k |[100]
6
,
jk = j
(134)
2 (k + 1)(j + 3/2) j6 GQ j k |[100] jk GG j k |[100]
where the diagonal ones,
QQ

j |[100] j6 QG j |[100]
j jj = 6 GQ
(135)
,
j |[100] GG j |[100]
j
coincide with the DIS anomalous dimensions.
The evolution operator in the MS-scheme leads already at NLO to a mixing of conformal
GPD moments. For integer conformal spin these moments evolve as

1 (1)k
F j , 2 , =
Ej k (, 0 ; )F k , 2 , 0 .
2
j
(136)
k=0
To NLO accuracy the evolution operator is expanded as15

Ej k (, 0 ; )

2
s () ab (1)
a
ab (1)
j k
+ O s
Aj (, 0 )j k + Bj k (, 0 )
ab P j j k +
=
2
a,b=
s ()
s (0 )
b k
0
(137)
The diagonal term Ejj (, 0 ; ) coincides with the evolution operator Ej (, 0 ) of the preceding section, where the diagonal anomalous dimensions j are the same in MS and CS
scheme. Also the projector a P j is the same as before and is defined in Eq. (122). The matrix
ab B(1) (, ), defined for j k = {2, 4, . . .} (while 0 otherwise), assumes the form analogous
0
jk
to (124)
ab
(1)
(1)
Bj k = ab Rj k (, 0 |1)a P j j k b P k ,
for j k even,
(138)
(1)
while j k can be expressed as

(1)
j k = (0) , g + d 0 (0) j k .
(139)
This equation holds for parity even and odd anomalous dimension matrices, where g j k and d j k
transform according to Eq. (131). In analogy to Eq. (134), the matrices g j k , and d j k can be read
off from the results given in Ref. [100]. Taking into account the usual properties of projectors,
15 If one would be interested to resum all logs rather than only the leading ones, one certainly has to deal also with
the resummation of the non-diagonal entries. This problem has been solved in the flavor non-singlet sector for = 1 in
Ref. [156].
278

(1)
one can conveniently rewrite ab Bj k (, 0 ) as

ab (1)
Bj k (, 0 ) = ab Rj k (, 0 |1) a j b k 0 b k a P j d j k b P k + a P j g j k b P k .
(140)
The NLO solution of the evolution operator convoluted with the Wilson coefficients can be
more easily treated as the evolution of the conformal GPD moments itself. Namely, the convolution, represented as an infinite series and understood as an analytic function of j , can be
expressed in two equivalent forms

j

1 (1)k

1 (1)j j 1 C j Q2 /2 , s ()
Ej k (, 0 ; )F k , 2 , 0
2
j =0
k=0

2 2

1 (1)
j j 1
=
1 (1)
C k Q / , s () Ekj (, 0 ; 1)
2
k=j
j =0

F j , 2 , 0 .
(141)
The form on the r.h.s. corresponds to shifting the evolution to Wilson coefficient which is
more convenient for numerical treatment of non-diagonal terms. In this representation it is also
obvious that in DVCS kinematics the mixing under evolution is not suppressed by powers of 2 .
The evolved Wilson coefficients can furthermore be decomposed as

C j Q2 /20 , , 0

= C j Q2 /2 , s () Ejj (, 0 ; 1)
+

C j +k+2 Q2 /2 , s () Ej +k+2,j (, 0 ; 1).
(142)
k=0
even
Of course, we understand that a consequent expansion in s to NLO will be performed. To have a

numerically more efficient treatment this sum can be transformed into a MellinBarnes integral.
Employing the fact that the only residue of cot(k/2) is 2/ for even integer values of k, we
straightforwardly arrive at

C j Q2 /2 , , 0

= C j Q2 /2 , s () Ejj (, 0 ; 1)
1
4i

k
dk cot
C j +k+2 Q2 /2 , s () Ej +k+2,j (, 0 ; 1),
2
c+i
(143)
ci
where 2 < c < 0.

The formulae for the flavor non-singlet sector (see, for example, Ref. [137]) are simply obtained by reducing the matrix valued quantities to real (or complex) valued ones, by performing
the replacements
(0)
j k NS j k = j k ,
g j k NS gj k = gj k ,
d j k NS dj k = dj k .
j NS j
(0)
j NS j ,
= j ,
P j 1,
(144)
279
We recall that in our convention the off-diagonal entries (j > k) are dressed with a factor Nj /Nk ,
e.g., NS j k = (Nj /Nk )QQ j k |[100] , and that only to NLO accuracy, considered here, NS j k coincides with the quarkquark entry j k .
5. Parameterization of GPDs
For the phenomenology of GPDs it is crucial that one has a realistic ansatz at hand. At present,
it is popular to use a GPD ansatz [33] that is based on the spectral representation of GPDs
[24,28], for a review on this subject see [157]. We think that our formalism offers the possibility
to approach this problem from a new perspective and that it finally leads to a more flexible
parameterization. It is also remarkable that the task of modelling the x dependence of GPDs by
using the constraints of the lowest moment [158160], i.e., j = 0 yields elastic form factors, turns
in our approach into the inverse problem. Namely, it is rather the (conformal) spin dependence
of the form factors that determines the x dependence, where j = 0 serves as an boundary
condition. Since it is timely to find a parameterization that is the most appropriate for a fitting
procedure, we here only briefly outline our considerations.
We recall that the GPD support can be separated into two different momentum fraction regions, both having a dual partonic interpretation, namely, as the exchange of partons in the
s-channel and as mesonlike configurations in the t -channel. Moreover, these regions are tied owing to the Poincar invariance, see, e.g., Ref. [98]. This well-understood support property and its
partonic interpretation might be considered a reflection of the conjectured duality between t - and
s-channel processes; for a review of its phenomenological application, see Ref. [161]. This point
of view opens the door to employ strong interaction phenomenology (combined with some basic
principles) in finding a realistic model ansatz for GPDs and partial extraction of its parameters.
We believe that it is a promising starting point to think in terms of t -channel exchanges combined
with SO(3) partial wave analysis and Regge phenomenology and then to perform the crossing
to the DVCS process. This approach is complementary to those which are based on modelling
GPDs in terms of partonic degrees of freedom, see, for instance, [4149]. In the next two sections
we outline the SO(3) partial wave analysis of DVCS and, based on an intuitive picture, propose
an ansatz for the partial wave amplitudes, appearing in conformal GPD moments. In the third
section we try to get insight into the properties of this ansatz, and, finally, in Eq. (165) we present
its approximated version convenient for numerical studies and used in subsequent sections. The
reader more interested in the applications might just skip to that point.
5.1. SO(3) partial wave decomposition of conformal GPD moments
It has already been proposed in Ref. [90] to decompose conformal moments of meson GPDs
into irreducible SO(3) representations, namely, in terms of Legendre polynomials labelled by the
orbital angular momentum quantum number. These functions enter the partial wave decomposition of the t -channel scattering amplitude and depend on the scattering angle cm , defined in
the center-of-mass frame, see Fig. 3(a). After crossing, cos becomes16 the inverse of the scaling variable , i.e., cos = 1/ [24,90]. The basis of Legendre polynomials has later been
16 For the sake of simplicity, here we neglect
Bjorken limit.
1 + M 2 /P 2 proportional corrections and those which die out in the
280
Fig. 3. The fusion of two photons into a pair of (massless) mesons in the center-of-mass frame (a), partonic spacetime
picture (b), and hadronic point of view (c). The thick arrows indicate the spin projection of the particles.
adopted for the conformal moments of nucleon GPDs, too [91]. Let us here start with the analysis
and discussion which basis of polynomials one should employ.
To avoid cumbersome technical details, we restrict ourselves here to massless hadrons, so
that the crossing relations between helicity amplitudes become simple: the signs of both fourmomenta and helicities of the crossed particles are to be changed. However, one has to be careful
with the virtual photon,17 which we will not cross, but whose virtuality causes a rotation of
helicity amplitudes, i.e., when going from t - to s-channel amplitudes all three helicities of the
virtual photon contribute whether it is crossed or not. Interestingly, the dominant contribution for
the kinematics we are interested arises if its helicity will flip, too. Hence, either in the s- or in
the t -channel one can take that both photons have the same helicity in the center-of-mass frame.
In the real world, one has to take more care of the rotation of helicity amplitudes, namely, a
given s-channel helicity amplitude is expressed as a linear combination of t -channel ones, and
for massive and/or virtual particles all helicities contribute.
The partial wave expansion of t -channel helicity amplitudes is given in terms of Wigner
J ( ), which are essentially expressed by Jacobi polynomials [162]. Here the quand-matrices d,
tum number J refers to the total angular momentum, and () is given by the helicity difference
of initial (final) state particles. In the DVCS process, h h () , the twist-two CFFs, considered in this paper, appear in the helicity amplitudes in which both photons are transversely
polarized and have the same helicity [24,107]. As discussed above, since one photon is virtual,
one can take that the photons in the t -channel process () hh also have the same helicities.
For the rest of this section, we take them to be 1, denoted as (the amplitude for
follows from reflection). Hence, for scalar (hh) and spin-1/2 particles (e.g., h h ) the Wigner
d-matrices with = 0 (photons) and = {+1, 0, 1} appear in the partial wave expansion:

J J
h h
hh
J
d0,1 d0,1
(145)
for
and d0,0 for
.
h h
h h
The SO(3) partial wave expansion of the t -channel helicity amplitudes yields those for the
crossed CFFs H(t) and H (t) , defined via Eqs. (5) and (6) by the replacement P1 P1 and
q2 q2 . The hadronic tensor in the t -channel is obtained by the same substitutions from
Eq. (3). The evaluation of the helicity amplitudes in the center-of-mass frame is straightforward.
17 We are indebted to M. Diehl for critique and discussion on this subtle point.
Taking into account the angular dependence arising from the spinor bilinears,
q1 q2
1 5
1 cos2
V (P1 ) =
,
U (P2 )
2
cos
(q1 q2 ) (P2 P1 )
281
(146)
we can read off the effective partial waves in the SO(3) expansion of the t -channel CFFs H(t)
and H (t) for massless hadrons:

(t)
cos , , s , Q

cos
1 (1)J (t)
d J ( ).
fJ s , , Q2
2 0,1
2
1
cos
J =1
(147)
Here the upper (lower) sign refers to the parity even (odd) case, in which the total angular momentum is even (odd). The property of the Wigner matrices ensures then that the CFFs as scalar
valued functions are invariant under reflection. The lowest partial wave that appears is for a total
angular momentum J = 2(1). Up to a phase factor, which is not indicated, the crossing relation
for helicity amplitudes simply reads

(t)
T1,1/2;1,1/2 s, t, Q2 = T1/2,1/2,1,1 t (t) , s (t) , Q2 , t (t) = s, s (t) = t
(148)
F
(t)
(2J + 1)
and leads with cos = 1/ to those for the CFFs:

F , , 2 , Q2 = F (t) cos = 1/, , s (t) = 2 , Q2
(149)
where the photon virtuality asymmetry = / , cf. Eq. (9), is invariant under crossing. Combining this relation with the general form of the OPE (28), where the Wilson coefficients, as
functions of , are invariant under crossing, we can read off within our conventions the crossing
relation for conformal moments [122,98]

Fj , 2 = j +1 Fj(t) cos = 1/, s (t) = 2 .
(150)
(t)
Here Fj are the conformal moments of generalized distribution amplitudes [24,163]. Finally, a
formal comparison of the crossed version of the partial wave expansion (147) with the OPE (28)
allows us to identify the partial waves

J
J
J
d0,1
()
( )
, 2 J j + 1 for j = {1, 3, 5, . . .},
dH
(151)
2 1
cos =1/

J

J
J
d0,1 ( )
, 1 J j + 1 for j = {0, 2, 4, . . .},
dH ()
(152)
2 1
cos =1/
in the SO(3) expansion of conformal GPD moments Hj and H j , where J is an even and odd
number, respectively. The symmetry of the Wigner d-matrix ensures that these partial waves are
always even polynomials in of order J 2 or J 1. Consequently, within odd (even) j we
see that Hj (H j ) is an even polynomial in , as it is required from time reversal invariance. The
order of these polynomials is j 1 and j for odd and even values of j , respectively. Here we
omit further technical details and now discuss the partial wave amplitudes.
5.2. Ansatz for SO(3) partial wave amplitudes
First, we would like to illustrate with two examples the potential power of crossing, which
allows us to consider the t -channel process rather than the s-channel one. For DVCS off
282
a (pseudo)scalar particle, in which the photon helicity is conserved, we have one twist-two
CFF H, which belongs to the parity even sector [121]. The expansion of the corresponding
conformal GPD moments is done in terms of the Wigner d-matrix

J
j +1 d0,0
( ) cos =1/ = j +1 PJ (1/)
0 J j + 1,
and J, (j + 1) even,
with
(153)
which is a polynomial in of order j + 1 expressed in terms of a Legendre polynomial

J ( ). The meaning of the quantum numbers in the crossed channel is obvious:
PJ (cos ) d0,0
two photons, e.g., travelling oppositely along the z axes, form a two-particle state with zero magnetic quantum number, projected on the z axes, and producing a pair of (pseudo)scalar hadrons,
cf. Fig. 3(a). The total angular momentum of the initial state entirely transfer to the orbital angular momentum of the hadrons and so the latter is equal to J . The minimal value of the orbital
angular momentum J , determined from the magnetic quantum number of the photons, is zero.
Let us have a closer look at the partonic subprocess, which is thought of as the production of
a quarkantiquark pair state, labelled by the conformal spin j + 2, and travelling close to the
light cone in different directions. Since at leading twist-two, only chirally even operators appear,
helicity conservation holds at short-distances. Namely, the quark and anti-quark have opposite
helicities and, obviously, their spins point in the same direction, see Fig. 3(b). The conformal
spin j + 2 must be larger than J + 1 and is an odd number in the parity even sector. Hence, all
odd conformal partial waves with J 1 j contribute to the SO(3) partial wave amplitude with
the total orbital angular momentum J . However, the quarkantiquark pair is also bound by confinement and so they also represent an intermediate mesonic state with spin J , which consists of
J 1 units of the orbital angular momentum and one unit of the total angular momentum arising
from the aligned spins of the quark and the anti-quark. This mesonic resonance then decays into
two scalar hadrons. The quark (antiquark) must be combined with an antiquark (quark), which is
picked up from the vacuum. To not alter the quantum numbers of the vacuum, however, a quark
antiquark pair must be picked up with opposite spin (magnetic quantum number), as shown in
Fig. 3(b). Suppose that in one hadron the spins of the quark and antiquark are opposite, then in
the other ones they are aligned. This would be in contradiction with a naive quark model picture
and it can be resolved only if one of the produced quarks flips his helicity due to non-perturbative
effects.
Let us come back to DVCS off a nucleon, where again s-channel helicity conservation for the
photons is required. Then the four CFFs H, . . . , E can be expanded with respect to the t -channel
J
J . However, in our fictional world of massless particles, the helicity
partial waves d0,1
and d0,0
J is absent. Hence, we are left only
non-conserving quantities E and E should vanish and so d0,0
which are expanded with respect
with final states which have opposite helicities, i.e., H and H,
to the partial waves (151), (152). We can again employ our intuitive picture, and as in the intermediate resonance before, the quark and the anti-quark form a spin-one state with the orbital
angular momentum J 1. By picking up two quarks and anti-quarks from the vacuum the final
spin-1/2 nucleon and anti-nucleon are formed and one would expect that they carry the helicity
of the produced quark and anti-quark, respectively. The corresponding conformal GPD moment
for odd conformal spin j + 2 is now a polynomial of order j 1. In the case that the helicity
of the produced quark or anti-quark is flipped owing to non-perturbative effects, the helicities of
J appears in the conformal GPD
the produced nucleons will be the same. The partial wave d0,0
moments Ej and also Hj , which are polynomials of order j + 1.
283
Let us emphasise that we have therefore observed a relation between the helicity flip and
the order of conformal GPD moments. In general, the conformal GPD moment Hj is an even
polynomial in , which is of order j + 1 for odd j . The minimal value of J is now given by
the magnetic quantum number of the final state and so it is one. We remind ourselves that in
the partial waves (151) a term of the order j + 1 in is absent and so it cannot appear in Hj .
However, if we would allow for helicity non-conserved quantities, such a term arises, owing to
J , which will then appear in E and H , see the helicity representation for
the partial wave d0,0
j
j
GPDs in Ref. [106]. We conclude again that the restoration of full polynomiality, namely, up to
order (j + 1) for odd j , appears owing to the non-perturbative interaction in which the helicity of
the parton is reversed. In other words, the breaking of chirality is encoded in the conformal GPD
moments, which for given odd conformal spin j + 2 are polynomials of order j + 1. Certainly,
the highest possible term in naturally arises as a part of the SO(3) partial waves. At this stage
we see no reason to treat such terms in a special way and to collect them into a separate so-called
D-term [120], which was introduced to complete the common spectral representation of GPDs.
As mentioned in Section 3.2.2, we suggest to cure the spectral representation, e.g., as proposed
in Ref. [121], rather than to add a D-term.
The partial wave expansion we wrote down for GPDs was borrowed from the two-photon
fusion into two hadrons. In the case of hard vector meson production, the leading twist-two
contributions arise from a longitudinally polarized photon and vector meson. It is easy to name
J
J matrices
and d0,0
the Wigner matrices, which are the same as for DVCS, namely, we have d0,1
in the SO(3) expansion of the conformal GPD moments Hj and Ej . The former ones, given by
3/2
sin( )CJ 1 (cos )/ J (J + 1), are expressed by the Gegenbauer polynomials with index =
3/2 and the latter ones are the Legendre polynomials, or, if one likes, Gegenbauer polynomials
J , while
with index = 1/2. Again, the helicity non-conserved quantity Ej is formed from d0,0
J and an admixture of the d J waves. If we now replace the vector meson by a
Hj contains d0,1
0,0
pseudoscalar ones, the same partial waves appear for the parity odd quantities H j and E j .
We realize that the SO(3) partial wave expansions of GPDs are universal, i.e., are the same
for the considered processes. Such an expansion has several advantages, e.g., for the analytic
continuation of even to odd j values and allows for a simple implementation of the normalization
at j = 0. It also leads in a natural way to a term of order j +1 in the polynomial Ej for odd values
of j . Moreover, it is convenient to have for helicity conserved quantities the same functions that
also appear in the conformal SO(2, 1) representation. In a conformally invariant world there
would appear only one SO(3) partial wave with J = j + 1.
Inspired by the hadronic view on the t -channel scattering process, see Fig. 3(c), we now propose an ansatz for the partonic partial wave amplitudes that appear in the SO(3) expansion of the
conformal GPD moments. Thereby, we rely on the Regge description of high-energy processes,
which states that the high-energy behavior of the s-channel process () h h is dominated by
linear Regge trajectories (t) of pomeron and meson exchanges in the t -channel. The strength of
the photonphoton-to-meson (pomeron) coupling is contained in a vertex factor fJj that depends
on the conformal spin, too. The conformal spin j + 2 is considered as a variable conjugated to the
partonic momentum fraction x. Thus, as in the case of mesonic distribution amplitudes that are
3/2
expanded with respect to the Gegenbauer polynomials Cj (x), fJj can be viewed in the partonic
language as a probability amplitude for finding a quarkantiquark pair state with conformal spin
j + 2 inside of a meson with given spin J . Furthermore, the partial wave amplitudes also contain
the propagator 1/(m2 (J ) t) 1/(J (t)) of the exchanged particles, as well as the impact
form factor, describing the interaction with the target. These form factors will be modelled by a
284
p-pole ansatz, i.e., monopole (p = 1), dipole (p = 2), and so on, with a J -dependent cutoff mass
squared M 2 (J ).
All together, we propose the following ansatz for the helicity non-flip conformal GPD moments in terms of partonic partial waves amplitudes, which we write down as a sum over the
angular momentum (we employ the fact that the polynomials are even):

Hj
2
2
,
,

0
H j
J

j +1

d0,1 ()
fJj
j +1J
odd,
for
j
=
=
(154)
J +1
even,
J (2 ) (1 2 )p d0,1
()
J =2,even
1,odd
M 2 (J )
J () J d J (1/)/ (2 1)
where for odd (even) j the sum runs over even (odd) J . Here d0,1
0,1
is the crossed version (151), (152) of the Wigner matrix, where the rotation matrix of the spinor
bilinears is taken off. It is simply our partonic toy model (49), defined for complex valued j in
Eq. (62):

(3/2)(3 + j ) j
j, j + 3 1
J =j +1
d0,1
(155)
2 F1
() = 1+j
2 .
2
2 (3/2 + j )
For our later convenience, the crossed d-matrices are normalized in the forward limit to one:
J
() = 1.
lim d,
(156)
The conformal GPD moments (154) are even polynomials in of order j 1 or j , as required.
3/2
l+1
We also note that they are build from even polynomials d0,1
(cos ) cos Cl (cos ), where
l = {0, 2, . . . , j 1} or {0, 2, . . . , j }, with eigenvalue +1 under parity transformation. For helicity non-conserved quantities Ej and E j , analogous anstze can be written down in terms of
J partial waves, too.
Legendre functions. Thereby Hj will get an admixture from d0,0
5.3. Modelling of conformal GPD moments
It is beyond the scope of this paper to present a thorough study of realistic anstze for GPD
moments. We would rather like to convince the reader that the proposed parameterization generically works and then use some simplified version for our numerical studies. One important aspect
is the skewness dependence and its approximations. Another, new one, is the implementation of
lattice results.
The problem of how different approximations of the skewness dependence will affect the
size of the corresponding CFF will be investigated within a toy model. It is similar to Eq. (49),
which describes the Compton scattering process at tree level or, in other words, within a noninteracting parton picture. To make it somewhat more realistic, we multiply it with the generically
valid Mellin moments of an unpolarized valencelike parton density 35x 1/2 (1 x)3 /32 (here
normalized to one):
toy
Hj ( ) =
(1/2 + j )(9/2) j +1
( = ).
d
(9/2 + j )(1/2) H
(157)
285
Fig. 4. The imaginary (left) and real (right) part of the CFF H(, 2 = 0) arising from the toy ansatz (157) for conformal GPD moments within different approximations (159): exact (solid), next-to-leading (dash-dotted), leading (dashed)
SO(3) partial waves and = 0 (dotted) term.
j +1
Here the partial wave dH (), given in Eq. (155), is normalized to one for vanishing skewness .
toy
For the sake of illustration, we now expand Hj ( ) with respect to Legendre polynomials:

j 2
2 1 + (1)j l j l2 (l + 3/2)(j + 1) l
j +1
j
dH () = d0,0 () +
d (),
4
2
2
(j + 3/2)(l + 1) 0,0
(158)
l=0
l () = (/2)l (l + 1)P (1/)/ (l + 1/2). The first term on the r.h.s. is the leading
where d0,0
l
partial wave, while all other partial waves, contained in the remaining sum, are suppressed by
powers of 2 . We will now study the numerical deviation in the CFF that is induced by an
j +1
approximation of the partial wave dH (). For this purpose we make three choices: we drop
the skewness dependence altogether, take the leading partial wave, and include also the next-toleading ones:
j +1
dH 0 = 1,
j +1
j
dH LO = d0,0 (),
(j 1)j 2 j 2
j +1
j
d0,0 ().
dH NLO = d0,0 () +
(4j 2 1)
(159)
Note that some care is needed in the truncation of exact partial waves. The approximated partial
j +1
wave dH NLO differs for j = 1 (and also j = 0) from the exact one by /27 (3/). This is
1
2
() (d0,0
()) that cancels the zero at j = 1 (or j = 0) in the expansion
induced by a pole in d0,0
coefficient. Finally, this leads to an addenda in the real part of the CFFs that must be subtracted.
In our toy example the subtraction term for the next-to-leading approximation reads 5/27.
toy
The outcome for the imaginary and real part of the CFF Hj ( ) is displayed in the left and
right panel of Fig. 4, respectively. In the next-to-leading approximation (dash-dotted) one can
hardly see a difference to the exact CFF (solid). If we take only the leading SO(3) partial wave
(dashed), we realize that the deviation from the exact CFF (solid) is small over the whole
region, in particular for the imaginary part. Furthermore, making the expansion in powers of 2
and retaining only the leading = 0 term (dotted) yields the imaginary part that approaches the
exact result already for 0.7. However, it has an unrealistic feature that it does not vanish
in the limit 1, as it should. Concerning the real part, both of these approximations start to
digress from the exact result for 0.3, where the deviation is larger for the expansion in .
To summarize, the expansion with respect to the angular momentum looks promising and
works rather well in the case that the leading pole factorizes. We have also found in the expansion
286
with respect to Legendre polynomials, which was used in Ref. [91], that the minimal version of
the dual model, i.e., taking leading and next-to-leading partial waves, yields also an astonishing
agreement with the exact result. We have also observed that in the experimentally accessible
kinematical region the expansion in practically coincides with the exact result, in particular
for the imaginary part. Nevertheless, there could be a drawback if the (leading) Regge poles
are non-factorizable as it is the case in our model ansatz (154). The next-to-leading term in the
partial wave expansion generates an artificial pole, e.g., 1/(j 1 (2 )), that is situated on
the r.h.s. of the leading Regge pole 1/(j + 1 (2 )). Such a pole leads to an addendum in
the MellinBarnes representation for CFFs with the same small -behavior as the leading pole.
Thus, we expect that the normalization of the CFFs, e.g., for 2 = 0, is governed by all terms in
the partial wave expansion. A closer look at this potential problem should be given somewhere
else. For the time being, we rely on the leading term in the expansion with respect to ,

Fj , 2 , 20 =

fj +1,j
1
+ O 2 ,
2
2

j + 1 ( ) (1 2 )p
(160)
Mj
which we will use in our numerical studies.

Next we fix the normalization of conformal GPD moments at = 0:

Fj = 0, 2 = 0, 20 =
fj +1,j
.
1 + j (0)
(161)
The conformal moments of helicity non-flip GPDs Hj and H j are then reduced to the Mellin
moments of unpolarized (q) and polarized (q) parton densities,
hj +1,j
=
1 + j (0)
1
dx x q(x, 0 ) and
0

h j +1,j
= dx x j q(x, 0 ),
1 + j (0)
1
(162)
respectively. They are parameterized with the guidance of Regge phenomenology, determining
its small x behavior to be x (0) , whereas counting rules suggest their large x behavior, parameterized as (1 x) for x 1. Corresponding to Eq. (162), the generic ansatz for parton densities
in the x space yields the following Mellin moments, e.g.,

q x, 20 = Nx 0 (1 x) hj +1,j = N 1 + j (0) B(1 0 + j, + 1), (163)
where B(a, b) = (a)(b)/ (a + b) is the Euler beta function. Note that the intercept 0 of
the Regge trajectory occurring in our ansatz and that obtained from a given fit of parton densities
must agree. Hence, the leading pole in the Mellin moments (163) is cancelled and replaced by the
corresponding trajectory.18 Certainly, a more realistic ansatz for parton densities, to be used in
global fits, could be obtained by a linear combination of such building blocks. In practice, either
one can take the Mellin moments of one of the standard parameterizations of parton densities
[310] or, if they are plagued by larger errors or theoretical uncertainties, one can perform a
simultaneous fit of exclusive (e.g., DVCS) and inclusive data. Here one has to bear in mind that
parton densities are scheme-dependent quantities.
18 Remaining non-leading poles at j = {2 + , 3 + , . . .} are considered an artifact of the parameterization and
0
0
appear as subleading contributions in the CFFs.
287
Further constraints for the conformal moments arise at j = 0. There they are given by the
so-called partonic form factor

Fj =0 , 2 =
N0
1
.
2
1 ( ) (1 22 )p
(164)
M0
After adjusting the flavor quantum numbers, the partonic form factors coincide with the measured
elastic form factors of the proton. The free parameters left, i.e., the cut-off mass M02 and the power
behavior p at large 2 can be taken from a fit to experimental data, where a refinement of our
parameterization might be necessary. We note that form factors are scheme independent, except
the axial one in the flavor singlet sector. Another advantage of the parameterization (154), which
we will not use here, is that the partonic partial wave amplitudes fJj are related to physical ones,
denoted as fJ (W 2 , , Q2 ), in the t -channel. The physical amplitudes fJ are given as a series
of partonic ones, where the sum runs over the conformal spin j + 2. Note that such a relation
depends on our scheme conventions, too.
Finally, plugging the normalization (163) into the ansatz (160), we end up with the following
simplified GPD parameterization:

j + 1 (0)
2
1
Fj , 2 , 20 = N B 1 (0) + j, + 1
t p +O .
2
j + 1 ( ) (1 2 )
(165)
Mj
We remark that for large ||2 , the counting rules predict a power-like falloff of form factors as
(1/||2 )ns , where ns is the number of spectators, while the large x behavior of parton densities
is (1 x)2ns 1 , i.e.,
p = ns 1,
= 2ns 1.
(166)
An inclusiveexclusive relation between the unpolarized DIS structure function W2 and the electromagnetic form factor F2 has also been derived by Drell and Yan within a field-theoretical
model that accounts for the dynamics of partons [164]. This result coincides with the counting
rules for valence quarks. However, one should be aware that for sea-quarks and gluons, these
counting rules might be modified; for a discussion see, e.g., Ref. [2]. For simplicity, we will not
account for that here.
Finally, we would like to demonstrate that our ansatz (165) can be easily adjusted to the experimental data on electromagnetic form factors and, moreover, that it is well suited to include
lattice data. In particular, present lattice measurements [3540] give insight into the functional
change of the 2 dependence with increasing conformal spin. This dependence arises from two
sources: the Regge trajectory and the impact form factor. As an example, we consider the conformal GPD moments of valence quarks, where the Regge trajectory is generically correctly
described by (t) = (0) + t with the intercept (0) = 1/2 and the slope = 1 GeV2 . For
two spectators the counting rules state that the impact form factor is a monopole, i.e., p = 1. For
its cutoff mass we naturally choose two times the proton mass M0 2Mp = 1.88 GeV. Hence,
our conformal GPD moments are written as

1
1
Hjval , 2 , 20
(167)
+ O 2 .
2
2
1 2 1 2
mj
Mj
Here we have introduced the monopole mass squared m2j = (1 (0) + j )/ = (1/2 + j ) GeV2 ,
which arises from the mesonic Regge trajectory. We might also assume that the spin dependence
288
Fig. 5. The 2 dependence of the conformal GPD moments (167) for valence quarks (left) and the spin dependence of the
effective dipole mass (170) (right). In the left panel the solid line displays the dipole fit (169) to the experimental data and
the dash-dotted one is our ansatz (167) with j = 0. The dashed and dotted lines show the changes with j = {1, 2, 3} for
our ansatz and a dipole fit within the dipole masses (170), where M 2 = (Mp )2 . In the right panel we show the effective
dipole masses (170) as a function of spin within M 2 : 0 (dashed), Mp2 (dash-dotted), (2Mp )2 (solid). The dotted line has
a slope that arises from M 2 = 2(2Mp )2 and is compatible with lattice measurements [165] for the flavor non-singlet
combination u d in the heavy-pion world, which was linearly extrapolated to the physical pion mass.
of the monopole mass squared Mj2 is linear

Mj2 = M02 + M 2 j.
(168)
The generic ansatz (167) leads to a satisfying description of the electromagnetic proton form
factor. In Fig. 5(a) we confront this ansatz (dash-dotted) with the dipole fit (solid) of the electromagnetic proton form factor F1 ,

1 + Q2 /1.26 GeV2
F1 Q2 = 2 =
,
2
Q2
(1 + Q2 )2 (1 + M
2)
mdipol
m2dipol = 0.71 GeV2 ,
(169)
and realize that they fairly agree. We also display the 2 dependence of the conformal moments
within the choice M 2 = (Mp )2 . A larger value of M 2 is compatible with the slope measured
on lattice in the heavy-pion world, as shown in Fig. 5(b) (dotted line). However, the intercept
differs (in fact, the lowest moment does not describe the F1 (Q2 ) data). The masses are extracted
from a dipole fit to the lattice data. To compare with our ansatz, we calculated the effective
masses
1

1
2
Mdipol = 2
(170)
,
+
1 + j (0) M02 + M 2 j
appearing in a dipole fit. For |2 | < 1 GeV2 such a refitting procedure only weakly modifies
the GPD moments, compare dashed and dotted lines in Fig. 5(a), and so this procedure is justified
to some extent.
6. Perturbative corrections of the DVCS cross section
This section is devoted to the numerical analysis of radiative corrections to CFFs. We will
concentrate on the CFF H, since it is the dominant contribution in most of the DVCS observables. The three remaining twist-two CFFs are usually suppressed by kinematical factors [68].
289
To reveal these CFFs from experimental data, it is therefore crucial to understand the theoretical uncertainties of H. Our findings can be qualitatively adapted for the helicity non-conserved
CFF E, which enters the parity even sector, too. In this sector we face a peculiarity that is related
to the appearance of the pomeron trajectory in high-energy scattering. Technically, it shows up as
an essential singularity of the evolution operator at j = 0, see anomalous dimensions (108) and
(109). Such a singularity is absent from the parity odd sector. Hence, our results in the small
region are not directly applicable to the study of radiative corrections of the two remaining parity
odd CFFs H and E.

Based on our model (154), we introduce in the next section a simplified generic ansatz for conformal GPD moments, which will serve us in our numerical studies of the experimentally accessible kinematical regions. In Section 6.2 we shortly discuss the features of flavor singlet and nonsinglet CFFs to LO accuracy. In the following two sections we then elaborately analyze the size
of radiative corrections to NLO and, finally, to NNLO accuracy. Thereby, independently of the
considered order and scheme, we take the same conformal GPD moments which thus leads to different CFFs. However, note that CFFs are physical observables and do not depend on our conventions. Therefore, conversely, the conformal GPD moments revealed from a measurement of the
physical CFFs will depend on both our scheme conventions and the approximation. Nevertheless,
the results of our analysis give us a measure for both the reparameterization of the GPD ansatz
needed to compensate convention change and for the convergency of the perturbation theory.
6.1. A simplified generic ansatz for conformal GPD moments
The kinematics of interest can be restricted to = 0.5, i.e., the Bjorken scaling variable
xBj = 2/(1 + ) is bounded by 2/3. For simplicity, we do not resum the partial waves and rely
on the leading term in the expansion (165).
Let us specify the other parameters in the ansatz (165) for a valence-like helicity non-flip
GPD H . The leading meson Regge trajectory is generically given by (t) = (0) + t with
(0) = 1/2 and = 1 GeV2 . For two spectators the counting rules (166) state that p = 1
and = 3. For the cut-off mass of the impact form factor we choose (168) with M0 = 2Mp =
2M = 1.88 GeV. Hence, we have for our GPD moments:

B(1/2 + j, 4)

1
1
+ O 2 ,
Hjval , 2 , 20 =
(171)
2
2
2

B(1/2, 4) 1
2 1 M 2 (4+j )
p
(1+2j ) GeV
which is normalized to one for j = 0 and 2 = 0.
The conformal GPD moments with the flavor non-singlet combination (A.7) or (A.8), which
is also relevant for DVCS, is built from valence and sea quarks. For four active quarks and the
SU(2) flavor symmetric sea, the sea quark part arises from the difference of charm and strange
(anti-)quarks, cf. Eq. (A.8). Suppose we are at the charm threshold and the charm sea is generated
dynamically. Then, essentially, only the (anti-) strange quark counts. We might assume that the
s anti-quarks have the same conformal GPD moments and so the breaking of SU(3) flavor
u,
d,
symmetry is described by one single parameter Rs/u , defined as the ratio of s to u anti-quarks.
For the purpose of illustration we consider two alternative cases: one without and one with sea
quark admixture:

NS
(172)
Hj , 2 , 20 = Hjval , 2 , 20 ,

Rs/u
NS
(173)
Hj , 2 , 20 = Hjval , 2 , 20
Hjsea , 2 , 20 .
2 + Rs/u
290
For the valence quark content we rely on Eq. (171) and the sea quark GPD moments are specified
below in the ansatz (175) with Nsea = 4/15. For the SU(3) flavor symmetry breaking parameter
we choose the often used value Rs/u = 1/2.
In the flavor singlet sector we need an ansatz for both quark and gluon conformal GPD moments:

Hj

2
2
, 2 , 20 ,
H j , , 0 =
G
Hj
u
Hj = Hjsea + Hj val + Hj val = Hjsea + 3Hjval .
(174)
The singlet quark combination consists of sea and valence quarks, cf. (A.7) and (A.8), and we
employ isospin symmetry to express the latter within the ansatz (171). Moreover, we rely on the
counting rules (166), i.e., G = 5, pG = 2, sea = 7, psea = 3, and take the same cut-off mass for
the impact form factors as for the valence quarks before. Hence, analogously to Eq. (171), the
generic ansatz (165) then reads:

B(1 sea (0) + j, 8)
1
Hjsea , 2 , 20 = Nsea
B(2 sea (0), 8) 1 sea2
(mj

G

2
Hj , 2 , 0
1
)2
(1
2
(Mjsea )2
)3

+ O 2 ,

B(1 G (0) + j, 6)
1
1
= NG
+ O 2 ,
2
2

B(2 G (0), 6) 1 G (1 G )2
2
2
(mj )
(175)
(176)
(Mj )
where MjG = Mjsea = Mp2 (4 + j ) and m2j = (1 (0) + j )/ is expressed in terms of the intercept and slope of the Regge trajectories, specified below.
The leading trajectory arises now from the pomeron exchange. We remind that in deeply
2
inelastic scattering the structure function F2 (1/xBj )(Q ) grows with increasing Q2 . Here the
exponent is governed by the intercept of the Regge trajectory = (0) 1, which is, in the
language of Regge phenomenology, that of the soft pomeron (for Q2 0):
P (t) = P (0) + P t,
P (0) 1, P = 0.25.
(177)
However, in hard processes the trajectory will effectively change due to evolution, which differently effects the behavior of quark and gluon parton densities. In particular, the value of (0)
increases, while that of decreases with growing resolution scale Q2 , e.g., see Ref. [166]. In
the flavor singlet sector the size of radiative corrections and the strength of evolution crucially
depends on the effective pomeron parameters, see for instance the variation of NLO corrections
obtained in [85]. In our numerical studies we shall consider two scenarios in which the radiative
corrections to NLO accuracy are respectively small and large. It is known that the corrections
to the quark sector are rather stable, while the main uncertainty arises from the gluons, which
enter in the perturbative description of CFFs at NLO. Small and large NLO corrections can be
obtained by choosing a softer and harder gluon, respectively:
soft gluon:
NG = 0.3,
hard gluon: NG = 0.4,
G (0) = sea (0) 0.2,
(178)
G (0) = sea (0) + 0.05.
(179)
= = 0.15, all at the input

Furthermore, we choose a realistic value for sea (0) = 1.1 and sea
G
2
2
scale Q0 = 2.5 GeV .
We remark that we normalize the sea quark (175) and gluon (176) moments at j = 1, so Nsea
and NG give the amount of momentum fraction carried by the considered parton species. Because
291
of the momentum sum rule (A.20), valid in the forward kinematics, we have the constraint
1
NG + Nsea +
dx x[uv + dv ](x) 1.
(180)
Within our toy ansatz for valence quark moments (171) we have for their momentum the generic
value 1/3 and so Nsea = 2/3 NG . We note, however, that the separate contributions will change
during the evolution. In the asymptotic limit Q , the evolution equation tells us that NG =
4CF /(4CF + nf ), i.e., that more than 50% of the longitudinal proton momentum is carried by
gluons. As it is experimentally verified, at a scale of a few GeV2 the gluons already carry about
40% of the momentum.
6.2. Compton form factors to LO accuracy
The CFFs S H and NS H are evaluated from the specified conformal GPD moments by the
MellinBarnes integral (58). Essentially, we employ here the same technique that is well established in deeply inelastic scattering, only the integrand is now more intricate. The integral does
not depend on the integration path, going from c i to c + i, as long as we do not cross any
singularities, see Fig. 2(a). In practice we use this property to get closer to the leading singularity, lying left of the integration path. Since the conformal GPD moments rapidly decrease with
growing conformal spin j , we can in practice cut the integration path to a finite length. Moreover,
the convergency can be improved by a separate rotation of the integration path in the upper and
lower half-plane in such a way that along the path the real part decreases, as shown in Fig. 2(b).
Nevertheless, it is always a good idea to exercise proper care in the choice of the integration path
and to check the numerical accuracy. Once this is done for given conformal GPD moments, the
numerical treatment is simple, fast, and stable even in NNLO. We use two different codes for
the numerical evaluation, one specifically written in F ORTRAN and, alternatively, the integration
routine from M ATHEMATICA.
For the evaluation of the CFFs to LO accuracy, we use the flavor non-singlet Wilson coefficients (91a) and (92) as well as the singlet ones (101a) and (102). The Q2 evolution is governed
by the flavor non-singlet operator (116) and singlet one (121)(123), approximated to LO. We
equate the factorization scale with the photon virtuality: 2 = Q2 . The various anstze, given
in Eqs. (172), (173), (175), and (176), are taken at the input scale Q20 = 2.5 GeV2 . The running
coupling in LO approximation is normalized to s (Q20 = 2.5 GeV2 )/ = 0.1, where the number
of active quarks is four.
We found it useful to describe the complex valued CFFs in polar coordinates,

H , 2 , Q2 = H , 2 , Q2 exp i , 2 , Q2 , < = arg(H) ,
(181)
rather than Cartesian ones. This avoids a discussion of radiative corrections in the vicinity of
zeros, appearing, e.g., in the real part of certain CFFs. Moreover, the polar coordinates reveal a
simple shape of CFFs in their functional dependence on and Q2 . This is demonstrated in Fig. 6,
where we display the modulus (left) and the phase (right) of the pure valence non-singlet (up) and
the hard gluon singlet (down) CFF, respectively. For the former one we choose the kinematical
region that covers the phase space of present fixed target experiments, namely 0.05 0.5
[0.1 xBj 0.65] and 1 GeV2 Q2 10 GeV2 . For the latter one we also include the phase
space that is explored in collider experiments, i.e., 106 0.5 [2 106 xBj 0.65] and
292
Fig. 6. The scaled moduli (a), (c) and phases (b), (d) of the flavor non-singlet (up) and singlet (down) CFF H are
displayed to LO accuracy versus Q2 and for fixed 2 = 0.25 GeV2 . At the input scale Q20 = 2.5 GeV2 we use the
anstze (172) and (179) for conformal GPD moments. The running coupling is normalized to s (Q20 )/ = 0.1, where
the number of active quarks is set to four.
1 GeV2 Q2 100 GeV2 . The simple shape of both the modulus and the phase is perhaps
a more general feature of CFFs and should not be considered as an artifact of our approximation [85]. It is for 1 GeV2 Q2 10 GeV2 almost independent on the photon virtuality. In
Section 6.3.1 we will have a closer look to the evolution. Here we remark only that both the
moduli and phases of CFFs are rather planar in fixed target kinematics. With increasing 1/
the evolution starts strongly to affect the modulus in the singlet sector. It grows with increasing scale Q2 , namely, in such a way that the resumed logarithmical scaling violations lead to
a power like change of its 1/ -dependence. This feature is well known from unpolarized DIS
and in agreement with experimental DVCS results [6567]. The phase is much less affected by
evolution and in the small region it is nearly independent of . It monotonously decreases from
its input value to be changed by less than 25% at Q2 = 100 GeV2 .
6.3. Size of NLO radiative corrections: CS versus MS scheme
Now we explore the radiative corrections to NLO accuracy and in particular the differences
between the CS and MS schemes, separately for the flavor non-singlet and singlet sector. Before
we do so let us explain the relation between these schemes and define the quantities that serve us
as measure for the size of perturbative corrections.
293
The perturbative expansion of a CFF in the CS scheme, e.g., for the flavor non-singlet case,
might be structurally written to NLO as

s (1)CS
F = C (0) +
(182)
+ C (0) A(1) E (0) F CS + O s2 .
C
2
Here E (0) denotes the evolution operator (116) in LO approximation and A(1) is the NLO correction. The convolution symbol indicates the integration over the complex valued conformal
spin j , which is for shortness not particularized, where the measure includes the appropriate
normalization and dependence, see (58) and (91a). By construction, in the forward kinematics
this scheme is identical with the MS one, see normalization condition (75). However, for DVCS
kinematics the form of Wilson coefficients and evolution operator in the MS scheme are already
at NLO modified by off-diagonal, i.e., proportional, terms, see Eqs. (72) and (73). Because of
the particularity of the DVCS process, where = , both Wilson coefficients and evolution operator are finally modified by an infinite sum of terms, which appears then in front of the conformal
GPD moments, cf. Eq. (141):

(0)

s (1)MS
(0)
(0) (1)
(0)
(1)
F= C +
(183)
E F MS + O s2 .
+C A +C B
C
2
Here C (0) B (1) contains off-diagonal part of the evolution operator, where symbolizes the
summation over the conformal spin, cf. Eq. (142). Since the physical observable F is independent
of our conventions, the conformal moments in both schemes are related to each other by a scheme
transformation. At the input scale Q2 = Q20 , where E (0) = 1, A(1) = 0, and B (1) = 0, we might
express this transformation by a finite factorization (or renormalization) constant z(1) :
s (1)
F MS = F CS +
(184)
C (1)CS = C (1)MS + C (0) z(1) .
z () F CS ,
2
j k z(1) contains only offFor (positive) integer conformal spin the triangular matrix zj(1)
k () =
jk
diagonal entries, i.e., j k + 2. Note that the change of Wilson coefficients is of course independent, while the skewness dependence of the conformal GPD moments is altered, which
is at least suppressed by a factor 2 . In particular, the two lowest GPD moments are untouched
by the scheme transformation, while the (positive non-vanishing integer) moments are modified
by s and 2 suppressed contributions:
s 2
F0MS = F0CS ,
for j = 3, 4, . . . .
F1MS = F1CS , and FjMS = FjCS +
O
2
(185)
Strictly spoken, the truncation of the perturbative expansion also induces a discrepancy in the
CFFs between these two schemes, which is beyond the approximation we are dealing with, i.e.,
of order s2 . However, the conformal GPD moments revealed from a given data set will differ to
order s . In the following we study the NLO corrections in both schemes within the same ansatz.
The resulting deviation in the CFFs can be viewed as a measure for the needed reparameterization
of conformal GPD moments by altering their dependencies.
We introduce now the quantities that we utilize as a measures of the scheme dependence and,
foremostly, as indicators of the convergency of the perturbation series. It is natural to employ for
this purpose the ratio of the CFF at order NP LO to the one at order NP 1 LO, where P = {0, 1, 2}
stands for LO, NLO, and NNLO order, respectively:

P
HP
2
2 2
P
2
2 2
2
2 2
,
Q
,
Q
,

,
Q
,

K
,

exp
i
Q
Q
Q0 .
0
0
HP 1
(186)
294
The phase difference

NP LO

H
P = arg
O sP
P
1
HN LO
(187)
is formally of order (s /2)P . If convergency holds, it diminishes in higher orders. Under this
circumstance the ratio of moduli
KP =
|HN
P LO
P 1
|HN LO |
approaches one, i.e., its deviation from one vanishes, too:

P K = K P 1 O sP .
(188)
(189)
If we suppose that the perturbative expansion is not ill-behaved, then the radiative corrections
should not overshoot the size itself of the CFF in a given order, i.e.,
NP LO

P 1
P 1
H
(190)
HN LO = r HN LO with r < 1.
If this inequality holds true, the phase difference (187) is geometrically constrained by the value
of r, otherwise it is independent. More precisely, we have the upper bound

sin P r P r for r 1.
(191)
2
Moreover, the triangle inequality, applied to Eq. (190), constrains the variation of the modulus,
namely, r | P K|. Employing the cos-theorem, we can appraise the radiative corrections (190),
quantified by the ratio r, from the variation of the phase and modulus. It will turn out that in our
analysis the phase difference | P | is always small, i.e., | P | /2. Hence, we can rely on
the expanded version of this theorem,

2

2
r 2 P K + 1 + P K P ,
(192)
which gives us a simple form of the constraint among the variations of the modulus and phase as
well as the size of radiative corrections. Note that in the case of a (very) small phase change the
variation of the modulus is roughly estimated to be | P K| r.
6.3.1. Flavor non-singlet sector
The flavor non-singlet CFF NS H is straightforwardly evaluated from the two anstze (172)
and (173) by means of the MellinBarnes integral (58). The Wilson coefficients, needed for our
NLO (P = 1) analysis, are listed for the CS and MS scheme in Sections 4.2.1 and 4.3, respectively, while the relevant expansion of the evolution operator can be read off from Sections 4.2.3
and 4.3. As said above, we combine the perturbative expansion of the Wilson coefficients with
that of the evolution operator in a consistent manner, where the leading logs are resummed, cf.
Eqs. (182) and (183). As long as it is not stated otherwise we equate the factorization f
and renormalization r scales with the photon virtuality Q. As input scale for the conformal
GPD moments we use as before Q20 = 2.5 GeV2 . This input scale serves us also to normalize
the coupling constant s (Q20 )/ = 0.1, where its running is described by the exact numerical
solution of the NLO renormalization group equation.
In Fig. 7 we display the relative NLO corrections (189) and (187) in the CS (dashed) and MS
(dash-dotted) scheme to the moduli (up) and phases (down) for the phase space of present fixed
295
Fig. 7. The relative NLO radiative corrections (189) and (187) to NS H for 2 = 0 (left) and 2 = 1 GeV2 (right)
are plotted versus for the moduli (up) and phases (down) within the CS (dashed) and MS (dash-dotted) scheme. Thin
and thick lines refer to a valence-like ansatz (172) and one with a sea-quark admixture (173), given at the input scale
Q20 = 2.5 GeV2 . We equated the scales f = r = Q and take the normalization condition s (Q20 )/ = 0.1.
target experiments. Thereby, we employ the pure valence-like ansatz (thin lines), i.e., Eq. (172),
and the one with a sea quark admixture (thick lines), given in Eq. (173). For the momentum transfer squared we take two extreme values, namely, 2 = 0 (left) and 2 = 1 GeV2
(right). Comparing the resulting radiative corrections for both choices, one realizes that the 2 dependencies only slightly influence their size.
Due to the radiative corrections the phases (lower panels) increase by a small amount, except
for the valence-like ansatz (thin lines) where we observe a decrease for smaller values of .
In any case, the absolute value of the phase differences | P | does not exceed 0.1 rad. The
influence of the NLO corrections on the moduli (upper panels) is more pronounced. Generally,
they moderately reduce the moduli, however, for the sea quark admixture ansatz in the CS (thick
dashed) there is a small increase at larger values of . The modulus is more affected for the
valence-like ansatz (thin) than for the one with a sea quark admixture (thick), while in the case
of the phase difference the situation is reversed.
Comparing the corrections in the CS (dashed) and MS (dash-dotted) scheme, one realizes
that in the former scheme they are smaller for the moduli, while the phase differences are almost independent of the specific choice. This is in agreement with the findings of Ref. [101],
where a slightly different ansatz has been chosen. As explained above, the difference between
the two schemes originates from the skewness dependence. For positive integer conformal spin,
we would count them in the conformal GPD moments as 2 effects, suppressed by s /2 . However, as we also spelled out, in DVCS kinematics we should not use = as an expansion
parameter, since the change of Wilson coefficients is -independent. Indeed, roughly spoken, the
moduli differences in both schemes are of the same order as the radiative corrections themselves
and nearly independent. This observation should be understood as a warning that 2 suppressed
terms in conformal GPD moments perhaps cannot be simply neglected by formal 2 counting.
296
Fig. 8. The modulus (up) and phase (down) changes of the evolved CFFs NS H are plotted for fixed Q2 = 1 GeV2 (left),
Q2 = 10 GeV2 (right), and 2 = 0.25 GeV2 versus in LO (dotted) and NLO: CS (dashed) and MS (dash-dotted)
scheme. The anstze and scale setting prescriptions are the same as in Fig. 7.
We consider now the evolution to LO and NLO accuracy, where we compare the modulus and
phase at a given scale with those at the input scale, quantified by the ratios:

2 2 |H(Q2 )|
2 2
H(Q2 )
K Q , Q0 =
(193)
1,
Q , Q0 = arg
.
|H(Q20 )|
H(Q20 )
Note that in contrast to the definitions (187) and (189), in which the variation is denoted by , here
we do not compare the CFFs in different order, but rather measure the strength of the evolution
within a given order. The evolution in the MS scheme is consistently treated, i.e., the mixing of
conformal GPD moments is taken into account. We remark, however, that this effect is tiny and
can be safely neglected. For instance, for the quantity
|NS H(Q2 )| |NS Hdia (Q2 )|
,
|NS H(Q2 )|
where the superscript dia stands for neglecting the non-diagonal parts in the anomalous dimension NS j k , we find in a broad range of and Q a value on the level of few per mil. The phase
differences are negligible, too.
In Fig. 8 we show the evolution effects for the same anstze as before, in LO (dotted) and
NLO for both the CS (dashed) and MS (dash-dotted) schemes. In the left and right panels we
plot the quantities (193) at a scale Q2 = 1 GeV2 and Q2 = 10 GeV2 , respectively. For simplicity, we do not perform any matching at the charm threshold. As already observed for the
radiative corrections at the input scale, see lower panels in Fig. 7, the phase differences in right
[left] panels in Fig. 8 are again rather small and lie for forward [backward] evolution in the interval 0.06(0.02) [0.020.06 ] rad. The signs tell us that the phases decrease during
the evolution to a larger scale. Radiative corrections amplify this effect, where the differences
between both schemes are again tiny. Comparing the upper left and right panels, we see that for
297
the valence-like ansatz in LO (thin dotted line) the evolution mildly affects the moduli, i.e., of
the order of about 5% at the edges of the phase space19 explored in the present fixed target
experiments.
Furthermore, the modulus crosses for low and high values of Q2 (upper left and right panels)
the zero line at almost the same point 0.12 to LO accuracy (thin dotted). This means that the
CFF in the vicinity of this point is nearly scale independent. As in DIS, the evolution predicts a
decreasing modulus for > 0 and increasing one for < 0 with growing Q2 . The value of 0 is
a function of 2 and depends on the input. In our example it is shifted by radiative corrections to
the right (thin dashed and dash-dotted lines). This balance effect arises from the fact that forward
evolution suppresses (enhances) the large (small) momentum fraction region.
The radiative corrections lead to an amplification of the evolution effects. Note that the Wilson
coefficients now also depend on Q2 and the perturbative corrections are getting smaller with
increasing Q2 . Now the variation of the CFF modulus reaches for the valence-like ansatz (thin
lines) the 10% [15%] (for small ) to 8% [6%] (for large ) level in the CS (dashed) [MS
(dash-dotted)] scheme. For the ansatz with sea quark admixture (thick lines) there is no crossing
point with the zero line anymore and so the balance-point is shifted outside of the discussed
kinematical region. Also here the strength of evolution grows by radiative corrections and varies
in the range from 6% (small ) to 15% (large ).
As we realized, in both anstze the NLO corrections within the MS (dash-dotted lines) and the
CS (dashed lines) change the LO prediction (dotted line). The differences, caused by the scheme
dependence, mainly arises from the different Wilson coefficients to NLO, which are multiplied
with the LO evolution operator. The contribution of the diagonal part of the NLO anomalous
dimensions is small, i.e., about one percent. As we spelled out above, the influence of the nondiagonal part, appearing in the MS scheme in the anomalous dimensions, is negligible.
We would like to briefly confront evolution effects with experimental measurements from
the Hall A experiment at Jefferson LAB [75], where scaling was reported. Within the lever arm
1.5 GeV2 Q2 2.5 GeV2 , we find for the valence-like ansatz that the scaling violation due to
the LO evolution is small, namely, the modulus of the CFF varies by about 1.2% for xBj = 0.36
(i.e., = 0.22) and 2 = 0.25 GeV2 , while a sea quark admixture can lead to a change of up
to 4%. In NLO the variation for the former ansatz is 0.5% [2%] for the CS [MS] scheme,
while for the later one we find 5.7% [4.3%]. If one naturally assumes that for xBj = 0.36 the valence components dominate, one might conclude that the observed scaling in Ref. [75] indicates
the smallness of higher twist contributions. However, in modelling of GPDs one usually realizes
that the role of sea quarks in the CFFs is more pronounced than in DIS structure functions. Interestingly, it has been argued that indeed the leading Regge trajectory essentially contributes to
DVCS even in the valence region, in contrast to DIS [167]. We only like to point out here that
even for fixed target kinematics a detailed view on scaling breaking effects for the net contribution to CFFs is necessary and that it might be used to constrain the GPD ansatz.
So far we have considered only the scale setting prescription f = r = Q. Change obtained
by choosing another prescription within the same input scale Q0 is often considered as an estimate for the possible size of higher order corrections. Let us have a closer look at this point
of view. A change of r modifies the size of the Wilson coefficients by formally inducing a
0 proportional contribution s2 /(2)2 that is multiplied with the NLO correction to the Wil19 For 2 1 GeV2 this evolution effect might increase to become of the order of 10% for smaller values of .
This is caused by the shift of the leading meson Regge pole to the left, approaching the j = 1 pole of the anomalous
dimension, see Eq. (95).
298
Table 1
Variation (194) in percent of the non-singlet CFFs NS H induced by separate factorization and renormalization scale
changes from Q2 /2 2Q2 . Here we used the valence-like [with sea quark admixture] conformal GPD moments (172)
[(173)] and set Q2 = 4 GeV2 and 2 = 0.25 GeV2
Variation, order/
0.05
0.1
0.25
0.5
f , LO
f , NLO (CS)
f , NLO (MS)
r , NLO (CS)
r , NLO (MS)
3.7 [6.9]
0.5 [0.8]
0.3 [0.6]
5.5 [1.3]
7.9 [3.9]
0.7 [8.5]
0.7 [0.7]
0.7 [0.3]
4.9 [0.6]
7.6 [3.2]
3.8 [9.5]
0.8 [0.3]
1.0 [0.3]
2.9 [0.4]
5.7 [2.2]
8.0 [10.9]
0.1 [0.8]
0.4 [0.2]
0.3 [1.6]
2.9 [0.8]
son coefficients themselves, see Eq. (91c). A modification of f essentially corresponds to the
difference between the expanded and non-expanded version of the evolution operator which is
formally also of higher order in s . However, besides non-leading log terms, it contains also
leading ones, e.g., proportional to s2 /(2)2 ln2 (f /Q0 ). Certainly, whenever a new entry appears in the next order that is completely independent of these quantities then the rough higher
order estimate can fail.
Let us explore these estimates in more detail by employing the definitions
i =
|H(Q2 |2i = 2Q2 )| |H(Q2 |2i = Q2 /2)|

|H(Q2 |2i = Q2 )|
(194)
where i is the factorization (i = f ) [renormalization (i = r)] scale and the renormalization [factorization] scale is fixed to be Q2 . At LO we can only change the prescription for the ambiguous
factorization scale setting. As one can read off from Table 1, the scale uncertainty to LO goes
from about 4% [7%] to 8% [11%] with increasing for the valence-like [with sea quark
admixture] ansatz. Comparing with the upper right panel in Fig. 8, we realize that this uncertainty reflects nothing else but the LO evolution itself. As we have expected, these numbers are
not correlated to the perturbative corrections. At NLO the factorization scale dependence is drastically reduced and is now only about 1% or even smaller in both schemes. At this order, the
renormalization scale dependence arises and the modulus of the CFF can vary of up to 6% [8%]
by changing the scale in the CS [MS] scheme. We will come back to these numbers and compare
them with the actual NNLO corrections in the CS scheme, evaluated in Section 6.4. We should
stress here that the uncertainties with respect to the factorization scale setting are maximized
at LO. Or, in other words, revealing GPDs from experimental data in this approximation means
that one does not know at which resolution scale 2f this information was extracted. Was it Q2 ,
Q2 /2, or . . . ?
6.3.2. Flavor singlet sector
For the numerical studies in the flavor singlet sector we use the same scale prescriptions and
normalization conditions as in the preceding section. As input we alternatively take the soft
and hard gluonic anstze (178) and (179), respectively. The Wilson coefficients for the CS are
listed in Eqs. (101a) and (101b) and the evolution operator can be read off from Eq. (121). The
results for the MS scheme are collected in Section 4.3.
In Fig. 9 we plot the relative NLO corrections (187) and (189) for the moduli (up) and phase
differences (down) at the input scale Q20 = 2.5 GeV2 , again for the two extreme values of the
momentum transfer squared: 2 = 0 (left) and 2 = 1 GeV2 (right). As in the non-singlet
case, the variation of the phase differences from LO to NLO is not large and does not exceed
299
Fig. 9. The relative NLO radiative corrections (187) and (189) are plotted versus for the modulus (up) and phase (down)
of S H for 2 = 0 (left) and 2 = 1 GeV2 (right): CS (dashed) and MS (dash-dotted) scheme. Thick (thin) lines refer
to the hard (soft) gluon parameterization, where the scale setting prescriptions are the same as in Fig. 7.
0.09 rad for 2 = 1 GeV2 and is even smaller for 2 = 0. However, the NLO corrections
to the moduli have now a wider variety. This is related to the fact that at NLO the gluons enter
the hard scattering part of the DVCS amplitude the first time. For the soft gluon ansatz (thin
lines) they lead to a small decrease of the CFF of about 1020% and 1530% for the CS and
MS scheme, respectively. In contrast, if the gluon is hard (thick lines) it cancels partly the sea
quark dominated LO contribution and reduces so drastically the modulus of the CFF. Since in
this scenario the gluonic part grows with decreasing faster than the sea quark one, the modulus
of the CFF monotonously decreases, too. For our ansatz the reduction reaches 80% at very small
and large 2 . As in the non-singlet case, we observe again that the radiative corrections are a
bit smaller at 2 = 0. Another similarity is that in the CS scheme they are up to 520% smaller
than in the MS one. Although the size of perturbative corrections can be very large, the phase
differences are still small. This is caused by the fact that the phase is dominated by the leading
pole of the ansatz. Let us stress that the rather large corrections to the hard scattering part, induced
by gluons, should not be considered an argument against the applicability of perturbation theory.
We study now the evolution effects to LO and NLO approximation. The ratios (193) are
plotted in Fig. 10 versus for two fixed values Q2 = 5 GeV2 and Q2 = 25 GeV2 , where
2 = 0.25 GeV2 . The moduli (up) in the soft gluon scenario (thin lines) are relatively mildly
affected by evolution and they grow with decreasing . Both the modification of the LO (dotted)
prediction and the difference between the CS (dashed) and MS (dash-dotted) schemes are rather
small. However, we remark that the NLO contributions to the anomalous dimensions are getting
large in the small region, e.g., about 100%, which is eventually compensated by the evolved
NLO Wilson coefficients. We also found that the off-diagonal entries in the anomalous dimensions cannot be neglected anymore. Their contribution to the net result can grow from a few
percent in the large region to over 25% in the small one. This is related to the fact that the offdiagonal entries (139) contain now j = 0 poles, which arise from the LO anomalous dimensions
300
Fig. 10. Evolution of the flavor singlet CFFs S H. The moduli (up) and phases (down) are plotted for fixed Q2 = 5 GeV2
(left), Q2 = 25 GeV2 (right), and 2 = 0.25 GeV2 versus in LO (dotted) and NLO for the CS (dashed) and MS
(dash-dotted) scheme. The anstze and scale setting prescriptions are the same as in Fig. 9.
Table 2
Relative changes (194) in percent of the singlet CFFs S H within the separate variation of the factorization and renormalization scale from Q2 /22Q2 . Here we used the soft [hard] conformal GPD moments (178) [(179)] and set Q2 = 4 GeV2
and 2 = 0.25 GeV2
Variation, order/
105
103
101
0.25
0.5
f , LO
f , NLO (CS)
f , NLO (MS)
r , NLO (CS)
r , NLO (MS)
13.2 [49.5]
26.8 [67.6]
32.0 [48.8]
9.3 [46.8]
13.8 [76.5]
9.4 [29.4]
15.8 [30.1]
19.1 [40.5]
7.8 [26.8]
12.1 [42.3]
2.1 [7.3]
2.8 [0.4]
3.6 [0.2]
6.5 [11.1]
10.9 [17.7]
1.5 [1.7]
0.5 [2.2]
0.6 [2.6]
4.6 [7.0]
8.7 [12.2]
4.8 [2.6]
0.3 [1.4]
0.0 [2.0]
2.4 [3.8]
6.1 [8.0]
(108) and (109). In the hard gluon scenario the evolution effects, the NLO corrections and
the scheme dependence are quite large. The NLO corrections to the evolution are dominated by
those to the anomalous dimensions. We again observe that the NLO corrections are smaller in the
CS scheme. The scheme dependence partly arises from the NLO Wilson coefficients, yielding in
the former scheme smaller corrections, which evolve with the LO evolution operator, however,
also due to the off-diagonal part in the anomalous dimensions. Corresponding to the evolution
effects that appear in the moduli, the phase differences in the soft scenario are much smaller
than in the hard one. However, also in the latter case they cannot be considered large. Again,
we see that at least within our anstze the phase is protected from radiative corrections, since
their leading pole is in the vicinity of j = 0.
Table 2 lists the changes of the CFF that come from the variation of the factorization and
renormalization scales, see Eq. (194). The first row demonstrates that the factorization scale
variation in LO is correlated with the evolution, compare with dotted lines in Fig. 10. In the
fixed target region the evolution and the associated variation is weak. However, approaching
301
the small region, its strength is growing, in particular for the hard gluon ansatz. One would
normally expect that the scale variation is getting smaller at NLO. But this is not the case for the
small region, rather the sign is reversed and its magnitude increases. This behavior completely
differs from the one in the fixed target kinematics and it tells us that the factorization logs in
the small region are enhanced. Hence, already from these NLO findings one might wonder
whether a perturbative treatment of the evolution in the usual manner is justified. The change of
the renormalization scale yields variations of the few to ten percent in the soft gluon scenario,
and, with decreasing , to much larger ones in the hard gluon scenario. These numbers reflect
simply the size of the NLO corrections for the former and latter scenario, respectively, and do
not necessarily indicate that the perturbative expansion of the Wilson coefficients is ill-defined.
Whether this is the case or not can be only clarified by a NNLO evaluation.
6.4. Radiative corrections beyond NLO
To investigate the radiative corrections in NNLO, we use the CS scheme within the Wilson coefficients (91a)(91c) and (101a)(101c), as well as the evolution operators (116) and (121). We
stress again that the perturbative expansion is consistently done as a power series in s /2 to the
order N2 LO, where the running of the coupling is described to the same order. We again calculate
CFF H, where the model ansatz for conformal GPD moments, scale setting, and normalization
of the running coupling are spelled out above. We remind that the mixing term, appearing at
three-loop level in the anomalous dimensions is unknown. Fortunately, we found that at NLO
the mixing term in the MS scheme is small (tiny) for flavor (non-)singlet CFFs in the fixed target
kinematics. Therefore, we expect that it is justified to neglect a NNLO mixing term for these
quantities in the CS scheme. Unfortunately, this is not true for the singlet part at smaller values
of , where we observed at NLO about 30% effect at = 105 and Q2 = 100 GeV2 . Roughly
speaking, we would presume that the mixing at NNLO is given by the contribution of the diagonal NLO anomalous dimensions times (0 )s /2 0.4, which is additionally suppressed by
the initial condition. All together, we expect that for small the neglected mixing contributes to
the net result at the 10% level.
In Fig. 11 we visualize the general features of the radiative corrections for fixed target and
collider experiments up to NNLO in the parity even sector. Thereby, we employ the flavor singlet CFF S H within the soft gluon ansatz (178) at 2 = 0.25 GeV2 . As pointed out in the
preceding section, in this ansatz both the modulus, scaled with (left panel) and the phase (right
panel) are mildly affected by NLO corrections (dashed). As it can be seen, the NNLO corrections (solid) are insignificant for both of them at the input scale Q2 = 2.5 GeV2 over the whole
region. As long as we stay away from the very small region, the perturbative prediction for the
evolution is stable, starting at NLO, too. But approaching the small region, NNLO corrections
are growing in size, which already shows up in a splitting of the = 103 NNLO and NLO trajectories, smaller for the modulus and larger one for the phase. The = 105 NNLO trajectory,
compared to the NLO one, is affected by rather large corrections, which are of the same size
as the NLO ones, but with a competing overall sign. With increasing Q2 the NNLO trajectory
approaches the LO one. Such an ill behavior reflects the competition of the Bjorken limit, i.e.,
Q2 , and the high energy limit, i.e., 1/ . Indeed, the expansion parameter is rather
ln(1/ )s (Q)/2 . Since the slopes of the NLO and NNLO trajectories are getting closer at large
values of Q2 , one can easily imagine that using a larger input scale would alleviate this problem.
Indeed, it entirely originates from the resummed poles in the anomalous dimensions at j = 0
and so it is universal (process independent) and the same one that appears for parton densities.
302
Fig. 11. The modulus (left) and the phase (right) of the rescaled singlet CFF S H(, Q2 ) versus Q2 . Here the soft
gluon ansatz is used within 2 = 0.25 GeV2 to LO (dotted), NLO (dash-dotted), and NNLO (solid).
Below we will come back to this issue. The changes of the phases with respect to Q2 is for any
given smaller than 0.03 rad and according to Eq. (191) they are bound by the ratio of moduli.
In the small region the phase approaches the value /2 and so the CFF is dominated by the
imaginary part. This value is driven by both the pole, which is in our ansatz (174) in the vicinity
of j = 0, and the essential singularity of the evolution operator at j = 0, resulting from poles in
the anomalous dimensions (108) and (109).
In Fig. 12 we present a more detailed view on the radiative corrections in the flavor nonsinglet sector for the moduli (up) and the phase differences (down). The left panels show the
radiative corrections at the input scale Q20 = 2.5 GeV2 , while in the right panels we evolve the
input to the scale Q2 = 10 GeV2 . This gives a measure of the radiative corrections arising from
the evolution operator. Let us first discuss those for the moduli that arise at the input scale, i.e.,
the left upper panel. The NLO corrections for both a valence-like ansatz (thin) and one with
a sea quark admixture (thick) have for the relevant fixed target kinematics a variance of about
20%. In NNLO this is reduced to the 5% level. More precisely, the radiative corrections reduce
the CFF both without and with sea quark ansatz by about 25%. If we evolve them to 10 GeV2
then the NNLO radiative corrections further reduce by 12%, while at NLO the variance of
them is of about 13%, cf. right upper panel. The radiative corrections to the phases are already
dictated by those to the moduli. Already to NLO, they are smaller than 0.05 [0.08 ] rad for the
ansatz without [with] sea quarks and shrink further with evolution. They become tiny at NNLO,
see lower panels. Obviously, there is an improvement in the perturbative expansion, which is
roughly of the same order we estimated from the variation of the scales, see Table 1. However,
a closer look to the separate contributions, arising from different color factors, shows that there
is a cancellation between the CF2 and the 0 proportional terms [101]. The latter is negative and
about two times larger than the former, positive one, Hence both of them partly compensate each
other. Without this delicate cancellation the NNLO corrections would be almost on the 10%
level. Let us mention that the scale dependencies are now almost independent and are of the
order of 1.5% and 3% for the factorization and renormalization scale, respectively.
In Fig. 13 we display the relative radiative corrections for the flavor singlet CFF in a manner
analogous to Fig. 12. However, we evolve the quantities in questions to a scale of 100 GeV2 ,
shown in the right panels. From the left panel, where corrections are given at the input scale,
303
Fig. 12. The relative NLO (dashed) and NNLO (solid) radiative corrections to the flavor non-singlet CFF NS H in the
CS scheme at the input scale Q20 = 2.5 GeV2 (left) and Q2 = 10 GeV2 (right) and 2 = 0.25. The moduli (189) and
phase differences (187) are shown in the upper and lower panels, respectively. The results without and with sea quarks
are shown as thin and thick lines, respectively, which are indistinguishable for NNLO moduli.
Fig. 13. The relative NLO (dashed) and NNLO (solid) radiative corrections in the CS scheme are plotted versus
for the modulus (up) and phase (down) of singlet S H for 2 = 0.25 at the input scale Q20 = 2.5 GeV2 (left) and
Q2 = 100 GeV2 (right). The results for the soft and hard gluon ansatz are shown as thin and thick lines, respectively.
we certainly realize that the large negative NLO corrections (thick dashed) to the modulus in
the hard gluon scenario are shrunk to less than 10% (thick solid), in particular in the small
304
Table 3
Variation (194) of S H in percent within the change of renormalization scale from Q2 /2 to 2Q2 . Here we used the soft
[hard] conformal moments (178) [(179)] and set Q20 = Q2 = 4 GeV2 and 2 = 0.25 GeV2
Order/
105
104
103
102
101
0.25
0.5
NLO
NNLO
2.4 [24.9]
1.6 [3.4]
2.8 [21.0]
0.6 [5.6]
3.5 [18.1]
0.3 [6.5]
5.0 [15.8]
0.6 [5.7]
5.8 [10.9]
2.2 [6.7]
4.4 [7.1]
3.5 [6.9]
2.4 [3.9]
3.7 [5.9]
region. For the soft gluon case the NNLO corrections (thin solid) are reduced to 5%. However,
for 0.5, the corrections are reduced only unessentially and are still around 5% and 10% at
NNLO level. The phase differences (lower panel) are becoming tiny at NNLO. If evolution is
now switched on, our findings drastically change. For 5 102 they are stabilized for the
moduli on the level of about 3% at Q2 = 100 GeV2 . However, they start to grow with decreasing
and reach at 105 the 20% level. It is remarkable that the relative sign of the NLO and
NNLO corrections change and that they are becoming independent of the input. This behavior
is also reflected in the phase differences, which decrease to 0.02 rad. As already explained
above, this breakdown of perturbation theory stems only from the anomalous dimensions and is
thus universal, i.e., process independent.
Finally, we comment on the scale dependencies. As it has been already seen in Table 2, the
variation within the factorization scale increases in the small region with the perturbative order. To NNLO, we find for instance at = 105 a variation of 44% [105%] for the soft [hard]
gluon ansatz. This is about two [1.5] times larger than that observed at NLO, where the sign is
alternating. This simply reflects the breakdown of perturbative expansion of the evolution operator, as we have already seen. The NLO estimates of the higher order corrections, obtained
by the variation of the renormalization scale, were for the soft gluon scenario comparable to
the actual NNLO result at the input scale. However, the corresponding estimates for the hard
gluon ansatz in the small region were too pessimistic, substantially overestimating the calculated NNLO corrections. Since our input scale Q20 = 2.5 GeV2 in Table 2 was lower than the
average scale Q2 = 4 GeV2 it might be that these large estimates are partly contaminated by the
factorization logs. In Table 3 we show the renormalization scale dependence, but now for the
input scale Q20 = 4 GeV2 . Compared to Table 2 the modifications in NLO are not large. We also
realize that the renormalization scale independency is improved at NNLO level.
To summarize our findings, we saw that NLO radiative corrections are moderate in the nonsinglet case but can be rather large in the singlet sector for a hard gluon ansatz. The factorization
scale dependence, substantial at LO, becomes for kinematics of fixed target experiments small
already at NLO. However, it is getting worse in the small region. Interestingly, we also observe
a scheme dependence at NLO of the order of 10% to 20%, which entirely arises due to skewness
effects, where the radiative corrections are more pronounced in the MS scheme than in the CS
one. At NNLO we have found that the perturbative corrections are getting reasonably small
at the input scale and for the evolution in the fixed target kinematics. Both the factorization and
renormalization scale dependencies are reduced to the level of a few percent. So far these findings
suggest that the perturbative expansion is a reliable tool. However, we saw that the perturbative
expansion of the evolution breaks down in the small region. Fortunately, this breakdown is
universal, and as long as one precisely defines the scheme and the approximation, perturbation
theory can be used as a tool to relate different processes. We will demonstrate this in the next
section.
305
7. Fitting procedure of experimental data

The DVCS data measured in fixed target [63,64] and collider [6567] experiments have been
confronted in the literature with theoretical predictions, e.g., color dipole model [168], collinear
factorization approach within an aligned jet model inspired GPD ansatz [169] and the minimal
dual GPD parameterization [91], see also Ref. [68] for a first analysis within the double distribution ansatz. Certainly, confronting model anstze with DVCS data is a too rigid approach
and should be considered only as a first step towards extracting GPD parameters within a given
ansatz. A fitting procedure for DVCS data in the small and large Q2 region has been proposed
within the double log approximation of the deeply virtual Compton scattering amplitude to LO
accuracy [166]. There a pomeron inspired GPD model was employed,20 where the singlet quarks
were dynamically generated. Thus, the number of fitting parameters could be reduced to three,
namely, normalization, slope-parameter, and input scale.
We will now demonstrate that the MellinBarnes representation of the CFFs is appropriate for a more general GPD fitting procedure. Technically, we use the standard fitting routine
M INUIT [171]. This routine calls a F ORTRAN code that evaluates the CFFs from conformal GPD
moments, depending on a few fitting parameters.
7.1. Setting the scene
The DVCS amplitude interferes with the BetheHeitler bremsstrahlung one and so we have
a rich selection of observables, mainly in the interference term. The decomposition in terms of
CFFs is generally challenging. We will deal here with the easiest case, namely, the fitting of small
xBj
= 2 data for the DVCS cross section. In this kinematics the interference term, integrated
over the azimuthal angle, can be neglected and the DVCS cross section can be extracted from
the photon leptoproduction one by subtracting the BetheHeitler bremsstrahlung cross section.
In the DVCS cross section we can safely neglect terms that are kinematically suppressed by 2 :

d
W, 2 , Q2
d2

4 2 W 2 2
2
2
2
2 , 2 , Q2
|E|
+
|
H|
|H|

Q4 W 2 + Q2
4Mp2
=
Q2
2W 2 +Q2
(195)
Here we have expressed the scaling variable in terms of the photon virtuality Q2 and the
photonproton center-of-mass energy W , defined by W 2 = (P1 + q1 )2 . The leading Regge tra arises from mesons with generic intercept (0) 1/2, which is less
jectory, appearing in H,
than the intercept P (0) 1 of the pomeron dominated CFFs H and E. Thus, the squared CFF
2 can be neglected, too, since it is approximately suppressed by one power of . More
|H|
care has to be taken about the remaining combination of H and E CFFs. Taking the mean
value of 2 = 0.17 GeV2 , which has been measured by the H1 Collaboration [67] for
|2 | < 1 GeV2 , we find that the helicity flip contribution is in the 2 integrated cross section
kinematically suppressed as
2
5 102 .
4Mp2
(196)
20 A Regge pole model for the virtual Compton scattering amplitude was also used for a fitting procedure in Ref. [170].
306
Hence, in this kinematical region, it might be justified to neglect the squared CFF |E|2 . But in
the differential cross section at larger values of 2 , |E|2 might contribute to some larger extent.
Since it is not possible to separate by the present data set the H and E contributions, we simplify
our analysis by neglecting the latter one. All together, the DVCS cross section reduces in the
small -region to:

d
W2
4 2
Q2
2
2
2
2
2
(197)
,
Q
|H|
,

,
Q
W,

=
.
d2
(2W 2 + Q2 )2 W 2 + Q2
2W 2 + Q2
As we have clearly pointed out in the preceding section, the perturbative expansion of the
evolution operator is ill-defined in the small xBj
= 2 region for the (singlet) parity even sector.
Nevertheless, we have argued that this should not affect the task of relating different processes
within perturbative QCD. One can ask the question: How to resum this alternating series of
ln(1/ )s /2 terms? An answer is certainly needed for the analysis of a large amount of high
precision data, as it is the case in DIS. Here, indeed, some progress has been recently reported,
see, e.g., Ref. [172]. Concerning the situation in DVCS, the solution of the problem would certainly improve our partonic interpretation of the nucleon content, however, is rather irrelevant
for the analysis of present experimental measurements, as we will see. This problem will also
affect the forward limit of conformal GPD moments, which provide the Mellin moments of
parton densities. Since for fitting parton densities to experimental data it is also necessary to
precisely specify the procedure, e.g., definition of perturbative expansion of evolution operator,
flavor scheme, and running of the coupling, we do not rely on any of the standard parameterizations. Rather, we fit the data ourselves, within our specifications, and then compare results with
a specific choice of parton density parameterization from literature. The perturbative expansion
of the DIS structure function F2 reads:

F2 xBj , Q
1
=
2i
c+i

j
dj xBj Q2S cj s Q2 q j Q2
ci

+ Q2NS NS cj s Q2 NS qj Q2 ,
(198)
where cj = ( cj , G cj ) and NS cj are the DIS Wilson coefficients corresponding to the structure
function F2 . They can be found in Ref. [141], where we set the spin label n = j + 1. Again, we
equate the factorization and renormalization scales with the photon virtuality Q2 and consistently
combine the perturbative expansion of the Wilson coefficients with the one of the evolution
operator. The evolution of the parton density moments is governed by the evolution equations
(115) and (119). These moments are related to the conformal GPD ones in the forward kinematics
(161), e.g., cf. (A.16),

H
q j 20 = G j = 0, 2 = 0, 20 .
(199)
Hj
We will employ in this kinematics our ansatz (175) and (176), see also Ref. [173], where we
neglect for simplicity the flavor non-singlet contribution. Moreover, instead of decomposing the
singlet quark contributions in valence and sea quarks, we use the effective parameterization

B(1 (0) + j, 8)
1
1
Hj , 2 , 20 = N
+ O 2 .
2
2
B(2 (0), 8) 1 (1 )3
(m )2
(M )2
j
(200)
307
From the inspection of the standard parameterizations of parton densities, we found that the
contributions from the flavor non-singlet sector generally do not exceed the 10% level. Here it is
effectively included in the parameterization of the singlet quark contribution. Pre-fitting the data
, and that M = M 0
we found that the fits are almost insensitive to the parameters
G
G
is the preferred value. According to the common fits to vector-meson electroproduction data in
= = 0.15 GeV2 . Moreover, we neglect the j -dependence in
the small region, we set
G
the cut-off masses Mj and MjG , i.e., put M = MG = 0. As relevant fitting parameters we
thus choose
N ,
(0),
M0 ,
NG ,
G (0),
M0G .
(201)
Obviously, in fits of the DIS structure function F2 we have only four parameters, where (0)
and G (0) actually determine the powers of xBj and N and NG the normalization. Note that
although in our anstze the conformal GPD moments are normalized to Hj=1 (, 2 = 0, 20 ) =
N and HjG=1 (, 2 = 0, 20 ) = NG , we do not impose the momentum sum rule N + NG = 1
here.21 This minimal parameterization is not so well suited to provide a high quality fit to the DIS
data. This is not our goal here; rather we would like to relate the DIS data with the DVCS data,
which have much larger error bars. The 2 slope of the DVCS amplitude is controlled by the cutoff masses M0 and M0G . The normalization and the -dependence of this amplitude at 2 = 0
is as in DIS fixed by the remaining parameters N , (0), NG , G (0). Within the assumption
that the skewness parameter is negligible in the conformal GPD moments, we also loose the
possibility to control the normalization of the DVCS amplitude by the skewness dependence. As
demonstrated above by inspection of radiative corrections in different schemes, compare dashed
and dash-dotted lines in Fig. 9, the differences in normalization, caused by the -dependence, is
almost skewness independent. Within our ansatz we found a 1020% effect for the modulus of
the amplitude, which means that the cross section would differ of about 2040%. The inclusion of
skewness dependence, controlled by a corresponding parameter, will be considered somewhere
else. In our simplified model ansatz we do not explicitly adjust the normalization of the DVCS
amplitude relative to DIS one. However, for given mean value 2 the normalization of the
DVCS amplitude, integrated over 2 , is controlled also by the parameters M0 and M0G , which
determine the 2 -slope.
7.2. Lessons from fits
In Fig. 14 we confront the outcome of a simultaneous 2 -fit to the DVCS (39 data points)
and DIS (85 data points) data in the CS scheme to NNLO accuracy. We equated the factorization and renormalization scale with the photon virtuality, used the conformal GPD moment
anstze at the input scale 20 = 4 GeV2 , fixed the number of flavors to nf = 4, and used for
s (2.5 GeV2 )/ = 0.0976.22 In the upper left panel we display the fit to the H1 [67] and ZEUS
[66] DVCS data versus 2 for fixed Q2 and W , while the upper right and lower left ones show
the DVCS cross section, integrated over |2 | < 1 GeV2 , versus Q2 and W dependence, where
21 Note that the whole momentum fraction region, i.e., 0 x 1, contributes to this rule, however, our fits constrain
only the small xBj region, where xBj x 1. For technical reasons, however, we assume a certain large x or j behavior.
A modification of this behavior, which is only weakly constrained in our fits, might be used to restore the momentum
sum rule. An improved treatment will be given somewhere else.
22 Employing the standard procedure for the NNLO running of coupling, we find that this value corresponds to
2 ) = 0.114, at the standard reference scale M = 91.18 GeV.
s (MZ
Z
308
Fig. 14. Simultaneous fit to the DVCS and DIS data in the CS scheme to NNLO. Upper left panel DVCS cross section
for Q2 = 4 GeV2 and W = 71 GeV (circles, dashed) as well as Q2 = 8 GeV2 and W = 82 GeV (squares, solid)
[67]. Upper right panel DVCS cross section (|2 | < 1 GeV2 ) versus Q2 for W = 82 GeV (H1, circles, dashed) and
W = 89 GeV (ZEUS, triangles, dash-dotted) [66]. Lower left panel DVCS cross section versus W Q2 = 4 GeV2 (H1,
circles, dashed), Q2 = 8 GeV2 (H1, squares, solid), and Q2 = 9.6 GeV2 (ZEUS, triangles, dash-dotted). Lower right
panel shows F2 (xBj , Q2 ) versus Q2 for xBj = {8 103 , 3.2 103 , 1.3 103 , 5 104 } [174].
Table 4
Parameters extracted from a simultaneous fit to the DVCS cross section and DIS structure function F2 , where s (MZ ),
= = 0.15 GeV2 , and M = M = 0 are fixed and 2 = 4 GeV2
G
G
0
Order (scheme)
s (MZ )
(0)
2
M
NG
G (0)
2
MG
2 /d.o.f.
2 2
LO
NLO (MS)
NLO (CS)
NNLO (CS)
0.130
0.116
0.116
0.114
0.157
0.172
0.167
0.167
1.17
1.14
1.14
1.14
0.228
1.93
1.34
1.17
0.527
0.472
0.535
0.571
1.25
1.08
1.09
1.07
0.263
4.45
1.59
1.39
100
109
95
91
0.85
0.92
0.80
0.77
38.5
4.2
2.2
2.2
the remaining variable W or Q2 is fixed. The fit to the DIS H1 data [174] is plotted in the lower
right panel, where for clarity not all points are displayed.23 As one realizes by eye inspection,
the normalization, scale- and 2 -dependency are separately well described.
From Table 4 one can see that the quality of these simultaneous fits for the 124 data points is
satisfying. For instance, for the NLO fit in CS scheme, where now s (2.5 GeV2 )/ = 0.1036,
23 Not shown, but used in fits, are H1 F data for x = {1.3 102 , 5 103 , 2 103 , 8 104 , 3.2 104 , 2
2
Bj
104 , 1.3 104 , 8 105 }.
309
we get 2 = 95, i.e., 2 /d.o.f. = 0.8. Taking into account the NNLO order corrections, the quality of the fit slightly improves, i.e., 2 /d.o.f. = 0.77. As one can realize by comparison of the
last two rows in Table 4, the resulting parameters remain stable. The largest modification appears
in the cut-off masses, which reduce from NLO to NNLO by about 13%. Thereby, contribution to
2 coming from the eight data points of the differential cross section (upper left panel in Fig. 14),
2 = 2.2 (last column in Table 4), indicates an equally good fit to the 2 -slope in NLO
denoted
2
and NNLO.
To understand better the double role of the cut-off masses in the fitting procedure, let us compare the LO parameters with the NLO ones, where their changes are much more pronounced.
Although in LO the fit to all data yields seemingly reasonable 2 /d.o.f. = 0.85, the 2 -slope
2 M 2 0.25 GeV2 ) turns out to be
(determined by the small squared cut-off masses M
G
2
2
30 GeV , which is incompatible with 6 GeV indicated by the data [67]. Or, in other
2 = 38.5 is unacceptably large. Therefore, we conclude that
words, to leading order accuracy
2
the relative normalization between the DIS structure function and the integrated DVCS cross
section is correctly reproduced in the fits by reducing the latter one by forcing the steeper 2 dependence. This also implies that at 2min 0 the normalization of the differential DVCS cross
section and the structure function F2 cannot be simultaneously described within our ansatz.
A separate fit to the DVCS data yields a slope that is getting compatible with the measured
one; however, the overall normalization is now deteriorated and so the quality of the fit is worse,
namely, 2 /d.o.f. = 4.8. Certainly, this is related to the fact that we have no control over the
skewness dependence of the conformal GPD moments.24 Therefore, non-trivial skewness dependence should be introduced in such a way to make |H|, and consequently the normalization
of the DVCS cross section, smaller. We remind, however, that the inclusion of the skewness
dependence within the spectral representation of GPDs is usually done in such a way that the
skewness effect leads to an increase of |H|.
Let us finally compare the NLO fits in the CS and MS schemes. The fit in the latter scheme
is compared to the former one a bit off. We find 2 /d.o.f. = 0.92, while in the CS scheme
2 /d.o.f. = 0.8. Also here the largest changes appear in the cut-off masses, in particular the gluon
one. However, the quality of the 2 -fit to the differential cross section is acceptable, namely,
2 = 4.2, compared to 2 = 2.2 in the CS scheme. As discussed in Section 6.3 above, these

2
2
schemes differ only in the skewness dependence of the conformal GPD moments. Instead of
adjusting the normalization by changing the skewness dependence, it is done within our ansatz by
increasing the slope. Hence, we must conclude that a more precise extraction of GPD parameters
requires the inclusion of the skewness effect for 2 = 0.
7.3. Comparison and partonic interpretation of the results
We would like now to confront our findings with the parton densities, as obtained from global
DIS fits. In Fig. 15, we plot the singlet quark and gluon distributions
c+i

1
q
H
2
x G
dj x j G j = 0, 2 = 0, 20 ,
x, 0 =
q
Hj
2i
(202)
ci
24 As a side effect, arising from this ill-defined fitting task, we observed that the central value of the resulting parameters,
in particular of the cut-off masses, are getting sensitive to the accuracy of numerics and they can vary inside the error
bands. In NLO and NNLO fits, by increasing the numerical accuracy, we observed a variation of the central values only
on the per mil level.
310
Fig. 15. The parton densities are shown for quarks (left) and gluons (right) at the input scale 20 = 4 GeV2 . The meaning
of the lines are the same as before: LO (dotted), NLO for MS (dash-dotted) and CS (dashed), as well as NNLO (solid), the
last two being indistinguishable in the quark case. The bands show Alekhins NLO parameterization with errors [175].
respectively, to LO (dotted), NLO for the MS (dash-dotted) and CS (dashed) as well as to NNLO
(solid) at the input scale 20 = 4 GeV2 . Note that the difference between the two schemes arises
purely due to the skewness dependence, which affects only the DVCS fit. As argued above, we do
not expect the outcome of, e.g., our simultaneous DVCS and DIS NLO fit, to coincide with any of
the standard parameterizations for parton densities. Indeed, we explicitly saw this for the parton
density fits of the H1 and ZEUS Collaborations. However, it turns out that the flavor singlet quark
parton density agrees very well with the parameterization of Alekhin [175], which is plotted
with error bands. Here the difference between the both schemes and NNLO order corrections are
rather small. As is well known, the gluon distribution is much less constrained by a pure DIS fit.
Here our central values lie outside the error band of the Alekhin parameterization. However, our
error band of the NLO fit in the CS scheme would overlap with Alekhins one. Note that we have
used the same settings for s as Alekhin, however, the procedures for the fits is slightly different.
As said above, we have neglected the flavor non-singlet contribution, took a fixed nf = 4 scheme
for both the evolution of the running coupling and the conformal GPD moments, and assumed a
generic j -dependence of the conformal GPD moments at the input scale. However, as we realize
a posteriori, these simplifications are justified within the error bands.
Although GPDs are amplitudes, it was shown that for = 0 they have a probabilistic interpretation in the infinite momentum frame within the parton picture [5356]. More precisely, their
Fourier transform with respect to the transversal degrees of freedom

i b

d 2

2 ,
e H x, = 0, 2 =
H (x, b) =
(203)
2
(2)
can be interpreted as parton densities that besides the longitudinal momentum fraction depend
The averaged squared distance of a parton from the center of the
also on the impact parameter b.
nucleon might be expressed by the slope of the corresponding GPD

b2 x, Q2 =
Q2 )

d b b2 H (x, b,
= 4B x, Q2 ,
2
Q )
d b H (x, b,
(204)
311
Fig. 16. The GPD slope, defined in Eq. (205), is shown for quarks (left) and gluons (right) at the input scale Q20 = 4 GeV2 .
The meaning of the lines are the same as in Fig. 15 and the band shows the errors for the NLO fit in the CS scheme.
where the slope is defined as

B x, Q

d
2
2
=
ln H x, = 0, , Q
.
d2
2 =0
(205)
In Fig. 16 we display the resulting slope for the flavor singlet quark combination (left) and gluon
(right) GPD, respectively. Here we exclude the LO result, which within our ansatz fails to describe the 2 dependence of the differentiated DVCS cross section. As observed before [166],
within the pomeron inspired ansatz, the slope parameter slightly increases with decreasing x
and is smaller for gluons than for quarks. As explained above, the difference between the CS
(dashed) and MS (dash-dotted) scheme is in fact induced by our ansatz being rigid with respect
to skewness. We would also expect that a more flexible ansatz would reduce the differences between NLO (dashed) and NNLO (solid) results. Compared to the LO analysis of Ref. [166], we
find, e.g., for x = 103 and Q2 = 4 GeV2 , that the central value of the slope to NLO accuracy
in the CS scheme is for quarks (gluons) about 35% (55%) smaller. In particular, for gluons is
our NLO analysis now compatible with the slope extracted from J / photo- or electroproduction, see, e.g., Refs. [176,177,30]. This process is dominated by the two gluon exchange and its
measured 2 slope of the differential cross section, which is nearly Q2 independent, is about
5 GeV2 . This is consistent with two times the value we extracted from our DVCS analysis in
the CS scheme, see right panel in Fig. 16. This slope yields the spatial averaged size squared for
gluons of
2

+0.08
2
2
b gluon x = 103 , Q2 = 4 GeV2 = 0.30+0.07
0.04 fm 0.330.04 fm
(206)
to NLO (NNLO) accuracy. The central value is about 20% (10%) smaller than the one of
Refs. [178,179], however, still compatible within errors. We consider this a further wink that
in the GPD phenomenology perturbative corrections should be taken into account.
312
8. Conclusions
In this article we have derived the leading twist-two Compton form factors represented as
MellinBarnes integrals in terms of conformal GPD moments with their evolution included.
Thereby, we have used the standard framework, known from DIS, which is based on the local
operator product expansion and dispersion relation techniques and we have confirmed known
results, derived using other methods. For instance, this representation can be obtained in a
straightforward way from the momentum fraction representation, which also allows other GPD
related processes, e.g., the hard electroproduction of mesons, to be represented by a Mellin
Barnes integral. Although the original motivation for this representation, the solution of the LO
evolution equation, is tied to conformal symmetry, we explicitly showed here that there is no
problem in using this MellinBarnes integral representation beyond LO within the standard MS
scheme, in which conformal symmetry is not explicitly manifested. This opens a new road for the
global analysis of experimental data within the perturbative GPD formalism to NLO accuracy.
We have combined the MellinBarnes representation with conformal symmetry predictions,
which would hold true if there existed a nontrivial fixed-point in QCD. Formally,in the perturbative sector one sets the function to zero. Unfortunately, the return to the real = 0 world is
plagued by an ambiguity, which we have shifted to the evolution equation where, consequently,
it induces at NNLO a mixing of conformal GPD moments. We have argued that this mixing
can be safely neglected in the fixed target kinematics, but will, however, influence the radiative
corrections for small values of Bjorken-like scaling parameter .
The outcome of our numerical analysis can be summarized as follows. To leading-order accuracy, the scale setting prescription is most problematic, and this ambiguity directly translates
into the ignorance of the scale that enters the (moments of) GPD. For fixed target kinematics,
this problem diminishes already in NLO. There radiative corrections are moderate in the flavor
non-singlet sector, whereas can be larger in the singlet one, due to the appearance of gluons. With
increasing experimental precision, one might also employ evolution to constrain the conformal
GPD anstze. To test the reliability of perturbation theory, we have studied NNLO corrections
and found that they are indeed small, 5% or even less.
In contrast to fixed target kinematics, in the small xBj region our studies have clarified the situation with perturbative expansion of the evolution operator being ill-defined in the kinematics of
interest, whereas the perturbative expansion of Wilson coefficients presents no difficulties. As in
DIS, this bad behavior arises from ln(1/ )s terms, induced by the j = 0 poles of the anomalous
dimensions in the parity even sector, where j + 2 is the conformal spin. Nevertheless, these poles
are universal and, as we have demonstrated, the large fluctuation of the scaling prediction within
the considered order does not influence the quality of fits, and, in particular, the possibility of
relating DVCS and DIS data. Hence, the problem of treatment or resummation of these large
corrections is relevant primarily to our partonic interpretation of the nucleon content. As long as
we precisely define the treatment of the evolution operator, perturbative QCD can be employed
as a tool for analyzing data in the small xBj region. According to this, the reported large-scale
dependence in the hard vector-meson electroproduction might not necessarily lead to the conclusion that this process cannot be analyzed within the GPD formalism. We also note that progress
in the resummation of ln(1/xBj )s corrections in DIS has recently been achieved by two groups,
see, e.g., Ref. [172].
Although we have studied only the parity even sector, one can imagine what happens in the
parity odd one. The analytic properties of Wilson coefficients in both sectors are similar and
the parity odd anomalous dimensions do not suffer from poles at j = 0. Hence we expect that
313
the radiative corrections have similar features in the fixed target kinematics as in the parity even
case and do not grow large in the small xBj region, which anyway can hardly be accessed in
experiments.
So far in the literature experimental data have been analyzed by comparing one or another
ansatz with measurements. In particular, in the small xBj region, where the Bjorken and the highenergy limit are competitors, the success of such an approach is based on trial and error. Indeed,
as in the unpolarized DIS case, data can only be successfully fitted by a kind of fine tuning
procedure of the relevant parameters within a clearly defined, however, ambiguous scheme. The
most important advantage of the MellinBarnes representation, as demonstrated in this paper
up to NNLO, is that it is adequate for building the fitting procedure of hard photon and meson
electroproduction data. Certainly, a great deal of work, e.g., resummation of SO(3) partial waves,
study of skewness dependence, and choosing an appropriate set of parameters, must be done to
release the full power of this framework.
To conclude, we would suggest the analysis of experimental data to NLO accuracy within
the standard MS scheme. This drastically reduces the theoretical uncertainties, present in the LO
approximation, particularly in relating different processes, e.g., hard photon and meson electroproduction. The MellinBarnes representation allows us to write flexible fitting routines, which
are fast enough and numerically stable, and thus it provides a reliable and systematic procedure
to for analyzing experimental data. This we consider the main step towards a global analysis of
experimental data, related to GPDs.
Note added
After the manuscript was submitted to hep-ph, the evaluation of the subtraction constant has
been also given in Eq. (22) of Ref. [180]. Our formula (47), approximated to LO, expressed
in momentum fraction space, and written down for CFF H, yields the aforementioned result
and contains a prescription for treating the appearing divergencies. Related work has been also
presented in Refs. [181,182].
Acknowledgements
This project has been supported by the German Research Foundation (DFG), Croatian Ministry of Science, Education and Sport under the contracts Nos. 119-0982930-1016 and 0980982930-2864, US National Science Foundation under grant No. PHY0456520, and EU project
Joint Research Activity 5: GPDs. The authors would like to thank the theory group at the University of Regensburg for its warm hospitality. D.M. is grateful for invitations at the Thomas Jefferson National Accelerator Facility and Service de Physique Nuclaire (Saclay). For both clarifying
and inspiring discussions we are indebted to I. Anikin, H. Avakian, M. Diehl, M. Garon,
V. Guzey, D. Ivanov, M. Kirch, A. Manashov, B. Pire, A. Schfer, L. Schoeffel, P. Schweigert,
L. Szymanowski, O.V. Teryaev, S. Wallon, and C. Weiss.
Appendix A. Normalization of Wilson coefficients and anomalous dimensions
Here we establish the normalization of the Wilson coefficients of the OPE (28) in the forward
kinematics

a I
Cj = 0, Q2 /2 = 1, s = a cjI (s ),
(A.1)
314
where a cjV (s ) and a cjA (s ) are the Wilson coefficients that appear in the perturbative description
of the DIS unpolarized and polarized structure function F1 and g1 , respectively, for = Q.
The hadronic DIS tensor is related to the Compton scattering tensor (1) by the optical theorem
1
(A.2)
m T (q, P = 2p, = 0),
2
where p = P1 = P2 . We recall that the DIS hadronic tensor for a longitudinally polarized target
might be written in terms of the structure functions F1 (xBj , Q2 ), FL (xBj , Q2 ), and g1 (xBj , Q2 ):
W =

p p

1
T
F1 xBj , Q2 +
FL xBj , Q2 iqp
g1 xBj , Q2 ,
W = g
pq
pq
(A.3)
where xBj = = Q2 /2M and = 1 is the polarization of the target with respect to the p
direction. Both F1 and g1 structure functions are expressed to leading power accuracy as

1
F1
x, Q2 =
g1
2
1
Q2a
a=u,u,...,G
dy
y
ac

a c
a

x Q2
q
, s ()
,
y, 2 , (A.4)
a q
y 2
where the flavor sum runs over quarks, anti-quarks, and gluons. Here ()ca are the partonic cross
sections, depending on the factorization scale . The polarized and unpolarized (anti-)parton
distributions are denoted as qa and qa [()qa = ()qa ], respectively, and Qa are the fractional
electrical charges
2
1
1
Q2a for gluons,
Qd = Qs = ,
Q2G =
Qu = Qc = ,
(A.5)
3
3
nf
a=u,d,...
while the coefficient functions are normalized as follows

d, . . .}
()a c = (1 x) + O(s ) for a = {u, u,
and
()G c = O(s ).
(A.6)
Since the evolution will yield a mixing of the quark singlet and gluon parton densities, it
is appropriate to introduce a group theoretical decomposition in flavor non-singlet and singlet
densities. For three active light quarks these combinations are
()NS q = 2()u q + 2()u q ()d q ()d q ()s q ()s q,
() q = ()u q + ()u q + ()d q + ()d q + ()s q + ()s q,
(A.7)
while for four active quarks they read:

()NS q = ()u q + ()u q ()d q ()d q ()s q ()s q + ()c q + ()c q,
(A.8)
() q = ()u q + ()u q + ()d q + ()d q + ()s q + ()s q + ()c q + ()c q.
The corresponding squared charge factors are defined in Eq. (27). In the singlet case the parton
densities with a = {+, } are eigenvectors of the evolution equation, i.e., all distributions satisfy
the evolution equation
2

d
()a q x, 2
d2

1

dy
x
=
()a P
, s () ()a q y, 2 for a = {NS, +, }
y
y
x
(A.9)
315
where ()a P are the splitting kernels. The rotation to this basis results into the replacement of
the flavor sum in (A.4).

.
(A.10)
a=u,u,...,G
a=NS,+,
Using the optical theorem (A.2) and the parameterization of the Compton tensor (3) and the
DIS hadronic tensor (A.3) we find that the structure functions are related to the CFFs by

1
F1
H
(A.11)
.
m ( = xBj i0, . . .)
xBj , Q2 =
H
g1
2
=0
Forming Mellin moments, we find by means of Eq. (36)

1
dx x
0

F1
x, Q2
g1

a V

aH
1
c
2
s (Q) a
= 0, 2 = 0, Q2 ,
Qa a A
=
c
H
2
j
j
(A.12)
a=NS,
and, on the other hand, in terms the parton densities and coefficient moments of Eq. (A.4)

1
dx x
0

F1
x, Q2
g1

a

aq

2
1
c
Q2a
(Q)
Q ,
s
a
a
c j
q j
2
(A.13)
a=NS,
with definitions
a
() qj Q
1
=
dx x () q x, Q ,
1
() cj (s ) =
a
dx x j ()a c(x, 1, s ).
0
From this we obtain the equalities

a
cj (s )a qj Q2 = a cjV (s )a Hj = 0, 2 = 0, Q2 ,

a cj (s )a qj Q2 = a cjA (s )a H j = 0, 2 = 0, Q2 .
(A.14)
(A.15)
The normalization of the conformal operators (21) and (23) is chosen so that their reduced matrix elements in the forward kinematics reduce to those used in DIS. Taking into account that
Eqs. (31) and (32) relate the forward matrix elements to the conformal moments Hj and H j , we
find, for instance, in the unpolarized case for odd j
a

Hj =0 =
1
j +1
P+
p|a OjV |p =
1
j +1
2p+
j
p| a (0)+ i D + a (0)|p = a qj + a qj
for a = {u, d, s},

j
1
1
a
a
Hj =0 = j +1 p|a OjV |p = j +1 p|(0)

+ i D + (0)|p = a qj
P+
2p+
for a = {NS, },
316

G

Hj =0 =
1
j +1
P+
p|G OjV |p =
1
j +1
2p+
p|G+ (0)(i D + )j 1 G+ (0)|p = G qj .
(A.16)
Hence, from the identity (A.15) we find the desired normalization for the Wilson coefficients
a V
cj (s ) = a cj (s )
and
a A
cj (s ) = a cj (s ).
Taking the forward limit in the evolution equation (81) and comparing it with the moments of
Eq. (A.9), we also establish from the identities (A.16) the definition of the anomalous dimensions
in terms of moments of the splitting kernel
1
dx x
ja
1
P (x, s ) = a jV (s ),
2
1
dx x j a P (x, s ) =
1a A
(s ).
2 j
(A.17)
Note that within our definitions, the momentum sum rule for j = 1 is established by
1
1
2
P2 , S2 | O1V |P1 , S1 + 2 P2 , S2 |G O1V |P1 , S1 = 2 P2 , S2 |++ |P1 , S1 ,
2
P+
P+
P+
(A.18)
which results in the scale independent expectation value of the energymomentum tensor
i
a
+ D + + Ga
++ =
(A.19)
+ G+ ,
2
projected on the plus light-cone components. In the forward case the sum rule (A.18) reduces to
q1 () + G q1 () =
1
p|++ |p 1.
2(p+ )2
(A.20)
As a consequence of the scale independence the anomalous dimensions satisfy the relation:
1V (s ) + G 1V (s ) = GG 1V (s ) + G 1V (s ) = 0.
(A.21)
Appendix B. Bases in the flavor singlet sector

The transformation of the gluon and quark singlet conformal operators to the basis of the
ones is defined in Eq. (24). In the following we drop the superscript I {V, A}. By making use
of the evolution equation it can be easily shown that the two-dimensional anomalous dimension
matrix is diagonalized by the rotation

+
d
j
0
j G j
1
1
)
U
=
U
(U
(B.1)
j
j
j (U j ) .
GG
G
0
d
j
j
j
d
U |=0 = 0.
Let us suppose that for a given scale 0 the inhomogeneous term vanishes, i.e., d
At this reference point the eigenvalues of the anomalous dimension matrix are

"

2
1 !
GG
j =
j + j j GG j + 4G j G j ,
(B.2)
2
and the rotation matrix reads
GG
j
j
1

G
j
s (0 ) .
U j (0 , 0 ) = +
(B.3)
j
j
1
G
j
317
The diagonalization of the evolution equation at an arbitrary scale can be now obtained by the
use of evolution
U j (, 0 ) = E j (, 0 )U j (0 , 0 )Ej1 (, 0 ),
where E 1 is the inverse of the evolution operator (120) and

+

exp{ 0 d
j (s ( ))}
E j (, 0 ) =
0
exp{ 0
(B.4)
d

j (s ( ))}
(B.5)
is the evolution operator in the {+, } basis.
Appendix C. Momentum fraction representation versus conformal moments

In the momentum fraction representation the Compton form factors are represented as convolution of the coefficient function with the corresponding GPD. In the singlet sector in which
quark ( O) and gluon (G O) operators mix under renormalization, we might introduce the vector
notation:
S

F , 2 , Q2 =
1

dx
C x/, Q2 /2 , s ()| F x, = , 2 , 2 .
Here the column vector

F
F= G
, F = {H, E, H , E}
F
(C.1)
(C.2)
contains the GPDs, and the row one, defined as C = ( C, (1/ )G C), consists of the hard scattering part that to LO accuracy reads

1
1
2
2
C x/, Q / , s ()| =
, 0 + O(s ).
(C.3)
x i
We remark that the dependence in C and G C enters only via the ratio x/ . Note also that the
u-channel contribution in the quark entry (C.3) has been reabsorbed into the symmetrized quark
singlet distribution

q
(C.4)
F x, , 2 , 2 =
F x, , 2 , 2 q F x, , 2 , 2 .
q=u,d,...
Here the second term in the square brackets with (+)-sign for H, E (H , E)-type
GPDs is
for x > related to the s-channel exchange of an anti-quark. The gluon GPDs have definite
symmetry property under the exchange of x x: G H and G E are even, while G H and G E are
odd.
C.1. Evaluation of conformal moments
The convolution formula (C.1) has already at LO the disadvantage that it contains a singularity at the cross-over point between the central region ( x ) and the outer region
( x 1), i.e., for x = = . Its treatment is defined by the i prescription, coming from
the Feynman propagator. The GPD is considered smooth at this point, but will generally not be
318
holomorphic [88]. The fact that both regions are dual to each other, up to a so-called D-term
contribution [120], makes the numerical treatment even more complicated. This motivated our
development of a more suitable formalism in [98].
To make contact with the conformal OPE, we expand the hard-scattering amplitude in terms
of Gegenbauer polynomials with indices 3/2 and 5/2 for quarks and gluons, respectively, and
introduce the conformal GPD moments, which formally leads to
S

F , 2 , Q2 = 2
j 1 C j Q2 /2 , s () F j , 2 , 2 .
(C.5)
j =0
The expansion coefficients C j can be calculated by the projection:

2j +1 (j + 5/2) 1
C j Q2 /2 , s () =
(3/2)(j + 4) 2
(j
1

dx C x, Q2 /2 , s ()| = 1
1
3/2
+ 3)[1 x 2 ]Cj
0
5/2
3[1 x 2 ]2 Cj 1
(x).
(C.6)
Note that we have here rescaled the integration variable with respect to and that the integral runs
only over the rescaled central region. The conformal moments of the singlet GPDs are defined as

F j , ,
(3/2)(j + 1) 1
= j
2 (j + 3/2) 2
1
1
dx j 1
3/2
Cj
0

F x, , 2 , 2 .

x
5/2
(3/j )Cj 1
0
(C.7)
Here j is an odd (even) non-negative integer for the (axial-)vector case.

In order to make use of the NLO MS results given in the momentum fraction representation, for quark part we determine the corresponding conformal moments using the results from
Ref. [137]. In particular, the conformal moments (128) for the axial-vector case can be read off
from Eq. (3.32) in Ref. [137], where the normalization factor (2j + 3)/(j + 1)(j + 2) must be
removed and the remaining expression multiplied by the color factor CF . The conformal moments (127) for the vector case one can easily recover from the hard-scattering amplitude given
in Ref. [154], Eqs. (14) and (15). Thereby one has only to evaluate the difference between the
vector and axial-vector results

CF
C V(1) x, Q/2 C A(1) x, Q/2 =
ln(1 x).
x
(C.8)
Employing Table 8 in Appendix C of Ref. [137], we find, after removing the normalization factor
2(2j + 3)/(j + 1)(j + 2), that the difference for the conformal moments is CF /(j + 1)(j + 2).
C.2. Conformal moments for the gluon part
Following the method for computing moments with respect to conformal partial waves (with
the index k) in the quark sector that was explained in detail in Appendix C of Ref. [137], we here
give the necessary results for the gluon sector.
319
Table 5
The conformal moments of some relevant functions
(5/2)
1
5/2
12Nk1
1
1x
k1
(5/2)
ln(1x)
1x k1
1
5/2 [1 S1 (k 1) S1 (k + 3)]
12Nk1
1 [ 1 1 + 1 1 ]
5/2
k+1
k+2
k+3
12Nk1 k
1
2
k
5/2 [(1) (k+1)(k+2) ]
12Nk1
(5/2)
ln(1x)
k1
x
(5/2)
k1
ln(1x)
2
x
2
(5/2)
ln (1x)
k1
2
2
1
[2S1 (k) + 2S1 (k + 2) 3]
5/2
12Nk1 (k+1)(k+2)
We introduce the notation similar to the one used in Ref. [137]

(5/2)
G(x) k1
1
dx G(x)
x 2 (1 x)2
5/2
Nk1
5/2
Ck1 (2x 1),
(C.9)
with
k(k + 3)
5/2
Nk1 =
3/2
Nk
3/2
Nk
(k + 1)(k + 2)
.
4(2k + 3)
(C.10)
(5/2)
(5/2)
It follows trivially that G(1 x)k1 = (1)k1 G(x)k1 , and one can easily make the correspondence to definitions of conformal moments given in (C.6):
G
Ck =
(5/2)
2k+1 (k + 5/2)
5/2
48Nk1 G C(2x 1) k1 .
(3/2)(k + 4)
(C.11)
It is convenient to use the following expression for the Gegenbauer polynomials:

x 2 (1 x)2
5/2
Nk1
= (1)
5/2
Ck1 (2x 1)
+ 3)
(1)i
k(k + 1)
k1 12(2k
k+1
i=0
k+1
i

k+i +1
x i+2 .
i +2
(C.12)
The evaluation of the conformal moments, i.e., in our case the evaluation of the expressions
'
(

k+1

g(x) (5/2)
k+i +1
k1 12(2k + 3)
i k+1
= (1)
(1)
i
i +2
x k1
k(k + 1)
i=0
1
x i+1 g(x),
(C.13)
and
'
g(x)
1x
((5/2)
k1
12(2k + 3)
=
(1)i
k(k + 1)
k+1
i=0
k+1
i
k+i +1
i +2
1
x i+1 g(1 x),
0
(C.14)
320
consists then in calculating the Mellin moments and performing the summation. The Mellin
moments for the functions we encounter and most of the non-trivial sums we are left with can be
found in [150].
In Table 5 we summarize the conformal moments of the functions relevant to our calculation.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
A. Mueller (Ed.), Perturbative Quantum Chromodynamics, World Scientific, Singapore, 1989.

S.J. Brodsky, M. Burkardt, I. Schmidt, Nucl. Phys. B 441 (1995) 197, hep-ph/9401328.
A.D. Martin, W.J. Stirling, R.S. Thorne, Phys. Lett. B 636 (2006) 259, hep-ph/0603143.
S. Alekhin, JETP Lett. 82 (2005) 628, hep-ph/0508248.
J. Pumplin, et al., JHEP 0207 (2002) 012, hep-ph/0201195.
M. Glck, E. Reya, A. Vogt, Eur. Phys. J. C 5 (1998) 461, hep-ph/9806404.
E. Leader, A.V. Sidorov, D.B. Stamenov, Phys. Rev. D 73 (2006) 034023, hep-ph/0512114.
J. Blmlein, H. Bottcher, Nucl. Phys. B 636 (2002) 225, hep-ph/0203155.
M. Gluck, E. Reya, M. Stratmann, W. Vogelsang, Phys. Rev. D 63 (2001) 094005, hep-ph/0011215.
T. Gehrmann, W.J. Stirling, Phys. Rev. D 53 (1996) 6100, hep-ph/9512406.
W.T. Giele, S. Keller, Phys. Rev. D 58 (1998) 094023, hep-ph/9803393.
S. Forte, L. Garrido, J.I. Latorre, A. Piccione, JHEP 0205 (2002) 062, hep-ph/0204232.
J.C. Rojo, The Neural network approach to parton distribution functions, hep-ph/0607122, PhD thesis, 2006.
G. Lepage, S. Brodsky, Phys. Rev. D 22 (1980) 2157.
S. Brodsky, G. Lepage, Phys. Rev. D 24 (1981) 1808.
A. Efremov, A. Radyushkin, Phys. Lett. B 94 (1980) 245.
A. Duncan, A. Mueller, Phys. Rev. D 21 (1980) 1636.
V. Chernyak, A. Zhitnitsky, Nucl. Phys. D 246 (1984) 52.
N. Isgur, C.L. Smith, Phys. Rev. Lett. 52 (1984) 1080.
N. Isgur, C.L. Smith, Phys. Lett. B 217 (1989) 535.
A. Radyushkin, Acta Phys. Pol. B 15 (1984) 403.
I.I. Balitsky, V.M. Braun, A.V. Kolesnichenko, Nucl. Phys. B 312 (1989) 509.
V. Braun, I. Filyanov, Z. Phys. C 48 (1990) 239.
D. Mller, D. Robaschik, B. Geyer, F.-M. Dittes, J. Horeji, Fortschr. Phys. 42 (1994) 101, hep-ph/9812448.
F.-M. Dittes, B. Geyer, D. Mller, D. Robaschik, J. Horeji, Phys. Lett. B 209 (1988) 325.
X. Ji, Phys. Rev. Lett. 78 (1997) 610, hep-ph/9603249.
X. Ji, Phys. Rev. D 55 (1997) 7114, hep-ph/9609381.
A. Radyushkin, Phys. Lett. B 380 (1996) 417, hep-ph/9604317.
S. Brodsky, L. Frankfurt, J. Gunion, A. Mueller, M. Strikman, Phys. Rev. D 50 (1994) 3134, hep-ph/9402283.
ZEUS Collaboration, J. Breitweg, et al., Eur. Phys. J. C 6 (1999) 603, hep-ex/9808020.
H1 Collaboration, C. Adloff, et al., Eur. Phys. J. C 13 (2000) 371, hep-ex/9902019.
ZEUS Collaboration, J. Breitweg, et al., Eur. Phys. J. C 12 (2000) 393, hep-ex/9908026.
A. Radyushkin, Phys. Lett. B 385 (1996) 333, hep-ph/9605431.
J. Collins, L. Frankfurt, M. Strikman, Phys. Rev. D 56 (1997) 2982, hep-ph/9611433.
LHPC Collaboration, P. Hgler, et al., Phys. Rev. D 68 (2003) 034505, hep-lat/0304018.
LHPC Collaboration, P. Hgler, et al., Eur. Phys. J. A 24S1 (2005) 29, hep-ph/0410017.
LHPC Collaboration, SESAM Collaboration, P. Hgler, et al., Phys. Rev. Lett. 93 (2004) 112001, hep-lat/0312014.
QCDSF Collaboration, M. Gckeler, et al., Phys. Rev. Lett. 92 (2004) 042002, hep-ph/0304249.
QCDSF Collaboration, M. Gckeler, et al., Phys. Lett. B 627 (2005) 113, hep-lat/0507001.
R.G. Edwards, et al., PoS LAT2006 (2006) 121, hep-lat/0610007.
X. Ji, W. Melnitchouk, X. Song, Phys. Rev. D 56 (1997) 5511, hep-ph/9702379.
M. Diehl, T. Feldmann, R. Jakob, P. Kroll, Nucl. Phys. B 596 (2001) 33, hep-ph/0009255.
S.J. Brodsky, M. Diehl, D.S. Hwang, Nucl. Phys. B 596 (2001) 99, hep-ph/0009254.
S. Scopetta, V. Vento, Eur. Phys. J. A 16 (2003) 527, hep-ph/0201265.
S. Boffi, B. Pasquini, M. Traini, Nucl. Phys. B 649 (2003) 243, hep-ph/0207340.
H.-M. Choi, C.-R. Ji, L.S. Kisslinger, Phys. Rev. D 64 (2001) 093006, hep-ph/0104117.
B.C. Tiburzi, G.A. Miller, Light front BetheSalpeter equation applied to form factors, Generalized parton distributions and generalized distribution amplitudes, hep-ph/0205109.
[48]
[49]
[50]
[51]
[52]
[53]
[54]
[55]
[56]
[57]
[58]
[59]
[60]
[61]
[62]
[63]
[64]
[65]
[66]
[67]
[68]
[69]
[70]
[71]
[72]
[73]
[74]
[75]
[76]
[77]
[78]
[79]
[80]
[81]
[82]
[83]
[84]
[85]
[86]
[87]
[88]
[89]
[90]
[91]
[92]
[93]
[94]
[95]
[96]
[97]
[98]
[99]
[100]
321
L. Theussl, S. Noguera, V. Vento, Eur. Phys. J. A 20 (2004) 483, nucl-th/0211036.

S. Ahmad, H. Honkanen, S. Liuti, S.K. Taneja, Phys. Rev. D 75 (2007) 094003, hep-ph/0611046.
M. Diehl, Phys. Rep. 388 (2003) 41, hep-ph/0307382.
A.V. Belitsky, A.V. Radyushkin, Phys. Rep. 418 (2005) 1, hep-ph/0504030.
J. Ralston, B. Pire, Phys. Rev. D 66 (2002) 111501, hep-ph/0110075.
M. Burkardt, Phys. Rev. D 62 (2000) 071503, hep-ph/0010082.
M. Burkardt, Int. J. Mod. Phys. A 18 (2003) 173, hep-ph/0207047.
M. Diehl, Eur. Phys. J. C 25 (2002) 223, hep-ph/0205208.
A.V. Belitsky, D. Mller, Nucl. Phys. A 711 (2002) 118, hep-ph/0206306.
A.V. Belitsky, X. Ji, F. Yuan, Phys. Rev. D 69 (2004) 074014, hep-ph/0307383.
M. Guidal, M. Vanderhaeghen, Phys. Rev. Lett. 90 (2003) 012001, hep-ph/0208275.
A.V. Belitsky, D. Mller, Phys. Rev. Lett. 90 (2003) 022001, hep-ph/0210313.
A.V. Belitsky, D. Mller, Phys. Rev. D 68 (2003) 116005, hep-ph/0307369.
S.J. Brodsky, D. Chakrabarti, A. Harindranath, A. Mukherjee, J.P. Vary, Phys. Lett. B 641 (2006) 440, hepph/0604262.
S.J. Brodsky, D. Chakrabarti, A. Harindranath, A. Mukherjee, J.P. Vary, Phys. Rev. D 75 (2007) 014003, hepph/0611159.
HERMES Collaboration, A. Airapetian, et al., Phys. Rev. Lett. 87 (2001) 182001, hep-ex/0106068.
CLAS Collaboration, S. Stepanyan, et al., Phys. Rev. Lett. 87 (2001) 182002, hep-ex/0107043.
H1 Collaboration, C. Adloff, et al., Phys. Lett. B 517 (2001) 47, hep-ex/0107005.
ZEUS Collaboration, S. Chekanov, et al., Phys. Lett. B 573 (2003) 46, hep-ex/0305028.
H1 Collaboration, A. Aktas, et al., Eur. Phys. J. C 44 (2005) 1, hep-ex/0505061.
A.V. Belitsky, D. Mller, A. Kirchner, Nucl. Phys. B 629 (2002) 323, hep-ph/0112108.
L. Frankfurt, W. Koepf, M. Strikman, Phys. Rev. D 54 (1996) 3194, hep-ph/9509311.
S.V. Goloskokov, P. Kroll, Eur. Phys. J. C 42 (2005) 281, hep-ph/0501242.
D.Y. Ivanov, L. Szymanowski, G. Krasnikov, JETP Lett. 80 (2004) 226, hep-ph/0407207.
HERMES Collaboration, A. Airapetian, et al., Phys. Rev. D 75 (2006) 011103, hep-ex/0605108.
HERMES Collaboration, Z. Ye, Transverse target-spin asymmetry associated with DVCS on the proton and a
resulting model-dependent constraint on the total angular momentum of quarks in the nucleon, hep-ex/0606061.
CLAS Collaboration, S. Chen, et al., Phys. Rev. Lett. 97 (2006) 072002, hep-ex/0605012.
Jefferson Lab Hall A, C.M. Camacho, et al., Phys. Rev. Lett. 97 (2006) 262002, nucl-ex/0607029.
D. Mller, A. Kirchner, Eur. Phys. J. C 32 (2003) 347, hep-ph/0302007.
A. Belitsky, D. Mller, Phys. Lett. B 417 (1997) 129, hep-ph/9709379.
L. Mankiewicz, G. Piller, E. Stein, M. Vnttinen, T. Weigl, Phys. Lett. B 425 (1998) 186, hep-ph/9712251.
X. Ji, J. Osborne, Phys. Rev. D 57 (1998) 1337, hep-ph/9707254.
X. Ji, J. Osborne, Phys. Rev. D 58 (1998) 094018, hep-ph/9801260.
A. Belitsky, D. Mller, Phys. Lett. B 486 (2000) 369, hep-ph/0005028.
A.V. Belitsky, D. Mller, Nucl. Phys. B 537 (1999) 397, hep-ph/9804379.
A. Belitsky, A. Freund, D. Mller, Nucl. Phys. B 574 (2000) 347, hep-ph/9912379.
A.V. Belitsky, A. Freund, D. Mller, Phys. Lett. B 493 (2000) 341, hep-ph/0008005.
A. Freund, M. McDermott, Phys. Rev. D 65 (2002) 074008, hep-ph/0106319.
A.V. Vinnikov, Code for prompt numerical computation of the leading order GPD evolution, hep-ph/0604248.
L. Schoeffel, private communication, 2006.
A. Radyushkin, Phys. Rev. D 56 (1997) 5524, hep-ph/9704207.
A. Belitsky, B. Geyer, D. Mller, A. Schfer, Phys. Lett. B 421 (1998) 312, hep-ph/9710427.
M. Polyakov, A. Shuvaev, On dual parametrizations of generalized parton distributions, hep-ph/0207153.
V. Guzey, T. Teckentrup, Phys. Rev. D 74 (2006) 054027, hep-ph/0607099.
A. Shuvaev, Phys. Rev. D 60 (1999) 116005, hep-ph/9902318.
J.D. Noritzsch, Phys. Rev. D 62 (2000) 054015, hep-ph/0004012.
I. Balitsky, V. Braun, Nucl. Phys. B 311 (1989) 541.
N. Kivel, L. Mankiewicz, Nucl. Phys. B 557 (1999) 271, hep-ph/9903531.
A. Manashov, M. Kirch, A. Schfer, Phys. Rev. Lett. 95 (2005) 012002, hep-ph/0503109.
M. Kirch, A. Manashov, A. Schfer, Phys. Rev. D 72 (2005) 114006, hep-ph/0509330.
D. Mller, A. Schfer, Nucl. Phys. B 739 (2006) 1, hep-ph/0509204.
A. Belitsky, D. Mller, L. Niedermeier, A. Schfer, Phys. Lett. B 437 (1998) 160, hep-ph/9806232.
A. Belitsky, D. Mller, L. Niedermeier, A. Schfer, Nucl. Phys. B 546 (1999) 279, hep-ph/9810275.
322
[101]
[102]
[103]
[104]
[105]
[106]
[107]
[108]
[109]
[110]
[111]
[112]
[113]
[114]
[115]
[116]
[117]
[118]
[119]
[120]
[121]
[122]
[123]
[124]
[125]
[126]
[127]
[128]
[129]
[130]
[131]
[132]
[133]
[134]
[135]
[136]
[137]
[138]
[139]
[140]
[141]
[142]
[143]
[144]
[145]
[146]
[147]
[148]
[149]
[150]
[151]
[152]
D. Mller, Phys. Lett. B 634 (2006) 227, hep-ph/0510109.

K. Kumericki, D. Mller, K. Passek-Kumericki, A. Schfer, Phys. Lett. B 648 (2007) 186, hep-ph/0605237.
Z. Chen, Nucl. Phys. B 525 (1997) 369, hep-ph/9705279.
A. Belitsky, D. Mller, L. Niedermeier, A. Schfer, Nucl. Phys. B 593 (2001) 289, hep-ph/0004059.
P. Hoodbhoy, X. Ji, Phys. Rev. D 58 (1998) 054006, hep-ph/9801369.
M. Diehl, Eur. Phys. J. C 19 (2001) 485, hep-ph/0101335.
M. Diehl, T. Gousset, B. Pire, J.P. Ralston, Phys. Lett. B 411 (1997) 193, hep-ph/9706344.
J. Bjorken, S. Drell, Relativistic Quantum Mechanics, McGrawHill, New York, 1964.
P. Roman, Introduction to Quantum Field Theory, Wiley, New York, 1969.
S. Ferrara, R. Gatto, A. Grillo, Phys. Lett. B 36 (1971) 124;
S. Ferrara, R. Gatto, A. Grillo, Phys. Lett. B 38 (1971) 188, Erratum.
S. Ferrara, R. Gatto, A. Grillo, Phys. Lett. B 38 (1972) 333.
S. Ferrara, R. Gatto, A. Grillo, Phys. Rev. D 5 (1972) 3102.
S. Ferrara, A. Grillo, G. Parisi, Nucl. Phys. B 54 (1973) 552.
D. Mller, Phys. Rev. D 58 (1998) 054005, hep-ph/9704406.
V.M. Braun, G.P. Korchemsky, D. Mller, Prog. Part. Nucl. Phys. 51 (2003) 312, hep-ph/0306057.
X. Ji, J. Phys. G 24 (1998) 1181, hep-ph/9807358.
F. Carlson, Sur une classe de sries de Taylor, PhD thesis, Uppsala University, 1914.
R.J. Eden, P.V. Landshoff, P.J. Olive, J.C. Polkinghorne, The Analytic S-Matrix, Cambridge Univ. Press, 1966,
p. 126.
O.V. Teryaev, Analytic properties of hard exclusive amplitudes, hep-ph/0510031.
M.V. Polyakov, C. Weiss, Phys. Rev. D 60 (1999) 114017, hep-ph/9902451.
A. Belitsky, D. Mller, A. Kirchner, A. Schfer, Phys. Rev. D 64 (2001) 116002, hep-ph/0011314.
O.V. Teryaev, Phys. Lett. B 510 (2001) 125, hep-ph/0102303.
M.J. Creutz, S.D. Drell, E.A. Paschos, Phys. Rev. 178 (1969) 2300.
M. Damashek, F.J. Gilman, Phys. Rev. D 1 (1970) 1319.
C.A. Dominguez, C. Ferro Fontan, R. Suaya, Phys. Rev. D 31 (1970) 365.
G. Schierholz, Phys. Rev. D 10 (1974) 1261.
S.J. Brodsky, F.E. Close, J.F. Gunion, Phys. Rev. D 5 (1972) 1384.
S.J. Brodsky, F.E. Close, J.F. Gunion, Phys. Rev. D 8 (1973) 3678.
M. Bander, Phys. Rev. D 5 (1972) 3274.
Y. Frishman, Phys. Rep. 13 (1974) 1.
J.M. Cornwall, D. Corrigan, R.E. Norton, Phys. Rev. Lett. 24 (1970) 1141.
J.M. Cornwall, D. Corrigan, R.E. Norton, Phys. Rev. D 3 (1971) 536.
F.-M. Dittes, A. Radyushkin, Phys. Lett. B 134 (1984) 359.
M. Sarmadi, Phys. Lett. B 143 (1984) 471.
S. Mikhailov, A. Radyushkin, Nucl. Phys. B 254 (1985) 89.
D. Mller, Phys. Rev. D 49 (1994) 2525.
B. Melic, D. Mller, K. Passek-Kumericki, Phys. Rev. D 68 (2003) 014013, hep-ph/0212346.
A. Belitsky, A. Schfer, Nucl. Phys. B 527 (1998) 235, hep-ph/9801252.
B. Melic, B. Niic, K. Passek, Phys. Rev. D 65 (2002) 053020, hep-ph/0107295.
G. Curci, W. Furmanski, R. Petronzio, Nucl. Phys. B 175 (1980) 27.
E.B. Zijlstra, W.L. van Neerven, Nucl. Phys. B 383 (1992) 525.
E. Zijlstra, W. van Neerven, Nucl. Phys. B 417 (1994) 61;
E. Zijlstra, W. van Neerven, Nucl. Phys. B 426 (1994) 245, Erratum;
E. Zijlstra, W. van Neerven, Nucl. Phys. B 773 (2007) 105, Erratum.
M. Gluck, E. Reya, A. Vogt, Z. Phys. C 48 (1990) 471.
W. Vogelsang, Nucl. Phys. B 475 (1996) 47, hep-ph/9603366.
R. Mertig, W. van Neerven, Z. Phys. C 70 (1996) 637, hep-ph/9506451.
A. Vogt, S. Moch, J.A.M. Vermaseren, Nucl. Phys. B 691 (2004) 129, hep-ph/0404111.
J. Blmlein, S. Kurth, Phys. Rev. D 60 (1999) 014018, hep-ph/9810241.
J.A.M. Vermaseren, Int. J. Mod. Phys. A 14 (1999) 2037, hep-ph/9806280.
J. Blmlein, S.-O. Moch, Phys. Lett. B 614 (2005) 53, hep-ph/0503188.
D. Maitre, Comput. Phys. Commun. 174 (2006) 222, hep-ph/0507152.
[153]
[154]
[155]
[156]
[157]
[158]
[159]
[160]
[161]
[162]
[163]
[164]
[165]
[166]
[167]
[168]
[169]
[170]
[171]
[172]
[173]
[174]
[175]
[176]
[177]
[178]
[179]
[180]
[181]
[182]
323
A. Belitsky, D. Mller, L. Niedermeier, A. Schfer, Phys. Lett. B 474 (2000) 163, hep-ph/9908337.
A.V. Belitsky, D. Mller, Nucl. Phys. B 527 (1998) 207, hep-ph/9802411.
A.P. Bakulev, N.G. Stefanis, Nucl. Phys. B 721 (2005) 50, hep-ph/0503045.
K. Goeke, M. Polyakov, M. Vanderhaeghen, Prog. Part. Nucl. Phys. 47 (2001) 401, hep-ph/0106012.
M. Diehl, T. Feldmann, R. Jakob, P. Kroll, Eur. Phys. J. C 39 (2005) 39, hep-ph/0408173.
M. Guidal, M.V. Polyakov, A.V. Radyushkin, M. Vanderhaeghen, Phys. Rev. D 72 (2005) 054013, hep-ph/
0410251.
C.F. Perdrisat, V. Punjabi, M. Vanderhaeghen, Prog. Part. Nucl. Phys. 59 (2007) 694, hep-ph/0612014.
W. Melnitchouk, R. Ent, C. Keppel, Phys. Rep. 406 (2005) 127, hep-ph/0501217.
E.P. Wigner, Group Theory and its Application to the Quantum Mechanics of Atomic Spectra, Academic Press,
New York, 1959.
M. Diehl, T. Gousset, B. Pire, O. Teryaev, Phys. Rev. Lett. 81 (1998) 1782, hep-ph/9805380.
S.D. Drell, T.-M. Yan, Phys. Rev. Lett. 24 (1970) 181.
QCDSF Collaboration, M. Gockeler, et al., Nucl. Phys. B (Proc. Suppl.) 140 (2005) 399, hep-lat/0409162.
D. Mller, Pomeron dominance in deeply virtual Compton scattering and the femto holographic image of the
proton, hep-ph/0605013.
A.P. Szczepaniak, J.T. Londergan, Phys. Lett. B 643 (2006) 17, hep-ph/0604266.
A. Donnachie, H.G. Dosch, Phys. Lett. B 502 (2001) 74, hep-ph/0010227.
A. Freund, M. McDermott, M. Strikman, Phys. Rev. D 67 (2003) 036001, hep-ph/0208160.
M. Capua, S. Fazio, R. Fiore, L. Jenkovszky, F. Paccanoni, Phys. Lett. B 645 (2007) 161, hep-ph/0605319.
F. James, Minuit reference manual, version 94.1 (CERN, 1994).
S. Forte, Int. J. Mod. Phys. A 21 (2006) 769, hep-ph/0509371.
A. Shuvaev, K. Golec-Biernat, A. Martin, M. Ryskin, Phys. Rev. D 60 (1999) 014015, hep-ph/9902410.
H1 Collaboration, S. Aid, et al., Nucl. Phys. B 470 (1996) 3, hep-ex/9603004.
S. Alekhin, Phys. Rev. D 68 (2003) 014002, hep-ph/0211096.
H1 Collaboration, C. Adloff, et al., Phys. Lett. B 483 (2000) 23, hep-ex/0003020.
ZEUS Collaboration, S. Chekanov, et al., Eur. Phys. J. C 24 (2002) 345, hep-ex/0201043.
M. Strikman, C. Weiss, Phys. Rev. D 69 (2004) 054012, hep-ph/0308191.
L. Frankfurt, M. Strikman, C. Weiss, Annu. Rev. Nucl. Part. Sci. 55 (2005) 403, hep-ph/0507286.
I.V. Anikin, O.V. Teryaev, Phys. Rev. D 76 (2007) 056007, arXiv: 0704.2185 [hep-ph].
M. Diehl, D.Y. Ivanov, Eur. Phys. J. C 52 (2007) 919, arXiv: 0707.0351 [hep-ph].
M.V. Polyakov, Tomography for amplitudes of hard exclusive processes, arXiv: 0707.2509 [hep-ph].

Finite Heisenberg groups from non-Abelian orbifold

quiver gauge theories
Benjamin A. Burrington a, , James T. Liu b , Leopoldo A. Pando Zayas b
a School of Physics and Astronomy, The Raymond and Beverly Sackler Faculty of Exact Sciences,
Tel Aviv University, Ramat Aviv 69978, Israel

b Michigan Center for Theoretical Physics, Randall Laboratory of Physics, The University of Michigan Ann Arbor,
MI 48109-1040, USA
Received 8 July 2007; received in revised form 27 October 2007; accepted 7 November 2007
Abstract
A large class of orbifold quiver gauge theories admits the action of finite Heisenberg groups of the form
i Heis(Zqi Zqi ). For an Abelian orbifold generated by , the Zqi shift generator in each Heisenberg
group is one cyclic factor of the Abelian group . For general non-Abelian , however, we find that the shift
generators are the cyclic factors in the Abelianization of . We explicitly show this for the case = (27),
where we construct the finite Heisenberg group symmetries of the field theory. These symmetries are dual
to brane number operators counting branes on homological torsion cycles. These brane number operators
therefore do not commute. We compare our field theory results with string theory states and find perfect
agreement.
PACS: 11.25.Tq; 11.25.Uv; 11.30.Fs
Keywords: AdS/CFT; Gauge/string duality; Quiver gauge theories; Orbifolds; Global symmetries; Heisenberg groups
1. Introduction
The AdS/CFT correspondence, which relates field theories with dual string (gravitational)
theories, is well on its way to becoming firmly established as an especially important new tool
for investigating field theories outside of perturbative control. As a strong/weak coupling duality,
E-mail address: bburring@umich.edu (B.A. Burrington).

doi:10.1016/j.nuclphysb.2007.11.004
B.A. Burrington et al. / Nuclear Physics B 794 (2008) 324347
325
AdS/CFT and its variants have opened up a whole new window into the exploration of strongly
coupled gauge theories through their weakly coupled string counterparts. Certainly, much of the
focus of gauge/string duality has been on the use of appropriate string duals as a way to address
important non-perturbative issues in gauge theories. However, given the limited technology for
studying string theory in non-trivial closed string backgrounds (i.e. curved backgrounds with RR
flux), one may ask whether it is possible to use AdS/CFT in the other direction to explore the
structure of string theory in such backgrounds starting from our understanding of gauge theories.
In fact, Gukov et al. in [1] did just this; based on properties of the dual orbifold quiver gauge
theory, they found the novel feature that brane number operators, counting branes wrapped on
torsion cycles,1 do not commute. In particular, they examined the duality between string theory
on AdS5 S 5 /Z3 and the quiver gauge theory corresponding to a stack of D3-branes placed at the
singular point of R1,3 C3 /Z3 , and demonstrated that the resulting quiver gauge theory admits
a set of discrete global symmetries that forms a Heisenberg group. Based on this, they concluded
that, on the gravitational side, F-string and D-string number operators do not commute when the
strings are wrapped on torsion cycles. In fact, these operators close on the number operator for
D3-branes wrapping a torsion 3-cycle, and it is this D3-brane number operator which plays the
part of the central extension in the Heisenberg group. We stress again that this non-commutativity
was first found in the field theory, and subsequently mapped into the dual supergravity/string
theory in [1]. It is therefore a way in which one can use AdS/CFT to predict a structure that the
proper classification of D-branes on non-trivial backgrounds must have.
A similar observation about fluxes on torsion cycles was recently made by Freed et al. in [2,3]
for generalized U(1) gauge theories. In this case, Poincar duality of the theory requires that two
gradings (electric and magnetic) of the Hilbert space be possible. Furthermore, in the presence
of torsion cycles, these two gradings cannot be simultaneously implemented: i.e. electric and
magnetic charges do not commute. As argued in [2], it is precisely this effect that is responsible
for the non-commutativity of charges found in [1].
The non-commutativity in [1] was found as a non-commutativity of global symmetries of the
quiver gauge theory dual. These discrete symmetries fall into two classes, with the first being
permutation type symmetries that map gauge groups into gauge groups (and correspondingly bifundamentals into bifundamentals) and the second being rephasing symmetries which act on the
links in such a manner that all trace type gauge invariant composite operators remain unchanged
under the rephasing. We will refer to the permutation symmetries here as A type symmetries and
the rephasing symmetries as either B (if they do not commute with permutations) or C (if they
are central) type symmetries.
In particular, for the S 5 /Z3 quiver theory considered in [1], the permutation symmetries are
generated by a shift operator A of order three which cyclically permutes the three nodes of the
quiver. (Note that this operation is simply the action of the Z3 orbifold group on the quiver itself.)
In addition, the rephasing symmetries are generated by an order three clock operator B. Taken
together, A and B do not commute, and form the Heisenberg group Heis(Z3 Z3 ) according to
A3 = B 3 = C 3 = 1,
AB = BAC,
(1.1)
where C is a central element associated with permutation invariant rephasings.

1 Here we will always refer to the topological torsion groups appearing in integer valued homology rather than notions relating to modifications of connections on a tangent bundle. To stress this, we will usually refer to these topological
cycles as torsion cycles.
326
The finite Heisenberg group identified above is the group of discrete global symmetries acting on the quiver gauge theory. However, as argued in [1], these symmetries ought to persist
in the dual description of strings on AdS5 S 5 /Z3 . In this dual string picture, the A operator is mapped to the number operator associated with wrapped F-strings, while the B and C
operators correspond to number operators associated with wrapped D-strings and D3-branes, respectively. Although S 5 by itself admits no non-trivial cycles, S 5 /Z3 admits the torsion cycles
H1 (S 5 /Z3 ; Z) = Z3 and H3 (S 5 /Z3 ; Z) = Z3 , which are precisely what are needed for strings
and D3-branes to wrap. In this case, the non-commutativity of (1.1) is a consequence of the
non-commutativity of fluxes [2,3] associated with these cycles. Furthermore, the fact that trace
type gauge invariant operators are inert under the B and C symmetries indicate that their dual
operators affect only states that are non-perturbative in the 1/N expansion, giving additional
justification to their association with D-branes.
It is worth reviewing some of the reasons that one associates the wrapped strings with the A
operator, and wrapped D-branes with the B and C operators. First we note that the trace operators
are exactly those that are leading order in the 1/N expansion because they can be constructed
using a small number of fields, n N . For example, a cubic superpotential piece Tr(1 2 3 )
is a product of only three fields, and hence corresponds to a low energy excitation. These
trace operators may therefore be described in terms of the low energy degrees of freedom on the
string/gravity side, namely fundamental strings. Furthermore, the B and C symmetries leave the
trace operators invariant by construction, and so only the A operations act on them. While the
B and C operations do not affect the trace operators, they do affect the det(i ) operators. These
determinant operators are products of N fields, and so represent an energetic states and are nonperturbative from the point of view of the 1/N expansion. The non-perturbative (in 1/N ) states
on the string theory side are given by D-branes, and so explains why the B and C operators are
associated with number operators for D-branes wrapped on cycles, but does not explain which
particular D-branes they are associated with. Nevertheless, it is generally straightforward to disentangle the B and C operators based on their commutation properties with the A operators. We
refer the reader to [1] for a detailed analysis in the Abelian Z3 case.
The work of [1] was recently extended by studying other orbifold theories [4], the inclusion of
fractional branes [5], and the effect of Seiberg duality on these global symmetries [6]. In all these
investigations, however, the orbifolds under consideration were Abelian, being Zn orbifolds of
either S 5 , T 1,1 or Y p,q . For the case without fractional branes, the resulting discrete symmetry
groups were all found to be isomorphic to Heis(Zn Zn ), which is exactly what one would
expect to have been inherited from the Zn action of the orbifold group.
Although the bulk of the work on orbifold models focuses on the Abelian case, in general nonAbelian orbifolds may also be constructed. Other than for added technical issues which arise in
the latter case, there is no fundamental distinction between Abelian and non-Abelian orbifolds.
Hence it is natural to expect that the identification of discrete global symmetries of Abelian orbifold quiver gauge theories carries over to non-Abelian orbifolds as well. Demonstrating that this
is in fact the case will be the focus of the present paper. In particular, we examine quiver gauge
theories that are dual to IIB string theory on AdS5 S 5 / , where now is some non-Abelian
discrete subgroup of SU(3), so that the gauge theory preserves at least N = 1 supersymmetry.
Although we only focus on orbifolds of the five-sphere, it is in principle possible to use the same
techniques for field theories dual to orbifolds of other spaces which admit isometries. One simply has to use the regular representation of the group to act on the gauge indices, which is the
natural way of making a geometric orbifold [7,8]. In the following, we will always be using
the regular representation of as the group acting on the gauge indices.
327
In the case where is Abelian, the A type permutation symmetries are easy to visualize
geometrically, as they simply correspond to the action of the group mapping the image D3branes to each other (so that F-strings stretched between D3-branes and their images close up to
an operation of A). This immediately suggests that the group of A type symmetries remains ,
even when extended to the non-Abelian case. As we show below, however, this identification is
not entirely correct; instead, the actual group generated by the A type symmetries turns out to be
the Abelianization of , which we denote by /[, ].
Unlike the A type symmetries, which are conceptually easy to visualize based on the action of
on the quiver, the B and C type rephasing symmetries are more difficult to identify. Motivated
by [1] as well as the dual string picture (where A and B are related by interchanging F-strings
and D-strings), it is natural to identify the group of B type symmetries with as well. However,
the explicit identification of the rephasing symmetries is significantly different, and rather more
intricate to work out. Nevertheless, as demonstrated in various examples, this general picture of
A and B type operations turns out to be correct. The A and B symmetries do not commute, and
instead close on a set of central elements given by C. Since A and B are both identified with ,
the group formed by these symmetries is just the finite Heisenberg group
Heis( , ).
(1.2)
This is identified as the group of discrete global symmetries of the S 5 / orbifold quiver gauge
theory.
Turning now to the dual string picture, we expect that the above quiver gauge theory may be
obtained by placing a stack of N D3-branes near the orbifold singularity of R1,3 C3 / . Since
the near-horizon space is AdS5 S 5 / , in this dual picture the Heisenberg group symmetries
(1.2) arise from the properties of strings and D3-branes wrapping appropriate cycles on S 5 / .
Since the homology of S 5 / is given by [9]

H1 S 5 / ; Z = ,
(1.3)
H3 S 5 / ; Z = ,
we indeed identify the appropriate cycles for F-strings, D-strings and D3-branes to wrap. In
particular, since the string duals of A and B are the number operators counting wrapped F-strings
and D-strings, they both must independently form groups isomorphic to H1 , which as we see is
just . This of course agrees with the quiver result.
The final part of the duality picture is to identify C with wrapped D3-branes, and hence the
group generated by C with H3 (which again is just ). This now allows us to make a stronger
statement on the form of the Heisenberg group (1.2). Since is Abelian, it may be decomposed
into a set of cyclic groups
= Za1 Za2 ,
(1.4)
where each factor is associated with a torsion cycle in H1

Because A, B and C are all
identified with , the central extension implicit in (1.2) may be more explicitly stated in the
decomposition
(S 5 / ).
Heis( , ) = Heis(Za1 , Za1 ) Heis(Za2 , Za2 ) .
(1.5)
Each individual factor in this decomposition is connected to the non-commutativity of fluxes on

that particular torsion cycle [2,3]. (Since we take to be a discrete subgroup of SU(3), there are
only a limited number of torsion cycles that may arise in practice.) Our main result can then be
stated as:
328
Quiver gauge theories obtained as worldvolume theories on a stack of N D3-branes placed

at the singularity of C3 / where is a (possibly non-Abelian) finite discrete subgroup of
SU(3) admit the action of global symmetries generating a group of the form (1.5) where the
factors Zai are given by the Abelianization of as in (1.4).
The structure of the paper is as follows. In Section 2 we review the construction of quiver
gauge theories obtained as the worldvolume theory on a stack of N D3-branes placed at the
conical singularity of C3 / , paying particular attention to the case where is non-Abelian.
This section also presents a general construction of the set of permutations A and rephasing
symmetries B and C. Section 3 describes how the field theory results match those of the dual
string theory states. Section 4 contains a number of explicit examples motivating our claims.
Finally, in Section 5, we present our conclusions and point out some open questions. We relegate
explicit details of the Abelianization of discrete subgroups of SU(3) to Appendix A and the
representations of (27) to Appendix B.
2. Discrete symmetries of orbifold quiver gauge theories
Before turning to the construction of the discrete symmetry operations A, B and C, we first
summarize the basic features of N = 1 quiver gauge theories corresponding to a stack of N
D3-branes at the singular point of the C3 / orbifold. We are, of course, especially interested in
the case where is non-Abelian. Pioneering work towards constructing these orbifold theories
began with [1012], and subsequently the quivers were more fully developed in [7,8].
2.1. N = 1 gauge theories from C3 / orbifolds
From our point of view, an N = 1 quiver is essentially a set of nodes representing gauge
groups and links which describe chiral multiplets transforming in the bifundamental representation. For a discrete orbifold group SU(3), the nodes of the quiver may be placed in
one-to-one correspondence with the irreducible representations ri of , and the corresponding
gauge groups are taken to be SU(ni N ) where ni = dim(ri ). In order to obtain the bifundamental
matter, we must choose an appropriate embedding of the orbifold action in C3 , corresponding to
a (possibly reducible) three-dimensional representation 3 of . The number of bifundamentals
stretching from node i to node j is then given by bij in the decomposition

3 ri =
(2.1)
bij rj .
j
From this definition, it is clear that bij is related to the familiar adjacency matrix via
aij = bij bj i .
(2.2)
Although not directly encoded in the quiver diagram itself, the (cubic) superpotential may
be fully determined from the properties of the orbifold group . In the notation of [8], the
superpotential is given by
fij ,fj k ,fki ij j k

Tr fij fj k fkiki ,
W=
hij k
(2.3)
where
f ,fj k ,fki
hijijk
= (Yfij )vi v j (Yfj k )vj v k (Yfki )vk v i .
(2.4)
329
Here fij = 1, . . . , bij labels the specific link showing up in the above decomposition, (2.1), and
(Yfij )vi v j is the corresponding ClebschGordan coefficient. In particular, let Vrvii denote the components (vi = 1, 2, . . . , ni ) of a vector transforming under the irreducible representation ri of .
The ClebschGordan coefficient (Yfij )vi v j is then defined by
v
Vrjj,fij = V3 Vrvii (Yfij )vi v j ,
(2.5)
where we make no distinction between superscript or subscript indices. Note that = 1, 2, 3

labels the components of the 3 representation.
2.2. Discrete symmetries from one-dimensional representations of
Given a quiver theory constructed as above, our goal is now to identify potential discrete
global symmetries of the quiver. These symmetries fall naturally into two categories:
1. Permutation symmetries which maps fields to fields and gauge groups to gauge groups. Motivated by the notation of [1], we label these permutations as A type symmetries.
2. Symmetries that rephase fields in such a way as to leave all trace type gauge invariant operators inert. These rephasing symmetries may be thought of as anomaly free discrete subgroups
of the global U(1) symmetries acting at each node of the quiver. These symmetries will be
denoted as either B type if they do not commute with the permutations or C type if they do.
2.2.1. Permutation symmetries
For a given quiver, the A type permutation symmetries are easy to visualize. If we view
the quiver as a directed graph, then the A type symmetries correspond to the subgroup of the
automorphism group of the quiver that leaves the superpotential invariant. In order to ensure that
this is a symmetry of the full theory, it is important that the superpotential remains invariant.
However, since this superpotential data is not manifest from the quiver diagram itself, additional
input must be considered when determining the group of permutations generated by A. As will be
seen in the examples below, this group could be substantially smaller than the full automorphism
group of the quiver diagram.
In order to systematically construct the A type symmetries, we first note that good permutations can only map amongst gauge groups of the same rank. Since each gauge group (node in the
quiver) is labeled by an irreducible representation ri of , this indicates that good permutations
must only map irreducible representations of the same dimension into each other. Such transformations are in fact naturally furnished by the one-dimensional representations of , which we
denote by 1 , where labels the particular representation. The action of 1 on the nodes of the
quiver follows directly from the tensor product map
1 ri = r(i) .
(2.6)
Note that, so long as ri is irreducible, then so is r(i) . This may easily be seen because multiplication by the one-dimensional representation on the left does not affect how one may or may not
block diagonalize a given matrix representation. As a result, the map (2.6) indeed gives rise to a
good permutation on the nodes of the quiver, taking node i to node (i).
What remains to be demonstrated is that the transformation induced by multiplication with 1
also yields a permutation of the bifundamental links consistent with invariance of the superpotential. To see that this is the case, we recall that all superpotential terms in the orbifolded theory
330
that are mapped to each other by the action of 1 descend from a single superpotential term in
the original un-orbifolded parent N = 4 theory. This turns out to be sufficient to guarantee that
the above action of the one-dimensional representations is a symmetry of the orbifold quiver.
To see this, we first examine the permutation of the links induced by 1 . Here we may simply
follow the decomposition rules for the tensor product

bij 1 rj =
bij r(j ) ,
3 r(i) = 3 (1 ri ) = 1 (3 ri ) =
(2.7)
j
which demonstrates that the matrices bij before and after the transformation are related by
b(i)(j ) = bij ,
(2.8)
and therefore so are the adjacency matrices. Since (i) is just the relabeling of node i, this
adjacency matrix relation is precisely what is desired for generating good permutations of the
links.
Of course, we must also ensure that the superpotential remains invariant under the mapping
induced by 1 . By examining (2.3) and (2.4), we see that this would be the case, so long as
h(i)(j )(k) = hij k
(2.9)
(where we have suppressed the fij labels for brevity). To prove this, we have to turn to the properties of the ClebschGordan decomposition. In a quantum mechanical notation, the Clebsch
Gordan coefficient Yvi v j corresponding to the decomposition 3 ri rj , and defined in (2.5),
may be written as the matrix element
Yvi v j = 3, ri ; , vi |3, ri ; rj , vj .
(2.10)
At the same time, the ClebschGordan coefficients for the (rather trivial) multiplication by the
one-dimensional representation 1 given in (2.6) are given by a 3 3 unitary matrix
Uvi v (i) = 1 , ri ; 1, vi |1 , ri ; r(i) , v(i) .
(2.11)
(Unitarity can be seen because we take 1 1 = 10 without any phases, where 1 is the complex
conjugate of 1 , and 10 is the trivial representation.) Inserting a complete set of states, we then
have
Yv(i) v (j ) = 3, r(i) ; , v(i) |3, r(i) ; r(j ) , v(j )

= 1 , ri ; r(i) , v(i) |1 , ri ; 1, vi 3, ri ; , vi |3, ri ; rj , vj
1 , rj ; 1, vj |1 , rj ; r(j ) , v(j )
= Uvi v (i) Yvi v j Uvj v (j ) .
(2.12)
Since Uvi v(i) is unitary, the superpotential relation (2.9) then follows directly from the definition (2.4). Thus we have now shown that permutations generated by the map (2.6) are indeed
good symmetries of the quiver gauge theory.
Finally, before turning to the rephasing symmetries, we make the key observation that the A
type permutation symmetries do indeed form a group. The reason for this is simply that the onedimensional representations of a finite group are all of the representations of the Abelianization
of that group, and the tensor product for the representation matrices acts as group multiplication
for these representations. We therefore conclude that the group formed by A type permutations
331
is given by
{A} = /[, ].
(2.13)
2.2.2. Rephasing symmetries

The second class of symmetries that may arise in the quiver theory are discrete (global) U(1)
transformations acting on the nodes of the quiver. As discussed in [5,6], these rephasing symmetries may be constructed by assigning discrete U(1) charges qi to node i. The SU(ni N ) adjoint
ij
gauge multiplets are of course inert under this transformation. However, the bifundamentals fij
stretching from the ith node to the j th node pick up a phase
fij qi qj fij ,
ij
ij
(2.14)
where is a primitive kth root of unity (to be determined below from the anomaly cancellation
conditions).
To ensure that the above transformation is a symmetry of the quiver theory, we must demand
that it both leaves the superpotential invariant and that it is anomaly free. Invariance of the superpotential is of course automatically satisfied, so only the anomaly condition comes in to restrict
the discrete charges. As demonstrated in [6], vanishing of the chiral anomaly at the ith node
demands that

aij qj
= 1,
(2.15)
where we recall that aij is the adjacency matrix, and qi qi ni N is the charge weighted by the
rank of the ith gauge group. Solutions to the above fall into two classes. The first class consists of
continuous global U(1) symmetries, which happens whenever qi is a zero eigenvector
of the adjacency matrix. The second class occurs whenever the components of the vector j aij qj share
a common divisor k (which can be taken to be integral by appropriate scaling of the charges). In
other words [6]

aij qj 0 (mod k) for k Z.
(2.16)
j
Given a set of charges qi solving the above equation, the chiral multiplets are then rephased
according to
ij
2i qi
(n
i
fij e kN
nj )
j
ij
(2.17)
fij .
Note, however, that since a transformation in the center of SU(ni N ) at node i may be written
ij
ij
as fij e2i/ni N fij , the weighted charges qi are only well defined mod k. In addition, this
demonstrates that such a rephasing symmetry is an order k element.
While (2.16) provides the only condition on the rephasing symmetries, it does not provide a
constructive procedure for obtaining the B type (non-central) and C type (central) transformations. Nevertheless, in practice, it is not too difficult to search for and obtain a consistent set of
charge assignments yielding the appropriate discrete transformations. This will be demonstrated
below in the examples. For now, we simply note that, for an A type transformation generated by
1 and a B type transformation specified by the charges qi , we may evaluate their commutator
expression
A1 B 1 AB:
ij
2i qi q(i) qj q(j )
( n
n
)
i
j
fij e kN
ij
fij .
(2.18)
332
Identifying this with a central element C (i.e. AB = BAC) shows that the U(1) charges si corresponding to C are given by
si = qi q(i)
(2.19)
(mod k).
Evaluating the commutator of A and C then gives rise to a potentially new symmetry element
D with U(1) charges given by
ti = si s(i)
(mod k)
= qi 2q(i) + q((i)) .
(2.20)
In order for C to be central, we demand that D is gauge equivalent to the identity. Perhaps the
simplest way for this to occur is for ti to vanish at each node. This gives rise to a sufficient
condition for C to be central
2q(i) = qi + q((i))
(2.21)
(mod k).
This condition, along with the anomaly requirement (2.16), may be used as a guide for constructing the appropriate B type rephasing symmetries of the quiver. We note, however, that
while (2.21) is a sufficient condition, there are other possibilities that make C central as well.
One case that often shows up is when the nodes are all of the same rank. In this case, instead of
demanding the vanishing of ti , it is sufficient to ensure that they have a common value
ti = tj
(mod k) when all ni = nj .
(2.22)
This case arises in particular for Abelian orbifolds of the form C3 /Zn .
Note that the charge condition (2.21) may be iterated to yield a general solution for the charges
q n (i) = nq(i) (n 1)qi
(mod k),
n = 1, 2, . . . ,
(2.23)
in terms of only two charges qi and q(i) . Recalling that this rephasing is an order k operation,
we may set n = k in the above expression to obtain k(q(i) qi ) = 0 (mod k), which is clearly
always satisfied for integer charges qi . This ensures the consistency of the charge requirement
(2.21).
2.3. The Heisenberg group
As indicated in (2.13), the A type permutation symmetries close to form a group isomorphic to , the Abelianization of the orbifold action . Ideally, we would be able to demonstrate
explicitly that the B type rephasing symmetries would also form a group isomorphic to . However, we have as yet been unable to show this in a general manner. Nevertheless, in practice, and
as indicated in the examples, given the proper identification of the A symmetries, it is straightforward to construct the appropriate B generators consistent with (2.16) and either (2.21) or (2.22).
(The C generators then follow as a direct consequence of commuting A and B.)
The construction of the B generators is guided by noting that since {A} is Abelian, it necessarily decomposes into a set of cyclic groups
{A} = = Za1 Za2 .
(2.24)
We may then focus on a single Zai group at a time. Being Abelian, this group is generated by
a single element (i.e. some particular 1 ) which we denote by Ai , and which cycles the nodes
of the quiver. The set of nodes of the quiver then fall into distinct orbits of Ai . In general, the
333
rephasing generator Bi corresponding to this Ai may be obtained by the linear charge assignment
given in (2.23), while in some cases iteration of (2.22) may be required. In any case, the related
central element Ci is given by the charge assignment of (2.19).
In this way, the complete group of global symmetries of the quiver is constructed as a direct
product of individual Heisenberg groups generated by the elements Ai Bi = Bi Ai Ci (where i
labels the group, and is not summed over). In other words, given the decomposition (2.24), the
discrete symmetry group takes the form
Heis(Za1 , Za1 ) Heis(Za2 , Za2 ) ,
(2.25)
which is just the decomposition (1.5) highlighted in the introduction.

3. String theory interpretation and torsion cycles
The result (2.25) takes on added physical significance when the quiver gauge theory is related
to the dual string picture. The general framework is of course clear: following the general ideas
of [1] and [4], we wish to identify the symmetry generators A, B and C in the field theory with
corresponding operators counting the number of wrapped F-strings, D-strings and D3-branes,
respectively, in the dual string theory.
Some subtleties arise, however, in making precise the field theory/string theory connection in
cases were is non-Abelian. As a result, we find it worthwhile to make a distinction between:
1. The orbifold quiver gauge theory.
2. The near horizon manifold S 5 / .
3. The string theory orbifold C3 / .
In the first case, we have demonstrated that the orbifold quiver gauge theory admits the finite
Heisenberg group (1.5) as a group of global symmetries of the field theory. Based on AdS/CFT,
this field theory ought to be dual to string theory on the horizon manifold S 5 / . To show that
the symmetries match on both sides of the duality, we need information on the homology classes
of S 5 / . After all, we expect the A transformations to be identified with operators counting the
number of F-strings in the dual string theory. The structure of eigenvalues of these operators is
clearly determined by the first homology class H1 (S 5 / ). Furthermore, based on S-duality, the
B transformations may be associated with operators counting the number of D-strings; these operators are also valued in H1 (S 5 / ). Finally, the C transformations match the operators counting
wrapped D3-branes and is determined by the third homology class, H3 (S 5 / ). While S 5 itself
has no non-trivial cycles, orbifolds of S 5 may admit torsion one and three-cycles, which are
exactly what is required to allow this duality to work.
There is a slight subtlety in the identification of the operators of C. Namely, it may seem
curious that the D3-brane number operator is somehow related to the F-string and D-string number operators. However, we note that Poincar duality relates the torsion free part of homology
groups by (Hp )T F = (H(dp) )T F , while the (cyclic) torsion parts of the homology are related
by Tors(Hp ) = Tors(H(dp1) ). We are in particular interested in the torsion parts; on the five
sphere, Tors(H1 ) = Tors(H(511) ) = Tors(H(3) ) (see for example [13]). Therefore, the homology classes of torsion cycles that F-strings or D-strings may wrap are isomorphic to the homology
classes of torsion cycles that D3-branes may wrap. As was shown explicitly in [9], in the more
general case of branes placed in generic toric singularities, this is no longer the case, and one can
have that H1 (H ) is not isomorphic to H3 (H ) where H is the near horizon space.
334
For S 5 / , on the other hand, the non-trivial homology is given by (1.3). As a result, the dual
picture of F-strings and D-strings wrapping torsion cycles gives rise to the identical Heisenberg
group (1.5) that was obtained in the field theory analysis. This statement is the extension of [1,4]
to the non-Abelian case.
Although the duality between the symmetries of the quiver gauge theory and the near horizon
manifold S 5 / is clear, the connection with string theory on C3 / is less so (at least in the
non-Abelian case). This is because a rigorous understanding of D3-branes near a non-Abelian
orbifold singularity involves the generalization of [1012] to the non-Abelian case, and this is as
yet incomplete. Here we simply make an observation on the structure of twisted sectors in string
theory on orbifolds. In string theory we are mainly familiar with global orbifolds by Abelian
groups such as ZN or ZN ZN . In this case, the space of twisted sectors is in correspondence
with elements of the group excluding the identity element. In general, however, twisted sectors
are in correspondence with conjugacy classes [14]. Consider a string field X whose boundary
conditions are twisted by an element g , namely X( +2) = gX( ). For any element h
we can act on the previous relation on the left: hX( + 2) = hgX( ) = (hgh1 ) hX( ). This
means that strings twisted by hgh1 are all in the same sector. Thus twisted sectors are defined
only up to conjugacy classes. That is, there is one sector per conjugacy class in .
Since the field theory states are classified by the Abelianization of , to make a connection
between the quiver and orbifold pictures, we must appropriately relate the conjugacy classes of
with the Abelianization of . In the non-Abelian case, however, this relation is not so clear.
Moreover, unlike the Abelianization of a group, which itself is a group, there is no natural group
structure on conjugacy classes. Nevertheless, we fully expect that (1.5) properly describes the
global symmetries of the theory on both the gauge theory and the string theory sides of the
duality.
4. Examples
As indicated above, we are interested in N = 1 quiver theories that may be obtained by orbifolding N = 4 super-YangMills by a group . To ensure N = 1 supersymmetry, must be
restricted to be a discrete subgroup of SU(3). In fact, all such discrete subgroups have been
classified [15], and many of the resulting quiver gauge theories have been described in [1618].
We follow the discussion of [16] concerning the non-Abelian discrete subgroups of SU(3).
The relevant subgroups which are not contained in SU(2) fall into two infinite series, (3n2 ) and
(6n2 ), as well as the exceptional subgroups (36), (60), (72), (168), (216), (360)
and (36 3), (60 3), (168 3), (216 3), (360 3). We note that the subgroups
in the infinite series are subgroups of SU(3) for n = 0 mod 3, and SU(3)/Z3 for n = 0 mod 3.
Similarly, the latter set of exceptional subgroups are subgroups of SU(3), while the former set
are subgroups of SU(3)/Z3 [16].
In Appendix A, we compute the Abelianization of several of these groups. The results are
given in Table 1. Note that the Abelianization of is not necessarily related to the center of
SU(3), as groups such as Z2 and Z4 may arise for . In most cases, is given by a single
cyclic group Zk , in which case the discrete symmetry group of the quiver is a single copy of
Heis(Zk Zk ). The case of (3n2 ) for n = 0 mod 3 is somewhat interesting, though, as its
Abelianization contains two cyclic factors. We will highlight this case below by considering
(27). However, we start with the familiar example of the Z3 orbifold in order to introduce the
language of discrete symmetries constructed from one-dimensional representations of .
335
Table 1
Some subgroups of SU(3) and their Abelianization
(3n2 )
Z3 Z3
Z3
Z2
Z4
Z4
(6n2 )
(36)
(36 3)
for n = 0 mod 3
for n = 0 mod 3
4.1. The Z3 quiver

The discrete symmetries of the Z3 quiver formed the basis of the analysis of Gukov, Rangamani and Witten in [1]. Although this group is Abelian, we find this example instructive as it
enables us to emphasize the rle of one-dimensional representations as generators of the A type
permutation symmetries.
The cyclic group Z3 has only one generator, which we call A, and which satisfies
A3 = 1.
(4.1)
Since Z3 is Abelian, it only has one-dimensional representations. For the same reason, the conjugacy classes are given by just the individual group elements. The character table is then
10
11
12
1
1
1
1
A
1
A2
1
2
(4.2)
where is a cube root of unity.

To construct the Z3 quiver, we must specify how it acts on the space C3 . This corresponds
to specifying an appropriate faithful three dimensional representation to act on the global index,
which is also required to be a subgroup of SU(3). We take this to be

0 0
A= 0 0 .
(4.3)
0 0
Now it is easy to see what happens. The three dimensional representation is a reducible representation: 3 = 11 11 11 . It is three copies of the same representation, and so in the quiver
we expect three arrows all pointing in the same direction from a given node. Also, we expect
a global SU(3) symmetry because one may recombine like representations into each other. We
will check this at the end of the calculation.2 The structure of the quiver is given by
3 1i = 1i+1 + 1i+1 + 1i+1 .
(4.4)
Thus the quiver, shown in Fig. 1, has three arrows pointing from one node to the next in the cyclic
i where a (b) labels the node that
order 10 11 12 . We label the bifundamental fields by Ca,b
2 The global SU(3) symmetry is a statement that there are three identical representations acting on the global index,
and this remains as a global symmetry. This is analogous to the unbroken gauge factors. Recall that in the regular
representation, an n-dimensional representation appears n times, and it is this fact that leads to the unbroken U(n) gauge
group. In the holographic limit, such symmetries become SU(nN ).
336
Fig. 1. The Z3 quiver.
the arrow points from (to), and i is a global SU(3) index labeling which arrow of the three that is
being referred to, and which directly corresponds to the global index in the N = 4 theory from
which the above orbifold theory descends.
To compute the superpotential, we note that the ClebschGordan coefficients are all trivial (taken to be unity in our conventions), as there is only one way of combining onedimensional representations. In this case, the superpotential is easy to deduce, and is simply
i C j C k . As claimed, there is a global SU(3) symmetry which directly corresponds
W = ij k C01
12 20
to mixing the three irreducible representations in the 3 (reducible) representation.
Given the quiver and the superpotential, we are now in a position to highlight the global
symmetries A, B and C. Starting with the A type shift symmetry, we recall that it is generated by
the action of the one-dimensional representations on the nodes and links of the quiver according
to (2.6). Taking 11 to generate the A symmetry, we see that it simply maps the nodes cyclicly
10 11 12 . The mapping of the fields is also easy, as all ClebschGordan coefficients are
trivial, and is given simply by
A:
i
i
Ca,b
Ca+1,b+1
(4.5)
(where the node labels are taken mod 3). This is clearly a symmetry of the superpotential.
To obtain the B and C rephasing symmetries, we start with the adjacency matrix

0
3 3
a = 3 0
(4.6)
3 .
3 3 0
The anomaly equation (2.16) takes the form
a v 0 (mod 3),
(4.7)
and can be satisfied by any vector v of integers. Since all gauge groups have the same rank, an
appropriate choice of charge vector satisfying (2.22) with ti = 1 mod 3 is given by v = (0, 0, 1).
On the fields, this corresponds to a rephasing symmetry
B:
i
i
C0,1
C0,1
,
i
i
1 C1,2
,
C1,2
i
i
1 C1,2
,
C2,0
(4.8)
where = exp(2i/(3N )). Note that this symmetry applied three times gives a member of the
center of the gauge group, and so is gauge equivalent to the identity.
337
Clearly A and B do not commute. They in fact close on the rephasing symmetry given by the
vector v = (1, 0, 1), acting on the fields as
C:
i
i
C0,1
,
C0,1
i
i
C1,2
C1,2
,
i
i
2 C2,0
.
C2,0
(4.9)
Although it is not obvious here, A and C commute (up to a gauge transformation). This is because
the charges needed to close the AC commutator are v = (2, 1, 1), which rephases in the exact
same way as v = (3, 0, 0). The latter is in the center of the gauge group because all charges are
divisible by three. Examination of (4.5), (4.8) and (4.9) indicates that the symmetry generators
obey the relations
A3 = B 3 = C 3 = 1,
AC = CA,
AB = BAC,
BC = CB,
(4.10)
up to the center of the gauge group [1]. As expected, this is just the finite Heisenberg group
Heis(Z3 Z3 ).
4.2. The (27) quiver
This particular orbifold has been much studied in the literature, and is one of the simplest
examples of a non-Abelian orbifold [1622]. Here we will try to be as detailed as possible, so
that the general structure of the discrete symmetries becomes apparent in this example.
The group theory details for (27) are given in Appendix B. Here we note that, for any
n = 0 mod 3, the group (3n2 ) is a subgroup of the full SU(3), and has n2 /3 1 threedimensional representations and 9 one-dimensional representations. The three-dimensional representations may be labeled by integer pairs i, j with 0 i, j n, along with the equivalence
3i,j = 3j,ij = 3ij,i (where labels are taken mod n). The 30,0 , 3n/3,n/3 and 32n/3,2n/3 representations are reducible, and fall apart into the nine one-dimensional representations, which we
may label as 1i,j as follows:
3in/3,in/3
1i,0 , 1i,1 , 1i,2 .
(4.11)
In the present case of n = 3, we only have two three-dimensional representations (which we

Using the 3 representation as the faithful representation that acts on the
therefore label 3 and 3).
global symmetry index, one may easily deduce the structure of the quiver (see the table of tensor
products in Appendix B). This is given in Fig. 2. We label the fields pointing from the 1i,j node
as Ai,j and those pointing to the 1i,j as Bi,j . In addition, we label the three other fields (pointing
as Ci , i = 0, 1, 2.
from the 3 to the 3)
The superpotential may be obtained using the ClebschGordan coefficients given in Appendix B. The result is
W = i,j,k Bi,j Ai,j Ck ,
(4.12)
with the coefficients i,j,k given in Table 2. More succinctly, we have

i,j,k = i1,k j i+1,k ,
where all indices are to be taken modulo 3, and where 3 = 1.
(4.13)
338
Fig. 2. The (27) quiver.
Table 2
The superpotential coefficients i,j,k multiplying Bi,j Ai,j Ck . Here = e2 i/3 is a cube root of unity
i,j,k
B0,j A0,j
C0
C1
C2
0
0
0
1
1
1
1
2
B1,j A1,j
1
2
0
0
0
1
1
1
B2,j A2,j
1
1
1
1
2
0
0
0
To construct the shift symmetries, we may begin with the Z3 symmetry generated by 11,0
Although the 3 and 3 nodes are inert
which takes 1i,j 1i+1,j as well as 3 3 and 3 3.
under this transformation, the Ci links are permuted. The action of this symmetry, which we
denote A, on the bifundamentals is thus
A:
Ai,j Ai+1,j ,
Bi,j Bi+1,j ,
Ck Ck+1 .
(4.14)
The Abelianization of (27) contains a second Z3 factor, and so we expect a second Z3 symmetry of the quiver. Noting that the nine one-dimensional representations form the group Z3 Z3
under ordinary multiplication (1i,j 1k,l = 1i+k,j +l ), we may take this second Z3 to be generated by the action of 10,1 . This acts on the second index of the Bi,j and Ai,j fields. However, this
339
is accompanied by additional rephasings as well. The action of this Z 3 , which we call A , is3
A :
Ai,j Ai,j +1 ,
Bi,j i Bi,j +1 ,
Ck 1k Ck .
(4.15)
We note that Z3 and Z 3 in (4.14) and (4.15) do not strictly commute. However, they commute up
to a rephasing of the fields
Ai,j Ai,j ,
Bi,j Bi,j ,
Ck 2 Ck ,
(4.16)
which is in the center of the gauge group associated with the 3 representation. This is because
is a third root of unity, and so is also a 3N th root of unity, thus ensuring that the above rephasing
is in the center of the SU(3N ) gauge group associated with the 3 node. Thus the (27) quiver
does in fact admit a Z3 Z3 symmetry. Incidentally, we note that the quiver diagram shown in
Fig. 2 actually admits an S9 symmetry permuting the nine singlet nodes. The superpotential,
however, is only invariant under the Z3 Z3 subgroup of the full S9 permutation group.
We now turn to the rephasing symmetries. Since the A type symmetries generate Z3 Z3 ,
we expect the B type rephasings to generate this identical group. These rephasings are relatively
easy to deduce from the results for the Z3 orbifold theory [1] given above. There, the appropriate
rephasings were given by the charges (0, 0, 1) on the three nodes of the quiver. Therefore, by
extension, we find the charge assignments for the nine SU(N ) nodes to be correctly given by
(0, 0, 0, 0, 0, 0, 1, 1, 1) for what we call B, and (0, 0, 1, 0, 0, 1, 0, 0, 1) for what we call B (the
SU(3N) nodes are assigned zero charge). This generates the rephasings
B:

B:
Bi,j 1 Bi,j ,
Ai,j Ai,j ,
Ai,j Ai,j ,
Bi,j
Bi,j ,
i = 2 only,
(4.17)
j = 2 only,
(4.18)
where the other fields are not rephased. Here = exp(2i/(3N )) is a primitive 3N th root of
unity.
It is clear that the primed symmetries commute with unprimed symmetries. However, A and
B do not commute, and A and B do not commute. Instead, they commute up to the respective
symmetries
C:

C:
Ai,j i1 Ai,j ,
Ai,j
j 1
Ai,j ,
Bi,j (i1) Bi,j ,

Bi,j
(j 1)
Bi,j .
(4.19)
(4.20)
Again, it is clear that primed and unprimed symmetries commute. It is also clear that B and C
(primed or not) commute. Finally, A and C also commute, but only up to the center of the gauge
group. In particular, the member of the center of the gauge group that closes the AC commutator
is
Ai,j 2 Ai,j ,
Ai,j Ai,j ,
Bi,j 2 Bi,j ,
Bi,j
Bi,j ,
i = 0,
i = 0,
(4.21)
and the member of the center of the gauge group that closes the A C commutator is
Ai,j 2 Ai,j ,
Ai,j Ai,j ,
Bi,j 2 Bi,j ,
Bi,j
Bi,j ,
j = 0,
j = 0,
(4.22)
3 Equivalently, one may choose to rephase A

i,j and leave Bi,j alone. This symmetry is pure gauge, given by simulta-
hence leaving the C fields unchanged.

neously using the center of the gauge groups 3 and 3,
340
where in both cases the Ck are unchanged. These two rephasings can be seen to correspond to
the center of the gauge groups by assigning charge 3 to each of the SU(3N ) nodes and then
assigning (3, 3, 3, 0, 0, 0, 0, 0, 0) and (3, 0, 0, 3, 0, 0, 3, 0, 0) charge vectors to the SU(N ) nodes,
respectively.
Thus the structure of the global symmetries is that each Z3 shift symmetry (A or A ) associated with a one-dimensional representation has a corresponding rephasing Z3 symmetry
(B or B ). These two symmetries close on a final Z3 symmetry (C or C ). All generators are
of order three (up to the center of the gauge group), and the primed and unprimed ones commute.
As a result, taken together, they form a direct product of two Heisenberg groups
Heis(Z3 Z3 ) Heis(Z3 Z3 ),
(4.23)
in agreement with the expectation from (1.5), where we note that the Abelianization of (27) is
just Z3 Z3 .
5. Conclusions
Our work demonstrates that quiver gauge theories obtained as worldvolume theories on a
stack of N D3-branes on the singular point of C3 / where is non-Abelian admit a discrete
group of global symmetries which may be expressed as a product of Heisenberg groups. It is
worth mentioning, however, that the actual matching which we perform is with string theory
on the near horizon manifold AdS5 S 5 / . This highlights the importance of the decoupling
limit which is accompanied with the decoupling of various U(1)s, some of them anomalous. We
believe a similar structure should exist in the general case of quiver gauge theories obtained as
worldvolume theories on a stack of N D3-branes placed at the singular point of a toric variety
which can be obtained as a non-Abelian orbifold of some other toric variety.
More generally, our series of papers [46] suggests that field theoretical methods might help in
answering in full generality the question of the spectrum of D-brane charges in string theories on
curved backgrounds and with fluxes whenever they admit field theory duals. It has clearly been
established that the Heisenberg group structure is present in a variety of situations, including
cascading theories and Seiberg dual phases. The larger question towards which our work points
is the computation of the spectrum on branes using AdS/CFT, therefore predicting the outcome
of the corresponding generalized cohomology theory classifying the D-brane charges in string
theory. For example, a question arises motivated by a recent comment made in [23] about twisted
K-theory being able to classify universality classes of baryonic vacua in the KlebanovStrassler
background. It is worth point out that the argument of [23] is entirely based on the geometry of
fluxes in the supergravity background.
Given our previous work with cascading quiver gauge theories [5], it is natural to suspect
that field theoretic methods along the lines described here should be relevant to understanding
the spectrum of D-brane charges in those backgrounds. Note that, to a large extent, the states
charged under the symmetries A, B and C are determinant operators in the field theory, also
referred to as baryonic. Here we would also like to stress that while we have considered certain
global symmetries, we have not considered all possible symmetries, and a full classification
and their dual interpretation would be enlightening. For example, the continuous U (1) baryon
number symmetries (associated with various fractional branes) do not commute with the shift
operators, and close on other baryon U (1) symmetries. We hope to return to some of these issues
in the future.
341
Acknowledgements
We are thankful to P. de Medeiros, Y.-H. He, C. Herzog, I. Kriz, D. Morrison and B. Uribe
for various comments about mathematical aspects of our work. J.T.L. is especially grateful to
S. Teraguchi for raising the issue of non-Abelian orbifolds and for suggesting that the previous
work on discrete symmetries may be extended to cover more general orbifold constructions. This
work is supported in part by the US Department of Energy under grant DE-FG02-95ER40899,
the National Science Foundation under grant PHY99-07949, the Israel Science Foundation, and
the GermanIsraeli Project (DIP) program (H.52). J.T.L. wishes to acknowledge the hospitality
of the Department of Physics at National Taiwan University and the Taiwan National Center
for Theoretical Sciences where this project was initiated, and the Kavli Institute for Theoretical
Physics where portions of this work were completed.
Appendix A. The Abelianization of G
We have seen that the Abelianization of the orbifold group (which we denote
/[, ]) plays a key rle in understanding the discrete global symmetries of the orbifold quiver
gauge theory. In particular, for a quiver constructed as an orbifold of N = 4 super-YangMills
by via the process in [8], (which is isomorphic to the group formed by the one-dimensional
representations of under ordinary multiplication) generates a set of permutation symmetries
mapping fields to fields and gauge groups to gauge groups. We expect this permutation mapping
to work for non-supersymmetric orbifolds as well as supersymmetric ones.4 However here we
restrict ourselves to the supersymmetric case. In this case, is a discrete subgroup of SU(3), and
the categorization of these subgroups is known [15]. In this appendix, we compute the Abelianization of for the following examples:

3n2 ,
(A.1)
6n2 ,
(36),
(36 3).
A.1. The case = (3n2 )
The group (3n2 ) is generated by
A3 = B n = C n = 1,
(A.2)
BA = AC 1 ,
(A.3)
and
CA = ABC 1
(with all other generators commuting). As a result, the commutator subgroup is generated by the
elements {B 1 C 1 , BC 2 }. This is equivalent to taking the generators

for n = 0 mod 3;
{BC, C} for n = 0 mod 3.
BC, C 3
(A.4)
4 In the non-supersymmetric case, one simply writes two different kinds of arrows in the quiver: one type for fermions,
and one type for bosons. Since the gauge factors for both the fermions and bosons are mapped in the same way, one still
expects the one-dimensional representations to generate symmetries of the resulting theory, again because the regular
representation is used on the gauge indices. One can similarly argue that the (non-super) potential that arises in these
cases is still preserved by these symmetries. It is interesting to see that this structure holds even for non-supersymmetric
orbifolds.
342
Since a generic (3n2 ) element can be written as

g = Aa B b C c = Aa (BC)b C cb ,
(A.5)
it is easy to see that the Abelianization of = (3n2 ) is

Z3 Z3 if n = 0 mod 3,
=
if n = 0 mod 3.
Z3
(A.6)
A.2. The case = (6n2 )

The group (6n2 ) is generated by
A21 = A32 = B n = C n = 1,
(A.7)
with
A2 A1 = A1 A1
2 ,
BA2 = A2 C
BA1 = A1 C,
CA2 = A2 BC
CA1 = A1 B,
1
(A.8)
Note that {A1 , A2 } generate the permutation group S3 . Thus

Zn ) S3 . A generic element of (6n2 ) may be written as
(6n2 )
is isomorphic to (Zn
g = Aa11 Aa22 B b C c ,
(A.9)
and the commutator subgroup is generated by

{A2 , B, C}.
(A.10)
As a result, the Abelianization of = (6n2 ) is simply the group generated by A1 . Thus

= Z2 ,
(A.11)
and is independent of any n mod 3 issues.

A.3. The case = (36)
The group (36) is generated by
V 4 = A3 = B 3 = 1,
(A.12)
along with
AV = V B,
BV = V A1 .
(A.13)
Elements of (36) may be written as

g = V v Aa B b .
(A.14)
Since the commutator subgroup is generated by

{A, B},
(A.15)
the Abelianization of = (36) is the group generated by V . This gives

= Z4 .
(A.16)
343
A.4. The case = (36 3)

The group (36 3) is a central extension of (36). It is generated by
V 4 = A3 = B 3 = C 3 = 1,
(A.17)
with C a central extension, so that

AB = BAC,
BV = V A1 .
AV = V B,
(A.18)
Elements of (36 3) may be written as

g = V v Aa B b C c .
(A.19)
Since A and B no longer commute, we ought to be a bit careful in determining the generators
of the commutator subgroup. From the above expressions, we see that the commutator subgroup
is generated by {A1 B, B 1 A1 , C}. Multiplying the first two generators in order (and using
A3 = 1) gives A. Since this is now part of the commutator subgroup, it can be multiplied with
A1 B to obtain B. As a result, the commutator subgroup is generated by
{A, B, C},
(A.20)
so that it is in fact isomorphic to the Heisenberg group (27). Taking a quotient of {V , A, B, C}

by {A, B, C} demonstrates that the Abelianization of = (36 3) is the group generated
by V . Thus
= Z4 .
(A.21)
It is not particularly surprising that this is the same result as the Abelianization of (36).
Appendix B. The group theory of (27)
Although the finite subgroups of SU(3) have been well studied [15,24,25], we will try to make
it accessible to the careful reader by displaying the basic group theoretic properties used in the
above calculations. Here, we consider the case where = (27). This group is a single group in
the series of groups (3n2 ) contained in SU(3). We take the presentation of (27) as follows:
A3 = B 3 = C 3 = 1,
AB = BAC,
AC = CA,
BC = CB.
(B.1)
This group has nine one-dimensional representations and two three-dimensional representations.
These are given in Table 3, where we have chosen a convenient set of labels.
Matching the eleven irreducible representations, the group (27) also has eleven conjugacy
classes. This gives rise to the (partial) character table
10,0
10,1
11,0
11,1
12,1
3
1
1
1
1
1
1
3
C
1
1
1
1
1
3
C2
1
1
1
1
1
3 2
A
1
A2
1
2
1
2
2
0
B
1
1
2
0
B2
1
1
2
2
AB
1
2
1
0
A2 B
1
2
AB 2
1
2
1
2
0
A2 B 2
1
2
2
1
0
(B.2)
344
Table 3
The eleven irreducible representations of (27). Here = e2 i/3 is a cube root of unity
Rep
10,0
10,1
10,2
11,0
11,1
11,2
12,0
12,1
12,2
2
1
2
1
0
0
1
0
0
1
1
1
1
2
2
2
1 0 0
0 0
1 0 2
1 0 0
0 2 0
1 0
1
1
1
1
1
1
1
1
1
1
0
0
1
0
0
0
1
0
0
1
0
0 0
0 0
0 0
2
0
0
0 2 0
0
0 2
The remaining characters are given by complex conjugation of the above representations and
characters. Explicitly, we take 2 in the characters, and map the labels for the representations by 3 3 and 1i,j 13i,3j , where these indices are taken mod 3. Also, note that the
conjugacy classes for the last eight columns above contain three elements: the element given, as
well as its product with C and C 2 .
The multiplication table for the above representations is given by
1k,l
3
3
1i,j
3
3
3 3 3
(1) (2)
(3)
1
i,j
i,j
1i+k,j +l
3
3
3
3
1i,j
3(1) 3(2) 3(3)
(B.3)
i,j
For our purposes, we desire the exact form of the superpotential, which may be obtained from
the appropriate set of ClebschGordan coefficients. In particular, we will need the coefficients
in the decomposition 3 ri rj . We will label the (trivial) vector that 1i,j acts on as i,j and
the vector that 3 acts on as triplets j . If multiple copies of a certain representation appear in a
product (on the left) or sum (on the right) of representations, an index in parentheses will appear
on both the representation and the vectors associated with it. The vectors transforming under
a barred representation will simply be labeled by putting a bar over the vectors (although one
should not take this as complex conjugation). In the following, we will show the direct product

(f )
(m)
(n)
and its resultant sum. We will display the sum ri rj = k fij rk ij and then use the
(fij )
notation rj
: (Vfijj ) with
v
vk
= Vrvii(m) Vrjj(n) (Yfij )vvij v k ,
V(f
ij )
(B.4)
so that one may directly read off the ClebschGordan coefficients. Again, we stress that the
numbers in parentheses will only be used to disambiguate two identical representations, and
further will be placed either as a superscript or subscript.
345
For tensor products of the one-dimensional representations with 3, we have

1 0,0
3 10,0 = 3 : 2 0,0 ,
3 0,0

1 0,1
3 10,1 = 3 : 2 0,1 ,
2 3 0,1

3 1,0
3 11,0 = 3 : 1 1,0 ,
2 1,0
2

3 1,1
,
3 11,1 = 3 :
1 1,1
2 1,1

2 2,1
2
3 12,1 = 3 : 3 2,1 ,
1 2,1

1 0,2
2
3 10,2 = 3 : 2 0,2 ,
3 0,2

2 2,0
3 12,0 = 3 : 3 2,0 ,
1 2,0

2 2,2
2
3 12,2 = 3 : 3 2,2 ,
1 2,2

3 1,2
2
3 11,2 = 3 : 1 1,2 .
2 1,2
(B.5)
Similarly, one can work out how the one-dimensional representations act on the 3 representation

1 0,0
2 0,0 ,
3 0,0

1 0,1
3 10,1 = 3 : 2 0,1 ,
2 3 0,1

2 1,0
3 11,0 = 3 : 3 1,0 ,
1 1,0

2 1,1
3 11,1 = 3 : 2 3 1,1 ,
1 1,1
2

3 2,1
,
3 12,1 = 3 :
1 2,1
2 2,1
3 10,0 = 3 :

1 0,2
2
3 10,2 = 3 : 2 0,2 ,
3 0,2

3 2,0
3 12,0 = 3 : 1 2,0 ,
2 2,0

3 2,2
3 12,2 = 3 :
,
1 2,2
2 2 2,2
2

2 1,2
3 11,2 = 3 : 3 1,2 .
1 1,2
Next, we display similarly the product of the 3 with itself:

3 ;
3(1) 3(2) = 3 (1)
3
(2)
(3)
(1) (2)
1 1
(1) (2)
:
3 (1)
2 2 ,
(1) (2)
3 3
(1) (2)
2 3
(1) (2)
:
3 (2)
3 1 ,
(1) (2)
1 2
(B.6)
346
(2)
(1)
3 2
(1) (2)
3 (3)
: 1 3 ,
(B.7)
(1) (2)
2 1
and of 3 with 3
3 3 = i,j 1i,j ;
1
10,0 : (1 1 + 2 2 + 3 3 ),
3

1
10,1 : 1 1 + 2 2 + 2 3 3 ,
3

1
10,2 : 2 1 1 + 2 2 + 3 3 ,
3
1
11,0 : (1 3 + 2 1 + 3 2 ),
3
1
12,0 : (1 2 + 2 3 + 3 1 ),
3

1
11,1 : 1 3 + 2 1 + 2 3 2 ,
3

1
12,2 : 1 3 + 2 2 1 + 3 2 ,
3

1
12,1 : 1 2 + 2 3 + 2 3 1 ,
3

1
11,2 : 1 2 + 2 3 + 2 3 1 .
3
In addition to these, we always take the ClebschGordan coefficients for the product of two onedimensional representations to be simply 1. Also, theClebschGordan
coefficient associated

with the product r r 10,0 is always taken to be (1/ dim r) fi fi in the usual way.
We take the expressions for the ClebschGordan coefficients to be commutative if two different representations are commutative, even though the tensor product of two matrices is not.
We also take that the ClebschGordan coefficients for the 3 3 can be chosen to be the same as
the 3 3 by simply barring all quantities and coefficients. (However they will not be needed.)
The above ClebschGordan coefficients are defined only up to a phase for each representation
in the sum. This means that, although we have displayed above that the phases of the 10,1 and
(10,1 ) = 10,2 representations appearing in 3 3 are related, they are in fact not related; we can
independently rephase the two representations. This corresponds directly to rephasing the fields.
We have chosen these phases to make a certain symmetry manifest in the resulting field theory.
We make one final note on this choice of ClebschGordan coefficients. While the tensor
product of three matrices is associative, the above choice for ClebschGordan coefficients is
not. However, this does not affect the physics. The above prescriptions define how to construct
singlets out of product representations. Any method for arriving at these will be physically
equivalent, as the number of linearly independent singlet functions is fixed for a given product. Therefore, the only possibility is that two different conventions are related by some unitary
transformation (taking that both conventions involve unitary ClebschGordan coefficients). This
347
corresponds to both the rephasing of the fields mentioned above, but also includes mixing different fields represented by arrows stretching between the same two nodes in the quiver. These
redefinitions are always available after constructing the field theory using a given prescription,
and as such different choices of ClebschGordan maps directly to field redefinition.
References
[1] S. Gukov, M. Rangamani, E. Witten, Dibaryons, strings, and branes in AdS orbifold models, JHEP 9812 (1998)
025, hep-th/9811048.
[2] D.S. Freed, G.W. Moore, G. Segal, The uncertainty of fluxes, Commun. Math. Phys. 271 (2007) 247, hepth/0605198.
[3] D.S. Freed, G.W. Moore, G. Segal, Heisenberg groups and noncommutative fluxes, Ann. Phys. 322 (2007) 236,
hep-th/0605200.
[4] B.A. Burrington, J.T. Liu, L.A. Pando Zayas, Finite Heisenberg groups in quiver gauge theories, Nucl. Phys. B 747
(2006) 436, hep-th/0602094.
[5] B.A. Burrington, J.T. Liu, L.A. Pando Zayas, Central extensions of finite Heisenberg groups in cascading quiver
gauge theories, Nucl. Phys. B 749 (2006) 245, hep-th/0603114.
[6] B.A. Burrington, J.T. Liu, M. Mahato, L.A. Pando Zayas, Finite Heisenberg groups and Seiberg dualities in quiver
gauge theories, Nucl. Phys. B 757 (2006) 1, hep-th/0604092.
[7] S. Kachru, E. Silverstein, 4d conformal theories and strings on orbifolds, Phys. Rev. Lett. 80 (1998) 4855, hepth/9802183.
[8] A.E. Lawrence, N. Nekrasov, C. Vafa, On conformal field theories in four dimensions, Nucl. Phys. B 533 (1998)
199, hep-th/9803015.
[9] D.R. Morrison, M.R. Plesser, Non-spherical horizons. I, Adv. Theor. Math. Phys. 3 (1999) 1, hep-th/9810201.
[10] M.R. Douglas, G.W. Moore, D-branes, quivers, and ALE Instantons, hep-th/9603167.
[11] M.R. Douglas, B.R. Greene, D.R. Morrison, Orbifold resolution by D-branes, Nucl. Phys. B 506 (1997) 84, hepth/9704151.
[12] M.R. Douglas, B.R. Greene, Metrics on D-brane orbifolds, Adv. Theor. Math. Phys. 1 (1998) 184, hep-th/9707214.
[13] A. Hatcher, Algebraic Topology, Cambridge Univ. Press, Cambridge, 2002, online at http://www.math.cornell.edu/
~hatcher/AT/ATpage.html.
[14] L.J. Dixon, J.A. Harvey, C. Vafa, E. Witten, Strings on orbifolds. 2, Nucl. Phys. B 274 (1986) 285.
[15] H.F. Blichfeldt, Finite Collineation Groups, University of Chicago Press, Chicago, 1917.
[16] A. Hanany, Y.H. He, Non-Abelian finite gauge theories, JHEP 9902 (1999) 013, hep-th/9811183.
[17] B. Feng, A. Hanany, Y.H. He, N. Prezas, Discrete torsion, non-Abelian orbifolds and the Schur multiplier,
JHEP 0101 (2001) 033, hep-th/0010023.
[18] B. Feng, A. Hanany, Y.H. He, N. Prezas, Discrete torsion, covering groups and quiver diagrams, JHEP 0104 (2001)
037, hep-th/0011192.
[19] T. Muto, D-branes on three-dimensional non-Abelian orbifolds, JHEP 9902 (1999) 008, hep-th/9811258.
[20] T. Muto, Brane cube realization of three-dimensional non-Abelian orbifolds, JHEP 0002 (2000) 026, hepth/9912273.
[21] T. Muto, Brane configurations for three-dimensional non-Abelian orbifolds, hep-th/9905230.
[22] H. Verlinde, M. Wijnholt, Building the standard model on a D3-brane, JHEP 0701 (2007) 106, hep-th/0508089.
[23] J. Evslin, What does(nt) K-theory classify?, hep-th/0610328.
[24] A. Bovier, M. Luling, D. Wyler, Finite subgroups of SU(3), J. Math. Phys. 22 (1981) 1543.
[25] W.M. Fairbairn, T. Fulton, W.H. Klink, Finite and disconnected subgroups of SU 3 and their applications to the
elementary particle spectrum, J. Math. Phys. 5 (1964) 1038.

Harmonicity in N = 4 supersymmetry and its

quantum anomaly
I. Antoniadis a,1 , S. Hohenegger a, , K.S. Narain b , E. Sokatchev c
a Department of Physics, CERN Theory Division, CH-1211 Geneva 23, Switzerland
b High Energy Section, The Abdus Salam International Center for Theoretical Physics,
Strada Costiera, 11-34014 Trieste, Italy

c Laboratoire dAnnecy-le-Vieux de Physique Thorique LAPTH 2 , B.P. 110, F-74941 Annecy-le-Vieux, France
Received 10 October 2007; accepted 7 November 2007

Abstract
The holomorphicity property of N = 1 superpotentials or of N = 2 F-terms involving vector multiplets is
generalized to the case of N = 4 1/2-BPS effective operators defined in harmonic superspace. The resulting
harmonicity equations are shown to control the moduli dependence of the couplings of higher-dimensional
operators involving powers of the N = 4 Weyl superfield, computed by N = 4 topological amplitudes.
These equations can also be derived on the string side, exhibiting an anomaly from world-sheet boundary
contributions that leads to recursion relations for the non-analytic part of the couplings.
PACS: 11.25.Mj; 11.30.Pb; 04.65.+e
Keywords: Topological amplitudes; Harmonic superspace; Supergravity; Compactification
1. Introduction
An important property of N = 2 F-terms involving vector multiplets is holomorphicity, implying that the corresponding couplings are holomorphic functions of the vector moduli. This
E-mail addresses: ignatios.antoniadis@cern.ch (I. Antoniadis), stefan.hohenegger@cern.ch (S. Hohenegger),

narain@ictp.trieste.it (K.S. Narain), emeri.sokatchev@cern.ch (E. Sokatchev).
1 On leave from CPHT (UMR CNRS 7644) Ecole Polytechnique, F-91128 Palaiseau.
2 UMR 5108 associe lUniversit de Savoie.
doi:10.1016/j.nuclphysb.2007.11.005
I. Antoniadis et al. / Nuclear Physics B 794 (2008) 348380
349
applies, for instance, to the couplings Fg of the higher dimensional F-terms W 2g , where W is the
self-dual (chiral) Weyl superfield, appearing in the string effective action [1]. On the other hand,
the couplings Fg are computed by the topological partition function of the N = 2 twisted Calabi
Yau -model associated to the six-dimensional compactification manifold of type II string theory
in four dimensions [1,2]. It turns out, however, that there is a holomorphic anomaly, related to
a violation of the conservation of the BRST current in the topological theory, implying that antichiral fields do not decouple at the quantum level [24]. The anomaly arises from boundary
contributions and takes the form of an equation that amounts to a recursion relation for the nonholomorphic part of the couplings Fg . From the point of view of the string effective action, it
arises from the quantum integration over massless states that is unavoidable when computing
on-shell physical amplitudes [1].
These couplings were generalized recently to 1/2-BPS terms of N = 4 compactifications, involving powers of the (superdescendant of the) N = 4 chiral Weyl superfield K ++ = D D W ,
where D are particular SU(4) projections of the spinor derivatives. We recall that the N = 4
gravity multiplet contains, besides the graviton and the four gravitini, six graviphotons, one complex graviscalar and four spin-1/2 Weyl fermions [5]. Moreover, there is an SU(4) R-symmetry,
transforming the gravitini in the fundamental and the graviphotons in the vector representation.
The (linearized on-shell) superfield K ++ satisfies 1/2-BPS shortening conditions. Its lowest
component is the (self-dual) graviphoton field strength and its next bosonic components are
the (self-dual) Riemann tensor and the second derivative of the graviscalar. Another basic 1/2BPS superfield in the N = 4 theory is the (linearized on-shell) vector multiplet Y ++ . Its lowest
component are the scalar moduli transforming in the vector representation of SU(4), like the
graviphoton field strengths.
(1)
(3)
(1)
In [5] two series of 1/2-BPS couplings were found: Fg K 2 K 2g and Fg K 2(g+1) . Here Fg
(3)
and Fg are functions of the N = 4 moduli vector multiplets Y ++ and of the SU(4) harmonic
variables that can again be computed by topological amplitudes on K3 T 2 of genus g and
(6d) 4g
g + 1, respectively. Actually, in six dimensions there is also the series Fg W6d , where W6d
is a similar Weyl superfield of the six-dimensional gravity multiplet and Fg(6d) is given by a
topological theory on K3 [6].
In this work, we study the question of what is the analog of N = 2 holomorphicity for
such 1/2-BPS N = 4 couplings. The main novelty of the generalization is that the relevant
R-symmetry group becomes non-Abelian, transforming non-trivially the superfields K ++ and
Y ++ . As a consequence, the notion of chirality of the N = 2 theory is replaced by Grassmann
analyticity (or 1/2-BPS shortness). The natural framework for studying this problem and covariantizing the expressions is then harmonic superspace [79]. By introducing SU(4) harmonic
variables one can define K ++ as a particular harmonic projection of the sixplet of superfields
Kij = Kj i = Di Dj W , associated to a corresponding 1/2-BPS subspace of the full N = 4
superspace. Supersymmetry then implies that the coupling coefficients Fg are functions of the
same harmonic projected vector superfields Y ++ living in the same 1/2-BPS subspace. Thus,
Fg (Y ++ ) is independent of the five remaining projections of the sixplet of the scalar moduli.
This defines a notion of analyticity that naturally generalizes N = 2 holomorphicity for the chiral N = 2 vector multiplets.
In this work, we show that the above property of analyticity can be formulated in terms of
a set of differential constraints on the couplings Fg of N = 4 1/2-BPS effective operators.
They express the property of the analytic functions Fg (Y ++ ) that, when expanded in powers
of the harmonic variables and the scalar fields, the coefficients should form symmetric trace-
350
less tensors of SO(6). This yields two non-trivial equations. The first requires one scalar and
one harmonic derivative to vanish and coincides with the so-called harmonicity equation found
previously in string computations, up to an anomaly [5,6,10]. The second involves two scalar
derivatives and gets modified in supergravity by an additive constant term due to the curvature
of the scalar kinetic terms. Both equations are checked by an explicit string computation for
(3)
Fg , which receives one-loop corrections on the heterotic side for all g, and are found to be
corrected by anomaly terms due to world-sheet boundary contributions that spoil the naive expectation of analyticity, in a way similar to the holomorphic anomaly equation of the N = 2
Fg s. The resulting equations are reduced again to recursion relations for the non-analytic part of
the moduli-dependent couplings.
We finally extend the above results to six-dimensional N = (1, 1) supersymmetry, where the
(3)
R-symmetry group is SO(4). In particular, we consider the decompactification limit of Fg that
2(g+1)
. These
gives rise to a new six-dimensional series of 1/2-BPS couplings of the form Fgdec W6d
couplings, although not exactly topological in six dimensions (the spacetime part is not decoupled), become topological upon compactification to four dimensions on a two-torus. Despite this
4g
(6d)
fact, Fgdec satisfy the same analyticity condition as Fg of W6d since they are both 1/2-BPS.
We then derive the corresponding analyticity equations, together with the anomaly terms.
The paper is organized as follows. In Section 2, we describe the analyticity conditions of the
1/2-BPS couplings in the case of global N = 4 supersymmetry. We first introduce the SU(4)
harmonic variables and the harmonic projected vector and Weyl superfields, Y ++ and K ++ respectively. We then derive the differential equations for the couplings Fg of the higher-derivative
operators involving powers of (K ++ )2 , as described above. In Section 3, we study the effects of
the curvature of the scalar manifold that parametrizes the coset SO(6, n)/SO(6) SO(n), where
n is the number of vector multiplets. We show in particular that the second-order derivative equation in the scalar fields gets modified by an additional term proportional to the Weyl weight of
the operator. In Section 4, we go to curved superspace in the framework of N = 4 conformal supergravity and derive the fully covariantized final expressions of the higher-derivative couplings.
In Section 5, we give a brief review of the N = 4 topological amplitudes in string theory and
(3)
recall the expression for Fg obtained from a one-loop string computation on the heterotic side.
In Section 6, we derive the harmonicity relation which is first order in the scalar field derivatives,
exhibiting a boundary anomaly that invalidates the expected vanishing result. In Section 7, we
obtain the second-order constraint which is also modified by an anomaly. In Section 8, we generalize our analysis to six dimensions. We first introduce the SO(4) harmonic variables and derive
the harmonicity equations for the couplings of the 1/2-BPS terms. We then consider the decom(3)
pactification limit of Fg and compute the two analyticity equations modified by the anomalous
terms. Finally, Section 9 contains some concluding remarks.
2. Global N = 4 supersymmetry
2.1. SU(4) harmonic variables
We consider N = 4 supersymmetry in four dimensions whose automorphism group is SU(4).
We introduce harmonic variables [79] on the coset SU(4)/S(U (2) U (2)) in the form of maa
trices (u+a
i , ui ) SU(4). They have an index i = 1, . . . , 4 transforming under the fundamental
irrep of SU(4) and indices a, a = 1, 2 of SU(2) SU(2) as well as U (1) charges 1. Together
351
a
with their complex conjugates u i+a = (u+a
ia = (u
i ), u
i ) they satisfy the unitarity conditions
i+b = ba ,
u+a
i u
a i
u
b = ba ,
i u
a i
u+a
ib = u
+b = 0,
i u
i u
a
u+a
+a + u
a = i
i u
i u
j
(2.1)
and the unit determinant condition
+b a b
ab a b
ij kl u+a
i uj uk ul =
(2.2)
(with 1234 = 12 = 12 = 12 = 1 2 = 1).

The harmonic functions have harmonic expansions homogeneous under the action of the subgroup S(U (2) U (2)). The harmonic expansions are organized in irreps of SU(4), keeping the
balance of projected indices so that the overall SU(2) SU(2) indices and the U (1) charge are
always the same. In what follows we shall often make use of functions depending on vector-like
combinations of SU(4) harmonics (i.e., with harmonics on SO(6)/SO(4) SO(2)) of the type
ij
M
(and their conjugates u M = uM
uM
ij = uj i , M = (++, , a a)
ij )
+a
+b
u++
ij = ui ab uj ,
a
b
u
ij = ui a b uj ,
a
uaija = u+a
[i uj ] ,
(2.3)
ij
M
where [ij ] denotes weighted antisymmetrization. They form SO(6) matrices uM
N = N uij where
M are the gamma matrices of SO(6). The vector-like harmonics satisfy algebraic conditions
expressing the fact that uM
N SO(6) and following from the conditions on the underlying SU(4)
harmonics:
ij kl
ukl = 4,
u++
ij
uaija ij kl ubklb = ab a b ,
ij kl ++
ij kl
ij kl a a
ij kl a a
ukl = u
ukl = u++
ukl = u
ukl = 0,
u++
ij
ij
ij
ij
++
a a
bb
u++
ij ukl + uij ukl 2uij ab a b ukl = ij kl .
(2.4)
An example of a harmonic function which we shall frequently encounter is ++ (u) = ij u++

ij +
++ mn
mn u++
ij ukl u++ + . The first component in this expansion is a 6 of SU(4) (or a vector of
SO(6)) ij = j i . The higher components give rise to higher-dimensional irreps, but we shall
not need them here.
The harmonic derivatives can be viewed as the covariant derivatives on the harmonic coset
SU(4)/S(U (2) U (2)), or equivalently, as the generators of the SU(4) algebra written down in
an S(U (2) U (2)) basis (see Section 3). This means that they are invariant under the left action
of the group SU(4), but covariant under the right action of the subgroup S(U (2) U (2)). They
can be split into generators of the subalgebra S(U (2) U (2)):

trace,
D+a +b = u+b
+a
i
u+a
u i+b
i

b
i
b
Da = ui
u a i
trace,
a
u b
u
i
ij kl
352

D0 =
u+a
i
i+a i
+a u
ui
u +a
a
u
i
u ia i
a
u a
ui

(2.5)
and of the coset:
D+a b = uib
Da +b = u+b
i
u i+a i ,
u+a
i
b
a
u
i
u ia
u i+b
(2.6)
The harmonic derivatives are differential operators preserving the defining algebraic constraints
(2.1), (2.2).
The derivatives (2.5) act homogeneously on the harmonic functions. For instance, the function
++ (u) above has no SU(2) SU(2) indices, but has U (1) charge, hence
D+a +b ++ (u) = Da b ++ (u) = 0,
D0 ++ (u) = 2 ++ (u).
(2.7)
The harmonic expansion of this function defines an infinitely reducible representation of SU(4).
It can be made irreducible by requiring that the raising operator Da +b annihilate the function:
Da +b ++ (u) = 0 ++ (u) = ij u++
ij .
(2.8)
In other words, such a function is a highest-weight state of the 6 of SU(4). The irreducibility
condition (2.8) is also called a condition for harmonic (H-) analyticity.
2.2. Grassmann analytic on-shell superfields
The introduction of harmonic variables allows us to define 1/2 BPS short or Grassmann
(G-) analytic3 superfields.4 They depend only on half of the Grassmann variables which can

be chosen to be +a = i u+a
ia i . One such superfield is the linearized on-shell
i and a = u
vector multiplet

Y ++ x , + , , u
+a
i +b
a b
= ij u++
ia i
b
ij + ab ui + u
+ +a +b ab F(+) + a b a b F() + derivative terms.
(2.9)
Here ij = 12 ij kl kl are the six real scalars, i are the four Majorana gluinos and F() is
the (anti)self-dual part of the gluon field strength. To exhibit manifest G-analyticity, one has to
choose the appropriate analytic basis in superspace,
x x + i +a +a + i a a ,
(2.10)
analogous to the familiar chiral basis. Note that the harmonic dependence here is cut down to
linear in the vector-like and fundamental harmonics. This is typical for on-shell multiplets which,
3 A more systematic derivation of the G-analytic superfields as functions on a coset of the N = 4 superconformal
algebra PSU(2, 2/4) will be given in Section 4.

4 The notion of Grassmann analyticity (with breaking of the R symmetry) was first proposed in [11] in the context of
the N = 2 hypermultiplet. Later on it was made R-symmetry covariant in the framework of N = 2 harmonic superspace
in [7]. The harmonic superspace description of the N = 3 off-shell vector multiplet was given in [8], and that of the
N = 4 on-shell vector multiplet in [9].
353
in addition to the G-analyticity condition, also satisfy the H-analyticity condition

Da +b Y ++ + , , u = 0.
(2.11)
Here the harmonic derivative is supersymmetrized by going to the manifestly G-analytic superspace coordinates (2.10). One can show that the ultrashort on-shell superfield (2.9) is the
solution to the simultaneous conditions for G- and H-analyticity [9,12].
Another example of a G-analytic superfield is the linearized on-shell Weyl multiplet. It is
obtained from the off-shell chiral Weyl superfield [14]

j
W i = + i
(2.12)
T(+)[ij ] + S(ij ) + .
Here is the physical scalar and T(+) is the self-dual part of the sixplet of graviphoton field
strengths, while S is an auxiliary field. On shell the latter must vanish, hence the additional
constraint
Di Dj W = 0.
(2.13)
Now, define the superfield (a superdescendant of W )
++
a b
= ( ) D
K
a Db W
(2.14)
where we have projected the SU(4) indices of Di Dj with the harmonics u ia u b . This superfield
is annihilated by half of the spinor derivatives and hence is 1/2 BPS short. Indeed, this is true
+a
i
for the projections D +a
since {D , Db } = 0 and D W = 0 (chirality). Further, hitting (2.14)
with Da we obtain zero as a consequence of the projection of the on-shell constraint (2.13) with
++ satisfies the G-analyticity constraints
u ia u b . We conclude that K
j
++
+a ++
D
a K = D K = 0
(2.15)
which imply that it depends on half of the s (bosons only):

++ +
K
, , u
ij
+a +b
ab R(+) + a b a b + .
= T(+) u++
ij +
In addition, the harmonic dependence of

from the condition for H-analyticity
++
K
(2.16)
is restricted to be linear. As in (2.11), this follows
++
= 0.
Da +b K
(2.17)
This is another example of an ultrashort superfield. Note, however, that it is not a primary object
but rather a superdescendant of the chiral on-shell Weyl multiplet.
Repeating the same steps, but this time starting with the antichiral superfield W ( ) we obtain
the other half of the on-shell Weyl multiplet. It is again described by an ultrashort superfield of
the same type,

++ +
K
, , u
b a b R() + +a +b ab + .
= T() u++
ij + a
ij
(2.18)
Note that in the N = 4 G-analytic superspace there exists a special conjugation combining
complex conjugation with a reflection on the harmonic coset, such that G-analyticity is preserved.
++ and K
++ , which implies, in particular, the reality condition

++ = K
In this sense Y ++ = Y
on the six scalars in Y .
354
2.3. Higher-derivative couplings

After having defined the G-analytic superfields (2.9) and (2.16), we now want to construct the
corresponding effective action couplings. Recently, by studying special higher loop scattering
processes in the gravitational sector of type II superstring theory compactified on K3 T 2 (or
the corresponding dual formulation of heterotic string on T 6 , as we will review in Section 5), the
following two terms were found in [5]:

++ ++ ++ ++ g 4(g1) ++
K K
YA , u ,
F1
S1 = d 4 x du d 4 + d 4 K
(2.19)
K

++ ++ g+1 4(g1) ++
YA , u ,
K
F2
S2 = d 4 x du d 4 + d 4 K
(2.20)
where A = 1 n is an SO(n) vector index labeling the coordinates of the coset of physical
scalars (see Section 3). In fact, if considered as g- and (g + 1)-loop contributions respectively,
both of these terms lead to so-called topological amplitudes, that is the corresponding physical
amplitudes are computed by correlation functions of the N = 4 topological string on K3 T 2 .
However, unlike the N = 2 case (see [1]) these correlation functions are not simply the topological partition function, but differ from it by additional operator insertions in the twisted version
of the theory. Actually, for convenience and notational simplicity, we changed the notation of
4(g1)
4(g1)
(1)
(3)
the two couplings from [5]: F1
corresponds to Fg , while F2
corresponds to Fg ,
with the upper index denoting the U (1) charge.
It is important to stress upon two points concerning the effective action terms (2.19) and
(2.20):
1. The Grassmann measure is G-analytic, i.e. it involves only half of the projected s, and so
must be the integrand, otherwise supersymmetry will be broken. This is why we have to use
the linearized on-shell superfields Y ++ and K ++ which are G-analytic like the measure.
2. The harmonic integral should produce an SU(4) invariant, i.e. it picks out the SU(4) singlet
part of the integrand. This is only possible if the latter is a chargeless harmonic function.

ij
kl
For example, f (u) = f0 + fkl u++
ij u++ + integrates to du f (u) = f0 , but a charged
function like f ++ = f ij u++
ij + will have a vanishing integral. Notice that for this reason
the harmonic integral should always be done last, after the Grassmann integrals, since the
latter are charged.
In our case (2.19), (2.20) the functions F1,2 carry U (1) charge 4(g 1) needed to compensate that of the factor K (+4(g + 1)) and of the Grassmann measure (8). Given the fact that
the argument Y ++ of F has a positive charge, we have to introduce a set of constant SU(4)
multispinors
p p
1 1
2p (u) (i1 ip )(j1 jp ) u ++
u ++
+
i j
i j
(2.21)
thus explicitly breaking SU(4).5 The dots denote higher-order terms in the harmonic expansion
of the coefficients (u) which will not be of interest for us, see below. Note that the product of
vector-like harmonics forms an irreducible representation of SO(6), a symmetric traceless tensor
5 The other possibility, which we do not consider here, would be to use singular functions involving inverse powers of
fields.
355
of rank p (recall that (u ++ )2 = 0, see (2.4)). In SU(4) notation this means that the indices i and
j of the coefficients are separately symmetrized, but antisymmetrized between i and j , i.e. we
are dealing with the irrep (0p0). In what follows this fact will be of crucial importance. So, we
consider the potential (m = 2(g 1); the SO(n) index A and the labels 1, 2 are suppressed)

n

2(m+n) (u) Y ++ .
F 2m Y ++ , u =
(2.22)
n=0
The factors K in (2.19) and (2.20) contribute, among others, the terms
++ m
++ m
+ 4
+ 4
2
2
2
T(+) ,
( )4 R(+)
R()
( )4 R(+)
()2 T(+)
,
(2.23)
respectively. The s saturate the superspace measure and are integrated out. The remainder has
a harmonic charge,

++ m
++
= T i1 j1 T im jm u++
T
(2.24)
i 1 j 1 ui m j m
which is compensated by the factor F in order to have a non-vanishing harmonic integral (i.e.,
an SU(4) singlet). Clearly, (2.24) is an irrep of SO(6), a symmetric traceless tensor of rank m.
This can be reformulated as the highest-weight condition (cf. (2.8))

m
Da +b T ++ = 0.
(2.25)
A similar condition holds for the entire effective action expressions (2.19) and (2.20) of the
graviphoton field strength superfield. The singlet needed for the harmonic integral is obtained by
combining (2.24) with the matching irrep in F . Consider the harmonic structure of F (all = 0):

im+n jm+n
i 1 j1
(i1 im+n )(j1 jm+n ) u ++
u ++
F 2m ++ , u =
n=0
++
(k1 (l1 kn )ln ) u++
k1 l1 ukn ln .
(2.26)
Here we have restricted the harmonic expansion (2.21) of the coefficient function 2(m+n) (u)
to the lowest-rank SO(6) irrep. The higher-rank terms are irrelevant due to the gauge invariance of the couplings (2.19), (2.20). Indeed, consider adding a total supersymmetrized harmonic
derivative Da +b (4g+2) ab ( + , , u) to the potential F 4(g1) . After integration by parts (the
G-analytic measure allows this), Da +b annihilates the on-shell superfield K ++ (recall (2.17)),
hence the gauge invariance of (2.19), (2.20) with the G-analytic parameter . By examining the
harmonic expansion of (0, 0, u) one can show that all the omitted terms in (2.26) can be gauged
away.
The gauge-fixed function (2.26) satisfies two differential conditions. The first one expresses
the fact that it is a function only of the projection ++ of the SO(6) vector of physical scalars:
F 2m = a a F 2m = 0.
(2.27)
This is yet another kind of analyticity condition (S-analyticity), this time with respect to the
scalars (which in fact are the coordinates on the curved manifold SO(6, n)/SO(6) SO(n), see
Section 3). The second one restricts the harmonic dependence
D+a b F 2m = ab bb
F 2m .
++
(2.28)
356
Note that if the right-hand side in (2.28) vanished, this would be a condition defining a lowestweight state of SU(4) (or an SO(6) tensor of rank m). However, the dependence on the scalars
makes the harmonic structure in (2.26) reducible.
i 1 j1
i m jm
u ++
needed to
From (2.26) we have to extract the irreducible harmonic structure u ++
match the conjugate structure in (2.24). It is obtained by contracting all the u++ in (2.26) with a
ij
[i j ]
subset of the u ++ , using u ++ u++
kl = 1/3k l + (see (2.1)). This confirms that the omitted
terms in the harmonic expansion of in (2.26) cannot contributethey contain higher-rank
SO(6) tensors. The result is the relevant part of the function F , or the reduced function
i 1 j1
im jm im+1 jm+1
F 2m =
(2.29)
(i1 im+n )(j1 jm+n ) u ++
u ++
im+n jm+n .
n
Notice the full symmetrization of the i and j indices of the tensor inherited from (2.26). As
required, the reduced function is manifestly H-analytic (i.e., SU(4) irreducible),
D+a b F 2m = 0.
(2.30)
++ )
of (2.26) is lost.
However, now the manifest S-analyticity (i.e., the dependence only on
It should be made clear that (2.29) is just a rearrangement of the harmonic expansion of the
gauge-fixed function F 2m . The information contained in this function is encoded in the fact that
the coefficients (i1 im+n )(j1 jm+n ) , which are the same in (2.26) and (2.29), form the SU(4) irrep
(0, m + n, 0).6 This information can be translated into two types of differential constraints on the
function F
. In general, the harmonic and scalar factors in (2.29) form the reducible representation
(0m0) np=1 (010) (0, m + n, 0) + . The relevant projection (0, m + n, 0) is obtained
by symmetrizing all the i and separately all the j indices (the antisymmetry of the is with the j s
is automatic). Any other irrep in this tensor product will have a subset of the is (and of the j s)
antisymmetrized. The product of two us is irreducible, (010) (010) (020) as follows form
the commuting nature of the SU(4) harmonics u i+a . The antisymmetrization of indices carried
by the us
and the s is ruled out by the so-called harmonicity condition:
pqrs
q
rs F = 0,
u +a A
(2.31)
where we have restored the SO(n) index A and suppressed the U (1) charge superscript 2m.
The above equation forbids the decomposition (100) (010) (001). This constraint involves
partial derivatives with respect to u + . Strictly speaking, such an operation is illegal in the harmonic formalism, since the variables u are not independent, as can be seen from (2.1), (2.2).
However the above equation can be rewritten using covariant harmonic derivatives introduced in
(2.5) and (2.6) as

1 +a
+a
b
+a
D
+
D
+
u
D
pqrs u+b
(2.32)
u
+b
0
q
q
b
rs F = 0.
2 q
A
Indeed, it is easy to see that this equation reduces to (2.31) since our function F explicitly involves only u + harmonics. The D0 term in (2.32) is just to remove the contribution from the
trace parts in D+b +a as defined in (2.5) which measures the total U (1) charge 2m of F . In the
following however we will continue to write the formula using partial derivatives with respect
to u + .
6 SU(4) irreps with Dynkin labels (0p0) are equivalent to rank p symmetric traceless tensor of SO(6).
357
Further, the antisymmetrization of indices carried by the s is ruled out by the constraint
pqrs
F =0
pq
A Brt
(2.33)
which forbids the decomposition (010) (010) (101) (000). In SO(6) (vector) notation
(2.33) reads
[M
A
M
A
F = 0,
BN]
F = 0.
BM
(2.34)
(2.35)
Here we do not take into account the fact that the physical scalars parametrize a curved manifold and hence the derivatives in (2.33) should be considered covariant with respect to the metric
of the manifold. In Section 3 we show that this leads to a modification of (2.35) by a term proportional to AB .
3. The coset of physical scalars
3.1. The coset SO(6, n)/SO(6) SO(n)
Here we briefly recall why the scalars of N = 4 Poincar supergravity describe the coset
space SO(6, n)/SO(6) SO(n) [1315].
N = 4 Poincar supergravity is obtained by coupling the off-shell Weyl multiplet to 6 + n
free vector multiplets. The first six are compensating multiplets (i.e. their kinetic terms have
the wrong sign), the remaining n are physical. Each vector multiplet supplies 6 scalars, so the
M and the other 6 n by M , where
total number is 6(6 + n). We denote the first 6 6 by N
A
M, N = 1, . . . , 6 and A = 1, . . . , n are SO(6) and SO(n) vector indices, respectively.
The Weyl multiplet contains an auxiliary field DMN = DN M , DMM = 0 in the 20
of SO(6)
SU(4). It serves as a Lagrange multiplier for the following quadratic combination of scalars:
M N

M N
DMN K
(3.1)
K A
A .
So, it imposes an algebraic constraint which eliminates 20 of the scalars. In addition, one makes a
Weyl (dilatation) gauge choice for the trace of the quadratic form in (3.1), thus fixing yet another
scalar. So, the resulting condition is (up to normalization)
M N
M N
K A
A = MN .
K
(3.2)
Notice that this condition is invariant under local SO(6) which allows to gauge away 15 additional scalars. Altogether 36 scalars are eliminated and the remaining 6n do indeed parametrize
the coset SO(6, n)/SO(6) SO(n).
M
Conditions (3.2) can be solved by first fixing an SO(6) gauge such that the 6 6 matrix N
T
becomes symmetric, = , after which one can write down

= I + T .
(3.3)
We can say that the 6n physical scalars are the unconstrained coordinates on the coset
SO(6, n)/SO(6) SO(n).
358
3.2. Harmonic description

The higher-derivative terms (2.19), (2.20) involve the function (potential) F defined on the
coset of physical scalars. The peculiarity of this function is that it depends only on a single pro++
jection YA++ = A
(x, u) + of the six-vectors of coset coordinates, obtained with the help
of the SU(4) harmonic variables. This is a typical example of an analytic harmonic realization of a coset space. Another, very similar example is that of the N = 4 superconformal group
PSU(2, 2/4) realized on the Grassmann analytic superfields (2.9) (see Section 4). Here we explain this coset construction, following closely the case of N = 2 superconformal symmetry and
Poincar supergravity [1618] and of N = 2 quaternionic sigma models [1921].
We start by writing down the algebra of SO(6, n) in a basis suitable for the forthcoming
introduction of the harmonic variables (2.1). The SO(n) generators are MAB = MBA and the
SO(6) ones are written in an S(U (2) U (2)) basis. Thus, the S(U (2) U (2)) generators are

Z+a +b Z+a +a = 0 ,
(3.4)
Za b Za a = 0 ,
Z0 ,
and the remaining generators of SU(4) are Z+a b and Za +b . Finally, the generators of the
coset SO(6, n)/SO(6) SO(n) are LAa a , LA++ and LA . Then the algebra of SO(6, n) takes
the form

c
[LAa a , LBbb ] = AB a b (ac Z+b) +c + ab (a c Zb)
+ ab a b MAB ,
[LAa a , LB++ ] = AB a b Z+a b ,

[LAa a , LB ] = AB ab Za +b ,
[LA++ , LB ] = AB Z0 + MAB ,

1
Z+a +b , LAcc = cb LAa c ab LAcc ,
2

1
Za b , LAcc = cb LAca ab LAcc ,

2
[Z0 , LA ] = 2LA ,

Z+a b , LA = bc LAa c ,

Za +b , LA++ = bc LAca ,

Z+a b , LAcc = ac ab LA++ ,

Za +b , LAcc = a c cb LA ,

Z+a +b , Z+c +d = cb Z+a +d ad Z+c +b ,

Za b , Zc d = cb Za d ad Zc b ,

Z0 , Z+a +b = 2Z+a +b ,

Z0 , Za +b = 2Za +b .
(3.5)
Now, we want to realize this algebra on a coset of the group SO(6, n). The standard coset
SO(6, n)/SO(6) SO(n) is obtained by putting all the generators M and Z in the coset denominator and leaving all the Ls in the coset with associated 6n coordinates :
SO(6, n) ++ a a
A , A , A .
(M, Z)
(3.6)
359
++
We wish to have an alternative S-analytic coset involving only the coordinates A
associated
with the generators LA++ . To this end we have to move the generators LAa a , LA to the coset
denominator. In doing this we encounter a problem: The SO(6) generator Z+a b converts LAa a
into the coset generator LA++ . In order to avoid this, we proceed to the harmonization of
which we treat as independent
the coset. This means to introduce an additional group SU(4)
of the SO(6) from the coset denominator. Let us denote its generators by T+a +b , Ta b , T0 ,
acts as an
T+a b and Ta +b , in complete analogy with SO(6). We assume that this extra SU(4)
external automorphism of (3.5), i.e. [T , Z] = Z, [T , L] = L. Then it is clear that the combination
Z+a b T+a b commutes with the generators of (3.5), in particular, with LAa a . So, to avoid
the above problem, we replace Z+a b in the coset denominator by this combination. The group
is itself realized on the harmonic coset SU(4)/S(

SU(4)
U
(2) U
(2)), which means that we have

to add the generators of the automorphism subgroup S(U
(2) U
(2)) to the coset denominator.
The result is a particular S-analytic realization of the coset

SO(6, n)
SU(4)
(M, La a , L , Z+ + , Z , Z0 , Z + , Z+
++ +a a
, wi , wi
A
T+ , T + + , T , T 0 )
(3.7)
++
associated with the SO(6, n) generators LA++ and by
parametrized by the coordinates A
+a
a
harmonics wi , wi (the latter differ from the usual SU(4) harmonics u (2.1), as explained
below).
++
This coset is analytic in the sense that we consider functions F (A
, w) on it which are
annihilated by the generators LAa a and LA . Then the algebra (3.5) implies
LAa a F = LA F = 0 MAB F = Z+a +b F = Za b F = Za +b F = 0,
(3.8)
i.e., F cannot carry SO(n) SU(2) SU(2) indices, but can have U (1) charges under both Z0
and T0 . In addition, we impose the coset defining constraint

Z+a b T+a b F = 0.
(3.9)
It leads to a particular mixing of the coordinates associated with the SO(6) generators Z and with
generators T . For this reason (3.7) is a semi-direct product (denoted by
the SU(4)
in (3.7)) of

the two cosets SO(6, n)/SO(6) SO(n) and SU(4)/S(U (2) U (2)).
The actual construction of the coset goes through the following steps. We first introduce a
harmonic variables u, harmonics I i
double harmonic space involving, in addition to the SU(4)
on SU(4) SO(6) satisfying the defining conditions (cf. (2.1))
I i i J = IJ ,
i I I j = i ,
I J KL I i J j K k L l = ij kl .
(3.10)
They undergo SU(4) transformations of two types: local (in the sense of SU(4) SO(6) from
the coset denominator) with parameter and rigid with parameter :
I i = I J J i + I j j i .
(3.11)
Our task now will be to make a change of variables from , u to z, w which are inert under
the rigid SU(4) and have simple transformation properties under the local SU(4). This will allow
us to impose the coset constraint (3.9) in a covariant way. We start by projecting the harmonics
360
with u, u:
+a +b = u +a I I i ui +b ,
+a b = u +a I I i ui b ,
a +b = u a I I i ui +b ,
a b = u a I I i ui b
(3.12)
and similarly for the conjugate matrix .

Next we make the following non-linear change of
variables (to simplify the notation, we suppress the SU(2) SU(2) indices; the position of the
U (1) charges allows to unambiguously restore them):

1
1

z + = + + +
= + ,
z = ,

1
1
z+ = +
(3.13)
= + + + ,
z+ + = + + .
These new variables satisfy an algebraic constraint following from the fact that SU(4), i.e.
det = 1. It can be used to eliminate, e.g. det z while the remaining z0 det z+ + can be
treated as the coordinate of the U (1) factor in S(U (2) U (2)) SU(4).
It is then not hard to check that the new variables z transform in the following way under the
local SU(4):
z + = + ,
z = z ,
z+ = + + z+ + z+ + ,
z+ + = + + z+ + ,
(3.14)
where = w I I J wJ and we have introduced the new harmonics
wi +a = ui +a + ui b zb +a ,
wi a = ui a ,
w a i = u a i za +b u +b i
w +a i = u +a i ,
(3.15)
with transformation laws
wi +a = wi b b +a ,
w +a i = 0,
wi a = 0,
w a i = a +b w +b i .
(3.16)
We point out that these new harmonics are not unitary anymore (i.e., w is not the conjugate of w),
but they still satisfy the same algebraic relations as the unitary harmonics u (2.1).
What we have achieved is that the new variables do not mix under the local SU(4) transformations with parameters . This allows us to eliminate all of the z variables (with the exception
of z0 ) in a covariant way, which corresponds to imposing the Z coset conditions from (3.8) and
the Z T condition (3.9).
3.3. Covariant constraints on the function F
Now we are able to see how the naive constraints (2.31), (2.34), (2.35) are modified due to
the curvature of the coset space (3.7) on which the reduced function F (2.29) lives. The origin of
these constraints can be traced back to the S-analyticity conditions satisfied by the gauge-fixed
function F (2.26). On the curved manifold they become covariant constraints (cf. (3.8)):
DAa a F = DA F = 0.
(3.17)
Here DAM are covariant derivatives generalizing the flat derivatives /. They satisfy the same
SO(6, n) algebra as the generators LAM .
361
Let us start with the constraint (2.35). The second-order derivative in it can be rewritten as
follows:

DAM DBM F = DA++ DB + DA DB++ 2DAa a ab a b DBbb F

= [DA , DB++ ]F = AB Z0 F,
(3.18)
where we have used the S-analyticity constraints (3.17) and the algebra (3.5). The function
F 4(g1) has two independent U (1) charges, one with respect to the generator T0 , T0 F 4(g1) =
4(g 1)F 4(g1) and the other for Z0 . For a reason which will become clear in the next section, the Z0 charge takes a different value, Z0 F = 4(g + 1)F . Thus, we have
DAM DBM F = 4(g + 1)AB F,
(3.19)
or, in SU(4) notation,

pqrs DApq DBrs F = 32(g + 1)AB F.
(3.20)
Further, the second-order derivative in (2.34) is replaced by DA[M DBN ] F . Due to the constraints (3.17), this operator has only two non-vanishing projections obtained by taking M = ++
and N = or N = a a.
The first choice yields back the constraint (3.18), while the second gives
rise to the commutator
[DAa a , DB++ ]F = AB a b Z+a b F
(3.21)
where Z is the covariant derivative replacing the generator Z. The effect of this is just a particular
SO(6) transformation of the coset coordinates, hence it is not really a constraint on the function.
Finally, in Eq. (2.31) (or (2.32)) the flat partial derivative with respect to scalars is replaced
by a covariant derivative
pqrs
q DArs F = 0.
u +a
(3.22)
We would like to point out that in the string theory analysis given in the following sections, the
differential equations are obtained on functions F which is the relevant part of F that survives
the harmonic space integrals. Indeed string theory amplitudes directly see F . The crucial step
used in Eq. (3.18) was that F does not depend on 5 combinations of moduli as is expressed in
the S-analyticity constraint (3.17). It is easy to see that F does not satisfy this S-analyticity constraint since it is obtained by making a certain SU(4) projection on F . Therefore the individual
steps in this derivation cannot be applied to F . However, the second-order differential operators
considered here are not sensitive to any particular SU(4) projection of F and therefore the final
equations are still true on F .
4. N = 4 conformal supersymmetry and supergravity
Here we show that the realization of G-analytic superfields of the type (2.9) as functions on a
particular coset of the N = 4 conformal superalgebra PSU(2, 2/4) is very similar to the bosonic
coset construction of the preceding section. This algebra involves the generators of Lorentz transformations (M ), translations (P ), conformal boosts (K ), dilatation (D), R symmetry SU(4)
j
(Zi ), Poincar supersymmetry (Qi and Q i ) and special conformal supersymmetry (Si and Si )
with anticommutation relations for the odd generators (schematically)
362
= P,
{Q, Q}
= K,
{S, S}
{Q, S} = M + D + Z.
(4.1)
The standard superspace corresponds to the coset

PSU(2, 2/4)
x , i , i
Z)
(M, K, D, S, S,
(4.2)
involving all the 16 Grassmann variables associated with the supersymmetry generators. In order
to obtain G-analytic superfields depending on half of these Grassmann variables, we add the
+a = Q
i u+a to the coset
SU(4) harmonic projections of the Q generators Qa = u ia Qi and Q
i
denominator, thus leaving only the odd coordinates +a and a in the coset. However, exactly
+ from the
as in the bosonic case of Section 3, the SU(4) generator Z+ converts Q and Q
. In order to avoid this, we introduce

coset denominator into the coset generators Q+ and Q

the external automorphism group SU(4) with generators T . Then the combination Z+ T+
commutes with all the Qs and thus can be safely put in the coset denominator7 :

PSU(2, 2/4)
SU(4)
+
(M, K, D, S, S, Q , Q , Z+ , Z , Z0 , Z + , Z+ T+ , T+ + , T , T0 )

x, + , , w .
(4.3)
Here the harmonics w are defined in exactly the same way as in Section 3, Eq. (3.15), replacing
the SO(6) harmonics by R-symmetry SU(4) harmonics. They transform as in (3.16) with the
parameter replaced by the G-analytic superparameter

b +a x, + , , w
= w b i i j wj +a + i +a b k + i w b i i +a + i b i wi +a
(4.4)
containing the parameters of the R-symmetry SU(4), k of conformal boosts and of special
conformal supersymmetry.
The basic G-analytic conformal superfield Y ++ (x, + , , w) (2.9) (with superconformal
harmonics w instead of u) describes the vector supermultiplet. It transforms with a G-analytic
superconformal weight factor
Y ++ = Y ++ (x
,
,
, w
) Y ++ (x, , , w) = Y ++ ,

x, + , , w = + k x + w +a i i j wj +a + i w +a i i +a + i i wi +a ,
b
(4.5)
where is the parameter of dilatations.8 The other G-analytic object we are discussing here is the
++ (2.14) of the Weyl multiplet. It is superconformal covariant due to the on-shell
descendant K
constraint and transforms with weight two, according to its scaling dimension, K ++ = 2K ++ .
The generalization to N = 4 conformal supergravity is done by replacing the parameters
b +a and by arbitrary G-analytic superfields. Poincar supergravity is obtained by coupling
7 Here we follow the formulation of N = 2 conformal supersymmetry of [16,18]. A somewhat different approach is
proposed in [9].
8 It can be shown that +a = D +a .
b
b
the Weyl multiplet to a set of six compensating vector multiplets (cf. (2.9))

kl +a +b
yij++ x, + , , w = yj++
i = ij wk wl ab + terms.
363
(4.6)
Here we see the 6 6 matrix of compensating scalars ijkl . Let us consider the following projections of yij++ with the harmonics w:
1
j
i
y0 = ab w +a
w +b yij++ .
2
It is easy to check that they transform as follows:
i
+a
w a yij++ ,
ya++
a = w
j
+b
ya++
y0 + ya++
a = ab a
a ,
y0 = y0 ,
so their ratio transforms as a compensator for the local superconformal transformations:

++
y
a a
= ab a +b .
y0
(4.7)
(4.8)
(4.9)
Then, with the help of this compensator we can define new harmonics inert under the local
superconformal transformations (notice the similarity with (3.15) and (3.16)):
vi
+a
= wi
+a
wi
v+a i = w +a i ,
v = v = 0.
y ++
b ab bb
y0
vi a = wi a ,
va i = w a i + ab
yb++
a
w +a i ,
y0
(4.10)
The role of the compensators is to completely absorb the local superconformal transformations. This allows us to use the parameter a +b in (4.9) fix a gauge in which ya++
a = 0, thus
identifying the harmonics v and w. This means, in particular, that the conformal SU(4) (gen (generators T in (4.3)). By the same logic, we can
erators Z in (4.3)) is identified with SU(4)
+
use the parameter of local SO(6) transformations in (3.14) to gauge away the compensator
z + . This results in the identification of the harmonics w with u. So, at the expense of manifest
covariance, the different SU(4) groups discussed above are reduced to a unique one, and the harmonics to the original ones (2.1). This gauge fixing procedure establishes a bridge between the
S-analytic coset (3.7) and the G-analytic coset (4.3).
Finally, we are ready for the superconformal covariantization of the higher-derivative terms
(2.19), (2.20). It is achieved in three steps. Firstly, we replace the explicit harmonics u in F (Y, u)
by the new inert ones v (however, the superfields Y still depend on the conformal harmonics w).
Secondly, we introduce weightless G-analytic superfields Y/y0 . In this way the potential F (Y, v)
becomes conformal invariant. Thirdly, we use the G-analytic density y0 to compensate the weight
4(g + 1) of the Weyl factor (the measure is weightless, as can be seen from its vanishing scaling
dimension). The result is

++ ++
S1 = d 4 x du d 4 + d 4 K
(y0 )4(g+1)
K

++ ++ g 4(g1) YA++
K K
F1
,v ,
y0
364

S2 =
++
++ ++ g+1
4(g1) YA
d 4 x du d 4 + d 4 K
K
(y0 )4(g+1) F2
,v .
y0
(4.11)
The presence of the density (y0 )4(g+1) in (4.11) explains why in (3.19) we took the value
Z0 F = 4(g + 1)F of the charge Z0 , different from that of the charge T0 . This density should
be viewed as part of the covariantized function F discussed at the end of Section 3. Then, F is
a function of the G-analytic superfields YA++ and yij++ and hence is a G-analytic superconformal
+ from the coset deobject itself. This means that it is annihilated by the supercharges Q , Q
nominator in (4.3). This is compatible with the condition of superconformal primarity (that the
object is annihilated by all the special superconformal charges S) only if the dilatation and Z0
weights of the object coincide [18,22]. Finally, the local SU(4) gauge-fixing procedure (elimination of the compensators) results in the identification of the Z0 charges from (3.7) and (4.3). The
automorphism charge T0 remains independent and, indeed, takes a different value.9
5. Topological amplitudesreview
In Sections 2 and 3 it was argued from the general structure (2.29) of the harmonic expansion
(1,3)
that they fulfill differential equations of first order (2.31)
of the supergravity amplitudes Fg
and second order (3.20) in the moduli of the internal compactification manifold (i.e. K3 T 2
for type II string theory). In this section, we would like to check these relations by applying them
directly to the string amplitudes. Since, as we have already pointed out, the latter are captured
by correlation functions of the topological string, it would be logical, to consider the twisted
version of the theory. However, here we are facing the problem that some of the moduli involved
in the K3 T 2 compactification are in fact part of the RamondRamond sector of the theory,
for which we have at present no representation in terms of the N = 4 superconformal algebra,
which is used to formulate the topological correlators. Besides that, the direct study of (2.31) and
(3.20) in the untwisted version of the type II string is quite cumbersome, since we would have to
deal with (in principle) an arbitrary high number of loops.
Fortunately, as was found in [5], the dual amplitudes of the couplings (2.20) in the heterotic
theory compactified on T 6 begin receiving corrections already at the 1-loop level, which are
relatively simple to compute. Therefore, for the purpose of checking (2.31) and (3.20), we will
focus on this amplitude which we review below.
After performing explicitly the superspace integrals of the 1/2-BPS F -type term (2.20) we
encounter among many different contributions a coupling of two self-dual Riemann tensors, two
graviscalars and 2g 2 graviphoton field strengths at (g + 1)-loop order

S2 =
++ 2g2
2
d 4 x Fg(3) R(+)
()2 T(+)
,
(5.1)
(3)
where we remind that Fg corresponds in the supergravity context to the reduced part of
4(g1)
. The corresponding heterotic string 1-loop torus amplitude can be formulated as the
F2
9 This situation is different from N = 2 superconformal symmetry where the relevant G-analytic superfields, e.g. the
hypermultiplet, have equal Z0 and T0 charges [18]. This can be explained by the different properties of the G-analytic
superspace measuresthe N = 2 measure has a conformal weight while the N = 4 one has not.
365
following two-dimensional integral over the fundamental domain F of the world-sheet torus

2
d 2g1
1 ij L 2g2 1 P 2 1 P 2
(HET)
Fg
(5.2)
=
Gg+1
q 2 L q 2 R .
u ++ Pij
2
24 2
L R
(6,22)
(P ,P )
In this expression = 1 + i2 is the Teichmller parameter of the torus, while q = e2i . Moreover, ( ) is the Dedekind eta-function given by
1
( ) = q 24

1 qn ,
(5.3)
n=1
and Gg+1 is defined via the following expansion of a generating functional for spacetime correlation functions

2h
h
h

1
2
1 2
2
2
1
d xi X X (xi )
d yj X X (yj )
G(, , )
(h!)2 2
=
h=0
j =1
i=1
2h Gh (, ).
(5.4)
h=1
In [23], this generating functional was calculated with the result

2i 3 2
2
G(, , ) =
exp
,
2
(,
)
where is the usual odd theta-function defined by
1
1 2
1
1
q 2 (n 2 ) e2i(z 2 )(n 2 ) .
(z, ) =
(5.5)
(5.6)
nZ
The most important property of Gg for our purposes is the fact that upon differentiation with
respect to it becomes Gg1 :
i
Gg = 2 Gg1 .
22
(5.7)
ij
SU(4)
S(U (2)U (2)) , which appear in the
Finally, PijL and PAR are the left- and
In (5.2), u ++ are precisely the harmonics of the coset

(HET)
in (2.29).
reduced harmonic expansion of the Fg
(6,22)
right-moving momenta of a
Narain-lattice describing the compactification of the heterotic
string on the T 6 torus. They encode the full dependence of the amplitude on the corresponding
6 22 = 132 moduli, which form the manifold
SO(6, 22)
(5.8)
,
SO(6) SO(22)
as explained in Section 3.1. The exact parameterization of the lattice momenta, however, will be
of no importance to our calculations and would involve the explicit construction of the worldsheet sigma model action, starting from the four-dimensional action of N = 4 supergravity
coupled to 22 vector multiplets. The left-moving momenta PijL are formulated in a complex
SU(4) basis and their square is given by
L 2 1 ij kl L L
P
(5.9)
= Pij Pkl ,
8
M=
366
which is manifestly real and SU(4) invariant. Moreover, in order to streamline our notation, we
will also introduce the following projection of the momenta
1 ij
L
u ++ PijL .
P++
2
(5.10)
6. First-order harmonicity relation

With the above setting, we are now in a position to discuss the harmonicity equation (2.31)
(HET)
satisfy the following relation
(or (3.22)). In [5], it was shown that Fg
ij kl
u i+1
u +2
Dkl,A Fg(HET) = 0,
(6.1)
up to an anomaly, which was calculated explicitly. The action of the differential Dij,A with
respect to the moduli ij,A can be analyzed in two different ways:
From the world-sheet point of view, it amounts to inserting the scalar vertex operator
mod.
Vij,A
=
1
Xij JA (z)eipX
2
(6.2)
into the correlation function, where Xij are the internal bosonic coordinates in an SU(4)
basis, satisfying the pseudo-reality condition
1
X ij = ij kl Xkl ,
2
(6.3)
and JA are the right-moving (Abelian) currents.

This approach is rather cumbersome, since the correlator corresponding to e.g. (6.1) contains
(2g + 3) vertices, for which all possible contractions need to be considered. We will therefore
rather resort to the following approach.
In terms of the (6,22) lattice momenta, the differentials act as infinitesimal Lorentz boosts10
Dij,A PklL = ij kl PAR ,
Dij,A PBR =
AB L
P .
2 ij
(6.4)
These rules were proved in [5] by an explicit world-sheet computation at the linearized level.
It can be easily checked that they in fact reproduce the algebra (3.5), up to normalization
factors. Moreover, they annihilate the SO(6, 22)-square of the lattice vectors
2 2
= 0.
Dij,A P L P R
(6.5)
(HET)
As we have seen in Section 2, the general harmonic expansion of Fg

suggests that (6.1) is in
fact merely a consequence of the stronger relation (2.31). The goal of this section is to explicitly
test the validity of (2.31) and to examine whether its right-hand side is modified by an anomaly
as it was the case for (6.1).
10 Note a factor of 2 misprint in Eq. (10.7) of [5] which had no effect in the subsequent analysis.
367
The computation is done in a straightforward way using the differentiation rules (6.4)
E1 ab ij kl
= ab
= 2ab
j
u +b
ij kl
u +b

j
u +b
1
PAR q 2 (P
Dkl,A Fg(HET)

Dkl,A

d 2 2g1
L 2g2 12 (P L )2 12 (P R )2
P++
G
q
q
g+1
2
24
L R
(P ,P )
d 2
24
L )2
2g1
Gg+1
L
P++
2g3
L
ij
(2g 2)u ++ 2 ij kl PklL P++
(P L ,P R )
q 2 (P
R )2
(6.6)
which was essentially already found in [5]. Using the simple identity
ab
u +b
L
P++
= u i+a PijL ,
(6.7)
we can easily calculate the harmonic partial derivative

2
d 2g1
E1 = 2
Gg+1
24 2

L
L
L ij kl L
3(2g 2)u i+a P++
(2g 2)2 u m
Pkl P++
+a Pmj
(P L ,P R )
L 2g4 R 1 (P L )2 1 (P R )2
L ij
++ P++
PA q 2
q 2
.
+ (2g 3)(2g 2)u m
+a Pmj u
Using furthermore the trivial relation
2
inkl PjLn PklL = 2ji P L ,
(6.8)
we can further simplify the expression

2

2 L 2g3
d
2g1
2g
E1 = 2(2g 2)u i+a
2g2
P++
Gg+1
22 P L
24
L R
(P ,P )
PAR q
1
L 2
2 (P )
1
R 2
2 (P )
q

= 4i(2g 2)u i+a

d 2
L 2g3 R 12 (P L )2 12 (P R )2
.
Gg+1
PA q
q
2g P++
24
L R
(P ,P )
At this point, we perform a partial integration in and use modular invariance together with the
exponential suppression in the infra-red region 2 , due to the presence of PL for g > 1,
to conclude that there are no boundary terms we have to worry about.11 The only contribution
therefore comes when the -derivative acts on Gg+1 . Using the identity (5.7) we get
2

d 2g2
i
L 2g3 R 12 (P L )2 12 (P R )2
E1 = 2(2g 2) u +a
(6.9)
P++
G
PA q
q
.
g
2
24
L R
(P ,P )
11 Note that the equation is trivially fulfilled in the case g = 1, since F (HET) is independent of the harmonic variables.
1
368
This result has to be contrasted with the expression

(HET)
D++,A Fg1 ,
(6.10)
where we have used the same projection as in (5.10)

1 ij
D++,A u ++ Dij,A .
2
The calculation follows much along the same lines as before and yields the result
2

d 2g2
(HET)
L 2g3 R 12 (P L )2 12 (P R )2
P++
2
Gg
PA q
q
.
D++,A Fg1 = 2
24
L R
(6.11)
(6.12)
(P ,P )
Comparing this result with (6.9), one concludes

ab ij kl
u +b
(HET)
Dkl,A Fg(HET) = (2g 2)u i+a D++,A Fg1 .
(6.13)
(HET)
Since the Fg1 , which appears on the right-hand side is of lower order in g than the initial one
we considered on the left-hand side, this term can be interpreted as an anomaly to the harmonicity
relation. This is justified by comparison to the holomorphic anomaly equation [14], where (for
the type II theory) the lower genus12 terms have their origin from boundary contributions in the
moduli space of genus g world-sheets.
As a trivial consistency check of this result, we can try to recover the weaker harmonicity
relation presented in [5], by applying a second partial differentiation with respect to u i+a to
(6.13) using the fact that it commutes with j Dkl,A
u +b
ab ij kl
u i+a u j+b
(HET)
Dkl,A Fg(HET) = 2(2g 2)(2g + 1)D++,A Fg1 ,
(6.14)
which is precisely the result found in [5].

7. Second-order constraint
In the same way as Eq. (2.31), we can now check relation (2.33) (or rather its counterparts
(3.20) and (3.21) taking into account the curvature of the moduli space) by directly applying
(HET)
. We use again the
the corresponding differential operator to the topological amplitude Fg
differentiation rules (6.4) to obtain
E2 ij km Dij,A Dkl,B Fg(HET)
2
d 2g1
= ij km Dij,A
Gg+1
24 2

1
L L 2g3 R 1 (P L )2 1 (P R )2
pq
L
PB q 2
q 2
.
(2g 2)klpq u ++ 22 Pkl P++ P++
2
L R
(P ,P )
12 In [2] scattering amplitudes of two (self-dual) Riemann tensors and (2g 2) graviphoton field strengths in type II
theory compactified on CalabiYau threefolds were considered. In these amplitudes, the number g corresponds to the
genus of the world-sheet Riemann surface.
369
Taking now the second scalar derivative, one has

2
d 2g1
Gg+1
E2 = ij km
24 2

L
pq L 2g3
22 ij kl P++
2 PklL ijpq u ++ P++
(P L ,P R )

L 1
L
1
pq
L
rs
L
(2g 3)ij rs u ++ 22 Pij P++
+ (2g 2)klpq u ++ 22 Pkl P++
2
2

L 2g4 R R 1 (P L )2 1 (P R )2
PA PB q 2
q 2
P++

AB ij km

2

L
d 2 2g1
pq
L 1
L
(2g
2)
P
G
P
u
2
P
g+1
klpq ++
2 kl
++
ij
2
24 2
L R
L 2g3 1 (P L )2 1 (P R )2
q2
q 2
.
P++
(P ,P )
Regrouping the terms furthermore

2
d 2g1
Gg+1
E2 =
24 2

L 2g2
L 2g3
122 lm P++
4(2g 2)2 PklL u km

++ P++
(P L ,P R )
L 2g3
L 2g2
pq
2 (2g 2)klpq ij km u ++ PijL P++
+ 4 2 22 ij km PklL PijL P++
1
L 2
R 2
PAR PBR q 2 (P ) q 2 (P )
L
AB
pq
(2g 2) ij km klpq PijL u ++ 22 ij km PijL PklL P++
+
2
L 2g3 1 (P L )2 1 (P R )2
q2
q 2
,
P++
and using the relations
2
L
= 2ji P L ,
klmi PklL Pmj
L i
1 klmi
pq
L pi

j ,
mjpq PklL u ++ + 2Ppj
u ++ = 2 P++
2
we obtain
2
d 2g
Gg+1
E2 = 4
24 2

2 L 2g2 R R 1 (P L )2 1 (P R )2
(2g + 1) 22 P L lm P++
PA PB q 2
q 2
(P L ,P R )
d 2 2g1
Gg+1
24 2

2 l L 2g2 1 (P L )2 1 (P R )2
q2
q 2
2g 22 P L m
P++
+ AB
(P L ,P R )
(7.1)
(7.2)
370
d 2 2g1
Gg+1
24 2

l
L 2g3 1 (P L )2 1 (P R )2
L
L pl
2 P++
m + (2g 2)Ppm
u++ P++
q2
q 2
.
AB
(7.3)
(P L ,P R )
At this point one can check that the first two lines as well as the last line, separately, are indeed
modular invariant (for the last line, this follows mainly from the presence of the harmonics).
Moreover, the first two lines can be written as differentiations with respect to the torus Teichmller parameter:
2
2g+1

d
L 2g2 R R 12 (P L )2 12 (P R )2
E2 = 8ilm
2
P++
Gg+1
PA PB q
q
24
L R

+ 2iAB lm

AB
(P ,P )
d 2
24
Gg+1
(P L ,P R )
2g L 2g2 1 (P L )2 1 (P R )2
P++
q2
q 2
2

l

d 2 2g1
L
L pl
2
P
G
+
(2g
2)P
u
g+1
++
++
m
pm
2
24
L R
(P ,P )
L 2g3 1 (P L )2 1 (P R )2
P++
q2
q 2
.
(7.4)
Since these terms are modular invariant, one is allowed to perform a partial integration, with
only hitting the factor Gg+1

E2 = 4 2 lm

d 2 2g1
L 2g2 R R 12 (P L )2 12 (P R )2
P++
G
PA PB q
q
g
2
24
L R

AB lm

AB
(P ,P )

L 2g2 12 (P L )2 12 (P R )2
P++
q
q
d 2
2g
Gg
24 2
(P L ,P R )

l

L
L pl
2 P++
m + (2g 2)Ppm
u++
d 2
2g1
Gg+1
24 2
(P L ,P R )
L 2g3 1 (P L )2 1 (P R )2
P++
q2
q 2
.
This expression can be contrasted with
(HET)
D++,A D++,B Fg1 ,
(7.5)
which can be computed using exactly the same rules as before

(HET)
D++,A D++,B Fg1

2

d 2g2
L 2g3 R 12 (P L )2 12 (P R )2
P++
= 2D++,A
2
Gg
PB q
q
24
L R

= 4 2
(P ,P )
d 2
2g1
Gg
24 2

(P L ,P R )
L
P++
2g2
PAR PBR q 2 (P
L )2
q 2 (P
R )2

AB
371

d 2 2g2
L 2g2 12 (P L )2 12 (P R )2
P++
2
Gg
q
q
.
24
L R
(P ,P )
From this, we conclude for the second-order constraint

ij km Dij,A Dkl,B Fg(HET)
(HET)
= lm D++,A D++,B Fg1
2lm AB Fg(HET)
2
d 2g1
(2g 2)AB
Gg+1
24 2
L 2g3 1 (P L )2 1 (P R )2
L pl
Ppm
u ++ P++
q2
q 2
.
(7.6)
(P L ,P R )
Notice that the last two terms are not of the form of an anomaly but are generic hard contributions to the equation. As one can easily see, they correspond however to an SU(4)( SO(6))
(HET)
, which is exactly what one expects according
rotation acting on the harmonics inside Fg
to (3.21).
On the other hand, (7.6) is vastly simplified when we contract its free SU(4) indices:
(HET)
ij kl Dij,A Dkl,B Fg(HET) = 4D++,A D++,B Fg1 4(g + 1)AB Fg(HET) .
(7.7)
Comparing this result to (3.20), we conclude that besides the anomalous term proportional to
(HET)
Fg1 the two relations indeed agree, up to an irrelevant normalization.
Finally, let us mention in passing that a second-order differentiation, which is antisymmetrized
in the SO(22) indices, is exactly vanishing
ij
[A
kl
B]
Fg(HET) = 0.
(7.8)
This can be seen most easily by representing the above expression as a correlator with two additional scalar vertices inserted and realizing that its right-moving part is given by

J[A (z)JB] (w)
(7.9)
,
which follows from the form of the scalar vertex operator (6.2). Since the right-moving currents
are Abelian, it follows that expression (7.9) is identically zero. Note in particular that in this case
there is not even an anomaly, and (7.8) remains in fact exact at the quantum level.
8. Harmonicity in six dimensions
8.1. The origin of the harmonicity constraint
In this subsection we summarize a few key points about six-dimensional harmonic superspace
and derive the corresponding harmonicity constraint. The discussion closely follows that of the
N = 4 case in four dimensions, therefore it is very brief.
We consider N = (1, 1) supersymmetry in six dimensions whose automorphism group is
SU(2)L SU(2)R . Let us introduce harmonic variables vaI for SU(2)L and vaI for SU(2)R , to
gether with their conjugates vIa = (vaI ) and vIa = (vaI ) . Here a, a are SU(2) doublet indices
372
while I, I = 1, 2 are projections onto the subgroup U (1)L U (1)R . They satisfy the completeness conditions
vaI vJa = JI ,
vIa vbI = ba
(8.1)
(and similarly for vaI ). Raising and lowering the indices with ab , I J , etc., we can rewrite the
non-trivial part of (8.1) as the unit determinant condition
ab va1 vb2 = 1.
(8.2)
In fact, the harmonics can be viewed as matrices of the corresponding SU(2) groups.
The harmonic functions are supposed to have harmonic expansions homogeneous under the
action of the subgroup U (1)L U (1)R . For example, a function of unit U (1)L U (1)R charges
has the expansion
1 1 1 2
va va vb vc + ,
11 (v) = a a va1 va1 + (abc)a va1 vb1 vc2 va1 + a(a bc)
(8.3)

so that in each term the number of v 1 exceeds by one the number of v 2 (the same for v 1,2 ).
Notice that due to the constraint (8.2) each component is an irrep of SU(2)L SU(2)R (i.e.,
only symmetrized indices appear). Effectively, such homogeneous functions live on the coset
SL2 SR2 = (SU(2)L /U (1)L ) (SU(2)R /U (1)R ).
The introduction of harmonic variables allows us to define G-analytic superfields which de
pend only on half of the Grassmann variables,13 e.g. on 1 = va1 a , 2 = v2a a = 1 . One such
short superfield describes the (on-shell) vector multiplet

Y 11 1 , 2 , v = a a va1 va1 + 1 a v2a + 2 a va1 + 2 1 F + .

(8.4)
Notice the conservation of the overall charges 1, 1 carried by the projected Grassmann variables or by the explicit harmonics projecting the component fields. This superfield is real in

11 = Y 11 , where is a combination of complex conjugation with a reflection on
the sense Y
S 2 S 2 preserving G-analyticity. In particular, this implies the reality of the first component,
( a a ) = ab a b bb .
Another short superfield of the same type describes the (on-shell) Weyl multiplet [6]

11 1
W , 2 , v = T a a va1 va1 + 1 2 R + ,
(8.5)
where (W 11 ) is in the adjoint of SU (4) ((W 11 ) = 0), (T a a ) = (T a a ) ( ) are the

graviphoton field strengths and R = R ( ) ( ) is the curvature.
In [6] the following term of the six-dimensional effective action was considered:

g
d 6 x dv d 4 1 d 4 2 1 2 3 4 1 2 3 4 W 11 1 W 11 2 W 11 3 W 11 4
11
F(1)4g4 (1)
4g4 Y , v .
(8.6)
In fact, what appears in (8.6) is the determinant of the 4 4 traceless matrix (W 11 ) . This is a
Lorentz invariant which breaks up into two independent invariants, [Tr(W 11 )2 ]2 and Tr(W 11 )4 .
We could use anyone of them to construct an effective action term similar to (8.6). However,
13 We use SU (4) SO(1, 5) chiral spinor notation with left-handed and right-handed
spinors.
373
upon decompactification of the four-dimensional couplings (2.19) and (2.20), one can show that
only the first of the two invariants contributes. We will eventually study this case in the next
subsection. The corresponding effective action term is

g+1
11
F(1)2g2 (1)
d 6 x dv d 4 1 d 4 2 W 11 W 11
(8.7)
2g2 Y , v .
The function F(1)m (1)

m (m = 2g 2) has to carry a negative (i.e. indices 1, 1 downstairs)
charges of each kind, in order to compensate that of the K factor (+4g) and of the Grassmann
measure (4). We consider functions of the type
F(1)m (1)
m =
11 n
(1)m+n (1)
,
m+n Y
(8.8)
n=0
where
a
a1
p a 1
p
(1)p (1)
p = (a1 ap )(a 1 a p ) v1 v1 v1 v1
a
(8.9)
introduces a set of constant SU(2)L SU(2)R multispinors, thus explicitly breaking the symmetry.
Let us examine the coupling (8.7) in some detail. First of all, from the term (W 11 W 11 )g+1 we
only consider contributions of the type
m
1 4
(2 )4 R 4 T 11 .
(8.10)
The Grassmann factor saturates the integrals. The harmonic dependence comes from the factor
11 m
= T (a1 (a 1 T am )a m ) va11 va1m va1 1 va1 m .

T
(8.11)
Notice that the projection with commuting harmonic variables forces symmetrization of the indices of the T s. Thus, this term contributes an irrep of each SU(2) of weight m. Since the
harmonic integral in (8.7) only sees the singlet part of the integrand, we have to find a matching
irrep in the F sector, so that together they can form a singlet. Let us look at a term from (8.8)
(where we replace the superfield Y by its first component ),
11 n
(1)m+n (1)
m+n
a
= (a1 am+n )(a 1 a m+n ) v1a1 v1 m+n v1a 1 v1 m+n

a
(b1 (b1 bn )bn ) vb11 vb1n vb1 vb1 .

1
(8.12)
The first factor involves only harmonics with upper SU(2) indices, the second only with lower
indices. Such products of harmonics are reducible. Using the defining conditions (8.1), we can
1 v b} , where {} denotes
decompose a reducible product of va1 with v1b as follows: va1 v1b = 1/2ab +v{a
1
the traceless part. Contracting the indices of all v 1 s and v 1 s with those of a subset of the v1 s
and v1 s, we can eliminate the v 1 s and v 1 s from (8.12). The result is the irrep of weight m of
j}
each SU(2) needed to match that in (8.11); any traceless combination v{i1 v1 will contribute to
an irrep of higher isospin without a match in (8.11), thus irrelevant for the harmonic integral. So,
we can reduce (8.8) to its relevant part
F=
(8.13)
(a1 am+n )(a 1 a m+n ) v1a1 v1am v1a 1 v1a m am+1 a m+1 am+n a m+n .
n
374
It is important to realize that the tensor in (8.13) has all its indices symmetrized. This is the
origin of the harmonicity constraint
F = 0.
(8.14)
a
v1 bb
It involves a partial derivative with respect to v1 . Strictly speaking, such an operation is illegal in
the harmonic formalism, since the variables v1 and v 1 are not independent, as can be seen from
(8.1). However, in (8.13) there are only v1 s left, so we can formally take such a derivative. In
fact, if needed, (8.14) can also be expressed using covariant harmonic derivatives as in (2.32).
In principle, we could go on and discuss the coset space SO(4, n)/SO(4) SO(n) parametrized by the scalars of the vector multiplets (8.4) in a manner similar to that of Section 3. The
conclusion would be a second-order constraint analogous to (3.18). However, in six dimensions
we do not have the setup of conformal supergravity of Section 4 which allowed us to fix the
value of the charge Z0 in (3.20). Therefore, we can make a prediction for the structure of this
constraint, but we cannot explain the precise value of the coefficient obtained from the string
calculation, see (8.31).
ab
8.2. Decompactification of four-dimensional amplitudes

8.2.1. Decompactification limit
In order to round up the six-dimensional discussion, let us now check the field theory predictions by direct string calculations for the decompactification of the topological amplitude (5.2)
from four to six dimensions, which corresponds to the coupling (8.7). Essentially, it was already
shown in [5] that upon decomposing T 6 into T 4 T 2 and the subsequent reduction of the (6,22)
lattice into
(6,22) (4,20) (2,2) ,
(8.15)
the weaker version of the first order harmonicity relation (6.14) is reduced to a relation for type II
string theory compactified on K3, proved in [10]. Below, we will check the stronger relation
(8.14) and compute its corresponding quantum anomaly.
L and its complex conjugate P L
In order to perform the reduction (8.15) we choose as in [5] P13
24
L
L and their complex conjugates
(6,22)
(2,2)
of
to be entirely in
and the remaining four P12 , P14
L , P L to form the (4,20) . In this way, the group SU(4) is reduced to its subgroup SU(2)
P34
L
23
SU(2)R where SU(2)L and SU(2)R are acting on the indices (1, 3) and (2, 4), respectively. In the
decompactification limit, P13 and P24 decouple and are dropped from the correlation function.
In this way, (4,20) lattice vectors are denoted by
(4,20),L
Pa b
(4,20),R
PA
with a, b = 1, 2
(8.16)
and the index of the right-moving lattice momenta takes now the values A = 1, . . . , 20. Moreover,
the square of the left-moving momenta will be denoted by
(4,20),L 2 1 (4,20),L (4,20),L a a b b
P
(8.17)
= Pa b
Pa b
1 2 1 2.
2 2
2 1 1
In order to make contact with the six-dimensional harmonic coordinates introduced in Section 8.1
we can assemble part of the SU(4) harmonics u +1 and u +2 into the harmonics of SU(2)L
SU(2)R with the identification
i

u +1 , u i1 v1a , v2a ,
(8.18)

u i+2 , u i2 v1a , v2a ,
375
(8.19)
and we recall that the SU(2)L SU(2)R harmonics satisfy the completeness condition (8.1) and
the unit determinant condition (8.2). Finally, the 1-loop heterotic amplitude (5.2) was shown
in [5] to take the following form after the decompactification of the T 2
2
a L b 2g2 1 (P L )2 1 (P R )2
d 2g2
v1 Pa b v1
Fgdec
2
Gg+1 (, )
q2
q 2
,
(8.20)
24
L R
(4,20)
(P ,P )
where from now on, we will drop the (4, 20) superscript of the lattice momenta and for further
convenience, we define the following shorthand notation

P1L1 v1a PaLb v1b ,
(8.21)
similar to the four-dimensional definition (5.10).
8.2.2. Harmonicity relation
We now study the six-dimensional harmonicity relation (8.14)
a b
D Fgdec ,
v1a a b,A
(8.22)
(4,20) lattice. We
where the covariant derivative Da b,A
is with respect to the moduli forming the
can again apply simple rules for the differentials acting on the lattice momenta, similar to (6.4):
AB L
P .
(8.23)
2 a a
The computation can then be performed in the same straightforward manner as in the fourdimensional case
2
2g2 1 L 2 1 R 2
d 2g2

P1L1
2
Gg+1
q 2 (P ) q 2 (P )
E1dec a b a Da b,A
24
v1
(P L ,P R )
2

d 2g2

(2g 2)v1c v1c ac b c + i( )P1L1 PaLb
= a b a
2
Gg+1
24
v
L R
L
R
Da a,A
Pbb = ab a b PA ,
R
Da a,A
PB =
(P ,P )
2g3 R 1 (P L )2 1 (P R )2
P1L1
PA q 2
q 2
.
The derivative with respect to the harmonic variable yields

2

2g3
d 2g2
dec
a b
(2g 2)v1c ac b a 22 v1c PaLb PcLa P1L1
G
E1 =
g+1
2
24
L R
(P ,P )

2g4
+ (2g 3) (2g 2)v1c v1c ac b c 22 PaLb P1L1 v1c PcLa P1L1
1
L 2
R 2
PAR q 2 (P ) q 2 (P )
2
d 2g2
= (2g 2)
Gg+1
24 2
2g3

1
L 2 1
R 2
P1L1
(2g 1)ac v1c + 22 v1c PcLa a b PaLb PAR q 2 (P ) q 2 (P ) .
(P L ,P R )
376
We can now make use of the identity

2
PbLa a b PaLb = P L ab ,
(8.24)
which simplifies the expression to

2
d 2g2
dec
Gg+1
E1 = (2g 2)
24 2
2g3
2
1
L 2 1
R 2
(2g 1) 22 P L PAR q 2 (P ) q 2 (P ) .
ac v1c P1L1

(P L ,P R )
The special form of this term allows for the following rewriting
2
d
dec
E1 = 2i(2g 2)
Gg+1
24
2g1 L 2g3 R 1 (P L )2 1 (P R )2
P11
,
ac v1c
PA q 2
q 2
2
L R
(P ,P )
while a partial integration in finally yields

2
d 2g3
E1dec = (2g 2)
Gg
24 2
2g3 R 1 (P L )2 1 (P R )2
ac v1c P1L1

PA q 2
q 2
,
(8.25)
(P L ,P R )
where we have once more made use of (5.7). We now confront this result with the following
expression
dec
dec
v1a v1a Da a,A
Fg1 ,
Fg1 D11,A
(8.26)
which can be evaluated exactly in the same way as (8.22)

2
2g4 1 L 2 1 R 2
d 2g4
P1L1
2
Gg (, )
q 2 (P ) q 2 (P )
D11,A
24
L R

= 2
(P ,P )
d 2
2g3
Gg
24 2
P1L1
2g3
PAR q 2 (P
L )2
q 2 (P
R )2
(P L ,P R )
Comparing this expression to (8.25), we conclude
a b
1
dec
D Fgdec = (2g 2)ab v1b D11,A
Fg1 .
2
v1a a b,A
(8.27)
Finally replacing the differentiation with respect to v1a by one with respect to v1a , we can derive
a similar equation
ab
1
b
dec
dec
Dba,A
Fg1 .
Fg = (2g 2)a b v1 D11,A
v1a
2
(8.28)
For both equations (8.27) and (8.28), the same considerations as in the four-dimensional case
imply that the right-hand side can be interpreted as an anomaly.
377
Notice that the left-hand side of (8.27) and (8.28) is exactly the harmonicity condition first
derived in [6]. There, however, corrections to the equation by boundary terms of the Riemann
surface as well as by certain contact terms in operator product expansions were neglected. In the
later work [10], the missing of these extra contributions was pointed out and it was suggested
that an additional contraction with harmonic coordinates would project out all extra terms. This
was demonstrated by a careful analysis in the topological twisted theory. Indeed, if we project
the free indices of (8.27) and (8.28) with v1 and v1 respectively, we find
v1a a b
1
dec
a
b
dec
Da b,A
Fg1 = 0,
Fg = (2g 2)v1 ab v1 D11,A
a
2
v1
v1a ab
1
b
dec
a
dec
Dba,A
Fg1 = 0,
Fg = (2g 2)v1 a b v1 D11,A
a
v1
2
in complete agreement with [10], serving as an additional check for our computation.
8.2.3. Second-order relation
Finally, we can also study the decompactification limit of the second-order constraint (7.7),
whose left-hand side becomes the following differential operator:
dec
ab a b Da a,A
Fg .
Dbb,B
(8.29)
Using again the differentiation rules (8.23), we can evaluate (8.29) in a straightforward way
dec
E2dec ab a b Da a,A
Fg
Dbb,B
2
d 2g2
= Da a,A
Gg+1
24 2
2g3

1
L 2 1
R 2

P1L1
(2g 2)v1a v1a 22 ab a b P1L1 PbLb PBR q 2 (P ) q 2 (P )
(P L ,P R )
d 2 2g2
Gg+1
24 2

2
2g2 1 (P L )2 1 (P R )2
2g2 222 P L PAR PBR P1L1
q2
q 2
= 4
(P L ,P R )
d 2 2g2
Gg+1
24 2

2 L 2g2 1 (P L )2 1 (P R )2
(2g 1) 22 P L
P11
q2
q 2
gAB Fgdec .
+ AB
(P L ,P R )
Following the same steps as before, we can re-write the first two lines as total derivatives with
respect to , namely
2
2g 2g2
1
d
L 2 1
R 2
dec
Gg+1
2 P1L1
E2 = 8i
24
L R

+ 2iAB
(P ,P )
d 2
G
24 g+1
(P L ,P R )
2g1 L 2g2 1 (P L )2 1 (P R )2
P11
gAB Fgdec ,
q2
q 2
378
which after a partial integration become

2
2g2
1
d 2g2
L 2 1
R 2
dec
2
P1L1
G
E2 = 4
g
2
24
(P L ,P R )
2
2g2 1 L 2 1 R 2
d 2g3
P1L1
AB
G
q 2 (P ) q 2 (P ) gAB Fgdec .
g
2
24
L R
(P ,P )
In order to simplify this result, we also evaluate the expression

dec
D11,A
D11,B
Fg1
2
2g3
1
d 2g3
L 2 1
R 2
P1L1
= 2D11,A
2
Gg
PBR q 2 (P ) q 2 (P )
24
(P L ,P R )
2
2g2
1
d 2g2
L 2 1
R 2
P1L1
= 4 2
2
Gg
PaLa PAR PBR q 2 (P ) q 2 (P )
24
(P L ,P R )
2
2g2
1
d 2g3
L 2 1
R 2
P1L1
AB
2
Gg
PaLa q 2 (P ) q 2 (P ) .
24
L R
(8.30)
(P ,P )
We can thus obtain the relation
dec
dec
dec
ab a b Da a,A
D11,B
Fg1 gAB Fg .
Fg = D11,A
Dbb,B
(8.31)
As already mentioned in Section 8.1, the general structure of this equation can be anticipated
from field theory, especially, the existence of the term proportional to AB on the right-hand
side of (8.31). However, due to the lack of the setup of conformal supergravity, we are not in
a position to predict the exact coefficient g, which is also different from the coefficient in the
four-dimensional analog of the second-order constraint (7.7).
Note finally, that the six-dimensional couplings Fgdec (8.20) of the 1/2-BPS effective operator
(8.7), although obtained by taking the decompactification limit of the N = 4 topological ampli(3)
tudes Fg , they are not given by the topological theory on K3. The reason is that in their exact
genus g + 1 type II expression, the det dI m factors from the spacetime coordinates do not
cancel. Thus, these couplings are semi-topological, in the sense that string oscillator modes do
not contribute, and upon compactification on a T 2 they become exactly topological.
9. Conclusions
In conclusion, in this work, we generalized the holomorphicity property of the N = 2 supersymmetric couplings involving vector multiplets to the moduli dependence of the N = 4
couplings of 1/2-BPS operators defined in harmonic superspace. An example of such operators
is provided by the two series found in [5], involving powers of the (superdescendant of the)
N = 4 chiral Weyl superfield K whose coupling-coefficients are functions of the N = 4 vector
moduli and are computed by the N = 4 topological string on K3 T 2 .
The resulting harmonicity or analyticity property is expressed in terms of two sets of differential constraints: the first requires the vanishing of one scalar and one harmonic derivatives, while
the second imposes two scalar (covariant) derivatives to give back the same coupling up to a multiplicative constant proportional to its (super)conformal weight. We verified these equations on
379
the string side using the explicit expressions for the couplings of one of the two series as 1-loop
heterotic integrals on T 6 .
We also extended the above analysis to N = 2 1/2-BPS terms in six dimensions and we
checked the resulting equations for the couplings obtained in the decompactification limit of the
four-dimensional N = 4 topological amplitudes considered before. In principle, our analysis can
be generalized in a straightforward way to the couplings of any 1/2-BPS operator of extended
supersymmetry in any spacetime dimension.
Acknowledgements
We have profited form enlightening discussions with N. Berkovits, B. de Wit, S. Ferrara and
E. Ivanov. This work was supported in part by the European Commission under the RTN contract
MRTN-CT-2004-503369, in part by the INTAS contract 03-51-6346 and in part by the French
Agence Nationale de la Recherche, contract ANR-06-BLAN-0142. The work of S.H. was supported by the Austrian Bundesministerium fr Wissenschaft und Forschung.
References
[1] I. Antoniadis, E. Gava, K.S. Narain, T.R. Taylor, Topological amplitudes in string theory, Nucl. Phys. B 413 (1994)
162, hep-th/9307158.
[2] M. Bershadsky, S. Cecotti, H. Ooguri, C. Vafa, KodairaSpencer theory of gravity and exact results for quantum
string amplitudes, Commun. Math. Phys. 165 (1994) 311, hep-th/9309140.
[3] S. Cecotti, P. Fendley, K.A. Intriligator, C. Vafa, A new supersymmetric index, Nucl. Phys. B 386 (1992) 405,
hep-th/9204102.
[4] M. Bershadsky, S. Cecotti, H. Ooguri, C. Vafa, Holomorphic anomalies in topological field theories, Nucl. Phys.
B 405 (1993) 279, hep-th/9302103.
[5] I. Antoniadis, S. Hohenegger, K.S. Narain, N = 4 topological amplitudes and string effective action, Nucl. Phys.
B 771 (2007) 40, hep-th/0610258.
[6] N. Berkovits, C. Vafa, N = 4 topological strings, Nucl. Phys. B 433 (1995) 123, hep-th/9407190.
[7] A. Galperin, E. Ivanov, S. Kalitsyn, V. Ogievetsky, E. Sokatchev, Unconstrained N = 2 matter, YangMills and
supergravity theories in harmonic superspace, Class. Quantum Grav. 1 (1984) 469.
[8] A. Galperin, E. Ivanov, S. Kalitsyn, V. Ogievetsky, E. Sokatchev, Unconstrained off-shell N = 3 supersymmetric
YangMills theory, Class. Quantum Grav. 2 (1985) 155.
[9] G.G. Hartwell, P.S. Howe, (N, P , Q) harmonic superspace, Int. J. Mod. Phys. A 10 (1995) 3901, hep-th/9412147;
P.S. Howe, G.G. Hartwell, A superspace survey, Class. Quantum Grav. 12 (1995) 1823.
[10] H. Ooguri, C. Vafa, All loop N = 2 string amplitudes, Nucl. Phys. B 451 (1995) 121, hep-th/9505183.
[11] A.S. Galperin, E.A. Ivanov, V.I. Ogievetsky, Grassmann analyticity and extended supersymmetries, JETP Lett. 33
(1981) 168, Pisma Zh. Eksp. Teor. Fiz. 33 (1981) 176.
[12] L. Andrianopoli, S. Ferrara, E. Sokatchev, B. Zupnik, Shortening of primary operators in N -extended SCFT(4) and
harmonic-superspace analyticity, Adv. Theor. Math. Phys. 4 (2000) 1149, hep-th/9912007.
[13] E. Cremmer, J. Scherk, S. Ferrara, Phys. Lett. B 74 (1978) 61.
[14] E. Bergshoeff, M. de Roo, B. de Wit, Extended conformal supergravity, Nucl. Phys. B 182 (1981) 173.
[15] M. de Roo, Matter coupling in N = 4 supergravity, Nucl. Phys. B 255 (1985) 515.
[16] A. Galperin, E. Ivanov, V. Ogievetsky, E. Sokatchev, Conformal invariance in harmonic superspace, preprint JINR
E2-85-363, (1985), in: I. Batalin, C.J. Isham, G. Vilkovisky (Eds.), Quantum Field Theory and Quantum Statistics,
vol. 2, Adam Hilger, Bristol, 1987, pp. 233248.
[17] A.S. Galperin, E.A. Ivanov, V.I. Ogievetsky, E. Sokatchev, N = 2 supergravity in superspace: Different versions
and matter couplings, Class. Quantum Grav. 4 (1987) 1255.
[18] A.S. Galperin, E.A. Ivanov, V.I. Ogievetsky, E.S. Sokatchev, Harmonic Superspace, Cambridge Univ. Press, Cambridge, UK, 2001, 306 p.
[19] J.A. Bagger, A.S. Galperin, E.A. Ivanov, V.I. Ogievetsky, Gauging N = 2 sigma models in harmonic superspace,
Nucl. Phys. B 303 (1988) 522.
380
[20] A. Galperin, E. Ivanov, O. Ogievetsky, Harmonic space and quaternionic manifolds, Ann. Phys. 230 (1994) 201,
hep-th/9212155.
[21] E. Ivanov, G. Valent, Quaternionic metrics from harmonic superspace: Lagrangian approach and quotient construction, Nucl. Phys. B 576 (2000) 543, hep-th/0001165.
[22] S. Ferrara, E. Sokatchev, Superconformal interpretation of BPS states in AdS geometries, Int. J. Theor. Phys. 40
(2001) 935, hep-th/0005151.
[23] I. Antoniadis, E. Gava, K.S. Narain, T.R. Taylor, N = 2 type II heterotic duality and higher derivative F terms, Nucl.
Phys. B 455 (1995) 109, hep-th/9507115.
Nuclear Physics B 794 [PM] (2008) 381401

Perturbed betagamma systems and complex geometry

Anton M. Zeitlin 1
Department of Mathematics, Yale University, 442 Dunham Lab, 10 Hillhouse Avenue, New Haven, CT 06511, USA
Received 29 July 2007; accepted 5 September 2007
Available online 11 September 2007
Abstract
We consider the equations, arising as the conformal invariance conditions of the perturbed curved beta
gamma system. These equations have the physical meaning of Einstein equations with a B-field and a
dilaton on a Hermitian manifold, where the B-field 2-form is imaginary and proportional to the canonical
form associated with Hermitian metric. We show that they decompose into linear and bilinear equations and
lead to the vanishing of the first Chern class of the manifold where the system is defined. We discuss the
relation of these equations to the generalized MaurerCartan structures related to BRST operator. Finally
we describe the relations of the generalized MaurerCartan bilinear operation and the Courant/Dorfman
brackets.
PACS: 11.25.Hf; 11.25.Wx; 02.40.Ky; 02.40.Tt; 04.20.Jb
Keywords: String theory; Conformal field theory; Sigma model
1. Introduction: Perturbations of systems and sigma models

The so-called (betagamma) systems recently drew a lot of attention. The interest to them
started in the mathematics papers [1,2] where they were considered in the context of the sheaves
of vertex operators and chiral de Rham complex. Soon after that, the obtained results appeared
to be very useful in applications to string physics, e.g. topological strings [3], the geometry of
(2, 0) models [4], mirror symmetry [5], and the pure spinor superstring formalism [6,7].
E-mail address: anton.zeitlin@yale.edu.
URLs: http://pantheon.yale.edu/~az84, http://www.ipme.ru/zam.html.
1 On leave of absence from the St.-Petersburg Division of Steklov Mathematical Institute.
doi:10.1016/j.nuclphysb.2007.09.002
382
A.M. Zeitlin / Nuclear Physics B 794 [PM] (2008) 381401
The system is a two-dimensional conformal field theory with the action:

1
i,
S0 =
d 2 z i
2h
(1)
where i and j are of conformal weights (1, 0) and (0, 0) correspondingly and h is some
constant. The action above leads to the following operator product between these two fields:
i (z1 )j (z2 )
hji
(2)
.
z1 z2
This model looks very simple, however, if we take into account the geometric aspects, namely
treating i (z) dz as the sections of (T M) T , where describes a map from Riemann
surface to the complex manifold M and prime denotes the holomorphic part of the appropriate cotangent bundle, the system becomes highly nontrivial. For example, one can ask
a question: how to keep operator products preserved after coordinate transformations or how to
keep the appropriate energymomentum tensor, well defined on the intersections of the appropriate coordinate patches (for a nice review see [7]). The answer to the first question is nontrivial
but the second one is quite simple. In order to be unaffected under the coordinate change, the
energymomentum tensor should be modified from the original one, T0 = h1 i i to the
following:
T = h1 i i 1/2 2 log ( ),
(3)
where is the density function for the holomorphic top form on the manifold M. This leads
to a globally defined conformal invariant theory. An obvious requirement for the possibility of
such modification is that this top form related to should be nonvanishing, or in other words the
manifold M should have the vanishing first Chern class [4,7].
The inclusion of the additional dilatonic term in the energymomentum tensor (3) yields the
modification of the action [8]:

1
i + 1 h sR (2) (s) log ( ) ,
S0 =
(4)
d 2 z i
2h
4
R (2)
where
is the curvature of the worldsheet metric sab . It is worth noting also that this term
entering the action is the secondary characteristic class associated with class c1 (M)c1 ().
It appears that the system considered above can be treated as an infinite-radius limit of
the certain nonlinear sigma-model [7,9]. Really, let us consider the action for the sigma model

1
+ h sR (2) (s) .
S=
(5)
d 2 z (G + B )X X
4h
We impose several conditions on the metric and the B-field. First of all we require the metric to be
Hermitian in the certain system of coordinates: Gik = Gik = 0, Gi k gi k . The other conditions
concern B-field:
gi k + Bi k = 0,
Bik = Bik = 0.
This means that the action (5) can be rewritten as follows:

1
i X j + 1/2h sR (2) (s)
S=
d 2 z gi j X
2h
(6)
(7)
383
and via the first-order formalism, this sigma model can be transformed [9] to the first-order one:

1
i + p X i g i j pi p + 1/2h sR (2) (s)0 ,
d 2 z pi X
Sg =
(8)
i
j
2h
where 0 = log g. The infinite radius limit of the sigma model therefore corresponds
to the action above without the perturbation operator g i j pi pj . Obviously, the identification of
the resulting model with the conformally invariant, well-defined system and its complex
conjugate, can be done by means of the identification i = pi and i = X i but only in the case
if 0 decomposes in the sum of holomorphic and antiholomorphic functions.
In Section 2, we examine the geometric and algebraic properties of the conformal invariance
conditions of the sigma model (i.e. beta-function) of the zero and first order in h of the theory
with action (7). These conditions are the Einstein equations with a B-field and a dilaton [1114]
equipped with constraints (6):
1
R = H H + 2 ,
4
H 2( )H = 0,
1
4( )2 4 + R + H H = 0.
12
(9)
Expressing them in terms of g i j and 0 , we find out that one of the equations is simply
i k 0 = 0,
(10)
what precisely means that 0 is given by the sum of holomorphic and antiholomorphic terms
which by construction lead to the existence of nonvanishing holomorphic top form and therefore
to the vanishing of the first Chern class of the manifold. The equations on g i j are either linear or
bilinear.
In Section 3, we discuss another algebraic structure, rudiments of which were given in [9,22].
Namely, in [9] there was introduced a conjecture (at first introduced by A.S. Losev [10]), that the
conformal invariance condition for the perturbed system (8) can be expressed as follows:

C1 (0) + C2 (0) , (0) + C3 (0) , (0) , (0) + = 0.
(11)
Here Cn are graded symmetric multilinear operations satisfying special quadratic relations generating L -like structure [15,16], such that C1 ( (0) ) = [Q , (0) ], where Q is the usual BRST
operator [17,18] associated with energymomentum tensor (3), and (0) is a differential polyno
mial in c, c ghost fields, of ghost number 2 and [b1 , [b1 , (0) ]] = h1 g i j pi pj . We explicitly

n (0)
construct the C2 operation and expanding (0) =
n=1 t n by the formal parameter t , we find
the agreement between (11) and the conformal invariance conditions on g i j , 0 we have derived
2
before, up to the order t . Moreover, we expect that the symmetries of (11) have the form:

(0) = C1 (0) + C2 (0) , (0) + C3 (0) , (0) , (0) + ,
(12)
where is infinitesimal and (0) is of ghost number 1. In the last part of Section 3.2, we obtain that
the symmetry under holomorphic transformations of the equations on g i j , 0 can be obtained in

such a way. In Section 3.3, we indicate some important properties of the bilinear operation C2
related to pure chiral data. Namely, we derive the relation of C2 to Courant [25] and Dorfman [26]
brackets.
384
In Conclusions we underline some important features of the above formalism and outline the
ways of further development.
2. Geometric aspects of conformal invariance conditions
In this section, we show that the conformal invariance conditions of the perturbed system
(i.e. Einstein equations) reduce to the bilinear system of equations on g i j . First of all, we define
some algebraic operations on the sections (T M T M), where we denoted by T M and T M
holomorphic and antiholomorphic tangent bundle T M = T M T M correspondingly.
Suppose g = g i j i j , h = hi j i j are bivector fields on some complex manifold M. If

one expands in some neighborhood g = I v I v I , h = J w J w J (sum over I and J can
be possibly infinite), where v I ,w J (v I , w J ) are (anti)holomorphic sections of T M (T M), one
can construct a new bivector field [9]:

Jg, hK =
(13)
v I , w J v I , w J .
I,J
It appears that this bivector field has the explicit expression in terms of g, h and therefore leads
to the following definition.
Definition 2.1. Let g, h (T M T M) written in components as g i j i j , hi j i j .

Then one can define symmetric bilinear operation
J , K : (T M T M) (T M T M) (T M T M)
(14)
written in components as follows:

Jg, hKk l k l g i j i j hk l + hi j i j g k l i g k j j hi l i hk j j g i l k l.
(15)
Property (Important observation). One can easily find the same structure in Kaehler geometry.
If metric associated with g is Kaehler, then the Ricci tensor R i j , expressed in terms of bilinear
i
j
vector field g i j associated with metric tensor is proportional to Jg, gK, more precisely
1
R i j (g) = Jg, gKi j .

2
(16)
Definition 2.2. Suppose the complex manifold M under consideration is equipped with nonvanishing holomorphic top degree form , which in local coordinates can be written as follows:
= ef (X) dX 1 dX D/2 , where D is the dimension of the manifold. Then one can define:
(1) A covariant divergence div : (T M) C(M) of a section v = v i i (T M), such
that
div v = 1 Lv = i v i + i f v i .
: (T M
(17)
T M)
(T M)
: (T M
and div
(2) A covariant divergences div
T M) (T M) of a bivector field such that in local coordinates it has the expression:

div g = j g i j + j f g i j i .
div g = i g i j + i f g i j j ,
(18)
385
We already know from Introduction that the conditions of conformal invariance for the perturbed system is equivalent to the system of Einstein equations with B-field and a dilaton.
Now we reexpress Einstein equations in terms of a bivector field g i j .

Proposition 2.1. The equations
1
R = H H
2 ,
4
H 2( )H = 0,
(19)
where metric, B-field, and a dilaton are expressed as follows:
Gi k = gi k ,
Bi k = gi k ,
= log g + 0 ,
(20)
are equivalent to the following system:
i k 0 = 0,
p dl 0 g lk = 0,
p dl 0 g kl = 0,
2g r l r lg i k 2r g i p p g r k g i l lds0 g s k g r k r dj 0 g j i
+ r g i k dj 0 g j r + p g ki dn0 g np = 0,
(21)
where di 0 g i j i g i j 2i 0 g i j and di 0 g ij i g j i 2i 0 g j i .
Proposition 2.2. The equation
1
(22)
H H = 0,
12
where metric, B-field, and dilaton are constrained by (20) and governed by Eqs. (19), is equivalent to the following one:
4( )2 4 + R +
di 0 dj 0 g j i = 0,
(23)
where di 0 dj 0 g j i (j 2j 0 )(i 2i 0 )g i j .
The proofs of the above two propositions are given in Appendix A.
Before reexpressing the equations on g i j in terms of the algebraic structures introduced above,
we note several interesting properties. The first observation is how the equations depend on the
bivector field. It is easy to see that the dependence is either linear or pure bilinear in g i j . This is
very unusual for the equations of Einstein type which are highly nonlinear in the general case.
Then one can notice that if there exists a global solution (on the whole manifold M) to the
equations from Proposition 2.1, then the first Chern class c1 (M) should vanish or, in other words,
the manifold M should possess a holomorphic top form. This is due to one of the equations from
the Proposition 2.1, namely i k 0 = 0. Really, density function e20 (since dilaton is a
well-defined function and det(gi j ) is a well-defined density function for the volume form, therefore e20 = e2 det(gi j ) is also a density function for the volume form) locally decomposes
which globally
into the product of holomorphic and antiholomorphic functions (X) and (
X)
serve as densities for the holomorphic and antiholomorphic volume forms and such that
= e20 dX 1 dX n dX 1 dX n .
(24)
386
Another property is that div (g) = di 0 g i j j and div (g) = dj 0 g j i i become (again, if the
Einstein equations from Proposition 2.1 admit the global solution) an antiholomorphic and
holomorphic sections of T M and T M correspondingly. Also, one can easily notice that this
condition automatically leads to the equation
di 0 dj 0 g j i = const.
(25)
This should not be strange since the equation

4( )2 4 + R +
1
H H = const
12
is a direct consequence of equation

1
+ 2 = 0.
R H H
4
(26)
(27)
So, in our case this is just a particular example of the general statement.
Finally, let us summarize and formulate the main result of this section as the following proposition.
Proposition 2.3. The system of Eqs. (19) and (22) with additional constraint (20) on the manifold M is equivalent to the following:
(1) There exists a nonvanishing holomorphic top degree form = (X) dX 1 dX D/2
on the manifold M (and therefore the first Chern class c1 (M) should vanish) such that
= e2(X,X)
(X)(
X)
g(X, X).
(28)
(T M)
(T M)
and div (g)

are respectively holomorphic
(2) Vector fields div (g)
and antiholomorphic.
(3) Bivector field g (T M T M) obeys the following two equations:
Jg, gK + Ldiv (g) g + Ldiv (g) g = 0,
div div (g) = 0,
(29)
where Ldiv (g) and Ldiv (g) are Lie derivatives with respect to the corresponding vector fields.
Remark. It is easy to see that when gi j is the Kaehler metric, then the bilinear equation on g
from (29) is automatically satisfied and Eqs. (19), (22) with constraint (20) reduce to linear ones.
3. Perturbed system and generalized MaurerCartan equations
3.1. Motivation
From Section 2 we learned that the conditions of conformal invariance for system per
turbed by g i j pi pj operator are given by bilinear and linear equations on the corresponding
bivector field. In paper [9] the conditions of conformal invariance for this first-order sigma model
were studied naively via cut-off regularization. The resulting equations which correspond to the
case when 0 = 0 (the additional dilaton field was ignored there) and g i j pi pj is a primary field
(i g i j = 0 and j g i j = 0) have the following form:

Jg, gK = 0.
(30)
c with operator products c(z)b(w)

Introducing the standard ghost fields b, c and b,
1
c(z)
b(w) z w , and BRST operator [17]:

1
JB , JB = jB dz jB d z ,
Q=
2i
jB = cT + :bcc:, jB = cT + :b c c:,
387
1
zw
and
(31)
where T , T are the holomorphic and antiholomorphic components of the energymomentum

tensor, we found out that the resulting equations can be rewritten as follows:

Q, (0) = 0,
(32)
lim P
(1) (0) (z) = 0,
h0
C,z
where (0) = cch

1 g i j pi pj , (1) = dz ch
1 g i j pi pj d z ch1 g i j pi pj , C,z is the contour
around point z, and P is the projection on the -independent terms.2 This suggests that the
complete system of equations given in Proposition 2.3 and also others entering beta function
with higher orders in h should be written in the MaurerCartan form:

Q, (0) + M (0) , (0) + = 0,
(33)
where

M (0) , (0) (z) P

(1) (0) (z)
(34)
C,z
is a bilinear operation and (0) , (1) are some modifications of (0) , (1) . In Section 3.2 we
define properly (0) , (1) , and M.
3.2. MaurerCartan form of conformal invariance conditions
3.2.1. Notation and conventions
Descent hierarchy and ghost fields. Throughout this section we will denote a conformal field
as a 0-form with the corresponding superscript if it is a differential polynomial in c, c ghost
fields, where the matter fields are coefficients, and denote the space of such fields as H 0 . We also
introduce the ghost number operator

Ng = (dz jg d z jg ),
(35)
where jg = bc and jg = b c.
If (0) is the eigenvalue of this operator with the eigenvalue n , we
say that this field is of ghost number n (it is obvious that it can be only nonnegative integer).
We associate with any field (0) depending on matter and ghost c, c fields the following 1-form
and 2-form:

(1) = dz b1 , 0 + d z b1 , (0) ,
(36)
(2) = dz d z b1 , b1 , (0) .
2 Here we point out that the bilinear operation similar to (32) was considered in [20] in the context very close to ours,
namely the comparison of the conditions of BRST-invariance of the conformal perturbation theory and the string field
theory equations of motion.
388
Operator products. We assume that all matter field operators enjoy the operator products of the
following type:
m

V (z)W (z ) =
n

(V , W )(r,s) (z )(z z )r (z z )s .
(37)
r= s=
Throughout this section we assume that all operators are ordered with respect to holomorphic
normal ordering. We neglect the sign of normal ordering if it does not lead to misunderstanding.
3.2.2. Bilinear operation and its properties
First of all we give the definition of the bilinear operation itself.
Definition 3.1. For any two fields (0) , (0) we define the bilinear operation M : H 0 H 0 H 0 ,

(0) (0)
1
(1) (0)
n n 1
(z) + (1)
(1) (0) (z),
P
P
M , (z) =
(38)
4i
4i
C,z
C,z
where P is a projection on the 0 term.

(0)
(0)
Remark (Comment about the projection operator). Really, for any 1 , 2 under con
(1) (0)
(0)
(0)
sideration C,z 1 2 (z) lies in the space of formal power series H 0 () = { | =

(0)
n (0)
0
n=k n , n H }. Therefore, the operator P is well defined and projects on the coefficient of 0 of the corresponding element of H 0 ().
Definition 3.1 leads to the following properties.
Proposition 3.1. Operation M satisfies the following relation:

Q, M (0) , (0) + M Q, (0) , (0) + (1)n M (0) , Q, (0) = 0.
(39)
Proof. First, we need to show

that BRST operator commutes with projection operator P. Really,
let us denote f (V , W )(z) = C,z dw V (w)W (z) for some operators V , W .

n
From (37) we know that f (V , W ) =
n=k fn (V , W ) . The projection operator acts as
follows: Pf (V , W ) = f0 (V , W ). Therefore we see that P[Q, f (V , W )] = [Q, Pf (V , W )] =
[Q, f0 (V , W )]. In such a way we see that BRST operator commutes with projection operator
and hence the relation (39) can be easily established by means of the simple formula [Q, (1) ] =
d (0) [Q, (0 )](1) . 2
Remark. Relation (39) is similar to the basic property of the string 2-products [15].
Proposition 3.2. The expression for the 2-form associated with M( (0) , (0) ) is given by the
following formula:
(2)

M (0) , (0)

1
1
=
(1) (2) (z) + (1)n n
(1) (2) (z) + d (1) ,
(40)
P
P
2i
2i
C,z
C,z
389
where as usual M( (0) , (0) )(2) = dz d z [b1 , [b1 , M( (0) , (0) )]], (1) is some 1-form, and
d is the de Rham differential.
The proof can be easily obtained from Proposition 2.1 of [22].
3.2.3. MaurerCartan structures and systems
We suggest (according to the hypothesis of A.S. Losev [10]) that the equations governing the
conformal invariance of system (8) can be summarized via the following generalized Maurer
Cartan equation:

C1 (0) + C2 (0) , (0) + C3 (0) , (0) , (0) + = 0
(41)
such that Cn are graded (with respect to ghost number) symmetric bilinear operations, C1 ( (0) ) =
[Q , (0) ] and C2 = 12 M, where Q is the BRST operator from the previous section such that
holomorphic and the antiholomorphic components of the energymomentum tensor are those for
accurately defined system (see Section 1):

Q = J ,

i 1/2 2 log ,
J = dz c h1 pi X i 1/2 2 log d z c h1 pi X
(42)
the 2-form (2) associated with (0) is the perturbation 2-form (2) = dz d z V such that
V = h1 g i j pi pj . We also put the following constraints on (0) :

b0 (0) = 0,
(43)
where b0 = b0 b0 .
This gives the following expression for (0) :
+ c(c + c)W
c(c
+ c)
W + 1/2c 2 cU 1/2c 2 cU
(0) = ccV
2 cY + (c + c)
2 cY + ,
c
2 cX + c 2 cX (c + c)
(44)
Y , Y are some matter fields and stand for the terms depending on
where W , W , U , U , X, X,
n c or m c such that n, m > 2. We will refer to the terms with U and U as dilatonic terms. More
U = f(X, X).
over we make these dilatonic terms only X, X-dependent,

that is U = f (X, X),
The W , W terms will be called in the following as gauge terms, so we will refer to the constraint
Y, Y b0 (0) = b0 (0) = 0 (which is equivalent to W = W = 0) as a gauge condition. The X, X,
terms have no physical interpretation, therefore we get rid of them reducing to the subspace.
Definition 3.2. The space S 0 consists of the elements (0) H 0 which enjoy three properties:
(1) n = 2,
(2) b0 (0) = 0,
(3) bi bj (0) = 0 if i + j > 1.
Remark. As we see the general form of the element from (0) S 0 is as follows:
+ c(c + c)W
c(c
+ c)
W + 1/2c 2 cU 1/2c 2 cU + ,
(0) = ccV
where stand for the terms depending on n c or m c such that n, m > 2.
(45)
390
If we expand the field (0) in the series of some formal parameter t : (0) =
the first two orders of expansion of Eq. (41) give the following equations:

(0)
Q , 1 = 0,
(0)
Q , 2
1 (0) (0)
+ M 1 , 1 = 0.
2
(0) n
n=1 n t , then
(46)
We show that (46) lead to Eqs. (21) and (23) expanded to the second order in the formal parameter. Let us formulate this as proposition.
Proposition 3.3. Let (0) be the element of S 0 , satisfying the following conditions: (2) = dz
i p , the dilatonic terms U (X, X)

= b1 b1 (0) , and U (X, X)
= b1 b1 (0) are
d z h1 g i j (X, X)p
j
respectively functions of X i (z), X i (z). Then (46) gives Eqs. (21) and (23) expanded up to the
i j
second order in the formal parameter t, such that the expansion of matter fields is: g i j = tg1 +
i j
+ (tU1 + t U 1 ) + t 2 (U2 + U 2 ) + ).
t 2 g + , 0 = 1/2( log()
2
The proof is given in Appendix B.

Remark 1. One could also put the condition that U , U are holomorphic and antiholomorphic
functions correspondingly. Really, we can consider Q + (1) as a deformation of a BRST current [21,22]. From the proof of Proposition 3.3 one can see that adding a total derivative to
(1) we can get rid of non(anti)holomorphic terms from U 1 (U1 ). Hence it is natural to put this
additional constraint.
Remark 2. Applying b1 b1 to (46) and using Proposition 3.2, we get the following equations:

1
Q , 1(2) = d1(1) ,
(47)
Q , 2(2) (z) +
1(1) 1(2) (z) = d2(1) (z)
P
2i
C,z
which can be interpreted as a conservation law for the deformed BRST current in the presence
of perturbation (2) [22].
Remark 3. From Proposition 3.1 we know that there are relations between operations C1 and C2
from (41):

C1 C1 (0) = 0,

C1 C2 (0) , (0) + C2 C1 (0) , (0) + (1)n C2 (0) , C1 (0) = 0
(48)
for any (0) , (0) H 0 . We hope that similar quadratic relations will hold for all Cn and generate
the homotopy Lie algebra [15,16], like it was for string products.
3.2.4. Symmetries of MaurerCartan equation
In Section 3.2.3 we made a conjecture that the conditions of conformal invariance for
model (8) coincide with a sort of MaurerCartan equation. However, we want our equation to
possess symmetries, more precisely we want them to be in the following form:

(0) = C1 (0) + C2 (0) , (0) + C3 (0) , (0) , (0) + ,
(49)
391
where is infinitesimal and (0) H 0 is of ghost number 1. We will see now that at the lowest
order in the parameter h they have precisely this form. Really, let us look on Eqs. (46). The first
equation of (46) due to the nilpotence of operator Q has the following symmetry:

(0)
(0)
1 = Q , 1 ,
(50)
(0)
where 1 is a zero form of ghost number 1 and is infinitesimal. Proposition 3.1 allows to
accompany this with the following transformation:
(0) (0)

(0)
(0)
2 = Q , 2 + M 1 , 1 ,
(51)
where n2 = 1 is again a 0-form of ghost number 1. Altogether they form a symmetry of (46). In
our case these symmetry transformations should correspond to the symmetries of (21), (23), i.e.
holomorphic coordinate transformations, therefore it is natural to give the following expression

n (0)
for (0) =
n=1 t n :

(0) = h1 cv i (X)pi cv i (X)p
(52)
i ,
where v i (X) are the components of the holomorphic section of T M. It appears that the secondorder approximation to symmetries considered above is very close to exact expression in our
case.
i p ,
Proposition 3.4. Let (0) S 0 satisfy the following conditions: (2) = dzd z h1 g i j (X, X)p
j
(0)
(0)
= b1 b1 , and U (X, X)
= b1 b1 . Then
the dilatonic terms be U (X, X)

(0) = Q , (0) + M (0) , (0) ,
(53)
where (0) is given by (52) and is infinitesimal, gives the following transformations:

(2) = dz d z h1 v g i j pi pj + O(h) ,

1
1
0 (X, X) =
div v(X) + div v (X) ,
2
2
(54)
where v g = (Lv g + Lv g) which coincides with the holomorphic coordinate transformation

of g and 0 = 1/2( log ()
+ U + U ).
The proof can be obtained easily by the direct calculation.
Remark. The appearance of the additional (noncovariant) terms of the higher order in h in (2)
after symmetry transformation (53) has deep roots in the very nature of curved systems (see
e.g. [1,7]). We will consider them elsewhere.
3.3. Relations of M-operation with Courant and Dorfman brackets
In this subsection we discuss a simple property of bilinear operation M which may lead to
some deep consequences in the understanding of the theory.
At first we remind the definition and point out some useful properties of Courant [25] and
Dorfman [26] brackets which appear to be very important in the theory of generalized complex
structures [23,24].
392
Definition 3.3. Let (vi , wi ) = vi + wi (i = 1, 2) be the sections of T M T M. Courant and

Dorfman brackets [ , ]c , [ , ]d : (T M T M) (T M T M) (T M T M) are defined
as follows:

1
(v1 , w1 ), (v2 , w2 ) c = [v1 , v2 ], Lv1 w2 Lv2 w1 d(iv1 w2 iv2 w1 ) ,
2

(v1 , w1 ), (v2 , w2 ) d = [v1 , v2 ], Lv1 w2 iv2 dw1 .
(55)
Now we give the properties which provide the differences between Courant and Dorfman
brackets and the Lie one.
Properties.
(1) The Dorfman bracket satisfies the Leibnitz rule, but it is not antisymmetric, moreover its
antisymmetrization leads to the Courant bracket, namely:

(v1 , w1 ), (v2 , w2 ) d + (v2 , w2 ), (v1 , w1 ) d = d (v1 , w1 ), (v2 , w2 ) ,
1
(56)
(v1 , w1 ), (v2 , w2 ) d (v2 , w2 ), (v1 , w1 ) d = (v2 , w2 ), (v1 , w1 ) c ,
2
where ((v1 , w1 ), (v2 , w2 )) = v1 w2 + v2 w1 is the symmetric bilinear form on (T M T M)
and d is the de Rham differential.
(2) The Courant bracket is antisymmetric but it does not satisfy Jacobi identity, more precisely
it satisfies Jacoby identity modulo exact (with respect to de Rham differential) term:

(v1 , w1 ), (v2 , w2 ), (v3 , w3 ) c c + (v3 , w3 ), (v1 , w1 ), (v2 , w2 ) c c

+ (v2 , w2 ), (v3 , w3 ), (v1 , w1 ) c c = dN (v1 , w1 ), (v2 , w2 ), (v3 , w3 ) ,
(57)
where N is the Nijenhuis operator [24].

For the proof and much other information on the subject including references see e.g. [24].
Recall that in Section 3.2.4 we studied the symmetries of the generalized MaurerCartan

equation (46). The symmetries were related to the 0-form (0) = h1 (cv i (X)pi cv i (X)p
i ),
i

where v (X) are the components of the holomorphic section of T M. Let us extend this 0-form
to the following one:

(0)
k
v,w
(58)
= h1 c v i (X)pi wk (X)X k h1 c v i pi w k (X)X
associated with a holomorphic section (v, w) = v i (X)i + wk (X)dX k (T M T M ).3 Let
us denote also M0 ( (0) , (0) ) = limh0 hM( (0) , (0) ) if such a limit exists.
Proposition 3.5. For any two holomorphic sections (vi , wi ) = vi + wi (T M T M )
(i = 1, 2),

(0)
M0 v(0)
(59)
= h[(v1 ,w1 ),(v2 ,w2 )]c .
, v(0)
1 ,w1
2 ,w2
3 We expect that (0) will generate holomorphic symmetries (not only coordinate invariance symmetry but also the
v,w
one associated with B-field) of the general first-order sigma model, see [9] and Conclusions of this paper.
393
The proof can be obtained by straightforward calculation.

This correspondence between M0 and Courant bracket leads to the following proposition.
(0)
(0)
Proposition 3.6. Let i = vi ,wi (i = 1, 2, 3) be the 0-forms associated with holomorphic sections (vi , wi ) T M T M . Then we have:
(0)
(0)
(0)
(0) (0)
(0) (0)
(0) (0)
+ M0 3 , M 1 , 2
+ M0 2 , M 3 , 1
M0 1 , M 2 , 3

= 0,
+ Q , N1,2,3 (X) + N 1,2,3 (X)
(60)
is its complex conjugate.
where N1,2,3 (X) = N ((v1 , w1 ), (v2 , w2 ), (v3 , w3 ))(X) and N 1,2,3 (X)
Proof. The proof is the immediate consequence of Proposition 3.5, property (57) of the Courant
i N 1,2,3 . 2
bracket, and the fact that [Q , N1,2,3 + N 1,2,3 ] = cX i i N1,2,3 + cX
i
(0)
Remark. We see that operation M acting on the space of the zero forms v,w is antisymmetric
and in the classical limit h 0 obeys Jacobi identity modulo Q -exact terms. This property
reminds the one unifying string 2- and 3-products leading to L -algebra [15]. Therefore if we
(0)
identify the operation M with 2-product, then 3-product for three vi ,wi 0-forms in the classical
limit gives Nijenhuis operator acting on the corresponding sections of T M T M.
From the definition of M it is evident that this operation is the supersymmetrization of the
following one:

1
(1) (0) ,
P
N (0) , (0) =
(61)
2i
C,z
that is M( (0) , (0) ) = 1/2(N ( (0) , (0) ) + (1)n n N ( (0) , (0) )). Therefore it is reasonable
to think about the correspondence of N0 = limh0 hN with the Dorfman bracket, in the sense of
the correspondence between M0 and the Courant one.
(0)
Proposition 3.7. Let us take v1 ,w1 (i = 1, 2) as in Proposition 3.5. Then the following holds:

(0)
= h[(v
, v(0)
,
N0 v(0)
1 ,w1
2 ,w2
1 ,w1 ),(v2 ,w2 )]d
(0)

= N0 v ,w , v(0),w + N0 v(0),w , v(0),w ,
Q , f12 (X) + f12 (X)
(62)
1 1
2 2
2 2
1 1
is its complex conjugate.
where f12 (X) = ((v1 , w1 ), (v2 , w2 ))(X) and f12 (X)
Remark 1. It is worth noting that relation (62) gives a very easy proof of the Leibnitz rule for
Dorfman bracket. The proof follows directly from the corresponding vertex operator algebra
axiom.
Remark 2. Similar statements to two propositions above were given in [19] in the context of
anomalous Poisson brackets in the first-order theories. Propositions 3.6 and 3.7 are the consequences of the general fact below.
(0)
Proposition 3.8. Let i = cJi (z) dz cJi (z) d z (i = 1, 2, 3), where Ji (z) (Ji (z)) are
(anti)holomorphic matter fields of conformal weight (1, 0) ((0, 1)). Then
394
(1) The N-operation is antisymmetric modulo BRST-exact term:

(0) (0)
(0) (0)
(0)
N 1 , 2 + N 2 , 1 = Q, 12 ,
(0)
(63)
(0)
where 12 is some 0-form of ghost number 0, and satisfies Leibnitz rule on i .

(2) The M-operation which is supersymmetrization of the N -operation, satisfies Jacobi identity for the fields of the form (0) modulo BRST-exact term:
(0)
(0)
(0) (0)
(0) (0)
M 1 , M 2 , 3
+ M 3 , M 1 , 2
(0) (0)
(0)
(0)
= Q, 123 ,
+ M 2 , M 3 , 1
(64)
(0)
where 123 is some 0-form of ghost number 0.

Proof. In order to prove the first part of the proposition one needs to consider the operator
product of Ji (z1 )Jj (z2 ):
(Ji , Jj )(2,0) (z2 ) (Ji , Jj )(1,0) (z2 )
+
,
(z1 z2 )
(z1 z2 )2
(Ji , Jj )(0,2) (z2 ) (Ji , Jj )(0,1) (z2 )
Ji (z1 )Jj (z2 )
+
.
(z1 z2 )
(z1 z2 )2
Ji (z1 )Jj (z2 )
(65)
Here (Ji , Jj )(2,0) and (Ji , Jj )(0,2) are the conformal fields of dimension (0, 0). They lead to the
relations:

Resz1 z2 Ji (z1 )c(z2 )Jj (z2 ) + Resz1 z2 Jj (z1 )c(z2 )Ji (z2 )
= c(z2 )L1 (Ji , Jj )(2,0) (z2 ),

z2 )Jj (z2 ) + Resz1 z2 Jj (z1 )c(z
2 )Ji (z2 )
Resz1 z2 Ji (z1 )c(
= c(z
2 )L 1 (Ji , Jj )(0,2) (z2 ).
(66)
(0)
This immediately leads to the first statement. Moreover one obtains that 12 = (J1 , J2 )(2,0) +
(J1 , J2 )(0,2) . To prove that Leibnitz rule holds, one just needs to use the appropriate axiom of
vertex operator algebra.
The second point of the proposition immediately follows if one rewrites RHS of (64) via
N -operation and uses the statements from point 1 of this proposition. 2
Remark. Similar statement to Proposition 3.8 was given in [2] in the context of the study of
chiral de Rham complex.
4. Conclusions and remarks
In this paper we have studied the curved system perturbed by the operator g i j pi pj . One
can also consider the general perturbation of conformal weight (1, 1):
j + b X i X
j + hR (2) (s)
Vgen = g i j pi pj ij pi X j ij pi X
ij
(67)
which, after integration over p-variables, transforms into the general string sigma model (5).
The same way we associated with g i j pi pj the bivector field, one can associate with pertur-
395
bation (67) an object from ((T M T M) (T M T M)). The conformal invariance

condition at one loop will no longer be bilinear but we suggest that the equations will be
described by means of the double commutator structure as in Definition 2.1 and its generalizations. Really, one can see that at the second order of perturbation theory in Vgen the term
(Vgen , Vgen )(1,1) leads to double commutator, where the commutators are replaced by the Dorfman brackets.
The formalism we developed in Section 2, allows us in principle to give the cohomological
meaning to the equations of conformal invariance, namely:

Q, (0) + C2 (0) , (0) + C3 (0) , (0) , (0) + = 0,
(68)
where Cn are graded (with respect to the ghost number) multilinear operations, satisfying
quadratic relations leading to L -like structure. In Section 2 we described C2 operation and
its properties. We expect that the next nontrivial operation C3 has the following form:

(0) (0) (0)
(1) (2) (0) (z),
C3 , , (z) = P
(69)
V,z
where the three-dimensional region V,z should depend on some parameter and P is the projection on the -independent term as in Definition 3.1.
We mention also that in comparison to sigma model, studying the perturbed system by
means of conformal perturbation theory looks somewhat more promising since the underlying
free CFT is the simplest possible. Again, in contrast to the usual sigma model throughout the
perturbation theory one can keep geometric structures not destroyed.
In the subsequent paper we will further develop the formalism related to M-operation
and apply it directly to the realistic sigma model (5) describing strings in background
fields.
Acknowledgements
The author is grateful to A.S. Losev for introduction in the subject and fruitful discussions. It
is important to mention that the hypotheses concerning the using of generalized MaurerCartan
equations and L -structures in the context of study of conditions of conformal invariance belong
to A.S. Losev. The author is very grateful to I.B. Frenkel, M. Kapranov and G. Zuckerman
for enumerous discussions on the subject and to I.B. Frenkel and N.Yu. Reshetikhin for their
permanent encouragement and support.
Appendix A
Proposition 2.1. The equations
1
R = H H
2 ,
4
H 2( )H = 0,
where metric, B-field, and a dilaton are expressed as follows:
Bi k = gi k ,
= log g + 0 ,
Gi k = gi k ,
(A.1)
(A.2)
(A.3)
396
are equivalent to the following system:
i k 0 = 0,
p dl 0 g lk = 0,
p dl 0 g kl = 0,
(A.4)
where di 0 g i j i g i j 2i 0 g i j and di 0 g ij i g j i 2i 0 g j i .
Proof. We use the following formula for the Ricci tensor [27]:
,
R = 1/2G G + ,
(A.5)
where
= G G ,
= ( + )
,
2
1
= G G log(G).
2
Remembering that Gi k = gi k and Bi k = gi k , this leads to:
1
R H H
+ 2
4
1
1
+ G G + ,
,
= H H
4
2
where
(A.6)
1

= ( + )
,
2
= G G + 2 0 .
Here 0 = log g and we denoted the determinant of matrix gi j by g. Now let us study the
, one has:
third term in (A.6): first, for the components of
1
rsi = g i k (r gks
+ s gkr
),
2
1
ris = g i k (s gr k k gr s ) and c.c.,
2
(A.7)
1
while all other components vanish. Therefore, one finds that i,r
, hence the third term
s = 2 Hs ir
2
in (A.6) provides the contribution of the H -type with an additional term in for = i and
= j:
1 k r

g r g l i + g l r r g k i g j p (k gpl
+ l gpk
)
4
1

k r
l i j p
= g k r r g l i g l r r g k i g j p (k gpl
l gpk
) g r g g l gpk
4
1
j
= H ikl Hkl + r g ik k g r j .
4
i,kl kl =
(A.8)
397
components can be rewritten in such

Thus we can see that Eqs. (A.1) with inhomogeneous (i, k)
a way:
1
k
R i k H i H
+ 2 i k
4

1 i k
+ k i
2
1 i l 0 s k 1 r k

r l
i k
i p
r k
= g r lg r g p g g lds g g r dj 0 g j i
2
2
1
1

+ r g i k dj 0 g j r + p g ki dn0 g np ,
2
2
= g r l r lg i k r g i p p g r k +
(A.9)
where we remind that = d 0 G = G + 2 0 G , while for homogeneous com k)

we get the following expression:
ponents (i, k and i,
1
j
R ij H i H + 2 i j
4

1
1

= i j + j i = g i p p dl 0 g lk + g k p p dl 0 g li .
2
2
(A.10)
+ 2 = 0 are equivalent to the equations on the

Hence the equations R 14 H H
bivector field g i j :

g i p p dl 0 g lk + g k p p dl 0 g li = 0 and c.c.
(A.11)
(A.12)
Now let us rewrite in a similar way the second series of equations, namely, the Maxwell-like
equations for the B-field (A.2). First of all we notice that by the antisymmetry of H and the
= 1 log(G),
formula
2

H = H +
(A.13)
H = gH .
Hence Eq. (A.2) is equivalent to

e20 H = 0.
(A.14)
Expressing H in terms of g i j and multiplying (A.14) on e20 , we arrive to the following system
of equations:

2g r l r lg i k 2r g i p p g r k g i l e20 s e20 lg s k g r k e20 j e20 r g j i


e20 l e20 g i p p g lk e20 g k p p g li = 0 and c.c.
(A.15)
(A.16)
Comparing the Eqs. (A.11) and (A.15) we get:

i k 0 = 0,
(A.17)
398
and then (A.12) and (A.16) lead to:
p dl 0 g lk = 0,
p dl 0 g l k = 0.
(A.18)
Thus the final system of the equations on g i j is:
p dl 0 g lk = 0,
p dl 0 g l k = 0,
i k 0 = 0.
(A.19)
Proposition 2.2. The equation

1
(A.20)
H H = 0,
12
where metric, B-field, and dilaton are constrained by (20) and governed by Eqs. (19), is equivalent to the following one:
4( )2 4 + R +
di 0 dj 0 g j i = 0,
(A.21)
where di 0 dj 0 g j i (j 2j 0 )(i 2i 0 )g i j .
Proof. Using Eq. (A.1) one can reduce (A.20) to
1
4( )2 2 H H = 0.
6
(A.22)
Let us reexpress each term from Eq. (A.22) by means of g i j and 0 :
4( )2 = 2g i k i log gk log g + 4g i k i log gk 0 + 4g i k k log gi 0
+ 4g i k k log gi 0 + 8g i k i 0 k 0 ,
(A.23)
2 = 2G +
= 2G 2G log g 2 G
= 2g i k i k log g p g pr r log g s g sk k log g
2p g pr r 0 2s g sk k 0 2g i k i log gk log g
1
H H
6
2g i k i log gk 0 2g i k k log gi 0 ,

= H lmn Hlmn = g mp p g l n g l r r g mn (l gmn m gl n )
(A.24)
= 2g l r r g mn m gl n 2g l r r g mn l gmn .
(A.25)
Now rewriting (A.20) we get:
2g i k i k log g p g pr r log g s g sk k log g
2g i k i log gk 0 2g i k k log gi 0 + 2g i k k log gi 0

+ 2g i k k log gi 0 + 8g i k i 0 k 0 = 0.
(A.26)
399
Let us consider the sum of this equation with (A.4) contracted with gi k :
2g r l r lg i k gi k + 2r g i p p g r k gi k + k ds0 g s k + i dj 0 g j i
+ r log gdj 0 g j r + p log gdn0 g np = 0.
(A.27)
In such a way, using the formula g = g i k gi k we get:
8g i k i 0 k 0 + k ds0 g s k + i dj 0 g j i = 0,
(A.28)
which can be rewritten as:
di 0 dj 0 g j i = 0.
(A.29)
Appendix B
Proposition 3.3. Let (0) be the element of S 0 , satisfying the following conditions: (2) = dz
i p , the dilatonic terms U (X, X)

= b1 b1 (0) , and U (X, X)
= b1 b1 (0) are
d z h1 g i j (X, X)p
j
respectively functions of X i (z), X i (z). Then (46) gives Eqs. (21) and (23) expanded up to the
i j
second order in the formal parameter t , such that the expansion of matter fields is: g i j = tg1 +
i j
+ (tU1 + t U 1 ) + t 2 (U2 + U 2 ) + ).
t 2 g + , 0 = 1/2( log()
2
Proof. The first equation of (46) leads to the following relations between U -, W -, and V -terms:
(L0 V1 )(z) V1 (z) + L1 W 1 + L 1 W1 = 0,

W 1 = 1/2 (L1 V1 ) + L 1 U1 ,
W1 = 1/2 (L 1 V1 ) + L1 U 1 ,
L1 W1 = 0,
L 1 W 1 = 0.
(B.1)
Therefore it is clear that

i j
i
W1 = 1/2dj
g1 pi 1/2X i U ,
i j
i U,
W 1 = 1/2di g1 pj 1/2X
i
(B.2)
where = 1/2 log( + ),

and hence (B.1) gives the equations:
i k (U1 + U 1 ) = 0,
i j
k dj
g1 = 0,
i j
k di g1 = 0,
i j
dj
di g1 = 0,
(B.3)
which coincide with (21), (23) at the first order in t .

The second equation of (46) leads to more complicated conditions:
(L0 V2 ) V2 1/2(V1 , V1 )(1,1) + 1/2(W 1 , V1 )(0,1) 1/2(V1 , W 1 )(0,1)
+ 1/2(W1 , V1 )(1,0) 1/2(V1 , W1 )(1,0) + L1 W 2 + L 1 W2 = 0,

W 2 = 1/2 (L1 V2 ) (V1 , V1 )(2,1) + (W 1 , V1 )(1,1) + (W1 , V1 )(2,0)

+ 1/2(U1 , V1 )(1,0) 1/2(V1 , U1 )(1,0) + L 1 U2 ,

W2 = 1/2 (L 1 V2 ) (V1 , V1 )(1,2) + (W1 , V1 )(1,1)

+ (W 1 , V1 )(0,2) + 1/2(U 1 , V1 )(0,1) 1/2(V1 , U 1 )(0,1) + L1 U 2 ,
(B.4)
(B.5)
400
2L1 W2 2L0 U2 + (U1 , W1 )(1,0) (W1 , U1 )(1,0) + 2(W1 , W1 )(2,0)

(V1 , W1 )(2,1) + (V1 , U1 )(1,1) + 2(W 1 , W1 )(1,1) (W 1 , U1 )(0,1) = 0,
2(L 1 W 2 ) 2L 0 U 2 + (U 1 , W 1 )(0,1) (W 1 , U 1 )(0,1) + 2(W 1 , W 1 )(0,2)
(V1 , W 1 )(1,2) + (V1 , U 1 )(1,1) + 2(W1 , W 1 )(1,1) (W1 , U 1 )(0,1) = 0.
Eqs. (B.5) give the following expression for W2 and W 2 :

i j
i j
i U2 + O(h),
W 2 = 1/2 di g2 pj + i (U1 + 1/2U 1 )g1 pj + X
i

i j
ij
i
W2 = 1/2 dj
g2 pi + i (U1 + 1/2U1 )g1 pj + X i U1 + O(h).
(B.6)
(B.7)
j gives the
Let us substitute the expressions for W2 , W 2 into (B.4). The coefficient of X i X
familiar equation
i k (U2 + U 2 ) = 0,
(B.8)
j , p X j , and pi p in (B.4) at the zeroth order in h lead to the folwhile coefficients of pi X

i
j
lowing:

p dl g2lk l(U1 + U 1 )g1lk = 0,
p dl 0 g2kl l (U1 + U 1 )g1kl = 0,
i p
ji
2g1r l r lg1i k 2r g1 p g1r k g1i l lds g1s k g1r k r dj

g1
jr
np
ki
+ r g1i k dj
g1 + p g1 dn g1 = 0.
(B.9)
Eqs. (B.8), (B.9) coincide with (21) at the order t 2 . Let us look on Eqs. (B.6) which we did not
touch before. We see that at the first order in h they both lead to the following relation:
i j
i j
di dj
g2 + 3/2g1 i j (U1 + U1 )
i j
i j
+ i (U1 + U 1 )dj
g1 + j (U1 + U1 )di g1 = 0,
(B.10)
which by means of condition (B.3) is equivalent to:
i j
i j
i j

di dj
g2 + i (U1 + U1 )dj g1 + j (U1 + U1 )di g1 = 0.
(B.11)
0 , and we see that this equation coincides with (23) at the

Using the relation between ,
2
order t . 2
References
[1] V. Gourbunov, F. Malikov, V. Schechtman, math.AG/0008154;
V. Gourbunov, F. Malikov, V. Schechtman, math.AG/0005201;
V. Gourbunov, F. Malikov, V. Schechtman, math.AG/0003170;
V. Gourbunov, F. Malikov, V. Schechtman, math.AG/9906117.
[2] F. Malikov, V. Schechtman, math.AG/9901065.
[3] A. Kapustin, hep-th/0504074;
P.A. Grassi, G. Policastro, E. Scheidegger, hep-th/0702044.
[4] E. Witten, hep-th/0504078.
[5] E. Frenkel, A. Losev, hep-th/0505131.
[6] N. Berkovits, hep-th/0001035.
[7] N. Nekrasov, hep-th/0511008.

[8] E.S. Fradkin, A.A. Tseytlin, Phys. Lett. B 160 (1985) 69;
E.S. Fradkin, A.A. Tseytlin, Nucl. Phys. B 261 (1985) 1.
[9] A.S. Losev, A. Marshakov, A.M. Zeitlin, Phys. Lett. B 633 (2006) 375, hep-th/0510065.
[10] A.S. Losev, unpublished; private communication.
[11] C.G. Callan, D. Friedan, E.J. Martinec, M.J. Perry, Nucl. Phys. B 262 (1985) 593;
C.G. Callan, I.R. Klebanov, M.J. Perry, Nucl. Phys. B 278 (1986) 78.
[12] T. Banks, D. Nemeshansky, A. Sen, Nucl. Phys. B 277 (1986) 67.
[13] A.M. Polyakov, Gauge Fields and Strings, Harwood Academic Publishers, 1987.
[14] C. Hull, P. Townsend, Nucl. Phys. B 274 (1986) 349.
[15] B. Zwiebach, Nucl. Phys. B 300 (1993) 33.
[16] T. Lada, J. Stasheff, hep-th/9209099.
[17] M. Kato, K. Ogawa, Nucl. Phys. B 212 (1983) 443;
D. Friedan, E.J. Martinec, S.H. Shenker, Phys. Lett. B 160 (1985) 55;
N. Ohta, Phys. Rev. D 33 (1986) 1681;
N. Ohta, Phys. Lett. B 179 (1986) 347;
K. Furuuchi, N. Ohta, Prog. Theor. Phys. 116 (2006) 601, hep-th/0607105.
[18] J. Polchinski, String Theory, Cambridge Univ. Press, 1998.
[19] A. Alekseev, T. Strobl, hep-th/0410183.
[20] A. Sen, Nucl. Phys. B 347 (1990) 270.
[21] E. Verlinde, hep-th/9202021.
[22] A.M. Zeitlin, Nucl. Phys. B 759 (2006) 370, hep-th/0610208.
[23] N. Hitchin, math.DG/0209099.
[24] M. Gualtieri, math.DG/0401221.
[25] T.J. Courant, Dirac manifolds, Trans. Amer. Math. Soc. 319 (1990) 631661.
[26] I. Dorfman, Phys. Lett. A 125 (1986) 240.
[27] V.A. Fock, Theory of Space, Time and Gravitation, Readers Digest Young Families, 1964.
401

Notes on D-branes and dualities in (p, q) minimal

superstring theory
Hirotaka Irie
Department of Physics, Kyoto University, Kyoto 606-8502, Japan
Received 10 July 2007; accepted 19 September 2007
Available online 25 September 2007
Abstract
We study boundary states in (p, q) minimal superstring theory, combining the explicit form of matter
wave functions. Within the modular bootstrap framework, Cardy states of (p, q) minimal superconformal
field theory are completely determined in both cases of the different supercharge combinations, and the
remaining consistency checks in the super-Liouville case are also performed. Using these boundary states,
we determine the explicit form of FZZT- and ZZ-brane boundary states both in type 0A and 0B GSO
projections. Annulus amplitudes of FZZT branes are evaluated and principal FZZT branes are identified.
In particular, we found that these principal FZZT branes do not satisfy Cardys consistency conditions for
each other and play a role of order/disorder parameters of the KramersWannier duality in spacetime of this
superstring theory.
PACS: 11.25.Hf; 11.25.Pm; 11.25.Uv
Keywords: Principle branes; Spacetime non-locality
1. Introduction and summary

Noncritical (super)string theories have been investigated as a useful laboratory of critical
string theory and have been discussed in various contexts: the worldsheet description [119],
matrix models [2032] and the string field theoretical descriptions [3339]. They have fewer degrees of freedom and we can make a detail study of many important properties shared with their
E-mail address: irie@gauge.scphys.kyoto-u.ac.jp.
doi:10.1016/j.nuclphysb.2007.09.015
H. Irie / Nuclear Physics B 794 [PM] (2008) 402428
403
critical counterparts. Moreover, since these string theories are described with many different formulations, we can investigate various aspects of stringy phenomena.
Here we further study (p, q) minimal superstring theory from the conformal field theory (or
Liouville theory) approach. Minimal superstring theory is one of the most tractable superstring
theories among these noncritical superstring theories. Its worldsheet description is defined with
N = 1 super-Liouville field theory [13] coupled to N = 1 (p, q) minimal superconformal field
theory [4043] with type 0 GSO projection (see [15,29] for its basic properties). From the recent
developments in Liouville field theory without boundaries [58] and with boundaries [911], the
boundary states of super-Liouville field theory were obtained in [12,13], and in the framework
of minimal superstring theory they have been extensively studied in [15,17,32], where the disk
amplitudes of corresponding D-branes were explicitly evaluated [15] and the pure-supergravity
case, (p, q) = (2, 4), was studied including annulus amplitudes [17,32]. For further investigation
of these D-branes, however, we need to know all the Cardy states of (p, q) minimal superconformal field theory with different supercharge combinations of left and right:
r )|B; = 0
(Gr iG
( = 1),
(1.1)
and we need to combine Cardy states of each SCFT in practice.

As is known from the original works on the conformally invariant boundary states [4446],
Cardy states have a one-to-one correspondence with the highest weight representations of the
open-channel Virasoro algebra. As is briefly reviewed in Section 2.1, Cardy states have the following form in this superconformal case1 :

iNS
iR
BhNS |iNSNS ; = 1
BhNS
|hNS =
|iRR ; = 1,
iR
iNS
|hR =
BhR
iNS
|iNSNS ; = +1
iR
iNS
BhR iR |iRR ; = +1.
(1.2)
We often use a notation like jRR in this paper to indicate the sector to which an index j belongs,
and |j ; is a corresponding Ishibashi state [45]. Three kinds of these wave functions Bh i (h and
and R sectors) are known from the work of [47],2 and they are written with
i are among NS, NS
the modular matrices Sh i of this theory [48] as
1 Sh iNS
BhNS iNS = NS
,
2 S0 iNS
NS
iR
BhNS
=
iR
ShNS

,
23/4 S iR
0NS

S hNS
= iR
BiR hNS
.
S0NS iNS
(1.3)
On the other hand, the form of the remaining wave functions BhR iR is still not known. One of the
reasons why these wave functions BhR iR have not been obtained is that they do not satisfy such a
simple relation with modular matrices Sh i like (1.3). Since any corresponding characters (those
sector) always vanish due to the fermion zero-modes in the cylindrical geometry (except for
in R
the Ramond ground states), there is not such a modular matrix Sh i among these R characters.
If one considers (p, q) minimal superconformal field theory as only a single SCFT, this does
not cause any problems in studying the boundary states as in the literature [49]. It is because
1 Since open strings between opposite boundaries are in Ramond sector, we make use of the convention that Cardy
states of = 1 (+1) are labeled with the highest weight in the NS (R) sector of the open-string super-Virasoro algebra.
2 The Cardy states of Ramond ground states (denoted as |
R in this paper) are not considered in [47]. This means
that only odd models are considered. In this paper, we also derive the formula of this case.
404
we impose the spin-model GSO projection [41] ( = 1 on RR states) and the above remaining states are always projected out. On the other hand, the case of the superstring theory is not
so. Since superstring theory is a combined superconformal field theory (Liouville, matters and
ghosts), the type 0 GSO projection cannot eliminate these contributions [15,29]. So this is the
first task of this paper: We will completely determine all the wave functions of Cardy states, including the remaining wave functions, in the case of (p, q) minimal superconformal field theory
(Section 2.2).
Actually, the way to obtain these remaining wave functions is very simple. All we have to do
is to consider the OPE algebra with the simplest degenerate primaries (1, 2)+ [40] in the sense of
open (or chiral) superconformal field theory and to consider the Cardy equations obtained from
the inner products with the Ramond Cardy state |(1, 2)R+ . The plus symbol in the subscript of
(1, 2)+ means that this is a primary operator of positive chirality. Essentially we use the following
fusion rule that could be read off from the super-Coulomb gas formalism in Ramond sector [43]:
N(1,2)+ ,(k,l)+ (r,s) = (k,l+1)+ (r,s) + (k,l1) (r,s) ,
(1.4)
where operators of both positive and negative chiralities come into the relations. This kind of
fusion rule is also found in the super-Liouville cases [12] and we actually show that all of the
results obtained within the conformal bootstrap methods [12] are consistent with our procedure.
This means that all the information about boundary states can be extracted from the modular
bootstrap methods also in the case of N = 1 superconformal field theory.3
In Section 2.3, we present the complete form of the boundary states in both cases of 0B and 0A
theory, and investigate their basic properties. As in the case of super-Liouville field theory [12],
the wave functions of = 1/+1 Cardy states in (p, q) minimal superconformal field theory
are not symmetric under the transformation of (1)fR (fR is the worldsheet fermion number),
even though the superconformally invariant boundary conditions (1.1) of = 1 are exchanged
for each other. The superstring Cardy state is also not an exception but we show that one of
the branes obtained from the transformation of (1)fR comes to play a role of fundamental
degrees of freedom, principal FZZT branes [15]. The principal = 1/+1 FZZT branes of
this theory are simply related under the transformation of (1)fR , and actually we show that
they do not satisfy the Cardy equations among each other. This is reminiscent of order/disorder
parameters of the KramersWannier duality. We actually argue that they are not mutually local
in the superstring spacetime, by evaluating annulus amplitudes (Section 3).
The organization of this paper is as follows: In Section 2, we summarize the Cardy states of
this theory. After we briefly review the definition of Cardy states in Section 2.1, we show the way
to derive the remaining wave functions in Section 2.2 and the boundary states of our superstring
theory are discussed in Section 2.3. In Section 3, we evaluate various annulus amplitudes of
FZZT branes. Section 4 is devoted to conclusion and discussion.
2. Boundary states of (p, q) minimal superstring theory
2.1. Cardy states in superconformal field theory
We now recall the definition of Cardy states [44] in the SCFT case to fix our notation. This
case (but only odd models) was studied under the spin-model GSO projection [47]. The following
3 Of cause, we need to perform the conformal bootstrap method to know some relations with variables in the boundary
action (e.g., boundary cosmological constants in Liouville theory).
405
discussion does not require any restrictions, and any GSO projections are not imposed form the
beginning.
Superconformally invariant boundary states |B; are defined as
r )|B; = 0
(Ln L n )|B; = (Gr iG
( = 1; n Z, r Z + ).
(2.1)
r ) are the left- (right-)handed super-Virasoro generators that satisfy

Here Ln (L n ) and Gr (G

c 3
n n n+m,0 ,
8
[Ln , Lm ] = (n m)Ln+m +
[Ln , Gr ] = (n r)Gn+r ,

c 2 1
r
r+s,0 ,
{Gr , Gs } = 2Lr+s +
2
4
(2.2)
and = 1/2 (or = 0) when |B; is in NSNS (or RR) sector. The closed string Hilbert space
can be expanded into irreducible Verma modules of super-Virasoro algebra or superconformal
families

i (z, z ) .
Vi V i
H(c) =
(2.3)
iNSNS,RR
iNSNS,RR
When the left and right Verma modules are isomorphic, Vi

= V i , one can find the Ishibashi state
[45] in these irreducible Verma modules,

|i; N U A|i; N ,
|i; =
(2.4)
N
which satisfies (2.1). Here A is an anti-unitary operator that commutes with the above superVirasoro generators, U is an automorphism of irreducible Verma module V i ,
U L n U1 = L n ,
r (1)fR ,
r U1 = iG
U G
U = U1 ,
(2.5)
r = G
r , L n = L n ) of a dual base i; N |,4
and |i; N is a hermitian conjugation (defined by G
i; N |j ; M = i|j N,M .
(2.6)
Note that the chirality on each states is given as5

|iNSNS ; = +|iNSNS ; ,
|iRR ; = |iRR ; ,
|RR ; = +|RR ; ,

(2.7)
(1)fR |i; .
and states of different are related as |i; =

representations in these Ishibashi states are given as
|jRR ; = |j + |j + + i|j |j + ,
Also note that RR zero-mode

(2.8)
and j + |j + = j |j = 1.
4 In the case of unitary CFTs, we can take orthonormal bases. Since we have to treat nonunitary CFTs in general, we
use dual bases. Our construction of these Ishibashi states is based on Watts technique noted in [50].
5 We can also define the chirality of the RamondRamond ground state as | ; = | ; .
RR
RR
406
Annulus amplitudes between these Ishibashi states are then given with the moduli parameter
q = q = e2t as
c
i; |q 2 (L0 16 ) q 2 (L0 16 ) |j ;
= Gij tri (
)f qL0 16 .
(2.9)
Here we denote Gij i |j as the Zamolodchikov metric of superconformal primary fields

i (z, z ). It is convenient to introduce dual superconformal primary fields i G ij j , then we
can write the formal completeness relation as

1 =
(2.10)
|i, G ij j, | =
|i, i , |.
i,j
Considering this relation, closed-channel amplitudes between general boundary states |;

i |i; i ; | are expressed as

c
; |q 2 (L0 16 ) q 2 (L0 16 ) |;
=

c
|i; i ;
| tri (
)f qL0 16 .
(2.11)
On the other hand, open channel amplitudes are given as a sum over the open channel Hilbert
(o)
space H with the boundaries. Therefore they must be expanded into a sum of Virasoro characters with non-negative integer n h [44]:
c
trH(o) q L0 16 =
n h trh q L0 16 ,
(2.12)
with q = e2/t . Note that the label h runs among irreducible Virasoro primary states belonging
to NS or R sector in open channel [47]. Comparing both expressions (2.11) and (2.12), we
obtain

c
h i
n Sh tri qL0 16 ,
|i; i ; |
0=
iNSNS,RR
0=
hNS
|i; i ; |

c
n h Sh i tri (1)f qL0 16 ,
(2.13)
hR
iNSNS,RR
where {Sl i } are modular matrices of the characters under S transformation, 1/ , which
we define as,
1
1
iR
ShNS iNS iNS ( )
ShNS
iR ( ),
2
2
2
i
iNS
iR

2

hR (1/ )
ShR i i ( ) =
ShR iNS
iNS
( ),
2
hNS (1/ )
R (1/ )

i
ShNS i i ( ) =
iNS
1
1

SR i i ( ) =
ShR iNS
h , .
iNS
( )
2 R R
2 2 i
(2.14)

NS
Noting that the characters in R sector must vanish due to fermion zero modes, except for the
Ramond ground states (here we define RR ( ) = 1), we obtain the following Cardy equations:
0 = |i; i ; |
n h Sh i
407
(i NSNS, RR),
hNS
0 = |i; i ; |
n h Sh i
hR
; |
0 = |RR ; RR
(i NSNS),
n R .
(2.15)
h=R
Then, following the usual procedure (identifying n h with the fusion number N h [44] under
the Verlinde formula [51] as a non-negative integer valued matrix representation of fusion algebra
[50] and considering the trivial relation N0NS+ h = h ), we can define the Cardy states |hNS
and |hR for each Virasoro highest weight hNS and hR in open channel:
iR
1 ShNS iNS
1 ShNS

|hNS =
|iNSNS ; = 1 3/4
|iRR ; = 1,
2
iR
2 iNS S0 iNS
S0NS
iR

NS
i

Sh iNS
2 hR R
R

|hR =
|iNSNS ; = +1 3/4
|iRR ; = +1,
2
iR
i
S0NS
iR
iNS S0NS NS
|R =
i

1 SR iNS
1 hR R

|iNSNS ; = +1 1/4
|iRR ; = +1,
2
2
iR
i
S0NS
iR
iNS S0NS NS
(2.16)
with R = 1(= SR R ) for the Cardy

state of the Ramond ground state, |R . Here we denote
R
the identity operator as 0, and

S0NS iNS
S0NS iNS is formally defined as it satisfies
S0NS iNS S0NS iNS ,
iR
S0NS

iR
iR S
S0NS
0NS

.
(2.17)
The normalization of is chosen as above for later convenience.

Here we consider the spin-model GSO projection ( = 1 on RR sector) [41] for odd models. In this case, the last wave functions h i of |h can be consistently dropped since all the
sector vanish (see also the Verlinde formula of this case [52]). This gives the
characters in R
previous result [47]6 :
iR
1 ShNS iNS
1 ShNS

|iNSNS ; = 1 3/4
|iRR ; = 1,
|hNS =
2
iR
2 iNS S0 iNS
S
iR
0

NS
NS
|hR+ =

Sh iNS
R
|iNSNS ; = +1.
i
iNS S0NS NS
(2.18)
For even models, since we should be careful about the Ramond ground state |R and need to
know the remaining wave functions h i , we will discuss this case in next subsection.
6 Note that our normalization of RR Ishibashi states is differed by
2 |iRR ; = 1there .
2 from [47], that is |iRR ; = 1here =
408
2.2. The wave functions in each SCFT

In this subsection, we determine the remaining wave functions in each superconformal field
theories. As we noted in Section 1, we consider the following fusion rule:
n(1,2)+ ,(k,l)+ (r,s) = 2N(1,2)+ ,(k,l)+ (r,s) = 2(k,l+1)+ (r,s) + 2(k,l1) (r,s)
(2.19)
of the simplest degenerate primary (1, 2)+ of Virasoro algebra [40]. The factor 2 in front of Ni,j k
comes from the degeneracy of Ramond zero-mode representations.7 We first reconsider the case
of super-Liouville field theory and see that this procedure actually reproduces the results of [12].
We then discuss the case of (p, q) minimal superconformal field theory. We also summarize the
corresponding modular matrices in each SCFT.
2.2.1. The super-Liouville case
The corresponding Cardy states were found in [12,13] and further discussed in [15,28,29,32].
They are expanded in the off-shell Hilbert space [4],

e(Q/2+i)(z,z) ,
V V =
H(c) =
(2.20)
NSNS,RR >0
NSNS,RR >0
and the Cardy states for non-degenerate representations, = Q2 /8 + 2 /2, are

1
d 2 cosh(2i ) A(L)
NS ()|NSNS ; = 1
2
0

(L)
1
3/4 2 cosh(2i ) AR ()|RR ; = 1 ,
2

(L)
|R = d 2 cosh(2i ) ANS ()|NSNS ; = +1
|NS =

(L)
2
2i
sinh(2i
)
A
()|
;
=
+1
,
RR
R
23/4
and those for degenerate representations = i(nb + m/b) are

(n, m)NS =

(n, m)R =

1
m
(L)
d 4 sinh(nb) sinh
ANS ()|NSNS ; = 1
b
2
0

1
m in
in
3/4 4 sinh nb +
sinh
2
b
2
2

(L)
AR ()|RR ; = 1 ,

m
d 4 sinh(nb) sinh
A(L)
NS ()|NSNS ; = +1
b
7 See e.g., [52]. This naturally comes from the Verlinde formula of super-Virasoro algebra.
(2.21)
409

2
m in
in
cosh
4i
sinh
nb
+
2
b
2
23/4

(L)
AR ()|RR ; = +1 ,
(L)
(2.22)
(L)
where ANS () and AR () are defined as

(1 ib) (1 i/b) i/b
= 1/ S0NS NS ,
2

(1/2 ib) (1/2 i/b) i/b
(L)
R .
= 1/ S0NS
AR ()

2
(L)
ANS ()
(2.23)
The modular bootstrap method, except for the R wave function , was studied in [12,13]. The
corresponding modular matrices are actually obtained as
NS
= S(L)
NS
= S(L)
R
NS
= 2 cosh(2i ),

NS

m
(L)
(L)
= S(n,m)R NS = 4 sinh(nb) sinh
,
S(n,m)NS
b

R
in
m in
(L)
sinh
S(n,m) = 4 sinh nb +
NS
2
b
2
S(L)
NS
(2.24)
form the characters:

(NS) ( ) = q
2
2
(NS)
( ),
(NS) ( ) = q
2
2

(NS)
( ),
(R) ( ) = q
2
2
(R)
0 ( )
(2.25)
for non-degenerate representations, and

(NS)
(NS)
(NS)
( ) i
( ),
2 (nb+m/b)
2 (nbm/b)

(NS)

(NS)
(NS)
( ) (1)mn i
( ),
(nb+m/b)
2
2 (nbm/b)
(R)
(R)
(R)
( ) i
( )
2 (nb+m/b)
2 (nbm/b)
(n,m) ( ) i
(n,m) ( ) i
(n,m) ( ) i
(2.26)
for degenerate representations. See Appendix A for our definition of basic modular functions
(like 0(NS) ( )) and its modular transformation.
The remaining wave functions were then obtained from the conformal bootstrap method
[12]. Here we show that they can also be obtained from the modular bootstrap method by considering the simplest degenerate primary operator (1, 2)+ and its fusion rule Ni,j k with operators
in R+ sector. Actually, the fusion rule between (1, 2)+ and non-degenerate primary R+ is controllable [6,7] and leads to
n(1,2)+ ,(n,m)+ h = 2N(1,2)+ ,(n,m)+ h = 2(n,m+1)+ h + 2(n,m1) h ,
n(1,2)+ ,R+ h = 2N(1,2)+ ,R+ h = 2( +ib/2)R+ h + 2( ib/2)R h .
(2.27)
Note that the chirality of the second terms is flipped due to the boundary Liouville action (see e.g.,
[12,13]) whose fermion number is odd, and that the factor 2 follows from the wave functions
of NSNS sector. The corresponding Cardy equations are
410

(1, 2)+ |RR ; = +1 RR ; = +1|(1, 2)+

1
R
R
= +S(1,3)NS
S(1,1)NS

2
1 (4i cosh(b) sinh(2/b)) (4i cosh(b) sinh(2/b))
=
,
4 cosh(b) cosh(/b)
2

(n, m)+ |RR ; = +1 RR ; = +1|(1, 2)+

1
R
R
= +S(n,m+1)NS
S(n,m1)NS

2
ni
1 (4i sinh(nb + ni
2 ) sinh(m/b 2 )) (4i cosh(b) sinh(2/b))
=
,
4 cosh(b) cosh(/b)
2

R+ |RR ; = +1 RR ; = +1|(1, 2)+

1
R
R
= +S( +ib/2)NS
S( ib/2)NS

2
1 (2i sinh(2i )) (4i cosh(b) sinh(2/b))
,
=
(2.28)
4 cosh(b) cosh(/b)
2
and solving these equation we can obtain

R R = 2i sinh(2i ),

in
m in
R
cosh
,
(n,m)R = 4i sinh nb +
2
b
2
(2.29)
and this correctly reproduces the results of [12].

2.2.2. The (p, q) minimal superconformal field theory case
Next we apply the above argument to the case of (p, q) minimal superconformal field theory. This theory can be classified into two categories: even and odd models [53]. The modular
and R sector were derived in [48] from the character formula [54],
matrices among NS, NS
(NS)
(NS)
(r,s) ( ) = 0
( )
(2npq+qrps)2
8pq
(2npq+qr+ps)2
8pq
nZ

(NS)

(NS)
(r,s) ( ) = 0
( )
(2npq+qr+ps)2
(2npq+qrps)2
8pq
8pq
,
(1)2pq q
(1)rs q
nZ
(R)
(R)
(r,s) ( ) = 0 ( )
(2npq+qrps)2
8pq
(2npq+qr+ps)2
8pq
nZ
2
(2npq+pq)2
1 (R)
(2npq)
(R)
q 8pq q 8pq
(only even model),
( p , q ) ( ) = 0 ( )
2
2 2
(2.30)
nZ
as

(rs)(r s )
4
r r
s s
S(r,s) (r ,s ) = (1) 2
sin
(q p) sin
(q p) .
pq
2p
2q
(2.31)
For the R wave function (r,s) (r ,s ) , we should know the fusion rule of open strings. One way
to know is the super-Coulomb gas formalism in Ramond sector [43]. Since the chirality of the
411
screening charges is odd, The fusion rule is also given in the following form:
n(1,2)+ ,(k,l)+ (r,s) = 2N(1,2)+ ,(k,l)+ (r,s) = 2(k,l+1)+ (r,s) + 2(k,l1) (r,s) .
This leads to the Cardy equations,

(1, 2)+ |(r, s)RR ; = +1 (r, s)RR ; = +1|(1, 2)+

1
(r,s)R
(r,s)R
= +S(1,3)NS
S(1,1)NS

2
4
r
2s
2
1 ( pq sin( 2p (q p)) sin( 2q (q p)))
=
,
r
s
(q p)) sin( 2q
(q p))
2 4pq sin( 2p

(k, l)+ |(r, s)RR ; = +1 (r, s)RR ; = +1|(1, 2)+

1
(r,s)R
(r,s)R
= +S(k,l+1)NS
S(k,l1)NS

2

(kl1)(rs)
kr
ls
4
1
2
(q p) sin
(q p)
= (1)
sin
pq
2p
2q
2

r
2s
4
(q p) sin
(q p)
sin
pq
2p
2q

1
4
r
s
sin
.
(q p) sin
(q p)
pq
2p
2q
(2.32)
(2.33)
Solving these equations, we obtain

(kl)(rs)+1
4
kr
ls
2
sin
(q p) sin
(q p) .
(k,l) (r,s) = (1)
pq
2p
2q
(2.34)
From this formula, we can see that any (k, l) = (p/2, q/2) Cardy states have no contribution
from the closed string (r, s) = (p/2, q/2) state. From this consideration, we also conclude
R iR = R iR = SR iR .
(2.35)
For an exercise of the later discussion, we also consider the spin-model GSO projections
[41] for even models. Here we first assume |RR ; = |RR ; . We then do not have to
reconsider the NS Cardy states |hNS and we obtain
iR
1 ShNS iNS
1 ShNS

|iNSNS ; = 1 3/4
|iRR ; = 1,
|hNS =
2
iR
2 iNS S0 iNS
S
iR
0NS

NS
|hR+ =

Sh iNS
R
|iNSNS ; = +1,
i
iNS S0NS NS
|R =

1
1 SR iNS
1

|iNSNS ; = +1 1/4
|RR ; = +1.
2
2
iR
i
S0NS
iNS S0NS NS
(2.36)
For the case of |RR ; = +|RR ; , the NS Cardy states |hNS and |R are no longer
Cardy states under the GSO projection. We should instead consider the following Cardy
states:
412
ShNS iNS
1

|hNS |hNS+ + |hNS =
|iNSNS ; = 1,
i
2
iNS S0 NS
NS

1
1 ShR iNS

|iNSNS ; = +1,
|hR |hR =
2
2 iNS S0 iNS
NS

1
1 SR iNS

|iNSNS ; = +1.
|R |R+ + |R =
2
2 iNS S0 iNS
(2.37)
NS
Note that the corresponding fusion rule in this case is that of generalized Verlinde formula
[52] (the fusion rule of super-Virasoro algebra) and the states propagating in the open channel are among super-Virasoro Verma module
(not only among the Virasoro sub-Verma module).
From this point of view, the factor 1/ 2 in the definition of the boundary states is necessary.
It will be seen that the former case is like the type 0B GSO projection and the later case is
like the type 0A GSO projection in minimal superstring theory. Although we can also consider other possibility (e.g., boundary states with ), we will not investigate this direction
here.
2.3. Boundary states in (p, q) minimal superstring theory
We now combine the above Cardy states. Since the boundary states in (p, q) minimal superstring theory have been discussed in [15,29,32], we first of all summarize the important things
given in the previous discussions.
1. Since the ghost Ishibashi states in RR sector are in (1/2, 3/2) picture8 and (1)f =
L
M
L
M
(1)fL +fR = (1)f +f +1 , we use the following type 0 GSO projection: (1)f +f = +1
(1) for type 0B (type 0A) [15,29].
M
2. Since (1)f |RR ; = +|RR ; , the closed-string contributions of Ramond ground
states are denied in the boundary states of = +1 ( = 1) branes in the type 0B (type 0A)
cases [15,29].
3. When we consider the case of negative < 0, we should use the transformation
and (L) (L) [15,32]. That is, the boundary states of < 0 are obtained with the following
replacement in the wave functions of RR emissions:

i
R
=
2
cosh(2i
)
2
cosh
2i
+
,
S(L)

NS
2

i
R
=
2i
sinh(2i
)
2i
sinh
2i
+
,
(L)
(2.38)
R
2
where = (1 sgn())/2. The corresponding boundary cosmological constants are chosen as
follows:

|| cosh(b ) ( = 1),
=
(2.39)
|| sinh(b ) ( = +1),
with the parameter = sgn() [15].
8 For a summary of ghost Ishibashi/Cardy states, see Appendix B.
413
Considering this, the boundary states for type 0B theory are given as9 :
Type 0B

, (k, l)NS ; = 1

(M) (r,s)

1
2 cosh(2i ) S(k,l)
d
ANS , (r, s) , (r, s)NSNS ; = 1
=
2
(r,s)NSNS
0

1
i
(M) (r,s)
d
AR , (r, s)
S(k,l)
2 cosh 2i +
2
2
(r,s)RR
0

, (r, s)RR ; = 1 (k + l 2Z),
(2.40)

, (k, l)R ; = +1

(M) (r,s)

1
2 cosh(2i ) S(k,l)
d
ANS , (r, s) , (r, s)NSNS ; = +1
=
2
(r,s)NSNS
0

1
i
(M) (r,s)
d
AR , (r, s)
(k,l)
2i sinh 2i +
2
2
(r,s)RR
0

, (r, s)RR ; = +1 (k + l 2Z + 1).
(2.41)
Here (r, s) NSNS, RR means that (r, s) runs among

NSNS:
1 r p 1,
1 s q 1,
qr ps > 0,
r + s 2Z,
RR:
1 r p 1,
1 s q 1,
qr ps 0,
r + s 2Z + 1.
(2.42)
(M) (r, s) (X = NS or R) and

We define AX (, (r, s)) A(L)
X () A

(M) (r,s)
A(M) (r, s) 1/ S(1,1)
(M) (r,s)
A(M) (r, s) A(M) (r, s) = 1/S(1,1)
The Ishibashi states are

, (r, s)XX ; = |XX ; (r, s)XX ; |GhXX ; .
(2.43)
(2.44)
The normalization of ghost Ishibashi/Cardy states is summarized in Appendix B. The boundary states of ZZ branes are obtained by replacing the Liouville wave functions, for example,

NS
NS
(L)
S(L)
(2.45)
= 2 cosh(2i ) S(n,m)NS
= 4 sinh(nb) sinh(m/b) .
NS
The normalization of = +1 boundary states is the same as that of = 1 boundary states.10 It
(L+M)
(z) form a doublet under the combination
is due to the fact that open-channel spin fields
(L)
(M)
of each Liouville and matter spin-field doublet, (z) and (z).11 The case of matter
Ramond ground states is also the same (the doublet is supplied from Liouville spin fields).
9 Note that the Ramond Cardy state is given as |, (p/2, q/2) = |, (p/2, q/2)
R
R .
10 This should be compared with (2.16), where the normalization of = +1 branes is 2 times bigger than that of
= 1 branes.
11 From this fact, the relation n k = N k holds in the superstring case.
ij
ij
414
From this expression, we can identify the corresponding principal FZZT branes [15]. For
= 1 FZZT branes, we can show the relations modulo BRST,

(kl)(mn)
; (k, l)NS ; = 1 =
+ i mb nb1 ; = 1, (1)
2
(2.46)
,

2
m,n
with | ; = 1, | ; (1, 1)NS ; = 1 following the arguments given in [15,16]. For
= +1 FZZT branes, on the other hand, there is not such a principal brane among the above
= +1 FZZT branes. Instead, we actually show that the following = +1 FZZT brane,
| ; = +1, (1)fR | ; = 1, ,
plays a role of a principal = +1 FZZT brane12 :

(kl)(mn)
+ i mb nb1 ; = +1, (1)
; (k, l)R ; = +1 =
2
.

2
m,n
(2.47)
(2.48)
Because this principal = +1 FZZT brane is not a Cardy state for the principal = 1 FZZT
branes, these principal branes cannot exist at the same time. Even though there is no open string
spectrum that propagates among these two branes, they are necessary for the construction of
all the spectrum of D-branes. Since these are simply related under the simple transformation
(1)fR , it is reminiscent of order/disorder parameters in the KramersWannier duality [55].13
Actually we argue in Section 3 that the corresponding annulus amplitudes are not mutually local
in spacetime.
The boundary states for type 0A theory are
Type 0A

, (k, l)NS ; = 1

=
(M) (r,s)

2 cosh(2i ) S(k,l)
ANS , (r, s) , (r, s); NSNS; = 1
(r,s)NSNS
(k + l 2Z),

, (k, l)R ; = +1

=
(2.49)
(M) (r,s)

2 cosh(2i ) S(k,l)
ANS , (r, s) , (r, s); NSNS; = +1
(r,s)NSNS
(k + l 2Z + 1),

; (p,
q)
R ; = +1
1
=
2

d
0
(2.50)
(M) (r,s)

2 cosh(2i ) S(k,l)
ANS , (r, s) , (r, s); NSNS; = +1
(r,s)NSNS
12 Note that, in the Liouville wave functions of RR sector, sinh turns to be cosh (cosh turns to be sinh if one
considers < 0) in this principal = +1 FZZT branes.
13 Note that this transformation is not the same as that of (or (L) (L) ).
415

i
q)
, (p,
q);
RR; = +1 .
d 2i sinh 2i +
AR , (p,
2
(2.51)
The first two kinds of boundary states are given
as14

, (k, l)X (0A) = 1 , (k, l)X (0B) + , (k, l)X (0B) .
+ >0
>0
>0
2
(2.52)
The last boundary state of the Ramond ground state is defined by the following fusion rule:
1
1
n( p , q )+ ,( p , q ) (r,s) = N( p , q )+ ,( p , q ) (r,s) =
2 2
2 2
2
2
2
2
2
2
(n,m) (r,s) ,
n=1,2,...,p1; m=1,3,...,q1
(2.53)
with the identification (r, s) (p r, q s). Under this normalization, we actually obtain the
following nontrivial identification in the pure-supergravity case, (p, q) = (2, 4):

, (1, 1)NS ; = 1 (0B) , (1, 2)R ; = +1 (0A) ,
>0
<0

(0B)
(0A)
, (1, 2)R = +1
, (1, 1)NS ; = 1
,
>0
<0
argued in [15,29,32].
Although the following branes are the principal FZZT branes of type 0A theory:

| ; = +1 (1)fR | ; = 1,
| ; = 1 , (1, 1)NS ; = 1 ,
(2.54)
(2.55)
the Cardy state of the Ramond ground state, | ; (p,

q);
= +1, cannot be written with the
above principal FZZT branes. So we must treat these things separately.
3. Annulus amplitudes of the principal FZZT branes
In this section, we evaluate the annulus amplitudes of FZZT branes. Annulus amplitudes of
ZZ branes are not considered here, since this kind of amplitude is obtained from those of the
principal FZZT branes (with the relation between ZZ and FZZT branes [14] in this case [15]).
Annulus amplitudes from the CFT approach has been studied in [14,16] (bosonic cases) and
[17,32] (fermionic case). Our technical procedure follows them, and we will not repeatedly write
such a thing. For the later convenience, we denote our amplitudes as

,

Z
, (k, l);
, (k
, l
)

1

c
1
c
dt , (k, l); = 1, q 2 (L0 24 ) q 2 (L0 24 )
, (k
, l
); = 1,

,

,

ZNSNS , (k, l);
, (k
, l
) +
ZRR , (k, l);
, (k
, l
) ,
(3.1)
14 This normalization gives the natural oscillator algebra, [ [0] , [0] ] = n

n+m , in the corresponding string field forn
m
mulation [38].
416
with q = e2t . Throughout this analysis, we use the following relations that can be shown with
the technique of [56]:

m2
2 2pq
km
lm
q sin
sin
pq
p
q
mZ
2
(2npq+qr+ps)2

(M) (r,s) (2npq+qrps)
8pq
8pq
S
(q
q
)
(r,s)NSNS
nZ
(k,l)
(k
2Z),
=
(2npq+qrps)2
(2npq+qr+ps)2

(M) (r,s)

8pq
8pq
S(k,l)
(1)npq (q
(1)rs q
)
(r,s)NSNS
nZ
(k l 2Z + 1),

m2
2
km
k
k
lm
2pq
q sin
+ sin
+
pq
p
2
q
2
mZ+pq/2
(M) (r,s)
S(k,l)
(2npq+qrps)2
8pq
(2npq+qr+ps)2
8pq
(k l 2Z),
(3.2)
(r,s)RR nZ
and the following useful formula that was used in [14]:

cos(Am) cosh((1 A))
=
,
sinh()
m2 + 2
(3.3)
mZ

mZ+1/2
cos(Am) sinh((1 A))

=
.
cosh()
m2 + 2
(3.4)
3.1. NSNS exchange amplitudes between FZZT branes

,
We now consider the general amplitudes of ZNSNS (, (k, l);

, (k
, l
)), so NSNS exchanges.
This kind of amplitudes does not essentially depend on and sgn(). The relation between 0A
and 0B is the following:

(0A),

(0B),

ZNSNS , (k, l);
, (k
, l
) = 2ZNSNS , (k, l);
, (k
, l
)

(k, l), (r, s) = (p/2, q/2) ,

(0A),+1
(0B),+1
ZNSNS
, (k, l);
, (p/2, q/2) = ZNSNS
, (k, l);
, (p/2, q/2)

(k, l) = (p/2, q/2) ,

(0A)+1,+1
ZNSNS
, (p/2, q/2);
, (p/2, q/2)

1 (0B)+1,+1
, (p/2, q/2);
, (p/2, q/2) .
= ZNSNS
2
(0B),
(3.5)
So we only consider 0B theory, ZNSNS (, (k, l);

, (k
, l
)). Then 0B amplitudes we consider
here are (i) NSNS amplitudes of
= +1 and (ii) NSNS amplitudes of
= 1. They can be
expressed as follows:
417
(0B),
(i)

, (k, l);
, (k
, l
)
ZNSNS
1
=
2
cosh(2i ) cosh(2i
)
d
sinh(b) sinh(/b)
0

dt
2
2
(2npq+qrps)2
8pq
(M) (r,s) (M) (r,s)

S(k
,l
)
(M) (r,s)
S(1,1)
S(k,l)
(r,s)NSNS
(2npq+qr+ps)2
8pq
(3.6)
nZ
(ii)
0

(0B),
ZNSNS
, (k, l);
, (k
, l
)

cosh(2i ) cosh(2i
)
d
sinh(b) sinh(/b)
1
=
2
0

dt
(M) (r,s) (M) (r,s)

S(k
,l
)
(M) (r,s)
S(1,1)
S(k,l)
(r,s)NSNS
2

(2npq+qr+ps)2
2 (2npq+qrps)
8pq
8pq
.
(1)npq q 2 q
(1)rs q
(3.7)
nZ
Although they seem to be different forms, they come to be a unified form. We can actually
,
reexpress the basic amplitudes ZNSNS (, (k, l);
, (1, 1)) by using (3.2) and (3.3) as follows:

,1
ZNSNS , (k, l);
, (1, 1)

km
lm
pq sin( p ) sin( q )
cosh(2i ) cosh(2i
)
= d
sinh(b) sinh(/b)
pq 2 + m2
mZ
cosh(2i pq ) cosh(2i pq )
l
sinh(( pk
p )) sinh( q )
sinh(/p) sinh(/q)
sinh()
(3.8)
By using the fusion relations for r s 2Z:

(M) (r,s)
S(k,l)
(M) (r,s)
S(k
,l
)
(M) (r,s)
S(1,1)
(M) (r,s)
m=k+k
1,k+k
3,...,kk
+1;
n=l+l
1,l+l
3,...,ll
+1
S(n,m)
(3.9)
we obtain the following general formula

,

ZNSNS , (k, l);
, (k
, l
)

d cosh(2i pq ) cosh(2i pq )
=
2
(sinh(/p) sinh(/q))2
l
k
l
sinh(( pk
p )) sinh( q ) sinh( p ) sinh( q )
.
sinh()
Following the arguments of [16], we rewrite this amplitude as
=
2

d
cosh 2i
cosh 2i
pq
pq
(3.10)
418
sinh( pk ) sinh( ql ) sinh( kp ) sinh( lq )
(sinh(/p) sinh(/q))2

cosh() cosh(k/p)
.
sinh()
sinh(k/p)

(3.11)
The first term in the parenthesis is the main part of this amplitude, since the second term is
actually a contribution from the unphysical poles of NSNS sector, = inp (n Z),15 as is in
the bosonic case [16]. The main part can be written with the amplitudes of the principal = 1
FZZT branes following the rule of (2.46) and (2.48):

,
ZNSNS,main , (k, l);

, (k
, l
)
(3.12)

i
i
,
=
ZNSNS,main + (mb n/b);
+ (m
b n
/b) ,
2
2
m,n;m ,n
where the above basic amplitudes satisfy

,
ZNSNS,main ( ;
) = ZNSNS,main ( ;
) = ZNSNS,main ( ;
).
,
(3.13)
The amplitudes of
= +1 principal FZZT branes can be evaluated with the procedure of
[16] as
,
ZNSNS ( ;
) =
sinh( +
) sinh(
)
1
pq
pq
ln
+
2
sinh(p
) sinh(p
)
pq
pq
cosh(2 ) cosh(2
)
1
pq
pq
.
= ln
p
p
cosh(2 pq ) cosh(2 pq )
(3.14)
At that time, we should be careful to treat the boundary cosmological constants [12] and the
uniformization parameter z [15]. We now define these parameters as

|| cosh(p
) ( = 1),
,
(3.15)
z = cosh ,
=
|| sinh(p
) ( = +1),
p q
where we use the label (p,
q)
that is more suitable from the point of view of two-matrix models
(or 2-component KP hierarchy) [38]:
(p,
q)
= (p/2, q/2) (even model),
= (p, q) (odd model).
Therefore we obtain the following form:
1 ln( 2zz
2 ) (even model),
2
,

ZNSNS ( ; ) = 1
ln( z22 z
22 ) (odd model).
2

(3.16)
(3.17)
15 It can be easily seen by recalling the correspondence with differential operator P2n = ( Lp )2n of 2-component KP
hierarchy [38]. From the viewpoints of string field formulation [3335,38], this contribution comes from the normal
ordering with respect to SL(2, C) invariant vacuum of [33].
Note that the dependence of is changed as

1
(( + 2 )1/p + ( 2 )1/p ) z( )
2

z= 1
1/p + ( 2 + )1/p ) z ( )
2
2 (( + + )
( = 1),
( = +1).
419
(3.18)
2
2
That is, of = 1 FZZT branes are simply related as =1
=+1
= .
The principal FZZT brane amplitudes of = 1 are somewhat more complicated but we
can say that the main parts are given as

12 ln(z z
)
(even model),
,
ZNSNS,main ( ; ) =
(3.19)
12 ln(z2 z
2 ) (odd model)
from the modulo-BRST equations (2.48). This means that if we have both brane operators (z)
of = 1 principal branes, they should have the following behavior16 :
=+1 (z
)=1 (z) (z
z)1/2 ,
(3.20)
since there are no RR exchange between these branes. In this sense, the principal = 1/+1
branes are not mutually local in z (or ) spacetime coordinate.17 That is, their square-root cut
cannot dissolve in the asymptotic weak coupling region, z (or ). Note that this is
due to the breaking of the Cardy consistency conditions. Actually the Cardy states of = +1
FZZT branes do not have such a behavior, because they are linear combinations of the principal
= +1 FZZT branes, the number of which is even, and then the square root dissolves in the
weak coupling region, z (or ). So we conclude that the principal = 1 FZZT
branes in minimal superstring theory can be interpreted as order/disorder parameters in superstring spacetime.
3.2. RR exchange amplitudes between FZZT branes
Next we consider RR exchange amplitudes. Note that annulus amplitudes between = 1
,
and = +1 branes must vanish, ZRR = 0. It is because the superconformal residual symmetry
is remained in cylinder and we need to insert a vertex operator to obtain non-zero results. Thus
we now neglect this contribution. The amplitudes we consider are (iii) RR-ground state exchange
amplitudes in 0A theory, (iv) RR amplitudes of = 1. First the case of (iii) is written as

p q
p q
+1,+1
,
;
,
,
,
(iii) ZRR
2 2
2 2

sinh(2i + 2 i) sinh(2i
+ 2 i)
1
=
d
2
cosh(b) cosh(/b)
0
2 (2npq)2
(2npq+pq)2
pq

dt
q 2 q 8pq q 8pq
4
0
nZ
16 Although we do not know the form of unphysical parts, these contributions are expected to be nothing but the normal
ordering with respect to SL(2, C) invariant vacuum of [33].

17 For the relations between a complex coordinate and the 2-dimensional spacetime coordinate ( , X ) of minimal
FF
L
string theory, see [35] and [30].
420
1
=
2

1
d sinh(2i pq + 2 i) sinh(2i pq + 2 i)
,
4 cosh(/p) cosh(/q)
sinh(2/pq)
(3.21)
and the remaining case of (iv) is

1,1
, (k, l);
, (k
, l
)
ZRR
1
=
2
cosh(2i + 2 i) cosh(2i
+ 2 i)
d
cosh(b) cosh(/b)
dt
2
2
(2npq+qrps)2
8pq
(2npq+qr+ps)2
8pq
(M) (r,s) (M)

S(k
,l
)
(M) (r,s)
S(1,1)
(r,s)
S(k,l)
(r,s)RR

,
(3.22)
nZ

+1,+1
ZRR
, (k, l);
, (k
, l
)
1
=
2

d
(i sinh(2i + 2 i))(i sinh(2i

+ 2 i))
cosh(b) cosh(/b)
0
(M) (r,s)
(k,l)
(M) (r,s)
S(1,1)
(r,s)RR
dt
(M) (r,s)
(k
,l
)
2
2
(2npq+qrps)2
8pq
(2npq+qr+ps)2
8pq

.
(3.23)
nZ
1,1
The basic amplitudes ZRR
(, (k, l);
, (1, 1)) (k + l 2Z) are

1,1
ZRR
, (k, l);
, (1, 1)

=
cosh(2i + i
2 ) cosh(2i +
cosh(b) cosh(/b)
i
2 )
pq

mZ+pq/2
sin( km
p +
k
lm
2 ) sin( q
pq 2 + m2
k
2 )
i

i
cosh(2i

pq + 2 ) cosh(2i pq + 2 )
1 d
2
cosh(/p) cosh(/q)
lk
k
l
l
(1) 2 sinh(( pk
p )+ 2 i) sinh( q 2 i)

(even model),
sinh()
=

i
+ i ) cosh(2i
cosh(2i
+
)

2
2
pq
pq
12 d
cosh(/p) cosh(/q)
lk
k
l
l
(1) 2 cosh(( pk
p )+ 2 i) sinh( q 2 i)
(odd model).
cosh()
(3.24)
421
By using the fusion relations for r s 2Z + 1,18

(M) (r,s)
S(k,l)
(M) (r,s)
S(k
,l
)
(M) (r,s)
S(1,1)
(M)
(k,l)
(r,s)
(M) (r,s)
S(1,1)
nm
2
m=k+k
1,k+k
3,...,kk
+1;
n=l+l
1,l+l
3,...,ll
+1
(M) (r,s)
(k
,l
)
(1)
(1)
nm
2
m=k+k
1,k+k
3,...,kk
+1;
n=l+l
1,l+l
3,...,ll
+1
(M) (r,s)
S(n,m)
(M) (r,s)
S(n,m)
we obtain the formula for the = 1 case,

1,1
ZRR
, (k, l);
, (k
, l
)
i

i
cosh(2i

pq + 2 ) cosh(2i pq + 2 )
1 d
2
(cosh(/p) cosh(/q))
k
l
l
k
sinh(( pk
p )+ 2 i) sinh( q 2 i) sinh( p 2 i) sinh( q 2 i)
sinh()
(even model),
=
i

i
cosh(2i

pq + 2 ) cosh(2i pq + 2 )
+ 1 d
(cosh(/p) cosh(/q))2
pk
k
l
l
k
cosh((
)+
p
2 i) sinh( q 2 i) sinh( p 2 i) sinh( q 2 i)

cosh()
(odd model),
for the case of = +1,

+1,+1
ZRR
, (k, l);
, (k
, l
)
i

i
i sinh(2i

pq + 2 )i sinh(2i pq + 2 )
1 d
(cosh(/p) cosh(/q))2
pk
k
l
l
k
sinh((
)+
p
2 i) sinh( q 2 i) sinh( p 2 i) sinh( q 2 i)
sinh()
(even model),
=
i

i
i sinh(2i

pq + 2 )i sinh(2i pq + 2 )
1 d
(cosh(/p) cosh(/q))
k
l
l
k
cosh(( pk
p )+ 2 i) sinh( q 2 i) sinh( p 2 i) sinh( q 2 i)
cosh()
(odd model).
(3.25)
(3.26)
(3.27)
Also in this case, we can separate the amplitudes into the sum of unphysical parts, = i(2n +
1)p,
and main parts. The main parts can be written with that of principal FZZT branes as

,
ZRR,main , (k, l);
, (k
, l
)

[(kl)(nm)] [(k
l
)(n
m
)]
+
2
2
=
(1)
n,m;n
,m

i
i
,
ZRR,main + (nb m/b);
+ (n
b m
/b) ,
2
2
18 Notice that (1)
[(kl)(k
l
)](rs)
+1
2
= (1)
[(k+k
)(l+l
)](rs)
2
(3.28)
422
and as is noted in Section 2.2, (2.47), one can find that

1,1
+1,+1
ZRR
( ;
) = ZRR
( ;
).
(3.29)
The amplitudes of principal FZZT branes are

(even model)
1,1
ZRR
( ;
)>0

1
1
cosh(p
) + cosh(p

)
= ln cosh( ) cosh( ) + ln
2
2
cosh(p
) cosh(p

)

1
1
+
= ln(z z
) + ln
,
2
2

(3.30)
1,1
( ;
)<0
ZRR

1
1
sinh(p
) + sinh(p

)
+
= ln cosh( ) cosh(
) ln
ln
sinh
2
2
sinh(p
) sinh(p

)
2

1
1
+
1
= ln(z z
) ln
(3.31)
ln zz
+ z2 1 z
2 1 1 ,
2
2

(odd model)
1,1
ZRR
( ;
)>0

1
1
cosh( ) cosh(
)
cosh(p
) + cosh(p

)
= ln
+ ln
2
cosh( ) + cosh(
)
2
cosh(p
) cosh(p

)

1
zz
+
1
= ln
+ ln
,
2
z+z
2

1,1
ZRR
( ;
)<0

1
1
sinh( ) sinh(
)
sinh(p
) + sinh(p

)
= ln
+
ln
2
sinh( ) + sinh(
)
2
sinh(p
) sinh(p

)

2

1
1
z 1 z
2 1
+
= ln
+ ln
.
2
2

z2 1 + z
2 1
(3.32)
(3.33)
We then summarize the full amplitudes (NSNS + RR) of the principal = 1 FZZT branes.
For 0B theory,
(even model)

1 + 1 2
z1 z2
1 1 2
1,1
Z1 ,2 (1 ; 2 )>0 =
(3.34)
ln
ln(1 + 2 ),
2
1 2
2
1 + 1 2
1 1 2
z1 z2
(1 ; 2 )<0 =
)
ln(
ln(1 + 2 )
Z1,1
1 ,2
2
1 2
2

1 2
(3.35)
ln z1 z2 z12 1 z22 1 1 ,
2
(odd model)

1 + 1 2
z1 z2
z1 + z2
1 1 2
1,1
Z1 ,2 (1 ; 2 )>0 =
(3.36)
ln
ln
+
,
2
1 2
2
1 + 2
423
Z1,1
(1 ; 2 )<0
1 ,2

z2 1 z2 1
z2 1 + z2 1
1 + 1 2
1 1 2
1
2
1
2
=
ln
ln
+
,
2
1 2
2
1 + 2
(3.37)
and for 0A theory,
ln(
(1 ; 2 ) =
Z1,1
1 ,2
zz
)
2
2
z2 z
2
ln( 2
2 )
(even model),
(odd model),
are obtained.
Also for the pure-supergravity case of (p, q) = (2, 4) (note that z and accidentally coincide,
z = = cosh ), we can actually show that

1 1 2
1,1
1 , (1, 1); 2 , (1, 1) >0 =
ln(1 + 2 ),
ZNSNS
2

1,1
ZNSNS
1 , (1, 1); 2 , (1, 1) <0

1 + 1 2 2
= ln cosh
2

1 2
= ln
1 || 22 || + 1 2 1 2 || .
2
(3.38)
Of course, this is the previous results of this case [17].

4. Conclusion and discussion
In this paper, we investigate the explicit form of boundary states in (p, q) minimal superstring
theory. For this purpose, we actually show the way to obtain all the wave functions of =
Cardy states within the modular bootstrap methods in N = 1 superconformal field theory. We
then identify the corresponding principal = 1 FZZT branes following the arguments given
in [15] and explicitly evaluate these annulus amplitudes. The principal = 1/+1 FZZT branes
of 0B theory are interpreted as order/disorder parameters of the KramersWannier duality.
From the analysis of [15], it was realized that among many different FZZT branes only a few
numbers of principal FZZT branes are important and they correspond to the fundamental degrees
of freedom of the theory. Since we can extract all the closed-string degrees of freedom from the
principal = 1 FZZT and its anti-FZZT branes [38], it is natural to think of the principal
= 1 FZZT brane as independent degrees of freedom.
In the case of Ising model, however, we can clearly see the relation with the fermion system
by introducing the disorder parameter. In this sense, it is interesting if we could find some more
general structures of minimal superstring theory, by considering how to describe the principal
= +1 FZZT branes in the exact nonperturbative formulations.
Of course this kind of duality is very familiar in conformal field theory, so worldsheet descriptions. A new thing about the duality in this case is that order/disorder parameters in minimal
superstring theory correspond to D-branes in spacetime (not worldsheet observables). Since this
structure is originated from the basic nature of the NSR formalism, it is interesting to investigate
what is the spacetime properties of NSR superstring theory.
424
Acknowledgements
The author thanks M. Fukuma for useful discussions and encouragements, and K. Okuyama,
K. Murakami, T. Takayanagi and M. Sato for useful conversations. The author also thanks
H. Hata for various supports in preparing this paper. This work was supported in part by the
Grant-in-Aid for the 21st Century COE Center for Diversity and Universality in Physics from
the Ministry of Education, Culture, Sports, Science and Technology (MEXT) of Japan. The author is also supported by Research Fellowships of the Japan Society for the Promotion of Science
(JSPS) for Young Scientists (No. 182672).
Appendix A. Summary of the basic modular functions
Here we summarize our convention of the basic characters:

3 ( )
1
q1/48
(NS)
n1/2
0 ( )
1+q
=
,
( )
( ) ( )
n=1

1/48

4 ( )
1
q

(NS)
1 qn1/2 =
,
0 ( )
( )
( ) ( )
n=1

1/24

2 2 ( )
q
(R)
1 + qn =
0 ( )
( )
( ) ( )
(A.1)
n=0
n
2i . ( ) is the corresponding
with Dedekind -function ( ) q1/24
a
n=1 (1 q ) and q = e
theta function. The modular transformations are

(1/ ) =
( ),
3 (1/ ) =
3 ( ),
i
i

4 (1/ ) =
(A.2)
2 (1/ ) =
2 ( ),
4 ( ).
i
i
Appendix B. The Ishibashi/Cardy states of superconformal ghost

It is useful to note about our convention and notation of the Ishibashi/Cardy states of superconformal ghost [57]. Here we especially consider the normalization of Cardy states. It
is convenient for ghost to be written in the form of [45]. We construct them with Watts
technique noted in [50] as

|q ; =
|q; N U A| 2 q; N
N
= U e
r<0 (r r +r r )
|q | 2 q
(B.1)
where q is a corresponding picture and |Ghq ; is defined with the proper Ishibashi state of bc
ghost, |bc, as |Ghq ; |q ; |bc. We use the following hermitian conjugation,
r = r ,
r = r ,
and we define the automorphism U to satisfy
(B.2)
425
U r U1 = ir ,
U r U1 = ir ,
U = (1)fR U1 ,
U = U (1)fR .
(B.3)
The normalization of states is defined from the open string character summed over the picture q
Hilbert space H(q) as follows:
1 ; |T (q)|1 ;
=+
q1/24
n1/2 )2
n1 (1 + q
=+

q 1/24
fR L0 c/24
,
=
tr
q
(1) (1)
H
n1/2 )2
n1 (1 + q
3/2 ; |T (q)|1/2 ;
=
q1/12
n 2
n1 (1 + q )

q 1/24
= trH(1) q L0 c/24 ,
n1/2
2
)
n1 (1 q
1 ; |T (q)(1)fR |1 ;
q1/24
n1/2 )2
n1 (1 q
=
=

q 1/12
= i trH(1/2) (1)fR q L0 c/24 ,
n
2
)
n1 (1 + q
1
(B.4)
where T (q) = q 2 (L0 c/24) q 2 (L0 c/24) The normalization of the second equation follows that
of the first equation.19 Note that the normalization (or the sign) of the third equation is required
from the spacetime statistics in superstring theory (open strings in R sector are fermions) and that
this negative sign of the character can be consistently obtained from the definition |Ghq ; =
(1)fR |Ghq ; of the closed-channel Ishibashi states.
References
[1] A.M. Polyakov, Quantum geometry of bosonic strings, Phys. Lett. B 103 (1981) 207;
A.M. Polyakov, Quantum geometry of fermionic strings, Phys. Lett. B 103 (1981) 211.
[2] V.G. Knizhnik, A.M. Polyakov, A.B. Zamolodchikov, Fractal structure of 2d-quantum gravity, Mod. Phys. Lett.
A 3 (1988) 819;
F. David, Conformal field theories coupled to 2-D gravity in the conformal gauge, Mod. Phys. Lett. A 3 (1988)
1651;
J. Distler, H. Kawai, Conformal field theory and 2-D quantum gravity, or whos afraid of Joseph Liouville?, Nucl.
Phys. B 321 (1989) 509.
[3] J. Distler, Z. Hlousek, H. Kawai, Super Liouville theory as a two-dimensional, superconformal supergravity theory,
Int. J. Mod. Phys. A 5 (1990) 391.
[4] N. Seiberg, Notes on quantum Liouville theory and quantum gravity, Prog. Theor. Phys. Suppl. 102 (1990) 319.
19 If we assume that the character in NS sector is expanded as q 1/24 + , then the character in NS
sector must be
1/24
+ .
expanded as q
426
[5] H. Dorn, H.J. Otto, Two and three point functions in Liouville theory, Nucl. Phys. B 429 (1994) 375, hep-th/
9403141.
[6] A.B. Zamolodchikov, Al.B. Zamolodchikov, Structure constants and conformal bootstrap in Liouville field theory,
Nucl. Phys. B 477 (1996) 577, hep-th/9506136.
[7] J. Teschner, On the Liouville three point function, Phys. Lett. B 363 (1995) 65, hep-th/9507109.
[8] R.H. Poghosian, Structure constants in the N = 1 super-Liouville field theory, Nucl. Phys. B 496 (1997) 451, hepth/9607120;
R.C. Rashkov, M. Stanishkov, Three-point correlation functions in N = 1 super Liouville theory, Phys. Lett. B 380
(1996) 49, hep-th/9602148.
[9] V. Fateev, A.B. Zamolodchikov, Al.B. Zamolodchikov, Boundary Liouville field theory. I: Boundary state and
boundary two-point function, hep-th/0001012;
J. Teschner, Remarks on Liouville theory with boundary, hep-th/0009138.
[10] A.B. Zamolodchikov, Al.B. Zamolodchikov, Liouville field theory on a pseudosphere, hep-th/0101152.
[11] B. Ponsot, J. Teschner, Boundary Liouville field theory: Boundary three point function, Nucl. Phys. B 622 (2002)
309, hep-th/0110244.
[12] T. Fukuda, K. Hosomichi, Super Liouville theory with boundary, Nucl. Phys. B 635 (2002) 215, hep-th/0202032.
[13] C. Ahn, C. Rim, M. Stanishkov, Exact one-point function of N = 1 super-Liouville theory with boundary, Nucl.
Phys. B 636 (2002) 497, hep-th/0202043.
[14] E.J. Martinec, The annular report on non-critical string theory, hep-th/0305148.
[15] N. Seiberg, D. Shih, Branes, rings and matrix models in minimal (super)string theory, JHEP 0402 (2004) 021,
hep-th/0312170.
[16] D. Kutasov, K. Okuyama, J.w. Park, N. Seiberg, D. Shih, Annulus amplitudes and ZZ branes in minimal string
theory, JHEP 0408 (2004) 026, hep-th/0406030.
[17] K. Okuyama, Annulus amplitudes in the minimal superstring, JHEP 0504 (2005) 002, hep-th/0503082.
[18] A. Basu, E.J. Martinec, Boundary ground ring in minimal string theory, Phys. Rev. D 72 (2005) 106007, hepth/0509142.
[19] A. Belavin, A. Zamolodchikov, Higher equations of motion in N = 1 SUSY Liouville field theory, JETP Lett. 84
(2006) 418, hep-th/0610316;
A. Belavin, V. Belavin, A. Neveu, A. Zamolodchikov, Bootstrap in supersymmetric Liouville field theory. I: NS
sector, hep-th/0703084;
V.A. Belavin, On the N = 1 super Liouville four-point functions, arXiv: 0705.1983 [hep-th].
[20] E. Brezin, V.A. Kazakov, Exactly solvable field theories of closed strings, Phys. Lett. B 236 (1990) 144;
M.R. Douglas, S.H. Shenker, Strings in less than one dimension, Nucl. Phys. B 335 (1990) 635;
D.J. Gross, A.A. Migdal, A nonperturbative treatment of two-dimensional quantum gravity, Nucl. Phys. B 340
(1990) 333.
[21] I.K. Kostov, Multiloop correlators for closed strings with discrete target space, Phys. Lett. B 266 (1991) 42;
I.K. Kostov, Strings with discrete target space, Nucl. Phys. B 376 (1992) 539, hep-th/9112059.
[22] G.W. Moore, N. Seiberg, M. Staudacher, From loops to states in 2-D quantum gravity, Nucl. Phys. B 362 (1991)
665.
[23] G.W. Moore, N. Seiberg, From loops to fields in 2-D quantum gravity, Int. J. Mod. Phys. A 7 (1992) 2601.
[24] J.M. Daul, V.A. Kazakov, I.K. Kostov, Rational theories of 2-D gravity from the two matrix model, Nucl. Phys.
B 409 (1993) 311, hep-th/9303093.
[25] D.J. Gross, E. Witten, Possible third order phase transition in the large N lattice gauge theory, Phys. Rev. D 21
(1980) 446;
V. Periwal, D. Shevitz, Unitary matrix models as exactly solvable string theories, Phys. Rev. Lett. 64 (1990) 1326;
V. Periwal, D. Shevitz, Exactly solvable unitary matrix models: Multicritical potentials and correlations, Nucl. Phys.
B 344 (1990) 731;
C.R. Nappi, Mod. Phys. Lett. A 5 (1990) 2773;
C. Crnkovic, M.R. Douglas, G.W. Moore, Int. J. Mod. Phys. A 7 (1992) 7693, hep-th/9108014;
T.J. Hollowood, L. Miramontes, A. Pasquinucci, C. Nappi, Nucl. Phys. B 373 (1992) 247, hep-th/9109046;
W. Ogura, Prog. Theor. Phys. 89 (1993) 1311, hep-th/9201018;
R.C. Brower, N. Deo, S. Jain, C.I. Tan, Nucl. Phys. B 405 (1993) 166, hep-th/9212127.
[26] S. Dalley, C.V. Johnson, T.R. Morris, Multicritical complex matrix models and nonperturbative 2-D quantum gravity, Nucl. Phys. B 368 (1992) 625;
S. Dalley, C.V. Johnson, T.R. Morris, A. Watterstam, Unitary matrix models and 2-D quantum gravity, Mod. Phys.
Lett. A 7 (1992) 2753, hep-th/9206060;
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
427
R. Lafrance, R.C. Myers, Flows for rectangular matrix models, Mod. Phys. Lett. A 9 (1994) 101, hep-th/9308113;
P. Di Francesco, Rectangular matrix models and combinatorics of colored graphs, Nucl. Phys. B 648 (2003) 461,
cond-mat/0208037.
T. Takayanagi, N. Toumbas, A matrix model dual of type 0B string theory in two dimensions, JHEP 0307 (2003)
064, hep-th/0307083.
M.R. Douglas, I.R. Klebanov, D. Kutasov, J. Maldacena, E. Martinec, N. Seiberg, A new hat for the c = 1 matrix
model, hep-th/0307195.
I.R. Klebanov, J. Maldacena, N. Seiberg, Unitary and complex matrix models as 1-d type 0 strings, Commun. Math.
Phys. 252 (2004) 275, hep-th/0309168.
J.M. Maldacena, G.W. Moore, N. Seiberg, D. Shih, Exact vs. semiclassical target space of the minimal string,
JHEP 0410 (2004) 020, hep-th/0408039.
D. Gaiotto, L. Rastelli, T. Takayanagi, Minimal superstrings and loop gas models, JHEP 0505 (2005) 029, hep-th/
0410121.
N. Seiberg, D. Shih, Flux vacua and branes of the minimal superstring, JHEP 0501 (2005) 055, hep-th/0412315.
M. Fukuma, S. Yahikozawa, Nonperturbative effects in noncritical strings with soliton backgrounds, Phys. Lett.
B 396 (1997) 97, hep-th/9609210.
M. Fukuma, S. Yahikozawa, Combinatorics of solitons in noncritical string theory, Phys. Lett. B 393 (1997) 316,
hep-th/9610199.
M. Fukuma, S. Yahikozawa, Comments on D-instantons in c < 1 strings, Phys. Lett. B 460 (1999) 71, hep-th/
9902169.
M. Fukuma, H. Irie, S. Seki, Comments on the D-instanton calculus in (p, p + 1) minimal string theory, Nucl. Phys.
B 728 (2005) 67, hep-th/0505253.
M. Fukuma, H. Irie, Y. Matsuo, Notes on the algebraic curves in (p, q) minimal string theory, JHEP 0609 (2006)
075, hep-th/0602274.
M. Fukuma, H. Irie, A string field theoretical description of (p, q) minimal superstrings, JHEP 0701 (2007) 037,
hep-th/0611045.
M. Fukuma, H. Irie, Supermatrix models and multi ZZ-brane partition functions in minimal superstring theories,
JHEP 0703 (2007) 101, hep-th/0701031.
A.A. Belavin, A.M. Polyakov, A.B. Zamolodchikov, Infinite conformal symmetry in two-dimensional quantum
field theory, Nucl. Phys. B 241 (1984) 333.
D. Friedan, E.J. Martinec, S.H. Shenker, Conformal invariance, supersymmetry and string theory, Nucl. Phys. B 271
(1986) 93.
D. Friedan, Z.a. Qiu, S.H. Shenker, Conformal invariance, unitarity and two-dimensional critical exponents, Phys.
Rev. Lett. 52 (1984) 1575;
H. Eichenherr, Minimal operator algebras in superconformal quantum field theory, Phys. Lett. B 151 (1985) 26;
M.A. Bershadsky, V.G. Knizhnik, M.G. Teitelman, Superconformal symmetry in two-dimensions, Phys. Lett. B 151
(1985) 31;
D. Friedan, Z.a. Qiu, S.H. Shenker, Superconformal invariance in two-dimensions and the tricritical Ising model,
Phys. Lett. B 151 (1985) 37.
G. Mussardo, G. Sotkov, M. Stanishkov, Ramond sector of the supersymmetric minimal models, Phys. Lett. B 195
(1987) 397;
G. Mussardo, G. Sotkov, H. Stanishkov, Fine structure of the supersymmetric operator product expansion algebras,
Nucl. Phys. B 305 (1988) 69.
J.L. Cardy, Conformal invariance and surface critical behavior, Nucl. Phys. B 240 (1984) 514.
N. Ishibashi, Mod. Phys. Lett. A 4 (1989) 251;
T. Onogi, N. Ishibashi, Mod. Phys. Lett. A 4 (1989) 161;
T. Onogi, N. Ishibashi, Mod. Phys. Lett. A 4 (1989) 885, Erratum.
J.L. Cardy, Boundary conditions, fusion rules and the Verlinde formula, Nucl. Phys. B 324 (1989) 581.
R.I. Nepomechie, Consistent superconformal boundary states, J. Phys. A 34 (2001) 6509, hep-th/0102010.
Y. Matsuo, S. Yahikozawa, Superconformal field theory with modular invariance on a torus, Phys. Lett. B 178 (1986)
211;
D. Kastor, Modular invariance in superconformal models, Nucl. Phys. B 280 (1987) 304.
G.Z. Toth, N = 1 supersymmetric boundary bootstrap, Nucl. Phys. B 676 (2004) 497, hep-th/0308146;
Z. Bajnok, L. Palla, G. Takacs, Spectrum of boundary states in N = 1 SUSY sine-Gordon theory, Nucl. Phys. B 644
(2002) 509, hep-th/0207099;
R.I. Nepomechie, The boundary supersymmetric sine-Gordon model revisited, Phys. Lett. B 509 (2001) 183, hepth/0103029;
428
[50]
[51]
[52]
[53]
[54]
[55]
[56]
[57]
C. Ahn, R.I. Nepomechie, J. Suzuki, Finite size effects in the spin-1 XXZ and supersymmetric sine-Gordon models
with Dirichlet boundary conditions, hep-th/0611136;
R.I. Nepomechie, C. Ahn, TBA boundary flows in the tricritical Ising field theory, Nucl. Phys. B 647 (2002) 433,
hep-th/0207012;
G. Feverati, P.A. Pearce, F. Ravanini, Exact (1, 3) boundary flows in the tricritical Ising model, Nucl. Phys. B 675
(2003) 469, hep-th/0308075;
C.r. Ahn, R.I. Nepomechie, The scaling supersymmetric YangLee model with boundary, Nucl. Phys. B 594 (2001)
660, hep-th/0009250;
M. Kormos, Boundary renormalization group flows of unitary superconformal minimal models, Nucl. Phys. B 744
(2006) 358, hep-th/0512085;
P. Dorey, I. Runkel, R. Tateo, G. Watts, g-function flow in perturbed boundary conformal field theories, Nucl. Phys.
B 578 (2000) 85, hep-th/9909216;
M. Kormos, Boundary renormalization group flows of the supersymmetric LeeYang model and its extensions,
Nucl. Phys. B 772 (2007) 227, hep-th/0701061;
S. Fredenhagen, D. Wellig, A common limit of super Liouville theory and minimal models, arXiv: 0706.1650 [hepth].
R.E. Behrend, P.A. Pearce, V.B. Petkova, J.B. Zuber, Boundary conditions in rational conformal field theories, Nucl.
Phys. B 570 (2000) 525;
R.E. Behrend, P.A. Pearce, V.B. Petkova, J.B. Zuber, Nucl. Phys. B 579 (2000) 707, hep-th/9908036.
E.P. Verlinde, Fusion rules and modular transformations in 2d conformal field theory, Nucl. Phys. B 300 (1988)
360.
W. Eholzer, R. Hubel, Fusion algebras of fermionic rational conformal field theories via a generalized Verlinde
formula, Nucl. Phys. B 414 (1994) 348, hep-th/9307031.
P. Di Francesco, H. Saleur, J.B. Zuber, Generalized Coulomb gas formalism for two-dimensional critical models
based on SU(2) coset construction, Nucl. Phys. B 300 (1988) 393.
P. Goddard, A. Kent, D.I. Olive, Unitary representations of the Virasoro and super Virasoro algebras, Commun.
Math. Phys. 103 (1986) 105.
H.A. Kramers, G.H. Wannier, Statistics of the two-dimensional ferromagnet. Part 1, Phys. Rev. 60 (1941) 252.
J.L. Cardy, Operator content of two-dimensional conformally invariant theories, Nucl. Phys. B 270 (1986) 186;
A. Cappelli, C. Itzykson, J.B. Zuber, Modular invariant partition functions in two-dimensions, Nucl. Phys. B 280
(1987) 445.
C.G. Callan, C. Lovelace, C.R. Nappi, S.A. Yost, String loop corrections to beta functions, Nucl. Phys. B 288 (1987)
525;
C.G. Callan, C. Lovelace, C.R. Nappi, S.A. Yost, Adding holes and crosscaps to the superstring, Nucl. Phys. B 293
(1987) 83;
C.G. Callan, C. Lovelace, C.R. Nappi, S.A. Yost, Loop corrections to conformal invariance for type 1 superstrings,
Phys. Lett. B 206 (1988) 41;
J. Polchinski, Y. Cai, Consistency of open superstring theories, Nucl. Phys. B 296 (1988) 91;
C.G. Callan, C. Lovelace, C.R. Nappi, S.A. Yost, Loop corrections to superstring equations of motion, Nucl. Phys.
B 308 (1988) 221.

On the multi-spin magnon and spike

solutions from membranes
P. Bozhilov a, , R.C. Rashkov b,1
a Institute for Nuclear Research and Nuclear Energy, Bulgarian Academy of Sciences, 1784 Sofia, Bulgaria
b Institute for Theoretical Physics, Vienna University of Technology, Wiedner Hauptstr. 8-10, 1040 Vienna, Austria
Received 10 August 2007; accepted 1 October 2007

Abstract
Recently important classes of solitonic string solutions were obtainedgiant magnons and single spikes.
In previous study we showed the existence of giant magnon-like membrane solutions and studied their
properties. In this paper we investigate classical rotating membranes representing analog of a specific class
of string spiky solutions. Using the reduction to the NeumannRosochatius integrable system we find analog
of the string single spike solutions. In contrast to the magnon-like solutions, this case is characterized with
finite difference of energy and winding number and finite spins as well.
PACS: 11.25.Mj
Keywords: M-theory; AdSCFT correspondence; Spin chains
1. Introduction
Recent advances indicate that integrable structures play an important role in relating string/Mtheory and the gauge theories. One of the conjectures of the correspondence is the direct relation
between the spectrum of anomalous dimensions of gauge invariant operators in the gauge theory
and the energy spectrum of the string/M-theory. Therefore, one of the important issues is the
dispersion relations in the string/M-theory. The techniques of integrable systems have therefore
become useful in studying this correspondence in details.
E-mail addresses: bozhilov@inrne.bas.bg (P. Bozhilov), rash@hep.itp.tuwien.ac.at (R.C. Rashkov).

1 On leave from Department of Physics, Sofia University, Bulgaria.
doi:10.1016/j.nuclphysb.2007.10.004
430
P. Bozhilov, R.C. Rashkov / Nuclear Physics B 794 [PM] (2008) 429441
The strong/weak nature of the duality between string/M-theory and gauge theory makes the
proof of the conjectured correspondence difficult. Basically, when the perturbative study is valid
on one side of the correspondence, one cannot trust the result from the other, strong coupled
theory. There exist however special regimes where one can do reliable calculations on both side
of the correspondence. Assuming integrability, in the strong-coupling limit the S matrix can be
interpreted as describing the two-body scattering of elementary excitations on the worldvolume.
When their worldvolume momenta become large, these excitations can be described as special
types of solitonic solutions, or giant magnons, and the interpolating region is described by the
dynamics of the so-called near-flat-space regime [1,2]. On the gauge theory side, the action of
the dilatation operator on single-trace operators is the same as that of a Hamiltonian acting on
the states of a certain spin chain [3]. This turns out to be of great advantage because one can
diagonalize the matrix of anomalous dimensions by using the algebraic Bethe ansatz technique
(see [4] for a nice review on the algebraic Bethe ansatz). In this picture the dynamics involves
diffractionless scattering encoded in the S matrix. Proving that the gauge and string/M theories
are identical in the planar limit therefore amounts to showing that the underlying physics of both
theories is governed by the same two-body scattering matrix.
This approach was developed mainly for checking string/gauge theory duality. It is still not
quite well understood in the case of M-theory/gauge theory correspondence due to the complexity
of the latter. It is worth however to study particular solutions of the above type for at least three
reasons. First, obtaining solutions of stringy type for membranes is interesting on its own and is
expected to shed light on some properties of the theory. Secondly, one can gain some insights
about how the correspondence should relate the spectrum to the gauge theory operators and
identify the latter, and finally it should be possible to reproduce the string result upon dimensional
reduction from eleven to ten dimensions.
Recently Hofman and Maldacena [1] were able to map spin chain magnon excitations to
specific rotating semiclassical string states on R S 2 . The relation between energy and angular
momentum for the one-spin giant magnon found in [1] is:

p
EJ =
(1.1)
sin
,
2
where p is the magnon momentum which on the string side is interpreted as a difference in
the angle (see [1] for details). These classical string solutions were further generalized to
include magnon bound states [58], as well as dynamics on the whole S 5 [9,10] and in fact a
method to construct classical string solutions describing superposition of arbitrary scattering and
bound states was found [11]. The semiclassical quantization of the giant magnon solution was
performed in [12].2
A class of classical string solutions with spikes may be of interest for the AdS/CFT correspondence as well. These spiky strings were constructed in the AdS5 subspace of AdS5 S 5 [26]
and it was argued that they correspond to single trace operators with a large number of derivatives. The spiky strings were generalized to include dynamics in the S 5 part of AdS5 S 5 [27]
and in [28] closed strings with kinks were considered. Some generalizations can be found also
in [29,30].
Leaded by the recent developments in string theory, we would like to study the existence of
so-called single spike solutions [3133] in the membrane case. As it is already known, one can
2 See also [1325].
431
find a consistent reduction of the membrane to the NeumannRosochatius integrable system [34]
and therefore one can expect that such solutions can be obtained in this case as well.
The paper is organized as follows. In the next section we give short review of the single spike
sector in the string case. To analyze the problem we use reduction to NeumannRosochatius
(NR) integrable system and reproduce some results from [3133]. Next we proceed with the
analysis of the membrane case. We use the same technique, namely reduction to the NR system,
to obtain the corresponding dispersion relations. We summarize our study in Section 4.
2. Spiky solutionsstring case
To obtain single spike solutions we will use the reduction to the NR system. This approach
is convenient because it will be used in the membrane case and then the comparison and dimensional reduction become transparent. We start with the Polyakov sigma model action supplied
with the corresponding Virasoro constraints.
2.1. The Lagrangian
To write the Lagrangian of the theory we use the following parametrization
xa ( ) = ra ( )eia ( ) ,
|xa |2 = ra 2 + ra2 a2 ,
= + .
(2.1)
Using parametrization in terms of complex variables as above one can find the Lagrangian

2

2
a
2
2
2 2
2 2

2 2
2
r
ra + ra a 2
L=
2
2 a a
a

+
(2.2)
ra2 1 .
a
Integrating once the equations of motion for a , we get

1
Ca

a = 2
+ a ,
2 ra2
(2.3)
where Ca are constants of motion. On other hand, for ra we get

2
2 ra
1
Ca2
2
+ 2
2 ra ra = 0,
2
3
ra
2 a
where we used the equations for (2.3).

The equations for ra can be then obtained from the effective Lagrangian

1
C2
2
2 2
2
r
r
1
.
2 2 ra 2 2 a 2 2 2
+
L=
a
ra
2 a a
a
a
(2.4)
(2.5)
These few steps were just to make link to string case where the corresponding solutions are
already known. One can easily find the corresponding Hamiltonian

1
Ca2
2
2
2 2
2 2
H=
(2.6)
+
r .
ra + 2
2 ra2 2 2 a a
a
432
2.2. Conformal constraints in NR

An important part of the analysis are Virasoro constraints

| Xa |2 + | Xa |2 = 2 ,
(2.7)

[ Xa X a + X a Xa ] = 0,
(2.8)
where we used
Xa = xa ( )eia ,
= + , xa ( + 2) = xa ( ).
(2.9)
One can use the ansatz (2.1) and the solutions (2.3) to find constraints in terms of the parameters.
Namely, one can write the final form of the Virasoro constraints as
2 + 2 2
,
H= 2
2

a Ca + 2 = 0.
(2.10)
(2.11)
The periodicity conditions on ra and a should be imposed and they are as follows
ra ( + 2) = ra ( ),
a ( + 2) = a ( ) + 2na ,
na integer.
(2.12)
2.3. Two-spin spiky solutions

What we have to do is to satisfy the constraints (2.10), (2.11) in a way that we get single spike
solutions. The two-spin case means that we reduce the system to the S 3 subspace, or r3 = 3 = 0
(also C3 = 0). In addition, the condition that one of the turning points of the string is at =
/2 imposes C2 = 0, = 1 C2 1 (we set = 1 and then || becomes a group velocity). The
equation for has two solutions: = 1 corresponding to a magnon type solution, and = C1
corresponding to single spike solution.
The single spike choice for and other parameters is3
= C1 ,
1 C1
1
= .
C1
2
We have to solve the following equation
2
du

(1 u) u u,
=u =2
2
d
1
(2.13)
(2.14)
where
C2
u = 1 , 2 = 12 22 ,
2
2
(C 2 12 ) du
(1 ) du
du
=
.
d = =
1
u
2 2 (1 u) u u
2C12 2 (1 u) u u
u = r12 = sin2 ,
3 For details see [3133].
(2.15)
The conserved quantities are

d
C1
1
1 =
d,
+
u
(1 2 )
(1 2 )

E = T d,

1
C1
J1 =
d
+
T
u d,
(1 2 )
(1 2 )

2
T
(1 u) d,
J2 =
(1 2 )
433
(2.16)
where 1 is the angular extension of the string

1 = 1 d.
Let us consider the following difference
2C1 T
arccos u =
,
E T 1 =
2
(2.17)
where
= 0 .
(2.18)
2
Unlike in the case of giant magnons, this difference is finite (and the energy is still large). This
means that the string has one spike, but is wounded many times along the equator of S 5 so that
the difference between the energy and winding number T 1 is finite.
For the spin J1 we get
2T 1
2T 1
cos 0 =
sin
J1 =
2

2
and analogously for J2 we find
(2.19)
2T 2
2T 2
cos 0 =
sin .
J2 =
2
2
Defining sin as in [31, Eq. (6.23)] (where it is actually wrong because 1 > 2 )
sin =
2
,
1
C1
sin 0 =
2
(2.20)
(2.21)
we one can write Ji as

J1 = 2T
1
sin ,
sin
J2 = 2T
sin
sin .
cos
(2.22)
Note that both Ji are finite.

Eliminating the auxiliary parameter one can obtain the following dispersion relations
,
E T 1 =
(2.23)
J1 = J22 + 2 sin2 .
(2.24)
434
We see that in the limit J2 = 0 we reproduce the one-spin case. One can define the anomalous
dimensions in this case (J2 = 0 and J1 = J ) as in the magnon considerations but subtracting the
winding number T 1 to get renormalized result
( sin ),
= (E T 1 ) J =
(2.25)
which reproduces the result of [3133].

3. Membranes on AdS4 S 7
Turning to the membrane case, let us first write down the gauge fixed membrane action and
constraints in diagonal worldvolume gauge, we are going to work with:

0 2
1
G
+
T
2
T
det
G
C
SM = d 3 LM = d 3
(3.1)
00
2
ij
2 012 ,
40
2

G00 + 20 T2 det Gij = 0,
(3.2)
G0i = 0.
(3.3)
They coincide with the frequently used gauge fixed Polyakov type action and constraints after
the identification 20 T2 = L = const, where 0 is Lagrange multiplier and T2 is the membrane
tension. In (3.1)(3.3), the fields induced on the membrane worldvolume Gmn and C012 are given
by
Gmn = gMN m X M n X N ,
C012 = cMNP 0 X M 1 X N 2 X P ,
m = / m , m = (0, i) = (0, 1, 2),

0 1 2
, , = (, 1 , 2 ), M = (0, 1, . . . , 10),
(3.4)
where gMN and cMNP are the components of the target space metric and 3-form gauge field
respectively.
Searching for membrane solutions in AdS4 S 7 analogous to string ones on AdS5 S 5 , we
should first eliminate the membrane interaction with the background 3-form field on AdS4 . To
make our choice, let us write down the background. It can be parameterized as follows

ds 2 = (2lp R)2 cosh2 dt 2 + d 2 + sinh2 d 2 + sin2 d 2 + 4 d72 ,
c(3) = (2lp R)3 sinh3 sin dt d d.
Since we want the membrane to have nonzero conserved energy and spin on AdS, the choice for
which the interaction with the c(3) field disappears is4 :
= 0 = const.
The metric of the corresponding subspace of AdS4 is

2
dssub
= (2lp R)2 cosh2 dt 2 + d 2 + sinh2 sin2 0 d 2

= (2lp R)2 cosh2 dt 2 + d 2 + sinh2 d( sin 0 )2 .
(3.5)
4 Of course, we can fix the angle instead of . We choose to fix because is one of the isometry coordinates in
the initial AdS4 space.
435
Hence, the appropriate membrane embedding into (3.5) and S 7 , analogous to the string embedding in AdS5 S 5 , is
m

Z = 2lp Rr m ei , = (0, 1), = (0 , 1 ) = (t, sin 0 ),
r r + 1 = 0, = (1, 1),

m
Wa = 4lp Rra m eia ( ) , a = (1, 2, 3, 4), ab ra rb 1 = 0.
(3.6)
For this embedding, the induced metric is given by

Gmn = (m Z n) Z + ab (m Wa n) W b
1

2
= (2lp R)
m r n r + r2 m n
,=0
+4
4

m ra n ra + ra2 m a n a
(3.7)
a=1
We will use the expression (3.7) for Gmn in (3.1), (3.2) and (3.3). Correspondingly, the membrane
Lagrangian becomes

L = LM + A r r + 1 + S (ab ra rb 1),
(3.8)
where A and S are Lagrange multipliers.
In this paper, we are interested in the following particular case of the membrane embedding (3.6)
Z0 = 2lp Rei ,
Z1 = 0,
(3.9)
which implies
r0 = 1,
r1 = 0,
0 = t = .
For this ansatz, the metric induced on the membrane worldvolume simplifies to
4

2
2
0 0
2
m ra n ra + ra m a n a m n (/2) ,
Gmn = (4lp R)
(3.10)
a=1
and the membrane Lagrangian is given by

4

2
ra 1 .
L = LM + S
(3.11)
a=1
The known most general membrane embedding in AdS4 S 7 leading to the Neumann
Rosochatius integrable system is [34]
for
Z0 = 2lp Rei ,
Z1 = 0,
= 1 + ,
= 2 + ,
Wa = 4lp Rra (, )ei[a +ga (,)] ,

, , , = const,
(3.12)
436
r1 = r1 ( ),
r2 = r2 ( ),
r3 = r3 () = a sin(b + c),
g1 = g1 ( ),
3 = 4 = ,
r4 = r4 () = a cos(b + c),
g2 = g2 ( ),
a, b, c, g3 , g4 = const,
a < 1,
= 0.
The above ansatz reduces the membrane Lagrangian L in (3.11) to

2
2

2
(2l
R)
p
LM
A 2 ( ra )2
=
0
a=1

2

2
a
A 2
+ A 2 ra2 ga
a2 ra2
A 2 2
A 2 2

2

2
2
2
2
ra 1 a ,
+ (/2) (a) + S
(3.13)
(3.14)
a=1
where

2
A 2 80 T2 lp Rab .
As shown in [34], LM
corresponds to the following NeumannRosochatius type Lagrangian
LM
NR

2

(2lp R)2 2
Ca2
A 2
2
2
2 2
a ra
A ( ra )
0
(A 2 2 )ra2 A 2 2
a=1

2

2
2
+ S
ra 1 a , Ca = const.
a=1
For the present case, the constraint G02 = 0 in (3.3) is satisfied identically, due to = 0. The
other two constraints in (3.2) and (3.3) can be written in the form
H
2

A 2 2 ( ra )2 +
a=1
=
2
Ca2
(A 2 2 )ra2
A 2
A 2 2

A 2 + 2
(/2)2 (a)2 ,
2
2
a Ca + (/2)2 (a)2 = 0,

a2 ra2
(3.15)
(3.16)
a=1
where ra must satisfy the condition

2

ra2 1 a 2 = 0.
(3.17)
a=1
Parameterizing the circle (3.17) by

1/2
1/2

cos ,
r2 = 1 a 2
sin ,
r1 = 1 a 2
one obtains that (3.17) is satisfied identically and (3.15) reduces to (prime is used for d/d )
2
1
(A 2
2 )2 (1 a 2 )
437
2

A + 2 (/2)2 (a)2

2

C22
C12
1
2
2
2
2
2
A 1 a 1 cos + 2 sin .
1 a 2 cos2 sin2
(3.18)
3.1. One-spin solutions

If we set C1 = 1 = 0, from (3.16) and (3.18) we get

2

A
sin2 0 sin2 sin2 sin2 1 ,
=
(A 2 2 ) sin
(3.19)
where
sin2 0 =
(/2)2 (a)2
,
22 (1 a 2 )
sin2 1 =
2
sin2 0 .
A 2
Eq. (3.19) gives the same spiky and giant magnon solutions as (5.11) of [31].
3.2. Two-spin solutions
Following [33], for obtaining two-spin, spiky and giant magnon solutions, we set C1 = 0
in (3.18) and taking into account (3.16), we receive

(1 a 2 )12
A 2 (22 12 )
(/2)2 (a)2
2
2
sin2
1+
=
(A 2 2 )2 sin2 (1 a 2 )(22 12 )
A 2 (/2)2 (a)2

2 [(/2)2 (a)2 ]2
4
sin .
(3.20)
A 2 (1 a 2 )2 22 (22 12 )
The solutions we are searching for, correspond to two particular choices of the parameters in the
above expression for 2 .
The first choice is to set in (3.20)

(/2)2 (a)2 = 1 a 2 22 .
This results in

A 22 12 cos
=
sin2 sin2 0 ,
A 2 2 sin
(3.21)
where
sin2 0 =
2 22
A 2 (22 12 )
Comparing (3.21) with (3.23) in [33], we see that this equation for leads to giant magnon
solution on S 3 .
The second appropriate choice is
(/2)2 (a)2 =
A 2 (1 a 2 )22
.
2
438
Then, (3.20) reduces to

A 22 12 cos
=
sin2 sin2 1 ,
A 2 2 sin
(3.22)
where
sin2 1 =
A 2 22
2 (22 12 )
Comparing (3.22) with (3.28) in [33], we see that this equation for leads to single spike solution
on S 3 , first obtained in [31] from the NambuGoto string action.
Let us finally note that both Eqs. (3.21) and (3.22) are of the type

A
22 12
du
K=
, u = sin2 ,
= u = 2K(1 u) u u,
d
A 2 2
corresponding to different values of the parameter u:
u = sin2 0 ,
or u = sin2 1 .
3.3. Conserved quantities

The energy E and the angular momenta Ja can be computed by using the equalities

L
L
2
E= d
,
Ja = d 2
, a = 1, 2, 3, 4.
(0 t)
(0 a )
For the embedding (3.12), (3.13), they are given by

(2lp R)2
d,
E=
0

(4lp R)2
d Ca + A 2 a ra2 , a = 1, 2.
Ja =
0
2
2
(A )
In order to reproduce the string case, we set = 0, which leads to J3 = J4 = 0.
On the solution for (u), E and Ja take the form
(2lp R)2
E=
0 K
1
u
du
,
(1 u) u u
(4lp R)2 A 2 (1 a 2 )1
J1 =
A 2 2 )
0 K(
J2 =
(4lp R)2
A 2 2 )
0 K(

A 2 1 a 2 2

1
u
du
,
u u

2
A 2 1 a 2 2
(/2)2
2
1
u
du
.
u u
1
u
du
(1 u) u u
(3.23)
439
From (3.23), for the giant magnon case, one obtains the energy-charge relation

2

2
J2
J1
E
=
+ 27 T2 (lp R)3 a 1 a 2 b cos2 0 .
2 1 a2
2 1 a2
The origin of the denominators 2 1 a 2 in the above equality can be explained as follows. The
7
5
factor of 2 is due to the different radii of AdS
4 and S , while the radii of AdS5 and S in type IIB
5
2
string background AdS5 S are equal. 1 a arises
because the NR system obtained from
membrane is restricted to lie on a circle with radius 1 a 2 , whereas in the string case this
radius equals one.
For the single spike solution using (3.23) again, one arrives at

2

J2 = J12 + 28 T2 (lp R)3 a 1 a 2 b cos2 1 .
(3.24)
Now following [10], we introduce the angular extension of the membrane as

1
C2

+ 2 d
g 2 = g2 d =
r22
A 2 2
and compute the difference E T g 2 . It turns out it is finite for

T = 26 T2 (lp R)3 a 1 a 2 b
and is given by
E T g 2 = 2T
1 .
2
(3.25)
Comparing (3.24), (3.25) with (2.24), (2.23), we see that the single spike dispersion relations
obtained from membrane and from string are of the same type.
4. Conclusions
It is well known that particular solitonic solutions in string/M-theory are important not only
on their own but also because of their relations to the gauge theories and for integrability issues
as well. The development in obtaining such solutions in string theory inspires the interest to
analogous investigations in M-theory. In this short Letter we investigated the issue of existence
of spike-like solutions in membrane case. These solutions extend the class of previously found
giant magnon solutions at the large energy scale. Concretely, these issues are interesting for the
following reasons. The results for type IIB strings should be reproduced taking an appropriate
dimensional reduction. Then the question whether the known string solutions have analog in
eleven-dimensional spacetime appears. The answer to this question for the giant magnon string
solutions is positive [35]. It is natural to extend the analysis to the case of so-called single spike
solutions.
First we presented a short review of the string case and demonstrated how one can find solutions of this type using the reduction to the NeumannRosochatius integrable system. In Section 3
we turn to the membrane case. The analysis shows that there is an analog of this class solutions
with the same properties. To show that, we used again the possibility to reduce the theory to a
NeumannRosochatius integrable system. The solutions are obtained by taking an appropriate
440
limit and the dispersion relations are found to be

2

J1
p
J
2
2
=
1a E
+ 2 sin2 ,
2
2
2
7 2
2 p

3
2
= 2 T2 (lp R) a 1 a b , = 0 ,
2
2
for the giant magnon solution, and

E T g 2 =
,

2
J1
J2
=
+ 2 sin2 ,
2
2
, = 1 ,
T =
2
2
for the single spike solution, in full analogy with the string case.
Although the membrane/gauge theory correspondence is not well developed, we hope that
these results will shed light on the duality. The main obstacle is the lack of clear principle
and mechanism to identify the gauge theory operators. Collecting evidences and analogies with
strings will show up hopefully such a principle. There are hopes that one can find an integrable
spin chain which governs this correspondence at least in the sectors that are known to be integrable but still a lot of work is ahead.
Acknowledgements
This work was supported in part by Bulgarian NSF grants F -1412/04 and V U -F -201/06.
R.R. acknowledges a Guest Professorship and warm hospitality at the Institute for Theoretical
Physics, Vienna University of Technology. Many thanks to Max Kreuzer and his group for fruitful
and stimulating atmosphere. The work of R.R. is partially supported by Austrian Research Fund
FWF grant #P19051-N16.
References
[1] D.M. Hofman, J. Maldacena, Giant magnons, J. Phys. A 39 (2006) 1309513118, hep-th/0604135.
[2] J. Maldacena, I. Swanson, Connecting giant magnons to the pp-wave: An interpolating limit of AdS5 S 5 , hepth/0612079.
[3] J.A. Minahan, K. Zarembo, The Bethe-ansatz for N = 4 super-YangMills, JHEP 0303 (2003) 013, hep-th/0212208.
[4] L.D. Faddeev, How algebraic Bethe Ansatz works for integrable model, hep-th/9605187.
[5] N. Dorey, Magnon bound states and the AdS/CFT correspondence, J. Phys. A 39 (2006) 1311913128, hepth/0604175.
[6] H.-Y. Chen, N. Dorey, K. Okamura, Dyonic giant magnons, JHEP 0609 (2006) 024, hep-th/0605155.
[7] G. Arutyunov, S. Frolov, M. Zamaklar, Finite-size effects from giant magnons, Nucl. Phys. B 778 (2007) 135,
hep-th/0606126.
[8] J.A. Minahan, A. Tirziu, A.A. Tseytlin, Infinite spin limit of semiclassical string states, JHEP 0608 (2006) 049,
hep-th/0606145.
[9] M. Spradlin, A. Volovich, Dressing the giant magnon, JHEP 0610 (2006) 012, hep-th/0607009.
[10] M. Kruczenski, J. Russo, A.A. Tseytlin, Spiky strings and giant magnons on S 5 , JHEP 0610 (2006) 002, hepth/0607044.
[11] C. Kalousios, M. Spradlin, A. Volovich, Dressing the giant magnon II, hep-th/0611033.
441
[12] G. Papathanasiou, M. Spradlin, Semiclassical quantization of the giant magnon, arXiv: 0704.2389 [hep-th].
[13] W.-H. Huang, Giant magnons under NSNS and Melvin fields, JHEP 0612 (2006) 040, hep-th/0607161.
[14] K. Okamura, R. Suzuki, A perspective on classical strings from complex sine-Gordon solitons, Phys. Rev. D 75
(2007) 046001, hep-th/0609026.
[15] S. Hirano, Fat magnon, hep-th/0610027.
[16] S. Ryang, Three-spin giant magnons in AdS5 S 5 , JHEP 0612 (2006) 043, hep-th/0610037.
[17] H.-Y. Chen, N. Dorey, K. Okamura, The asymptotic spectrum of the N = 4 super-YangMills spin chain, hepth/0610295.
[18] P. Bozhilov, A note on two-spin magnon-like energy-charge relations from M-theory viewpoint, hep-th/0612175.
[19] J.A. Minahan, Zero modes for the giant magnon, JHEP 0702 (2007) 048, hep-th/0701005.
[20] D. Astolfi, V. Forini, G. Grignani, G.W. Semenoff, Gauge invariant finite size spectrum of the giant magnon, hepth/0702043.
[21] B. Vicedo, Giant magnons and singular curves, hep-th/0703180.
[22] J. Kluson, R.R. Nayak, K.L. Panigrahi, Giant magnon in NS5-brane background, hep-th/0703244.
[23] C.A.S. Young, q-Deformed supersymmetry and dynamic magnon representations, arXiv: 0704.2069 [hep-th].
[24] H.-Y. Chen, N. Dorey, R.F. Lima Matos, Quantum scattering of giant magnons, arXiv: 0707.0668 [hep-th].
[25] D. Berenstein, S.E. Vazquez, Giant magnon bound states from strongly coupled N = 4 SYM, arXiv: 0707.4669
[hep-th].
[26] M. Kruczenski, Spiky strings and single trace operators in gauge theories, JHEP 0508 (2005) 014, hep-th/0410226.
[27] S. Ryang, Wound and rotating strings in AdS5 S 5 , JHEP 0508 (2005) 047, hep-th/0503239.
[28] T. McLoughlin, X. Wu, Kinky strings in AdS5 S 5 , JHEP 0608 (2006) 063, hep-th/0604193.
[29] C.-S. Chu, G. Georgiou, V.V. Khoze, Magnons, classical strings and beta-deformations, JHEP 0611 (2006) 093,
hep-th/0606220.
[30] N.P. Bobev, R.C. Rashkov, Multispin giant magnons, Phys. Rev. D 74 (2006) 046011, hep-th/0607018.
[31] R. Ishizeki, M. Kruczenski, Single spike solutions for strings on S2 and S3, arXiv: 0705.2429 [hep-th].
[32] A.E. Mosaffa, B. Safarzadeh, Dual spikes; new spiky string solutions, arXiv: 0705.3131 [hep-th].
[33] N.P. Bobev, R.C. Rashkov, Spiky strings, giant magnons and beta-deformations, arXiv: 0706.0442 [hep-th].
[34] P. Bozhilov, Neumann and NeumannRosochatius integrable systems from membranes on AdS4 S 7 , arXiv:
0704.3082 [hep-th].
[35] P. Bozhilov, R.C. Rashkov, Magnon-like dispersion relation from M-theory, Nucl. Phys. B 768 (2007) 193208,
hep-th/0607116.

Interactions for a collection of spin-two fields

intermediated by a massless p-form
C. Bizdadea, E.M. Cioroianu, D. Cornea, E. Diaconu,
S.O. Saliu , S.C. Sararu
Faculty of Physics, University of Craiova, 13 A. I. Cuza Str., Craiova 200585, Romania
Received 24 June 2007; received in revised form 11 September 2007; accepted 4 October 2007
Abstract
Under the general hypotheses of locality, smoothness of interactions in the coupling constant, Poincar
invariance, Lorentz covariance, and preservation of the number of derivatives on each field, we investigate
the cross-couplings of one or several spin-two fields to a massless p-form. Two complementary cases arise.
The first case is related to the standard interactions from General Relativity, but the second case describes a
new, special type of couplings in D = p + 2 spacetime dimensions, which break the PT-invariance. Nevertheless, no consistent, indirect cross-interactions among different gravitons with a positively defined metric
in internal space can be constructed.
PACS: 11.10.Ef
Keywords: BRST symmetry; Local BRST cohomology; Interactions in field theory; Spin-two fields
1. Introduction
Theories involving one or several spin-two fields have raised a constant interest over the last
thirty years, especially at the level of direct or intermediated graviton interactions [118]. In this
context more results on the impossibility of consistent cross-couplings among different gravi* Corresponding author.
E-mail addresses: bizdadea@central.ucv.ro (C. Bizdadea), manache@central.ucv.ro (E.M. Cioroianu),

dcornea@central.ucv.ro (D. Cornea), ediaconu@central.ucv.ro (E. Diaconu), osaliu@central.ucv.ro (S.O. Saliu),
scsararu@central.ucv.ro (S.C. Sararu).
doi:10.1016/j.nuclphysb.2007.10.007
C. Bizdadea et al. / Nuclear Physics B 794 [PM] (2008) 442494
443
tons have been obtained, either without other fields [14] or in the presence of a scalar field [14],
a Dirac spinor [15], or respectively of a massive RaritaSchwinger field [16]. All these no-go
results have been deduced under some specific hypotheses, always including the preservation
of the derivative order of each field equation with respect to its free limit (derivative order assumption). Through their implications, these findings support the common belief that the only
consistent interactions in graviton theories require a single spin-two field and are subject to the
standard prescriptions of General Relativity (meaning diffeomorphisms for the gauge transformations of the graviton and diffeomorphism algebra for the gauge algebra of the interacting
theory). This idea is also strengthened by the confirmation of the uniqueness of EinsteinHilbert
action [14] having the PauliFierz model as its free limit or the uniqueness of N = 1, D = 4
SUGRA action [17] allowing for a PauliFierz field and a massless RaritaSchwinger spinor
as the corresponding uncoupled limit. Indirect arguments are thus presented in favour of ruling
out N > 8 extended supergravity theories since they require more than one spin-two field. It is
nevertheless known that the relaxation of the derivative order condition may lead to exotic couplings for one or a collection of spin-two fields [18], which are no longer mastered by General
Relativity.
Our paper submits to the same topic, of constructing spin-two field(s) couplings, initially in
the presence of a massless vector field and then of a p-form, with p > 1. We employ a systematic
approach to the construction of interactions in gauge theories [1921], based on the cohomological reformulation of Lagrangian BRST symmetry [2226]. In this approach interactions result
from the analysis of consistent deformations of the generator of the Lagrangian BRST symmetry
(known as the solution of the master equation) by means of specific cohomological techniques,
relying on local BRST cohomology [27,28]. The emerging deformations, and hence also the interactions, are constructed under the general hypotheses of locality, smoothness in the coupling
constant, Poincar invariance, Lorentz covariance, and derivative order assumption. In this specific situation the derivative order assumption requires that the interaction vertices contain at
most two spacetime derivatives of the fields, but does not restrict the polynomial order in the
undifferentiated fields either in the Lagrangian or in the gauge symmetries. Our analysis envisages three steps, which introduce gradually the situations under investigation, according to the
complexity of their cohomological content.
Initially, we consider the case of couplings between a single PauliFierz field [29,30] and a
massless vector field. In this setting we compute the coupling terms to order two in the coupling
constant k and find two distinct solutions. The first solution leads to the full cross-coupling
Lagrangian in all D > 2
(int)
LI

1
gg g F F + k q1 3D 1 2 3 V1 F2 3
4

+ q2 5D 1 2 3 4 5 V1 F2 3 F4 5 ,
which respects the standard rules of General Relativity. The second solution is more unusual:
it lives only in D = 3, produces polynomials of order two in the coupling constant (and not
series, like in the first case), and the couplings are mixing-component terms that can be written
in terms of a deformed field strength (of the massless vector-field) as
1
(int)

LII
= F F
, F
= F + 2k [ h] .
4
By contrast to General Relativity, where all the gauge symmetries are deformed, here only those
of the vector field are modified by terms of order one in the coupling constant that involve
444
the PauliFierz gauge parameters, while the spin-two field keeps its original gauge symmetries,
namely the linearized version of diffeomorphisms. To our knowledge, this is the first situation
where the linearized version of the spin-two field allows for nontrivial couplings, other than
those subject to General Relativity, which fulfill all the working hypotheses, including that on
the derivative order.
Next, we focus on the investigation of cross-interactions among different gravitons intermediated by a massless vector field. In view of this, we start from a finite sum of PauliFierz actions
with a positively defined metric in internal space and a massless vector field. The cohomological
analysis reveals again two cases. The former is related to the standard graviton-vector field interactions from General Relativity and exhibits no consistent cross-interactions among different
gravitons (with a positively defined metric in internal space) in the presence of a massless vector
field. At most one graviton can be coupled to the vector field via a Lagrangian similar to LI(int) ,
while each of the other spin-two fields may interact only with itself through an EinsteinHilbert
action with a cosmological term. The latter case seems to describe some new type of couplings
in D = 3, which appear to allow for cross-couplings among different gravitons. The coupled
Lagrangian is, like in the case of a single graviton, a polynomial of order two in the coupling
constant, obtained by deforming the vector field strength
1
(int)
L II
= F F ,
4
F = F + 2k
n

y3A [ hA
] ,
A=1
where y3A are some arbitrary, non-vanishing real constants. Nevertheless, these cross-couplings
can be decoupled through an orthogonal, linear transformation of the spin-two fields, in terms of
(int)
(int)
which L II becomes nothing but LII , with h replaced for instance by the first transformed
spin-two field from the collection. In consequence, these case also leads to no indirect crosscouplings between different gravitons.
Then, we show that all the new results obtained in the case a massless vector field can be
generalized to an arbitrary p-form. More precisely, if one starts from a free action describing an
Abelian p-form and a single PauliFierz field, then it is possible to construct some new deformations in D = p + 2 that are consistent to all orders in the coupling constant and are not subject
to the rules of General Relativity. It is important to remark that all the working hypotheses, including the derivative order assumption, are fulfilled. There are several physical consequences
of these couplings, such as the appearance of a constant linearized scalar curvature if one allows
for a cosmological term or the modification of the initial (p + 1)-order conservation law for the
p-form by terms containing the spin-two field. Regarding a collection of spin-two fields, we find
that the deformed Lagrangian does not allow for cross-couplings between different gravitons
intermediated by a p-form, either in the setting of General Relativity or in the special, (p + 2)dimensional situation.
This paper is organized in seven sections. In Section 2 we construct the BRST symmetry of a free model with a single PauliFierz field and one massless vector field. Section 3
briefly addresses the deformation procedure based on the BRST symmetry. In Sections 4 and 5
we compute the deformations corresponding to a vector field and one or respectively several
spin-two fields, and emphasize the Lagrangian formulation of the resulting theories. Section 6
discusses the generalization of the previous results to the case of couplings between one or several
gravitons and an arbitrary p-form gauge field. Section 7 ends the paper with the main conclusions.
445
2. BRST symmetry of the free model

Our starting point is represented by a free Lagrangian action, written as the sum between the
linearized HilbertEinstein action (also known as the PauliFierz action) and Maxwells action
in D > 2 spacetime dimensions

1
S0L [h , V ] = d D x ( h ) h + h h
2

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

(PF)
(vect)
.
d D x L 0 + L0
(1)
The restriction D > 2 is required by the spin-two field action, which is known to reduce to a total
derivative in D = 2. Throughout the paper we work with the flat metric of mostly plus signature, = ( + +). In the above h denotes the trace of the PauliFierz field, h = h ,
and F represents the Abelian field-strength of the massless vector field (F [ V] ). The
theory described by action (1) possesses an Abelian and irreducible generating set of gauge
transformations
h = (
) ,
V =
,
(2)
with
and
bosonic gauge parameters. The notation [ . . . ] (or ( . . . )) signifies antisymmetry (or symmetry) with respect to all indices between brackets without normalization factors
(i.e., the independent terms appear only once and are not multiplied by overall numerical factors).
In order to construct the BRST symmetry for action (1), it is necessary to introduce the
field/ghost and antifield spectra

0 = (h , V ),
(3)
0 = h , V ,

1
1 = ( , ),
(4)
= , .
The fermionic ghosts 1 are associated with the gauge parameters
1 = {
,
} respectively
and the star variables represent the antifields of the corresponding fields/ghosts. (According to
the standard rule of the BRST method, the Grassmann parity of a given antifield is opposite
to that of the corresponding field/ghost.) Since the gauge generators are field-independent and
irreducible, it follows that the BRST differential decomposes into
s = + ,
(5)
where is the KoszulTate differential and denotes the exterior longitudinal derivative.
The KoszulTate differential is graded in terms of the antighost number (agh, agh() = 1,
agh( ) = 0) and enforces a resolution of the algebra of smooth functions defined on the stationary surface of field equations for action (1), C (), : S0L / 0 = 0. The exterior longitudinal derivative is graded in terms of the pure ghost number (pgh, pgh( ) = 1, pgh() = 0) and
is correlated with the original gauge symmetry via its cohomology in pure ghost number zero
computed in C (), which is isomorphic to the algebra of physical observables for this free
theory. These two degrees of the BRST generators are valued as

agh 0 = agh(1 ) = 0,
(6)
agh 1 = 2,
agh 0 = 1,

0
1
pgh 0 = pgh
= 0,
pgh(1 ) = 1,
= 0.
pgh
(7)
446
The overall degree that grades the BRST complex is named ghost number (gh) and is defined
like the difference between the pure ghost number and the antighost number, such that gh(s) =
gh() = gh( ) = 1. The actions of the operators and (taken to act as right differentials) on
the BRST generators read as
h = 2H ,
V = F ,
= 2 h
= V
(8)
,
(9)
= 0,
1 = 0,
(10)
= 0,
(11)
h = ( ) ,
= 0,
= 0,
V = ,
= 0.
(12)
(13)
In the above H is the linearized Einstein tensor

1
H = K K,
(14)
2
with K and K the linearized Ricci tensor and the linearized scalar curvature respectively, both
obtained from the linearized Riemann tensor
1
K| = ( h + h h h ),
2
from its trace and double trace respectively
K = K| ,
K = K| .
(15)
(16)
The BRST differential is known to have a canonical action in a structure named antibracket
which is obtained by considering the fields/ghosts
and denoted by the symbol (,) (s = (, S)),
conjugated respectively to the corresponding antifields. The generator of the BRST symmetry
is a bosonic functional of ghost number zero, which is solution to the classical master equation
S)
= 0. The full solution to the master equation for the free model under study reads as
(S,

S = S0L [h , V ] + d D x h ( ) + V
(17)
and encodes all the information on the gauge structure of the theory (1)(2).
3. Brief review of the deformation procedure
We begin with a free gauge theory, described by a Lagrangian action S0L [ 0 ], invariant
S L
under some gauge transformations
0 = Z 0 1
1 , i.e. 00 Z 0 1 = 0, and consider the
problem of constructing consistent interactions among the fields 0 such that the couplings preserve the field spectrum and the original number of gauge symmetries. This matter is addressed
by means of reformulating the problem of constructing consistent interactions as a deformation
problem of the solution to the master equation corresponding to the free theory [1921]. Such
a reformulation is possible due to the fact that the solution to the master equation contains all the
information on the gauge structure of the theory. If an interacting gauge theory can be consistently constructed, then the solution S to the master equation associated with the free theory,
S)
= 0, can be deformed into a solution S
(S,
S S = S + kS1 + k 2 S2 + = S + k
447

d D x a + k2
dDx b +
(18)
of the master equation for the deformed theory

(S, S) = 0,
(19)
such that both the ghost and antifield spectra of the initial theory are preserved. The projection
of Eq. (19) on the various orders in the coupling constant k leads to the equivalent tower of
equations
S)
= 0,
(S,
= 0,
2(S1 , S)
+ (S1 , S1 ) = 0,
2(S2 , S)
..
.
(20)
(21)
(22)
Eq. (20) is fulfilled by hypothesis. The next equation requires that the first-order deformation of
the solution to the master equation, S1 , is a co-cycle of the free BRST differential s, sS1 = 0.
However, only cohomologically nontrivial solutions to (21) should be taken into account, since
the BRST-exact ones can be eliminated by some (in general nonlinear) field redefinitions. This
means that S1 pertains to the ghost number zero cohomological space of s, H 0 (s), which is nonempty because it is isomorphic to the space of physical observables of the free theory. It has
been shown (by of the triviality of the antibracket map in the cohomology of the BRST differential) that there are no obstructions in finding solutions to the remaining equations, namely (22),
etc. However, the resulting interactions may be nonlocal and there might even appear obstructions if one insists on their locality. The analysis of these obstructions can be done with the help
of cohomological techniques.
4. Consistent interactions between the spin-two field and a massless vector field
4.1. Standard material: Basic cohomologies
The aim of this section is to investigate the cross-couplings that can be introduced between the
spin-two field and a massless vector field. This matter is addressed in the context of the antifieldBRST deformation procedure described in the above and relies on computing the solutions to
Eqs. (21)(22), etc., with the help of the BRST cohomology of the free theory. The deformations are obtained under the following (reasonable) assumptions: smoothness in the deformation
parameter, locality, Lorentz covariance, Poincar invariance, and the presence of at most two
derivatives in the coupled Lagrangian. Smoothness in the deformation parameter refers to the
fact that the deformed solution to the master equation, (18), is smooth in the coupling constant
k and reduces to the original solution, (17), in the free limit k = 0. The hypothesis on the deformed theory to be Poincar invariant means that one does not allow an explicit dependence on
the spacetime coordinates into the deformed solution to the master equation. The requirement
concerning the maximum number of derivatives allowed to enter the deformed Lagrangian is
frequently imposed in the literature; for instance, see the case of couplings between the Pauli
Fierz and the massless RaritaSchwinger fields [17] or of cross-interactions for a collection of
448
PauliFierz fields [14]. If we make the notation S1 =

first-order deformation, takes the local form
sa = m ,
gh(a) = 0,
d D x a, then Eq. (21), which controls the
(a) = 0,
(23)
for some local current m . It shows that the non-integrated density of the first-order deformation pertains to the local cohomology of the free BRST differential in ghost number zero,
a H 0 (s|d), where d denotes the exterior spacetime differential. The solution to (23) is unique
up to s-exact pieces plus divergences
a a + sb + n ,
(24)
with gh(b) = 1, (b) = 1, gh(n ) = 0, and (n ) = 0. At the same time, if the general solution
of (23) is found to be completely trivial, a = sb + n , then it can be made to vanish, a = 0.
In order to analyze equation (23) we develop a according to the antighost number
a=
I
agh(ai ) = i,
ai ,
gh(ai ) = 0,
(ai ) = 0,
(25)
i=0
and assume, without loss of generality, that decomposition (25) stops at some finite value of I .
This can be shown for instance like in Appendix A of [14]. Replacing decomposition (25) into
(23) and projecting it on the various values of the antighost number by means of (5), we obtain
that (23) is equivalent with the tower of equations
aI = mI ,
(26)
aI + aI 1 = mI 1 ,
ai + ai1 = mi1 ,
where (mi )i=0,I are some local
(27)
1 i I 1,
(28)
agh(mi ) = i.
currents, with
Moreover, according to the general
result from [14] in the absence of collection indices, Eq. (26) can be replaced in strictly positive
antighost numbers by
aI = 0,
I > 0.
(29)
Due to the second-order nilpotency of ( 2 = 0), the solution to (29) is unique up to -exact
contributions
a I a I + bI ,
agh(bI ) = I,
pgh(bI ) = I 1,
(bI ) = 1.
(30)
Meanwhile, if it turns out that aI reduces to -exact terms, aI = bI , then it can be made to
vanish, aI = 0. In other words, the nontriviality of the first-order deformation a is translated at
its highest antighost number component into the requirement that aI H I ( ), where H I ( )
denotes the cohomology of the exterior longitudinal derivative in pure ghost number equal
to I . So, in order to solve equation (23) (equivalent with (29) and (27)(28), we need to compute
the cohomology of , H ( ), and, as it will be made clear below, also the local cohomology of ,
H (|d).
Using the results on the cohomology of in the PauliFierz sector [14] as well as definitions
(11)(13), we can state that H ( ) is generated on the one hand by 0 , 1 , F , and K ,
together with their spacetime derivatives and, on the other hand, by the undifferentiated ghosts
and as well as by their antisymmetric first-order derivatives [ ] . (The spacetime derivatives
of are -exact, in agreement with the latter definition from (12), and the same is valid for the
449
derivatives of of order two and higher.) So, the most general (and nontrivial) solution to (29)
can be written, up to -exact contributions, as

aI = I [F ], [K ], 0 , 1 eI (, , [ ] ),
(31)
where the notation f ([q]) means that f depends on q and its derivatives up to a finite order,
while eI denotes the elements of a basis in the space of polynomials with pure ghost number I
in , , and [ ] . The objects I (obviously nontrivial in H 0 ( )) were taken to have a finite
antighost number and a bounded number of derivatives, and therefore they are polynomials in
the antifields, in the linearized Riemann tensor K , and in the field-strength F as well as
in their subsequent derivatives. They are required to fulfill the property agh(I ) = I in order to
ensure that the ghost number of aI is equal to zero. Due to their -closeness, I = 0, and to
their polynomial character, I will be called invariant polynomials. In antighost number zero the
invariant polynomials are polynomials in the linearized Riemann tensor, in the field-strength of
the Abelian field, and in their derivatives. The result that one can replace Eq. (26) with (29) is a
consequence of the triviality of the cohomology of the exterior spacetime differential in the space
of invariant polynomials in strictly positive antighost numbers. For more details, see Section A.1
from [14].
Inserting (31) in (27), we obtain that a necessary (but not sufficient) condition for the existence
of (nontrivial) solutions aI 1 is that the invariant polynomials I are (nontrivial) objects from
the local cohomology of the KoszulTate differential H (|d) in antighost number I > 0 and in
pure ghost number zero

I = jI 1 ,
(32)
pgh jI 1 = 0.
agh jI 1 = I 1,
We recall that the local cohomology H (|d) is completely trivial in both strictly positive
antighost and pure ghost numbers (for instance, see Theorem 5.4 from [27] and also [28]). Using
the fact that the Cauchy order of the free theory under study is equal to two, the general results
from [27] and [28], according to which the local cohomology of the KoszulTate differential
in pure ghost number zero is trivial in antighost numbers strictly greater than its Cauchy order,
ensure that
HJ (|d) = 0,
J > 2,
(33)
where HJ (|d) denotes the local cohomology of the KoszulTate differential in antighost number J and in pure ghost number zero. It can be shown that any invariant polynomial that is trivial
in HJ (|d) with J 2 can be taken to be trivial also in HJinv (|d). (HJinv (|d) denotes the invariant characteristic cohomology in antighost number J the local cohomology of the KoszulTate
differential in the space of invariant polynomials.) Thus:

J = bJ +1 + cJ , agh(J ) = J 2 J = J +1 + J ,
(34)
with both J +1 and J invariant polynomials. Results (34) and (33) yield the conclusion that the
invariant characteristic cohomology is trivial in antighost numbers strictly greater than two
HJinv (|d) = 0,
J > 2.
(35)
By proceeding in the same manner like in [14] and [31], it can be proved that the spaces H2 (|d)
and H2inv (|d) are spanned by

H2 (|d), H2inv (|d) : , .
(36)
450
In contrast to the groups (HJ (|d))J 2 and (HJinv (|d))J 2 , which are finite-dimensional, the
cohomology H1 (|d) in pure ghost number zero, known to be related to global symmetries and
ordinary conservation laws, is infinite-dimensional since the theory is free. Fortunately, it will
not be needed in the sequel.
The previous results on H (|d) and H inv (|d) in strictly positive antighost numbers are
important because they control the obstructions of removing the antifields from the first-order
deformation. Based on formulas (33)(35), one can eliminate all the pieces of antighost number
strictly greater than two from the non-integrated density of the first-order deformation by adding
only trivial terms. Consequently, one can take (without loss of nontrivial objects) I 2 into the
decomposition (25). (The proof of this statement can be realized like in Section A.3 from [14].)
In addition, the last representative reads as in (31), where the invariant polynomial is necessarily
a nontrivial object from H2inv (|d) if I = 2 and from H1 (|d) if I = 1, respectively.
4.2. Computation of first-order deformations
Assuming I = 2, the non-integrated density of the first-order deformation (25) becomes
a = a0 + a1 + a2 .
(37)
We can further decompose a in a natural manner as

a = a (PF) + a (int) + a (vect) ,
a (PF)
(38)
a (int)
where
contains only fields/ghosts/antifields from the PauliFierz sector,
mixes both
fields, and a (vect) involves only the vector field sector. The component a (PF) is completely known
[14] and satisfies by itself an equation of the type (23). It admits a decomposition similar to (37)
a (PF) = a0(PF) + a1(PF) + a2(PF) ,
(39)
where
f
[ ] ,
2

(PF)
a1 = f h h [ h] ,
a2(PF) =
(40)
(41)
(PF)
and a0 is the cubic vertex of the EinsteinHilbert Lagrangian multiplied by a real constant f
plus a cosmological term1
(PF)
a0
(EH-cubic)
= f a0
2h,
(42)
with the cosmological constant. Due to the fact that a (int) and a (vect) contain different sorts of
fields, it follows that they are subject to two separate equations
sa (vect) = m(vect) ,
(43)
= m
(44)
sa
(int)
(int)
1 The terms a (PF) and a (PF) given in (40) and (41) differ from those present in [14] (in the absence of collection
2
1
(PF)
(PF)
indices) by a -exact and respectively a -exact contribution. However, the difference between our a2 + a1
and
(PF)
that from [14] is a s-exact modulo d quantity. The associated a0
is nevertheless the same in both formulations. As a
consequence, a (PF) and the first-order deformation from [14] belong to the same cohomological class from H 0 (s|d).
451
for some local m s. It is known (for instance, see [32]) that the general solution to (43) reduces
to its component of antighost number zero and reads as

D
1 2 3 ...2j 2j +1
qj 2j
V1 F2 3 F2j 2j +1 ,
a (vect) = a0(vect) =
(45)
+1
j >0
with qj some real constants. Selecting from (45) only the terms with maximum two spacetime
derivatives, we conclude that we must ask qj = 0 for all j > 2, so
(vect)
a (vect) = a0
= q1 3D V F + q2 5D V F F .
(46)
D signifies the Kronecker symbol. In the sequel we analyze the general solution to
The notation m
Eq. (44).
Due to (37), we should consider that the general solution to (44) stops at antighost number
two, a (int) = a0(int) + a1(int) + a2(int) . Eq. (44) is equivalent to the fact that the components of a (int)
are subject to Eqs. (29) and (27)(28) with I = 2 and a replaced by a (int) . It can be shown that
there exist no such solutions ending at antighost number two. For the sake of simplicity, we omit
(int)
the proof of this result, which is mainly based on showing that there is no nontrivial a2 yielding
(int)
a consistent a0 . In view of this finding, we approach the next situation, where the solution to
(44) stops at antighost number one
a (int) = a0(int) + a1(int) ,
(47)
such that the components on the right-hand side of (47) are subject to the equations
a1(int) = 0,
(int)
a1
(int)
+ a0
(48)
(int)
= m0
.
(49)
In agreement with (31) for I = 1 and the discussion from the end of Section 4.1, the general
solution to (48) is (up to trivial, -exact contributions)
(int)
a1
= 1 + 1 + 1 [ ] ,
(50)
where 1 , 1 , and 1 are nontrivial invariant polynomials from H1 (|d) (but not necessarily from H1inv (|d)) in order to produce a consistent a0(int) . Because they are nontrivial invariant
polynomials of antighost number one, we can always assume that they are linear in the undifferentiated antifields V and h , such that (50) becomes

(int)
a1 = V M + M + M [ ]

+ h N + N + N [ ] ,
(51)
where all the coefficients, denoted by M or N , must be -closed quantities, and therefore they
may depend on F , K| , and their derivatives. In addition, these tensors are subject to the
symmetry/antisymmetry properties
M = M ,
N = N ,
N = N ,
N = N = N .
(52)
(53)
At this point we recall the hypothesis on the derivative order of the deformed Lagrangian, which
(int)
imposes that a0 as solution to (49) contains at most two spacetime derivatives of the fields.
452
Then, relation (51), Eq. (49), and definitions (8)(13) yield the following results: (A) none of the
M- or N -type tensors entering (51) are allowed to depend on K| or its derivatives; (B) M
and N cannot involve either F or its derivatives, and therefore they are non-derivative,
constant tensors; (C) the tensors M , M , N , and N may depend on F (and not on its
derivatives), but only in a linear manner. These results are synthesized by the formulas
M = C + C F ,
M = C + C F ,
(54)
N = D + D F ,
N = D + D F
M = C ,
N = D ,
(55)
(56)
D, or D are non-derivative, constant tensors, subject to

where the quantities denoted by C, C,
some symmetry/antisymmetry properties such that (52) and (53) are fulfilled. Since we work in
D > 2 spacetime dimensions, the only choice that complies with the above mentioned properties
and leads to consistent cross-couplings between the PauliFierz field and the vector field is2 :
C = 0,
C = 0,
C = y1 ,
p
C = ( ),
2
D = D = D = 0,
D = y2 ,
C = y3 3D .
D = 0,
(57)
(58)
(59)
(60)
Substituting (57)(60) in (54)(56) and the resulting expressions in (51), we obtain

(int)
a1
= y1 V + y2 h + y3 3D V [ ] + pV F ,
where h = h . Acting with on (61), we infer

(int)
a1 = (D 2)y2 V [ h] y3 3D F [ h]

p
1
F F h + F F h + u
2
4

+ y1 + (D 2)y2 V .
(61)
(62)
Comparing (62) with (49) and observing that (61) already contains a term of the type V , it
(int)
follows that a1 is consistent at antighost number zero if and only if
y1 + (D 2)y2 = 0.
(63)
(int)
D
2 Strictly speaking, there is a non-vanishing solution C
the term
= z3 , which adds to a1
D
z3 V F . Even if consistent, this term would lead to selfinteractions in the Maxwell sector. How(int)
is restricted by hypothesis to provide only cross-couplings between the PauliFierz field and the
ever, a1
electromagnetic field, so this term must be removed from this context by setting z = 0. Apparently, there
(int)
are two more possibilities, C = z 4D and D = z 3D , which add to a1
the terms
(z 3D h F z 4D V F ) . They are not eligible to enter a1 since the corresponding invariant polynomial, z 3D h F z 4D V F , does not belong to H 1 (|d), such that they cannot lead to consistent
(int)
(int)
pieces in a0
unless z = 0 = z .
453
Replacing (63) into (61) and (62), we get finally

(int)
a1
(int)
a0
= y2 h (D 2)V + y3 3D V [ ] + pV F ,
= (D 2)y2 V [ h] + y3 3D F [ h]

p
1
(int)
+
F F h + F F h + a 0 ,
2
4
(int)
where a 0
(64)
(65)
is the general solution to the homogeneous equation
a 0(int) = m
(int) .
(66)
(int)
Such solutions correspond to a 1 = 0 and thus they cannot deform either the gauge algebra or the gauge transformations, but only the Lagrangian at order one in the coupling constant.
There are two main types of solutions to (66). The first one corresponds to m
(int) = 0 and is
given by gauge-invariant, non-integrated densities constructed from the original fields and their
spacetime derivatives. According to (31) for both pure ghost and antighost numbers equal to
zero, they are given by a 0(int) = a 0(int) ([F ], [K| ]), up to the conditions that they describe
true cross-couplings between the two types of fields and cannot be written in a divergence-like
form. Unfortunately, this type of solutions must depend simultaneously at least on the linearized
Riemann tensor and on the Abelian field strength in order to provide cross-couplings, so they
would lead to terms with at least three derivatives in the deformed Lagrangian. By virtue of the
derivative order assumption, they must be discarded by setting a 0(int) = 0. The second kind of
solutions is associated with m
(int)
= 0 in (66), being understood that we maintain the require(int)
ments on a 0 to contain maximum two derivatives of the fields and to describe cross-couplings.
In order to simplify the presentation, we omit the technical aspects regarding the analysis of these
(int)
solutions. The main result is that, without loss of generality, we can take a 0 = 0 in (65). Very
briefly, we mention that the procedure used for obtaining this result relies on decomposing a 0(int)
(int)
along the number of derivatives, a 0 = 0 + 1 + 2 , where i contains exactly i derivatives
of the fields. As a consequence, Eq. (66) becomes equivalent to three independent equations,
one for each component. The terms 0 and respectively 1 are ruled out because they cannot
produce cross-couplings. As for 2 , it requires the existence of a non-derivative, real, constant
tensor C |; , which displays the generalized symmetry properties of the Riemann tensor with
respect to its first four indices and is simultaneously antisymmetric in its last three indices. Since
there are no such tensors in any D 3 spacetime dimension, we must discard 2 , which finally
(int)
leaves us with a 0 = 0.
Replacing (64), (65), and a 0(int) = 0 in (47), we obtain the concrete form of the general solution
a (int) to (44). We can still remove certain trivial, s-exact modulo d terms from the resulting a (int) .
Indeed, we have that

1
a (int) = a (int) + s p V + V V h
(67)
+ t ,
2
such that, in agreement with the discussion made in the beginning of this section, we can work
with
454
(int)

1

=a
+ s p V + V V h
t
2

y2 h + (D 2) V + V [ h]

+ y3 3D V [ ] + F [ h]

1
+ p V V [ ] + 2( V ) h
2

1
+ F
2[ h] V + F h 4F h
8
(int)
(68)
instead of a (int) .
In view of the results (39), (46), and (68) we conclude that the most general, nontrivial firstorder deformation of the solution to the master equation corresponding to action (1) and to its
gauge transformations (2), which complies with all the working hypotheses, is expressed by
(PF)
S 1 = S1
where
(PF)
S1
(PF)
(69)
and
(int)
S1
(int)
+ S1

d xa
(PF)
(PF)
(PF)
,
d D x a2 + a1 + a0
(70)

d D x a (int) + a (vect)

d D x y2 h + (D 2) V + V [ h]

+ y3 3D V [ ] + F [ h]

1
+ p V V [ ] + 2( V ) h
2

1
+ F 2[ h] V + F h 4F h
8

+ q1 3D V F + q2 5D V F F .
(71)
Thus, the first-order deformation of the solution to the master equation for the model under study
(PF)
is parameterized by seven independent, real constants, namely f and corresponding to S1
(int)
(see (40), (41), and (42)) together with p, y2 , y3 3D , q1 3D , and q2 5D associated with S1 .
4.3. Computation of second-order deformations
Here, we approach the construction of the second-order deformation of the solution to the
master equation, governed by Eq. (22). Replacing (69) into (22) we find that it becomes equivalent to the equations
(PF) (PF) (int) (int) (PF)
(PF)
+ S1 , S1
+ 2sS2 = 0,
S 1 , S1
(72)
(PF) (int) (int) (int) (int)
(int)
2 S 1 , S1
(73)
+ S1 , S1
+ 2sS2 = 0,

(int)
455
(int)
where (S1 , S1 )(PF) comprises only BRST generators from the PauliFierz sector and each
(int) (int)
term from (S1 , S1 )(int) contains at least one BRST generator from the one-form sector. By
writing down (72) and (73), it is understood that the second-order deformation decomposes as
(PF)
S2 = S 2
(int)
+ S2
(74)
(PF)
(int)
where S2 represents the component from the PauliFierz sector and S2 signifies the complementary part.
Initially, we analyze Eq. (72). It is known from the literature (for instance, see [14] in the
(PF)
absence of collection indices) that there exists S2 (f 2 , f ) such that

(PF) (PF)
(PF) 2
+ 2sS2
f , f = 0,
S 1 , S1
(75)
where
(PF) 2
S2

(EH-quartic)
f , f = f 2 S2
+f

1
d D x h h h2 ,
2
(76)
(EH-quartic)
the second-order EinsteinHilbert deformation, including the quartic vertex of
with S2
the EinsteinHilbert Lagrangian. On the other hand, direct computation based on (71) leads to

(int) (int) (PF)

(D 2)2 2
D
S1 , S1
= 2s d x y22
h h h
4

[ ]

D
2 D [ ]
+ y2 y3 (D 2)3 h
h + y3 3 h
[ h]
(PF) 2
(PF)
(PF) 2
y2 + S2 (y2 y3 ) + S2
y3 ,
2s S2
where we used the obvious notations
2

(PF) 2
2 (D 2)
y2 = y2
S2
d D x h2 h h ,
4

(PF)
S2 (y2 y3 ) = y2 y3 (D 2)3D d D x [ h] h ,

(PF) 2
S2
y3 = y32 3D d D x [ h] [ h] .
Taking into account relations (75)(77) it follows that (72) becomes equivalent with
(PF) (PF) 2

(PF) 2
(PF)
(PF) 2
s S2 S2
f , f + S2
y2 + S2 (y2 y3 ) + S2
y3
= 0,
(77)
(78)
(79)
(80)
(81)
(PF)
which allows us to determine the component S2 from the second-order deformation (74), up
to trivial, s-exact contributions,3 in the form

S2(PF) = S2(PF) f 2 , f + S2(PF) y22 + S2(PF) (y2 y3 ) + S2(PF) y32 .
(82)
3 Strictly speaking, we must add to (82) the nontrivial solution F to the homogeneous equation sF = 0. However, this
solution brings nothing new and can always be absorbed into the full deformed solution to the master equation S (actually
(PF)
(PF)
in S1 ) through a convenient redefinition of the coupling constant and of the other constants that parameterize S1 .
For instance, see Section 7 from [14].
456
Next, we pass to Eq. (73). If we denote by (int) and b(int) the non-integrated densities of
(PF) (int)
(int) (int)
(int)
2(S1 , S1 ) + (S1 , S1 )(int) and S2 respectively,

2 S1 , S1
(83)
+ S1 , S1
d D x (int) ,

(int)
S2 d D x b(int) ,
(84)
then the local form of Eq. (73) reads as
(int) = 2sb(int) + n ,
(85)
where

gh (int) = 1,

gh b(int) = 0,

gh n = 1,
(86)
for some local currents n . By direct computation, from (70) and (71) we deduce that (int)
decomposes as
(int) =
2
(int)
(int)
= I,
agh I
(87)
I =0
where
(int)
(int)

= p p h (p + f )V [ ] + w2 ,
(88)

= p p h (p + f )V [ ]

1
2
+ p V
( V )h + ([ h] )V
2

1
1
3
V h[ (] ) V ( [ )h] h h
4
4
4

1
+ p(p + f )V V ([ h] + [ h] ) h h
2

y3 3D V f h [ ] + (2p + f ) [ h]

+ py2 V (D 2)h V y2 h f (h + 2V )

2(p + f ) V p(p + f )V F [ ]

+ (2p + f )V y3 3D [ ] [ ]
+ y2 (D 2)([ ] ) + w1 ,
(89)
(int)
0

1
1
= p V
( V )h + ([ h] )V V h[ (] )
2
4

3
16
1
y3 q1 3D h
V ( [ )h] h h +
4
4
D2

+ p(p + f )V V ([ h] + [ h] ) h h
2
2
457

p 2 [ ]
([ h] )V 2 [ h] h[ ] V
V h
8

[ ]
+ h V h[ ] V + F h h [ (] V ) [ h ] V

1
+ F h [ ] h V + p 2 F F h h + F h2 2h h
16

1
h [ h] V h[ ] V + F h h F h h
2

1
1
+
[ h] V h[ (] V ) h + p(p + f ) F F
4
4

1
+ F F h h + q1 p3D hV F 2h V F + h V F
4

+ q2 p5D hV F F 4h V F F + 2h V F F

D
[ ]
2
16y3 q1 3 V h (D 2)(D 1)y2 V V

+
4q1 y2 3D (D 2) F 6q2 y2 5D F F

1
1
+ p(p + f ) F F + F F h [ ] 2[ h ]
2
4
(int)

(int)
+ y2 f A0 0 0 1 + pB0
0 0 1 4D
(int)

(int)
+ y3 3D f C0
0 0 1 + pD0 0 0 1 + w0 .
(int)
(int)
(int)
(90)
(int)
In (90) A0 , B0 , C0 , and D0 are linear in their arguments; for instance, the notation
A0(int) ( 0 0 1 ) means that each term from A0(int) contains two spacetime derivatives and
is simultaneously quadratic in the fields 0 from (3) and linear in the ghosts 1 from (4).
Replacing decomposition (87) into Eq. (85) and using (5), one can assume, without loss of

(int)
generality, that b(int) and n stop at antighost number three: b(int) = 3I =0 bI , n = 3I =0 nI .
inv
However, it can be shown in a direct manner (based on the result H3 (|d) = 0) that one can
(int)
take b3 = 0, so we can work with
b(int) =
2
(int)
bI
(int)
= I,
agh bI
(91)
I =0
n =
2
nI ,

agh nI = I.
(92)
I =0
The above expansions inserted into Eq. (85) produce the equivalent equations
2(int) = 2 b2(int) + n2 ,
(int)
(int)
(int)
1 = 2 b2 + b1
+ n1 ,
(int)
(int)
(int)
0 = 2 b1 + b0
+ n0 .
(93)
(94)
(95)
At this stage it is useful to make the notations

(int)
b2

1
(int)
= p p h (p + f )V [ ] + b2 ,
2
(96)
458

(int)
b1
(int)
b0

1 2
1
= p V
( V )h + ([ h] )V
2
2

1
1
3
V h[ (] ) V ( [ )h] h h
4
4
4

1
p(p + f )V V ([ h] + [ h] ) h h
4

1
+ y3 3D V f h [ ] + (2p + f ) [ h]
2

1
1
py2 V (D 2)h V + y2 h f (h + 2V )
2
2
8
(int)
y3 q1 3D h + b1 ,
2(p + f ) V
D2
(97)

p 2 [ ]
([ h] )V 2 [ h] h[ ] V
V h
16

+ h [ ] V h[ ] V + F h h [ (] V ) [ h ] V

1

1
+ F h [ ] h V p 2 F F h h + F h2 2h h
2
16

1
h [ h] V h[ (] V ) + F h h F h h
2

1
1
+
[ h] V h[ (] V ) h p(p + f ) F F
4
8

1
1
+ F F h h q1 p3D hV F 2h V F + h V F
4
2

1
q2 p5D hV F F 4h V F F + 2h V F F
2
1
(int)
+ 8y3 q1 3D V [ h] + (D 2)(D 1)y22 V V + b0 .
(98)
2
(int)
Using the above notations and recalling the expressions (88)(90) of I

(equivalent to (85)) become
b2
(int)
, Eqs. (93)(95)
= 2 ,
(int)
+ b1 = 1 + 1 ,
2
1
(int)
(int)
b1 + b0 = 0 + 0 ,
2
where
1

I = wI nI , I = 0, 2,
2
b2
(int)

1 = V p(p + f )F [ ] + (2p + f ) y3 3D [ ] [ ]

+ y2 (D 2)([ ] ) ,
(99)
(100)
(101)
(102)
(103)
459

0 = y3 3D V f h [ ] + (2p + f ) [ h]

py2 V (D 2)h V + y2 h f (h + 2V )

2(p + f ) V 4q1 y2 3D (D 2) F 6q2 y2 5DF F

1
1
+ p(p + f ) F F + F F h [ ] 2[ h ]
2
4
(int)

(int)
0 0
+ y2 f A0 1 + pB0
0 0 1 4D
(int)

(int)
+ y3 3D f C0
(104)
0 0 1 + pD0 0 0 1 .
One can replace (99) with
b2(int) = 0,
(105)
such that (85) is in fact equivalent to (105) and (100)(101). So far, we have shown that the
second-order deformation of the solution to the master equation, (74), is completely known once
we manage to solve equations (105) and (100)(101). This is our next concern.
(int)
(int)
From (100) we obtain a necessary condition for the existence of b2 and b1 , namely
1 = 2 + 1 + l1 ,
(106)
agh(l1 ) = 1,
pgh(l1 ) = 2.
where agh(2 ) = 2 = pgh(2 ), agh(1 ) = 1 = pgh(1 ),

It is essential
to remark that all the functions 2 , 1 , and l1 must be local since otherwise we cannot obtain
local second-order deformations from (100). Assuming (106) holds, we act with on it and use
its nilpotency and its anticommutation with , which yields

1 = (1 ) + l1 .
(107)
Without entering technical details, we mention that the validity of (107) can be checked by means
of standard cohomological arguments. In fact, after direct manipulations of (103), it can be shown
that (107) (and thus also (106)) requires that the following conditions are simultaneously satisfied:
1
F F + F F = ,
(108)
4
F = ,
(109)
,
[ h] =

[ ]
[ h] h [ h] h = .
(110)
(111)
All the quantities denoted by or must be local; their locality is essential in obtaining local
deformations, which is one of the main working hypotheses of our paper. One can explicitly
reveal locality obstructions to each of these conditions. For instance, assuming that Eq. (108) is
satisfied for some local and taking its divergence, it follows that the relation

1
F F + F F =
(112)
4
should also take place. On the other hand, it is easy to see that

1
F F + F F = V F .
4
(113)
460
Since V F obviously is not a divergence of a local function, Eq. (112) cannot hold for some
local , so neither does (108). Acting in a similar manner with respect to Eq. (109), we infer
F = V
= ( ), such that (109) cannot be satisfied for some local . Related to
h
),
(110), if we apply on it and then take its trace, we obtain [ h] = D2
= (
and hence (110) is not valid for some local . Concerning equation (111), it can be shown
directly that its left-hand side reads as (h h ) + u , with u

= 0 and u
= u1 for
some local u1 , so (111) also fails to be true. Combining these last results, it follows that (107)
(and hence also (106)) cannot hold locally unless 1 = 0, which yields
p(p + f ) = 0,
(2p + f )y3 3D
(114)
= 0,
(115)
(2p + f )y2 = 0.
(116)
There are three relevant solutions to the above equations4

Case I:
p = f
= 0,
y2 = 0 = y3 3D ,
Case II:
p = f = 0,
D = 3,
(118)
Case III:
p = f = 0,
D > 3,
(119)
D > 2,
(117)
which require an individual treatment.

4.3.1. Case I: General Relativity
According to (117), the first-order deformation (69) is parameterized in this situation by four
real constants, namely, f , , q1 3D , and q2 5D . For the sake of simplicity we set f = 1, so
p = 1, such that the S1 (see (70) with the components (40), (41), and (42) plus (71)) takes the
concrete form
(I)
(PF)
(int)
S 1 = S 1 + S1

1
(EH-cubic)
d D x [ ] + h h [ h] + a0
2h
2

1
+ d D x + V V [ ] + 2( V ) h
2

1
2[ h] V + F h 4F h
F
8

D
D
+ q1 3 V F + q2 5
(120)
V F F .
Replacing (117) into (103) and (104), we find that
1 = 0,
0 = 0,
(121)
4 By relevant solution we mean that the resulting deformations lead to a maximum number of consistent couplings
and gauge symmetries. For instance, another possible solution to (117)(119) is p = 0, f
= 0, y2 = 0, y3 3D = 0. This
case is not relevant since it would mean to allow the EinsteinHilbert selfinteractions of the graviton, but forbid: (i) the
standard couplings graviton-photon and (ii) the diffeomorphism sector of the vector field gauge symmetries prescribed
by General Relativity.
461
such that Eqs. (105) and (100)(101) become

(int)
b2 = 0,
(122)
(int)
(int)
b2 + b1 = 1 ,
(int)
b1
(int)
+ b0
(123)
= 0 .
(124)
These equations have already been considered in Section 4.2 at the construction of the first-order
(int)
deformation, so their solutions can be absorbed into S1 from (120) by a suitable redefinition
of the constants p, q1 , and q2 . In conclusion, we can work with
(int)
b2 = 0,
(int)
b1 = 0,
(int)
b0 = 0.
(125)
Inserting the previous results together with (117) for f = 1 in (96), (97), and (98) and then the re(int)
sulting expressions in (91), we complete the interacting component S2 from the second-order
deformation of the solution to the master equation, in agreement with notation (84). Particular(PF)
izing (76) and (78)(80) to the case (117) for f = 1, we also infer S2 with the help of relation
(int)
(PF)
(76). Putting together these expressions of S2 and S2 via formula (74), we can state that the
full second-order deformation to the master equation in case I reads as
S2(I) = S2(PF) + S2(int)

1
(EH-quartic)
S2
+ d D x h h h2
2

1
1
d D x h + V ( V )h + ([ h] )V
2
2

1
1
3
V h[ (] ) V ( [ )h] h h
4
4
4

1
+ F h h [ (] V ) [ h ] V + F h [ ] h V
8

+ V [ h] ([ h] )V

2 [ h] h[ ] V + h [ ] V h[ ] V

1
1
+F
F h h + F h h F h h + F h2 2h h
2
16

1
[ h] V h[ (] V ) h
h [ h] V h[ (] V ) +
4

q1 3D hV F 2h V F + h V F

q2 5D hV F F 4h V F F + 2h V F F .
(126)
The deformation procedure goes on indefinitely in the sense that it produces an infinite number
of nontrivial higher-order components of the deformed solution to the master equation
Sn(I)
= 0,
for all n > 0.
(127)
Nevertheless, we will see in Section 4.4.1 that the first two deformations derived so far for Case I
are enough in order to describe the overall deformed theory at all orders in the coupling constant,
462
which turns out to describe nothing but the standard graviton-vector interactions from General
Relativity.
4.3.2. Case II: New solutions in D = 3
In this situation we substitute (118) into (103) and (104) and obtain that5

1 = 0,
0 = 4y2 q1 F + 3 .
(128)
(int)
(int)
Thus, from (101) we obtain a necessary condition for the existence of b1 and b0 , namely
0 = 1 + 0 + l0 ,
(129)
where agh(1 ) = 1 = pgh(1 ), agh(0 ) = 0 = pgh(0 ), agh(l0 ) = 0, pgh(l0 ) = 1. We insist that
all the quantities 1 , 0 , and l0 from (129) must be local in order to render a local second-order
deformation via (101). This is the second place where we analyze the possible obstructions in
finding local deformations. It is clear from (128) that 0 is a nontrivial element from H 1 ( )
of antighost number zero, 0 = 0, since it is written as 0 = 0M (F )e1M , where 0M are
invariant polynomials not depending on the antifields and e1M are the elements of a basis in
the space of polynomials with pure ghost number one in and . The latter term from the
right-hand side of (128) is derivative-free while the non-vanishing actions of and contain at
least one derivative, so it cannot be written as in (129) and, as a consequence, we must require
y2 = 0. (From the latter definition in (12) we have that ( V ) = , so we can indeed write
= (1 V ). But 1 V is not local, so this solution must be discarded.) Regarding the
former term, proportional with F , since agh(1 ) = 1, it follows that 1 is linear in the
antifields 0 = (h , V ). On behalf of definitions (8), it would produce in (129) terms with
two spacetime derivatives. But F contains only pieces with at most one derivative, so
the locality assumption requires 1 = 0 in (129), such that this becomes
4y2 q1 F = 0 + l0 .
(130)
From definitions (12) it is clear now that (130) cannot hold for some local 0 and l0 . By virtue
of the above discussion we must impose 0 = 0, which is equivalent with the supplementary
conditions
y2 q1 = 0,
y2 = 0,
(131)
displaying two relevant solutions

y2 = 0,
(132)
q1 = 0 = .
(133)
Thus, the second case admits two subcases, deserving separate analyses.
Case II.1 results from (118) and (132), so it corresponds to the choice
D = 3,
p = f = q2 5D = y2 = 0.
(134)
We observe that the deformations lie in three spacetime dimensions and are parameterized by
three constants, namely , y3 , and q1 . Under these circumstances, the first-order deformation S1
5 Note that in D = 3 we have q D = 0.
2 5
463
(see (70) with the components (40), (41), and (42) plus (71), all particularized to (134)) is expressed by
S1(II.1) = S1(PF) + S1(int)

3
2 d x h + d 3 x y3 V [ ] + F [ h] + q1 V F .
(135)
Substituting relations (134) into (103) and (104), we find that
1 = 0,
0 = 0,
(136)
so the discussion from Section 4.3.1 applies here as well and we can take
(int)
b2 = 0,
(int)
b1 = 0,
(int)
b0 = 0
(137)
in (96), (97), and (98). Consequently, with the help of formulas (74), (82) (with the components
(78)(80)), (84), and (91) (with the components (96)(98)) written in the presence of conditions
(134) and (137) we determine the second-order deformation in the form
S2(II.1) = S2(PF) + S2(int)

y32 d 3 x [ h] [ h] + 8y3 q1 d 3 x h + V [ h] .
(138)
(II.1)
Next, we approach the consistency of S2 , i.e. we solve the equation introducing the third-order
deformation of the solution to the master equation
(II.1) (II.1)
(II.1)
S1 , S 2
(139)
+ sS3
= 0.
By direct computation we obtain

(II.1) (II.1)
2
3
[ ]
= s 4y3 q1 d x h h + 48y3 q1 d 3 x .
S1 , S2
(140)
Substituting the last result into (139) we arrive at

(II.1)
2
3
[ ]
+ 4y3 q1 d x h h + 48y3 q1 d 3 x = 0.
s S3
(141)
The last equation possesses local solutions if and only if the integrand of the last term from the
left-hand side of (141) is written in a s-exact modulo d form from local functions. We discussed
a similar term in the beginning of Section 4.3.2 (see the second term on the right-hand side of
(128) and Eq. (129)) and concluded that it cannot be written in a s-exact modulo d form from
local functions until its coefficient vanishes. Then, we can state that (141) holds if and only if
y3 q1 = 0.
(142)
The relevant solutions to the above equation are6

y3
= 0,

= 0,
q1 = 0,
(143)
6 The solution y = 0 and q

= 0 yields no couplings: the original gauge transformations (2) are maintained and two
3
1
gauge-invariant terms are added to the starting Lagrangian (1): 2kh and kq1 3D V F .
464
y3
= 0,
q1
= 0,
= 0.
(144)
Thus, the first subcase from Case II splits again into two complementary situations.
In Case II.1.1, where (134) and (143) hold simultaneously,
p = f = q2 5D = y2 = q1 = 0,
D = 3,
y3
= 0,

= 0,
(145)
we have that the deformed solution to the master equation is parameterized by two constants,
y3 and . Its first two components result from (135) and (138) where we set q1 = 0 and read as

(II.1.1)
= d 3 x 2h + y3 V [ ] + F [ h] ,
S1
(146)

(II.1.1)
= y32 d 3 x [ h] [ h] .
S2
(147)
(II.1.1)
Consequently, (S1
(II.1.1)
sS3
(II.1.1)
, S2
) = 0, so (139) becomes
= 0,
(148)
whose solution can be taken to be trivial

(II.1.1)
S3
=0
(149)
(the solution to the homogeneous equation (148) can be absorbed into (146) by a suitable redefinition of the involved constants). Inserting (149) into the next deformation equation
1 (II.1.1) (II.1.1) (II.1.1) (II.1.1)
(II.1.1)
+ S1
+ sS4
, S2
, S3
=0
S
2 2
(150)
and observing that (S2(II.1.1) , S2(II.1.1) ) = 0, we can again take

(II.1.1)
S4
= 0.
(151)
It is easy to see that in fact we can set

Sn(II.1.1) = 0,
for all n > 2.
(152)
We can therefore conclude that in Case II.1.1, described by conditions (145), the deformation
procedure stops nontrivially at a finite step (n = 2) and the deformed solution to the master
equation, consistent to all orders in the deformation parameter, takes the form
(II.1.1)
(II.1.1)
+ k 2 S2
S (II.1.1) = S + kS1

1
d 3 x L(PF)
F F + h ( ) + V 2kh
0
4

[ ]

+ ky3 V + F [ h] + k 2 y32 [ h] [ h] ,
(PF)
(153)
where L0 is the PauliFierz Lagrangian.

We choose not to expose in detail the remaining possibilities, whose investigation is merely
technical, but simply state their main conclusions. Thus, in Case II.1.2, where (134) and (144)
are assumed to take place concurrently, the deformed solution to the master equation is parameterized by y3 and q1 and starts like in (135) and (138) where we set = 0. There appear no
obstructions in solving the higher-order deformation equations, of order three and four, while that
465
of order five requires the supplementary condition y33 q12 = 0. Its relevant solution is q1 = 0 since
in the opposite situation, y3 = 0, there are no cross-couplings at all between the graviton and the
vector field: the original gauge transformations are not affected and the Lagrangian is modified
by an Abelian ChernSimons term kq1 V F . Based on q1 = 0, it can be shown that all the
deformations of order three or higher can be made to vanish, such that the resulting deformed
solution to the master equation precisely reduces to a particular solution of Case II.1.1: it is
expressed by (153) for = 0. Regarding Case II.2, it is pictured by conditions (118) and (133),
so the deformations live again in a three-dimensional spacetime, being parameterized by y2
and y3 . The first-order deformation reduces to (71) where we set D = 3 and p = 0 = q1 . There
are no obstructions in finding the deformation of order two in the coupling constant, but the
existence of the third-order deformation imposes the additional condition y2 = 0, which further
implies that all the deformations of order three or higher are trivial. Therefore, the fully deformed
solution to the master equation is nothing but the same particular solution from Case II.1.1,
being equal to (153) for = 0.
Combining all the results exposed so far, we can state that the most general solution of the
deformation procedure in Case II is provided by a three-dimensional, consistent solution to the
master equation that stops at the second order in the deformation parameter, is parameterized by
y3 and , and reads as in (153). We will argue in Section 4.4.2 that this solution describes a new
mechanism for coupling a spin-two field to a massless vector field in D = 3, which is completely
different from the standard one, based on General Relativity prescriptions.
4.3.3. Case III: Nothing new
Case III is subject to conditions (119), so it is valid only in D > 3 spacetime dimensions,7
being parameterized in the first instance by y2 , q2 5D , and . In agreement with (119), formulas
(103) and (104) will be

1 = 0,
(154)
0 = 2y2 3q2 5D F F + 2D ,
(int)
(int)
such that (101) yields the same necessary condition for the existence of b1 and b0 like in
Case II
0 = 1 + 0 + l0 ,
(155)
where agh(1 ) = 1 = pgh(1 ), agh(0 ) = 0 = pgh(0 ), agh(l0 ) = 0, pgh(l0 ) = 1. The locality
of the second-order deformation requires that all 1 , 0 , and l0 are local functions. From (154)
and definitions (8) and (12) it is obvious that (155) cannot be satisfied for some local 1 , 0 , and
l0 until we set 0 = 0, which further demands

y2 q2 5D = 0,
y2 = 0.
(156)
There are obviously two complementary solutions to these equations

q2 5D = 0,
= 0,
y2 = 0.
(157)
(158)
Once more, we try to simplify the presentation by avoiding the technical details involved
and mentioning only the key points. Case III.1, described by (119) and (157), is parameterized
7 Note that D > 3 implies automatically y D = 0 = q D .
3 3
1 3
466
by a single constant, y2 , such that the first-order deformation reduces to the component of (71)
proportional with this parameter

(III.1)
S1
(159)
= y2 d D x h + (D 2) V + V [ h] .
(III.1)
The second-order deformation, S2

, is then easily obtained from the observation that 1 = 0 =
0 , so Eqs. (105) and (100)(101) reduce, like in Case I, to (122)(124), whose solution can be
taken to vanish, like in (125). Consequently, the non-integrated density of S2(III.1) contains only
terms of antighost number zero and reduces to the integrand of (78) plus the terms proportional
with y22 from (98). It is easy to show that the existence of a local third-order deformation requires
(III.1)
y2 = 0, so Case III.1 leads to no nontrivial deformations, Sn
= 0, for all n 1. Case III.2,
pictured by (119) and (158), is parameterized by q2 5D and (see also footnote 7). Only the
first-order deformation is found nontrivial, being equal to

(III.2)
S1
(160)
= d D x 2h + q2 5D V F F .
Analyzing (160), we can state that Case III.2 is not interesting since the deformation procedure
does not modify the original gauge transformations (2), but mainly adds to the original action (1)
two gauge-invariant terms: a cosmological one and a generalized Abelian ChernSimons action.
In conclusion, Case III brings no new information on the possible couplings between a spin-two
field and a massless one-form.
4.4. Analysis of the deformed theory
The main aim of this section is to give an appropriate interpretation of the Lagrangian formulation of the deformed theories obtained previously from the deformation of the solution to
the master equation. We will analyze the first two cases separately since we have seen that the
third one gives nothing interesting. It is useful to recall the relationship between some quantities
appearing in the deformed solution of the master equation, S, and the associated gauge theory:
the component of antighost number zero from the former is nothing but the Lagrangian action
of the coupled model, the piece of antighost number one provides the gauge transformations of
the deformed theory, and the terms of antighost number two contain the structure functions defined by the commutators among the deformed gauge transformations. More precisely, the gauge
transformations of the coupled theory result from the terms of antighost number one present in S
(generically written as 0 Z 0 1 1 ) by replacing the ghosts with the gauge parameters
1 ,
0 = Z 0 1
1 . The functions
Z 0 1 = Z 0 1 + kZ1 0 1 + k 2 Z2 0 1 +
(161)
define the gauge generators of the coupled model, where the components Z 0 1 are responsible
for the original gauge transformations.
4.4.1. Case I: Standard couplings from General Relativity
We discussed in detail in Section 4.3.1 a first case of obtaining consistent interactions between
a PauliFierz field and a vector field. This is defined by conditions (117), in which situation the
deformed solution to the master equation starts like
S (I) = S + kS1 + k 2 S2 +
(PF)
(PF)
(int)
(int)
+ k 2 S 2 + S2
+ ,
= S + k S1 + S1
(I)
467
(I)
(162)
where S,
and
read as in (17), (120), and (126) respectively.
In order to identify the main ingredients of the coupled model in the first case we use the
result proved in Section 5 of [15], according to which the local BRST cohomologies of the
PauliFierz model and of the linearized version of vielbein formulation of spin-two field theory
are isomorphic. Because the local BRST cohomology (in ghost numbers zero and one) controls
the deformation procedure, it results that this isomorphism allows one to pass in a consistent
manner from the PauliFierz model to the linearized version of the vielbein formulation and
conversely during the deformation procedure. Nevertheless, the linearized vielbein formulation
possesses more fields (the antisymmetric part of the linearized vielbein) and more gauge parameters (Lorentz parameters) than the PauliFierz model. The switch from the former version to the
latter is realized via the above mentioned isomorphism by imposing some partial gauge-fixing
conditions, chosen to annihilate the antisymmetric components of the vielbein. An appropriate
interpretation of the Lagrangian description of the interacting theory in Case I requires the generalized expression of these partial gauge-fixing conditions [33]
(I)
S1 ,
(I)
S2
[a eb] = 0
(163)
a up to the second order in the coupling

and the development of the vielbein ea and of its inverse e
constant in terms of the PauliFierz field
k
3k 2
(0)
(1)
(2)
+
h h + ,
ea = ea + k ea + k 2 ea + = a h
(164)
2 a
8 a
k
k2
(0)
(1)
(2)
a
= ea + k ea + k 2 ea + = a + ha ha h + .
e
(165)
2
8
The expansion of the inverse of the metric

tensor g and of the square root from the minus
determinant of the metric tensor g = det g in terms of the PauliFierz field,

(0)
(1)
(2)
g = g + k g + k 2 g + = kh + k 2 h
h + ,
(0)
(1)
(2)
g = g + k g + k 2 g +

k
k2 2
=1+ h+
h 2h h + ,
2
8
will also be necessary in what follows. We note that the metric tensor is
g = + kh .
(166)
(167)
(168)
(int)
The interacting Lagrangian at order one in the coupling constant, L1 , is the non-integrated
density of the piece of antighost number zero from the first-order deformation in the interacting
sector, S1(int) . Using (120) and expansions (164)(167), we can write
(int)
L1

1
1
1
= F [ h] V F F h + F F h
4
8
2
+ q1 3D V F + q2 5D V F F

(0)
1 (0) (0) (1) (0) (1) (0) (1) (0) (0) (0)
g g g + g g g + g g g
F F
4
468

(0) (1)
(0) (0) (0) (1) (0)
g g g
F F + F F
(0) (0)
(0) (0) (0) (0)
+ q1 3D g ea11 ea22 ea33 a1 a2 a3 V1 F2 3
(0)
(0) (0)
(0) (0)
(0)
+ q2 5D g ea11 ea55 a1 a2 a3 a4 a5 V1 F2 3 F4 5 ,
(169)
where
(0)

(0)
a
Va ,
F = [ e]
(0)
(0)
V = ea Va ,
(1)

(1)
a
Va .
F = [ e]
(int)
(170)
(int)
Along the same line, the interacting Lagrangian at order two, L2 , results from S2 at antighost
number zero. Taking into account formula (126) and expansions (164)(167), we have that
(2)
(2) (0)
1 (0) (0) (0) (0)
g g g
F F + F F
4
(0) (0) (1)
(1) (0)
(0) (1) (0)
(1) (0) (0) (0) (1)
+
g g g + g g g + g g g F F + F F
(0) (0) (2)
(0) (2) (0)
(0) (1) (1)
g g g + g g g + g g g
+
(1) (0) (1)
(1) (1) (0)
(2) (0) (0) (0) (0)
+ g g g + g g g + g g g F F
(1) (0) (0) (0) (0) (0)
+ q1 3D a1 a2 a3 g ea1 ea2 ea3 V1 F2 3
L2(int)
(0) (0)
(1) (0)
(1) (0) (0) (0) (0)
(0) (0) (1)
(0) (0) (0)
+ g ea11 ea22 ea33 V1 F2 3 + 2ea11 ea22 ea33 V1 F2 3 + ea11 ea22 ea33 V1 F2 3

(1) (0)
(0)
(0) (1)
(0) (0)
(0) (0) (0)
(0)
+ ea11 ea22 ea33 V1 F2 3 + q2 5D a1 a2 a3 a4 a5 g ea11 ea55 V1 F2 3 F4 5
(0)
(0)
(0) (0)
(0) (0)
(0) (1) (0)
(0)
(0)
(0) (1)
+ g ea11 ea22 ea55 V1 F2 3 F4 5 + 4ea11 ea44 ea55 V1 F2 3 F4 5

(0)
(1)
(1) (0)
(0) (0)
(0)
(0)
(0)
(0)
+ ea11 ea55 V1 F2 3 F4 5 + 2ea11 ea55 V1 F2 3 F4 5 ,
(171)
(0)
with
(1)
(1)
V = ea Va ,
(2)

(2)
a
Va .
F = [ e]
(int)
(int)
From the expressions of L1 and L2

interacting Lagrangian in Case I
(int)
LI
(vect)
= L0
(int)
+ kL1
(int)
+ k 2 L2
(172)
, we observe that the first three terms from the full
(173)
coincide with the first orders of the Lagrangian describing the standard vector field-graviton
cross-couplings from General Relativity

1
gg g F F + k q1 3D 1 2 3 V1 F2 3
4

+ q2 5D 1 2 3 4 5 V1 F2 3 F4 5 ,
L(vector-graviton) =
(174)
469
where the fully deformed field strength F and the Levi-Civita symbol with curved indices
1 ...D are given by
(1)
(2)
a (0)
Va F + k F + k 2 F +
F = [ e]
(0)

(1)

(2)

a
a
a
= [ e]
Va + k[ e]
Va + k 2 [ e]
Va + ,
1 ...D =
gea11 eaDD a1 ...aD .
(175)
(176)
The self-interactions of the PauliFierz field at orders one and two in the coupling constant,
(PF)
(PF)
(PF)
L1,2 , result from the terms of antighost number zero present in S1 and S2 (see (120) and
(126)), so the full Lagrangian describing the self-interactions of the graviton in Case I starts like
(PF)
(PF)
(PF)
L (PF)
= L0 + kL1 + k 2 L2 + ,
I
(177)
(PF)
where L0 is the PauliFierz Lagrangian. Using (166)(168), one finds that the first three terms
(PF)
from L I are nothing but the first orders of the EinsteinHilbert Lagrangian with a cosmological
term [14]

2
(178)
g R 2k 2 ,
2
k
where R is the full scalar curvature.
As explained in the beginning of this section, the terms present in (162) (see (17), (120), and
(126)) that are linear in the antifields V provide the deformed gauge transformations of the
vector field

k 3k 2
(I)
h h +
V = h +
2
8

1
1
k
1
+ [
] + k ([ h] )
+ h [ ]
+ (
[ )h] V +
2
4
8
8

2
3
k 3k
+ ( V ) k h +
(179)
h h +
.
2
8
L(EH) =
In the last formula the indices of the one-form, even if written in Latin letters, are flat. In standard,
Latin notation the above gauge transformations can be written as
(0)
(1)
(2)
(I) Va =
Va + k
Va + k 2
Va + ,
where the first orders of the gauge transformations read as
(0)
(0)
(1)
(1)
(0)
(0)
(2)
(2)
(1)
(1)
Va = ea
,
(180)
Va = ea
+
ab V b + ( Va )
,
Va = ea
+
ab V b + ( Va )

(181)
(182)
and the various orders of the gauge parameters are expressed by

(0)
=

a a ,
=
a h
a,
2
(1)
(183)
470
ab = [a
b] ,
(184)
2
1
1
1
(1)
ab =
c [a hb]c + hc[a b]
c + (c
[a )hcb] .
(185)
4
8
8
Based on the above notations, we can re-write the gauge transformations of the vector field with
a flat index as
(0)

(0)

(1)
(1)
(I) Va = ea + k ea +
+ k
ab + k
ab + V b
(0)

(1)
+ k( Va )
+ k
+ .
(186)
(0)
(0)
(1)
The gauge parameters

ab and
ab are precisely the first two terms from the Lorentz parameters
expressed in terms of the flat parameters
a via the partial gauge-fixing (163). Indeed, (163) leads
to

[a eb] = 0.
(187)
Using
ea =
ea ea
+
a b eb
and inserting (164) together with the expansions

(0)
(1)
k
=
+ k
+ = a ha +
a ,
2
(0)
(1)
ab =
ab + k
ab + ,
(188)
(189)
(190)
in (187), we arrive precisely at (184) and (185). At this point it is easy to see that the first orders
of the gauge transformations (186) coincide with those arising from the perturbative expansion
of the formula
(I) Va = ea
+ k
ab V b + k( Va )
.
(191)
Concerning the vector field with a curved index V

a
V = e
Va ,
(192)
its gauge transformations will be correctly described at the level of the first orders in the coupling
constant by the well-known gauge transformations

(I) V =
+ k
V + k( V )

(193)
of the vector field (in interaction with the EinsteinHilbert graviton) from General Relativity.
Finally, from the terms present in (162) linear in the PauliFierz antifields h (see (17), (120),
and (126)) one infers that the deformed gauge transformations of the metric tensor (168) reproduce the first orders of diffeomorphisms
(I) g = k
(;) ,
(194)
where
; is the (full) covariant derivative of
.
So far, we argued that in the first case the consistent interactions between a graviton and a
vector field are described in all D > 2 dimensions by the first orders of the Lagrangian and gauge
transformations prescribed by the standard rules of General Relativity (see (174), (178), (193),
471
and (194)). Our result follows as a consequence of applying a cohomological procedure based on
the free BRST symmetry in the presence of a few natural assumptions: locality, smoothness in
the coupling constant, Poincar invariance, Lorentz covariance, and preservation of the number
of derivatives on each field. General covariance was not imposed a priori, but was gained in
a natural way from the cohomological setting developed here under the previously mentioned
hypotheses. It can be shown that formulas (174), (178), (193), and (194) apply in fact to all
orders in the coupling constant, so we can state that the fully interacting Lagrangian action in
Case I reads as

1
2
L(I)
D
gg g F F
S [g , V ] = d x 2 g R 2k 2
4
k

D

D 1 2 3 4 5
1
2
3
+ k q1 3
V1 F2 3 + q2 5
V1 F2 3 F4 5
(195)
and is invariant under the deformed gauge transformations

(I) g = k
(;) ,
(I) V =
+ k
V + k( V )
.
(196)
The validity of (195) and (196) to all orders in the coupling constant can be done by developing
the same technique used in Section 7 of [14].
4.4.2. Case II: New couplings in D = 3
As discussed in Section 4.3.2, the second case of interest allowing for nontrivial, consistent
couplings between a PauliFierz field and a vector field is pictured by the deformed solution to
the master equation given in (153). We can re-write the deformation in a more convenient way by
adding to (153) some s-exact terms, since we know that this does not affect the physical content
of the coupled model (see (24)). Because the most general couplings in Case II are obtained in
Case II.1.1, described by conditions (145), we will denote the deformed solution (153) to which
we add the previously mentioned s-exact terms and where we set y3 = 1 by S (II)

S (II) S (II.1.1) y =1 s 2k 2 d 3 x h h +
3

1
= d 3 x L(PF)
F F 2kh kF [ h] + 2k 2 [ h] [ h]
0
4

+ h ( ) + V + k [ ] .
(197)
Essentially, it is not trivial and is consistent to all orders in the coupling constant, namely
(II) (II)
S ,S
(198)
= 0.
From the terms of antighost number zero we deduce the Lagrangian action of the coupled model

1
S L(II) [h , V ] = d 3 x L(PF)
F F 2kh
0
4

kF [ h] + 2k 2 [ h] [ h] ,
(199)
(PF)
where L0 is the PauliFierz Lagrangian and is the cosmological constant. The component
of antighost number one provides the gauge symmetries of (199) (see the discussion from the
472
preamble of Section 4.4)
(II) h = (
) ,
(II) V =
+ k [
] .
(200)
The absence of terms of antighost number strictly greater than one shows that the above gauge
transformations are independent (irreducible) and their algebra remains Abelian, like the original
one. Action (199) can be set in a more suggestive form by introducing a deformed field strength

F
= F + 2k [ h] ,
in terms of which we can write

1
(PF)
S L(II) [h , V ] = d 3 x L0 2kh F
F
.
4
(201)
(202)
Under this form, action (202) is manifestly invariant under the gauge transformations (200):
its first two terms are known to be invariant under linearized diffeomorphisms and the third is
gauge-invariant under (200) since the deformed field strength is so
(II) F
= 0.
(203)
This result is new and will be generalized in Section 6 to the case of couplings between a
graviton and an arbitrary p-form. In conclusion, this case yields another possibility to establish nontrivial couplings between the PauliFierz field and a vector field. It is complementary
to Case I (General Relativity) and is valid only in D = 3. The resulting Lagrangian action and
gauge transformations are not series in the coupling constant. The Lagrangian contains pieces
of maximum order two in the coupling constant, which are mixing-component terms (there is
no interaction vertex at least cubic in the fields) and emphasize the deformation of the standard
Abelian field strength of the vector field like in (201). Concerning the new gauge transformations, only those of the massless vector field are modified at order one in the coupling constant
by adding to the original U (1) gauge symmetry a term linear in the antisymmetric first-order
derivatives of the PauliFierz gauge parameters. As a consequence, the gauge algebra, defined
by the commutators among the deformed gauge transformations, remains Abelian, just like for
the free theory. We cannot stress enough that these two Cases (I and II) cannot coexist, even in
D = 3, due to the consistency conditions (114)(116).
5. No cross-couplings in multi-graviton theories intermediated by a vector field
As it has been proved in [14], there are no direct cross-couplings that can be introduced among
a finite collection of gravitons and also no cross-couplings among different gravitons intermediated by a scalar field. Similar conclusions have been drawn in [15,16] related to the couplings
between a finite collection of spin-two fields and a Dirac or a massive RaritaSchwinger field. In
this section, under the same hypotheses like before, namely, locality, smoothness in the coupling
constant, Poincar invariance, Lorentz covariance, and preservation of the number of derivatives
on each field, we investigate the existence of cross-couplings among different gravitons intermediated by a massless vector field. The Greek field indices are (Lorentz) flat: they are lowered and
raised with a flat metric of mostly plus signature, = ( + +).
473
5.1. First- and second-order deformations. Consistency conditions

5.1.1. Generalities
We start now from a finite sum of PauliFierz actions and a single Maxwell action in D > 2

S0L hA
, V =

1
A
d D x hA
hA + hA h
2

1

1
A
A
h hA + h hA F F
,
2
4
(204)
where hA is the trace of the PauliFierz field hA (hA = hA ), with A = 1, n and n > 1. The
collection indices A, B, etc., are raised and lowered with a quadratic form kAB that determines
a positively-defined metric in the internal space. It can always be normalized to AB by a simple
linear field redefinition, so from now on we take kAB = AB and rewrite (204) as
n

(PF)

(vect)
A
S0L hA
(205)
dDx
L0 hA
,
, V =
, h + L0
A=1
(PF)
A
where L0 (hA
, h ) is the PauliFierz Lagrangian for the graviton A. Action (204) is invariant under the gauge transformations
A
hA
= (
) ,
V =
.
(206)
The BRST complex comprises the fields, ghosts, and antifields

0 = hA
, V ,

0 = hA , V ,
A
1 =
, ,

1 = A , ,
(207)
(208)
whose degrees are the same like in the case of a single PauliFierz field. The BRST differential
decomposes exactly like in (5) and its components act on the BRST generators via the relations
hA
V = F ,
= 2HA ,
A = 2 hA ,
0 = 0,
1
= 0,
A
hA
= ( ) ,
A

= 0,
= V
= 0,
= 0,
V = ,
= 0,
(209)
,
(210)
(211)
(212)
(213)
(214)
where HA = KA 12 KA is the linearized Einstein tensor of the PauliFierz field hA . The

solution to the master equation for this free model takes the simple form

L A
A
S = S0 h , V + d D x hA ( )
(215)
+ V .
474
5.1.2. First-order deformation

The first-order deformation of the solution to the master equation decomposes like in the case
of a single graviton in a sum of three independent components
a = a (PF) + a (int) + a (vect) .
(216)
The first-order deformation in the PauliFierz sector, a (PF) , can be shown to expand as
a (PF) = a 2(PF) + a 1(PF) + a 0(PF) ,
(217)
where
(PF)
a 2
1 A B
C
= fBC
A [ ]
,
2
(218)
A some real constants. The requirement that a

2(PF) produces a consistent a 1(PF) as solution
with fBC
(PF)
(PF)
(PF)
A to be symmetric with
1
restricts the coefficients fBC
to the equation a 2 + a 1 = m
A ) [14]8
respect to their lower indices (commutativity of the algebra defined by fBC
A
A
fBC
= fCB
.
(219)
Based on (219), it follows that

A
B
C
a 1(PF) = fBC
B hC
hA
[ h] .
(PF)
(PF)
(220)
(PF)
(PF)
Asking that a 1
provides a consistent a 0
as solution to the equation a 1 + a 0 =
(PF)
D
D
m
0
further constrains the coefficients with lowered indices, fABC = kAD fBC
AD fBC ,
9
to be fully symmetric [14]
1
fABC = f(ABC) .
3
(221)
(PF)
From (221) we obtain that a 0

coefficients fABC by aabc )
(PF)
a 0
(cubic)ABC
= fABC a 0
coincides with that from [14] (where it is denoted by a0 and the
2A hA ,
(222)
(cubic)ABC
where a 0
contains only vertices that are cubic in the PauliFierz fields and reduce
to the cubic EinsteinHilbert vertex in the absence of collection indices. A play the role of
cosmological constants. Employing exactly the same line like in Section 4.2, we find that the
first-order deformation giving the cross-couplings between the gravitons and the vector fields
ends at antighost number one
(int)
a (int) = a 1
(int)
+ a 0
(223)
where

a 1(int) = y2A hA (D 2)V A

+ y3A 3D V [ A + pA V F A ,
]
8 The term (218) differs from that corresponding to [14] through a -exact term, which does not affect (219).
9 The piece (220) differs from that corresponding to [14] through a -exact term, which does not change (221).
(224)

(int)
a 0
A D
[
= (D 2)y2A V [ hA
] + y3 3 F hA

pA
1
+
F hA
F F hA
+ F
2
4
475
(225)
and y2A , y3A together with pA are some arbitrary, real constants. Like in Section 4.2, we eliminate
some s-exact modulo d terms from a (int) and work with

1 A
(int)
(int)
A
a
(226)
= a
+ s pA V + V V h
t .
2
The component a (vect) coincides with that from Section 4.2 (see (46))
a (vect) = a (vect) = q1 3D V F + q2 5D V F F .
(227)
Putting together (217) and (223)(227) with the help of (216), we can write the first-order
deformation of the solution to the master for a single vector field and a collection of PauliFierz
fields like
(PF)
(int)
S1 = S1 + S1 ,
where
(PF)
(PF)
(PF)
d D x a 2 + a 1 + a 0

1 A B
D
C
A
B
C
B hC
A [ ]
+ fBC
hA
= d x fBC
[ h]
2

+ fABC a 0(cubic)ABC 2A hA ,
(PF)
S1
(228)
(229)

d D x a (int) + a (vect)

A
= d D x y2A hA + (D 2) V A + V [ h]
(int)
S1

]
]
+ y3A 3D V [ A + F [ hA

1
A
A
V V [ ]
+ 2( V )A hA
+ pA

2

A

1
A
A
2[ h] V + F h 4F h
+ F
8

D
D
+ q1 3 V F + q2 5
V F F .
(230)
A , , y , y AD ,
It is parameterized by seven types of real, constant coefficients, namely fBC
A 2A 3 3
D
D
A
pA , q1 3 , and q2 5 , with fBC fully symmetric (see (221)).
5.1.3. Consistency of the first-order deformation

Next, we investigate the consistency of the first-order deformation, expressed by Eq. (22),
with S1,2 replaced by S1,2
(S1 , S1 ) + 2s S2 = 0.
(231)
476
We decompose the second-order deformation as

(PF)
(int)
S2 = S2 + S2 ,
(232)
(PF)
(int)
where S2 is responsible only for the self-interactions of the PauliFierz fields and S2 for
the cross-couplings between the gravitons and the vector field. Using (228), we find that (231)
becomes equivalent with two independent equations
(PF) (PF) (int) (int) (PF)
(PF)
S1 , S1
(233)
+ S1 , S1
+ 2s S2 = 0,
(int)
2 S1 , S1
(234)
+ 2s S2 = 0,
+ S1 , S1
(int) (int)
where (S1 , S1 )(PF) contains only PauliFierz BRST generators and each term of
(int)
(int)
(S1 , S1 )(int) includes at least one BRST generator from the Maxwell sector.
Initially, we analyze the existence of S2(PF) , governed by Eq. (233). By direct computation we
find
(int) (int) (PF)
S1 , S1

(D 2)2 A B
= 2s d D x y2A y2B
h h hA hB
4

]
]
+ y2A y3B 3D (D 2) hA [ hB + y3A y3B 3D [ hA [ hB
]

2s S2(PF) (y2A y2B ) + S2(PF) y2A y3B + S2(PF) y3A y3B ,
(235)
where

(D 2)2
(PF)
d D x hA hB hA hB
S2 (y2A y2B ) = y2A y2B
,
4

]
(PF)
B
B D
y2A y3 = y2A y3 3 (D 2) d D x hA [ hB ,
S2

]
(PF) A B
y3 y3 = y3A y3B 3D d D x [ hA [ hB
S2
] .
Replacing (235) into (233), it becomes equivalent to

(PF)

(PF) (PF)
(PF)
(PF)
(PF) A B
+ 2s S
y2A y3B S
y3 y3 = 0,
, S
S
(y2A y2B ) S
S
1
(236)
(237)
(238)
(239)
(PF)
(PF) (PF)
(PF)
so the existence of S2 requires that (S1 , S1 ) is s-exact, where S1 reads as in (229). It
C to satisfy
has been shown in [14] (Section 5.4) that this requirement restricts the coefficients fAB
the supplementary conditions
D
E
fA[B
fC]D
= 0.
(240)
C define the structure

Combining (219), (221), and (240), we conclude that the coefficients fAB
constants of a real, commutative, symmetric, and associative (finite-dimensional) algebra. The
analysis realized in [14] (Section 6) shows that such an algebra has a trivial structure: it is a
C = 0 whenever two indices are different
direct sum of one-dimensional ideals. Therefore, fAB
C
fAB
= 0,
if (A
= B
or B
= C
or
C
= A).
(241)
477
For notational simplicity, we denote fABC for A = B = C by

fAAA fA
(242)
without summation over A.
Using (241), it follows that (S1 , S1 ) cannot couple different gravitons: it will be written as
a sum of s-exact terms, each term involving a single graviton
n

(PF) (PF)
A 2
1
(EH
quartic)A
D
A
A
S1 , S1
= 2s
f A fA S 2
+ A d x h
h h
2
(PF)
2s
(PF)
A=1
n

(PF) 2
S2
fA , f A A .
(243)
A=1
(EH-quartic)A
Each S2
is the second-order EinsteinHilbert deformation in the sector of the graviton A. It includes the quartic EinsteinHilbert Lagrangian for the field hA
and is written only
A
A
in terms of the BRST generators from the A sector, namely h , , and their antifields. Also,
(PF)
it is important to note that (241) restricts S1 to have the same property (see (229)) of being
written as a sum of individual components, each component involving a single graviton sector

n

1
A
S1(PF) =
+ hA A hA
fA d D x A A [ ]
2
A=1

n
(EH-cubic)A
D
A
d
+
a
A [ hA
x
h
(244)
.
A
]
0
A=1
Now, a 0(EH-cubic)A is nothing but the cubic EinsteinHilbert Lagrangian involving only the graviton field hA
. Substituting (243) into (239) we find the equation

n

(PF)
(PF)
(PF)
(PF)
(PF)
s S
y2A y3B S
y3A y3B
fA2 , fA A = 0,
S
(y2A y2B ) S
S
2
A=1
(245)
s S2(PF)
whose solution reads as (up to the solution of the homogeneous equation,

be incorporated into (244) by a suitable redefinition of the constants involved)
= 0, which can
n

(PF)
(PF)
(PF)
(PF) A B
(PF) 2
y2A y3B + S2
y3 y3 +
fA , f A A .
S2 = S2 (y2A y2B ) + S2
S2
(246)
A=1
Inspecting (244) and (246), we observe that the latter component contains at this stage three
pieces that mix different graviton sectors, namely those proportional with yiA yj B for i, j = 2, 3
and A
= B.
Next, we approach the solution S2(int) to Eq. (234). We act like in Section 4.3. If we make the
notations

(int)
2 S1(PF) , S1(int) + S1(int) , S1(int)
(247)
d D x (int) ,

(int)
S2 d D x b (int) ,
(248)
478
then Eq. (234) takes the local form

(int) = 2s b (int) + n .
(249)
Developing (int) according to the antighost number, we obtain that

(int) =
2
I(int) ,

agh I(int) = I,
I = 0, 2,
(250)
I =0
with

(int)
2 = pA pB A hB
A
C
B
+ w 2 ,
fAB pC + pA pB V [ ]
(251)

A

C
(int)
B
1 = pA pB A hB
fAB pC + pA pB V [ ]

B
1
( V )hA B + [ hA
+ pA pB V
] V
2

1
B 3 A B
1 A
B
A
V h[ ] V [ h] h h
4
4
4

B

1 C
A
B
A
B
+ fAB
pC + pA pB V V [ hA
hA
h
] + [ h]
2

A

C A
B
C
3D V y3C fAB
[ hB
h [ ]
+ 2y3B pA + y3C fAB
]

B
AB V
+ y2A pB V (D 2)hA

B

A
C
B
C
B
y2C fAB h + 2V 2 y2A pB + y2C fAB V
h
C

B
fAB pC + pA pB V F A [ ]
+ V y3A pB + y3B pA
D

C
A
3
[ ]
[ B] + y2A pB + y2B pA
+ y3C fAB

C
A B
(D 2) [ ]
+ w 1 ,
+ y2C fAB
(252)

B
1
(int)
0 = pA pB V ( V )hA B + [ hA
] V
2

B 3 A B
1
1
A
A
h] h h
V h[ ] B V [
4
4
4

B
1 C
A
+ fAB pC + pA pB V V [ hA
] + [ h]
2

16
A
B
A
B
D A
h h +
y3A q1 3 h
D2

[ A] B
pA pB [ A]
+
[ hB
h[ ] V
V h
] V 2 h
8

[ ] B
V h[ ] V + F hA hB [ (] V ) [ hB ] V
+ hA

1
A
B
B
F hA hB
F hA
+ F h [ ] h V + pA pB F
h +
16
479

A
B
[ hB
2hA hB
h
] V h[ (] V )

1

B
1 A B
A
V
h
(
V
)
h
F h h F hA hB +
[ hA
]
[
2
4

1 C
1
B
+ fAB
pC + pA pB F F + F F hA
h
4
4

A
+ q1 3D pA hA V F 2hA
V F + h V F

A
+ q2 5D pA hA V F F 4h V F F + 2hA
V F F

A
D
A
16y3A q1 3 V [ h] (D 2)(D 1)y2A y2 V V
4q1 3D y2A (D 2) F A 6q2 5D y2A F F A

1 C
1
+ fAB pC + pA pB F F + F F hA [ B] 2[ hA ] B
2
4

A
A (int)BC
+ y2A 4D + fBC A0
0 0 1
A (int)BC

(int)AB
+ pB B 0
0 0 1 + y3A 3D fBC
0 0 1
C 0
(int)AB
+ pB D 0
(253)
0 0 1 + w 0 .
, B 0
, C 0
, and D 0
are linear in their arguments,
In (253) the functions A 0
just like in (90).
Acting exactly like in the case of a single graviton (see Section 4.3), we deduce that b (int) and
n from (249) can be taken to stop at antighost number two and one respectively
(int)BC
b (int) =
2
(int)
bI ,
(int)AB
(int)
= I,
agh bI
(int)BC
(int)AB
I = 0, 2,
(254)
I =0
n =
1
n I ,

agh n I = I,
I = 0, 1.
(255)
I =0
If we make the notations

A

C
1
B
(int) ,
b2(int) = pA pB A hB
fAB pC + pA pB V [ ] + b2
2
(256)

B

b(int) = pA pB V ( V )hA B + 1 [ hA
] V
1
2
2

B 3 A B
1
1 A
B
A
V h[ ] V [ h] h h
4
4
4

B
1 C
A
B
A
B
fAB pC + pA pB V V [ hA
hA
h
] + [ h]
4
A

1 D
C A
B
C
[ hB
h [ ]
+ 2y3B pA + y3C fAB
+ 3 V y3C fAB
]
2

1
B
AB V
y2A pB V (D 2)hA

2
480

B

1
C
C
h + 2V B 2 y2A pB + y2C fAB
V B
+ hA y2C fAB
2
8
(int)
y3A q1 3D hA + b1 ,
D2
(257)

[ A] B
pA pB [ A]
(int)
[ hB
h[ ] V
V h
b0 =
] V 2 h
16

[ ] B
+ hA
V h[ ] V + F hA hB [ (] V ) [ hB ] V

+ F hA [ ] hB
V

pA pB
1
B
F
F hA hB 2hA hB
F hA
h +
2
16

1 A B
B
hA
[ hB
F h h F hA hB
] V h[ ] V +
2

B
1
A
A
+
[ h] V h[ ] V h
4

1 C
1
B
fAB
pC + pA pB F F + F F hA
h
8
4

pA
q1 3D hA V F 2hA
V F + h V F
2

pA
A
q2 5D hA V F F 4h V F F + 2hA
V F F
2

1
(int)
+ 8y3A q1 3D V [ hA
(258)
] + (D 2)(D 1) y2A y2 V V + b0
2
and take into account expansions (254)(255) and (5), then Eq. (249) becomes equivalent with
the tower of equations
(int)
b2
= 0,
(259)
(int)
+ b1 = 1 + 1 ,
2
1
(int)
(int)
b1 + b0 = 0 + 0 ,
2
(int)
b2
(260)
(261)
where I = 12 (w I n I ) and
C

B
pC + pA pB F A [ ]
1 = V fAB
D

C
A
3
[ ]
[ B]
+ y3A pB + y3B pA + y3C fAB

C
A B
(D 2) [ ]
,
+ y2A pB + y2B pA + y2C fAB
(262)

A

C A
B
C
[ hB
h [ ]
+ 2y3B pA + y3C fAB
0 = 3D V y3C fAB
]

B
AB V
y2A pB V (D 2)hA

B

C
C
h + 2V B 2 y2A pB + y2C fAB
V B
+ hA y2C fAB
4q1 y2A 3D (D 2) F A 6q2 y2A 5D F F A

1 C
1
+ fAB pC + pA pB F F + F F hA [ B] 2[ hA ] B
2
4
481

(int)AB
A (int)BC
A0
0 0 1 + pB B 0
0 0 1
+ y2A 4DA + fBC
A (int)BC

+ y3A 3D fBC
(263)
0 0 1 + pB D 0(int)AB 0 0 1 .
C 0
The component S2(int) , given by (248), is thus completely determined once we compute b (int) ,
(int)
which expands as in (254). The only unknown components from b (int) are (bI )I =0,2 appearing in formulas (256)(258). They are subject to Eqs. (259)(261). In conclusion, the final step
(int)
needed in order to construct S2 is to solve Eqs. (259)(261).
(int)
(int)
Related to Eq. (260), we observe that the existence of b2 and b1 requires that (262) must
be written as
1 = 2 + 1 + l1 ,
(264)
where 2 , 1 , and l1 exhibit the same properties like the corresponding unhatted quantities from
(106). We require that the second-order deformation is local, so 2 , 1 , and l1 must be local
functions. Assuming (264) is fulfilled, we apply on it and find the necessary condition

1 = ( 1 ) + l1 .
(265)
We do not insist on the investigation of Eq. (265), which can be done by standard cohomological techniques, but simply state that it can be shown to hold if the following conditions are
simultaneously satisfied
1
F F + F F = ,
4
F = ,
[ hA
]
A
=
,

[ A] B
[ B]
[ hA
h = AB .
h
] h
(266)
(267)
(268)
(269)
All the quantities denoted by or must be local in order to produce local deformations. It is
easy to see, by arguments similar to those exposed in the end of the preamble of Section 4.3, that
none of Eqs. (266)(269) is fulfilled (for local functions), so (265), and therefore (264), cannot
hold unless
1 = 0,
(270)
which further implies the following equations

C
fAB
pC + pA pB = 0,

C
y3C 3D = 0,
pA y3B + pB y3A + fAB
(272)
C
y2C = 0.
pA y2B + pB y2A + fAB
(273)
(271)
C are not arbitrary. They have been restricted previously to define

We recall that the constants fAB
the structure constants of a real, commutative, symmetric, and associative (finite-dimensional)
algebra, so in addition they satisfy relations (241).
Let us analyze briefly the solutions to (271)(273). Taking into account (241) and recalling
(242), Eqs. (271)(273) become equivalent to
482
pA pB = 0,
for all A
= B,
(274)
for all A
= B,
(275)
pA y2B + pB y2A = 0,
for all A
= B,
(276)
pA (fA + pA ) = 0,
without summation over A,
(277)
(278)
(279)
(pA y3B + pB y3A )3D
(fA + 2pA )y3A 3D
= 0,
= 0,
(fA + 2pA )y2A = 0,
Unlike Section 4.3, where we searched only the solutions relevant from the point of view of
deformations, here we must discuss all the solutions, since our aim is to see whether they allow
or not cross-couplings among different gravitons. Inspecting (274)(279), we observe that there
appear two complementary cases related to the pA s: either at least one is non-vanishing, say p1 ,
or all the pA s vanish. In Case I
p1
= 0,
(280)
so from (277) for A = 1 it follows that at least f1 is non-vanishing

f1 = p1
= 0,
(281)
while (274) restricts all the other pB s to vanish

pB = 0,
B = 2, n.
(282)
Thus, (275) and (276) for A = 1 and B

= 1 imply
p1 y3B 3D = 0,
p1 y2B = 0,
B = 2, n,
(283)
while (278) and (279) for A = 1 together with (281) lead to

p1 y31 3D = 0,
p1 y21 = 0.
(284)
The last two sets of equations, (283) and (284), display a unique solution
y3A 3D = 0 = y2A ,
A = 1, n.
(285)
In Case II
pA = 0,
A = 1, n,
(286)
Eqs. (274)(277) are identically satisfied, while the other two take the simple form
fA y3A 3D = 0,
(287)
fA y2A = 0,
(288)
Therefore, we have a single option, namely the set {1, 2, . . . , n} is divided into two complemen A ) with A
= A and (f = 0, y3A D = 0, y2A = 0). Re-ordering
tary subsets such that A = (A,
A
3
the indices we can always write
fA = 0,
A = 1, m,
y3A 3D = 0 = y2A ,
A = m + 1, n.
The above solution contains two limit situations: m = n and m = 0.
(289)
483
5.2. Main cases. Coupled theories

5.2.1. Case I: No-go results in General Relativity
As we have discussed previously, the first case is governed by the solution

p1 = f1
= 0,
(pB )B=2,n = 0,
y3A 3D A=1,n = 0 = (y2A )A=1,n ,
(290)
so the deformed solution to the master equation in all D > 2 spacetime dimensions is maximally parameterized by (fA )A=1,n , p1 = f1
= 0, (A )A=1,n , q1 3D , and q2 5D . Of course, it is
possible that some of fB (for B
= 1), A , q1 , or q2 vanish. Inserting (290) into (263) we find
0 = 0.
(291)
Combining this result with (270) we observe that the tower of Eqs. (259)(261) takes the homogeneous form
(int)
= 0,
b2
(int)
(int)
+ b1 = 1 ,
(293)
(int)
b1
(int)
+ b0
= 0 ,
(294)
b2(int) = b1(int) = b0(int) = 0
(295)
b2
(292)
so we can take
(int)
and incorporate the homogeneous solution into the first-order deformation S1 (see (230))
through a suitable redefinition of the parameterizing constants. At this point we act like in Sections 4.3.1 and 4.4.1. Replacing (295) and (290) into (230), (243), (244), (246), and (256)(258)
and regrouping the terms from (228) and (232) with the help of (248) and (254), we find that
there are no cross-couplings among different gravitons intermediated by the vector field. The
vector field gets coupled to a single graviton (the first one in our convention) and the resulting
interactions fit the rules prescribed by General Relativity.
The Lagrangian formulation of the coupled model can be completed by imposing some gaugefixing conditions similar to (163), one for each graviton sector. If in addition we make the
convention
f1 = 1 = p1 ,
(296)
then the fully deformed solution to the master equation

(I)
(I)
S (I) = S + k S1 + k 2 S2 + ,
S
(297)
is the free solution (215), leads to a Lagrangian action in which a single graviton
where
(A = 1) couples to the vector field V according to the standard coupling from General Relativity,
while each of the other gravitons (B = 2, n) interacts only with itself according to an Einstein
Hilbert action (or possibly a PauliFierz action if fB = 0) with a cosmological term. Accordingly,
in Case I we obtain the Lagrangian action

2
L(I) A
D
h , V = d x 2 g 1 R 1 2k 2 1
S
k

1
1 1
g 1 g 1 g 1 F
F + k q1 3D 11 2 3 V11 F12 3
4
484
+ q2 5D 11 2 3 4 5 V11 F12 3 F14 5

+
n
2
d x 2
kB
D
B=2
g B
R 2kkB B
B
n
1

B
,
, V1 +
S L(I) g
S L(EH) g
(298)
B=2
1 are curved with the vielbein fields from the first graviton sector
where V1 and F
1a
1a
1
Va ,
F
= [ e]
Va ,
V1 = e

1 e 1D a1 ...aD .
11 2 ...D = g 1 ea1
aD
1
(299)
(300)
The notations R A and g A (A = 1, n) denote the full scalar curvature and the determinant of the
A =
A
metric tensor g
+ kA h (without summation over A) from the Ath graviton sector
respectively, while kB = kfB , B = 2, n. The final conclusion is that in the first case there is no
cross-interaction among different gravitons to all orders in the coupling constant.
5.2.2. Case II: No-go results for the new couplings in D = 3
The second case is subject to the conditions
(fA )A=1,m
= 0,
(pA )A=1,n = 0,

y3A 3D A =m+1,n = 0 = (y2A )A =m+1,n ,
(301)
so the deformed solution to the master equation is maximally parameterized by (fA )A =m+1,n ,
(y3A 3D )A=1,m
, (y2A )A=1,m
, (A )A=1,n , q1 3D , and q2 5D . Substituting (301) into (263), it fol
lows that
0 = 4q1 3D y2A (D 2) F A 6q2 5D y2A F F A

m

A
y2A 4D.
(302)
A=1
Reasoning exactly like in the case of formulas (128) and (154), we deduce that Eq. (261) demands
an equation of the type (129), 0 = 1 + 0 + l0 , which cannot be satisfied for local 1 ,
0 , and l0 unless
0 = 0,
(303)
which further requires

q1 3D y2A
= 0,
A=1,m
q2 5D y2A
= 0,
A=1,m
m

y2A A = 0.
(304)
A=1
Clearly, there are two distinct solutions to the above equations

q1 3D
= 0 = q2 5D ,
m

y2A A = 0,
A=1
(305)
485
A = 1, m,
y2A = 0,
(306)
deserving separate analyses. In each subcase (270) and (303) hold, such that Eqs. (259)(261)
take the homogeneous form (292)(294), whose solution can be taken of the form (295).
5.2.2.1. Case II.1 From (301) and (305) we observe that the deformed solution to the mas,
ter equation is maximally parameterized in this situation by (fA )A =m+1,n , (y3A 3D )A=1,m
(y2A )A=1,m
, and (A )A=1,n , where in addition the first m cosmological constants are restricted
to satisfy the condition

m

y2A A = 0.
(307)
A=1
Consequently, the first- and second-order deformations of the solution to the master equation,
(228) and (232), read as

n

1

A
d D x fA A A [ ]
+ hA A hA
S1(II.1) =
2
A =m+1

(EH-cubic)A
A
A
A
[ h] + a 0
2A h
+
m

d D x y2A hA + (D 2) V A + V [ hA
]
A=1
]
+ y3A 3D V [ A
(II.1)
S2
=

n

fA
A =m+1
m
]
+ F [ hA 2A hA
(EH-quartic)A
fA S2

D
d x y2A y2B
B=1
A,

+ A
(308)

1 A 2
A A
d x h
h h
2
D
(D 2)2 A B

h h hA hB
]
]
+ y2A y3B 3D (D 2) hA [ hB + y3A y3B 3D [ hA [ hB
]
m

1
2
+ (D 2)(D 1)
(y2A )
d D x V V
2

(309)
A=1
respectively. The third-order deformation results from the equation

(II.1) (II.1)
(II.1)
S1 , S2
+ s S3
= 0.
(310)
If we make the notations

1

(EH-)A
A
d D x fA A A [ ]
+ hA A hA
S1
2

(EH-cubic)A
A

+
a
A [ hA
h
2
,
A
]
0
(311)
486

(EH-)A
S2

fA
(EH-quartic)A
fA S2

+ A

1 A 2
A A
d x h
h h
,
2
D
(312)
(EH-)A
(EH-)A
then we observe that S1
and S2
are nothing but the first- and second-order components respectively (in the coupling constant) of the solution to the master equation corresponding
to the full EinsteinHilbert theory in the presence of a cosmological constant for the graviton A .
Therefore,
n

n

n
(EH-)A
(EH-)A
(EH-)B
(313)
S1
,
S2
S3
= s
,
A =m+1
B =m+1
A =m+1
)A
(EHis the third-order component of the solution to the master equation associated
where S3
with the full EinsteinHilbert theory with a cosmological term in the graviton sector A . By direct
computation we then infer that
(II.1) (II.1)
S1 , S2

m
m

n

4
(EH-)A
A
D B
y2A y3
y3B 3 h
S3
=s
D2
=m+1
A
A=1
B=1
m

(y2A )2 (D 2)(D 1)
+
A=1
m
B=1

(D 2)

dDx
y2B hB 2V B + y3B 3D F B
2

,
(314)
such that the existence of local solutions to Eq. (310) demands that (hB 2V B ) and
F B are s-exact modulo d quantities from local functions for each B = 1, m. It is easy
to show that none of them has this property, so we must set
m

2
(315)
(y ) y = 0, B = 1, m,
2A
A=1
m
2B

2
(y2A )
B = 1, m.
y3B 3D = 0,
(316)
A=1
The solution to these equations,

y2B = 0,
B = 1, m,
(317)
solves in addition equation (307). Substituting (317) into (314) and then in (310) we find the
equivalent equation

n

(II.1)
(EH
)A
S3
= 0,
s S3
(318)
A =m+1
whose solution can be chosen, without loss of generality, of the form

(II.1)
S3
=
n

A =m+1
(EH-)A
S3
(319)
487
(EH-)A
We recall that S3
gathers the contributions of order three in the coupling constant from
the solution of the master equation corresponding to the full EinsteinHilbert action with a cosmological constant for the graviton A .
Putting together the results expressed by formulas (301), (305), and (317) we conclude that in
Case II.1 the consistency of the deformed solution to the master equation requires the conditions
(pA )A=1,n = 0 = (y2A )A=1,n ,

(fA )A=1,m
= 0,

D
D
D
q 1 3 = 0 = q 2 5 .
y3A 3 A =m+1,n = 0,
(320)
(321)
The full deformed solution to the master equation S (II.1) reads as

(II.1)
(II.1)
(II.1)
S (II.1) = S + k S1
+ k 2 S2
+ k 3 S3
+
(322)
(with S the solution of the master equation for the free model, (215)) and it is maximally parameterized by (fA )A =m+1,n , (y3A 3D )A=1,m
, and the cosmological constants (A )A=1,n . Taking
into account relations (215), (308), (309), (319) and notations (311)(312), we can decompose
S (II.1) as a sum between two basic parts
n

(II.1)
(EH-)A
S
(323)
=
S
+ S (special)
A =m+1
that are independent one of the other. The first part decomposes into (n m) components that
are all series in the constant coupling k

(EH-)A
(EH-)A
(EH-)A
S (EH-)A = S A + kS1
+ k 2 S2
+ k 3 S2
+ ,
with

S A
(PF)
A
A
A
d D x L0 hA
( )
, h + h
(324)

(PF)
A

(EH-)A represents
and L0 (hA
, h ) the PauliFierz Lagrangian for the graviton A . Each S
a copy of the solution to the master equation for the full EinsteinHilbert theory with a cosmo

logical constant associated with the graviton field hA
(A = m + 1, n), so they cannot produce
couplings among different gravitons. We emphasize that none of the (n m) gravitons gets coupled to the vector field V . Let us analyze in more detail the second part. It stops at order two in
the coupling constant

m

(PF) A
A
A
(special)
D
A
A
S
d x L0 h , h 2kA h + h
=
( )
A=1

1
A
d x F F + V + k
y3A 3D V [ ]
4
A=1

m

A
A B D
A
B
2
+k
y3 y3 3 [ h] [ h ]
+ F [ h]
(325)
B=1
A,
and in D = 3 spacetime dimensions seems to mix different spin-two fields via the terms from the
last (double) sum in the right-hand side of (325) with A

= B.
488
In order to focus in more detail on (325) we take the limit situation m = n (so A A) in the
conditions (320)(321) and work in D = 3, such that the entire deformed solution to the master equation, S (II.1) , consistent to all orders in the coupling constant, reduces to (325). We can
express S (special) in a nicer form by acting in a manner similar to that followed in Section 4.4.2.
Based on the observation that the deformed solution to the master equation is unique up to addition of s-exact terms, in the sequel we work with

n

A B

(special)
2
3
A
B
A
B
m=n s 2k
S
d x y 3 y3 h
h +
D=3

=
A,B=1
n

(PF) A

1
d 3 x F F + V +
L0 h , hA
4
A=1

A
A
A
A
A
2kA h + h
V [ ]
( ) + ky3
F [ hA
]

n

B
y3A y3B [ hA
.
+ 2k 2
] [ h]
(326)
A,B=1
The part of antighost number zero gives the Lagrangian action of the coupled model

S L(II.1) hA
, V

n

(PF) A
1
3
A
A
= d x F F +
L0 h , hA
F [ hA
2kA h ky3
]
4
A=1

n

2
A B
A
B
+ 2k
(327)
y3 y3 [ h] [ h]
A,B=1
and the terms of antighost one provide its gauge symmetries

A
(II.1) hA
= (
) ,
(II.1) V =
+ k
n

A
y3A [
]
.
(328)
A=1
This Lagrangian action can be brought to a simpler form by redefining the field strength of the
vector field as
F = F + 2k
n

y3A [ hA
] ,
(329)
A=1
in terms of which

=
S L(II.1) hA
, V

3
d x
n

(PF)
A
A
L0 hA
, h 2kA h
A=1

1
F F
.
4
(330)
The absence of terms of antighost number strictly greater than one indicates that the deformed
gauge symmetries (328) are independent and Abelian (their commutators close everywhere in the
space of field histories). We remark that this case corresponds to the situation from Section 4.4.2
(in the absence of internal PauliFierz indices), where we obtained a result complementary
to the usual couplings prescribed by General Relativity. The gauge symmetries of the vector
field are modified by terms proportional with the antisymmetric first-order derivatives of the
489
PauliFierz gauge parameters, while the gravitons keep their original gauge symmetries. The invariance of S L(II.1) under (328) is ensured by the gauge invariance of the deformed field strength,
(II.1)
F = 0.
Unfortunately, action (330) does not describe in fact cross-couplings between different spintwo fields. In order to make this observation clear, let us denote by Y the matrix of elements
y3A y3B . It is simple to see that the rank of Y is equal to one. By an orthogonal transformation M
we can always find a matrix Y of the form
Y = M T Y M,
with
MT
(331)
the transposed of M, such that
Y 11 =
n

y3A
2
2 ,
A=1

Y 1A = Y B 1 = Y A B = 0,
A , B = 2, n.
(332)
If we make the notation

y A = M AC y3C ,
(333)
then relation (332) implies

y A = 1A .
(334)
Now, we make the field redefinition

AC C
h ,
hA
= M
(335)
with M AC the elements of M. This transformation of the spin-two fields leaves

hA
) invariant and, moreover, based on the above results, we obtain
n
(PF) A
A=1 L0 (h ,
n

B
y3A y3B [ hA
= 2 [ h 1] [ h 1] ,
] [ h]
(336)
A,B=1
n
y3A F [ hA
F [ h 1] ,
] =
(337)
A=1
such that (330) becomes

S

L(II.1)
h A
, V

=
d 3x
n

(PF)
A
A
L0 h A
, h 2k A h
A=1

1
F F
,
4
(338)
where
A = B M BA ,
F = F + 2k [ h 1] .
(339)
Action (338) is invariant under the gauge transformations

(II.1) A
h
A
= (
)
,
(II.1) V =
+ k [
]
,
(340)
where
A =
B M BA .
(341)
490
We observe that action (338) decouples into action (199) (derived in Section 4.4.2) for the first
spin-two field (A = 1) and a sum of PauliFierz actions with cosmological terms for the remaining (n 1) spin-two fields. In conclusion, we cannot couple different spin-two fields even outside
the framework of General Relativity.
5.2.2.2. Case II.2 Now, we start from conditions (301) and (306), such that the deformed solution to the master equation is maximally parameterized in this situation by (fA )A =m+1,n ,
(y3A 3D )A=1,m
, (A )A=1,n , q1 3D , and q2 5D . Without entering unnecessary details, we only men
tion that this case is similar to Case II.1.2 in the absence of PauliFierz internal indices, briefly
discussed in the final part of Section 4.3.2. The consistency of the deformed solution to the master equation goes on unobstructed up to order five in the coupling constant, where the existence
(II.1.2)
requires a condition of the type y33 q12 = 0, namely
of a local S5
m

2
2
q1
(342)
(y ) y 3D = 0, B = 1, m.
3A
3B
A=1
There are two main possibilities, none of them leading to cross-couplings between different spintwo fields. Thus, if we take D
= 3, then no couplings among different gravitons are allowed since
the Lagrangian of the interacting model is a sum of independent EinsteinHilbert Lagrangians
with cosmological terms for the last (n m) gravitons (none of them coupled to the vector field),
a sum of PauliFierz Lagrangians plus simple cosmological terms 2kA hA for the first m
gravitons and the Maxwell Lagrangian supplemented by the generalized Abelian ChernSimons
density kq2 5D V F F . If D = 3, then either q1 = 0, in which situation we re-obtain
the case from the previous section, described by formula (323), where we have shown that there
are no cross-couplings between different gravitons, or (y3A )A=1,m
= 0, such that again no cross
couplings are permitted and the resulting Lagrangian is like in the above for D
= 3 (after formula
(342)) up to replacing the density kq2 5D V F F with the standard Abelian Chern
Simons term kq1 V F .
6. Generalization to an arbitrary p-form
The results obtained so far in the presence of a massless vector field can be generalized to the
case of deformations for one or several gravitons and an arbitrary p-form gauge field.
In the case of a single graviton the starting point is the sum between the PauliFierz action
and the Lagrangian action of an Abelian p-form with p > 1

1
(PF)
F1 ...p+1 F 1 ...p+1 ,
S0L [h , V1 ...p ] = d D x L0
(343)
2 (p + 1)!
in D p + 1 spacetime dimensions, with F1 ...p+1 the Abelian field strength of the p-form
gauge field V1 ...p
F1 ...p+1 = [1 V2 ...p+1 ] .
(344)
This action is known to be invariant under the gauge transformations
h = (
) ,
V1 ...p = [1
(p)
.
2 ...p]
(345)
Unlike the Maxwell field (p = 1), the gauge transformations of the p-form for p > 1 are off-shell
reducible of order (p 1). This property has strong implications at the level of the BRST complex
491
and of the BRST cohomology in the form sector: a whole tower of ghosts and of antifields will be
required in order to incorporate the reducibility, only the ghost of maximum pure ghost number,
p, will enter H ( ), and the local characteristic cohomology will be richer in the sense that (33)
and (35) become [31]
HJ (|d) = 0 = HJinv (|d),
J > p + 1.
(346)
In spite of these new cohomological ingredients, which complicate the analysis of deformations,
the results from Sections 4.4.1 and 4.4.2 can still be generalized.
Thus, two complementary cases are revealed. One describes the standard graviton-p-form interactions from General Relativity and leads to a Lagrangian action similar to (195) up to replacing (1/4)g g F F with the expression (2(p +1)!)1 g 1 1 g p+1 p+1 F1 ...p+1 F1 ...p+1
and, if p is odd, also the terms containing 3D 1 2 3 and 5D 1 2 3 4 5 with some densities
D
D
1 ...2p+1 and 3p+2
1 ...3p+2 respectively (if p is even, the terms proportional
involving 2p+1
with either q1 or q2 must be suppressed). The other case emphasizes that it is possible to construct some new deformations in D = p + 2, describing a spin two-field coupled to a p-form
and having (343) and (345) as a free limit, which are consistent to all orders in the coupling
constant and are not subject to the rules of General Relativity. Their source is a generalization
of the terms proportional with y3 from the first-order deformation (71)

1 1 ...p [ ]
(int)
p+2
1 ...p [ ]
x V
+ F
h .
S1 (y3 ) = y3 1 ...p d
(347)
p!
Performing the necessary computations, we find the Lagrangian action

1
(PF)
L
p+2

1 ...p+1
S [h , V1 ...p ] = d
x L0 2kh
F
F
, (348)
2 (p + 1)! 1 ...p+1
where the field strength of the p-form is deformed as
F 1 ...p+1 = F1 ...p+1 + 2()p+1 ky3 1 ...p+1 [ h] .
(349)
This action is fully invariant under the original PauliFierz gauge transformations and
V1 ...p = [1
(p)
+ ky3 1 ...p [
] .
2 ...p]
(350)
The gauge algebra remains Abelian and the reducibility of (350) is not affected by these couplings: the associated functions and relations are the initial ones.
It is important to notice that all the standard hypotheses imposed to consistent deformations
are fulfilled. Indeed, in the free limit (k = 0) the field strength (349) is restored to its original form
(344), the cosmological term 2kh is destroyed, and the PauliFierz gauge parameters
are
discarded from the gauge transformations
V1 ...p , leaving us with the original action (343)
and initial gauge transformations (345). Also, the spacetime locality, Lorentz covariance, and
Poincar invariance of action (348) are obvious. Likewise, the smoothness of the deformed theory
in the coupling constant is ensured by the polynomial behaviour of (348) and (350) with respect
to k: the action is a polynomial of order two and the gauge transformations are polynomials of
order one. Furthermore, the differential order of the coupled field equations is preserved with
respect to that of the free equations (derivative order assumption), being equal to two, as it can
be observed from the concrete form of the EulerLagrange derivatives of action (348)
492
S L [h , V1 ...p ]
h
=
S0L [h , V1 ...p ]
2k
h
ky3
()p 1 ...p+1 ( ) F 1 ...p+1 + 2 1 ...p+2 1 F 2 ...p+2

(p + 1)!

h + 1 + 4k 2 y32 h 1 + 2k 2 y32 ( h)

+ 1 + 4k 2 y32 h h 2k
ky3
+ ()p+1
1 ...p+1 ( ) F1 ...p+1 ,
(p + 1)!
S L [h , V1 ...p ]
V1 ...p
=
=
(351)
1
F 1 ...p
p!
S0L [h , V1 ...p ]
V1 ...p
1
F 1 ...p
p!
ky3 1 ...p
[ h]
p!

ky3 1 ...p [ h] .
(352)
It is truly remarkable that these new couplings comply with the derivative order assumption.
Let us analyze the main physical consequences of these new couplings. First, we investigate
some direct outcomes of the field equations, obtained by equating (351) and (352) to zero. By
taking the trace of the field equations for the graviton, S L /h = 0, we infer the equivalent
equation
K=
2k(p + 2)
,
p + (p + 1)4k 2 y32
(353)
where K is the linearized scalar curvature, K = h h. Due to the presence of the cosmological constant, the linearized scalar curvature is a non-vanishing constant. The field equation
of the p-form, S L /V1 ...p = 0, is nothing but a nontrivial conservation law of order (p + 1),
F 1 ...p = 0, where the associated current is precisely the deformed field strength (349). It
is not a usual conservation law because it results from some rigid symmetries of the solution to
the master equation for the coupled theory. The main difference between the free theory (343)
and the coupled one is that the (p + 1)-order conservation law of the latter contains a nontrivial
component from the PauliFierz sector, 2()p+1 ky3 1 ...p [ h] . Another interesting observation is that, unlike the free limit (343), the field equations of the coupled model admit to
be written in a compact form. Indeed, it can be shown that both field equations, S L /h = 0
and S L /V1 ...p = 0, are completely equivalent with the following expression of the first-order
derivatives of the Abelian field strength (344)

()p
1
2k h
1 ...p+1 2ky3 [ h] +
F1 ...p+1 =
2
ky3

1 + 4k 2 y32 h + 1 + 2k 2 y32 ( h)

2 2
1 + 4k y3 h h .
(354)
493
The direct as well as the converse implication results from simple algebraic manipulations of the
coupled field equations or respectively of (354) and also by means of the identity
1 ...p F1 ...p =
()p+1 1 ...p+1 [ ]
F1 ...p+1 ,
p+1
(355)
valid in D = p + 2.
Regarding a collection of spin-two fields and a p-form, it can be used a line similar to that
employed in Section 5. Thus, it can be shown that two complementary cases are again unfolded.
One is similar to the situation discussed in Section 5.2.1 and the other with the result from
Section 5.2.2. In both cases there are no cross-couplings among different spin-two fields intermediated by a p-form gauge field: the p-form couples to a single spin-two field.
7. Conclusion
To conclude with, in this paper we have investigated the couplings between a single spin-two
field or a collection of such fields (described in the free limit by a sum of PauliFierz actions)
and a massless p-form using the powerful setting based on local BRST cohomology. Under the
hypotheses of locality, smoothness in the coupling constant, Poincar invariance, Lorentz covariance, and preservation of the number of derivatives on each field (plus positivity of the metric
in the internal space in the case of a collection of spin-two fields), we found two complementary situations. One submits to the well-known prescriptions of General Relativity, but the other
situation discloses some new type of couplings in (p + 2) spacetime dimensions, which only
modify the gauge symmetries of the p-form. It is remarkable that these (p + 2)-dimensional
cross-couplings comply with the derivative order assumption, unlike other situations from the
literature. Unfortunately, in the case of a collection of spin-two fields none of these coupled
theories allows for (indirect) cross-couplings between different gravitons.
Acknowledgements
The authors are partially supported by the European Commission FP6 program MRTNCT-2004-005104 and by the type A grant 305/2004 with the Romanian National Council for
Academic Scientific Research (C.N.C.S.I.S.) and the Romanian Ministry of Education and Research (M.E.C.). The authors thank the referee for his/her valuable comments and suggestions.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
C. Cutler, R.M. Wald, Class. Quantum Grav. 4 (1987) 1267.

R.M. Wald, Class. Quantum Grav. 4 (1987) 1279.
A. Hindawi, B.A. Ovrut, D. Waldram, Phys. Rev. D 53 (1996) 5583.
S.C. Anco, Ann. Phys. (N.Y.) 270 (1998) 52.
S.N. Gupta, Phys. Rev. 96 (1954) 1683.
R.H. Kraichnan, Phys. Rev. 98 (1955) 1118.
S. Weinberg, Phys. Rev. 138 (1965) B988.
S. Deser, Gen. Relativ. Gravit. 1 (1970) 9.
D.G. Boulware, S. Deser, Ann. Phys. (N.Y.) 89 (1975) 193.
J. Fang, C. Fronsdal, J. Math. Phys. 20 (1979) 2264.
F.A. Berends, G.J.H. Burgers, H. Van Dam, Z. Phys. C 24 (1984) 247.
R.M. Wald, Phys. Rev. D 33 (1986) 3613.
R.P. Feynman, F.B. Morinigo, W.G. Wagner, Feynman Lectures on Gravitation, AddisonWesley, Reading, 1995.
494
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
N. Boulanger, T. Damour, L. Gualtieri, M. Henneaux, Nucl. Phys. B 597 (2001) 127.

C. Bizdadea, E.M. Cioroianu, A.C. Lungu, S.O. Saliu, JHEP 0502 (2005) 016.
C. Bizdadea, E.M. Cioroianu, D. Cornea, S.O. Saliu, S.C. Sararu, Eur. Phys. J. C 48 (2006) 265.
N. Boulanger, M. Esole, Class. Quantum Grav. 19 (2002) 2107.
N. Boulanger, L. Gualtieri, Class. Quantum Grav. 18 (2001) 1485.
G. Barnich, M. Henneaux, Phys. Lett. B 311 (1993) 123.
M. Henneaux, Contemp. Math. 219 (1997) 93.
J.A. Garcia, B. Knaepen, Phys. Lett. B 441 (1998) 198.
I.A. Batalin, G.A. Vilkovisky, Phys. Lett. B 102 (1981) 27.
I.A. Batalin, G.A. Vilkovisky, Phys. Rev. D 28 (1983) 2567;
I.A. Batalin, G.A. Vilkovisky, Phys. Rev. D 30 (1984) 508, Erratum.
I.A. Batalin, G.A. Vilkovisky, J. Math. Phys. 26 (1985) 172.
M. Henneaux, Nucl. Phys. B (Proc. Suppl.) 18A (1990) 47.
M. Henneaux, C. Teitelboim, Quantization of Gauge Systems, Princeton Univ. Press, Princeton, 1992.
G. Barnich, F. Brandt, M. Henneaux, Commun. Math. Phys. 174 (1995) 57.
G. Barnich, F. Brandt, M. Henneaux, Phys. Rep. 338 (2000) 439.
W. Pauli, M. Fierz, Helv. Phys. Acta 12 (1939) 297.
M. Fierz, W. Pauli, Proc. R. Soc. London A 173 (1939) 211.
M. Henneaux, B. Knaepen, C. Schomblond, Commun. Math. Phys. 186 (1997) 137.
M. Henneaux, B. Knaepen, Phys. Rev. D 56 (1997) 6076.
W. Siegel, Fields, hep-th/9912205.

N = 1/2 supergravity with matter

in four Euclidean dimensions
Tomoya Hatanaka, Sergei V. Ketov
Department of Physics, Tokyo Metropolitan University, 1-1 Minami-osawa, Hachioji-shi, Tokyo 192-0397, Japan
Received 1 August 2007; received in revised form 13 October 2007; accepted 15 October 2007
Abstract
An N = 1/2 supergravity in four Euclidean spacetime dimensions, coupled to both vector- and scalarmultiplet matter, is constructed for the first time. We begin with the standard N = (1, 1) conformally
extended supergravity in four Euclidean dimensions, and freeze out the graviphoton field strength to an
arbitrary (fixed) self-dual field (the so-called C-deformation). Though a consistency of such procedure with
local supersymmetry is not guaranteed, we find a simple consistent set of algebraic constraints that reduce
the local supersymmetry by 3/4 and eliminate the corresponding gravitini. The final field theory (after the
superconformal gauge-fixing) has the residual local N = (0, 1/2) or just N = 1/2 supersymmetry with only
one chiral gravitino as the corresponding gauge field. Our theory is not Lorentz-invariant because of
the non-vanishing self-dual graviphoton vacuum expectation value, which is common to the C-deformed
N = 1/2 rigidly supersymmetric field theories constructed in a non-anticommutative superspace.
PACS: 04.65.+e; 12.60.Jv
Keywords: Supergravity; Gravi-photon
1. Introduction
A construction of new supergravity theories is apparently complete after a lot of work done
in the pastsee e.g. Refs. [1,2]. However, it is merely apparent, because some recent developments in field theory, strings and gravity offer new opportunities for even further generalizations
Supported in part by the Japanese Society for Promotion of Science (JSPS).
E-mail addresses: hatanaka-tomoya@c.metro-u.ac.jp (T. Hatanaka), ketov@phys.metro-u.ac.jp (S.V. Ketov).

doi:10.1016/j.nuclphysb.2007.10.020
496
T. Hatanaka, S.V. Ketov / Nuclear Physics B 794 [PM] (2008) 495511
of supergravities, by relaxing some of the symmetry requirements. It offers new perspectives to

various physical applications, such as (i) partial supersymmetry breaking, and (ii) brane supersymmetry reduction.
One of the recent developments is a noncommutative gravity. Though the idea of replacing the
ordinary field product by the noncommutative Moyal (star) product is not new [3], its implementation is not unique, while it often leads to a complexification of the metric and the appearance of
ghosts (see, however, Ref. [4] for possible cures). The appearance of infinitely many interaction
vertices with unlimited powers of momenta is the necessary feature of those noncommutative
gravity models.
String theory can teach us more about noncommutative gravity (see e.g. Ref. [5]), as well
as about supersymmetry [6]. In particular, as was observed by Ooguri and Vafa [6], the superworldvolume of a supersymmetric D-brane in a constant RamondRamond type flux gives rise
to the remarkable new structure in the corresponding superspace, which is now called NonAntiCommutativity (NAC). It means that the fermionic superspace coordinates are no longer
Grassmann (i.e. they no longer anticommute), but satisfy a Clifford algebra. In other words, the
impact of the RamondRamond flux on the D-brane dynamics can be described by the nonanticommutativity in the D-brane superworldvolume.
As regards a D3-brane, a 10-dimensional (self-dual) five-form flux upon compactification to
four dimensions gives rise to the (self-dual) graviphoton flux in the D3-brane 4-dimensional
worldvolume [6]. In its turn, the non-anticommutativity in superspace can be described by the
(MoyalWeyl type) non-anticommutative star product amongst superfields. It results in a construction of the NAC deformed supersymmetric field theories with partially broken supersymmetry, pioneered by Seiberg [7], in four Euclidean dimensions. Unlike bosonic noncommutativity,
the NAC supersymmetric field theories usually have only a limited number of new interaction
terms, without higher derivatives, while their Lagrangians can often be written down in closed
form. As a matter of fact, all recent studies of the NAC supersymmetric field theories, following Ref. [7], were limited to rigid supersymmetry, i.e. without gravity or supergravity (see e.g.
Ref. [8] and references therein).
Given the relation between a non-anticommutativity and a non-vanishing (self-dual) vacuum
expectation value of a graviphoton in four dimensions, it is quite natural to explore further possibilities for a construction of new supergravities, by freezing a graviphoton field in an extended
supergravity theory (with matter). The minimally extended Poincar supergravity in four dimensions, that has a graviphoton as the superpartner of a graviton, is the N = (1, 1) or just N = 2
supergravity. The structure of N = 2 supergravity with matter was given in detail in Refs. [9,10].
In our earlier paper [11], a toy model of the four-dimensional N = 1/2 supergravity with
a fixed self-dual graviphoton expectation value was constructed by freezing out the graviphoton
field strength in the standard N = (1, 1) extended supergravity with two non-chiral gravitini [12].
Our supergravity model [11] has local N = (0, 12 ) supersymmetry. Consistency of the model [11]
requires the expectation value of the graviphoton field strength to be equal to the self-dual (bilinear) gravitino condensate.
An extension of the construction [11] to a matter-coupled N = 1/2 supergravity is not automatic since more consistency conditions have to be satisfied. In this paper we report our results
of such construction, by presenting the Lagrangian and the local supersymmetry transformation
laws of the N = 1/2 supergravity in four Euclidean dimensions, coupled to vector supermultiplets and scalar supermultiplets, with all the fermionic terms included.
Our paper is organized as follows: in Section 2 we briefly describe the contents of N = 2
conformal supergravity multiplets in the N = 2 superconformal tensor calculus [9]. In Section 3
497
we introduce the 2-component notation for spinors in four Euclidean dimensions. In Section 4
we describe our way of freezing of a graviphoton field, by imposing a self-duality condition on
the graviphoton field strength, and then studying its consistency and the surviving symmetries.1
The residual superconformal transformations are given in Section 5. The superconformal gaugefixing and elimination of the auxiliary fields are discussed in Section 6. Our Lagrangian of the
N = 1/2 Euclidean supergravity with matter is given in Section 7. Section 8 is our conclusion.
Appendices AC are devoted to further notation and some technical details. The notation [9]
for hypermultiplets is briefly summarized in Appendix A. In Appendix B we collect the N = 2
superconformal transformation laws [9] which is the starting point of our construction. In Appendix C we quote the so-called decomposition rules [9] needed in passing from the conformal
supergravity to the Poincar supergravity.
2. N = 2 supergravity field components
The last paper of Ref. [9] is the pre-requisite to our construction. So, instead of copying the
equations of Ref. [9] here, we merely review the methods of Ref. [9], and concentrate on the
differences between our construction and that of Ref. [9].
First, our construction can only be defined in four Euclidean dimensions, not in Minkowski
spacetime as in Ref. [9]. As is well known, the change of signature has important implications
on the structure of field representations, especially on spinors. For instance, minimal spinor
representations in Minkowski signature are given by real (Majorana) spinors or complex chiral spinors, while the chiral and anti-chiral parts of a Majorana spinor are related by complex
conjugation. Accordingly, a number of supersymmetries in Minkowski spacetime is measured
by a number of Majorana supercharges. In four Euclidean dimensions, Majorana spinors do not
exist [15], whereas the chiral and anti-chiral spinors are independent. Hence, the numbers of left
and right (chiral and anti-chiral) Euclidean supercharges need not be the same, while the minimal
choice is obviously given by one chiral or anti-chiral supercharge. We call it (1/2, 0) or (0, 1/2)
SUSY, respectively, or simply N = 1/2 supersymmetry.
Second, in order to make the chiral supersymmetry manifest, we are going to use the 2-component notation for spinors, which is best suitable for our purposes. So we rewrite the results
of Ref. [9] obtained in the 4-component spinor notation, to the 2-component notation in four
Euclidean dimensions (see Section 3).
The N = 2 superconformal tensor calculus gives us a systematic method for constructing the
N = 2 superconformal and super-Poincar-invariant Lagrangians and the N = 2 transformation
laws (see e.g. Ref. [14] for a review). It provides us with (i) the off-shell N = 2 supermultiplets,
as the representations of N = 2 local superconformal algebra, together with the transformation
laws of their field components, which form a closed algebra, (ii) the multiplication rules for a
construction of new representations, and (iii) the density formulas describing the superconformal
invariants.
In order to get the super-Poincar Lagrangian and the transformation laws, one has to fix
the truly superconformal symmetries, while keeping the super-Poincar ones. It is often called
gauge fixing. The gauge fixing conditions give rise to the decomposition laws relating the truly
superconformal transformation parameters to the super-Poincar transformation parameters
see Refs. [9,14] for details.
1 As regards a consistent reduction of the N = 2 matter-coupled supergravity to N = 1 matter-coupled supergravity,
including all fermionic terms, see Ref. [13].
498
The off-shell N = 2 superconformal multiplets are given by [9]

a Weyl multiplet,
a vector multiplet,
a hypermultiplet.
The N = 2 Weyl multiplet has 24 + 24 independent field components:
e a , i , b , A , V i j , Tab ij , i , D
(2.1)
a is vierbein, and the gravitino doublet i is the gauge field of N = 2 local supersymwhere e
metry. The gauge fields of other superconformal symmetries are b for dilatations, A for chiral
U (1) rotations, and V i j for chiral SU(2) rotations. We also need the auxiliary fields: the bosonic
ij
tensor Tab and the real scalar D, and a fermionic (spinor) doublet i .
a and i are going to represent physical degrees of freedom, the V i is the antiOnly e
j
ij
Hermitian traceless matrix in its SU(2) indices i, j , while Tab is the real tensor antisymmetric in
its both SU(2) and Lorentz index pairs.
An N = 2 vector multiplet has 8 + 8 independent field components:
(X, i , W , Yij )
(2.2)
where X is a complex scalar, i is a spinor doublet, W is a vector gauge field, and Yij is a real
SU(2) auxiliary triplet.
We consider the vector gauge fields to be Lie-algebra valued, with the generators tI obeying
an algebra
[tI , tJ ] = fI J K tK
(2.3)
where we have introduced the Lie algebra structure constants fI J K . So, the vector multiplet components carry the extra (gauge) index, I, J, . . . = 0, 1, . . . , n. We choose I = 0 for a graviphoton
(Abelian) gauge field, so we also define I, J, . . . = 1, 2, . . . , n.
The hypermultiplet physical fields are given by
Ai ,
(2.4)
where Ai is a scalar doublet and is a complex spinor. The hypermultiplets are supposed to
belong to a representation of the non-Abelian gauge group. The index i = 1, 2 is associated
to the SU(2) automorphism group of the N = 2 supersymmetry algebra, whereas the index
= 1, 2, . . . , 2r is the representation index with respect to the non-Abelian gauge group. See
Appendix A for more about the hypermultiplet notation.
The N = 2 superconformal transformation laws in the 2-component Euclidean notation are
collected in Appendix B.
A consistent Wick rotation of a field theory with fermions from four Minkowski dimensions to
four Euclidean dimensions is described in detail in Ref. [15]. In Ref. [9] the spacetime signature
(+++) was used, which is now going to be Wick-rotated to (++++) by setting x4 = it. As
regards the vector gauge fields, it implies A (AE , iAE
4 ). As is argued in Ref. [15], one should
5
change 4 iE , and use gamma matrices and a charge conjugation matrix in four Euclidean
499
dimensions (see e.g. Appendix in Ref. [1]).2 So our definition of the Dirac conjugation is
i = (i ) iE5 .
(2.5)
For instance, the Majorana condition is modified as follows:

i T
CE = (i ) iE5 .
(2.6)
It is worth mentioning that this condition is not a reality condition for spinors. Nevertheless, we
can still use this condition for constructing the Euclidean version of a given supergravity theory.
To avoid confusion, we sometimes append a script (E) for Euclidean fields or matrices, and a
script (M) for their Minkowski counterparts.
3. Euclidean 2-component spinor notation
We use lower case Greek letters for curved space vector indices, , , . . . = 1, 2, 3, 4, lower
case Latin indices for flat (tangent) space vector indices, a, b, . . . = 1, 2, 3, 4, and capital Latin
letters for (anti-)chiral spinor indices (dotted or undotted), A, B, . . . = 1, 2.
Gamma matrices in four Euclidean dimensions satisfy an algebra
{a , b } = 2ab ,
{5 , a } = 0.
An explicit representation of the Euclidean gamma matrices is as follows [1]:

0
ikAB
0 1
1 0
,
4 =
,
5E =
,
k =
1 0
0 1
ik AB
0
where k are Pauli matrices, k = 1, 2, 3. In addition, we define the matrices

abA

0
1 ( a b b a )A C
0
C
ab =
=:
0
ab A C
4
0
( b b a )A C
(3.1)
(3.2)
(3.3)
which are anti-Hermitian,

(ab ) = ba ,
(3.4)
in terms of their self-dual and anti-self-dual combinations,

1 abcd
1 abcd A
cd A B = ab A B and
cd B = abA B .
2
2
The Euclidean charge conjugation matrix is given by

0
AB
i2
0
C = 4 2 =
=
0 i2
0
AB
(3.5)
(3.6)
so that
T
C a C 1 = a ,
AB

0
C = C T = C ,
C =
,
0
A B
C = C T ,
(3.7)
(3.8)
2 In Ref. [15] the Dirac conjugation includes a factor of i, while the Lagrangian excludes a factor of i, but we are going
to use the opposite notation.
500
C C = CC = I.
(3.9)
When changing the notation [9] by representing all the 4-component spinors in terms of their
2-component constituents in Euclidean space, it is important to observe the chirality of each
4-component spinor given by the position (up or down) of its SU(2) index (i, j = 1, 2), e.g.,
IB

i
0
iI =:
(3.10)
,
,
iI =: iI
B
0
T

I T
I

iI
I
Mi
EiI CE =: 0, iI A ,
Ei
CE =: Ai
,0
M
(3.11)
and

iB
0
=:
,
,
i =:
i B
0

i T
i

A
i
M
.
E
CE =: A
,0 ,
iM (iE )T CE =: 0, i
In order to avoid double counting, their complex conjugates are given by

I
iI

iI
iiA
I
=
= iiA
,0 ,
,
0

I
I
0

i = 0, i iI A ,
i =
,
i iI A

i
i
0
A
,
,
= 0, i i
=
A
i i

i

iA
i
(i ) =
,0 .
,
(i ) = iA
0
(3.12)
(3.13)
(3.14)
(3.15)
(3.16)
(3.17)
The SU(2) indices are contracted by ij and ij ,

ij ij = 2,
12 = 21 = 12 = 21 = 1,
(ij ) = ik j l kl = ij .
(3.18)
Two-component spinor indices are contracted as follows:

A = B BA ,
A = B B A ,
A = AB B ,
AB
a DC
= a AC B D ,
aAB
B D AC
A = AB B ,
(3.19)
(3.20)
a DC
,
where we have
AB = BA ,
AB BC = A C ,
12 = 12 = 1 2 = 12 = 1.
(3.21)
We use the following book-keeping notation:
a A aAB B ,
B
a A a AB
b
C
a b A aAB BC
,
a b A a bB C C ,
AB
D D ,
D D .
(3.22)
A fully antisymmetric Levi-Civita symbol abcd is normalized by 1234 = 1. The (anti-)selfdual parts of an antisymmetric tensor Tab are3
1
Tab = abcd T cd ,
2
Tab
= (Tab Tab ).
2
3 Our sign convention here is opposite to that in Refs. [1] and [11].
(3.23)
501
4. Consistent freezing of a graviphoton field

Our basic idea is to eliminate a graviphoton field from the N = 2 matter-coupled supergravity,
by assigning it a fixed value, say, its vacuum expectation value (VEV). Generally speaking, it is
going to break supersymmetry, because the graviphoton is a field component of N = 2 supergravity multiplet. We show in this section that, when assigning a self-dual vacuum expectation value
to a graviphoton, the N = 2 local supersymmetry can be consistently broken to an N = 1/2 local
supersymmetry in four Euclidean dimensions, and in the presence of a generic N = 2 matter,
thus generalizing the earlier results [11] obtained for pure N = 2 supergravity without matter.
The graviphoton field is identified with W0 (x) after imposing the gauge condition on X 0 by
using the local chiral U (1) rotations [9]:
X 0 = X 0 > 0.
(4.1)
We prefer to impose the gauge condition (4.1) after our truncation procedure described in this
section. The graviphoton field strength is given by
0
= W0 W0 .
F
(4.2)
We now freeze the graviphoton field by imposing the self-dual constraints:

0+
C (x) and
F
0
F
0,
(4.3)
where C (x) is a fixed self-dual antisymmetric tensor field (VEV) with an arbitrary x-dependence.
Our strategy is to look for the residual local supersymmetry that keeps our assignment (4.3)
0+ and F 0 to be unchanged. As was shown in Ref. [11] in the
invariant, i.e. that leaves both F
case of the N = 2 supergravity without matter, we already have to fix 3/4 of Q-supersymmetry
by eliminating three (out of four) infinitesimal local Q-supersymmetry chiral spinor parameters
as follows:
1 2 2 0.
(4.4)
Then we may only hope that the remaining local N = 1/2 supersymmetry with the chiral spinor
parameter 1 (x) remains to be a symmetry of the theory after investigating all the consistency
conditions originating from the C-deformation (4.3):
0
= 0.
1 F
(4.5)
The N = 1/2 Q-supersymmetry transformation law of W0 can be read off from Eq. (B.6):
1 W0 = i 1 20 21 2 X 0 .
(4.6)
It implies the following consistency condition:

20 A 2i 2A X 0 = 0.
(4.7)
This algebraic condition can be easily solved, thus eliminating an independent gravitino field
2A , in terms of the other (matter) fields.
Eq. (4.7) is also not invariant under the N = 1/2 Q-supersymmetry, so that we have to impose
its invariance for consistency. By using the N = 1/2 Q-supersymmetry transformation laws
m
= i 1 m 1 ,
1 e
1 20A = 0,
1 2A = V2 1 1A ,
1 X 0 = 0,
(4.8)
502
we find from Eqs. (B.1), (B.3), (B.4) and (B.7) that the constraint (4.7) yields yet another constraint
1 20 V2 1 X 0 = 0.
(4.9)
This is again an algebraic equation, while it can be easily solved for V2 1 , thus eliminating that
field in terms of the other fields.
Generally speaking, further consistency requirements might lead to the infinite and increasingly complicated set of constraints, but it does not happen in our case! We find that Eq. (4.9) is
invariant under the N = 1/2 Q-supersymmetry because of
1 1 = 0 and 1 V2 1 = 0
(4.10)
which easily follow from Eqs. (B.2) and (B.16), respectively.

Having found the consistent short (finite) set of algebraic constraints that are invariant under
the N = 1/2 local Q-supersymmetry, we can easily check what are the other residual symmetries
in the list of the N = 2 superconformal transformation laws (see Appendices B and C), which
also leave the constraints invariant. We then find that all our constraints (4.3), (4.7) and (4.9) are
still invariant under the full S-supersymmetry and the chiral U (1) transformations, whereas the
local SU(2) automorphisms of the N = 2 supersymmetry algebra are broken: i j = i j = 0.
The residual S-supersymmetry implies that the original decomposition laws (see Ref. [9] and
Appendix C) are still valid, being subject to the conditions 1 = 2 = 2 = 0. It means that our
construction does not affect the S-gauge, K-gauge and D-gauge, as described in Ref. [9]. So we
are going to use the same gauges when passing to the Poincar supergravity.
The rest of our construction of the N = 1/2 matter-coupled supergravity is pretty straightforward (though tedious!), by inserting the equations derived in this section into the results of
Ref. [9], making gauge-fixing, and deriving the transformation rules and the Lagrangian of the
(Poincar) N = 1/2 supergravity with matter. Our results are summarized in Sections 5, 6 and 7.
5. N = 1/2 supergravity transformation laws
Here we summarize the residual local superconformal symmetry transformation laws of the
independent field components with respect to the Q-supersymmetry with the parameters (1 ),
the S-supersymmetry with the parameters (i ) and ( i ), and the chiral U (1) symmetry with the
parameter (A ).
(i) As regards a vierbein and gravitini, we find
e a = i 1 a 1 ,
A 1

1A = +i 1 iA 1A ,
2

A
A 1

1
21 m
1 + i 2 iA 2A ,
2A = iTm
2
2

1 mn
1
1A = + 2 1A mnA B 1B i A 1A + V1 1 1A i( 1 )A
2
2
1
+ iA 1A .
2
503
As is clear from those equations, the gravitino field 1A is the gauge field of the residual N =
1/2 local supersymmetry.
(ii) As regards the vector multiplet components, we find (I > 0)
X I = iA X I ,
X I = +1 1I + iA X I ,
WI = i 1 2I 21 2 X I ,

A

1
( + iA )X I gfJ K I WJ X K k kI
1I A = 2i 1
2
1
+ 2X I 1A iA 1I A ,
2
1
IA
I A
2 = +2X 2 iA 2I A ,
2
1
A
1I
A
11I

= +Y 1 + 2X I 1A + iA 1I A ,
2
1
A +I
2I A = +Y 21I 1A + 1B Bmn
Fmn + 2gfJ K I X J X K 1A + 2X I 2A + iA 2I A .
2
(iii) As regards the hypermultiplet components, we find (see Appendix A for the notation)
A1 = 2 1 ,
A2 = 0,
A1 = 0,
A2 = +2 1 ,

i
A = +i 1 A D A1 + Ai iA + A A ,
2
i
i
A = 2gX A2 1A + Ai A
A A
2
where we have used the abbreviations (B.13), (B.14) and (B.15).
6. Gauge-fixing and eliminating auxiliary fields
In the superconformal tensor calculus, the N = 1 or N = 2 matter-coupled (Poincar) supergravity is obtained from the N = 1 or N = 2 conformal supergravity, respectively, by imposing
certain gauges, in order to fix the truly superconformal symmetries, while keeping the superPoincar symmetries. In this process the residual supersymmetry transformations are deformed
by the compensating transformations needed to restore the gauges [9]. We follow here the same
pattern by eliminating the truly superconformal N = 1/2 symmetries after imposing our set of
constraints (Section 4), with the help of the gauges similar to that of Ref. [9]. Then we eliminate
the auxiliary fields by using their algebraic equations of motion. The Lagrangian of the resulting
N = 1/2 supergravity is given in the next section.
The K-gauge to fix the conformal boosts is given by
b = 0.
(6.1)
504
The D-gauge to fix dilatations (and to get the standard normalization of the Einstein term in
the Lagrangian) is given by
NI J X I X J = 1,
Ai d Ai = 2.
(6.2)
The S-gauge to fix S-supersymmetry is

X I NI J AiJ
= 0,
X I NI J iJ A = 0,
Ai d A = 0,
A d Ai = 0
(6.3)
where we have used the notation [9] NI J = 14 (FI J + FI J ) in terms of a homogeneous function
F (X I ) of degree two in X I . Here the subscripts I, J, . . . stand for the derivatives with respect to
X I , X J , . . . , respectively. The function F (X I ) obeys the relations
1
FI (X) = FI J (X)X J ,
F (X) = FI (X)X I ,
2
FI J K (X) = FI J KL (X)X L .
FI J K (X)X K = 0,
As a result, the S-supersymmetry parameters can be written down in terms of the N = 1/2
Q-supersymmetry parameter 1 as follows:

1
I

1I
I
2J
,
+ d
1A = 1 A NI J 1 2
8

1
2A = 1 A NI J 2I 1J ,
4
1
1
A
=
+2gd
1A A2 X A ,

2
A
= 2gd 1A A1 X A d 1 A .
Similarly, as regards the chiral U (1) rotations, we find
i
A = NI J 1 1I X J .
2
The important part of the superconformal tensor calculus is a construction of the (superconformally) invariant actions. As regards the N = 2 case, the invariant action for vector multiplets
is given by Eq. (3.9) of Ref. [9], whereas the invariant action of hypermultiplets is given by
Eq. (3.29) of Ref. [9]. The full invariant action for vector- and hypermultiplets coupled to N = 2
conformal supergravity is a sum of Eq. (3.9), a ChernSimons coupling (3.16), and Eq. (3.29) of
Ref. [9]. We use the same N = 2 superconformally invariant action as our starting point. Hence,
we are still in a position to fix the algebraic field equations of the auxiliary fields that follow from
the full action [9] after taking into account our constraint (Section 4).4
As regards the chiral U (1) gauge fields, we find

1

i
A = NI J X I X J X I X J NI J iI iJ d .
2
8
(6.4)
4 It is always assumed here that we are working in Euclidean space. Hence, the results of Ref. [9] are to be reformulated
in the Euclidean signature with the 2-component notation for spinorssee Section 3.
505
As regards the chiral SU(2) gauge fields, the V2 1 is already fixed (Section 4) by the constraint (4.9) whose solution is

1
V2 1 = 0 1 20 .
(6.5)
X
The remaining SU(2) gauge fields are given by (when i = 2 and j = 1)

i
i j
Vi j = d Aj Ai Aj Ai NI J j I i J + i NI J kI kJ . (6.6)
2
4
+ ij after the C-deformation (F 0+ = C ) is determined by the algebraic equation
The T
+ ij
+ J
NI J X I X J T
ij 4NI 0 X I C 4NI J X I F
1
8d FI J K X K ij iI j J = 0,
4
0 = 0) we find
whereas for the Tij
after the C-deformation (F
(6.7)
J
NI J X I X J Tij
ij 4NI J X I F
1
+ 8d + FI J K X K ij iI jJ = 0.
(6.8)
4
Finally, the vector multiplet auxiliary fields obey the algebraic field equations

1
1
NI J YijJ + gd Ak ki tI Aj
FI J K iJ jK + FI J K ik j l kJ lK = 0,
4
32
(6.9)
and

1
1
NI J Y ij J gd Aj ik tI Ak +
FI J K iJ jK + FI J K ik j l kJ lK = 0.
4
32
(6.10)
7. Lagrangian
A derivation of the full Lagrangian of the new N = 1/2 supergravity with vector- and hypermultiplets is now fully straightforward, so we merely present our final result.
Let be the covariant derivative with respect to the non-Abelian gauge transformations and
local Lorentz rotations, but not w.r.t. the chiral U (1) and SU(2) rotations. The gravitino field
2 is not independent, but a solution to Eq. (4.7). In our Lagrangian 2 is just the notation for
2A = 2Xi 0 ( 20 )A .
The N = 1/2 supergravity Lagrangian has the following structure:
L = Lkin + L4-fermi + Lcontact + Lgauge + LF + Laux ,
whose separate terms read as follows:
1
e1 Lkin = R d Ai Ai + NI J X I X J
2
+ ie1 i i
i
i
NI J iI iJ + NI J iI m iJ
4
4

+ 2id + 2id ,
(7.1)
506

1
NI J e1 i j ij k lJ kl
16

1
+ NI J e1 j I i ij lK k kl
16

1
1
+ NI J e1 i j ij i k lI X J k l X I X J kl
8
2

1
1
NI J e1 i j ij i lI k X J + k l kl X I X J kl
8
2

i
+ FI J K i kI lJ jK ij kl
48

i
FI J K kI i j K lJ ij kl
48

d i i + j ij

+ d i i + j ij
e1 L4-fermi = +

1
FI J KL iI kJ jK lL ij kl
192

1
FI J KL iI kJ j K lL ij kl ,
192

i
e1 Lcontact = + e1 i i NI J X I X J X I X J
4

i
+ e1 i j d Aj Ai Ai Aj
2
1
1
NI J i iJ X I NI J iJ i X I
2
2
+ 2d i Ai + 2d i Ai

1

1
+ NI J iI i j jJ NI J i iI j J j
8
8
i
i
+ FI J K iI iK X J + FI J K iK iI X J ,
16
16

i
3
1
1 I J
K
K L M
e Lgauge = gCI,J K e
W W W gfLM W W
6
8
2 i
2
+ 4g d A X X Ai g NI J fKL I X K X L fMN J X M X N
1
1
gNI J iI fKL J X K jL ij + gNI J j L X K fKL J iI ij
2
2
+ 4gd Ai j ij 4gd Ai j ij
4gd X + 4gd X
i
i
gNI J i j I ij fKL J X K X L gNI J jI i ij fKL J X K X L
2
2
+ 4igd i Aj X ij + 4igd i X Aj ij
+ igd i k Aj Ak ij + igd k i Ak Aj ij
j k 2gd j i X Ak Ai j k ,
+ 2gd i j Ai Ak X
507
where CI,J K are the real coefficient functions defined in terms of the input function F (X) by
considering the non-Abelian gauge transformations of the latter (with the gauge coupling constant g) [9]:
F = gFJ fI K J X K I igI CI,J K X J X K .
(7.2)
The coefficient functions CI,J K are symmetric in the last two indices, and obey a relation [9]
CI,J K + CJ,KL + CK,I J = 0.
(7.3)
The LF part of the Lagrangian (7.1) is given by

1
J GI
NI 0 C + NI J F J NI J G
4

1

+ N00 C + G0 + N0J F J C + G0
8

1
I
+ NI0 C + G0 + NIJ F J F
8
iI j J
1
+K
+
+ FI J K G+K
FI J 0 C + FI J K F
ij
32
J I ij
1
j i
FI J K F K + FI J K GK
32
1
1
I
(F00 F00 )C C (FI FI0 )F
C
32
16
1
I
J
(FIJ FIJ )e1 F
F
64
where we have introduced the non-Abelian vector field strength
e1 LF =
I
= WI WI gfJK I WJ WK
F
(7.4)
and the book-keeping notation

i iI
j ij iI j ij ( )
2

+ X I i j ij X I i j ij .
GI :=
(7.5)
Finally, the last term Laux in Eq. (7.1) is given by

1
1
e1 Laux = A A V i j V j i + NI J YijI Y ij J
4
8
+ ij 2
ij 2
1
1
NI J X I X J Tij
NI J X I X J T
ij .
64
64
The gravitino field 2 enters the Lagrangian (7.1) algebraically, so it may be eliminated via
its non-propagating field equation. The bosonic part of the matrix multiplying 2 in its field
equation has an inverse, due to the identity

1
2 C A g C A C D DA
(7.6)
A B = + C B
3
so that there is a unique solution for 2 .
508
8. Conclusion
We formulated the N = 1/2 supergravity coupled to both vector and scalar matter multiplets in four Euclidean dimensions. The gauge field of the local N = (0, 12 ) supersymmetry is
given by a single chiral gravitino. The C-deformation, originally introduced in the context of
supersymmetric D-branes with RR-type fluxes, was the main tool of our purely field-theoretical
construction. The new matter-coupled N = 1/2 supergravity with matter is not invariant under
the Euclidean analogue of Lorentz rotations in four dimensions, due to the explicit presence of a
fixed (self-dual) graviphoton background. In addition, the Lagrangian we constructed (Section 7)
is not Hermitian, thus hampering immediate physical applications of the proposed new supergravity theory. However, those are the common problems of all recently constructed N = 1/2
supersymmetric field theories, either with rigid or local N = 1/2 supersymmetry.
When compared to our earlier construction of the pure N = 1/2 supergravity without matter [11], in the matter-coupled N = 1/2 supergravity we observe no formation of the gravitino
condensate. Instead, we got one more constraint on the gravitini in terms of the matter fields.
Being the first construction of that type, our N = 1/2 supergravity with matter is unlikely to
be the most general one having a local N = 1/2 supersymmetry. The use of the N = 2 superconformal tensor calculus was essential in our construction, while we still have the vector multiplet
scalars parameterizing a special Khler manifold, and the hypermultiplet scalars parameterizing
a quaternionic (projective) manifold or its quaternionic quotient. It is rather straightforward to
generalize our results to the case of arbitrary (quaternionic) hypermultiplet couplings by using
the same N = 2 superconformal calculus and the results of Ref. [16]. However, we cannot exclude the existence of a much larger class of the invariant actions with merely N = 1/2 local
supersymmetry, which are not derivable from the N = 2 invariant actions we used. It would
be interesting to find such additional invariants for a construction of the most general N = 1/2
supergravity matter couplings.
We are unaware of any supergravity model to be constructed in a (curved) non-anticommutative superspace, so a relation of our N = 1/2 matter-coupled supergravity to the non-anticommutative superspace remains unclear to us.
It may also be of interest to study the conditions of further (spontaneous) breaking of local N = 1 supersymmetry by analyzing the vacuum expectation values of the supersymmetry
transformations of the fermionic fields.5
Acknowledgements
T.H. would like to thank Satoru Saito for useful discussions and encouragement.
This work is partially supported by the Japanese Society for Promotion of Science (JSPS)
under the Grant-in-Aid programme for scientific research, and the bilateral GermanJapanese
exchange programme under the auspices of JSPS and DFG (Deutsche Forschungsgemeinschaft).
Appendix A. Notation for hypermultiplets
We follow the notation introduced in Ref. [9]. In particular, the hypermultiplets (Ai , )
belong to a representation of the YangMills group. The scalars Ai obey a reality condition
5 See e.g. Ref. [17], as regards partial breaking of N = 2 local supersymmetry.
Ai Ai = ij Aj
509
(A.1)
where the matrices are used for raising and lowering of Greek indices. A consistency requires

.
=
(A.2)
When writing the action, it is convenient to introduce a matrix d as
d := [ ]
(A.3)
where is the real multiplication tensor in the sense

.
=
The matrix
(A.4)
has the properties
d d = d
(Hermitian),
d = d
(quaternionic).
(A.5)
Appendix B. N = 2 superconformal transformation laws

In this appendix we summarize the relevant part of the N = 2 superconformal transformation
laws [9] in our 2-component spinor notation, in four Euclidean dimensions:
(i) as regards the N = 2 Weyl multiplet (physical) components,
e a = i i a i i i a i ,

1 mn
1
iA = +2 iA
mn A B iB + iA iA + V i j j A
2
2

1 ij m A
1
A
iTm j + i i iA iA + i j j A ,
2
2

1 mn
1
B
i A = +2 i A mnA i B i A i A + Vi j j A
2
2

1 +
1
m j A i( i )A + iA i A + i j j A ;
+ iTmij
2
2
(ii) as regards the N = 2 vector multiplet components,
X I = i iI iA X I ,
X I = +i iI + iA X I ,
= i i jI ij
(B.3)
(B.5)
i
1
+ 2X I iA iA iI A + i j jI A ,
2

i A
1
iI
A

= 2i
( i A )X I gfJ K I WJ X K + k kI
2
(B.2)
(B.4)
+ i j I ij + 2 i j X I ij 2i j X I ij ,

A
1 k I
I J K
IA
I
( + iA )X gfJ K W X k
i = 2i i
2
I J K
I jA
mnA
I
jB
2gfJ K X X ij j A
+ Yij +
B Fmn ij
WI
(B.1)
(B.6)
(B.7)
+I ik
2gfJ K I X J X K ik kA
+ Y ikI kA kB mn B A Fmn
1
+ 2X I i A + iA iI A + i k kI A ;
2
(B.8)
510
(iii) and as regards hypermultiplets,

Ai = 2 i 2 ij j + i j Aj ,
= 2 2 j + j A ,

i
j
A = +i i A D Ai + 2gAi X ij A + Ai iA + A A ,
2

i
i
A = +i i A D Ai + 2gX Ai ij j A + Ai A
A A ,
2
where we have used the following abbreviations:
A
ij
(B.9)
(B.10)
(B.11)
(B.12)
1
F I := F I + G I ,
F I := F I X I Tij ij ,
(B.13)
4
1
D Ai := Ai + Vi j Aj gW Ai + i + ij j ,
(B.14)
2
1
D Ai = Ai + Aj V i j gAi W + i + ij j .
(B.15)
2
As for the auxiliary fields, we have e.g.

V i j = 3i i j il j m m l + 2 i j i j + il j m l m l ] m

1
i j 3i k k k k + 2 k k + k k k k k k .
2
(B.16)
Appendix C. Decomposition rules
After the C-deformation (Section 3) and the gauge-fixing of the truly superconformal symmetries (Section 6), the Q-supersymmetry transformations are modified by the compensating
S-supersymmetry-, chiral U (1)- and chiral SU(2)-transformations, whose parameters are given
by

k
iA = +2gd ij A
Ak X Aj + d ij j A

1

1 j I
j
I
jJ
kJ
(C.1)
,
+ i d
j A NI J i i k
4
2

i
ij
ij
A
= +2gd k A Aj X A + d j A

j 1

1 i I

I
iJ
kJ
i
(C.2)
+ A NI J j j k
+ j d ,
4
2

1
A = + iNI J i Ii X J + i iI X J ,
(C.3)
2

1

2
i j = +2
d B j i j i B k k
c
2

k
1
j
d Bi i Bk
,
(C.4)
2
where the convenient parametrization is given by [9]
511

i I := iI Z I i0 =
IiA
0

,

1
c := 4 Ai a Ai a
= d Ba Ba ,
a, b = 1, 2,
, = 3, . . . , 2r.
Z I :=
XI
,
X0

:= Ba a =
Ba := A1i a Ai ,
0
A

,
(C.5)
References
[1] P. van Nieuwenhuizen, Phys. Rep. 68 (1981) 189.
[2] S.W. Hawking, M. Rocek (Eds.), Superspace and Supergravity, Cambridge Univ. Press, 1981;
S.J. Gates Jr., M.T. Grisaru, M. Rocek, W. Siegel, Superspace or One Thousand and One Lessons in Supersymmetry,
Benjamin/Cummings Publ., 1983;
A. Salam, E. Sezgin (Eds.), Supergravities in Diverse Dimensions, Elsevier/World Scientific Publ., 1989.
[3] A.H. Chamseddine, G. Felder, J. Frlich, Commun. Math. Phys. 155 (1993) 205, hep-th/9209044;
J. Madore, J. Mourad, Int. J. Mod. Phys. D 3 (1994) 221, gr-qc/9307030;
M. Chaichian, P.P. Kulish, K. Nishijima, A. Tureanu, Phys. Lett. B 604 (2004) 98, hep-th/0408069;
P. Aschieri, C. Blohmann, M. Dimitrijevic, F. Meyer, P. Schupp, J. Wess, Class. Quantum Grav. 22 (2005), hepth/0504183.
[4] H. Nishino, S. Rajpoot, Phys. Lett. B 532 (2004) 334, hep-th/0107216;
B.M. Zupnik, Class. Quantum Grav. 24 (2007) 15, hep-th/0512231.
[5] L. Alvarez-Gaum, F. Meyer, M.A. Vazquez-Mozo, Nucl. Phys. B 753 (2006) 92, hep-th/0605113.
[6] H. Ooguri, C. Vafa, Adv. Theor. Math. Phys. 7 (2003) 53, hep-th/0302109;
H. Ooguri, C. Vafa, Adv. Theor. Math. Phys. 7 (2004) 405, hep-th/0303063.
[7] N. Seiberg, JHEP 0306 (2003) 010, hep-th/0305248.
[8] T. Hatanaka, S.V. Ketov, S. Sasaki, Phys. Lett. B 619 (2005) 352, hep-th/0504191;
T. Hatanaka, S.V. Ketov, Y. Kobayashi, S. Sasaki, Nucl. Phys. B 716 (2005) 88, hep-th/0502026;
T. Hatanaka, S.V. Ketov, Y. Kobayashi, S. Sasaki, Nucl. Phys. B 726 (2005) 481, hep-th/0506071.
[9] B. de Wit, J.W. van Holten, A. van Proeyen, Nucl. Phys. B 167 (1980) 186;
B. de Wit, P.G. Lauwers, R. Philippe, S.Q. Su, A. van Proeyen, Phys. Lett. B 134 (1984) 37;
B. de Wit, A. van Proeyen, Nucl. Phys. B 245 (1984) 89;
B. de Wit, P.G. Lauwers, A. van Proeyen, Nucl. Phys. B 255 (1985) 569.
[10] R. DAuria, S. Ferrara, P. Fr, Nucl. Phys. B 359 (1991) 705;
L. Andrianopoli, M. Bertolini, A. Ceresole, R. DAuria, S. Ferrara, P. Fr, Nucl. Phys. B 476 (1996) 397, hep-th/
9603004.
[11] T. Hatanaka, S.V. Ketov, Class. Quantum Grav. 23 (2006) L45, hep-th/0602115.
[12] S. Ferrara, P. van Nieuwenhuizen, Phys. Rev. Lett. 37 (1976) 1669.
[13] L. Andrianopoli, R. DAuria, S. Ferrara, Nucl. Phys. B 628 (2002) 387, hep-th/0112192.
[14] A. van Proeyen, Superconformal tensor calculus in N = 1 and N = 2 supergravity, in: Supersymmetry and Supergravity 1983, World Scientific, 1983, p. 578.
[15] P. van Nieuwenhuizen, A. Waldron, Phys. Lett. B 389 (1996) 29, hep-th/9608174;
P. van Nieuwenhuizen, A. Waldron, A continuous Wick rotation for spinor fields and supersymmetry in Euclidean space, in: Gauge Theories, Applied Supersymmetry and Quantum Gravity, World Scientific, Singapore, 1997,
p. 394, hep-th/9611043;
U. Theis, P. van Nieuwenhuizen, Class. Quantum Grav. 18 (2001) 5469, hep-th/0108204.
[16] B. de Wit, B. Kleijn, S. Vandoren, Nucl. Phys. B 568 (2000) 475, hep-th/9909228.
[17] S. Cecotti, L. Girardello, M. Porrati, Phys. Lett. B 168 (1986) 83;
S. Ferrara, L. Girardello, M. Porrati, Phys. Lett. B 366 (1996) 155, hep-th/9510074;
S. Ferrara, L. Girardello, M. Porrati, Phys. Lett. B 376 (1996) 275, hep-th/9512180;
L. Andrianopoli, R. DAuria, S. Ferrara, M.A. Lledo, JHEP 0301 (2003) 045, hep-th/0212236;
H. Itoyama, K. Maruyoshi, Int. J. Mod. Phys. A 21 (2006) 6191, hep-th/0603180.

Reduced Hamiltonian for intersecting shells

Francesco Fiamberti a , Pietro Menotti b,
a Dipartimento di Fisica, Universit di Milano and INFN, Sezione di Milano, via Celoria 16, I-20133, Italy
b Dipartimento di Fisica, Universit di Pisa and INFN, Sezione di Pisa, Largo B. Pontecorvo 3, I-56127, Italy
Received 10 September 2007; accepted 7 November 2007

Abstract
The gauge usually adopted for extracting the reduced Hamiltonian of a thin spherical shell of matter in
general relativity, becomes singular when dealing with two or more intersecting shells. We introduce here a
more general class of gauges which is apt for dealing with intersecting shells. As an application we give the
Hamiltonian treatment of two intersecting shells, both massive and massless. Such a formulation is applied
to the computation of the semiclassical tunneling probability of two shells. The probability for the emission
of two shells is simply the product of the separate probabilities thus showing no correlation in the emission
probabilities in this model.
PACS: 11.15.Kc; 04.60.-m; 04.70.Dy
Keywords: Black hole; Semiclassical
1. Introduction
A lot of work has been done in the subject of thin spherical shells of matter in general relativity [1,2] and several applications given [35]. An interesting application of the mechanics of
thin shells has been the semiclassical treatment of the black hole radiation [68]. An important
role in such a treatment is played by the choice of the gauge. In the original treatment [6] a gauge
was adopted in which the radial component of the metric is equal to the radial coordinate except
for an arbitrarily small region around the shell. Such a gauge choice was put in mathematically
Work supported in part by M.U.R.S.T.
* Corresponding author at: Largo B. Pontecorvo 3, 56127 Pisa, Italy. Tel.: +39 050 2214 885; fax: +39 050 2214 887.
E-mail addresses: francesco.fiamberti@mi.infn.it (F. Fiamberti), menotti@df.unipi.it (P. Menotti).

doi:10.1016/j.nuclphysb.2007.11.003
F. Fiamberti, P. Menotti / Nuclear Physics B 794 [PM] (2008) 512537
513
clear terms in [9,10] where the reduced canonical momentum is extracted through a well-defined
limiting process. However such a limit gauge becomes singular when two or more shells are
present and intersect. In addition also in the simple instance in which only one shell is present,
it would be nice to have a procedure in which no limit process is present as the gauge choice
should be completely arbitrary provided it is free of coordinate singularities. Moreover even in
the one shell case the procedure for extracting the canonical momentum is usually considered as
a very complicated one. Here we shall give a treatment which greatly simplifies the derivation of
the reduced action, does not require any limiting procedure and can be applied to the treatment
of two or more shells which intersect.
All the problem is to derive the reduced action, i.e., an action in terms of the shell coordinates
and some appropriate conjugate momenta. We shall keep the formalism for the two shell case as
close as possible to the one shell treatment.
We shall perform all the treatment for a massive dust shell and the simpler case of a null
shell can be derived as a particular case. As an application of the developed formalism we rederive the well-known Drayt Hooft and Redmount relations for the intersection of light-like
shells [11,12].
The main motivation which originated the works [6,8,13] is the computation of the semiclassical tunneling amplitude, which is related to the black hole radiation in which one takes into
account the effect of energy loss by the black hole. It is remarkable that such a simple model
reproduces all the correct features of the Hawking radiation, giving also some corrections due
to energy conservation. For approaches more kinematical in nature see, e.g., [14]. The adoption of more general gauges allows the dynamical treatment of two intersecting shells without
encountering singularities. The formalism developed here is applied to the computation of the
semiclassical emission probability of two shells; it was in fact suggested [6] that correlations
could show up in the multiple shell emission. We find however that such a model gives no correlation among the probabilities of the two emitted shells.
The paper is structured as follows.
In Section 2 we lay down the formalism by exploiting some peculiar properties of a function F
strictly related to the momenta canonically conjugate to the metric functions. It is then just a
simple matter of partial integrations to extract the reduced canonically conjugate momentum
without employing any limit process. This is done in Section 3. A new term containing the
time derivative of the mass of the remnant black hole appears; if such a mass is considered as
a datum of the problem the result agrees with those of the original massless case of Kraus and
Wilczek and with the massive case result obtained through a limiting process in [9]. Then we
discuss the derivation of the equations of motion. In [6] and [9] the variational principle was
applied by varying the total mass of the system which we shall denote by H . In [8] the attitude
was adopted of keeping the total mass of the system fixed and varying instead the mass of the
remnant black hole. Here we show that both procedures can be applied to obtain the equations
of motion; depending however on the choice of the gauge one procedure is far more complicated
than the other and we give them both.
In Section 4 we discuss more general gauge choices and derive the equation of motion in
the inner gauge. We shall not consider in the present paper complex gauges or complex gauge
transformations [15].
In Section 5 we discuss the analytic properties of the conjugate momentum; this is of interest
because the imaginary part of the conjugate momentum is responsible for the tunneling amplitude and in determining the Hawking temperature both via a simple mechanical model or more
precisely by working out the semiclassical wave functions on which to expand the matter quan-
514
tum field [16]. We remark that the result for the tunneling probability is independent of the mass
of the shell but depends only on the initial and final energy of the black hole.
In Section 6 we extend the treatment to two shells in the outer gauge writing down the explicit
expression of the two reduced canonical momenta.
In Section 7 we derive the equations of motion for both shells from the reduced two shell
action. For simplicity this is done for the massless case; the general treatment for massive shells
is given in Appendix C.
In Section 8 the developed formalism is applied to give a very simple treatment of two
shells which intersect. In the massless case we rederive the well-known relations of Dray and
t Hooft [11] and Redmount [12].
In Section 9 we consider the problem of computing the tunneling probability for the emission
of two shells which in the process can intersect. To this end one has to compute the imaginary
part of the action along the analytically continued solution of the equations of motion. In this
connection a helpful integrability result is proven which allows to compute the action along a
specially chosen trajectory on the reduced coordinate space. Such a result allows on one hand to
prove the independence of the result from the deformation defining the gauge and on the other
hand it allows to compute explicitly such imaginary part. The final outcome is that in all instances
(massive or massless shells) the result again depends only on the initial and final values of the
masses of the black hole and the expression coincides with the one obtained in the one shell
case. The interest in studying the two shell system was pointed out in [6] in order to investigate
possible correlations among the emitted shells. Here we simply find that the two shells, even
if they interact with an exchange of masses, are emitted with a probability which is simply the
product of the probabilities for single shell emissions and thus that the model does not predict
any correlation among the emitted shells.
In Section 10 we summarize the obtained results.
2. The action
As usual we write the metric for a spherically symmetric configuration in the ADM form
[3,6,9]

2
ds 2 = N 2 dt 2 + L2 dr + N r dt + R 2 d 2 .
(1)
We shall work on a finite region of spacetime (ti , tf ) (r0 , rm ). On the two initial and final
surfaces we give the intrinsic metric by specifying R(r, ti ) and L(r, ti ) and similarly R(r, tf ) and
L(r, tf ).
The complete action in Hamiltonian form, boundary terms included is [3,9,17]
tf
S = Sshell +
dt
ti
tf
+
ti
rm

dr L L + R R NHt N r Hr
r0

N RR rm
r
dt N L L +
,
L
r0
(2)
515
where
tf
Sshell =
dt p r .
(3)
ti
Ht and Hr are the constraints which are reported in Appendix A, r is the shell position and
p its canonical conjugate momentum. Action (2) is immediately generalized to a finite number
of shells. Sshell as given by Eq. (3) refers to a dust shell even though generalizations to more
complicated equations of state have been considered [2,3,18].
Varying the action w.r.t. N and N r gives the vanishing of the constraints while the variations
w.r.t. R, L, R , L give the equations of motion of the gravitational field [3,9] which for completeness are reported in Appendix A. The functions R, L, N , N r are continuous in r while R ,
L , N , N r , L , R can have finite discontinuities at the shell position [9]. In [9] it was proven
that the equation of motion of a massive shell cannot be obtained from a true variational procedure, in the sense that the obtained expression for p is discontinuous at r = r . In the same
paper it was remarked that for consistency the average value of the r.h.s. of the equation for p
has to be taken. For the readers convenience we give in Appendix A the explicit proof that the
equations of motion of matter, i.e., r and p can be deduced from the already obtained equations
of motion for the gravitational field combined with the constraints. The equation for r does not
pose any problem while for the equation for p one deduces algebraically that the r.h.s. contains
the average of the derivatives of L, N , N r across the shell. In Appendix A we also show directly
that such discontinuity is absent in the massless case [10].
Already in [3,6] it was pointed out that in the region free of matter, as a consequence of the
two constraints the quantity M
L2
R RR 2
+
2R
2
2L2
is constant in r and this allows to solve for the momenta [3,6]

2
2M
R
1+
RW,
L = R
L
R
M=
(4)
(5)
L[(R/L)(R /L) + (R /L)2 1 + M/R]

.
(6)
W
In the one shell problem we shall call the value of such M, M for r < r and H for r > r . The
function

2
2
2M
R
2M
R
R

1+
1+
+ RR log
R f (R) (7)
F = RL
L
R
L
L
R
R =
has the property of generating the conjugate momenta as follows

F
,
L
F
F
F
R =
.
=
R
R r R
L =
(8)
(9)
The total derivative fr(R) = R f (R) of the arbitrary function f (R) does not contribute to the
momenta. The function F will play a major role in the subsequent developments. A large freedom
516
is left in the choice of the gauge. With regard to L we shall adopt the usual gauge L = 1. It will
be useful in the following developments to choose the arbitrary function f (R) such that F = 0
for R = 1. Such a requirement fixes f (R) uniquely:

F (R, R , M) = RW (R, R , M) + RR L(R, R , M) B(R, M) ,
(10)
where

2M
R 2 1 +
;
R

L(R, R , M) = log R W (R, R , M)
W (R, R , M) =
(11)
and

B(R, M) =

2M
2M
+ log 1
.
R
R
(12)
The function F (R, R , M) has the following useful properties

F (R, R , M)
= R(L B);
R
L
1
= .
R
W
(13)
Other properties of F will be written when needed.

In the following section we shall choose R = r for r > r and also R = r for r < r l so that
F vanishes identically outside the interval r l < r < r . We shall call this class of gauges outer
gauges. In Section 4 we shall consider more general gauges, e.g., the gauge R = r for r < r
and also R = r for r > r + l which we shall call inner gauges. However contrary to what is
done in [6,9] we will not take any limit l 0 and prove that the results are independent of the
deformation of R in the region r l < r < r (or r < r < r + l for the inner gauges) provided
R satisfies the constraint at r = r . We shall call these regions (r l, r ) for the outer gauge and
(r , r + l) for the inner gauge, the deformation regions.
The variation of S which produces the equations of motion has to be taken, as it is well
known, by keeping the metric and in particular R and L fixed at the boundaries. The variation of
the boundary terms gives
N r (rm )L (rm ) + N r (r0 )L (r0 ).
The N, N r can be obtained from the two equations of motion for the gravitational field

L
L R
R = N
+ Nr ;
+ N r R.
0=N
R
R
R2
(14)
(15)
Using
the deformation region (r l, r ) we have N r =
these it is easily proved that for r outside

2M
r
N 2H
r for r > r , N = const and N = N
r for r < r l, N = const where the two constants
as a rule differ. Thus the variation of the boundary term is
N(rm )H + N (r0 )M.
(16)
In the next section we shall connect N (rm ) with N (r0 ) being N (r) not constant in the deformation region.
517
3. The one shell effective action in the outer gauge

As outlined in the previous section we shall choose
V (t)
R(r, t) = r +
r (t)
r
0

V (t)
r r (t) dr = r +
g r r (t)
r
(17)
having support (l, 0), (0) = 1 and smooth in l and

0
(r) dr = 0.
(18)
As a consequence the deformation g(r) has support in (l, 0) and g (0 ) = 1. Such R satisfies the discontinuity requirements at r = r which are imposed by the constraints. In fact the
constraints (see Appendix A) impose the following discontinuity relations at r (we recall that we
chose L 1)

V
R = ; V = p 2 + m2
(19)
R
and
L = p.
(20)
In the outer gauge the bulk gravitational action becomes

tf
Sg =
(21)
Ig dt,
ti
where, keeping in mind that L 1

F
F
R dr
Ig = (L L + R R) dr = R R dr =
R r R
r0
r0
r (t)
=
dF
dr M(t)
dt
r0
d
=
dt
r (t)
r0
r0

F
F r (t)
R
dr
M
R r0

r (t)
r (t)
F
F
F dr M(t)
,
dr r (t)F +
R
M
R r (t)
r0
(22)
r0
where we used the fact that F vanishes at r = r0 .

Adding Ishell = p r and neglecting the total time derivative which does not contribute to the
equations of motion, we obtain for the reduced action in the outer gauge

tf
r (t)

F
r
m
dr + N r L + N RR r dt,
pc r M(t)
0
M
ti
r0
(23)
518
where using (19), (20)

1 F
pc = F r (t)
+ p
R

r (t) R r (t)

r + p 2 + m2 p 2H r
= 2M r 2H r r log
(24)
.
r 2M r
A few comments are in order: (1) No limit l 0 is necessary for obtaining pc of Eq. (24) which
holds for any deformation g. (2) The M(t)

term is important, as we shall see, if we consider the
variational problem in which M is varied. On the other hand, if we consider M as a datum of the
problem and vary H the contribution M(t)

is absent. (3) p in Eq. (24) is a function of r , H and
M as given by the discontinuity relation (20) equivalent to the implicit equation

2H
m2
p
.
H M =V +
(25)
2r
r
We discuss now these issues in more detail. Let us consider at first M as a datum of the problem
and vary H . As shown in Appendix A in order to be consistent with the gravitational equations
M has to be constant in time. This is the situation examined in [9] where the expression (24)
for pc was derived by a limit process in which l 0. From Eq. (23) we see that the equation of
motion for r is given by
pc
N(rm ) = 0,
r
(26)
H
where N(rm ) can be replaced by N (r ) being N (r) in the outer gauge constant for r > r . The
computation of Eq. (26) keeping in mind the implicit definition of p Eq. (25) gives the correct
equation of motion for the massive shell

p
2H
p
r
r = N (r ) N (r ) =
(27)
N (r ).
V
V
r
The outline of the calculation is done in Appendix B.
Alternatively one can consider H = const as a datum of the problem and vary M(t). We
remark that as shown in Appendix A the datum M or H is consistent with the gravitational
equations only if H and M are constant in time. Nevertheless in the variational problem H and
M have to be considered as functions of time, because the constraints tell us only that M and H
are constant in r. Only after deriving the equation of motion one can insert the consequences of
the gravitational equations of motion.
The variation of M(t) in the outer gauge is a far more complicated procedure, due to the
presence of M(t),
but gives rise to the same result obtained by varying H and keeping M fixed.
As this will be useful to understand the two shell reduced dynamics to be developed in Section 6
we go into it with some detail. In this case the M term plays a major role; in fact the equation of
motion now takes the form
d
pc
+
r
M dt
r (t)
r0
F
L
dr + N r (r0 )
(r0 ) = 0,
M
M
(28)
where due to the vanishing of g(x) for x < l we have L (r0 ) = 2Mr0 . N r (r) on the other
hand is obtained by solving the two coupled equations (15) with the condition that for r > r ,
N r (r) equals
2H
r
519
, having normalized N = 1 for r > r . One easily finds that for r < r
r
N (r) = W
r
R
2H r
R
dr +
.
M
2H r + p
(29)
Taking into account that

2

F
2F
R
d
+
=

dr R M
MR
M
(30)
we have
d
dt
r (t)
F
dr =
M
r0
r
r0

F
2F
R
dr + r
+
R
R
M
M MR r
and Eq. (28) becomes

2F
2H r
F
p
c
+
+ r
+ R
r
= 0.
M
M
R M r
2H r + p
(31)
(32)
From the expression for pc given by the first line of Eq. (24) we obtain

2H r
r p V R
(33)
+
= 0,

M W M r
2H r + p

+ pr and R (r ) = 1 + V /r . Using
where W (r ) = R 2 (r ) 1 + 2M/R(r ) = 2H
r
1
p

=
p
M
2H
V
r
(34)
we obtain Eq. (27) again. We remark once more that no limit process l 0 is necessary for all
these developments.
To summarize, in the present section we have derived the reduced action for the one shell
problem in the outer gauge with an arbitrary deformation. One can vary H (the exterior ADM
mass) considering the interior mass as given, or one can vary the interior mass M considering
the exterior mass H as given, or even one can vary both M and H always obtaining the correct
equations of motion. Whenever M is varied the M term in Eq. (28) plays a crucial role. All the
results do not depend on the deformation g.
4. More general gauges
It is of interest to examine more general gauges given by Eq. (17) where g does not necessarily
vanish for positive argument, i.e., we can consider g(x) with g(x) = 0 for |x| > l, g(0) = 0 and
g (+0) g (0) = 1, thus again satisfying the constraint (19). In this case F (R, R , H ) does
not vanish identically for r > r and the bulk action (21) is given by the time integral of
d
Ig =
dt
rm
r0
F dr M(t)
r (t)
r0
F
dr H (t)
M
rm
r (t)

F
F r (t)+
.
dr + r (t)F +
R
H
R r (t)
(35)
520
The pc is easily computed using Eq. (35) and one gets immediately the general form for the
canonical momentum pc
pc = r (L B),
(36)
where L = L(r + ) L(r ) and similarly for B and the reduced action becomes
tf
S=
pc r M(t)
ti
r (t)
F
dr H (t)
M
rm
r (t)
r0

F
r
r m
dr + N L + N RR r dt.
0
H
(37)
We shall call inner gauge the one characterized by g(x) = 0 for x < 0.
Due to the similarity with the treatment of Section 3 we shall go through rather quickly. Now
F vanishes identically for r < r and the action takes the form

tf
S=
dt
pci r
rm
r (t)
ti

F
r
r m
dr + N L + N RR r ,
0
H
where pci is given by

R F
i
pc = F +
+ p
r R r +
(38)
(39)
whose explicit value is

pci
2M r

2H r r log
2H r
r V + p 2M r
r

(40)
and p,
again determined by the discontinuity equation (20), is given by the implicit equation

2M
m2
p
H M =V
(41)
2r
r
which is different from Eq. (25). Now the simple procedure is the one in which one varies M
keeping H as a fixed datum of
the problem. This time the solution of the system Eq. (15) gives
simply N = const and N r = N
2M
r
for r0 < r < r .
5. The analytic properties of pc

We saw that in the outer gauge
pc =
2M r

r + V p 2H r
2H r r log
.
r 2M r
The solution of Eq. (25) for p is

2H
A
A2 (1
p
r
=
r
1 2H
r
2H m2
)
r r 2
(42)
(43)
521
where
A=
H M
m2
2.
r
2r
(44)
If we want p to describe an outgoing shell we must choose the plus sign in front of the square
root. Moreover the shell reaches r = + only if H M > m as expected.
The logarithm has branch points at zero and infinity and thus we must investigate for which
values or r such values are reached. At r = 2H , p has a simple pole with positive residue; then
the numerator goes to zero and below 2H it becomes

V
2H
p
1
(45)
,
r
r
r
where here V is the absolute value of the square root. Expression (45) is negative irrespective
of the sign of p and stays so because p is no longer singular. In order to compute the tunneling
amplitude, below r = 2H we have to use the prescription [8] r 2H r 2H i and as a
consequence the pc below 2H acquires the imaginary part i r . Below r = 2M the denominator of the argument of the logarithm in Eq. (42) becomes negative so that the argument of the
logarithm reverts to positive values. Thus the classically forbidden region is 2M < r < 2H
independent of m and of the deformation g and the integral of the imaginary part of pc for any
deformation g is
2H

Im pc dr =

2
2
r dr = 2 H M = 4 M +
(46)
2M
with = H M which is the ParikhWilczek result [8]. A more profound way to relate the
result for pc to the formula for the Hawking radiation is to use (24) to compute semiclassically
the modes on which to expand the quantum field, and then proceed as usual by means of the
Bogoliubov transformation. This was done in [6,16]. A more direct particle like interpretation of
(46) was given, e.g., in [19].
Similarly one can discuss the analytic properties of the conjugate momentum pci in the inner
gauge. We have

1 2H
r
i

pc = 2M r 2H r r log
(47)
.
p
1 Vr 2M
+
r
r
This time the solution of Eq. (41) gives for p

2
2M
A
A2 (1 2M
) mr 2
p
r
r
,
=
r
1 2M
r
(48)
where now
A=
m2
H M
+ 2.
r
2r
(49)
Notice that for m = 0 we have pc = pci . To describe an outgoing shell the square root in (48) has
to be taken with the positive sign.
522
The main point is the discussion of the sign of the term

V
2M p
2H

1
(50)
+ = R (r + ) R 2 (r + ) 1 +
,
r
r
r
r
where

2M p
2H
= R 2 (r + ) 1 +
.
(51)
r
r
r
For m = 0 at r = 2H Eq. (51) is negative so that at r = 2H

V
2M p
2H

1
(52)
+ = R (r + ) + R 2 (r + ) 1 +
r
r
r
r
is positive, being in Eq. (52) the square root on the r.h.s. understood as the positive determination
of the square root. The same happens for m = 0 provided m < H M which is the condition for
> 0 such a term stays positive
the shell to be able to reach r = +. For r < 2H as 1 + 2H
r
irrespective of the sign of R , until R diverges. This happens at 2M where p given by Eq. (48)
has a simple pole with positive residue. Thus below 2M Eq. (52) reverts to

V
2M p
2H

1
(53)
+ = R (r + ) R 2 (r + ) 1 +
r
r
r
r
which is negative irrespective of the sign of R (r + ). The conclusion is that pci takes the imaginary part i r in the interval 2M, 2H as it happens for pc .
6. The two shell reduced action
From now on we shall work in the outer gauge. We denote with r1 and r2 the coordinates
of the first and second shell r1 < r2 . The value of M for r < r1 will be denoted by M, for
r1 < r < r2 by M0 and for r > r2 will be denoted by H as before. In extending the treatment to
two interacting shells we shall keep the formalism as close as possible to the one developed in
Section 3. The most relevant difference is that in any gauge the intermediate mass M0 and the
total mass H always intervene dynamically. We shall consider the mass M as a datum of the
problem.
For the metric component R we shall use
V2
V1
; v1 =
,
(54)
R(r2 )
R(r1 )
where h(x) has the same properties of g(x) described in Section 3 (actually we could use the
same function). Both g and h vanish for positive argument and thus we are working in an outer
gauge according to the definition of Section 3.
The action is given by the time integral of

I = p 2 r 2 + p 1 r 1 + dr R R + b.t.
(55)
R(r) = r + v2 g(r r2 ) + v1 h(r r1 );
v2 =
Breaking the integration range from r0 to r1 and from r1 to r2 and using Eq. (9) for R and the
same technique as used for the one shell case, we reach the expression
d
p 2 r 2 + p 1 r 1 +
dt
r2
r0
F dr M 0
r2
r1
F
dr r 2 F (r2 )
M0
523

F
F
R (r2 ) + r 1 F (r1 ) + R (r1 )

R
R
d
= p 1 r 1 +
dt
r2
F dr M 0
r2
r1
r0

F
F
0
r 2 + r 1 F (r1 ) + R (r1 ),
dr + pc2
M0
R
(56)
0 is given by Eq. (24) with M replaced by M , and stays for the jump across the
where pc2
0
discontinuity at r1 . We recall that F depends only on R, R and M and the partial derivative w.r.t.
these variables will have the usual meaning. For the other quantities appearing in the calculations
we recall that the independent variables are M, M0 , H , r1 , r2 and when taking partial derivatives
we shall consider the remaining variables as fixed. In Eq. (56) we have, using Eq. (13), denoting
with the discontinuity across r1 and with the bar the average at r1

r 1 p 1 + r 1 F + F R
R
F
F
R + R
= r 1 p 1 + r 1 F +
R
R

+ RR(
+ RR(L
L B)
= r 1 R R(L B) + R R(L B)
B)
T

T

D,
D + M 0 R(r1 ) r1
= r 1 pc1 + r 2 p c2 + H R(r1 ) r1
H
M0
(57)
where
D = R(L B);
pc1 = R (r1 + )D;
T = log v2 ;
T

d
p c2 = R (r1 + ) 1 D + R(r1 ) r1
R(r1 ) r1 D
D=
r2
d r2
(58)
(59)
having used
R = r 1 R
(60)
dT

v1
.
R = r 1 r 2 (R 1) + R(r1 ) r1
2
dt
(61)
and
Summing up the reduced action for the two shell system is given, boundary terms included, by
the time integral of
T

T

D
D + M 0 R(r1 ) r1
r 1 pc1 + r 2 pc2 + H R(r1 ) r1
H
M0
d
+
dt
r2
r0
F dr M 0
r2
r1
r

F
dr + N r L + N RR rm ,
0
M0
(62)
0 + p with p 0 given by Eq. (24) with M replaced by M and p by Eq. (59).

where pc2 = pc2
c2
0
c2
c2
With regard to Eq. (62) we notice that irrespective of the gauge used both terms in M 0 and H
appear in the action. Moreover pc1 and pc2 depend both on r1 and r2 .
524
7. The two shell equations of motion

From the reduced action (62) we can derive the equations of motion for the two shells. This
is of some importance in order to show the consistency of the scheme. We recall that action (62)
has been derived in the outer gauge, i.e., R(r) = r for r > r2 (r2 > r1 ). While in the one shell
problem formulated in the outer gauge H does not appear, in the two shell problem it is always
present. Again we consider M = const as a datum of the problem.
In the variational procedure we can vary r1 , r2 , H and M0 independently. We shall start
varying H but keeping all other parameters fixed. For the sake of simplicity we shall deal here
with the massless case m1 = m2 = 0. In Appendix C we give the general derivation for m1 and
m2 different from zero. The most important fact that occurs when we vary H keeping M0 fixed is
that the terms proportional to r 1 cancel in the variation. The simplifying feature of the massless
case is that L = 0 so that in Eq. (58) D = RB, being B function only of R and M and not
of R and thus only of R and M0 being M a datum of the problem. The coefficient of r 1 , taking
into account the following relation, easily derived from (13),
B
1
F
+ RR
=
M
M W R
(63)
is found proportional to
T D

T

R (r1 + )D R (r1 + ) 1
D R(r1 ) r1
R (r1 + ),
H
H
H R
where we used

dT
dR(r1 )
= r 1 R (r1 + ) r 2 R (r1 + ) 1 +
R(r1 ) r1
dt
dt
(64)
(65)
and we took into account that T does not depend on r1 and that on the equations of motion
H = M 0 = 0. Now employing the relations

R (r1 + )
T
(66)
=
R (r1 + ) 1
H
H
we see that expression (64) vanishes. Thus we are left only with the r 2 terms. We know already
0 term give the correct equation of motion and thus we have
that the boundary term and the pc2
simply to prove that the r 2 terms originating from

T
d
(67)
R(r1 ) r1
D
dt
H

R(r1 )
T
=
R(r1 ) r1 ;
H
H
cancel
p c2
H ,
i.e.,
T
T

D
R (r1 + ) 1 D + R(r1 ) r1
D = 0.
R(r1 ) r1
H
r2
r2
H
(68)
Using relations (66) we have that Eq. (68) is satisfied. Such a result is expected as the exterior
shell parameterized by r2 moves irrespective of the dynamics which develops at lower values
of r until r1 crosses r2 .
Now we vary M0 keeping H fixed. We have no boundary term contribution because also M
is constant and we find the equation
525

r 1 pc1 + r 2 pc2 d R(r1 ) r1 T D + r 2 F (r2 )
M0
M0 dt
M0
M0
F (r1 + )
r 1
+
M0
r2
r1

2 F r2
R

R
dr + R
= 0.
M0
M0 R r1 +
(69)
Using
expression
(29) for N r (r), relation (15) for N (r) and relation (63) we obtain with =
2H r /( 2H r + p)
1

r 1 + N r (r1 ) N (r1 )
R (r1 + ) W (r1 + )

0
pc2
F
2F
R
+ r 2
+
+
+ r 2
M0
M0 M0 R r2
1 R

(r1 + )
R (r1 + ) W (r1 + )
W

T
p c2

2 F
r 2
D = 0.
+ r 2
R(r1 ) r1
M0 M0 R r1 +
r2
M0
(70)
The first line is the equation of motion for r1 , the second line vanishes being the equation of
motion for r2 (see Eq. (32) of the one shell problem) while exploiting the relation
B
2F
1
+R
=

MR
M W (W R )
(71)
also derived from (13) and substituting into r 2 the value given by the equations of motion derived
previously we find that the third line simplifies with the fourth. Thus we are left with the relation
r 1 + N r (r1 ) N (r1 ) = 0
(72)
which is the correct equation of motion for the interior shell.

8. Exchange relations
In the present section we consider in the above developed formalism the intersection of two
shells during their motion. In the massless case we shall rederive the well-known relations of
Dray and t Hooft [11] and Redmount [12] between the masses characteristic of the three regions
before and after the collision. We shall denote by t0 the instant of collision, r0 the coordinate
of the collision. It is well known that some hypothesis has to be done on the dynamics of the
collision which should tell us which are the masses of the two shells after the collision. Once
these are given the problem is reduced to that of a two particle collision in special relativity.
The formulas which we develop below give the intermediate mass M0 after the collision. The
simplest assumption is that of the transparent crossing, i.e., after the collision shell 1 carries
along with unchanged mass m1 and the same for shell 2. A further simplification occurs in the
massless case, i.e., when the two massless shells go over to two massless shells. We develop
first the general formalism which applies also when we have a change in the masses during the
collision and then specialize to particular situations.
Just before the collision, i.e., at t = t0 assuming that r1 (t0 ) < r2 (t0 ) we have for
the momentum L
526
L (r1 0, t0 ) = L (r1 + 0, t0 ) + p 1 = L (r2 0, t0 ) + p 1

= L (r2 + 0, t0 ) + p 1 + p 2
(73)
and after the collision, i.e., at t = t0 + we have with r2 (t0 + ) < r1 (t0 + ),
L (r2 0, t0 + ) = L (r2 + 0, t0 + ) + p 2 = L (r1 0, t0 + ) + p 2
= L (r1 + 0, t0 + ) + p 1 + p 2
(74)
and at the collision r2 = r1 = r0 . The main point in treating the crossing is to realize that the
sum V1 + V2 has to be continuous in time as it represents the discontinuity of R at the time
of crossing. In fact just before the crossing we have for r < r0 , choosing for simplicity in (54)
g(x) = h(x),
R(r) = r +
(V1 + V2 )
g(r r0 )
r0
(75)
while immediately after we have

R(r) = r +
(V1 + V2 )
g(r r0 ).
r0
(76)
If V1 + V2 = V1 + V2 we would have a discontinuous evolution of the metric for r < r0 which is
incompatible with the equations of motion of the gravitational field. As M is unchanged during
the time evolution V1 + V2 = V1 + V2 implies that L (r1 0, t0 ) = L (r2 0, t0 + ) which

combined with the fact that L (r2 + 0, t0 ) = L (r1 + 0, t0 + ) = 2H r0 gives the further
relation p 1 + p 2 = p 1 + p 2 . Thus we have the same kinematics as in the two particle collision
in special relativity. In the case of conservation of the masses m1 and m2 we have the same
kinematics as that of an elastic collision in 1 + 1 dimensions. A discussion of special cases with
massive shells using a different formalism has been given in [20].
The relation between L (r1 0, t0 ) and L (r2 + 0, t0 )

2M
V1 + V2 2
1+
= 2H r0 + p 1 + p 2
1+
r0
(77)
r0
r0
gives the equation
m21 + m22 + 2(V1 V2 p 1 p 2 ) + 2(V1 + V2 )r0 + 2M r0

= 2H r0 + 2(p 1 + p 2 ) 2H r0 ,
(78)
where M0 is given by

m22
2H
H M0 = V2 +
p 2
.
2r0
r0
From the knowledge of p 1 one derives M0 from

m1 2
2H

H M0 = V1 +
p 1
.
2r0
r0
(79)
(80)
For transparent crossing, i.e., mj = mj and p j = p j , M0 is given by (80) with p 1 = p 1 . The
massless case is most easily treated. In this case in order for the shells to intersect they must
527
move in opposite directions, the exterior one with negative and the interior one with positive
velocity. The conservation of V1 + V2 and p 1 + p 2 in the massless case has two solutions which
are physically equivalent, i.e., p 1 = p 1 , p 2 = p 2 . The two shells intersect only if p 2 < 0 and
p 1 > 0 so that V2 = p 2 and V1 = p 1 . From Eq. (79) we have
p 2 =
H M0

1 + 2H
r
(81)
and from
p 1
p 1 =
= p 1
H M0
.
1 2H
r
(82)
Eq. (78) now becomes

H r0 + 2H r0 (p 1 + p 2 ) = r0 (p 1 p 2 ) + M r0 2p 1 p 2
(83)
and substituting here Eqs. (81), (82) we obtain

H r0 + M r0 2H M = M0 r0 2M0 M0 + M0 r0
(84)
which is the well-known Drayt Hooft and Redmount relation [11,12].

9. Integrability of the form pc1 d r1 + pc2 d r2 and the two shell tunneling probability
We are interested in computing the action for the two shell system, i.e., the integral
tf
ti
dt (pc1 r 1 + pc2 r c2 ) =
r1f
,r2f
(pc1 d r1 + pc2 d r2 )
(85)
r1i ,r2i
on the solution of the equations of motion. This is of interest in the computation of the semiclassical wave function and the tunneling probability in the two shell case. We shall prove explicitly
that the form pc1 d r1 + pc2 d r2 is closed, i.e., integrable. Such a result will be very useful in the
actual computation and in showing the independence of the results of the deformation g. In fact
we can reduce the computation to the integral on a simple path on which the two momenta pc1
and pc2 take a simpler form.
The integrability result is similar to a theorem of analytical mechanics [21,22] stating that for
a system with two degrees of freedom, in presence of a constant of motion the form p1 dq1 +
p2 dq2 where the pj are expressed in terms of the energy and of the constant of motion, is a
closed form. Here the setting is somewhat different as we are dealing with an effective action
with two degrees of freedom, arising from the action of a constrained system. We recall that for
r1 < r2 , M0 like H is a constant of motion in virtue of the equations of motion of the gravitational
field. The pcj are functions of H , M0 , r1 , r2 in addition to the fixed datum of the problem M.
The effective action takes the form

dt (pc1 r 1 + pc2 r 2 + pH H + pM0 M 0 ) + b.t.,
(86)
where b.t. is the boundary term (see Eq. (2)) which depends only on H , N r and not on rj . The
value of N r is supplied by the solution of the gravitational equations of motion. Thus varying
528
w.r.t. r1 we have
dpc1 pc1 pc2 pH pM0
+ r 2
+
+ r 1
H+
M0 .
0=
(87)
dt
r1
r1
r1
r1
We show in Appendix A that the constraints combined with the equations for the gravitational
field impose 0 = H = M 0 and thus we have
pc1 pc2
(88)
= 0.
r2
r1
The meaning of the procedure is that the consistency of the variational principle imposes Eq. (88).
On the other hand, Eq. (88) can be verified also from the explicit expression of pc1 and pc2 given
in Section 6. If the crossing of the shells occurs outside the region 2M, 2H we can choose the
path as to keep the two deformations non-overlapping in such a region; then pc1 takes the simple
form (24) with H substituted by M0 and pc2 again the form (24) with M substituted by M0 and
thus one obtains for the imaginary part of integral (85)
1

1
(2M0 )2 (2M)2 + (2H )2 (2M0 )2 = (2H )2 (2M)2 ,
(89)
2
2
i.e., the two shells are emitted independently. If the crossing occurs at r0 with 2M < r0 < 2H ,
with, e.g., r1 < r2 before the crossing, we choose the path r1 = r2 before the crossing and
r2 = r1 after the crossing. For clearness sake we examine at first the problem for the crossing
of a null shell with a massive shell, even when the mass of the massive shell changes. In order to
reduce the integration path to the described one, one must first, given the two initial points with
R(r1i ) < 2M and R(r2i ) < 2M, bring them together (r1 will denote the position of the massless
shell). For r1i < r2i this is done by integrating along the line r2 = const with r1 varying from r1i
to r2i . The contribution is
r
2i
r
2i
R (r1 + )D d r1 .
pc1 d r1 =
r1i
But
R (r1 )
(90)
r1i
is real and
D = R(r1 )

1 2M0

2M0
2M
R(r1 )

+ log
.
2M
R(r1 )
R(r1 )
1 R(
r )
(91)
For R(r1 ) < 2M < 2M0 Eq. (91) is real and thus the contribution (90) is real. If on the other
hand r2i < r1i we integrate pc1 from r1i to r2i + keeping r2 fixed. This time we recall that
0
+ p c1 ,
pc1 = pc1
where
0
pc1
(92)
is real because we are outside the interval (2M0 , 2H ), and
T

D(r2 )
p c1 = R (r2 + ) 1 D(r2 ) + R(r2 ) r2
(93)
r1
with T now given by log v1 . Again all the items in Eq. (93) are real. Thus we can start, e.g., with
r1i = r2i . The integration along the path r1 = r2 up to r0 is very simple because

1 + V2 r p2 2H
2M
2H
r

log
pc2 + pc1 = r
(94)
r
r
1 2M
r
529
whose complete discussion has already been done in Section 5. The gap is 2M, 2H . More
difficult is the analysis for r0 < r < 2H . Now we have

V2 p 2

2H
1
+
2M
2H
r
r

+ pc1
= r
pc2
(95)
log
r
r
2M
1
r
with
H M0
.
1 2H
r
p 1 =
(96)
Moreover
p 2
r2
AW (r2 + ) + R (r2

+ ) A2 (1
2M0 m22
) 2
r2
r
2
2M0
r2
(97)
where
A=
M0 M
m2
22
r2
2r2
and

W (r2 + ) =
(98)

p 1 2
2M0
1+
.
1+
r2
r2
(99)
p 1 diverges with positive residue at r1 = 2H and thus also W (r2 + ) and p 2 of Eq. (97) diverge
like p 1 . As a consequence the analytic continuation of V2 p 2 below r = 2H is negative which
makes the numerator of the argument of the logarithm in Eq. (95) negative. We show now that
such numerator stays negative all the way for r < 2H . The argument of the square root in (97)
never vanishes so that at r2 = 2M0 the numerator in (97) reduces to

A W (r2 + ) + R (r2 + ) .
(100)
We can now explicitly compute W (r2 + ) and R (r2 + ) at r2 = 2M0 . At such a point we have

(H M0 )/r2 1
H

R (r2 + ) = 1 +
(101)
=
1
2
M0
1 2H
r2
while
2H
1 H M0

+
=
W (r2 + ) =
2M0
r2 1 2H
r2

1
H
= 1
2
M0
H

H
1 M0 1

+
M0
2 1 H
M
0
(102)
which means that there is no pole in p 2 at r = 2M0 . Thus p 2 is regular below 2H and V2 p 2
cannot change sign.
530
In this way we proved that for the crossing of a null shell and a massive shell, into a null shell
and another massive shell even with a change of mass, the imaginary part of the integral (85) is
still given by Eq. (89).
The reasoning in the general case of both m1 and m2 different from zero is dealt with similarly.
It is sufficient to examine the case r1 < r2 the other case being now equivalent. We shall first
discuss the integration along the path r1 + = r2 up to r0 . For r1 + = r2 = r we have

V2 p 2 +V1 p 1

2H
1
+
2M
2H
r
r

log
pc2 + pc1 = r
(103)
r
r
2M
1
r
with

m22
2H
p 2
H M0 = V2 +
2r2
r2
(104)
which as before makes p 2 diverge at r2 = 2H . For p 1 we have

2

0 m1
A1 W (r1 + 0) + R (r1 + 0) A21 (1 2M
)
2
r1
r1
p 1
=
2M
0
r1
1 r
(105)
with
A1 =
m2
M0 M
12 .
r1
2r1
(106)
p 2 diverges with positive residue at r = 2H and p 1 also diverges like p 2 as

R (r1 + 0) = 1 +
V2
r1
(107)
and W (r1 + 0) as given by Eq. (11) also diverges like p 2 . Then for r just below 2H we have both
V2 p 2 < 0 and V1 p 1 < 0. As before the main point is in proving that p 1 is regular below
2H , i.e., does not diverge at r = 2M0 . To this end we must examine the numerator of Eq. (105)
at r = 2M0 . We have for r = 2M0

m2
H
1
2
W (2M0 + 0) =
1 +
(108)
>0
H
2
M0
8M02 ( M
+
1)
0
as H > M0 . With regard to R (r1 + 0) which is negative for r1 = 2H it cannot change sign
for 2M0 < r < 2H because at the point where R (r1 + 0) vanishes

2M0
W (r1 + 0) = R 2 (r1 + 0) 1 +
(109)
r1

would become imaginary while from W (r1 + 0) = 2H /r1 + p 2 /r1 we have that W is real. Then
at r = 2M0 the numerator in Eq. (105) vanishes and p 1 is regular all the way below 2H . Thus
we are in the same situation as in the previously discussed case. Below r = 2M the argument of
the logarithm
in Eq. (103) becomes positive again due to the change in sign of the denominator

1 2M/r . The integration for r > r0 is treated simply by exchanging m1 with m2 .
531
Finally we deal with the integral

r
1i +
pc2 d r2
(110)
r2i
which makes the two initial points coalesce. As in the previously discussed case of one massless
shell, all the point is in proving that D is real. In addition to the contribution of B which we
already proved to be real, we have now the contribution of L. We have already proven that at
r = r2

2M0

R (r) R (r)2 1 +
(111)
= R (r) W (r)
R(r)
equals
V2 p 2
1+
r2
r2

2H
r2
(112)
which is negative for r2 < 2H and thus in particular for r2 < 2M. In the interval r1 < r < r2 <
2M the term (111) cannot change sign because 1 + 2M0 /R is positive. Moreover we have
R (r1 ) W (r1 ) = R (r1 + ) W (r1 + ) +
V1
p 1
.
R(r1 ) R(r1 )
(113)
But we proved after Eq. (107) that below 2H , the analytic continuation of V1 p 1 is negative,
implying that (113) is negative, like R (r1 + ) W (r1 + ). The outcome is that the discontinuity
of L at r1 , being given by the logarithm of a positive number, is real. This concludes the proof in
the case of the emission of two massive shells.
10. Conclusions
The main issue of the present paper is the treatment of two intersecting shells of matter or
radiation in general relativity, in a formalism, apt to compute the tunneling amplitude for the
emission of two shells. In order to do so it is necessary to adopt a gauge which is more general
than the one used [6,9] in the treatment of a single shell. In the usual treatments of the tunneling
amplitude for a single shell, a limit gauge is adopted. Already at the level of a single shell we
show that the complete action contains a term in which the mass of the remnant black hole
plays a dynamical role. Such a term is unimportant if the variation of the action is taken with
respect to the total mass of the system, keeping the mass of the remnant as a datum of the
problem, but becomes essential if one varies the mass of the remnant keeping the total mass of
the system fixed as done, e.g., in [8]. The reduced canonical momentum even in the single shell
instance is gauge dependent but the tunneling probability turns out to be independent of such
a choice. All the treatment is performed in the general massive case, the massless one being a
special case. The tunneling results are independent of the mass of the shell. The adoption of
a general non-limit gauge allows the extension of the formalism to two or more shells. In this
instance both the intermediate mass and the total mass become dynamical variables in the sense
that the reduced action contains terms proportional to the time derivative of them. We show
how to derive the equations of motion of both the interior and exterior shells by varying the
532
reduced action and this is done both in the massless and in the more complicated massive cases.
With regard to the computation of the tunneling amplitude it is possible to prove an integrability
theorem which allows to deform the trajectory in coordinate space to a contour which drastically
simplifies the computation. Firstly one proves in such a way that the result is independent of the
deformation defining the gauge introduced in Section 2, and secondly one finds in the general
massive or massless case that the tunneling probability is given simply by the product of the
tunneling probabilities for the independent emission of the two shells. Such a circumstance is
interpreted [13] as the fact that in this model we have no information encoded in the radiation
emitted by the black hole.
Acknowledgements
One of us (P.M.) is grateful to M. Porrati and V.S. Rychkov for stimulating discussions.
Appendix A
For completeness we report here some formulas which are useful in the text. The constraints
are given by [3,6,9]
r ),
Hr = R R L L p(r

LL2
RR R
RR L L R L
Ht =
p 2 L2 + m2 (r r ).
+
+
+
L
2L
R
2
2R 2
L2
From the constraints it follows [3] that the quantity
(A.1)
2
L2
R R (R )2
+
2R
2
2L2
is constant in the regions of r where there are no sources as there
M=
R
L
Ht
Hr .
L
RL
The equations of motion for the gravitational field are [9]

LL R
L = N
+ NrL ,
2
R
R
N
L
+ N r R,
R =
R
2

L2
N
R
2p 2 (r r )
N RR
L =
+1+
+ N r L ,
2
2
2
L
R
L
3
2
2
2
L p L + m

LL2
L R
R
NR
R = N
+ N r R .
3
2
L
L
R
R
M =
(A.2)
(A.3)
(A.4)
(A.5)
(A.6)
(A.7)
(A.8)
The equations of motion for r and p follow from the above equations for R, L, R and L and
the constraints. In fact using the relation
dR(r )
r ) + R (r )r
r + ) + R (r + )r = R(
= R(
dt
(A.9)
533
we have
r R + R = 0.
(A.10)
Using now the equation of motion (A.6) and the constraints

R (r ) =
where V =
r =
V
;
R
p
L (r ) = ,
L
(A.11)
p 2 + m2 L2 , one obtains
N (r )p

N r (r ).
2
2
2
L(r ) p + m L (r )
(A.12)
Using the relation

dL(r )
r + ) = L (r )r + L(
r ) = L r + L
= L (r + )r + L(
dt
(A.13)
R
N R(r )
+ N r (r )R
L(r )
L(r )
(A.14)
and
r R = N (r )
derived from (A.8) and
dL
dL(r )
L L(r )
p =
dt
dt
one obtains
p =

N
N(r )p 2 L

p 2 + m2 L2 (r ) + p N r .
2
2
2
2
L (r ) p + m L (r ) L(r )
(A.15)
(A.16)
The occurrence of the average values L = [L (r + ) + L (r )]/2, etc., in the previous
equation, was pointed out and discussed at the level of the variational principle in [9]. In the
massless case m = 0 however, using again relations (A.9), (A.13), (A.14) one proves that the
r.h.s. of Eq. (A.16) has no discontinuity, i.e., there is no need to take the average value in the
r.h.s. of Eq. (A.16). In fact let us consider the discontinuity of the zero mass version of the r.h.s.
of Eq. (A.16)

N L N
r
+N
p
(A.17)
L
L2
being the sign of p,
i.e.,

N L N
r
+
N
.
p
L
L2
(A.18)
From Eq. (A.5) and the equation of motion (A.12) we have

N
NL
N
L = R 2 L LN r
L
R
R
and from Eq. (A.8)
(A.19)
N R = RN + N R .
(A.20)
534
Substituting (A.19) and (A.20) into Eq. (A.18) and taking into account that L = p/L
and
R = p/R
we have that expression (A.18) vanishes. This fact was discussed at the level of
the variational principle in [10].
From the equations of motion (A.5)(A.8) it follows that in the region where there are no
sources
dM
R
L
R
(A.21)
= N 3 Hr N r Ht N r
Hr ,
dt
L
RL
L
i.e., combining with (A.4), we have that in the region free of sources M is constant both in r and
in t .
Appendix B
Equation of motion for r .
(1) Outer gauge.
In this case

pc = R(L B) = r L(r )
2H
+
r

2M
2M
+ log 1
.
r
r
(B.1)
Using

2H 1
p
p
p
= 1+
H
V
r
2H r
and
L
1
(r ) =
=
R
W (r )
(B.2)
2H
p
+
r
r
1
we have

r
r p
pc
=
+
H
2H
2H V
p
V

2H
r
(B.3)

2H
r
1
(B.4)
from which

r = N(rm ) p 2H .
V
r
(B.5)
(2) Inner gauge.

This time we have

2M
m2
p
H M =V
2r
r
and

pci
= r L(r + )
2H
+
r
(B.6)

2M
2H
log 1
.
r
r
(B.7)
Using

p
2M 1
p
p
= 1
M
V
r
2M r
and
1
L
(r + ) =
=

R
W (r + )
we have
pci
=
M
2M p
r
r
(B.8)
1

r
2M 1
2M 1
p p
p
,
1
=
2M
V V
r
V
r
i.e.,
r =
2M
r
535

N (r0 ).
(B.9)
(B.10)
(B.11)
Appendix C
In this appendix we derive the equations of motion for the exterior and interior shells from the
reduced action (62) for masses m1 and m2 different from zero.
First we consider the variation H = 0 and M0 = 0. Under such a variation the coefficient
of the r 1 term is given, using
dT
R
= (R 1)
H
dH
R
dT
= (R r1 )
;
H
dH
dT
multiplied by
by (W (r1 + )W (r1 ))1 dH

v1
v1
R W (r1 )(R 1) + W (r1 + ) R 1 + (R r1 )
+ (R 1)
R
R

v1 v1
(R r1 ) W (r1 )R + W (r1 + ) R +
R
R +
,
R
R
(C.1)
(C.2)
where R, R , R stay for R(r1 ), R (r1 + ), R (r1 + ). From

W (r1 ) = W (r1 + ) +
p 1
R(r1 )
(C.3)
we find
W (r1 + )
v1

R (r1 + )
p 1
R(r1 )
(C.4)
which substituted into Eq. (C.2) makes it vanish. This is an expected result as the motion of the
exterior shell must not depend on the dynamics which develops at smaller radiuses, but only on
the two masses H and M0 .
With regard to the terms proportional to r 2 in addition to
0
p
r 2 c2
H
(C.5)
536
we have the term given by

1
dT
(C.6)
W (r1 + )W (r1 ) dH
multiplied by

T
W (r1 )(R 1)
(R 1) + (R r1 )
r2

v1
v1
(R
1)
(R r1 ) +
+ W (r1 + ) R 1 +
R
R

dT
dT
(R r1 ) W (r1 )
(R 1) R + W (r1 + )
(R 1) R
d r2
d r2

v1
dT
v1
dT

+
(R r1 ) +
+
(R
1)
,
(C.7)
R + 1 +
R
R
d r2
R
d r2
where again R, R , R stay for R(r1 ), R (r1 + ), R (r1 + ). As a consequence of Eq. (C.4) the
above expression vanishes. Adding the contribution of the boundary term we have
0
p
r 2 c2 N(rm ) = 0
(C.8)
H
which is the single shell equation of motion (26).
The equation of motion of the interior shell is obtained from the variation H = 0, M0 = 0.
In this case we have no contribution form the boundary terms. For the coefficient of r 1 we obtain
1
dT
1
(R 1)
+ R R
W (r1 + )
W (r1 + ) dM0

v1
v1
v1
dT
1
(R
1)
+
(R r1 ) +
R 1 +
+
W (r1 ) dM0
R
R
M0

dT
R
v
v
1
1
1
(R r1 )
(C.9)
R
R
+
R +
R +
dM0
W (r1 + ) W (r1 )
R
R
having used the relation (see Eq. (13))
2F
F
1
= .

MR
M
W
Substituting in Eq. (C.9) the relation
R
p 1
)
p 1 (1 + RW (
v1
r1 +)
=

M0 R(R (r1 + )p 1 V1 W (r1 + ))
such a coefficient of r 1 becomes simply
1
p 1
r1
V1 R (
+ ) W (r1 + )
(C.10)
(C.11)
(C.12)
which is the non-zero mass generalization of the coefficient of r 1 appearing in Eq. (70). Then
using Eqs. (15), (29) for N (r1 ) and N r (r1 ) we have
1

p 1
p 1
r
R (r + ) W (r + )
0=
r 1 N (r1 ) + N (r1 )
V1
V1
537

0
pc2
2 F V2
F
+
+
M0 M0 R r2
M0
1
r1 + )
p 1 R(
p 1
R (r + ) W (r + )
V1
V1 W (r + )
+ r 2
2
dT

p c2
r1 + ) F (r1 + ) r 2 d
R(
)
D
,
R(
r
(C.13)
1
1
M0
M0 R
d r2
dM0
where = 2H r /( 2H r + p).
The first line is simply the equation of motion for r1 , the second
line vanishes being simply the equation of motion for r2 due to the variation of M0 (see Eq. (32))
while the sum of the third and fourth lines vanishes in virtue of relation (C.4).
+ r 2
References
[1] W. Israel, Nuovo Cimento B 44 (1966) 1;
W. Israel, Nuovo Cimento B 48 (1967) 463, Erratum.
[2] P. Hajicek, Phys. Rev. D 56 (1997) 4706, arXiv: gr-qc/9706022;
P. Hajicek, Phys. Rev. D 58 (1998) 084005, arXiv: gr-qc/9804010;
P. Hajicek, I. Kouletsis, Class. Quantum Grav. 19 (2002) 2529, arXiv: gr-qc/0112060;
P. Hajicek, I. Kouletsis, Class. Quantum Grav. 19 (2002) 2551, arXiv: gr-qc/0112061;
J. Bicak, P. Hajicek, Phys. Rev. D 68 (2003) 104016, arXiv: gr-qc/0308013.
[3] W. Fischler, D. Morgan, J. Polchinski, Phys. Rev. D 41 (1990) 2638;
W. Fischler, D. Morgan, J. Polchinski, Phys. Rev. D 42 (1990) 4042.
[4] E. Farhi, A. Guth, J. Guven, Nucl. Phys. B 339 (1990) 417.
[5] S. Ansoldi, A. Aurilia, R. Balbinot, E. Spallucci, Class. Quantum Grav. 14 (1997) 2727, arXiv: gr-qc/9706081.
[6] P. Kraus, F. Wilczek, Nucl. Phys. B 433 (1995) 403, arXiv: gr-qc/9408003.
[7] P. Kraus, F. Wilczek, Nucl. Phys. B 437 (1995) 231, arXiv: hep-th/9411219.
[8] M. Parikh, F. Wilczek, Phys. Rev. Lett. 85 (2000) 5042, arXiv: hep-th/9907001.
[9] J.L. Friedman, J. Louko, S. Winters-Hilt, Phys. Rev. D 56 (1997) 7674, arXiv: gr-qc/9706051.
[10] J. Louko, B.F. Whiting, J.L. Friedman, Phys. Rev. D 57 (1998) 2279, arXiv: gr-qc/9708012.
[11] T. Dray, G. t Hooft, Commun. Math. Phys. 99 (1985) 613.
[12] I.H. Redmount, Prog. Theor. Phys. 73 (1985) 1401.
[13] M. Parikh, arXiv: hep-th/0402166.
[14] V.A. Berezin, A. Boyarsky, A.Y. Neronov, Grav. Cosmol. 5 (1999) 16, arXiv: gr-qc/0605099;
M. Angheben, M. Nadalini, L. Vanzo, S. Zerbini, JHEP 0505 (2005) 014, arXiv: hep-th/0503081;
A.J.M. Medved, E.C. Vagenas, Mod. Phys. Lett. A 20 (2005) 2449, arXiv: gr-qc/0504113.
[15] B. Chowdhury, arXiv: hep-th/0605197;
P. Mitra, arXiv: hep-th/061265.
[16] E. Keski-Vakkuri, P. Kraus, Nucl. Phys. B 491 (1997) 249, arXiv: hep-th/9610045.
[17] S.W. Hawking, C.J. Hunter, Class. Quantum Grav. 13 (1996) 2735, arXiv: gr-qc/9603050.
[18] S. Goncalves, Phys. Rev. D 66 (2002) 084021, arXiv: gr-qc/0212124.
[19] A.J. Hamilton, D. Kabat, M.K. Parikh, JHEP 0407 (2004) 024, arXiv: hep-th/0311180.
[20] D. Nunez, H.P. Oliveira, J. Salim, Class. Quantum Grav. 10 (1993) 1117, arXiv: gr-qc/9302003.
[21] E.T. Whittaker, A treatise on the Analytical Mechanics of Particles and Rigid Bodies, Cambridge Univ. Press,
Cambridge, UK, 1947 (Chapter X).
[22] V.I. Arnold, Mathematical Methods of Classical Mechanics, Springer, Berlin, 1989 (Chapter IX).

The discrete spectrum of the D = 11 bosonic M5-brane

I. Martin a, , L. Navarro b , A.J. Prez b , A. Restuccia a
a Departamento de Fsica, Universidad Simn Bolvar, Apartado 89000, Caracas 1080-A, Venezuela
b Departamento de Matemticas, Universidad Simn Bolvar, Apartado 89000, Caracas 1080-A, Venezuela
Received 25 May 2007; received in revised form 13 October 2007; accepted 14 November 2007
Abstract
We prove that the spectrum of a regularized M5-brane in D = 11 target space is discrete with eigenvalues
extending to . The proof includes the same result for the spectra of regularized bosonic p-branes in
general.
PACS: 11.25.-W; 11.10.Ki; 11.10.Ef
Keywords: p-branes; Super membrane; M5-brane spectrum
1. Introduction
The understanding of the spectral properties of the supermembrane and super M5-brane in 11
dimensions are important steps towards the non-perturbative analysis of M-theory. The SU(N )
regularized Hamiltonian of the supermembrane on a D = 11 Minkowski target space has a
continuous spectrum [1], see also [25]. The supermembrane on a D = 11 target space with
a compact sector is expected to have also a continuous spectrum [6]. But, the D = 11 supermembrane wrapped in an irreducible way on a compact sector of the target space, i.e., with a
topological condition on configuration space yielding a nontrivial central charge has a discrete
spectrum and its ground state has a strictly positive energy [712].
In all cases the bosonic Hamiltonian has a discrete spectrum. However, they are qualitatively
different in a crucial way. In the latter case, the central charge generate mass terms implying that
E-mail addresses: isbeliam@usb.ve (I. Martin), lnavarro@ma.usb.ve (L. Navarro), ajperez@ma.usb.ve (A.J. Prez),
arestu@usb.ve (A. Restuccia).
doi:10.1016/j.nuclphysb.2007.11.012
I. Martin et al. / Nuclear Physics B 794 [PM] (2008) 538551
539
the potential on configuration space tends to infinity when it approaches infinity in this space.
Moreover, this qualitative property of the spectrum remains unchanged in the supersymmetric
theory with nontrivial central charges. In the former case [1], the potential presents zero point
valleys extending to infinity on configuration space. In this case the membrane admits as physical
configurations stringy spikes that make the supermembrane spectrum continuous, despite the
fact that its bosonic part on its own would not produce a continuous spectrum. Although the
bosonic potential is zero on the minima of the valleys, the walls of the valleys get closer as they
approach infinity in a way that the quantum mechanical wave function cannot escape to infinity.
The precise mathematical meaning of this property was explained in [13]. It is formulated in
terms of the integral of the potential on a fixed sized cell, defined in the sense of Molchanov,
Mazya and Shubin [14,15], when the center of the cell approaches infinity on configuration
space. The potential integral of the cell in the directions of zero potential is bounded below by
the potential of an harmonic oscillator ensuring that the integral goes to infinity when the cell is
moved to infinity in those directions. This bound from below is lost once the fermionic part of
the potential is added, as a consequence the supermembrane spectrum becomes continuous. All
the above results refer to regularized Hamiltonians. More recently, it has been shown [10,16] that
the exact bosonic Hamiltonian for the case of the supermembrane with nontrivial charges, has a
discrete spectrum. This was achieved with a precise definition of the configuration space in terms
of Sobolev spaces. It was also proven that the spectrum of the SU(N ) regularized model of the
semiclassical Hamiltonian converges to the spectrum of the exact Hamiltonian when N tends to
infinity.
In the case of the M5-brane no results have been reported concerning its spectrum. In [17] a
semiclassical analysis of the spectrum of the M5-brane was performed. The M5-brane covariant
action was first obtained in [18,19] and a gauge fixed action version was obtained independently
in [20]. In [21] a formulation was obtained for its Hamiltonian in terms of first class constraints
only. Also, its BRST structure and the existence of string and membrane spikes were shown.
Later, the NambuPoisson structure of the M5-brane was introduced in [22]. Also, a general
analysis of such structure for p-branes was presented in [23]. In this paper, following [13,22]
we prove that the spectrum of the regularized M5-brane and that one of any p-brane is discrete.
The proof in general applies to many matrix models associated to such theories. In Section 2,
the algebraic structure of the M5-brane Hamiltonian is presented for completeness. In Section 3,
we present the SU(N ) regularized version of the M5-brane Hamiltonian exploiting the Nambu
Poisson structure underlying it. In Section 4, using appropriate theorems of spectral analysis, we
show that generally p-branes matrix models present discrete spectra. With this result at hand, the
discreteness of the spectrum of the regularized bosonic M5-brane Hamiltonian follows easily. In
Section 5, we present conclusions.
2. The algebraic structure of M5-brane Hamiltonian
We start recalling the bosonic M5-brane Hamiltonian on a D = 11 Minkowski target space in
the light cone gauge obtained in [21],
1
H = M M + 2g + l l + 5i 5i + j j + ,
2
where

1
l = P + H
2
(1)
(2)
540
and
1
H =
H ,
6
H = B + B + B .
5i , j ,
(3)
(4)
are the Lagrange multipliers associated to remaining constraints
5i = P 5i H 5i = 0,
(5)
= P
(6)
[] = [
= 0, i = 1, 2, 3, 4,

1
1
M
M ] X + V] = 0,
4
W
(7)
where
V =
l l .
M
(8)
XM
and
are the canonical conjugate momenta to
and B respectively, g is the determinant of the induced metric. Eqs. (5) and (6) are the first class constraints generating the local
gauge symmetry associated to the antisymmetric field while (7) is the first class constraint generating volume preserving diffeomorphisms. X M are the light cone gauge transverse coordinates
on the target space.
W is a scalar density introduced in the gauge fixing procedure. It represents the determinant
of an intrinsic metric over the spatial world volume of the brane. In our notation caps Latin letters
are transverse light cone gauge indices M, N = 1, . . . , 9, Greek ones are spatial world volume
indices, and small Latin letters denote spatial world volume indices on a 4-dimensional spatial
submanifold.
The elimination of second class constraints from the formulation in [18,19] and [20] to produce a canonical Hamiltonian with only first class constraints, was achieved at the price of
loosing the manifest 5-dimensional spatial covariance. In this way, the spatial world volume splits
into M5 = M4 M1 . We will exploit that decomposition in our analysis of the Hamiltonian. We
will assume M4 has a symplectic structure with 0 being its associated nondegenerate closed
2-form. It is assumed that M4 and M1 are compact manifolds. Integrals over M5 = M4 M1 are

W d 1 d 2 d 3 d 4 d 5 ,
M5
M5
with
0 0 =
W d 1 d 2 d 3 d 4
Let us analyze the Hamiltonian density term by term. We first notice that g, the determinant
of the induced metric, may be re-expressed in a straightforward manner as a squared five entries
bracket, a NambuPoisson bracket,
1
1 M N P Q R
2
g =
1 ,...,5
1 ,...,5 g1 1 . . . g5 5 =
(9)
.
X ,X ,X ,X ,X
5!
5!
Let us consider now the third term dependent on the antisymmetric field B . It is invariant
under the action of the first class constraints (5) and (6). To eliminate part of these constraints,
we proceed to make a partial gauge fixing on B , following [21] we take
B5i = 0,
(10)
541
which, together with the constraint (6) allow us a canonical reduction of the Hamiltonian (1).
Notice that the contribution of this partial gauge fixing to the functional measure is 1. We are
then left with the constraint
j P ij + 5 H 5i = 0,
i, j = 1, 2, 3, 4,
(11)
which generates the gauge symmetry on the 2-form B

Bij = i j j i .
(12)
B as a 2-form over M4 may be decomposed using the Hodge decomposition theorem into an
exact form plus a co-exact form plus an harmonic form. Its exact part is canonically conjugate to
the co-exact part of P ij , that is, calling i bj the exact part of Bij we have
ij

P i bj = i P ij bj ,
(13)
then an admissible gauge fixing is to set bj = 0 to eliminate the exact form and the i P ij from
the constraints.
We are then left with the co-exact part of B. It is directly related to l 5i
l 5i =
ij kl j Bkl .
Noticing that
l 5i
(14)
is divergenceless, it may always be rewritten without loosing generality as
l 5i =
5ij kl j [a k b l c] ,
a, b, c = 1, 2, 3.
(15)
This decomposition in terms of scalars is always valid locally for any four-dimensional divergenceless smooth vectorial density. a , a = 1, 2, 3 represent the three degrees of freedom of the
co-exact part of Bij .
Now we decompose the tensor density l ij into
l ij =
j i [a b c] +
j ikl kl ,
(16)
where is a closed 2-form.

It is now possible, following the Darbouxs theorem, to express kl in terms of the canonical
2-form 0 over M4 . In fact the area preserving diffeomorphisms homotopic to the identity are
generated by with infinitesimal parameter or equivalently generated by

1
M
= M X + V ,
(17)
4
with infinitesimal parameter given by

1
= W
W
satisfying identically

W = 0.
(18)
(19)
This volume preserving restriction leaves the four spatial parameters associated to M4
unconstrained. So, we are allowed to use the Darboux procedure to fix to 0 . We should be
left still with one free parameter since the local degrees of freedom of are only three. Indeed,
that is the case since the corresponding gauge fixing procedure allows to eliminate l 5i from the
constraints in the following way,
542
From (7) we have

1
1
(20)
M X M + V = U,
4
W
where U is an auxiliary field. We notice in V , the product of the dependent terms is zero. In
particular,
0
,
Vi = 4l 5j ij
(21)
l 5j
it allows to eliminate
the equation for = 5:
from the equation for = i in terms of U , which is determined from
0 XM
j ikl kl
M i
,
W
we are then left with the remaining constraint

M i X M
5j
j ikl 0
j l =
kl j
= 0.
W
0
l 5j =
j ikl kl
i U +
(22)
(23)
The kinetic term associated to this gauge fixing is a total time derivative and, since 0 is time
independent, it can be eliminated from the action. Finally, we are left with a complicated l l
term but to prove the discreteness of the spectrum it will become irrelevant as we will see in the
following sections. The final gauge fixing corresponding to the symplectomorphisms preserving
0 is performed by taking the Lagrange multiplier of the associated first class constraint to be
zero, the ghosts fields decouple from the action.
3. Regularization of the M5-brane
After fixing to 0 we may resolve the volume-preserving constraint for a , a = 1, 2, 3. We
are then left still with one constraint,

M j X M
0
ij kl kl
(24)
i
= 0.
W
The left hand member generates the symplectomorphisms preserving 0 . The full fivedimensional diffeomorphisms have been reduced to only that generator. We are then left with
a formulation in terms of X M and its conjugate momenta M , invariant under symplectomorphisms. The antisymmetric field B and its conjugate momenta P have been reduced to
0 , there is no local dynamics related to them. All the dynamics may be expressed in terms of
(X M , M ).
In order to obtain a regularization of the Hamiltonian, we express X M (, ) and M in terms
of a complete orthonormal basis over M4 M1 , {Ya ( 1 , 2 , 3 , 4 , 5 )}, in the Hilbert space of
L2 functions for M4 and a Fourier basis for the M1 manifold.
X M (, ) = X a M ( )Ya ( ), a = 1, 2, . . . , ,
a
( )Ya ( ).
M (, ) = W M
(25)
(26)
Since for every pair a, b

0
ij kl kl
i Ya j Yb ,
(27)
543
is a scalar function over M4 , we may reexpress it in terms of the basis, and obtain the symplectic
bracket
ij kl 0
{Ya , Yb } = kl i Ya j Yb = fabc Yc ,
W
where fabc is given by
ij kl 0
kl
i Ya j Yb Yc = fabc
(28)
(29)
M4
and satisfy the Jacobi identity. These are the structure constants of the symplectomorphisms
preserving 0 . Furthermore, we may introduce
Ya Yb = Cabd Yd ,
(30)
it is again valid since Ya Yb is also a scalar function over M4 . We get

Ya Yb Yd = Cabd ,
(31)
M4
which becomes totally symmetric in a, b, d. The other natural bracket in the formalism [22] is
the Nambu one,
1
{A, B, C, D, E} =
A B C D E
W
(32)
in particular the scalar

g
{Ya , Yb , Yc , Yd , Ye } = fabcde Yg ,
(33)
where

{Ya , Yb , Yc , Yd , Ye }Yg = fabcdeg ,
(34)
M5
is a totally antisymmetric tensor satisfying the generalized Jacobi identity [2427]. By construction we have the following relation for the compact base manifold we are considering

g
c e h g
ina fbc
fde Cce Cha ,
fabcde =
(35)
antisymm(a,b,c,d,e)
where n2a is an eigenvalue of the Laplacian on M1 . The right-hand side satisfies the generalized
Jacobi identity by construction.
We now consider a regularization of the M5-brane Hamiltonian by truncation of the infinite
dimensional basis, that is, a = 1, 2, . . . , N . We require in addition that in the remaining configc and C
uration space there exists brackets to have an intrinsic definition of the parameters fab
abc
entering in the theory. In the large N limit the corresponding structure constants should be the
area preserving ones. If so, we will have in the large N limit the generalized Jacobi identity for
g
fabcde . Meanwhile, it is not necessary to require a generalized Jacobi identity for the truncated
theory. In the discreteness proof presented here we do not use any algebraic properties of the
g
brackets. It is valid for any truncation in terms of some constants fabcde . However, if we require
544
an intrinsic meaning for the truncated theory, the algebraic structure should be present. To be
precise:
Let M4 be in particular the product of two Riemann surfaces M4 = . We denote the
local coordinates on and by 1 , 2 and 1 , 2 respectively. We take the symplectic nondegenerate 2-form as

1 , 2 , 1 , 2 = 1 1 , 2 d 1 d 2 + 2 1 , 2 d 1 d 2 ,
where 1 and 2 are the symplectic 2-forms on and respectively. Now since M5 = M4
M1 , an orthonormal basis on M5 may be expressed as

Ya 1 , 2 Yb 1 , 2 hn 5
where Ya and Yb are orthonormal basis on and respectively. hn is a Fourier basis on M1 .
The brackets

e g
e g
{Ya Yb hm , Yc Yd hn } = fac
(36)
C + Cac
f hm+n Ye Yg , with
bd
bd
e g
(Ya Yb hm )(Yc Yd hn ) = Cac
Cbd hm+n Ye Yg .
(37)
We have denoted by f and C the structure constants defined on each Riemann surface in the
usual way [5], i.e.,
1
e
Ye ,
1 (1 Ya 2 Yc 2 Ya 1 Yc ) = fac
2
e
Ya Yc = Cac
Ye .
It is known [5] that the structure constants of SU(N ), i.e., f N , C N satisfy

lim f N = f,
N
lim C N = C.
N
It then follows that the structure constants of SU(N ) SU(N ), i.e., f N C N + C N fN and
in the large N limit converge to the structure constants in (36) and (37).
We have then a truncated model with an intrinsic algebraic structure which in the large N
limit converges to the symplectic algebraic structure on M4
g
We notice that the Nambu structure constants fabcde of the symplectomorphisms satisfy the
following properties

{Ya , Yb , Yc , Yd , Ye }2 = fabcdeg f abcdeg > 0
C N C N
M5
and more generally

g
M gh = fabcde f abcdeh ,
(38)
is a strictly positive matrix. This is the only assumption we will require on the truncated theory
g
for the associated fabcde .
All the interacting terms in the Hamiltonian density may now be rewritten using X a M ( ),
their conjugate momenta and the structure constants fabc and Cabd . Integrating on the spatial
coordinates we arrive to a quantum mechanical Hamiltonian.
For the purpose of analyzing the spectrum of this regularized Hamiltonian, we will concentrate
on the first interacting terms that do not include l , this procedure is perfectly justified since
1
1
H = M M + 2g + l l M M + 2g = H0
2
2
(39)
545
so, in what follows the spectrum of H0 is studied. We will prove that H0 has a discrete spectrum
n , with n when n . The min max theorem assures the same qualitative spectrum
for H as it will be shown in the next section. The regularized Hamiltonian density H0 will
correspond to a p = 5-brane.
4. The spectrum of the quantum mechanical Hamiltonian for regularized p-brane
potentials
Let us consider the Schrdinger operator
2
a1
aL b
HL = + VL (X) = + XM
XM
f
,
L a1 ...aL
1
(40)
where L is the degree of the brane considered, Mi = 1, . . . , K, ai = 1, . . . , N , K L, N L,

a RKN and f b
X = XM
a1 ...aL is a constant tensor totally antisymmetric in a1 , . . . , aL and it is not
singular, i.e.,
if
a1 b
a1
XM
fa1 ,a2 ...aL = 0 then XM
= 0.
(41)
We introduce
Hl = + Vl (X) 1 l L,
where
Vl =

M1
=M2
=...
=Ml
1Mi K
(42)
a 1

al b
a l b
a1
XM
XM 1 . . . X M
.
. . . XM
f
f
l a1 ...al bl+1 ...bL
l a 1 ...a l bl+1 ...bL
1
In this section we prove that Hl , 1 l L has a discrete spectrum, but first let us recall some
mathematical definitions and theorems needed.
The definition of capacity of a compact set in Rn is an important ingredient in the Mazya and
Shubin generalization of Molchanovs theorem [14] on the necessary and sufficient conditions
for the discreteness of the spectrum of the Schrdinger operator. For details and the full-general
version of the Mazya and Shubin theorem see [15].
Definition 1. Let n 3, F Rn be a compact, and Lipc (Rn ) the set of all real-valued functions
with compact support satisfying a uniform Lipschitz condition in Rn . The Wieners capacity of
F is defined by

2
n
u(x) dx u Lip R , u|F = 1 .
cap(F ) = capRn (F ) = inf
c

Rn
In physical terms the capacity of the set F Rn is defined as the electrostatic energy over Rn
when the electrostatic potential is set to 1 on F .
Definition 2. Let Gd Rn be an open and bounded star-shaped set of diameter d, let (0, 1).
The negligibility class N (Gd ; Rn ) consists of all compact sets F Gd satisfying cap(F )
cap(Gd ).
For example we can take Gd to be a n-cube or a ball in Rn . In what follow we refer to Gd \ F
as a cell.
546
Theorem 1 (Mazya and Shubin). Let V L1loc (Rn ), V 0.

Necessity: If the spectrum of + V in L2 (Rn ) is discrete then for every function :
(0, +) (0, 1) and every d > 0

V (x) dx + as Gd .
inf
(43)
F N (Gd ;Rn )
Gd \F
Sufficiency: Let a function d (d) (0, 1) be defined for d > 0 in a neighborhood of 0 and
satisfying
lim sup d 2 (d) = +.
d0
Assume that there exists d0 > 0 such that (43) holds for every d (0, d0 ). Then the spectrum
of + V in L2 (Rn ) is discrete.
Remark 1. It follows from the previous theorem that a necessary condition for the discreteness
of spectrum of + V is

(44)
V (x) dx as Gd .
Gd
Let us recall that K. Friedrichs (see [15] for further references) proved that the spectrum of
the Schrdinger operator + V in L2 (Rn ) with a locally integrable potential V is discrete
provided V (x) as |x| .
In what follow we will denote the n-dimensional Lebesgue measure by Vol().
Lemma 1. For each given ball Gd = Gd (x0 ) centered at x0 and radius d > 0.
inf
F N (Gd ;Rn )
Vol(Gd \ F ) > 0.
Proof. Let V (x) = |x|. Then by Friedrichs theorem the spectrum of + V is discrete, so by
Theorem 1 we have

inf
V (x) dx as |x0 | .
F N (Gd ;Rn )
Gd \F
Now Gd \F V (x) dx (|x0 | + d) Vol(Gd \ F ) implies that

inf
V (x) dx |x0 | + d
inf
Vol(Gd \ F )
F N (Gd ;Rn )
Gd \F
F N (Gd ;Rn )
from which follows that infF N (Gd ;Rn ) Vol(Gd \ F ) > 0, as we claimed.
The minmax principle is useful in the proof of the next proposition, so we state it for completeness.
547
Theorem 2 (Minmax principle). Let H be a self-adjoint operator that is bounded from below,
i.e., H cI for some c. Define
n (H ) =
sup
UH (1 , 2 . . . n1 ),
1 ,2 ...n1
where
UH (1 , 2 . . . m ) = inf(, H ) when = 1
and [1 , 2 . . . m ] .
[1 , 2 . . . m ] is shorthand for {|(, i ) = 0, i = 1, 2, . . . , m}. The i are not necessarily

independent.
Then, for each fixed n, either:
(a) there are n eigenvalues (counting degenerate eigenvalues a number of times equal to there
multiplicity) below the bottom of the essential spectrum, and n (H ) is the nth eigenvalue
counting multiplicity;
or
(b) n is the bottom of the essential spectrum, i.e., n = inf{ | ess (H )}, and in that case
n = n+1 = n+2 = and there are at most n 1 eigenvalues (counting multiplicity)
below n .
For proof of the min-max theorem and further reading on the subject see [28].
Remark 2. As a consequence of this theorem if A and B are self-adjoint operators bounded from
below and if A B, then n (A) n (B). In this case if n (A) when n then n (B)
will also tend to infinity. This, in turn, means that if A has a discrete spectrum with a compact
resolvent then B will also have a discrete spectrum with a compact resolvent.
Now, we prove a theorem concerning the above defined operator Hl . In what follow we use
DS for discrete spectrum.
Proposition 1. Let
a
2
al b
Vl =
XM11 XM
f
,
l a1 ...al bl+1 ...bL
1Mi K
with M1
= M2
=
= Ml .
Then the following sequence holds:

Hl has DS + Vl has DS Hl+1
has
DS.
Proof. A. Hl has DS + Vl has DS.

Let Gd = Gd (X0 ) RKN be a ball centered at X0 and radius d > 0, let F N (Gd ; Rn ).
Then X = X0 + for all X in the cell Gd \ F . Let F be the set of all such . Then the necessary
condition of Theorem 1 implies that

inf Vl (X0 , ) d as |X0 | ,
(45)
F
548
We can rewrite the potential as

Vl =
Nl

Pj2 (X0 , ),
(46)
j =1
where Pj , j = 1, . . . , Nl , are polynomials in with coefficients depending on X0 . Using the

GramSchmidt process it is possible to rewrite
Pj (X0 , ) =
nj

aj k (X0 )j k ( ),
k=1
where {j k ( )} is a finite system of orthonormal polynomials depending on F , i.e.,

j k ( )im ( ) d = (j,k)(i,m) (the Kroneker delta)
F
and aj k (X0 ) are its corresponding coefficients.

Note that for any system {j k } there exists MF > 0 such that |j k ( )| MF for all (j, k) and
all F .
Hence
nj

2
2
2
Pj (X0 , ) =
aj k (X0 )F =: Pj F and
Vl d =
Pj 2F .
(47)
k=1
It is possible to choose N0 max{Nl , nj : j = 1, . . . , Nl } independent of F , then using

nj 4 4
( nk=1 ak )2 n nk=1 ak2 twice, we have Pj4 N03 k=1
aj k j k ( ), therefore

Pj4 d N03
4
F Pj d
j4k ( ) d N03 MF2
aj4k
k=1
i.e.,
nj

N04 MF2
Because of
L2 (

F)
3/2
Vl d
aj2k
N04 MF2
2
Vl (X0 , ) d
(48)
for all 0, using Schwarz inequality twice we obtain:

1/2
Vl d

1/2
Vl2 d
Now using (48) and (49) we have:

1/2

F
2
k=1

Pj 4F
Vl d
nj

Then from this and (46) and (47)
Vl
aj4k N03 MF2
k=1
N03 MF2 Pj 4F .
Vl2 (X0 , ) d
nj

N02 MF
1/2
Vl
F
d.
(49)
549
And from (45) we concluded that

inf
F N (Gd ;Rn )
MF > 0 and
+
Vl
has DS.
B. + Vl has DS Hl+1 has DS.

Now we show that + Vl+1 also has a
DS using the minmax principle. We start with
+ Vl+1 and get a bound in terms of + Vl . We rewrite Vl+1 as

a1
a 1

a
a
XM1 . . . XMl+1
XM1 . . . XMl+1
fb
fb
Vl+1 =
l+1 a1 ...al+1 bl+2 ...bL
l+1 a 1 ...a l+1 bl+2 ...bL
M1
=M2
=...
=Ml+1

M
M1
=M2
=...
=M
a 1
a
. . . XMl l fab1 ...al cbl+2 ...bL XM
1
a1
XM
1

a
. . . XMl l fab1 ...a l cb
l+2 ...bL
c
c
XM
XM
and notice that it is a sum of harmonic oscillator potentials with a matrix coefficient not involving

2
XM . Let M = N
c=0 (X c )2 , then for any k, 0 < k < 1, we have
M
+ Vl+1 = (1 k)() +
+

M1
=...
=Ml
=M
kM
a 1
c c
al b
a l b
a1
. . . XM
f
f
XM
XM1 . . . X M
L XM XM
l a1 ...al c...bL
l a 1 ...a l c...b
1

1 (M) c c
M + Acc XM XM
= (1 k)() + k
k
M

(1 k)() +
k tr A(M) (1 k) +
M

k
Vl .
(1 k)
(50)
k
Vl ,
By hypothesis Vl satisfies the sufficient condition of Theorem 1, then so does (1k)
hence the right-hand side of (50) has a DS. It implies that +Vl+1 also has a DS by Theorem 2.
So the proposition is proved. 2
Remark 3. If + Vl has DS then + Vl has DS if Vl is locally bounded, i.e., for all

compact set K there exists MK > 0 such that Vl (X) MK for all X K, because of Lemma 1
and
2

Vl dX Vol(Gd \ F )
Vl dX.
Gd \F
Gd \F
With the latter proposition proved it is straightforward to show that:

Proposition 2. Hl has a discrete spectrum, for all 1 l L.
Proof. H1 is the Hamiltonian for a harmonic oscillator, it then has a DS. One makes use of the
sequence proved in Proposition 1 to conclude that Hl has a DS. 2
550
To show that the regularized M5-brane has a discrete spectrum with eigenvalues tending to
infinity, we need to prove first that the p-brane has the maximum of its eigenvalues going to
infinity, since the proof of discreteness is not enough to prove a compact resolvent.
Proposition 3. The spectrum of Hl with 1 l L has eigenvalues n satisfying
n when n .
(51)
Proof. H1 is the Hamiltonian for a harmonic oscillator, it hasa compact resolvent, hence its
spectrum satisfies (51). We now consider any of the potentials Vl in the sequence of Proposition 1. Let us denote it V . It is of the form

V (X) = R n W ( ), where W ( ) = 1
(52)
R, are polar coordinates. We consider the neighborhood
of zeros of W ( ), i.e.,

= : W ( ) <
,
> 0.
We then define V
(X) = R n W
( ) where
X complement of
,
W
( ) = W ( ),
X
.
W
( ) =
,
(53)
(54)
(55)
For any
> 0, V
(X) as |X| , hence the spectrum of H
= + V
satisfy (51).
In fact, using the minmax theorem is easy to see that if we define a Hamiltonian Hwell =
+ W , where W is a potential well c constant in the region inside a ball S and 0 outside
it. And the ball S is taken in such a way that V
c outside it (this is always possible since
V
(X) as |X| ) then V
c + W so that n (H
) c + n (Hwell ). Using the fact that
n (Hwell ) 1 with n N , for some N , since W is a bounded potential of compact support, we
get n (H
) c 1 if n N but since c is arbitrary n (H
) as n . So, the spectrum
of H
satisfy (51).
When
0, the spectrum of H
could become continuous but we already know from Proposition 1 that it is discrete. It then follows property (51) also for the case when
0. 2
Proposition 4. The Hamiltonian of the regularized bosonic M5-brane H satisfies property (51).
Proof. We have H > H4 , from the above propositions H4 has a discrete spectrum satisfying
property (51). Theorem 2 ensures then n (H ) > n (H4 ) so the spectrum of H also satisfies
(51). H has a compact resolvent in the Hilbert space obtained by the completion of the subspace
generated by its eigenfunctions. 2
5. Conclusions
We showed the discreteness of the spectrum of the M5-brane and, in general, of p-brane theories. The condition is obtained from the Molchanov, Mazya and Shubin necessary and sufficient
condition on the potential of a Schrdinger operator to have a discrete spectrum. The criteria is
expressed not in terms of the behaviour of the potentials at each point, but by a mean value, on
the configuration space. The mean value in the sense of Molchanov considers the integral of the
551
potential on a finite region of configuration space. It can be naturally associated to a discretization of configuration space in the quantum theory. We found that the mean value in the direction
of the valleys where the potential is zero, at large distances in the configuration space, is the same
as that of a harmonic oscillator. Also, using the minmax principle it was shown that the discrete
spectra had eigenvalues running to infinity showing that their respective resolvent operators are
compact.
Acknowledgements
A.R. would like to thank Perimeter Institute for kind hospitality. Also, A.R. and I.M. are
grateful to E. Planchart and V. Strauss (USB), L. Boulton (Heriot-Watt U.), F. Cachazo (Perimeter
Institute) and M.P. Garcia del Moral (Turin U.) for fruitful discussions. This work was supported
by PROSUL under contract CNPq 490134/2006-8 and Decanato de Investigaciones y Desarrollo
(DID-USB), Proyecto G-11.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
B. de Wit, M. Luscher, H. Nicolai, Nucl. Phys. B 320 (1989) 135.

H. Nicolai, R. Helling, hep-th/9809103.
M. Lscher, Nucl. Phys. B 219 (1983) 233261.
B. de Wit, J. Hoppe, H. Nicolai, Nucl. Phys. B 305 (1988) 545.
B. de Wit, Marquard, H. Nicolai, Commun. Math. Phys. 128 (1990) 3962.
B. de Wit, K. Peeters, J. Plefka, Phys. Lett. B 409 (1997) 117123.
M.P. Garcia del Moral, A. Restuccia, Phys. Rev. D 66 (2002) 045023, hep-th/0103261.
L.S. Boulton, A. Restuccia, Nucl. Phys. B 724 (2005) 380396.
L.S. Boulton, M.P. Garcia del Moral, I. Martin, A. Restuccia, Class. Quantum Grav. 19 (2002) 2951, hep-th/
0109153.
L. Boulton, M.P. Garcia del Moral, A. Restuccia, Nucl. Phys. B 671 (2003) 343358, hep-th/0211047.
J. Bellorin, A. Restuccia, Nucl. Phys. B 737 (2006) 190208, hep-th/0510259.
I. Martin, A. Restuccia, Nucl. Phys. B 622 (2002) 240, hep-th/0108046.
M.P. Garcia del Moral, L. Navarro, A.J. Perez, A. Restuccia, Nucl. Phys. B 765 (2007) 287298, hep-th/0607234.
A.M. Molchanov, Proc. Moscow Math. Soc. 2 (1953) 169199 (in Russian).
V. Mazya, M. Shubin, Ann. Math. 162 (2005) 919942.
L. Boulton, M.P. Garcia del Moral, A. Restuccia, hep-th/0609054.
J. Maldacena, M.M. Sheikh-Jabbari, M. Van Raamsdonk, hep-th/0211139.
D.S. Paolo Pasti, M. Tonin, Phys. Lett. B 398 (1997) 41.
I. Bandos, K. Lechner, A. Nurmagambetov, P. Pasti, D. Sorokin, M. Tonin, Phys. Rev. Lett. 78 (1997) 4332.
C.P.M. Aganagic, J. Park, J. Schwarz, Nucl. Phys. B 496 (1997) 191.
A. De Castro, A. Restuccia, Nucl. Phys. B 617 (2001) 215, hep-th/0103123.
A. De Castro, M.P. Garcia del Moral, I. Martin, A. Restuccia, Phys. Lett. B 584 (2004) 171177.
J. Hoppe, Helv. Phys. Acta 70 (1997) 302.
Y. Nambu, Phys. Rev. D 7 (1973) 2405.
L. Takhtajan, Commun. Math. Phys. 160 (1994) 295, hep-th/9301111.
J. Grabowski, G. Marmo, math.DG/9902127.
J. Grabowski, G. Marmo, math.DG/9902128.
M. Reed, B. Simon, Methods of Modern Mathematical Physics, vol. IV: Analysis of Operators, Academic Press,
1978.

Some Implications of Perturbative Approach To Ads/Cft Correspondence

Transféré par

Informations du document

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Some Implications of Perturbative Approach To Ads/Cft Correspondence

Transféré par

Droits d'auteur :

Formats disponibles

Nuclear Physics B 794 (2008) 112

Some implications of perturbative approach

Wako, Saitama 351-0198, Japan

E-mail addresses: hkawai@gauge.scphys.kyoto-u.ac.jp (H. Kawai), suyama@gauge.scphys.kyoto-u.ac.jp

H. Kawai, T. Suyama / Nuclear Physics B 794 (2008) 112

H. Kawai, T. Suyama / Nuclear Physics B 794 (2008) 112

Fig. 1. Worldsheet configuration for the regularized amplitude.

Fig. 2. Feynman diagram corresponding to the worldsheet in Fig. 1.

H. Kawai, T. Suyama / Nuclear Physics B 794 (2008) 112

is XD = 0. Therefore, xD should be integrated in the worldsheet path-integral which indicates

Fig. 3. A Feynman diagram which could have an IR divergence.

H. Kawai, T. Suyama / Nuclear Physics B 794 (2008) 112

where i are the SO(8) gamma matrices.

H. Kawai, T. Suyama / Nuclear Physics B 794 (2008) 112

Fig. 4. LSZ-like reduction. The cross represents the insertion of S or |B0 .

is invariant under the scale transformation. Note that

describes the uniform distribution of the D-instanton.

where Fn contains the contributions from worldsheets with n boundaries.

H. Kawai, T. Suyama / Nuclear Physics B 794 (2008) 112

Since a single D-instanton is separated from the hypersurface along which

H. Kawai, T. Suyama / Nuclear Physics B 794 (2008) 112

is used. In other words, the free energy transforms as

In the region 4 , the metric (2.38) behaves as

H. Kawai, T. Suyama / Nuclear Physics B 794 (2008) 112

Fig. 5. Scale transformation in the T-dual picture.

H. Kawai, T. Suyama / Nuclear Physics B 794 (2008) 112

where w is the winding number, and

H. Kawai, T. Suyama / Nuclear Physics B 794 (2008) 112

H. Kawai, T. Suyama / Nuclear Physics B 794 (2008) 112

Nuclear Physics B 794 (2008) 1327

Stationary black holes and attractor mechanism

Seoul 151-747, South Korea

E-mail addresses: dumitru@th.phys.titech.ac.jp (D. Astefanesei), yavar@phya.snu.ac.kr (H. Yavartanoo).

D. Astefanesei, H. Yavartanoo / Nuclear Physics B 794 (2008) 1327

D. Astefanesei, H. Yavartanoo / Nuclear Physics B 794 (2008) 1327

D. Astefanesei, H. Yavartanoo / Nuclear Physics B 794 (2008) 1327

Symmetries: we assume time independence and axisymmetry;

D. Astefanesei, H. Yavartanoo / Nuclear Physics B 794 (2008) 1327

Finiteness of certain physical quantities;

D. Astefanesei, H. Yavartanoo / Nuclear Physics B 794 (2008) 1327

D. Astefanesei, H. Yavartanoo / Nuclear Physics B 794 (2008) 1327

Requiring F 2 remains finite at the horizon implies , , 0. We take , , = 0 so that

D. Astefanesei, H. Yavartanoo / Nuclear Physics B 794 (2008) 1327

where Veff ( i ) is a function of scalars fields i given by

Veff (i ) = f AB QA fAC P C QB fBD P D + fAB P A P B .

Thus, the black hole entropy is given by S = F (

D. Astefanesei, H. Yavartanoo / Nuclear Physics B 794 (2008) 1327

The field strength ansatz is

The attractor equations are:

4.2. Stationary black holes

D. Astefanesei, H. Yavartanoo / Nuclear Physics B 794 (2008) 1327

Based on this observation, a generalized entropy function was proposed in [9]

D. Astefanesei, H. Yavartanoo / Nuclear Physics B 794 (2008) 1327

D. Astefanesei, H. Yavartanoo / Nuclear Physics B 794 (2008) 1327

where ( ) = at ( ) ()a ( ).

D. Astefanesei, H. Yavartanoo / Nuclear Physics B 794 (2008) 1327

Maxwells equations are

From the r-component of Maxwells equation we obtain

which can be integrated to give

D. Astefanesei, H. Yavartanoo / Nuclear Physics B 794 (2008) 1327

D. Astefanesei, H. Yavartanoo / Nuclear Physics B 794 (2008) 1327

K. Saraikin, C. Vafa, Non-supersymmetric black holes and topological strings, hep-th/0703214;

Nuclear Physics B 794 (2008) 2845

Fig. 4. LSZ-like reduction. The cross represents the insertion of S or |B0 .

{Lff Z , Rff Z } = 2Tf3 2ef xW ff ,

{Lqq W , Rqq W } = { 2cW Vqq , 0},